diff --git a/aiter/ops/triton/flash_attn_triton_amd/__init__.py b/aiter/ops/triton/flash_attn_triton_amd/__init__.py new file mode 100644 index 0000000000..11a1f81c22 --- /dev/null +++ b/aiter/ops/triton/flash_attn_triton_amd/__init__.py @@ -0,0 +1,4 @@ +from . import interface_fa as flash_attn_2 +from . import interface_fa_v3 as flash_attn_3 + +__all__ = ["flash_attn_2", "flash_attn_3"] diff --git a/aiter/ops/triton/flash_attn_triton_amd/bench.py b/aiter/ops/triton/flash_attn_triton_amd/bench.py new file mode 100755 index 0000000000..7e3d17f611 --- /dev/null +++ b/aiter/ops/triton/flash_attn_triton_amd/bench.py @@ -0,0 +1,1763 @@ +import os +import sys +import torch +import triton +import time +import argparse +import itertools +import logging +import warnings +import datetime +import pandas as pd +from logging import warning +from typing import Dict, List, Literal, Optional, Tuple +from dataclasses import dataclass +from functools import lru_cache +from utils import get_arch, input_helper + +DEBUG = False + +ENV_FLAGS = [ + "FLASH_ATTENTION_TRITON_AMD_ENABLE", + "FLASH_ATTENTION_TRITON_AMD_AUTOTUNE", + "FLASH_ATTENTION_TRITON_AMD_DEBUG", +] + +FUNCTIONS = [ + "flash_attn_func", + "flash_attn_fp8_func", + "flash_attn_kvpacked_func", + "flash_attn_qkvpacked_func", + "flash_attn_qkvpacked_fp8_func", + "flash_attn_varlen_func", + "flash_attn_varlen_fp8_func", + "flash_attn_varlen_kvpacked_func", + "flash_attn_varlen_qkvpacked_func", + "flash_attn_varlen_qkvpacked_fp8_func", + "flash_attn_with_kvcache", +] + +SUPPORTED_DTYPES = { + "flash_attn_func": [torch.float16], + "flash_attn_fp8_func": [torch.float8_e4m3fnuz], + "flash_attn_kvpacked_func": [torch.float16], + "flash_attn_qkvpacked_func": [torch.float16], + "flash_attn_qkvpacked_fp8_func": [torch.float16], + "flash_attn_varlen_func": [torch.float16], + "flash_attn_varlen_fp8_func": [torch.float8_e4m3fnuz], + "flash_attn_varlen_kvpacked_func": [torch.float16], + "flash_attn_varlen_qkvpacked_func": [torch.float16], + "flash_attn_varlen_qkvpacked_fp8_func": [torch.float16], + "flash_attn_with_kvcache": [torch.float16], +} + +SUPPORTED_BACKENDS = { + "flash_attn_func": ["ck", "triton"], + "flash_attn_fp8_func": ["triton"], + "flash_attn_kvpacked_func": ["ck", "triton"], + "flash_attn_qkvpacked_func": ["ck", "triton"], + "flash_attn_qkvpacked_fp8_func": ["triton"], + "flash_attn_varlen_func": ["ck", "triton"], + "flash_attn_varlen_fp8_func": ["triton"], + "flash_attn_varlen_kvpacked_func": ["ck", "triton"], + "flash_attn_varlen_qkvpacked_func": ["ck", "triton"], + "flash_attn_varlen_qkvpacked_fp8_func": ["triton"], + "flash_attn_with_kvcache": ["ck", "triton"], +} + +VALID_MODES = ["fwd", "bwd", "full"] +SUPPORTED_MODES = { + "flash_attn_func": ["fwd", "bwd", "full"], + "flash_attn_fp8_func": ["fwd", "bwd", "full"], + "flash_attn_kvpacked_func": ["fwd", "bwd", "full"], + "flash_attn_qkvpacked_func": ["fwd", "bwd", "full"], + "flash_attn_qkvpacked_fp8_func": ["fwd", "bwd", "full"], + "flash_attn_varlen_func": ["fwd", "bwd", "full"], + "flash_attn_varlen_fp8_func": ["fwd", "bwd", "full"], + "flash_attn_varlen_kvpacked_func": ["fwd", "bwd", "full"], + "flash_attn_varlen_qkvpacked_func": ["fwd", "bwd", "full"], + "flash_attn_varlen_qkvpacked_fp8_func": ["fwd", "bwd", "full"], + "flash_attn_with_kvcache": ["fwd"], +} + + +# Add a global variable for verbose mode +VERBOSE = False + + +@dataclass +class EnvVariableConfig: + key: str + values: List[str] + backend: Optional[Literal["triton", "ck"]] = None + + +ENV_VARIABLE_CONFIGS: List[EnvVariableConfig] = [ + # EnvVariableConfig(key="BWD_MODE", values=["split", "fused_atomics", "fused_no_atomics"], backend="triton"), +] + + +class FunctionConfig: + def __init__( + self, + fn_name: str, + mode: Literal["fwd", "bwd", "full"], + dtype, + backend: Literal["triton", "ck"], + env_config: Dict, + ): + self.fn_name = fn_name + self.mode: Literal["fwd", "bwd", "full"] = mode + self.dtype = dtype + self.backend: Literal["triton", "ck"] = backend + self.arch = get_arch() + self.env_configs = env_config + + def __str__(self): + # extract base dtype name if it's a torch dtype + dtype_str = str(self.dtype) + if "torch." in dtype_str: + dtype_str = dtype_str.split(".")[-1] + + if len(self.env_configs) > 0: + env_str = "" + for env_key, env_value in self.env_configs.items(): + env_str += f"{env_key}={env_value}" + return f"{self.fn_name}_{self.mode}_{dtype_str}_{self.backend}_{self.arch}_{env_str}" + else: + return f"{self.fn_name}_{self.mode}_{dtype_str}_{self.backend}_{self.arch}" + + def column_name(self): + return f"{self}_ms" + + +def generate_fn_inputs( + fn_name: str, + BATCH: int, + HQ: int, + HK: int, + N_CTX_Q: int, + N_CTX_K: int, + D_HEAD: int, + CAUSAL: bool, + DROPOUT_P: float, + dtype: torch.dtype, + device: Literal["cpu", "cuda"], +): + if fn_name == "flash_attn_func": + return input_helper( + BATCH, + HQ, + HK, + N_CTX_Q, + N_CTX_K, + D_HEAD, + CAUSAL, + DROPOUT_P, + dtype, + layout="bshd", + device=device, + ) + elif fn_name == "flash_attn_kvpacked_func": + return input_helper( + BATCH, + HQ, + HK, + N_CTX_Q, + N_CTX_K, + D_HEAD, + CAUSAL, + DROPOUT_P, + dtype, + layout="bshd", + packing="kv", + device=device, + ) + elif fn_name == "flash_attn_qkvpacked_func": + return input_helper( + BATCH, + HQ, + HK, + N_CTX_Q, + N_CTX_K, + D_HEAD, + CAUSAL, + DROPOUT_P, + dtype, + layout="bshd", + packing="qkv", + device=device, + ) + elif fn_name == "flash_attn_varlen_func": + return input_helper( + BATCH, + HQ, + HK, + N_CTX_Q, + N_CTX_K, + D_HEAD, + CAUSAL, + DROPOUT_P, + dtype, + layout="thd", + device=device, + ) + elif fn_name == "flash_attn_varlen_kvpacked_func": + return input_helper( + BATCH, + HQ, + HK, + N_CTX_Q, + N_CTX_K, + D_HEAD, + CAUSAL, + DROPOUT_P, + dtype, + layout="thd", + packing="kv", + device=device, + ) + elif fn_name == "flash_attn_varlen_qkvpacked_func": + return input_helper( + BATCH, + HQ, + HK, + N_CTX_Q, + N_CTX_K, + D_HEAD, + CAUSAL, + DROPOUT_P, + dtype, + layout="thd", + packing="qkv", + device=device, + ) + elif fn_name == "flash_attn_with_kvcache": + return input_helper( + BATCH, + HQ, + HK, + N_CTX_Q, + N_CTX_K, + D_HEAD, + CAUSAL, + DROPOUT_P, + dtype, + layout="bshd", + device=device, + ) + elif fn_name == "flash_attn_fp8_func": + return input_helper( + BATCH, + HQ, + HK, + N_CTX_Q, + N_CTX_K, + D_HEAD, + CAUSAL, + DROPOUT_P, + dtype, + layout="bshd", + device=device, + ) + elif fn_name == "flash_attn_qkvpacked_fp8_func": + return input_helper( + BATCH, + HQ, + HK, + N_CTX_Q, + N_CTX_K, + D_HEAD, + CAUSAL, + DROPOUT_P, + dtype, + layout="bshd", + packing="qkv", + device=device, + ) + elif fn_name == "flash_attn_varlen_fp8_func": + return input_helper( + BATCH, + HQ, + HK, + N_CTX_Q, + N_CTX_K, + D_HEAD, + CAUSAL, + DROPOUT_P, + dtype, + layout="thd", + device=device, + ) + elif fn_name == "flash_attn_varlen_qkvpacked_fp8_func": + return input_helper( + BATCH, + HQ, + HK, + N_CTX_Q, + N_CTX_K, + D_HEAD, + CAUSAL, + DROPOUT_P, + dtype, + layout="thd", + packing="qkv", + device=device, + ) + else: + valid_fn_names = ", ".join(FUNCTIONS) + raise ValueError( + f"{fn_name} should be one of the following functions. {valid_fn_names}" + ) + + +def estimate_memory(config): + batch, hq, hk, sq, sk, d_head, causal, dropout = config + memory_estimate = batch * (hq * sq + hk * sk) * d_head * 4 # bytes + return memory_estimate + + +def generate_benchmark_configs( + is_varlen: bool, packing: Optional[Literal["kv", "qkv"]] +): + """ + generates a small number of configs that cover the parameter space well + """ + + # define all parameter options as lists + batch_sizes = [1, 64] + if packing == "qkv": + hq_values = hk_values = [2, 8] + sq_values = sk_values = [256, 8192] + else: + if ( + is_varlen + ): # make sure the seqlen is greater than the batchsize so that subsequences are greater than 0 + hq_values = [16, 32] # test mqa/gqa + hk_values = [8, 16] + sq_values = [128, 512] + sk_values = [512, 2024] + else: + hq_values = [64, 128] # test mqa/gqa + hk_values = [16, 64] + sq_values = [4, 4096] + sk_values = [4096, 16384] # test large k values for inference perf + d_head_values = [64, 128] + causal_values = [True, False] # most models usual causal True + dropout_values = [0.0, 0.1] + + # generate all fn_configs without inputs + input_configs = [] + + # one big loop to generate configs + for batch in batch_sizes: + for hq in hq_values: + for hk in hk_values: + for sq in sq_values: + for sk in sk_values: + for d_head in d_head_values: + for causal in causal_values: + for dropout in dropout_values: + # filter configs + input_config = ( + batch, + hq, + hk, + sq, + sk, + d_head, + causal, + dropout, + ) + + # skip if memory usage would be too high + if ( + estimate_memory(input_config) + > 8 * 1024 * 1024 * 1024 + ): # 8 GB limit + continue + + # we need hq to be a multiple of hk + if hq % hk != 0: + continue + + # for qkvpacked functions, q and k must have same dimensions + if packing == "qkv" and (sq != sk or hq != hk): + continue + + input_configs.append(input_config) + + return input_configs + + +def create_benchmark_fn( + flash_attn, fn_name, fn_input, mode: Literal["fwd", "bwd", "full"] +): + if DEBUG: + print("create_benchmark_fn") + print("flash_attn:", flash_attn) + print("fn_name:", fn_name) + print("fn_input:", len(fn_input)) + print("mode:", mode) + + if fn_name == "flash_attn_func": + q, k, v, do, metadata = fn_input + if mode == "fwd": + + def flash_attn_bench_fn(): + out, lse, S_dmask = flash_attn.flash_attn_func( + q, + k, + v, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + return out + + elif mode == "bwd": + out, lse, S_dmask = flash_attn.flash_attn_func( + q, + k, + v, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + + def flash_attn_bench_fn(): + dq, dk, dv = torch.autograd.grad(out, (q, k, v), do, retain_graph=True) + return dq, dk, dv + + elif mode == "full": + + def flash_attn_bench_fn(): + out, lse, S_dmask = flash_attn.flash_attn_func( + q, + k, + v, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + dq, dk, dv = torch.autograd.grad(out, (q, k, v), do, retain_graph=True) + return dq, dk, dv + + else: + raise ValueError(f"Unsupported benchmarking mode: {mode}") + + return flash_attn_bench_fn + + elif fn_name == "flash_attn_kvpacked_func": + q, kv, do, metadata = fn_input + if mode == "fwd": + + def flash_attn_kvpacked_bench_fn(): + out, lse, S_dmask = flash_attn.flash_attn_kvpacked_func( + q, + kv, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + return out + + elif mode == "bwd": + out, lse, S_dmask = flash_attn.flash_attn_kvpacked_func( + q, + kv, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + + def flash_attn_kvpacked_bench_fn(): + dq, dkv = torch.autograd.grad(out, (q, kv), do, retain_graph=True) + return dq, dkv + + elif mode == "full": + + def flash_attn_kvpacked_bench_fn(): + out, lse, S_dmask = flash_attn.flash_attn_kvpacked_func( + q, + kv, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + dq, dkv = torch.autograd.grad(out, (q, kv), do, retain_graph=True) + return dq, dkv + + else: + raise ValueError(f"Unsupported benchmarking mode: {mode}") + + return flash_attn_kvpacked_bench_fn + elif fn_name == "flash_attn_qkvpacked_func": + qkv, do, metadata = fn_input + if mode == "fwd": + + def flash_attn_qkvpacked_bench_fn(): + out, lse, S_dmask = flash_attn.flash_attn_qkvpacked_func( + qkv, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + return out + + elif mode == "bwd": + out, lse, S_dmask = flash_attn.flash_attn_qkvpacked_func( + qkv, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + + def flash_attn_qkvpacked_bench_fn(): + dqkv = torch.autograd.grad(out, (qkv), do, retain_graph=True) + return dqkv + + elif mode == "full": + + def flash_attn_qkvpacked_bench_fn(): + out, lse, S_dmask = flash_attn.flash_attn_qkvpacked_func( + qkv, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + dqkv = torch.autograd.grad(out, (qkv), do, retain_graph=True) + return dqkv + + else: + raise ValueError(f"Unsupported benchmarking mode: {mode}") + + return flash_attn_qkvpacked_bench_fn + elif fn_name == "flash_attn_varlen_func": + q_unpad, k_unpad, v_unpad, do_unpad, metadata = fn_input + if mode == "fwd": + + def flash_attn_varlen_bench_fn(): + out_unpad, lse, S_dmask = flash_attn.flash_attn_varlen_func( + q_unpad, + k_unpad, + v_unpad, + metadata.cu_seqlens_q, + metadata.cu_seqlens_k, + metadata.max_seqlens_q, + metadata.max_seqlens_k, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + return out_unpad + + elif mode == "bwd": + out_unpad, lse, S_dmask = flash_attn.flash_attn_varlen_func( + q_unpad, + k_unpad, + v_unpad, + metadata.cu_seqlens_q, + metadata.cu_seqlens_k, + metadata.max_seqlens_q, + metadata.max_seqlens_k, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + + def flash_attn_varlen_bench_fn(): + dq_unpad, dk_unpad, dv_unpad = torch.autograd.grad( + out_unpad, (q_unpad, k_unpad, v_unpad), do_unpad, retain_graph=True + ) + return dq_unpad, dk_unpad, dv_unpad + + elif mode == "full": + + def flash_attn_varlen_bench_fn(): + out_unpad, lse, S_dmask = flash_attn.flash_attn_varlen_func( + q_unpad, + k_unpad, + v_unpad, + metadata.cu_seqlens_q, + metadata.cu_seqlens_k, + metadata.max_seqlens_q, + metadata.max_seqlens_k, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + dq_unpad, dk_unpad, dv_unpad = torch.autograd.grad( + out_unpad, (q_unpad, k_unpad, v_unpad), do_unpad, retain_graph=True + ) + return dq_unpad, dk_unpad, dv_unpad + + else: + raise ValueError(f"Unsupported benchmarking mode: {mode}") + + return flash_attn_varlen_bench_fn + elif fn_name == "flash_attn_varlen_kvpacked_func": + q_unpad, kv_unpad, do_unpad, metadata = fn_input + if mode == "fwd": + + def flash_attn_varlen_kvpacked_bench_fn(): + out_unpad, lse, S_dmask = flash_attn.flash_attn_varlen_kvpacked_func( + q_unpad, + kv_unpad, + metadata.cu_seqlens_q, + metadata.cu_seqlens_k, + metadata.max_seqlens_q, + metadata.max_seqlens_k, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + return out_unpad + + elif mode == "bwd": + out_unpad, lse, S_dmask = flash_attn.flash_attn_varlen_kvpacked_func( + q_unpad, + kv_unpad, + metadata.cu_seqlens_q, + metadata.cu_seqlens_k, + metadata.max_seqlens_q, + metadata.max_seqlens_k, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + + def flash_attn_varlen_kvpacked_bench_fn(): + dq_unpad, dkv_unpad = torch.autograd.grad( + out_unpad, (q_unpad, kv_unpad), do_unpad, retain_graph=True + ) + return dq_unpad, dkv_unpad + + elif mode == "full": + + def flash_attn_varlen_kvpacked_bench_fn(): + out_unpad, lse, S_dmask = flash_attn.flash_attn_varlen_kvpacked_func( + q_unpad, + kv_unpad, + metadata.cu_seqlens_q, + metadata.cu_seqlens_k, + metadata.max_seqlens_q, + metadata.max_seqlens_k, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + dq_unpad, dkv_unpad = torch.autograd.grad( + out_unpad, (q_unpad, kv_unpad), do_unpad, retain_graph=True + ) + return dq_unpad, dkv_unpad + + else: + raise ValueError(f"Unsupported benchmarking mode: {mode}") + + return flash_attn_varlen_kvpacked_bench_fn + elif fn_name == "flash_attn_varlen_qkvpacked_func": + qkv_unpad, do_unpad, metadata = fn_input + if mode == "fwd": + + def flash_attn_varlen_qkvpacked_bench_fn(): + out_unpad, lse, S_dmask = flash_attn.flash_attn_varlen_qkvpacked_func( + qkv_unpad, + metadata.cu_seqlens_q, + metadata.max_seqlens_q, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + return out_unpad + + elif mode == "bwd": + out_unpad, lse, S_dmask = flash_attn.flash_attn_varlen_qkvpacked_func( + qkv_unpad, + metadata.cu_seqlens_q, + metadata.max_seqlens_q, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + + def flash_attn_varlen_qkvpacked_bench_fn(): + dqkv_unpad = torch.autograd.grad( + out_unpad, (qkv_unpad), do_unpad, retain_graph=True + ) + return dqkv_unpad + + elif mode == "full": + + def flash_attn_varlen_qkvpacked_bench_fn(): + out_unpad, lse, S_dmask = flash_attn.flash_attn_varlen_qkvpacked_func( + qkv_unpad, + metadata.cu_seqlens_q, + metadata.max_seqlens_q, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + dqkv_unpad = torch.autograd.grad( + out_unpad, (qkv_unpad), do_unpad, retain_graph=True + ) + return dqkv_unpad + + else: + raise ValueError(f"Unsupported benchmarking mode: {mode}") + + return flash_attn_varlen_qkvpacked_bench_fn + elif fn_name == "flash_attn_with_kvcache": + q, k_cache, v_cache, _, metadata = fn_input + if mode == "fwd": + + def flash_attn_with_kvcache_bench_fn(): + out = flash_attn.flash_attn_with_kvcache( + q, + k_cache, + v_cache, + None, + None, + rotary_cos=None, + rotary_sin=None, + cache_seqlens=None, + cache_batch_idx=None, + cache_leftpad=None, + block_table=None, + causal=metadata.causal, + window_size=(-1, -1), + rotary_interleaved=False, + alibi_slopes=None, + num_splits=0, + ) + return out + + else: + raise ValueError(f"Unsupported benchmarking mode: {mode}") + + return flash_attn_with_kvcache_bench_fn + elif fn_name == "flash_attn_fp8_func": + (q, descale_q), (k, descale_k), (v, descale_v), (do, descale_do), metadata = ( + fn_input + ) + if mode == "fwd": + + def flash_attn_f8_bench_fn(): + out, lse, S_dmask = flash_attn.flash_attn_fp8_func( + q, + k, + v, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + return out + + elif mode == "bwd": + out, lse, S_dmask = flash_attn.flash_attn_fp8_func( + q, + k, + v, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + + def flash_attn_f8_bench_fn(): + dq, dk, dv = torch.autograd.grad(out, (q, k, v), do, retain_graph=True) + return dq, dk, dv + + elif mode == "full": + + def flash_attn_f8_bench_fn(): + out, lse, S_dmask = flash_attn.flash_attn_fp8_func( + q, + k, + v, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + dq, dk, dv = torch.autograd.grad(out, (q, k, v), do, retain_graph=True) + return dq, dk, dv + + else: + raise ValueError(f"Unsupported benchmarking mode: {mode}") + + return flash_attn_f8_bench_fn + elif fn_name == "flash_attn_qkvpacked_fp8_func": + qkv, do, metadata = fn_input + if mode == "fwd": + + def flash_attn_qkvpacked_fp8_bench_fn(): + out, lse, S_dmask = flash_attn.flash_attn_qkvpacked_fp8_func( + qkv, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + return out + + elif mode == "bwd": + out, lse, S_dmask = flash_attn.flash_attn_qkvpacked_fp8_func( + qkv, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + + def flash_attn_qkvpacked_fp8_bench_fn(): + dqkv = torch.autograd.grad(out, (qkv), do, retain_graph=True) + return dqkv + + elif mode == "full": + + def flash_attn_qkvpacked_fp8_bench_fn(): + out, lse, S_dmask = flash_attn.flash_attn_qkvpacked_fp8_func( + qkv, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + dqkv = torch.autograd.grad(out, (qkv), do, retain_graph=True) + return dqkv + + else: + raise ValueError(f"Unsupported benchmarking mode: {mode}") + + return flash_attn_qkvpacked_fp8_bench_fn + elif fn_name == "flash_attn_varlen_fp8_func": + ( + (q_unpad, descale_q), + (k_unpad, descale_k), + (v_unpad, descale_v), + (do_unpad, descale_do), + metadata, + ) = fn_input + if mode == "fwd": + + def flash_attn_varlen_fp8_bench_fn(): + out_unpad, lse, S_dmask = flash_attn.flash_attn_varlen_fp8_func( + q_unpad, + k_unpad, + v_unpad, + metadata.cu_seqlens_q, + metadata.cu_seqlens_k, + metadata.max_seqlens_q, + metadata.max_seqlens_k, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + return out_unpad + + elif mode == "bwd": + out_unpad, lse, S_dmask = flash_attn.flash_attn_varlen_fp8_func( + q_unpad, + k_unpad, + v_unpad, + metadata.cu_seqlens_q, + metadata.cu_seqlens_k, + metadata.max_seqlens_q, + metadata.max_seqlens_k, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + + def flash_attn_varlen_fp8_bench_fn(): + dq_unpad, dk_unpad, dv_unpad = torch.autograd.grad( + out_unpad, (q_unpad, k_unpad, v_unpad), do_unpad, retain_graph=True + ) + return dq_unpad, dk_unpad, dv_unpad + + elif mode == "full": + + def flash_attn_varlen_fp8_bench_fn(): + out_unpad, lse, S_dmask = flash_attn.flash_attn_varlen_fp8_func( + q_unpad, + k_unpad, + v_unpad, + metadata.cu_seqlens_q, + metadata.cu_seqlens_k, + metadata.max_seqlens_q, + metadata.max_seqlens_k, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + dq_unpad, dk_unpad, dv_unpad = torch.autograd.grad( + out_unpad, (q_unpad, k_unpad, v_unpad), do_unpad, retain_graph=True + ) + return dq_unpad, dk_unpad, dv_unpad + + else: + raise ValueError(f"Unsupported benchmarking mode: {mode}") + + return flash_attn_varlen_fp8_bench_fn + elif fn_name == "flash_attn_varlen_qkvpacked_fp8_func": + qkv_unpad, do_unpad, metadata = fn_input + if mode == "fwd": + + def flash_attn_varlen_qkvpacked_fp8_bench_fn(): + out_unpad, lse, S_dmask = ( + flash_attn.flash_attn_varlen_qkvpacked_fp8_func( + qkv_unpad, + metadata.cu_seqlens_q, + metadata.max_seqlens_q, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + ) + return out_unpad + + elif mode == "bwd": + out_unpad, lse, S_dmask = flash_attn.flash_attn_varlen_qkvpacked_fp8_func( + qkv_unpad, + metadata.cu_seqlens_q, + metadata.max_seqlens_q, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + + def flash_attn_varlen_qkvpacked_fp8_bench_fn(): + dqkv_unpad = torch.autograd.grad( + out_unpad, (qkv_unpad), do_unpad, retain_graph=True + ) + return dqkv_unpad + + elif mode == "full": + + def flash_attn_varlen_qkvpacked_fp8_bench_fn(): + out_unpad, lse, S_dmask = ( + flash_attn.flash_attn_varlen_qkvpacked_fp8_func( + qkv_unpad, + metadata.cu_seqlens_q, + metadata.max_seqlens_q, + metadata.dropout_p, + causal=metadata.causal, + window_size=(-1, -1), + softcap=0.0, + alibi_slopes=None, + deterministic=False, + return_attn_probs=True, + ) + ) + dqkv_unpad = torch.autograd.grad( + out_unpad, (qkv_unpad), do_unpad, retain_graph=True + ) + return dqkv_unpad + + else: + raise ValueError(f"Unsupported benchmarking mode: {mode}") + + return flash_attn_varlen_qkvpacked_fp8_bench_fn + else: + valid_fn_names = ", ".join(FUNCTIONS) + raise ValueError( + f"{fn_name} should be one of the following functions. {valid_fn_names}" + ) + + +def get_packing_type(fn_name: str) -> Optional[Literal["kv", "qkv"]]: + if "_kvpacked" in fn_name: + packing = "kv" + elif "_qkvpacked" in fn_name: + packing = "qkv" + else: + packing = None + + return packing + + +def load_flash_attn_module(backend: Literal["triton", "ck"], env_configs: Dict = {}): + """ + Load the flash_attn module with the specified backend configuration + """ + global VERBOSE + + # remove any existing env variables first + for key in ENV_FLAGS: + if key in os.environ: + del os.environ[key] + + # set environment variable for the desired backend + if backend == "triton": + os.environ["FLASH_ATTENTION_TRITON_AMD_ENABLE"] = "TRUE" + os.environ["FLASH_ATTENTION_TRITON_AMD_DEBUG"] = "0" + os.environ["FLASH_ATTENTION_TRITON_AMD_AUTOTUNE"] = "1" + elif backend == "ck": + os.environ["FLASH_ATTENTION_TRITON_AMD_ENABLE"] = "FALSE" + else: + raise ValueError(f"Unknown backend {backend}") + + # add custom env configs + add_env_configs(env_configs) + + if VERBOSE: # Only print if both local and global verbose are True + print(f"Loading flash_attn module with {backend} backend.") + + # Remove any existing flash_attn modules from sys.modules + for module_name in list(sys.modules.keys()): + if module_name.startswith("flash_attn"): + del sys.modules[module_name] + + # Clear CUDA cache + torch.cuda.empty_cache() + + # Import and return the module + import flash_attn + + # disable triton printing from autotuning + if not VERBOSE: + os.environ["TRITON_PRINT_AUTOTUNING"] = "0" + + return flash_attn + + +def add_env_configs(env_config: Dict): + for env_key, env_value in env_config.items(): + if env_key in os.environ: + del os.environ[ + env_key + ] # remove previous version so that env key is the latest key added + os.environ[env_key] = env_value + + +def run_benchmark(func_config: FunctionConfig, input_configs): + """ + Runs the benchmark for the provided function configuration with the given input configurations. + """ + global VERBOSE + + # extract function configuration parameters + fn_name = func_config.fn_name + mode = func_config.mode + dtype = func_config.dtype + backend = func_config.backend + + # load flash attention module + flash_attn_module = load_flash_attn_module(backend, func_config.env_configs) + + # start timing the benchmark + start_time = time.time() + if VERBOSE: + print(f"Benchmarking {func_config} ...") + else: + print(f"Running {fn_name} ({mode}, {backend})...", end="", flush=True) + + # Setup benchmark configurations + bench_configs = [ + triton.testing.Benchmark( + x_names=[ + "BATCH", + "HQ", + "HK", + "N_CTX_Q", + "N_CTX_K", + "D_HEAD", + "CAUSAL", + "DROPOUT", + ], + x_vals=list(input_configs.keys()), + line_arg="provider", + line_vals=["triton"], + line_names=["Time (ms)"], + styles=[("red", "-")], + ylabel="ms", + plot_name=f"benchmark-{func_config}", + args={}, + ) + ] + + @triton.testing.perf_report(bench_configs) + def bench_function( + BATCH, + HQ, + HK, + N_CTX_Q, + N_CTX_K, + D_HEAD, + CAUSAL, + DROPOUT, + provider, + device="cuda", + ): + if DEBUG: + print("BATCH:", BATCH) + print("HQ:", HQ) + print("HK:", HK) + print("N_CTX_Q:", N_CTX_Q) + print("N_CTX_Q:", N_CTX_Q) + print("D_HEAD:", D_HEAD) + print("CAUSAL:", CAUSAL) + print("DROPOUT:", DROPOUT) + print("mode:", mode) + print("provider:", provider) + print("device:", device) + fn_input = input_configs[ + (BATCH, HQ, HK, N_CTX_Q, N_CTX_K, D_HEAD, CAUSAL, DROPOUT) + ] + benchmark_fn = create_benchmark_fn(flash_attn_module, fn_name, fn_input, mode) + + # run the benchmark + ms = triton.testing.do_bench(benchmark_fn, warmup=25, rep=100) + return ms + + df = bench_function.run(return_df=True)[0] + + # set the column name to reflect the function configuration + df = df.rename(columns={"Time (ms)": func_config.column_name()}) + + # calculate and print elapsed time + elapsed_time = time.time() - start_time + + return df, elapsed_time + + +def filter_modes(requested_modes, fn_name, supported_modes_for_fn): + modes_to_run = [] + if requested_modes: + for mode in requested_modes: + if mode in supported_modes_for_fn: + modes_to_run.append(mode) + else: + warning( + f"Mode '{mode}' requested but not supported by function '{fn_name}'. Skipping this mode for this function." + ) + else: + modes_to_run = ["full" if "full" in supported_modes_for_fn else "fwd"] + return modes_to_run + + +def get_env_value_combinations( + current_backend: Optional[Literal["triton", "ck"]], +) -> List[Dict[str, str]]: + # filter environment variations applicable to the current backend + applicable_variations = [ + var_config + for var_config in ENV_VARIABLE_CONFIGS + if var_config.backend is None or var_config.backend == current_backend + ] + + if not applicable_variations: + # no applicable variations, return list with empty dict + return [{}] + + # prepare keys and value lists + variation_keys = [v.key for v in applicable_variations] + variation_value_lists = [v.values for v in applicable_variations] + + # generate all combinations as dictionaries directly + env_configs = [] + for value_combination in itertools.product(*variation_value_lists): + env_configs.append(dict(zip(variation_keys, value_combination))) + + return env_configs + + +def get_input_config_set(config_type): + if config_type == "llama": + # batch, hq, hk, sq, sk, d_head, causal, dropout + input_configs = [ + # LLaMA 3 8B + (4, 32, 8, 8192, 8192, 128, True, 0.0), + # LLaMA 3 70B + (4, 64, 8, 8192, 8192, 128, True, 0.0), + ] + else: + raise ValueError(f"Unknown input config: {config_type}") + + return input_configs + + +def available_backends(): + """Check which backends are available by trying to load them.""" + available = [] + + for backend in ["triton", "ck"]: + try: + # try loading the module with this backend + load_flash_attn_module(backend) + available.append(backend) + except Exception as e: + # backend not available, just continue + if DEBUG: + print(f"Backend {backend} not available: {e}") + + if not available: + raise ValueError( + "No backends are available. Please check your flash_attn installation." + ) + + return available + + +# 2. Simplify get_fn_params to remove the backend filtering logic here +@lru_cache() +def get_fn_params(fn_name): + # get params for fn + packing = get_packing_type(fn_name) + is_varlen = True if "varlen" in fn_name else False + is_fp8 = True if "fp8" in fn_name else False + supported_dtypes = SUPPORTED_DTYPES.get(fn_name, [torch.float16]) + supported_backends = SUPPORTED_BACKENDS.get( + fn_name, ["triton"] + ) # just get what the function supports + supports_backward = False if fn_name in ["flash_attn_with_kvcache"] else True + supported_modes = SUPPORTED_MODES.get(fn_name, ["fwd"]) + device = "cuda" + + # get supported env configs for each backend + supported_env_configs = {} + for backend in supported_backends: + supported_env_configs[backend] = get_env_value_combinations(backend) + + # check backward pass support + if not supports_backward: + warning( + f"{fn_name} does not have a backward pass so benching forward pass only." + ) + + return ( + is_varlen, + is_fp8, + packing, + supported_dtypes, + supported_backends, + supported_modes, + supported_env_configs, + device, + ) + + +# 3. Create a new simpler function to validate and filter backends +def validate_backends(requested_backends, supported_backends, fn_name): + """Validate that requested backends are available and supported.""" + # get actually available backends + available = available_backends() + + # determine which backends to use + if requested_backends: + # user specified backends - validate them + valid_backends = [] + for backend in requested_backends: + if backend not in available: + warning( + f"Backend '{backend}' is not available on this system. Skipping." + ) + continue + if backend not in supported_backends: + warning( + f"Backend '{backend}' is not supported by function '{fn_name}'. Skipping." + ) + continue + valid_backends.append(backend) + + if not valid_backends: + raise ValueError( + f"None of the requested backends {requested_backends} are available and supported for {fn_name}" + ) + + return valid_backends + else: + # no backends specified - use all available and supported + valid_backends = [b for b in supported_backends if b in available] + + if not valid_backends: + raise ValueError( + f"No available backends found for {fn_name}. Function supports {supported_backends} but only {available} are available." + ) + + return valid_backends + + +# 4. Update process_args to use the new validate_backends function +def process_args(): + """ + Parses command-line arguments and returns function configs and input configs. + """ + global VERBOSE + + # create parser + parser = argparse.ArgumentParser( + prog="Benchmark FlashAttention", + allow_abbrev=False, + ) + # functions + parser.add_argument( + "-benchmark_fn", + type=str, + nargs="*", + choices=FUNCTIONS, + required=True, + help=f"Function(s) to benchmark", + ) + parser.add_argument( + "--mode", + type=str, + nargs="*", + choices=VALID_MODES, + default=["fwd", "bwd"], + help=f"Benchmarking mode(s) to run. Default: fwd, bwd", + ) + parser.add_argument( + "--backend", + type=str, + nargs="*", + choices=["triton", "ck"], + default=["triton"], + help="Backend(s) to run. Default: triton", + ) + parser.add_argument( + "--output", + type=str, + choices=["ms", "tflops"], + default="tflops", + help="Output metric type: ms (milliseconds) or tflops (TFLOPS). Default: tflops", + ) + parser.add_argument( + "--format", + type=str, + choices=["csv", "markdown"], + default="csv", + help="Output file format: csv or markdown. Default: csv", + ) + parser.add_argument( + "--verbose", + "-v", + action="store_true", + help="Enable verbose output (show autotuning details)", + ) + # config + parser.add_argument("-b", type=int, default=None, help="Batch size") + parser.add_argument("-hq", type=int, default=None, help="Q Number of heads") + parser.add_argument("-hk", type=int, default=None, help="K and V Number of heads") + parser.add_argument("-sq", type=int, default=None, help="Q Sequence Length") + parser.add_argument("-sk", type=int, default=None, help="K and V Sequence Length") + parser.add_argument("-d", type=int, default=None, help="Head Dimension") + parser.add_argument("-causal", action="store_true", default=None, help="Causal") + parser.add_argument("-dropout", type=float, default=None, help="Dropout") + + # parse args + args = parser.parse_args() + + # Set global verbose flag + VERBOSE = args.verbose + + # parse function args + benchmark_fns = args.benchmark_fn + requested_modes = args.mode + requested_backends = args.backend + output_type: Literal["ms", "tflops"] = args.output + output_format: Literal["csv", "markdown"] = args.format + + # generate function configurations and input configurations separately + all_function_configs = [] + all_input_configs = {} # Maps function config -> input configs + + for fn_name in benchmark_fns: + ( + is_varlen, + is_fp8, + packing, + supported_dtypes, + supported_backends, + supported_modes_for_fn, + supported_env_configs, + device, + ) = get_fn_params(fn_name) + + # Generate or use custom input configurations + if args.b or args.hq or args.hk or args.sq or args.sk or args.d: + assert ( + args.b and args.hq and args.sq and args.d + ), "if custom config is specified, please provide at least batch, number of Q heads, Q sequence length, and head size." + + batch = args.b + hq = args.hq + hk = args.hk if args.hk is not None else args.hq + sq = args.sq + sk = args.sk if args.sk is not None else args.sq + d_head = args.d + causal = args.causal if args.causal is not None else False + dropout = args.dropout if args.dropout is not None else 0.0 + input_configs = [(batch, hq, hk, sq, sk, d_head, causal, dropout)] + else: + input_configs = get_input_config_set("llama") + + # filter by mode + modes_to_run = filter_modes(requested_modes, fn_name, supported_modes_for_fn) + if not modes_to_run: + warning( + f"No valid modes to run for function '{fn_name}' based on request and function support. Skipping this function." + ) + continue + + # validate and filter backends + try: + backends_to_run = validate_backends( + requested_backends, supported_backends, fn_name + ) + except ValueError as e: + warning(str(e)) + continue + + # create a function config for each backend and dtype combination + for backend in backends_to_run: + for dtype in supported_dtypes: + for mode in modes_to_run: + for env_config in supported_env_configs[backend]: + func_config = FunctionConfig( + fn_name, mode, dtype, backend, env_config + ) + all_function_configs.append(func_config) + + # Generate inputs for this function configuration + fn_inputs = {} + for input_config in input_configs: + fn_inputs[input_config] = generate_fn_inputs( + fn_name, *input_config, dtype, device + ) + + all_input_configs[func_config] = fn_inputs + + return all_function_configs, all_input_configs, output_type, output_format + + +def check_environment_variables(): + for key in ENV_FLAGS: + if key in os.environ: + raise ValueError( + f"Running with {key} environment variable is not recommended for the benching script. Use --help to see how to use the benching script." + ) + + +def compute_flops(batch, hq, hk, sq, sk, d_head, causal): + # 2 FLOPs per multiply‑add + if causal: + valid_pairs = ( + (sk * (sk + 1)) // 2 if sq > sk else sq * sk - (sq * (sq - 1)) // 2 + ) + else: + valid_pairs = sq * sk + return 2 * batch * hq * valid_pairs * d_head + + +# see ref, https://github.com/ROCm/aiter/blob/jukorhon/mha-bwd/op_benchmarks/triton/bench_mha.py +def _flops_single_row(row: pd.Series, mode: str) -> float: + b, hq, d_head = int(row["BATCH"]), int(row["HQ"]), int(row["D_HEAD"]) + sq, sk = int(row["N_CTX_Q"]), int(row["N_CTX_K"]) + causal = bool(row["CAUSAL"]) + + # -------- number of (query, key) products per head ---------------- + if not causal: + valid_pairs = sq * sk + else: # triangular mask + if sq > sk: + valid_pairs = sk * (sk + 1) // 2 + (sq - sk) * sk + else: # sq <= sk + valid_pairs = sq * (sq + 1) // 2 + + # one matmul FLOPs (mul + add) = 2 · m · n · k + flops_per_matmul = 2.0 * b * hq * valid_pairs * d_head + total_flops = 2.0 * flops_per_matmul # 2 matmuls in forward + + if mode == "fwd": + pass + elif mode == "bwd": + total_flops *= 2.5 # 2·bwd + 0.5·recompute + elif mode == "full": + total_flops *= 3.5 # fwd + bwd + else: + raise ValueError(f"unknown mode {mode}") + + return total_flops + + +def add_tflops_columns(df: pd.DataFrame, func_cfg: FunctionConfig) -> pd.DataFrame: + ms_col = func_cfg.column_name() + tf_col = ms_col.replace("_ms", "_tflops") + flops = df.apply(_flops_single_row, axis=1, mode=func_cfg.mode) + df[tf_col] = flops / df[ms_col] * 1e-9 + return df + + +def generate_output_filename(function_configs, output_type, output_format): + # create a timestamp + timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S") + + # simple filename format + base_filename = f"benchmark_{timestamp}" + + if output_format == "csv": + return base_filename + ".csv" + else: # markdown + return base_filename + ".md" + + +def main(): + """ + Main function to run benchmarks. + """ + global VERBOSE + + # check environment variables + check_environment_variables() + + # start timing the entire benchmarking process + total_start_time = time.time() + + # process args to get function configs and input configs + function_configs, all_input_configs, output_type, output_format = process_args() + + # Print summary of what will be benchmarked (always show this) + print(f"\nBenchmarking {len(function_configs)} configuration(s):") + unique_fns = set(fc.fn_name for fc in function_configs) + print(f" Functions: {', '.join(unique_fns)}") + unique_backends = set(fc.backend for fc in function_configs) + print(f" Backends: {', '.join(unique_backends)}") + unique_modes = set(fc.mode for fc in function_configs) + print(f" Modes: {', '.join(unique_modes)}") + print() + + # run benchmarks for each function configuration + combined_ms_df = None + combined_tf_df = None + input_cols = [ + "BATCH", + "HQ", + "HK", + "N_CTX_Q", + "N_CTX_K", + "D_HEAD", + "CAUSAL", + "DROPOUT", + ] + + for i, func_config in enumerate(function_configs, 1): + # Progress indicator + if not VERBOSE: + print(f"[{i}/{len(function_configs)}] ", end="") + + # run benchmark with the input configs for this function config + input_configs = all_input_configs[func_config] + df, elapsed_time = run_benchmark(func_config, input_configs) + + if VERBOSE: + print( + f"Total time for benchmarking {func_config.fn_name} in {func_config.mode} mode with {func_config.dtype}: {elapsed_time:.2f} seconds" + ) + + # add to combined table + df = add_tflops_columns(df, func_config) + ms_cols = [c for c in df.columns if c.endswith("_ms")] + tf_cols = [c for c in df.columns if c.endswith("_tflops")] + + ms_df = df[input_cols + ms_cols] + tf_df = df[input_cols + tf_cols] + + if combined_ms_df is None: + combined_ms_df = ms_df + combined_tf_df = tf_df + else: + combined_ms_df = combined_ms_df.merge(ms_df, on=input_cols, how="outer") + combined_tf_df = combined_tf_df.merge(tf_df, on=input_cols, how="outer") + + # print new line to seperate the combined data information from the benchmark specific information + print() + + # print total time for all benchmarks + total_elapsed_time = time.time() - total_start_time + print(f"Total benchmark time: {total_elapsed_time:.1f} seconds") + + # save combined data and make comparisons if we have multiple function configs + has_multiple_func_configs = False # len(function_configs) > 1 + if has_multiple_func_configs: + if len(function_configs) == 2: + func1 = function_configs[0] + func2 = function_configs[1] + + # construct column names for the timing results + col1 = func1.column_name() + col2 = func2.column_name() + + # Check if we're comparing triton vs ck (in either order) + is_triton_vs_ck = (func1.backend == "triton" and func2.backend == "ck") or ( + func1.backend == "ck" and func2.backend == "triton" + ) + + # For triton vs ck comparisons + if is_triton_vs_ck: + # For triton vs ck comparisons, always make triton the baseline + if func1.backend == "triton" and func2.backend == "ck": + triton_col = col1 + ck_col = col2 + ratio_col = f"ck_to_triton_ratio" + else: + triton_col = col2 + ck_col = col1 + ratio_col = f"ck_to_triton_ratio" + + # Calculate ratio: ck_time / triton_time (values > 1 mean triton is faster) + combined_ms_df[ratio_col] = ( + combined_ms_df[ck_col] / combined_ms_df[triton_col] + ) + + # print explanation + print(f"Comparison Results (triton vs ck):") + print( + f"Ratio values: values > 1 mean triton is faster (by that factor), values < 1 mean ck is faster" + ) + + # output based on selected metric + if output_type == "ms": + if combined_ms_df is not None: + filename = generate_output_filename(function_configs, "ms", output_format) + print(f"\nCombined wall-time (ms) table:") + print(combined_ms_df) + + if output_format == "csv": + combined_ms_df.to_csv(filename, index=False) + print(f"Results saved to: {filename}") + else: # markdown + with open(filename, "w") as f: + f.write(combined_ms_df.to_markdown(index=False, floatfmt=".2f")) + print(f"Results saved to: {filename}") + else: # output_type == "tflops" + if combined_tf_df is not None: + filename = generate_output_filename( + function_configs, "tflops", output_format + ) + print(f"\nCombined throughput (TFLOPs) table:") + print(combined_tf_df) + + if output_format == "csv": + combined_tf_df.to_csv(filename, index=False) + print(f"Results saved to: {filename}") + else: # markdown + with open(filename, "w") as f: + f.write(combined_tf_df.to_markdown(index=False, floatfmt=".2f")) + print(f"Results saved to: {filename}") + + +if __name__ == "__main__": + main() diff --git a/aiter/ops/triton/flash_attn_triton_amd/bwd_prefill_fused_atomics.py b/aiter/ops/triton/flash_attn_triton_amd/bwd_prefill_fused_atomics.py new file mode 100755 index 0000000000..d742d9b712 --- /dev/null +++ b/aiter/ops/triton/flash_attn_triton_amd/bwd_prefill_fused_atomics.py @@ -0,0 +1,2578 @@ +import torch +import triton +import triton.language as tl +from .utils import compute_fp8_scaling_factors, DEBUG, is_fp8 + +from typing import Optional, Tuple + + +# This function computes delta given output Out and gradient DO +# Here is the I/O shape: +# Out: (batch, nhead_q, max_seqlens_q, headDim) +# DO: (batch, nhead_q, max_seqlens_q, headDim) +# Delta: (batch, nheads_q, max_seqlens_q), same as softmax_lse defined at +@triton.jit +def _bwd_preprocess( + o_ptr, + do_ptr, # noqa: E741 + delta_ptr, + stride_o_b, + stride_o_h, + stride_o_m, + stride_o_k, + stride_delta_b, + stride_delta_h, + stride_delta_m, + stride_descale_do_z, + cu_seqlens_q, + max_seqlen_q, + descale_do_ptr, + BLOCK_M: tl.constexpr, + BLOCK_D_MODEL: tl.constexpr, + BLOCK_D_MODEL_POW2: tl.constexpr, + IS_VARLEN: tl.constexpr, + IS_FP8: tl.constexpr, +): + pid_m = tl.program_id(0) # seqlen + bid = tl.program_id(1) # batch + hid = tl.program_id(2) # head + + # Handle varlen + q_start = 0 + seqlen_q = max_seqlen_q + if IS_VARLEN: + q_start = tl.load(cu_seqlens_q + bid) + q_end = tl.load(cu_seqlens_q + bid + 1) + seqlen_q = q_end - q_start + else: + q_start = 0 + seqlen_q = max_seqlen_q + + # Compute offsets + offs_m = pid_m * BLOCK_M + tl.arange(0, BLOCK_M) + offs_k = tl.arange(0, BLOCK_D_MODEL_POW2) + + # Offset O/DO by batch, head and q_start + offs = ( + bid * stride_o_b + + hid * stride_o_h + + q_start * stride_o_m + + offs_m[:, None] * stride_o_m + + offs_k[None, :] * stride_o_k + ) + + # create masks + mask_m = offs_m < seqlen_q + mask = mask_m[:, None] + PADDED_HEAD: tl.constexpr = BLOCK_D_MODEL != BLOCK_D_MODEL_POW2 + if PADDED_HEAD: + mask &= offs_k[None, :] < BLOCK_D_MODEL + + # load [BLOCK_M, BLOCK_D_MODEL_POW2] + o = tl.load(o_ptr + offs, mask=mask, other=0.0) + do = tl.load(do_ptr + offs, mask=mask, other=0.0) + + # compute and write-back to delta + if IS_FP8: + descale_do = tl.load(descale_do_ptr + bid * stride_descale_do_z + hid) + + # NOTE: do is in the fp8 range and o is not in fp8 + delta = tl.sum(o.to(tl.float32) * (do.to(tl.float32) * descale_do), axis=1) + else: + delta = tl.sum(o.to(tl.float32) * do.to(tl.float32), axis=1) + + offs_delta = ( + bid * stride_delta_b + + hid * stride_delta_h + + q_start * stride_delta_m + + offs_m * stride_delta_m + ) + tl.store(delta_ptr + offs_delta, delta, mask=mask_m) + + +@triton.jit +def _bwd_dq_inner( + dq, + q, + K, + V, + do, + m, + Delta, + sm_scale, + stride_qm, + stride_qk, + stride_kn, + stride_kk, + stride_vn, + stride_vk, + stride_dropout_m, + stride_dropout_n, + stride_deltam, + seqlen_q, + seqlen_k, + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, + start_m, + start_n, + end_n, + num_steps, + descale_q, + descale_k, + descale_v, + descale_do, + BLOCK_M: tl.constexpr, + BLOCK_N: tl.constexpr, + BLOCK_D_MODEL: tl.constexpr, + BLOCK_D_MODEL_POW2: tl.constexpr, + MASK: tl.constexpr, + ENABLE_DROPOUT: tl.constexpr, + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, +): + RCP_LN2: tl.constexpr = 1.4426950408889634 + + PADDED_HEAD: tl.constexpr = BLOCK_D_MODEL != BLOCK_D_MODEL_POW2 + delta_qk = seqlen_q - seqlen_k + offs_m = start_m + tl.arange(0, BLOCK_M) + offs_n = start_n + tl.arange(0, BLOCK_N) + offs_k = tl.arange(0, BLOCK_D_MODEL_POW2) + + # mask to make sure not OOB of seqlen_q + mask_m = offs_m < seqlen_q + + kT_ptrs = K + offs_n[None, :] * stride_kn + offs_k[:, None] * stride_kk + vT_ptrs = V + offs_n[None, :] * stride_vn + offs_k[:, None] * stride_vk + + # D (= delta) is pre-divided by ds_scale. + Di = tl.load(Delta + offs_m * stride_deltam, mask=mask_m, other=0.0) + + curr_n = start_n + step_n = BLOCK_N + curr_philox_offset = batch_philox_offset + curr_dropout_offset = dropout_offset + for blk_idx in range(num_steps): + offs_n = curr_n + tl.arange(0, BLOCK_N) + # end_n is needed because the end of causal True might not be perfectly + # aligned with the end of the block + mask_n = offs_n < end_n + mask_kT = mask_n[None, :] + mask_mn = mask_m[:, None] & (offs_n[None, :] < end_n) + if PADDED_HEAD: + mask_kT &= offs_k[:, None] < BLOCK_D_MODEL + + kT = tl.load(kT_ptrs, mask=mask_kT, other=0.0) + vT = tl.load(vT_ptrs, mask=mask_kT, other=0.0) + + # dropout + if ENABLE_DROPOUT: + philox_offs = ( + curr_philox_offset + + offs_m[:, None] * stride_dropout_m + + offs_n[None, :] * stride_dropout_n + ) + rand_vals = tl.rand(philox_seed, philox_offs) + dropout_mask = rand_vals > dropout_p + dropout_scale = 1 / (1 - dropout_p) + + # qk + if IS_FP8: + qk = tl.dot(q, kT) * descale_q * descale_k + else: + qk = tl.dot(q, kT) + p = tl.math.exp2(qk * sm_scale * RCP_LN2 - m * RCP_LN2) + + if MASK: + causal_mask = (offs_m[:, None] - delta_qk) >= offs_n[None, :] + mask = causal_mask * mask_mn + p = tl.where(mask, p, 0.0) + + # dp + if IS_FP8: + dp = tl.dot(do, vT) * descale_do * descale_v + else: + dp = tl.dot(do, vT) + + if ENABLE_DROPOUT: + dp = tl.where(dropout_mask, dp, 0.0) * dropout_scale + + # ds + delta_i = Di[:, None] + ds = p * (dp - delta_i) + + # dq + # NOTE: We need to de-scale dq in the end, because kT was pre-scaled. + if IS_FP8: + scale_ds, descale_ds = compute_fp8_scaling_factors(ds, FP8_MAX) + dq += ( + tl.dot((ds * scale_ds).to(kT.type.element_ty), tl.trans(kT)) + * descale_ds + * descale_k + ) + else: + dq += tl.dot(ds.to(kT.type.element_ty), tl.trans(kT)) + + curr_n += step_n + kT_ptrs += step_n * stride_kn + vT_ptrs += step_n * stride_vn + return dq + + +@triton.jit +def _bwd_dkdv_inner( + dk, + dv, + Q, + k, + v, + DO, + M, + D, + sm_scale, + stride_q_m, + stride_q_k, + stride_do_m, + stride_do_k, + stride_dropout_m, + stride_dropout_n, + stride_deltam, + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, + seqlen_q, + seqlen_k, + start_n, + start_m, + num_steps, + descale_q, + descale_k, + descale_v, + descale_do, + BLOCK_M: tl.constexpr, + BLOCK_N: tl.constexpr, + BLOCK_D_MODEL: tl.constexpr, + BLOCK_D_MODEL_POW2: tl.constexpr, + MASK: tl.constexpr, + ENABLE_DROPOUT: tl.constexpr, + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, +): + PADDED_HEAD: tl.constexpr = BLOCK_D_MODEL != BLOCK_D_MODEL_POW2 + delta_qk = seqlen_q - seqlen_k + offs_m = start_m + tl.arange(0, BLOCK_M) + offs_n = start_n + tl.arange(0, BLOCK_N) + offs_k = tl.arange(0, BLOCK_D_MODEL_POW2) + + # mask to make sure not OOB of seqlen_q + mask_n = offs_n < seqlen_k + qT_ptrs = ( + Q + offs_m[None, :] * stride_q_m + offs_k[:, None] * stride_q_k + ) # [BLOCK_D_MODEL_POW2, BLOCK_M] + do_ptrs = DO + offs_m[:, None] * stride_do_m + offs_k[None, :] * stride_do_k + curr_m = start_m + step_m = BLOCK_M + curr_philox_offset = batch_philox_offset + curr_dropout_offset = dropout_offset + RCP_LN2: tl.constexpr = 1.4426950408889634 + + # Iterate over blocks(BLOCK_M size) of Q while calculating + # a fixed block(BLOCK_N) of dk and dv. Note, during backward + # pass P has to be recomputed. However, this kernel computes + # dV and dK, so we compute we need P^T and S^T. See backward pass + # equations + # + # From Flash Attention Paper: + # ForwardPass: S = QkT, P=softmax(S), O=PV + # + # BackwardPass equations + # dV = P^TdO + # dP = dOV^T + # dS = dsoftmax(dP) + # dQ = dSK + # dK = QdS^T + for blk_idx in range(num_steps): + offs_m = curr_m + tl.arange(0, BLOCK_M) + mask_m = offs_m < seqlen_q + mask_qT = mask_m[None, :] + mask_do = mask_m[:, None] + mask_nm = mask_n[:, None] & (offs_m[None, :] < seqlen_q) + if PADDED_HEAD: + mask_qT &= offs_k[:, None] < BLOCK_D_MODEL + mask_do &= offs_k[None, :] < BLOCK_D_MODEL + + # load qT + qT = tl.load(qT_ptrs, mask=mask_qT, other=0.0) + + # dropout + if ENABLE_DROPOUT: + # NOTE: dropout is transposed because it is used to mask pT + philox_offs = ( + curr_philox_offset + + offs_m[None, :] * stride_dropout_m + + offs_n[:, None] * stride_dropout_n + ) + rand_vals = tl.rand(philox_seed, philox_offs) + dropout_mask = rand_vals > dropout_p + dropout_scale = 1.0 / (1 - dropout_p) + + # Load M + m = tl.load(M + offs_m * stride_deltam, mask=mask_m, other=0.0) + + # Compute qkT + if IS_FP8: + qkT = tl.dot(k, qT) * descale_q * descale_k + else: + qkT = tl.dot(k, qT) + + # Compute pT(use m and also apply sm_scale) + pT = tl.math.exp(qkT * sm_scale - m[None, :]) + + if MASK: + causal_mask = (offs_m[None, :] - delta_qk) >= offs_n[:, None] + mask = causal_mask & mask_nm + pT = tl.where(mask, pT, 0.0) + + # load DO + do = tl.load(do_ptrs, mask=mask_do, other=0.0) + + # dV + if ENABLE_DROPOUT: + pT_dropout = tl.where(dropout_mask, pT, 0.0) * dropout_scale + if IS_FP8: + scale_p_dropout, descale_p_dropout = compute_fp8_scaling_factors( + pT_dropout, FP8_MAX + ) + dv += ( + tl.dot((pT_dropout * scale_p_dropout).to(do.type.element_ty), do) + * descale_p_dropout + * descale_do + ) + else: + dv += tl.dot(pT_dropout.to(do.type.element_ty), do) + else: + if IS_FP8: + scale_pT, descale_pT = compute_fp8_scaling_factors(pT, FP8_MAX) + dv += ( + tl.dot((pT * scale_pT).to(do.type.element_ty), do) + * descale_pT + * descale_do + ) + else: + dv += tl.dot(pT.to(do.type.element_ty), do) + + # Load delta + Di = tl.load(D + offs_m * stride_deltam, mask=mask_m) + + # Compute dP and dS + if IS_FP8: + dpT = tl.dot(v, tl.trans(do)) * descale_v * descale_do + else: + dpT = tl.dot(v, tl.trans(do)) + + if ENABLE_DROPOUT: + dpT = tl.where(dropout_mask, dpT, 0.0) * dropout_scale + + delta_i = Di[None, :] + dsT = pT * (dpT - delta_i) + + # compute dk + if IS_FP8: + scale_dsT, descale_dsT = compute_fp8_scaling_factors(dsT, FP8_MAX) + dk += ( + tl.dot((dsT * scale_dsT).to(qT.type.element_ty), tl.trans(qT)) + * descale_dsT + * descale_q + ) + else: + dk += tl.dot(dsT.to(qT.type.element_ty), tl.trans(qT)) + + # increment pointers + curr_m += step_m + qT_ptrs += step_m * stride_q_m + do_ptrs += step_m * stride_do_m + + return dk, dv + + +@triton.jit +def _bwd_dkdvdq_inner( + dk, + dv, + Q, + k, + v, + DO, + DQ, + M, + D, + sm_scale, + stride_q_m, + stride_q_k, + stride_dq_m, + stride_dq_k, + stride_do_m, + stride_do_k, + stride_dropout_m, + stride_dropout_n, + stride_deltam, + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, + seqlen_q, + seqlen_k, + start_n, + start_m, + num_steps, + descale_q, + descale_k, + descale_v, + descale_do, + BLOCK_M: tl.constexpr, + BLOCK_N: tl.constexpr, + BLOCK_D_MODEL: tl.constexpr, + BLOCK_D_MODEL_POW2: tl.constexpr, + MASK: tl.constexpr, + ENABLE_DROPOUT: tl.constexpr, + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, + workgroup_id: tl.int32, +): + PADDED_HEAD: tl.constexpr = BLOCK_D_MODEL != BLOCK_D_MODEL_POW2 + delta_qk = seqlen_q - seqlen_k + offs_m = start_m + tl.arange(0, BLOCK_M) + offs_n = start_n + tl.arange(0, BLOCK_N) + offs_k = tl.arange(0, BLOCK_D_MODEL_POW2) + + # mask to make sure not OOB of seqlen_q + mask_n = offs_n < seqlen_k + + qT_ptrs_start = ( + Q + offs_m[None, :] * stride_q_m + offs_k[:, None] * stride_q_k + ) # [BLOCK_D_MODEL_POW2, BLOCK_M] + dq_ptrs_start = ( + DQ + offs_m[:, None] * stride_dq_m + offs_k[None, :] * stride_dq_k + ) # [BLOCK_M, BLOCK_D_MODEL_POW2] + + do_ptrs_start = DO + offs_m[:, None] * stride_do_m + offs_k[None, :] * stride_do_k + curr_m = start_m + step_m = BLOCK_M + curr_philox_offset = batch_philox_offset + curr_dropout_offset = dropout_offset + RCP_LN2: tl.constexpr = 1.4426950408889634 + + # Iterate over blocks(BLOCK_M size) of Q while calculating + # a fixed block(BLOCK_N) of dk and dv. Note, during backward + # pass P has to be recomputed. However, this kernel computes + # dV and dK, so we compute we need P^T and S^T. See backward pass + # equations + # + # From Flash Attention Paper: + # ForwardPass: S = QkT, P=softmax(S), O=PV + # + # BackwardPass equations + # dV = P^TdO + # dP = dOV^T + # dS = dsoftmax(dP) + # dQ = dSK + # dK = QdS^T + + # Compute a starting index and step based on workgroup_id + # Use a simple hash-like function to spread out the starting points + start_idx = ( + workgroup_id * 17 + ) % num_steps # 17 is an arbitrary prime to spread indices + # Ensure step is coprime with num_steps to visit all indices exactly once + step = 1 # 3 if num_steps > 1 or num_steps==3 else 1 # coprime with num_steps + + for iter in range(num_steps): + # Compute the permuted block index + blk_idx = (start_idx + iter * step) % num_steps + + curr_m = start_m + blk_idx * step_m + qT_ptrs = qT_ptrs_start + blk_idx * step_m * stride_q_m + dq_ptrs = dq_ptrs_start + blk_idx * step_m * stride_dq_m + do_ptrs = do_ptrs_start + blk_idx * step_m * stride_do_m + + offs_m = curr_m + tl.arange(0, BLOCK_M) + mask_m = offs_m < seqlen_q + mask_qT = mask_m[None, :] + mask_do = mask_m[:, None] + mask_nm = mask_n[:, None] & (offs_m[None, :] < seqlen_q) + + if PADDED_HEAD: + mask_qT &= offs_k[:, None] < BLOCK_D_MODEL + mask_do &= offs_k[None, :] < BLOCK_D_MODEL + + # load qT + qT = tl.load(qT_ptrs, mask=mask_qT, other=0.0) + + # dropout + if ENABLE_DROPOUT: + # NOTE: dropout is transposed because it is used to mask pT + philox_offs = ( + curr_philox_offset + + offs_m[None, :] * stride_dropout_m + + offs_n[:, None] * stride_dropout_n + ) + rand_vals = tl.rand(philox_seed, philox_offs) + dropout_mask = rand_vals > dropout_p + dropout_scale = 1.0 / (1 - dropout_p) + + # Load M + m = tl.load(M + offs_m * stride_deltam, mask=mask_m, other=0.0) + + # Compute qkT + if IS_FP8: + qkT = tl.dot(k, qT) * descale_q * descale_k + else: + qkT = tl.dot(k, qT) + + # Compute pT(use m and also apply sm_scale) + pT = tl.math.exp(qkT * sm_scale - m[None, :]) + + if MASK: + causal_mask = (offs_m[None, :] - delta_qk) >= (offs_n[:, None]) + mask = causal_mask & mask_nm + pT = tl.where(mask, pT, 0.0) + + # load DO + do = tl.load(do_ptrs, mask=mask_do, other=0.0) + + # dV + if ENABLE_DROPOUT: + pT_dropout = tl.where(dropout_mask, pT, 0.0) * dropout_scale + if IS_FP8: + scale_p_dropout, descale_p_dropout = compute_fp8_scaling_factors( + pT_dropout, FP8_MAX + ) + dv += ( + tl.dot((pT_dropout * scale_p_dropout).to(do.type.element_ty), do) + * descale_p_dropout + * descale_do + ) + else: + dv += tl.dot(pT_dropout.to(do.type.element_ty), do) + else: + if IS_FP8: + scale_pT, descale_pT = compute_fp8_scaling_factors(pT, FP8_MAX) + dv += ( + tl.dot((pT * scale_pT).to(do.type.element_ty), do) + * descale_pT + * descale_do + ) + else: + dv += tl.dot(pT.to(do.type.element_ty), do) + + # Load delta + Di = tl.load(D + offs_m * stride_deltam, mask=mask_m) + + # Compute dP and dS + if IS_FP8: + dpT = tl.dot(v, tl.trans(do)) * descale_v * descale_do + else: + dpT = tl.dot(v, tl.trans(do)) + + if ENABLE_DROPOUT: + dpT = tl.where(dropout_mask, dpT, 0.0) * dropout_scale + + delta_i = Di[None, :] + dsT = pT * (dpT - delta_i) + + # compute dk + if IS_FP8: + scale_dsT, descale_dsT = compute_fp8_scaling_factors(dsT, FP8_MAX) + dk += ( + tl.dot((dsT * scale_dsT).to(qT.type.element_ty), tl.trans(qT)) + * descale_dsT + * descale_q + ) + else: + dk += tl.dot(dsT.to(qT.type.element_ty), tl.trans(qT)) + + # We can compute the dq_partial here and do a atomic add to the correct memory location + # NOTE: Possible problems with the atomic add: contention, is inside a loop which has achieved bad perf before + # (BLOCK_M, BLOCK_N) x (BLOCK_N, D) + if IS_FP8: + dq_partial = ( + tl.dot((dsT * scale_dsT).to(k.dtype).T, k) * descale_dsT * descale_k + ) + else: + dq_partial = tl.dot(dsT.to(k.dtype).T, k) + tl.atomic_add( + dq_ptrs, + dq_partial * sm_scale, + mask=mask_m[:, None], + sem="relaxed", + ) + + return dk, dv + + +@triton.jit +def _bwd_kernel_dkdvdq_causal( + q_ptr, + k_ptr, + v_ptr, + sm_scale, + do_ptr, + dk_ptr, + dv_ptr, + dq_ptr, + m_ptr, + delta_ptr, + stride_q_b, + stride_q_h, + stride_q_m, + stride_q_k, + stride_k_b, + stride_k_h, + stride_k_n, + stride_k_k, + stride_v_b, + stride_v_h, + stride_v_n, + stride_v_k, + stride_dk_b, + stride_dk_h, + stride_dk_n, + stride_dk_k, + stride_dq_b, + stride_dq_h, + stride_dq_m, + stride_dq_k, + stride_delta_b, + stride_delta_h, + stride_delta_m, + stride_do_b, + stride_do_h, + stride_do_m, + stride_do_k, + stride_dropout_b, + stride_dropout_h, + stride_dropout_m, + stride_dropout_n, + stride_descale_q_z, + stride_descale_k_z, + stride_descale_v_z, + stride_descale_do_z, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_mask, + dropout_p, + philox_seed, + philox_offset_base, + descale_q_ptr, + descale_k_ptr, + descale_v_ptr, + descale_do_ptr, + NUM_Q_HEADS: tl.constexpr, + NUM_K_HEADS: tl.constexpr, + BATCH, + NUM_K_PIDS, + BLOCK_M: tl.constexpr, + BLOCK_N: tl.constexpr, + BLK_SLICE_FACTOR: tl.constexpr, + BLOCK_D_MODEL: tl.constexpr, + BLOCK_D_MODEL_POW2: tl.constexpr, + ENABLE_DROPOUT: tl.constexpr, + IS_VARLEN: tl.constexpr, + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, +): + wid = tl.program_id(0) # workgoup id: 0, ..., NUM_K_PIDS * BATCH * NUM_K_HEADS - 1 + + # workgroups get launched first along batch dim, then in head_k dim, and then in seq k block dim + batch_idx = wid % BATCH + head_k_idx = wid // BATCH % NUM_K_HEADS + seq_k_blk_idx = wid // (BATCH * NUM_K_HEADS) % NUM_K_PIDS + + # Determine q and k start along with seqlen_q and seqlen_k + q_start = 0 + k_start = 0 + seqlen_q = max_seqlen_q + seqlen_k = max_seqlen_k + if IS_VARLEN: + q_start = tl.load(cu_seqlens_q + batch_idx) + q_end = tl.load(cu_seqlens_q + batch_idx + 1) + k_start = tl.load(cu_seqlens_k + batch_idx) + k_end = tl.load(cu_seqlens_k + batch_idx + 1) + seqlen_q = q_end - q_start + seqlen_k = k_end - k_start + + dk = tl.zeros([BLOCK_N, BLOCK_D_MODEL_POW2], dtype=tl.float32) + dv = tl.zeros([BLOCK_N, BLOCK_D_MODEL_POW2], dtype=tl.float32) + + # Figure out causal starting block since we have seqlen_q >=< seqlen_k. + # Unlike forward pass where we tile on M dim and iterate on N dim, so that + # we can skip some M blocks, in backward pass, we tile on the N dim for kv + # and iterate over the M. In this way, we cannot skip N blocks, but only to + # determine the starting M blocks to skip some initial blocks masked by + # causal. + delta_qk = seqlen_q - seqlen_k + + # q > k: diretcly skip all the way until the start of causal block + start_delta_q_gt_k = delta_qk + + # q < k: some blocks will have no Masked block, other needs to re-calc + # starting position + # delta_qk is negative so flip it, only multiple of BLOCK_N can skip the + # masked op + num_blocks_skip = -delta_qk // BLOCK_N + delta_aligned = (num_blocks_skip + 1) * BLOCK_N + delta_qk + start_delta_q_lt_k = delta_aligned // BLOCK_M * BLOCK_M + if delta_qk >= 0: + start_delta = delta_qk + else: + start_delta = start_delta_q_lt_k + + start_n = seq_k_blk_idx * BLOCK_N + + offs_k = tl.arange(0, BLOCK_D_MODEL_POW2) + offs_n = start_n + tl.arange(0, BLOCK_N) + # Mask for loading K and V + mask_kv = offs_n[:, None] < seqlen_k + PADDED_HEAD: tl.constexpr = BLOCK_D_MODEL != BLOCK_D_MODEL_POW2 + if PADDED_HEAD: + mask_k = offs_k < BLOCK_D_MODEL + mask_kv &= mask_k[None, :] + + GROUP_SIZE = NUM_Q_HEADS // NUM_K_HEADS + adj_k = ( + batch_idx * stride_k_b + + head_k_idx * stride_k_h + + k_start * stride_k_n + + offs_n[:, None] * stride_k_n + + offs_k[None, :] * stride_k_k + ) + adj_v = ( + batch_idx * stride_v_b + + head_k_idx * stride_v_h + + k_start * stride_v_n + + offs_n[:, None] * stride_v_n + + offs_k[None, :] * stride_v_k + ) + # load K and V: they stay in SRAM throughout the inner loop. + k = tl.load(k_ptr + adj_k, mask=mask_kv, other=0.0) + v = tl.load(v_ptr + adj_v, mask=mask_kv, other=0.0) + + # If MQA / GQA, set the K and V head offsets appropriately. + for head_q_idx in range( + head_k_idx * GROUP_SIZE, head_k_idx * GROUP_SIZE + GROUP_SIZE + ): + if delta_qk >= 0: + start_m = start_n + start_delta + len_m = BLOCK_N + else: + start_m = max(start_n + delta_qk, 0) + start_m = (start_m // BLOCK_M) * BLOCK_M + # because we might shift the masked blocks up, we are deeper into + # the masked out region, so we would potentially increase the total + # steps with masked operation to get out of it + residue_m = max(start_n + delta_qk - start_m, 0) + len_m = BLOCK_N + residue_m + + # offset input and output tensor by batch and Q/K heads + adj_q = batch_idx * stride_q_b + head_q_idx * stride_q_h + q_start * stride_q_m + adj_dq = ( + batch_idx * stride_dq_b + head_q_idx * stride_dq_h + q_start * stride_dq_m + ) + + q_ptr_adj = q_ptr + adj_q + dq_ptr_adj = dq_ptr + adj_dq + + adj_do = ( + batch_idx * stride_do_b + head_q_idx * stride_do_h + q_start * stride_do_m + ) + do_ptr_adj = do_ptr + adj_do + adj_delta = ( + batch_idx * stride_delta_b + + head_q_idx * stride_delta_h + + q_start * stride_delta_m + ) + m_ptr_adj = m_ptr + adj_delta + delta_ptr_adj = delta_ptr + adj_delta + + # batch_philox_offset is the ACTUALLY dropout offset + # dropout_offset is for debug purpose and will be removed later + batch_philox_offset = 0 + dropout_offset = 0 + if ENABLE_DROPOUT: + batch_philox_offset = ( + philox_offset_base + + batch_idx * stride_dropout_b + + head_q_idx * stride_dropout_h + ) + dropout_offset = ( + dropout_mask + + batch_idx * stride_dropout_b + + head_q_idx * stride_dropout_h + ) + + MASK_BLOCK_M: tl.constexpr = BLOCK_M // BLK_SLICE_FACTOR + # bound the masked operation to q len so it does not have to wast cycles + len_m = min(len_m, seqlen_q) + num_steps = tl.cdiv(len_m, MASK_BLOCK_M) + + # when q < k, we may skip the initial masked op + # if seq_k_blk_idx < num_blocks_skip: + # num_steps = 0 + + if IS_FP8: + descale_q = tl.load( + descale_q_ptr + batch_idx * stride_descale_q_z + head_q_idx + ) + descale_k = tl.load( + descale_k_ptr + batch_idx * stride_descale_k_z + head_k_idx + ) + descale_v = tl.load( + descale_v_ptr + batch_idx * stride_descale_v_z + head_k_idx + ) + descale_do = tl.load( + descale_do_ptr + batch_idx * stride_descale_do_z + head_q_idx + ) + else: + descale_q, descale_k, descale_v, descale_do = 1.0, 1.0, 1.0, 1.0 + + # if unaligned start_m is negative, the current N-tile has no block on the + # diagonal of causal mask, so everything have no causal mask + dk, dv = _bwd_dkdvdq_inner( + dk, + dv, # output tensors + q_ptr_adj, + k, + v, + do_ptr_adj, + dq_ptr_adj, + m_ptr_adj, + delta_ptr_adj, + sm_scale, # input tensors + stride_q_m, + stride_q_k, # strides for q + stride_dq_m, + stride_dq_k, # strides for q + stride_do_m, + stride_do_k, # strides for o + stride_dropout_m, + stride_dropout_n, # strides for dropout + stride_delta_m, + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, # + seqlen_q, + seqlen_k, # max sequence length for q and k + start_n, + start_m, + num_steps, # iteration numbers + descale_q, + descale_k, + descale_v, + descale_do, # fp8 descale factors from user + MASK_BLOCK_M, + BLOCK_N, # block dim + BLOCK_D_MODEL, + BLOCK_D_MODEL_POW2, # head dim + MASK=True, # causal masking + ENABLE_DROPOUT=ENABLE_DROPOUT, # activate dropout + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + workgroup_id=seq_k_blk_idx, + ) + + start_m += num_steps * MASK_BLOCK_M + num_steps = tl.cdiv(seqlen_q - start_m, BLOCK_M) + end_m = start_m + num_steps * BLOCK_M + + dk, dv = _bwd_dkdvdq_inner( + dk, + dv, # output tensors + q_ptr_adj, + k, + v, + do_ptr_adj, + dq_ptr_adj, + m_ptr_adj, + delta_ptr_adj, + sm_scale, # input tensors + stride_q_m, + stride_q_k, # strides for q + stride_dq_m, + stride_dq_k, # strides for dq + stride_do_m, + stride_do_k, # strides for o + stride_dropout_m, + stride_dropout_n, # strides for dropout + stride_delta_m, + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, # + seqlen_q, + seqlen_k, # max sequence length for q and k + start_n, + start_m, + num_steps, # iteration numbers + descale_q, + descale_k, + descale_v, + descale_do, # fp8 descale factors from user + BLOCK_M, + BLOCK_N, # block dim + BLOCK_D_MODEL, + BLOCK_D_MODEL_POW2, # head dim + MASK=False, # causal masking + ENABLE_DROPOUT=ENABLE_DROPOUT, # activate dropout + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + workgroup_id=seq_k_blk_idx, + ) + + # Write back dV and dK. + offs_dkdv = ( + batch_idx * stride_dk_b + + head_k_idx * stride_dk_h + + k_start * stride_dk_n + + offs_n[:, None] * stride_dk_n + + offs_k[None, :] * stride_dk_k + ) + tl.store(dv_ptr + offs_dkdv, dv, mask=mask_kv) + dk *= sm_scale + tl.store(dk_ptr + offs_dkdv, dk, mask=mask_kv) + + +@triton.jit +def _bwd_kernel_dkdv_causal( + q_ptr, + k_ptr, + v_ptr, + sm_scale, + do_ptr, + dk_ptr, + dv_ptr, + m_ptr, + delta_ptr, + stride_q_b, + stride_q_h, + stride_q_m, + stride_q_k, + stride_k_b, + stride_k_h, + stride_k_n, + stride_k_k, + stride_v_b, + stride_v_h, + stride_v_n, + stride_v_k, + stride_dk_b, + stride_dk_h, + stride_dk_n, + stride_dk_k, + stride_delta_b, + stride_delta_h, + stride_delta_m, + stride_do_b, + stride_do_h, + stride_do_m, + stride_do_k, + stride_dropout_b, + stride_dropout_h, + stride_dropout_m, + stride_dropout_n, + stride_descale_q_z, + stride_descale_k_z, + stride_descale_v_z, + stride_descale_do_z, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_mask, + dropout_p, + philox_seed, + philox_offset_base, + descale_q_ptr, + descale_k_ptr, + descale_v_ptr, + descale_do_ptr, + NUM_Q_HEADS: tl.constexpr, + NUM_K_HEADS: tl.constexpr, + BLOCK_M: tl.constexpr, + BLOCK_N: tl.constexpr, + BLK_SLICE_FACTOR: tl.constexpr, + BLOCK_D_MODEL: tl.constexpr, + BLOCK_D_MODEL_POW2: tl.constexpr, + ENABLE_DROPOUT: tl.constexpr, + IS_VARLEN: tl.constexpr, + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, +): + # seq block, batch, head_k + seq_k_blk_idx = tl.program_id(0) + batch_idx = tl.program_id(1) + head_k_idx = tl.program_id(2) + + # Determine q and k start along with seqlen_q and seqlen_k + q_start = 0 + k_start = 0 + seqlen_q = max_seqlen_q + seqlen_k = max_seqlen_k + if IS_VARLEN: + q_start = tl.load(cu_seqlens_q + batch_idx) + q_end = tl.load(cu_seqlens_q + batch_idx + 1) + k_start = tl.load(cu_seqlens_k + batch_idx) + k_end = tl.load(cu_seqlens_k + batch_idx + 1) + seqlen_q = q_end - q_start + seqlen_k = k_end - k_start + + dk = tl.zeros([BLOCK_N, BLOCK_D_MODEL_POW2], dtype=tl.float32) + dv = tl.zeros([BLOCK_N, BLOCK_D_MODEL_POW2], dtype=tl.float32) + + # Figure out causal starting block since we have seqlen_q >=< seqlen_k. + # Unlike forward pass where we tile on M dim and iterate on N dim, so that + # we can skip some M blocks, in backward pass, we tile on the N dim for kv + # and iterate over the M. In this way, we cannot skip N blocks, but only to + # determine the starting M blocks to skip some initial blocks masked by + # causal. + delta_qk = seqlen_q - seqlen_k + + # q > k: diretcly skip all the way until the start of causal block + start_delta_q_gt_k = delta_qk + + # q < k: some blocks will have no Masked block, other needs to re-calc + # starting position + # delta_qk is negative so flip it, only multiple of BLOCK_N can skip the + # masked op + num_blocks_skip = -delta_qk // BLOCK_N + delta_aligned = (num_blocks_skip + 1) * BLOCK_N + delta_qk + start_delta_q_lt_k = delta_aligned // BLOCK_M * BLOCK_M + if delta_qk >= 0: + start_delta = delta_qk + else: + start_delta = start_delta_q_lt_k + + start_n = seq_k_blk_idx * BLOCK_N + + offs_k = tl.arange(0, BLOCK_D_MODEL_POW2) + offs_n = start_n + tl.arange(0, BLOCK_N) + # Mask for loading K and V + mask_kv = offs_n[:, None] < seqlen_k + PADDED_HEAD: tl.constexpr = BLOCK_D_MODEL != BLOCK_D_MODEL_POW2 + if PADDED_HEAD: + mask_k = offs_k < BLOCK_D_MODEL + mask_kv &= mask_k[None, :] + + GROUP_SIZE = NUM_Q_HEADS // NUM_K_HEADS + adj_k = ( + batch_idx * stride_k_b + + head_k_idx * stride_k_h + + k_start * stride_k_n + + offs_n[:, None] * stride_k_n + + offs_k[None, :] * stride_k_k + ) + adj_v = ( + batch_idx * stride_v_b + + head_k_idx * stride_v_h + + k_start * stride_v_n + + offs_n[:, None] * stride_v_n + + offs_k[None, :] * stride_v_k + ) + # load K and V: they stay in SRAM throughout the inner loop. + k = tl.load(k_ptr + adj_k, mask=mask_kv, other=0.0) + v = tl.load(v_ptr + adj_v, mask=mask_kv, other=0.0) + + # If MQA / GQA, set the K and V head offsets appropriately. + for head_q_idx in range( + head_k_idx * GROUP_SIZE, head_k_idx * GROUP_SIZE + GROUP_SIZE + ): + if delta_qk >= 0: + start_m = start_n + start_delta + len_m = BLOCK_N + else: + start_m = max(start_n + delta_qk, 0) + start_m = start_m // BLOCK_M * BLOCK_M + # because we might shift the masked blocks up, we are deeper into + # the masked out region, so we would potentially increase the total + # steps with masked operation to get out of it + residue_m = max(start_n + delta_qk - start_m, 0) + len_m = BLOCK_N + residue_m + + # offset input and output tensor by batch and Q/K heads + adj_q = batch_idx * stride_q_b + head_q_idx * stride_q_h + q_start * stride_q_m + q_ptr_adj = q_ptr + adj_q + adj_do = ( + batch_idx * stride_do_b + head_q_idx * stride_do_h + q_start * stride_do_m + ) + do_ptr_adj = do_ptr + adj_do + adj_delta = ( + batch_idx * stride_delta_b + + head_q_idx * stride_delta_h + + q_start * stride_delta_m + ) + m_ptr_adj = m_ptr + adj_delta + delta_ptr_adj = delta_ptr + adj_delta + + # batch_philox_offset is the ACTUALLY dropout offset + # dropout_offset is for debug purpose and will be removed later + batch_philox_offset = 0 + dropout_offset = 0 + if ENABLE_DROPOUT: + batch_philox_offset = ( + philox_offset_base + + batch_idx * stride_dropout_b + + head_q_idx * stride_dropout_h + ) + dropout_offset = ( + dropout_mask + + batch_idx * stride_dropout_b + + head_q_idx * stride_dropout_h + ) + + MASK_BLOCK_M: tl.constexpr = BLOCK_M // BLK_SLICE_FACTOR + # bound the masked operation to q len so it does not have to wast cycles + len_m = min(len_m, seqlen_q) + num_steps = tl.cdiv(len_m, MASK_BLOCK_M) + # when q < k, we may skip the initial masked op + if seq_k_blk_idx < num_blocks_skip: + num_steps = 0 + + if IS_FP8: + descale_q = tl.load( + descale_q_ptr + batch_idx * stride_descale_q_z + head_q_idx + ) + descale_k = tl.load( + descale_k_ptr + batch_idx * stride_descale_k_z + head_k_idx + ) + descale_v = tl.load( + descale_v_ptr + batch_idx * stride_descale_v_z + head_k_idx + ) + descale_do = tl.load( + descale_do_ptr + batch_idx * stride_descale_do_z + head_q_idx + ) + else: + descale_q, descale_k, descale_v, descale_do = 1.0, 1.0, 1.0, 1.0 + + # if start_m is negative, the current N-tile has no block on the + # diagonal of causal mask, so everything have no causal mask + dk, dv = _bwd_dkdv_inner( + dk, + dv, # output tensors + q_ptr_adj, + k, + v, + do_ptr_adj, + m_ptr_adj, + delta_ptr_adj, + sm_scale, # input tensors + stride_q_m, + stride_q_k, # strides for q + stride_do_m, + stride_do_k, # strides for o + stride_dropout_m, + stride_dropout_n, # strides for dropout + stride_delta_m, + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, # + seqlen_q, + seqlen_k, # max sequence length for q and k + start_n, + start_m, + num_steps, # iteration numbers + descale_q, + descale_k, + descale_v, + descale_do, # fp8 descale factors from user + MASK_BLOCK_M, + BLOCK_N, # block dim + BLOCK_D_MODEL, + BLOCK_D_MODEL_POW2, # head dim + MASK=True, # causal masking + ENABLE_DROPOUT=ENABLE_DROPOUT, # activate dropout + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + ) + start_m += num_steps * MASK_BLOCK_M + num_steps = tl.cdiv(seqlen_q - start_m, BLOCK_M) + end_m = start_m + num_steps * BLOCK_M + + dk, dv = _bwd_dkdv_inner( + dk, + dv, # output tensors + q_ptr_adj, + k, + v, + do_ptr_adj, + m_ptr_adj, + delta_ptr_adj, + sm_scale, # input tensors + stride_q_m, + stride_q_k, # strides for q + stride_do_m, + stride_do_k, # strides for o + stride_dropout_m, + stride_dropout_n, # strides for dropout + stride_delta_m, + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, # + seqlen_q, + seqlen_k, # max sequence length for q and k + start_n, + start_m, + num_steps, # iteration numbers + descale_q, + descale_k, + descale_v, + descale_do, # fp8 descale factors from user + BLOCK_M, + BLOCK_N, # block dim + BLOCK_D_MODEL, + BLOCK_D_MODEL_POW2, # head dim + MASK=False, # causal masking + ENABLE_DROPOUT=ENABLE_DROPOUT, # activate dropout + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + ) + + # Write back dV and dK. + offs_dkdv = ( + batch_idx * stride_dk_b + + head_k_idx * stride_dk_h + + k_start * stride_dk_n + + offs_n[:, None] * stride_dk_n + + offs_k[None, :] * stride_dk_k + ) + tl.store(dv_ptr + offs_dkdv, dv, mask=mask_kv) + dk *= sm_scale + tl.store(dk_ptr + offs_dkdv, dk, mask=mask_kv) + + +@triton.jit +def _bwd_kernel_dq_causal( + q_ptr, + k_ptr, + v_ptr, + sm_scale, + do_ptr, + dq_ptr, + m_ptr, + delta_ptr, + stride_q_b, + stride_q_h, + stride_q_m, + stride_q_k, + stride_k_b, + stride_k_h, + stride_k_n, + stride_k_k, + stride_v_b, + stride_v_h, + stride_v_n, + stride_v_k, + stride_dq_b, + stride_dq_h, + stride_dq_m, + stride_dq_k, + stride_delta_b, + stride_delta_h, + stride_delta_m, + stride_do_b, + stride_do_h, + stride_do_m, + stride_do_k, + stride_dropout_b, + stride_dropout_h, + stride_dropout_m, + stride_dropout_n, + stride_descale_q_z, + stride_descale_k_z, + stride_descale_v_z, + stride_descale_do_z, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_mask, + dropout_p, + philox_seed, + philox_offset_base, + descale_q_ptr, + descale_k_ptr, + descale_v_ptr, + descale_do_ptr, + NUM_Q_HEADS: tl.constexpr, + NUM_K_HEADS: tl.constexpr, + BLOCK_M: tl.constexpr, + BLOCK_N: tl.constexpr, + BLK_SLICE_FACTOR: tl.constexpr, + BLOCK_D_MODEL: tl.constexpr, + BLOCK_D_MODEL_POW2: tl.constexpr, + ENABLE_DROPOUT: tl.constexpr, + IS_VARLEN: tl.constexpr, + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, +): + seq_q_blk_idx = tl.program_id(0) + batch_idx = tl.program_id(1) + head_k_idx = tl.program_id(2) + + q_start = 0 + k_start = 0 + seqlen_q = max_seqlen_q + seqlen_k = max_seqlen_k + if IS_VARLEN: + q_start = tl.load(cu_seqlens_q + batch_idx) + q_end = tl.load(cu_seqlens_q + batch_idx + 1) + k_start = tl.load(cu_seqlens_k + batch_idx) + k_end = tl.load(cu_seqlens_k + batch_idx + 1) + seqlen_q = q_end - q_start + seqlen_k = k_end - k_start + + # Figure out causal starting block since we have seqlen_q <=> seqlen_k. + # Unlike forward pass where we tile on M dim and iterate on N dim, so that + # we can skip some M blocks, in backward pass, we tile on the N dim for kv + # and iterate over the M. In this way, we cannot skip N blocks, but only to + # determine the starting M blocks to skip some initial blocks masked by + # causal. + # DQ tiles on M dim and iterate on N dim, so we there could be some tiles we + # can simply skip and we need to adjust starting position. + start_m = seq_q_blk_idx * BLOCK_M + # seqlen_q > seqlen_k, no need to process these tile for dq + delta_qk = seqlen_q - seqlen_k + if start_m + BLOCK_M < delta_qk: + return + + offs_k = tl.arange(0, BLOCK_D_MODEL_POW2) + offs_m = start_m + tl.arange(0, BLOCK_M) + # Mask for loading K and V + mask_q = offs_m[:, None] < seqlen_q + PADDED_HEAD: tl.constexpr = BLOCK_D_MODEL != BLOCK_D_MODEL_POW2 + if PADDED_HEAD: + mask_k = offs_k < BLOCK_D_MODEL + mask_q &= mask_k[None, :] + offs_q = offs_m[:, None] * stride_q_m + offs_k[None, :] * stride_q_k + offs_do = offs_m[:, None] * stride_do_m + offs_k[None, :] * stride_do_k + adj_k = batch_idx * stride_k_b + head_k_idx * stride_k_h + k_start * stride_k_n + adj_v = batch_idx * stride_v_b + head_k_idx * stride_v_h + k_start * stride_v_n + k_ptr_adj = k_ptr + v_ptr_adj = v_ptr + k_ptr_adj += adj_k + v_ptr_adj += adj_v + + # If MQA / GQA, set the K and V head offsets appropriately. + GROUP_SIZE = NUM_Q_HEADS // NUM_K_HEADS + for head_q_idx in range( + head_k_idx * GROUP_SIZE, head_k_idx * GROUP_SIZE + GROUP_SIZE + ): + # seqlen_q < seqlen_k: delta_qk more kv tokens are added at the front + # for every M-tile + end_n = start_m + BLOCK_M - delta_qk + # clamp end_n at [0, seqlen_k] + end_n = max(min(end_n, seqlen_k), 0) + + # offset input and output tensor by batch and Q/K heads + adj_q = batch_idx * stride_q_b + head_q_idx * stride_q_h + q_start * stride_q_m + adj_do = ( + batch_idx * stride_do_b + head_q_idx * stride_do_h + q_start * stride_do_m + ) + adj_delta = ( + batch_idx * stride_delta_b + + head_q_idx * stride_delta_h + + q_start * stride_delta_m + ) + delta_ptr_adj = delta_ptr + adj_delta + + # batch_philox_offset is the ACTUALLY dropout offset + # dropout_offset is for debug purpose and will be removed later + batch_philox_offset = 0 + dropout_offset = 0 + if ENABLE_DROPOUT: + batch_philox_offset = ( + philox_offset_base + + batch_idx * stride_dropout_b + + head_q_idx * stride_dropout_h + ) + dropout_offset = ( + dropout_mask + + batch_idx * stride_dropout_b + + head_q_idx * stride_dropout_h + ) + + q = tl.load(q_ptr + adj_q + offs_q, mask=mask_q, other=0.0) + do = tl.load(do_ptr + adj_do + offs_do, mask=mask_q, other=0.0) + m = tl.load(m_ptr + adj_delta + offs_m * stride_delta_m, mask=offs_m < seqlen_q) + m = m[:, None] + + MASK_BLOCK_N: tl.constexpr = BLOCK_N // BLK_SLICE_FACTOR + # start can only be 0 at minimum + start_n = max(end_n - BLOCK_M, 0) + num_steps = tl.cdiv(end_n - start_n, MASK_BLOCK_N) + + if IS_FP8: + descale_q = tl.load( + descale_q_ptr + batch_idx * stride_descale_q_z + head_q_idx + ) + descale_k = tl.load( + descale_k_ptr + batch_idx * stride_descale_k_z + head_k_idx + ) + descale_v = tl.load( + descale_v_ptr + batch_idx * stride_descale_v_z + head_k_idx + ) + descale_do = tl.load( + descale_do_ptr + batch_idx * stride_descale_do_z + head_q_idx + ) + else: + descale_q, descale_k, descale_v, descale_do = 1.0, 1.0, 1.0, 1.0 + + dq = tl.zeros([BLOCK_M, BLOCK_D_MODEL_POW2], dtype=tl.float32) + # Compute dQ for masked (diagonal) blocks. + # NOTE: This code scans each row of QK^T backward (from right to left, + # but inside each call to _bwd_dq_inner, from left to right), but that's + # not due to anything important. I just wanted to reuse the loop + # structure for dK & dV above as much as possible. + dq = _bwd_dq_inner( + dq, + q, + k_ptr_adj, + v_ptr_adj, + do, + m, + delta_ptr_adj, + sm_scale, + stride_q_m, + stride_q_k, + stride_k_n, + stride_k_k, + stride_v_n, + stride_v_k, + stride_dropout_m, + stride_dropout_n, + stride_delta_m, + seqlen_q, + seqlen_k, + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, + start_m, + start_n, + end_n, + num_steps, + descale_q, + descale_k, + descale_v, + descale_do, + BLOCK_M, + MASK_BLOCK_N, + BLOCK_D_MODEL, + BLOCK_D_MODEL_POW2, + MASK=True, + ENABLE_DROPOUT=ENABLE_DROPOUT, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + ) + end_n -= num_steps * MASK_BLOCK_N + num_steps = tl.cdiv(end_n, BLOCK_N) + start_n = max(end_n - num_steps * BLOCK_N, 0) + dq = _bwd_dq_inner( + dq, + q, + k_ptr_adj, + v_ptr_adj, + do, + m, + delta_ptr_adj, + sm_scale, + stride_q_m, + stride_q_k, + stride_k_n, + stride_k_k, + stride_v_n, + stride_v_k, + stride_dropout_m, + stride_dropout_n, + stride_delta_m, + seqlen_q, + seqlen_k, + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, + start_m, + start_n, + end_n, + num_steps, + descale_q, + descale_k, + descale_v, + descale_do, + BLOCK_M, + BLOCK_N, + BLOCK_D_MODEL, + BLOCK_D_MODEL_POW2, + MASK=False, + ENABLE_DROPOUT=ENABLE_DROPOUT, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + ) + # Write back dQ. + offs_dq = ( + batch_idx * stride_dq_b + + head_q_idx * stride_dq_h + + q_start * stride_dq_m + + offs_m[:, None] * stride_dq_m + + offs_k[None, :] * stride_dq_k + ) + dq *= sm_scale + tl.store(dq_ptr + offs_dq, dq, mask=mask_q) + + +@triton.jit +def _bwd_kernel_dkdvdq_noncausal( + Q, + K, + V, + sm_scale, + DO, + DK, + DV, + DQ, + M, + Delta, + stride_qb, + stride_qh, + stride_qm, + stride_qk, + stride_kb, + stride_kh, + stride_kn, + stride_kk, + stride_vb, + stride_vh, + stride_vn, + stride_vk, + stride_dkb, + stride_dkh, + stride_dkn, + stride_dkk, + stride_dqb, + stride_dqh, + stride_dqm, + stride_dqk, + stride_deltab, + stride_deltah, + stride_deltam, + stride_dob, + stride_doh, + stride_dom, + stride_dok, + stride_dropoutb, + stride_dropouth, + stride_dropoutm, + stride_dropoutn, + stride_descale_q_z, + stride_descale_k_z, + stride_descale_v_z, + stride_descale_do_z, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_mask, + dropout_p, + philox_seed, + philox_offset, + descale_q_ptr, + descale_k_ptr, + descale_v_ptr, + descale_do_ptr, + NUM_Q_HEADS: tl.constexpr, + NUM_K_HEADS: tl.constexpr, + BATCH, + NUM_K_PIDS, + BLOCK_M: tl.constexpr, + BLOCK_N: tl.constexpr, + BLK_SLICE_FACTOR: tl.constexpr, + BLOCK_D_MODEL: tl.constexpr, + BLOCK_D_MODEL_POW2: tl.constexpr, + ENABLE_DROPOUT: tl.constexpr, + IS_VARLEN: tl.constexpr, + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, +): + # workgroup id + wid = tl.program_id(0) # 0, ..., NUM_K_PIDS * BATCH * NUM_K_HEADS - 1 + + # Workgroups get launched first along batch dim, then in head_k dim, and then in seq k block dim + # This is in order to avoid contention for the tl.atomic_add (inside _bwd_dkdvdq_inner) that happens between workgroups that share the same batch and head_k. + bid = wid % BATCH + hkid = wid // BATCH % NUM_K_HEADS + pid = wid // (BATCH * NUM_K_HEADS) % NUM_K_PIDS + + q_start = 0 + k_start = 0 + seqlen_q = max_seqlen_q + seqlen_k = max_seqlen_k + + if IS_VARLEN: + q_start = tl.load(cu_seqlens_q + bid) + q_end = tl.load(cu_seqlens_q + bid + 1) + k_start = tl.load(cu_seqlens_k + bid) + k_end = tl.load(cu_seqlens_k + bid + 1) + seqlen_q = q_end - q_start + seqlen_k = k_end - k_start + + dk = tl.zeros([BLOCK_N, BLOCK_D_MODEL_POW2], dtype=tl.float32) + dv = tl.zeros([BLOCK_N, BLOCK_D_MODEL_POW2], dtype=tl.float32) + + start_n = pid * BLOCK_N + + offs_k = tl.arange(0, BLOCK_D_MODEL_POW2) + offs_n = start_n + tl.arange(0, BLOCK_N) + mask_kv = offs_n[:, None] < seqlen_k + PADDED_HEAD: tl.constexpr = BLOCK_D_MODEL != BLOCK_D_MODEL_POW2 + if PADDED_HEAD: + mask_kv &= offs_k < BLOCK_D_MODEL + + GROUP_SIZE = NUM_Q_HEADS // NUM_K_HEADS + adj_k = ( + bid * stride_kb + + hkid * stride_kh + + k_start * stride_kn + + offs_n[:, None] * stride_kn + + offs_k[None, :] * stride_kk + ) + adj_v = ( + bid * stride_vb + + hkid * stride_vh + + k_start * stride_vn + + offs_n[:, None] * stride_vn + + offs_k[None, :] * stride_vk + ) + + k = tl.load(K + adj_k, mask=mask_kv, other=0.0) + v = tl.load(V + adj_v, mask=mask_kv, other=0.0) + + for hqid in range(hkid * GROUP_SIZE, hkid * GROUP_SIZE + GROUP_SIZE): + adj_q = bid * stride_qb + hqid * stride_qh + q_start * stride_qm + adj_dq = bid * stride_dqb + hqid * stride_dqh + q_start * stride_dqm + + Q_ptr = Q + adj_q + DQ_ptr = DQ + adj_dq + + adj_do = bid * stride_dob + hqid * stride_doh + q_start * stride_dom + DO_ptr = DO + adj_do + adj_delta = bid * stride_deltab + hqid * stride_deltah + q_start * stride_deltam + M_ptr = M + adj_delta + Delta_ptr = Delta + adj_delta + + # dropout + batch_philox_offset = 0 + dropout_offset = 0 + if ENABLE_DROPOUT: + batch_philox_offset = ( + philox_offset + bid * stride_dropoutb + hqid * stride_dropouth + ) + dropout_offset = ( + dropout_mask + bid * stride_dropoutb + hqid * stride_dropouth + ) + + if IS_FP8: + descale_q = tl.load(descale_q_ptr + bid * stride_descale_q_z + hqid) + descale_k = tl.load(descale_k_ptr + bid * stride_descale_k_z + hkid) + descale_v = tl.load(descale_v_ptr + bid * stride_descale_v_z + hkid) + descale_do = tl.load(descale_do_ptr + bid * stride_descale_do_z + hqid) + else: + descale_q, descale_k, descale_v, descale_do = 1.0, 1.0, 1.0, 1.0 + + start_m = 0 + num_steps = tl.cdiv(seqlen_q, BLOCK_M) + + dk, dv = _bwd_dkdvdq_inner( + dk, + dv, + Q_ptr, + k, + v, + DO_ptr, + DQ_ptr, + M_ptr, + Delta_ptr, + sm_scale, + stride_qm, + stride_qk, + stride_dqm, + stride_dqk, + stride_dom, + stride_dok, + stride_dropoutm, + stride_dropoutn, + stride_deltam, + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, + seqlen_q, + seqlen_k, + start_n, + start_m, + num_steps, + descale_q, + descale_k, + descale_v, + descale_do, + BLOCK_M, + BLOCK_N, + BLOCK_D_MODEL, + BLOCK_D_MODEL_POW2, + MASK=False, + ENABLE_DROPOUT=ENABLE_DROPOUT, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + workgroup_id=pid, + ) + + adj_dkdv = ( + bid * stride_dkb + + hkid * stride_dkh + + k_start * stride_dkn + + offs_n[:, None] * stride_dkn + + offs_k[None, :] * stride_dkk + ) + tl.store(DV + adj_dkdv, dv, mask=mask_kv) + dk *= sm_scale + tl.store(DK + adj_dkdv, dk, mask=mask_kv) + + +@triton.jit +def _bwd_kernel_dkdv_noncausal( + Q, + K, + V, + sm_scale, + DO, + DK, + DV, + M, + Delta, + stride_qb, + stride_qh, + stride_qm, + stride_qk, + stride_kb, + stride_kh, + stride_kn, + stride_kk, + stride_vb, + stride_vh, + stride_vn, + stride_vk, + stride_dkb, + stride_dkh, + stride_dkn, + stride_dkk, + stride_deltab, + stride_deltah, + stride_deltam, + stride_dob, + stride_doh, + stride_dom, + stride_dok, + stride_dropoutb, + stride_dropouth, + stride_dropoutm, + stride_dropoutn, + stride_descale_q_z, + stride_descale_k_z, + stride_descale_v_z, + stride_descale_do_z, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_mask, + dropout_p, + philox_seed, + philox_offset, + descale_q_ptr, + descale_k_ptr, + descale_v_ptr, + descale_do_ptr, + NUM_Q_HEADS: tl.constexpr, + NUM_K_HEADS: tl.constexpr, + BLOCK_M: tl.constexpr, + BLOCK_N: tl.constexpr, + BLK_SLICE_FACTOR: tl.constexpr, + BLOCK_D_MODEL: tl.constexpr, + BLOCK_D_MODEL_POW2: tl.constexpr, + ENABLE_DROPOUT: tl.constexpr, + IS_VARLEN: tl.constexpr, + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, +): + pid = tl.program_id(0) + bid = tl.program_id(1) + hkid = tl.program_id(2) + + q_start = 0 + k_start = 0 + seqlen_q = max_seqlen_q + seqlen_k = max_seqlen_k + + if IS_VARLEN: + q_start = tl.load(cu_seqlens_q + bid) + q_end = tl.load(cu_seqlens_q + bid + 1) + k_start = tl.load(cu_seqlens_k + bid) + k_end = tl.load(cu_seqlens_k + bid + 1) + seqlen_q = q_end - q_start + seqlen_k = k_end - k_start + + dk = tl.zeros([BLOCK_N, BLOCK_D_MODEL_POW2], dtype=tl.float32) + dv = tl.zeros([BLOCK_N, BLOCK_D_MODEL_POW2], dtype=tl.float32) + + start_n = pid * BLOCK_N + + offs_k = tl.arange(0, BLOCK_D_MODEL_POW2) + offs_n = start_n + tl.arange(0, BLOCK_N) + mask_kv = offs_n[:, None] < seqlen_k + PADDED_HEAD: tl.constexpr = BLOCK_D_MODEL != BLOCK_D_MODEL_POW2 + if PADDED_HEAD: + mask_kv &= offs_k < BLOCK_D_MODEL + + GROUP_SIZE = NUM_Q_HEADS // NUM_K_HEADS + adj_k = ( + bid * stride_kb + + hkid * stride_kh + + k_start * stride_kn + + offs_n[:, None] * stride_kn + + offs_k[None, :] * stride_kk + ) + adj_v = ( + bid * stride_vb + + hkid * stride_vh + + k_start * stride_vn + + offs_n[:, None] * stride_vn + + offs_k[None, :] * stride_vk + ) + + k = tl.load(K + adj_k, mask=mask_kv, other=0.0) + v = tl.load(V + adj_v, mask=mask_kv, other=0.0) + + for hqid in range(hkid * GROUP_SIZE, hkid * GROUP_SIZE + GROUP_SIZE): + adj_q = bid * stride_qb + hqid * stride_qh + q_start * stride_qm + Q_ptr = Q + adj_q + adj_do = bid * stride_dob + hqid * stride_doh + q_start * stride_dom + DO_ptr = DO + adj_do + adj_delta = bid * stride_deltab + hqid * stride_deltah + q_start * stride_deltam + M_ptr = M + adj_delta + Delta_ptr = Delta + adj_delta + + # dropout + batch_philox_offset = 0 + dropout_offset = 0 + if ENABLE_DROPOUT: + batch_philox_offset = ( + philox_offset + bid * stride_dropoutb + hqid * stride_dropouth + ) + dropout_offset = ( + dropout_mask + bid * stride_dropoutb + hqid * stride_dropouth + ) + + if IS_FP8: + descale_q = tl.load(descale_q_ptr + bid * stride_descale_q_z + hqid) + descale_k = tl.load(descale_k_ptr + bid * stride_descale_k_z + hkid) + descale_v = tl.load(descale_v_ptr + bid * stride_descale_v_z + hkid) + descale_do = tl.load(descale_do_ptr + bid * stride_descale_do_z + hqid) + else: + descale_q, descale_k, descale_v, descale_do = 1.0, 1.0, 1.0, 1.0 + + start_m = 0 + num_steps = tl.cdiv(seqlen_q, BLOCK_M) + dk, dv = _bwd_dkdv_inner( + dk, + dv, + Q_ptr, + k, + v, + DO_ptr, + M_ptr, + Delta_ptr, + sm_scale, + stride_qm, + stride_qk, + stride_dom, + stride_dok, + stride_dropoutm, + stride_dropoutn, + stride_deltam, + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, + seqlen_q, + seqlen_k, + start_n, + start_m, + num_steps, + descale_q, + descale_k, + descale_v, + descale_do, + BLOCK_M, + BLOCK_N, + BLOCK_D_MODEL, + BLOCK_D_MODEL_POW2, + MASK=False, + ENABLE_DROPOUT=ENABLE_DROPOUT, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + ) + + adj_dkdv = ( + bid * stride_dkb + + hkid * stride_dkh + + k_start * stride_dkn + + offs_n[:, None] * stride_dkn + + offs_k[None, :] * stride_dkk + ) + tl.store(DV + adj_dkdv, dv, mask=mask_kv) + dk *= sm_scale + tl.store(DK + adj_dkdv, dk, mask=mask_kv) + + +@triton.jit +def _bwd_kernel_dq_noncausal( + Q, + K, + V, + sm_scale, + DO, + DQ, + M, + delta, + stride_qb, + stride_qh, + stride_qm, + stride_qk, + stride_kb, + stride_kh, + stride_kn, + stride_kk, + stride_vb, + stride_vh, + stride_vn, + stride_vk, + stride_dqb, + stride_dqh, + stride_dqm, + stride_dqk, + stride_deltab, + stride_deltah, + stride_deltam, + stride_dob, + stride_doh, + stride_dom, + stride_dok, + stride_dropoutb, + stride_dropouth, + stride_dropoutm, + stride_dropoutn, + stride_descale_q_z, + stride_descale_k_z, + stride_descale_v_z, + stride_descale_do_z, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_mask, + dropout_p, + philox_seed, + philox_offset_base, + descale_q_ptr, + descale_k_ptr, + descale_v_ptr, + descale_do_ptr, + NUM_Q_HEADS: tl.constexpr, + NUM_K_HEADS: tl.constexpr, + BLOCK_M: tl.constexpr, + BLOCK_N: tl.constexpr, + BLK_SLICE_FACTOR: tl.constexpr, + BLOCK_D_MODEL: tl.constexpr, + BLOCK_D_MODEL_POW2: tl.constexpr, + ENABLE_DROPOUT: tl.constexpr, + IS_VARLEN: tl.constexpr, + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, +): + pid = tl.program_id(0) # seqlen + bid = tl.program_id(1) # batch + hkid = tl.program_id(2) # head_k + + q_start = 0 + k_start = 0 + seqlen_q = max_seqlen_q + seqlen_k = max_seqlen_k + + if IS_VARLEN: + # Compute actual sequence lengths + q_start = tl.load(cu_seqlens_q + bid) + q_end = tl.load(cu_seqlens_q + bid + 1) + k_start = tl.load(cu_seqlens_k + bid) + k_end = tl.load(cu_seqlens_k + bid + 1) + seqlen_q = q_end - q_start + seqlen_k = k_end - k_start + + start_m = pid * BLOCK_M + + offs_k = tl.arange(0, BLOCK_D_MODEL_POW2) + offs_m = start_m + tl.arange(0, BLOCK_M) + + # mask for loading K and V + mask_q = offs_m[:, None] < seqlen_q + PADDED_HEAD: tl.constexpr = BLOCK_D_MODEL != BLOCK_D_MODEL_POW2 + if PADDED_HEAD: + mask_k = offs_k < BLOCK_D_MODEL + mask_q &= mask_k[None, :] + offs_q = offs_m[:, None] * stride_qm + offs_k[None, :] * stride_qk + offs_do = offs_m[:, None] * stride_dom + offs_k[None, :] * stride_dok + adj_k = bid * stride_kb + hkid * stride_kh + k_start * stride_kn + adj_v = bid * stride_vb + hkid * stride_vh + k_start * stride_vn + K += adj_k + V += adj_v + + GROUP_SIZE = NUM_Q_HEADS // NUM_K_HEADS + for hqid in range(hkid * GROUP_SIZE, hkid * GROUP_SIZE + GROUP_SIZE): + adj_q = bid * stride_qb + hqid * stride_qh + q_start * stride_qm + adj_do = bid * stride_dob + hqid * stride_doh + q_start * stride_dom + adj_delta = bid * stride_deltab + hqid * stride_deltah + q_start * stride_deltam + delta_ptr = delta + adj_delta + + batch_philox_offset = 0 + dropout_offset = 0 + if ENABLE_DROPOUT: + batch_philox_offset = ( + philox_offset_base + bid * stride_dropoutb + hqid * stride_dropouth + ) + dropout_offset = ( + dropout_mask + bid * stride_dropoutb + hqid * stride_dropouth + ) + + q = tl.load(Q + adj_q + offs_q, mask=mask_q, other=0.0) + do = tl.load(DO + adj_do + offs_do, mask=mask_q, other=0.0) + m = tl.load(M + adj_delta + offs_m * stride_deltam, mask=offs_m < seqlen_q) + m = m[:, None] + + # FP8 + if IS_FP8: + descale_q = tl.load(descale_q_ptr + bid * stride_descale_q_z + hqid) + descale_k = tl.load(descale_k_ptr + bid * stride_descale_k_z + hkid) + descale_v = tl.load(descale_v_ptr + bid * stride_descale_v_z + hkid) + descale_do = tl.load(descale_do_ptr + bid * stride_descale_do_z + hqid) + else: + descale_q, descale_k, descale_v, descale_do = 1.0, 1.0, 1.0, 1.0 + + start_n = 0 + end_n = seqlen_k + num_steps = tl.cdiv(seqlen_k, BLOCK_N) + dq = tl.zeros([BLOCK_M, BLOCK_D_MODEL_POW2], dtype=tl.float32) + dq = _bwd_dq_inner( + dq, + q, + K, + V, + do, + m, + delta_ptr, + sm_scale, + stride_qm, + stride_qk, + stride_kn, + stride_kk, + stride_vn, + stride_vk, + stride_dropoutm, + stride_dropoutn, + stride_deltam, + seqlen_q, + seqlen_k, + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, + start_m, + start_n, + end_n, + num_steps, + descale_q, + descale_k, + descale_v, + descale_do, + BLOCK_M, + BLOCK_N, + BLOCK_D_MODEL, + BLOCK_D_MODEL_POW2, + MASK=False, + ENABLE_DROPOUT=ENABLE_DROPOUT, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + ) + + adj_dq = bid * stride_dqb + hqid * stride_dqh + q_start * stride_dqm + offs_dq = offs_m[:, None] * stride_dqm + offs_k[None, :] * stride_dqk + dq *= sm_scale + tl.store(DQ + adj_dq + offs_dq, dq, mask=mask_q) + + +def attention_prefill_backward_triton_fused_atomics_impl( + do: torch.Tensor, + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + o: torch.Tensor, + softmax_lse: torch.Tensor, + dq: torch.Tensor, + dk: torch.Tensor, + dv: torch.Tensor, + sm_scale: float, + alibi_slopes: Optional[torch.Tensor], + causal: bool, + cu_seqlens_q: Optional[torch.Tensor], + cu_seqlens_k: Optional[torch.Tensor], + max_seqlen_q: int, + max_seqlen_k: int, + dropout_p: float, + philox_seed: Optional[int] = 0, + philox_offset: Optional[int] = 0, + descale_q: Optional[torch.Tensor] = None, + descale_k: Optional[torch.Tensor] = None, + descale_v: Optional[torch.Tensor] = None, + descale_do: Optional[torch.Tensor] = None, + fused: bool = False, + # seqused for FA v3 (currently ignored in this implementation) + seqused_q: Optional[torch.Tensor] = None, + seqused_k: Optional[torch.Tensor] = None, +): + IS_FP8 = is_fp8(q) + if IS_FP8: + FP8_MAX = torch.finfo(q.dtype).max + descale_strides = ( + descale_q.stride(0), + descale_k.stride(0), + descale_v.stride(0), + descale_do.stride(0), + ) + + if DEBUG: + print(f"FP8 path triggered in bwd_prefill_fused_atomics.py") + else: + FP8_MAX = None + stride_descale_q_z = stride_descale_k_z = stride_descale_v_z = ( + stride_descale_do_z + ) = None + descale_strides = ( + stride_descale_q_z, + stride_descale_k_z, + stride_descale_v_z, + stride_descale_do_z, + ) + + IS_VARLEN = True if cu_seqlens_q is not None else False + + # get strides and shape + if IS_VARLEN: + # Layout for q,k,v is thd ie [total tokens, num_head, head_dim] + batch, seqlen_q, num_q_heads, head_sz = ( + len(cu_seqlens_q) - 1, + max_seqlen_q, + q.shape[1], + q.shape[2], + ) + seqlen_k, num_k_heads = max_seqlen_k, k.shape[1] + q_strides = (0, q.stride(1), q.stride(0), q.stride(2)) + q_strides = (0, q.stride(1), q.stride(0), q.stride(2)) + k_strides = (0, k.stride(1), k.stride(0), k.stride(2)) + v_strides = (0, v.stride(1), v.stride(0), v.stride(2)) + o_strides = (0, o.stride(1), o.stride(0), o.stride(2)) + dq_strides = (0, dq.stride(1), dq.stride(0), dq.stride(2)) + dk_strides = (0, dk.stride(1), dk.stride(0), dk.stride(2)) + dv_strides = (0, dv.stride(1), dv.stride(0), dv.stride(2)) + do_strides = (0, do.stride(1), do.stride(0), do.stride(2)) + else: + # Layout for q,k,v is bshd ie [batch, seq_len, num_head, head_dim] + batch, seqlen_q, num_q_heads, head_sz = q.shape + seqlen_k, num_k_heads = k.shape[1], k.shape[2] + q_strides = (q.stride(0), q.stride(2), q.stride(1), q.stride(3)) + k_strides = (k.stride(0), k.stride(2), k.stride(1), k.stride(3)) + v_strides = (v.stride(0), v.stride(2), v.stride(1), v.stride(3)) + o_strides = (o.stride(0), o.stride(2), o.stride(1), o.stride(3)) + dq_strides = (dq.stride(0), dq.stride(2), dq.stride(1), dq.stride(3)) + dk_strides = (dk.stride(0), dk.stride(2), dk.stride(1), dk.stride(3)) + dv_strides = (dv.stride(0), dv.stride(2), dv.stride(1), dv.stride(3)) + do_strides = (do.stride(0), do.stride(2), do.stride(1), do.stride(3)) + + # BLOCK_D_MODEL, BLOCK_D_MODEL_POW2 + # padding for head_dim. Power of 2 or 16 + BLOCK_D_MODEL_POW2 = triton.next_power_of_2(head_sz) + BLOCK_D_MODEL_POW2 = max(BLOCK_D_MODEL_POW2, 16) + + # Configs + # PRE_BLOCK, BLOCK_M1, BLOCK_N1, BLOCK_M2, BLOCK_N2 + # BLK_SLICE_FACTOR + NUM_WARPS, NUM_STAGES = 4, 1 + WAVES_PER_EU = 1 + PRE_BLOCK = 128 + # BLOCK_M1, BLOCK_N1, BLOCK_M2, BLOCK_N2 = 32, 128, 128, 32 + BLOCK_M1, BLOCK_N1, BLOCK_M2, BLOCK_N2 = 64, 64, 64, 16 + BLK_SLICE_FACTOR = 2 + + # init delta + delta = torch.zeros_like(softmax_lse) + if IS_VARLEN: + # [total_tokens, num_q_heads, seqlen_q] + delta_strides = (0, delta.stride(1), delta.stride(0)) + else: + # [batch, num_q_heads, seqlen_q] + delta_strides = delta.stride() + + # preprocess + # compute D(delta) = rowsum(dO*O). Note, multiplication is element-wise. + pre_grid = (triton.cdiv(max_seqlen_q, PRE_BLOCK), batch, num_q_heads) + _bwd_preprocess[pre_grid]( + o, + do, + delta, + *o_strides, + *delta_strides, + descale_strides[3], + cu_seqlens_q, + max_seqlen_q, + descale_do, + BLOCK_M=PRE_BLOCK, + BLOCK_D_MODEL=head_sz, + BLOCK_D_MODEL_POW2=BLOCK_D_MODEL_POW2, + IS_VARLEN=IS_VARLEN, + IS_FP8=IS_FP8, + ) + + # dropout_mask + use_dropout = dropout_p > 0.0 + if use_dropout: + dropout_mask = torch.zeros( + (batch, num_q_heads, max_seqlen_q, max_seqlen_k), + device=q.device, + dtype=torch.float32, + ) + dropout_strides = dropout_mask.stride() + else: + dropout_mask = None + dropout_strides = (0, 0, 0, 0) + + grid_dkdv = ((max_seqlen_k + BLOCK_N1 - 1) // BLOCK_N1, batch, num_k_heads) + grid_dq = ((max_seqlen_q + BLOCK_M2 - 1) // BLOCK_M2, batch, num_k_heads) + + if ( + fused + ): # fuses dk, dv, dq computations into one kernel by computing the dq using atomic adds between workgroups + + BLOCK_N = ( + 128 if BLOCK_D_MODEL_POW2 < 160 else 64 + ) # larger head sizes lead to oom + config = { + "BLOCK_M": 32, + "BLOCK_N": BLOCK_N, + "num_warps": 4, + "num_stages": 1, + "waves_per_eu": 1, + "BLK_SLICE_FACTOR": 2, + } + + num_k_pids = (max_seqlen_k + BLOCK_N - 1) // BLOCK_N + grid_dkdvdq = (batch * num_k_heads * num_k_pids,) + + if causal: + _bwd_kernel_dkdvdq_causal[grid_dkdvdq]( + q, + k, + v, + sm_scale, + do, + dk, + dv, + dq, + softmax_lse, + delta, + *q_strides, + *k_strides, + *v_strides, + *dk_strides, + *dq_strides, + *delta_strides, + *do_strides, + *dropout_strides, + *descale_strides, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_mask, + dropout_p, + philox_seed, + philox_offset, + descale_q, + descale_k, + descale_v, + descale_do, + NUM_Q_HEADS=num_q_heads, + NUM_K_HEADS=num_k_heads, + BATCH=batch, + NUM_K_PIDS=num_k_pids, + BLOCK_D_MODEL=head_sz, + BLOCK_D_MODEL_POW2=BLOCK_D_MODEL_POW2, + ENABLE_DROPOUT=use_dropout, + IS_VARLEN=IS_VARLEN, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + **config, + ) + else: + _bwd_kernel_dkdvdq_noncausal[grid_dkdvdq]( + q, + k, + v, + sm_scale, + do, + dk, + dv, + dq, + softmax_lse, + delta, + *q_strides, + *k_strides, + *v_strides, + *dk_strides, + *dq_strides, + *delta_strides, + *do_strides, + *dropout_strides, + *descale_strides, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_mask, + dropout_p, + philox_seed, + philox_offset, + descale_q, + descale_k, + descale_v, + descale_do, + NUM_Q_HEADS=num_q_heads, + NUM_K_HEADS=num_k_heads, + BATCH=batch, + NUM_K_PIDS=num_k_pids, + BLOCK_D_MODEL=head_sz, + BLOCK_D_MODEL_POW2=BLOCK_D_MODEL_POW2, + ENABLE_DROPOUT=use_dropout, + IS_VARLEN=IS_VARLEN, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + **config, + ) + + return delta + + # split kernels solution: one kernel computes dk, dv and the other computes dq + + if causal: + _bwd_kernel_dkdv_causal[grid_dkdv]( + q, + k, + v, + sm_scale, + do, + dk, + dv, + softmax_lse, + delta, + *q_strides, + *k_strides, + *v_strides, + *dk_strides, + *delta_strides, + *do_strides, + *dropout_strides, + *descale_strides, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_mask, + dropout_p, + philox_seed, + philox_offset, + descale_q, + descale_k, + descale_v, + descale_do, + NUM_Q_HEADS=num_q_heads, + NUM_K_HEADS=num_k_heads, + BLOCK_M=BLOCK_M1, + BLOCK_N=BLOCK_N1, + BLK_SLICE_FACTOR=BLK_SLICE_FACTOR, + BLOCK_D_MODEL=head_sz, + BLOCK_D_MODEL_POW2=BLOCK_D_MODEL_POW2, + ENABLE_DROPOUT=use_dropout, + IS_VARLEN=IS_VARLEN, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + num_warps=NUM_WARPS, + num_stages=NUM_STAGES, + waves_per_eu=WAVES_PER_EU, + ) + _bwd_kernel_dq_causal[grid_dq]( + q, + k, + v, + sm_scale, + do, + dq, + softmax_lse, + delta, + *q_strides, + *k_strides, + *v_strides, + *dq_strides, + *delta_strides, + *do_strides, + *dropout_strides, + *descale_strides, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_mask, + dropout_p, + philox_seed, + philox_offset, + descale_q, + descale_k, + descale_v, + descale_do, + NUM_Q_HEADS=num_q_heads, + NUM_K_HEADS=num_k_heads, + BLOCK_M=BLOCK_M2, + BLOCK_N=BLOCK_N2, + BLK_SLICE_FACTOR=BLK_SLICE_FACTOR, + BLOCK_D_MODEL=head_sz, + BLOCK_D_MODEL_POW2=BLOCK_D_MODEL_POW2, + ENABLE_DROPOUT=use_dropout, + IS_VARLEN=IS_VARLEN, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + num_warps=NUM_WARPS, + num_stages=NUM_STAGES, + waves_per_eu=WAVES_PER_EU, + ) + else: + _bwd_kernel_dkdv_noncausal[grid_dkdv]( + q, + k, + v, + sm_scale, + do, + dk, + dv, + softmax_lse, + delta, + *q_strides, + *k_strides, + *v_strides, + *dk_strides, + *delta_strides, + *do_strides, + *dropout_strides, + *descale_strides, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_mask, + dropout_p, + philox_seed, + philox_offset, + descale_q, + descale_k, + descale_v, + descale_do, + NUM_Q_HEADS=num_q_heads, + NUM_K_HEADS=num_k_heads, + BLOCK_M=BLOCK_M1, + BLOCK_N=BLOCK_N1, + BLK_SLICE_FACTOR=BLK_SLICE_FACTOR, + BLOCK_D_MODEL=head_sz, + BLOCK_D_MODEL_POW2=BLOCK_D_MODEL_POW2, + ENABLE_DROPOUT=use_dropout, + IS_VARLEN=IS_VARLEN, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + num_warps=NUM_WARPS, + num_stages=NUM_STAGES, + waves_per_eu=WAVES_PER_EU, + ) + + _bwd_kernel_dq_noncausal[grid_dq]( + q, + k, + v, + sm_scale, + do, + dq, + softmax_lse, + delta, + *q_strides, + *k_strides, + *v_strides, + *dq_strides, + *delta_strides, + *do_strides, + *dropout_strides, + *descale_strides, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_mask, + dropout_p, + philox_seed, + philox_offset, + descale_q, + descale_k, + descale_v, + descale_do, + NUM_Q_HEADS=num_q_heads, + NUM_K_HEADS=num_k_heads, + BLOCK_M=BLOCK_M2, + BLOCK_N=BLOCK_N2, + BLK_SLICE_FACTOR=BLK_SLICE_FACTOR, + BLOCK_D_MODEL=head_sz, + BLOCK_D_MODEL_POW2=BLOCK_D_MODEL_POW2, + ENABLE_DROPOUT=use_dropout, + IS_VARLEN=IS_VARLEN, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + num_warps=NUM_WARPS, + num_stages=NUM_STAGES, + waves_per_eu=WAVES_PER_EU, + ) + + return delta diff --git a/aiter/ops/triton/flash_attn_triton_amd/bwd_prefill_fused_no_atomics.py b/aiter/ops/triton/flash_attn_triton_amd/bwd_prefill_fused_no_atomics.py new file mode 100755 index 0000000000..31aaaf7003 --- /dev/null +++ b/aiter/ops/triton/flash_attn_triton_amd/bwd_prefill_fused_no_atomics.py @@ -0,0 +1,2457 @@ +import os +import torch +import triton # type: ignore +import triton.language as tl # type: ignore +from typing import Literal, Optional +from .utils import ( + AUTOTUNE, + DROPOUT_USE_PYTORCH, + DROPOUT_DUMP, + DEBUG, + compute_fp8_scaling_factors, + create_dropout_mask, + create_dropout_mask_varlen, + is_cdna, + is_fp8, + is_rdna, + round_multiple, +) + +# Import atomics implementation (kept for kernel launch path) and expose a unified wrapper below. +from .bwd_prefill_fused_atomics import ( + attention_prefill_backward_triton_fused_atomics_impl, +) + +# NOTE: triton fails to import tl.constexprs so create them here for the file +tl_DROPOUT_USE_PYTORCH: tl.constexpr = triton.language.constexpr(DROPOUT_USE_PYTORCH) +tl_DROPOUT_DUMP: tl.constexpr = triton.language.constexpr(DROPOUT_DUMP) + + +def get_autotune_configs(): + if False: + if is_cdna(): + # shared meta-parameters + NUM_STAGES = 1 + NUM_WARPS = 4 + WAVES_PER_EU = 2 + MATRIX_INSTR_NONKDIM = 16 + + preprocess_autotune_configs = [ + triton.Config( + { + "PRE_BLOCK": 128, + "waves_per_eu": WAVES_PER_EU, + "matrix_instr_nonkdim": MATRIX_INSTR_NONKDIM, + }, + num_stages=NUM_STAGES, + num_warps=NUM_WARPS, + ), # og config + triton.Config( + { + "PRE_BLOCK": 64, + "waves_per_eu": WAVES_PER_EU, + "matrix_instr_nonkdim": MATRIX_INSTR_NONKDIM, + }, + num_stages=NUM_STAGES, + num_warps=NUM_WARPS, + ), + triton.Config( + { + "PRE_BLOCK": 32, + "waves_per_eu": WAVES_PER_EU, + "matrix_instr_nonkdim": MATRIX_INSTR_NONKDIM, + }, + num_stages=NUM_STAGES, + num_warps=NUM_WARPS, + ), + triton.Config( + { + "PRE_BLOCK": 16, + "waves_per_eu": WAVES_PER_EU, + "matrix_instr_nonkdim": MATRIX_INSTR_NONKDIM, + }, + num_stages=NUM_STAGES, + num_warps=NUM_WARPS, + ), + ] + preprocess_autotune_keys = [ + "IS_CAUSAL", + "dropout_p", + "MAX_SEQLENS_Q", + "MAX_SEQLENS_K", + "ACTUAL_HEAD_DIM_QK", + "ACTUAL_HEAD_DIM_V", + "IS_VARLEN", + "HQ", + "HK", + ] + causal_autotune_configs = [ + triton.Config( + { + "BLOCK_M1": 32, + "BLOCK_N1": 128, + "BLOCK_M2": 128, + "BLOCK_N2": 32, + "BLK_SLICE_FACTOR": 2, + "waves_per_eu": WAVES_PER_EU, + "matrix_instr_nonkdim": MATRIX_INSTR_NONKDIM, + }, + num_stages=NUM_STAGES, + num_warps=NUM_WARPS, + ), # og config + triton.Config( + { + "BLOCK_M1": 16, + "BLOCK_N1": 128, + "BLOCK_M2": 128, + "BLOCK_N2": 16, + "BLK_SLICE_FACTOR": 2, + "waves_per_eu": WAVES_PER_EU, + "matrix_instr_nonkdim": MATRIX_INSTR_NONKDIM, + }, + num_stages=NUM_STAGES, + num_warps=NUM_WARPS, + ), + triton.Config( + { + "BLOCK_M1": 16, + "BLOCK_N1": 64, + "BLOCK_M2": 64, + "BLOCK_N2": 16, + "BLK_SLICE_FACTOR": 2, + "waves_per_eu": WAVES_PER_EU, + "matrix_instr_nonkdim": MATRIX_INSTR_NONKDIM, + }, + num_stages=NUM_STAGES, + num_warps=NUM_WARPS, + ), + triton.Config( + { + "BLOCK_M1": 32, + "BLOCK_N1": 64, + "BLOCK_M2": 64, + "BLOCK_N2": 32, + "BLK_SLICE_FACTOR": 2, + "waves_per_eu": WAVES_PER_EU, + "matrix_instr_nonkdim": MATRIX_INSTR_NONKDIM, + }, + num_stages=NUM_STAGES, + num_warps=NUM_WARPS, + ), + ] + causal_autotune_keys = [ + "IS_CAUSAL", + "dropout_p", + "MAX_SEQLENS_Q", + "MAX_SEQLENS_K", + "ACTUAL_HEAD_DIM_QK", + "ACTUAL_HEAD_DIM_V", + "IS_VARLEN", + "HQ", + "HK", + ] + noncausal_autotune_configs = [ + triton.Config( + { + "BLOCK_M1": 32, + "BLOCK_N1": 128, + "BLOCK_M2": 128, + "BLOCK_N2": 32, + "BLK_SLICE_FACTOR": 2, + "waves_per_eu": WAVES_PER_EU, + "matrix_instr_nonkdim": MATRIX_INSTR_NONKDIM, + }, + num_stages=NUM_STAGES, + num_warps=NUM_WARPS, + ), # og config + triton.Config( + { + "BLOCK_M1": 16, + "BLOCK_N1": 128, + "BLOCK_M2": 128, + "BLOCK_N2": 16, + "BLK_SLICE_FACTOR": 2, + "waves_per_eu": WAVES_PER_EU, + "matrix_instr_nonkdim": MATRIX_INSTR_NONKDIM, + }, + num_stages=NUM_STAGES, + num_warps=NUM_WARPS, + ), + triton.Config( + { + "BLOCK_M1": 16, + "BLOCK_N1": 64, + "BLOCK_M2": 64, + "BLOCK_N2": 16, + "BLK_SLICE_FACTOR": 2, + "waves_per_eu": WAVES_PER_EU, + "matrix_instr_nonkdim": MATRIX_INSTR_NONKDIM, + }, + num_stages=NUM_STAGES, + num_warps=NUM_WARPS, + ), + triton.Config( + { + "BLOCK_M1": 32, + "BLOCK_N1": 64, + "BLOCK_M2": 64, + "BLOCK_N2": 32, + "BLK_SLICE_FACTOR": 2, + "waves_per_eu": WAVES_PER_EU, + "matrix_instr_nonkdim": MATRIX_INSTR_NONKDIM, + }, + num_stages=NUM_STAGES, + num_warps=NUM_WARPS, + ), + ] + noncausal_autotune_keys = [ + "IS_CAUSAL", + "dropout_p", + "MAX_SEQLENS_Q", + "MAX_SEQLENS_K", + "ACTUAL_HEAD_DIM_QK", + "ACTUAL_HEAD_DIM_V", + "IS_VARLEN", + "HQ", + "HK", + ] + + return ( + (preprocess_autotune_configs, preprocess_autotune_keys), + (causal_autotune_configs, causal_autotune_keys), + (noncausal_autotune_configs, noncausal_autotune_keys), + ) + else: + raise ValueError("Unknown Device Type") + else: + # meta-parameters + # TODO: fix num_stages later + NUM_WARPS, NUM_STAGES = 4, 1 + WAVES_PER_EU = 1 + PRE_BLOCK = 128 + BLOCK_M1, BLOCK_N1, BLOCK_M2, BLOCK_N2 = 32, 128, 128, 32 + BLK_SLICE_FACTOR = 2 + + assert BLOCK_N1 == BLOCK_M2 + + # configs for the kernels + preprocess_autotune_configs = [ + triton.Config( + {"PRE_BLOCK": PRE_BLOCK, "waves_per_eu": WAVES_PER_EU}, + num_stages=NUM_STAGES, + num_warps=NUM_WARPS, + ), + ] + preprocess_autotune_keys = [ + "max_seqlen_q", + "ACTUAL_HEAD_DIM_V", + "IS_VARLEN", + ] + causal_autotune_configs = [ + triton.Config( + { + "BLOCK_M1": BLOCK_M1, + "BLOCK_N1": BLOCK_N1, + "BLOCK_M2": BLOCK_M2, + "BLOCK_N2": BLOCK_N2, + "BLK_SLICE_FACTOR": BLK_SLICE_FACTOR, + "waves_per_eu": WAVES_PER_EU, + }, + num_stages=NUM_STAGES, + num_warps=NUM_WARPS, + ), + ] + causal_autotune_keys = [ + "dropout_p", + "max_seqlen_q", + "max_seqlen_k", + "ACTUAL_HEAD_DIM_QK", + "ACTUAL_HEAD_DIM_V", + "IS_VARLEN", + "HQ", + "HK", + ] + noncausal_autotune_configs = [ + triton.Config( + { + "BLOCK_M1": BLOCK_M1, + "BLOCK_N1": BLOCK_N1, + "BLOCK_M2": BLOCK_M2, + "BLOCK_N2": BLOCK_N2, + "BLK_SLICE_FACTOR": BLK_SLICE_FACTOR, + "waves_per_eu": WAVES_PER_EU, + }, + num_stages=NUM_STAGES, + num_warps=NUM_WARPS, + ), + ] + noncausal_autotune_keys = [ + "dropout_p", + "max_seqlen_q", + "max_seqlen_k", + "ACTUAL_HEAD_DIM_QK", + "ACTUAL_HEAD_DIM_V", + "IS_VARLEN", + "HQ", + "HK", + ] + return ( + (preprocess_autotune_configs, preprocess_autotune_keys), + (causal_autotune_configs, causal_autotune_keys), + (noncausal_autotune_configs, noncausal_autotune_keys), + ) + + +( + (preprocess_autotune_configs, preprocess_autotune_keys), + (causal_autotune_configs, causal_autotune_keys), + (noncausal_autotune_configs, noncausal_autotune_keys), +) = get_autotune_configs() + + +# This function computes delta given output Out and gradient DO +# Here is the I/O shape: +# Out: (batch, nhead_q, max_seqlens_q, headDim) +# DO: (batch, nhead_q, max_seqlens_q, headDim) +# Delta: (batch, nheads_q, max_seqlens_q) +@triton.autotune( + configs=preprocess_autotune_configs, + key=preprocess_autotune_keys, + use_cuda_graph=True, +) +@triton.jit +def _bwd_preprocess( + O, + DO, # noqa: E741 + Delta, + stride_ob, + stride_oh, + stride_om, + stride_od, + stride_dob, + stride_doh, + stride_dom, + stride_dod, + stride_delta_b, + stride_delta_h, + stride_delta_m, + stride_descale_do_z, + cu_seqlens_q, + max_seqlen_q, + Descale_do, + PRE_BLOCK: tl.constexpr, + HEAD_DIM_V: tl.constexpr, + ACTUAL_HEAD_DIM_V: tl.constexpr, + IS_VARLEN: tl.constexpr, + IS_FP8: tl.constexpr, +): + pid_m = tl.program_id(0) + bid = tl.program_id(1) + hid = tl.program_id(2) + # Handle varlen + if IS_VARLEN: + q_start = tl.load(cu_seqlens_q + bid) + q_end = tl.load(cu_seqlens_q + bid + 1) + seqlen_q = q_end - q_start + else: + q_start = 0 + seqlen_q = max_seqlen_q + + # Compute offsets + offs_m = pid_m * PRE_BLOCK + tl.arange(0, PRE_BLOCK) + offs_d = tl.arange(0, HEAD_DIM_V) + # pointer offsets for O & DO + off_o = ( + bid * stride_ob + + hid * stride_oh + + q_start * stride_om + + offs_m[:, None] * stride_om + + offs_d[None, :] * stride_od + ) # noqa: E741 + off_do = ( + bid * stride_dob + + hid * stride_doh + + q_start * stride_dom + + offs_m[:, None] * stride_dom + + offs_d[None, :] * stride_dod + ) + + # create masks + mask_m = offs_m < seqlen_q + mask_md = mask_m[:, None] + PADDED_HEAD_V: tl.constexpr = ACTUAL_HEAD_DIM_V != HEAD_DIM_V + if PADDED_HEAD_V: + mask_md &= offs_d[None, :] < ACTUAL_HEAD_DIM_V + # load + o = tl.load(O + off_o, mask=mask_md, other=0.0) + do = tl.load(DO + off_do, mask=mask_md, other=0.0) + # compute and write-back to delta + if IS_FP8: + off_descale_do = bid * stride_descale_do_z + hid + descale_do = tl.load(Descale_do + off_descale_do) + + # NOTE: do is in the fp8 range and o is not in fp8 + delta = tl.sum(o.to(tl.float32) * (do.to(tl.float32) * descale_do), axis=1) + else: + delta = tl.sum(o.to(tl.float32) * do.to(tl.float32), axis=1) + off_delta = ( + bid * stride_delta_b + + hid * stride_delta_h + + q_start * stride_delta_m + + offs_m * stride_delta_m + ) + tl.store(Delta + off_delta, delta, mask=mask_m) + + +# The main inner-loop logic for computing dK and dV. +@triton.jit +def _bwd_dkdv_inner( + dk, + dv, # output + Q, + k, + v, + DO, + M, + D, + sm_scale, # input tensor + stride_qm, + stride_qk, + stride_dom, + stride_dok, + stride_dropoutm, + stride_dropoutn, + stride_lse_m, + stride_delta_m, + BLOCK_M: tl.constexpr, # 16 + BLOCK_N: tl.constexpr, # 128 + HEAD_DIM_QK: tl.constexpr, # + HEAD_DIM_V: tl.constexpr, # + ACTUAL_HEAD_DIM_QK: tl.constexpr, # + ACTUAL_HEAD_DIM_V: tl.constexpr, # + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, + alibi_slope, + seqlen_q, + seqlen_k, # max sequence length for q and k + # Filled in by the wrapper. + start_n, + start_m, + num_steps, # iteration numbers + descale_q, + descale_k, + descale_v, + descale_do, # fp8 descale factors from user + MASK: tl.constexpr, # causal masking, only apply to tiles on mask diagonal + ENABLE_DROPOUT: tl.constexpr, # activate dropout + USE_ALIBI: tl.constexpr, + USE_EXP2: tl.constexpr, # activate exp2 + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, + DEBUG_TRITON: tl.constexpr, + DEBUG_TRITON_DETAIL: tl.constexpr, +): + # if HEAD_DIM is padded + PADDED_HEAD_QK: tl.constexpr = ACTUAL_HEAD_DIM_QK != HEAD_DIM_QK + PADDED_HEAD_V: tl.constexpr = ACTUAL_HEAD_DIM_V != HEAD_DIM_V + delta_qk = seqlen_q - seqlen_k + offs_m = start_m + tl.arange(0, BLOCK_M) # start_m + (0, 15) + offs_n = start_n + tl.arange(0, BLOCK_N) # start_m + (0, 127) + offs_k_qk = tl.arange(0, HEAD_DIM_QK) + offs_k_v = tl.arange(0, HEAD_DIM_V) + # mask to make sure not OOB of seqlen_q + mask_n = offs_n < seqlen_k + # Q and DO are (seqlen_q, head_dim) + # qT_ptrs = (1, BLOCK_M) + (HEAD_DIM_QK, 1), transpose of q + qT_ptrs = Q + offs_m[None, :] * stride_qm + offs_k_qk[:, None] * stride_qk + # do_ptrs = (BLOCK_M, 1) + (1, HEAD_DIM_V), NOT transposed + do_ptrs = DO + offs_m[:, None] * stride_dom + offs_k_v[None, :] * stride_dok + # BLOCK_N must be a multiple of BLOCK_M, otherwise the code wouldn't work. + tl.static_assert(BLOCK_N % BLOCK_M == 0) + curr_m = start_m + step_m = BLOCK_M + curr_philox_offset = batch_philox_offset + curr_dropout_offset = dropout_offset + RCP_LN2: tl.constexpr = 1.4426950408889634 # = 1.0 / ln(2) + + for blk_idx in range(num_steps): + if DEBUG_TRITON: + print(f"iter {blk_idx}: curr_m = {curr_m}") # noqa: E701 + offs_m = curr_m + tl.arange(0, BLOCK_M) + # update the mask because offs_m advanced + mask_m = offs_m < seqlen_q + mask_qT = mask_m[None, :] + mask_do = mask_m[:, None] + mask_nm = mask_n[:, None] & (offs_m[None, :] < seqlen_q) + if PADDED_HEAD_QK: + mask_qT &= offs_k_qk[:, None] < ACTUAL_HEAD_DIM_QK + if PADDED_HEAD_V: + mask_do &= offs_k_v[None, :] < ACTUAL_HEAD_DIM_V + qT = tl.load(qT_ptrs, mask=mask_qT, other=0.0) + # generate dropout mask + if ENABLE_DROPOUT: + # NOTE: dropout is transposed because it is used to mask pT + philox_offs = ( + curr_philox_offset + + offs_m[None, :] * stride_dropoutm + + offs_n[:, None] * stride_dropoutn + ) + if tl_DROPOUT_USE_PYTORCH: + dropout_offs = ( + offs_m[None, :] * stride_dropoutm + + offs_n[:, None] * stride_dropoutn + ) + dropout_mask = tl.load(curr_dropout_offset + dropout_offs, mask=mask_nm) + else: + rand_vals = tl.rand(philox_seed, philox_offs) + dropout_mask = rand_vals > dropout_p + dropout_scale = 1.0 / (1 - dropout_p) + # Load m before computing qk to reduce pipeline stall. + m = tl.load(M + offs_m * stride_lse_m, mask=mask_m, other=0.0) + if IS_FP8: + qkT = tl.dot(k, qT) * descale_q * descale_k + else: + qkT = tl.dot(k, qT) + qkT_scaled = qkT * sm_scale + + if USE_ALIBI: + relative_pos_block = offs_n[:, None] + seqlen_q - seqlen_k - offs_m[None, :] + alibi_block = -1 * alibi_slope * tl.abs(relative_pos_block) + qkT_scaled += alibi_block + + if DEBUG_TRITON_DETAIL: + if start_n == 256: + print(f"qT: {qT.shape}\n", qT) + print(f"k: {k.shape}\n", k) + print(f"qkT scaled: {qkT.shape}\n", qkT_scaled) + # TODO: remove the scaling of m later when we removed re-scaling in fwd + if USE_EXP2: + pT = tl.math.exp2(qkT_scaled * RCP_LN2 - m[None, :] * RCP_LN2) + else: + pT = tl.math.exp(qkT_scaled - m[None, :]) + + # Autoregressive masking. + if MASK: + # offset offs_m with delta_qk since the causal mask starts at + # bottom right of the (seqlen_q, seqlen_k) matrix + causal_mask = (offs_m[None, :] - delta_qk) >= offs_n[:, None] + mask = causal_mask & mask_nm + if DEBUG_TRITON_DETAIL: + if start_n == 256: + print(f"causal_mask: {causal_mask.shape}\n", causal_mask) + print( + f"qkT after causal: {qkT.shape}\n", + tl.where(causal_mask, qkT * sm_scale, 0.0), + ) + pT = tl.where(mask, pT, 0.0) + do = tl.load(do_ptrs, mask=mask_do, other=0.0) + # Compute dV. + if ENABLE_DROPOUT: + pT_dropout = tl.where(dropout_mask, pT, 0.0) * dropout_scale + if IS_FP8: + scale_p_dropout, descale_p_dropout = compute_fp8_scaling_factors( + pT_dropout, FP8_MAX + ) + dv += ( + tl.dot((pT_dropout * scale_p_dropout).to(do.type.element_ty), do) + * descale_p_dropout + * descale_do + ) + else: + dv += tl.dot(pT_dropout.to(do.type.element_ty), do) + else: + if IS_FP8: + scale_pT, descale_pT = compute_fp8_scaling_factors(pT, FP8_MAX) + dv += ( + tl.dot((pT * scale_pT).to(do.type.element_ty), do) + * descale_pT + * descale_do + ) + else: + dv += tl.dot(pT.to(do.type.element_ty), do) + + if DEBUG_TRITON_DETAIL: + if start_n == 256: + print(f"pT: {pT.shape}\n", pT) + # D (= delta) is pre-divided by ds_scale. + Di = tl.load(D + offs_m * stride_delta_m, mask=mask_m) + # Compute dP and dS. + if IS_FP8: + dpT = tl.dot(v, tl.trans(do)) * descale_v * descale_do + else: + dpT = tl.dot(v, tl.trans(do)) + if ENABLE_DROPOUT: + dpT = tl.where(dropout_mask, dpT, 0.0) * dropout_scale + delta_i = Di[None, :] + dsT = pT * (dpT - delta_i) + if IS_FP8: + scale_dsT, descale_dsT = compute_fp8_scaling_factors(dsT, FP8_MAX) + dk += ( + tl.dot((dsT * scale_dsT).to(qT.type.element_ty), tl.trans(qT)) + * descale_dsT + * descale_q + ) + else: + dk += tl.dot(dsT.to(qT.type.element_ty), tl.trans(qT)) + # Increment pointers. + curr_m += step_m + qT_ptrs += step_m * stride_qm + do_ptrs += step_m * stride_dom + return dk, dv + + +# the main inner-loop logic for computing dQ +@triton.jit +def _bwd_dq_inner( + dq, # output + q, + K, + V, + do, + m, + Delta, + sm_scale, # input + # shared by Q/K/V. + stride_qm, + stride_qk, + stride_kn, + stride_kk, + stride_vn, + stride_vk, + stride_dropoutm, + stride_dropoutn, # stride for dropout + stride_lse_m, + stride_delta_m, + seqlen_q, + seqlen_k, # + BLOCK_M2: tl.constexpr, # + BLOCK_N2: tl.constexpr, # + HEAD_DIM_QK: tl.constexpr, + HEAD_DIM_V: tl.constexpr, + ACTUAL_HEAD_DIM_QK: tl.constexpr, + ACTUAL_HEAD_DIM_V: tl.constexpr, # + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, + alibi_slope, + # Filled in by the wrapper. + start_m, + start_n, + end_n, + num_steps, # + descale_q, + descale_k, + descale_v, + descale_do, # fp8 descale factors from user + MASK: tl.constexpr, + ENABLE_DROPOUT: tl.constexpr, + USE_ALIBI: tl.constexpr, + USE_EXP2: tl.constexpr, + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, + DEBUG_TRITON: tl.constexpr, + DEBUG_TRITON_DETAIL: tl.constexpr, +): + # if HEAD_DIM is padded + PADDED_HEAD_QK: tl.constexpr = ACTUAL_HEAD_DIM_QK != HEAD_DIM_QK + PADDED_HEAD_V: tl.constexpr = ACTUAL_HEAD_DIM_V != HEAD_DIM_V + delta_qk = seqlen_q - seqlen_k + offs_m = start_m + tl.arange(0, BLOCK_M2) + offs_n = start_n + tl.arange(0, BLOCK_N2) + offs_k_qk = tl.arange(0, HEAD_DIM_QK) + offs_k_v = tl.arange(0, HEAD_DIM_V) + + # mask to make sure not OOB of seqlen_q + mask_m = offs_m < seqlen_q + + kT_ptrs = K + offs_n[None, :] * stride_kn + offs_k_qk[:, None] * stride_kk + vT_ptrs = V + offs_n[None, :] * stride_vn + offs_k_v[:, None] * stride_vk + # D (= delta) is pre-divided by ds_scale. + Di = tl.load(Delta + offs_m * stride_delta_m, mask=mask_m, other=0.0) + # BLOCK_M2 must be a multiple of BLOCK_N2, otherwise the code wouldn't work. + tl.static_assert(BLOCK_M2 % BLOCK_N2 == 0) + curr_n = start_n + step_n = BLOCK_N2 + curr_philox_offset = batch_philox_offset + curr_dropout_offset = dropout_offset + RCP_LN2: tl.constexpr = 1.4426950408889634 # = 1.0 / ln(2) + for blk_idx in range(num_steps): + if DEBUG_TRITON: + print(f"iter {blk_idx}: curr_n = {curr_n}") # noqa: E701 + offs_n = curr_n + tl.arange(0, BLOCK_N2) + # end_n is needed because the end of causal True might not be perfectly + # aligned with the end of the block + mask_n = offs_n < end_n + if DEBUG_TRITON_DETAIL: + print( + f"start_n = {start_n}, end_n = {end_n}, offs_n: {offs_n.shape}\n{offs_n}" + ) # noqa: E701 + if DEBUG_TRITON_DETAIL: + print(f"mask_n: {mask_n.shape}\n{mask_n}") # noqa: E701 + mask_kT = mask_n[None, :] + mask_vT = mask_n[None, :] + mask_mn = mask_m[:, None] & (offs_n[None, :] < end_n) + if PADDED_HEAD_QK: + mask_kT &= offs_k_qk[:, None] < ACTUAL_HEAD_DIM_QK + if PADDED_HEAD_V: + mask_vT &= offs_k_v[:, None] < ACTUAL_HEAD_DIM_V + + kT = tl.load(kT_ptrs, mask=mask_kT, other=0.0) + vT = tl.load(vT_ptrs, mask=mask_vT, other=0.0) + + if ENABLE_DROPOUT: + # NOTE: dropout is transposed because it is used to mask pT + philox_offs = ( + curr_philox_offset + + offs_m[:, None] * stride_dropoutm + + offs_n[None, :] * stride_dropoutn + ) + if tl_DROPOUT_USE_PYTORCH: + dropout_offs = ( + offs_m[:, None] * stride_dropoutm + + offs_n[None, :] * stride_dropoutn + ) + dropout_mask = tl.load(curr_dropout_offset + dropout_offs, mask=mask_mn) + else: + rand_vals = tl.rand(philox_seed, philox_offs) + dropout_mask = rand_vals > dropout_p + dropout_scale = 1 / (1 - dropout_p) + + if IS_FP8: + qk = tl.dot(q, kT) * descale_q * descale_k + else: + qk = tl.dot(q, kT) + qk_scaled = qk * sm_scale + + if USE_ALIBI: + relative_pos_block = offs_m[:, None] + seqlen_k - seqlen_q - offs_n[None, :] + alibi_block = -1 * alibi_slope * tl.abs(relative_pos_block) + qk_scaled += alibi_block + + if DEBUG_TRITON_DETAIL: + print(f"qk scaled: {qk.shape}\n", qk_scaled) # noqa: E701 + if USE_EXP2: + p = tl.math.exp2(qk_scaled * RCP_LN2 - m * RCP_LN2) + else: + p = tl.math.exp(qk_scaled - m) + + # Autoregressive masking. + if MASK: + causal_mask = (offs_m[:, None] - delta_qk) >= offs_n[None, :] + mask = causal_mask & mask_mn + p = tl.where(mask, p, 0.0) + # Compute dP and dS. + if IS_FP8: + dp = tl.dot(do, vT) * descale_do * descale_v + else: + dp = tl.dot(do, vT) + if ENABLE_DROPOUT: + dp = tl.where(dropout_mask, dp, 0.0) * dropout_scale + delta_i = Di[:, None] + ds = p * (dp - delta_i) + # Compute dQ. + # NOTE: We need to de-scale dq in the end, because kT was pre-scaled. + if IS_FP8: + scale_ds, descale_ds = compute_fp8_scaling_factors(ds, FP8_MAX) + dq += ( + tl.dot((ds * scale_ds).to(kT.type.element_ty), tl.trans(kT)) + * descale_ds + * descale_k + ) + else: + dq += tl.dot(ds.to(kT.type.element_ty), tl.trans(kT)) + # Increment pointers. + curr_n += step_n + kT_ptrs += step_n * stride_kn + vT_ptrs += step_n * stride_vn + return dq + + +@triton.autotune( + configs=causal_autotune_configs, + key=causal_autotune_keys, + use_cuda_graph=True, +) +@triton.jit +def bwd_kernel_causal( # grid = (nheads_k, tl.cdiv(max_seqlen_q // BLOCK_M2), batch) + Q, + K, + V, + sm_scale, + DO, + DQ, + DK, + DV, + M, + Delta, + stride_qb, + stride_qh, + stride_qm, + stride_qd, + stride_kb, + stride_kh, + stride_kn, + stride_kd, + stride_vb, + stride_vh, + stride_vn, + stride_vd, + stride_dqb, + stride_dqh, + stride_dqm, + stride_dqd, + stride_dkb, + stride_dkh, + stride_dkn, + stride_dkd, + stride_dvb, + stride_dvh, + stride_dvn, + stride_dvd, + stride_lse_b, + stride_lse_h, + stride_lse_m, + stride_delta_b, + stride_delta_h, + stride_delta_m, + stride_dob, + stride_doh, + stride_dom, + stride_dod, + stride_dropoutb, + stride_dropouth, + stride_dropoutm, + stride_dropoutn, + stride_descale_q_z, + stride_descale_k_z, + stride_descale_v_z, + stride_descale_do_z, + stride_az, + stride_ah, + HQ, + HK, + cu_seqlens_q, + cu_seqlens_k, + seqused_q, + seqused_k, # Add seqused parameters + max_seqlen_q, + max_seqlen_k, + Dropout_mask, + dropout_p, + philox_seed, + philox_offset_base, + Alibi_slopes, + Descale_q, + Descale_k, + Descale_v, + Descale_do, + BLOCK_M1: tl.constexpr, + BLOCK_N1: tl.constexpr, + BLOCK_M2: tl.constexpr, + BLOCK_N2: tl.constexpr, + BLK_SLICE_FACTOR: tl.constexpr, + HEAD_DIM_QK: tl.constexpr, + HEAD_DIM_V: tl.constexpr, + ACTUAL_HEAD_DIM_QK: tl.constexpr, + ACTUAL_HEAD_DIM_V: tl.constexpr, + ENABLE_DROPOUT: tl.constexpr, + IS_VARLEN: tl.constexpr, + USE_ALIBI: tl.constexpr, + USE_EXP2: tl.constexpr, + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, + FP8_OUTPUT: tl.constexpr, + USE_SEQUSED: tl.constexpr, # Add flag for seqused + DEBUG_TRITON: tl.constexpr, + DEBUG_TRITON_DETAIL: tl.constexpr, +): + # program ids + hkid = tl.program_id(0) + pid = tl.program_id(1) + bid = tl.program_id(2) + if DEBUG_TRITON: + print(f"\npid: {pid}, bid: {bid}, hkid: {hkid}") # noqa: E701 + # figure out varlen start and end + q_start = 0 + k_start = 0 + seqlen_q = max_seqlen_q + seqlen_k = max_seqlen_k + if IS_VARLEN: + # Compute actual sequence lengths + q_start = tl.load(cu_seqlens_q + bid) + q_end = tl.load(cu_seqlens_q + bid + 1) + k_start = tl.load(cu_seqlens_k + bid) + k_end = tl.load(cu_seqlens_k + bid + 1) + + # If seqused is provided, use it to limit the actual sequence length + if USE_SEQUSED: + actual_seqlen_q = ( + tl.load(seqused_q + bid) if seqused_q is not None else q_end - q_start + ) + seqlen_q = tl.minimum(actual_seqlen_q, q_end - q_start) + actual_seqlen_k = ( + tl.load(seqused_k + bid) if seqused_k is not None else k_end - k_start + ) + seqlen_k = tl.minimum(actual_seqlen_k, k_end - k_start) + else: + seqlen_q = q_end - q_start + seqlen_k = k_end - k_start + + delta_qk = seqlen_q - seqlen_k + if DEBUG_TRITON: + print(f"delta_qk = {delta_qk}") # noqa: E701 + PADDED_HEAD_QK: tl.constexpr = ACTUAL_HEAD_DIM_QK != HEAD_DIM_QK + PADDED_HEAD_V: tl.constexpr = ACTUAL_HEAD_DIM_V != HEAD_DIM_V + offs_d_qk = tl.arange(0, HEAD_DIM_QK) + offs_d_v = tl.arange(0, HEAD_DIM_V) + GROUP_SIZE: tl.constexpr = HQ // HK + + # align the delta_qk + start_n = pid * BLOCK_N1 + if start_n < seqlen_k: + # This section does dk and dv + dk = tl.zeros([BLOCK_N1, HEAD_DIM_QK], dtype=tl.float32) + dv = tl.zeros([BLOCK_N1, HEAD_DIM_V], dtype=tl.float32) + + # q > k: diretcly skip all the way until the start of causal block + start_delta_q_gt_k = delta_qk + # q < k: some blocks will have no Masked block, other needs to re-calc + # starting position + # delta_qk is negative so flip it, only multiple of BLOCK_N can skip the + # masked op + num_blocks_skip = -delta_qk // BLOCK_N1 + delta_aligned = (num_blocks_skip + 1) * BLOCK_N1 + delta_qk + start_delta_q_lt_k = delta_aligned // BLOCK_M1 * BLOCK_M1 + if delta_qk >= 0: + start_delta = delta_qk + if DEBUG_TRITON: + print( + f"q >= k: start_delta = delta_qk aligned to BLOCK_M = {start_delta_q_gt_k}" + ) # noqa: E701 + else: + start_delta = start_delta_q_lt_k + if DEBUG_TRITON: + print( + f"q < k: start_delta = residue btw multiple BLOCK_N and delta_qk = {delta_aligned} = aligned to BLOCK_M = {start_delta_q_lt_k}" + ) # noqa: E701 + + offs_n = start_n + tl.arange(0, BLOCK_N1) + # Mask for loading K and V + mask_k = offs_n[:, None] < seqlen_k + mask_v = offs_n[:, None] < seqlen_k + if PADDED_HEAD_QK: + mask_d_qk = offs_d_qk < ACTUAL_HEAD_DIM_QK + mask_k &= mask_d_qk[None, :] + if PADDED_HEAD_V: + mask_d_v = offs_d_v < ACTUAL_HEAD_DIM_V + mask_v &= mask_d_v[None, :] + + # K/V tensors not changed for the group + adj_k = ( + bid * stride_kb + + hkid * stride_kh + + k_start * stride_kn + + offs_n[:, None] * stride_kn + + offs_d_qk[None, :] * stride_kd + ) + adj_v = ( + bid * stride_vb + + hkid * stride_vh + + k_start * stride_vn + + offs_n[:, None] * stride_vn + + offs_d_v[None, :] * stride_vd + ) + # load K and V: they stay in SRAM throughout the inner loop. + k = tl.load(K + adj_k, mask=mask_k, other=0.0) + v = tl.load(V + adj_v, mask=mask_v, other=0.0) + # If MQA / GQA, set the K and V head offsets appropriately. + # hqid = hkid + for hqid in range(hkid * GROUP_SIZE, hkid * GROUP_SIZE + GROUP_SIZE): + if delta_qk >= 0: + start_m = start_n + start_delta + len_m = BLOCK_N1 + else: + start_m = max(start_n + delta_qk, 0) + start_m = start_m // BLOCK_M1 * BLOCK_M1 + # because we might shift the masked blocks up, we are deeper into + # the masked out region, so we would potentially increase the total + # steps with masked operation to get out of it + residue_m = max(start_n + delta_qk - start_m, 0) + len_m = BLOCK_N1 + residue_m + if DEBUG_TRITON: + print(f"residue_m = {residue_m}") # noqa: E701 + + # offset input and output tensor by batch and Q/K heads + adj_q = bid * stride_qb + hqid * stride_qh + q_start * stride_qm + Q_ptr = Q + adj_q + adj_do = bid * stride_dob + hqid * stride_doh + q_start * stride_dom + DO_ptr = DO + adj_do + adj_delta = ( + bid * stride_delta_b + hqid * stride_delta_h + q_start * stride_delta_m + ) + Delta_ptr = Delta + adj_delta + adj_m = bid * stride_lse_b + hqid * stride_lse_h + q_start * stride_lse_m + M_ptr = M + adj_m + + if USE_ALIBI: + alibi_offset = bid * stride_az + hqid * stride_ah + alibi_slope = tl.load(Alibi_slopes + alibi_offset) + else: + alibi_slope = None + + # batch_philox_offset is the ACTUALLY dropout offset + # dropout_offset is for debug purpose and will be removed later + batch_philox_offset = 0 + dropout_offset = 0 + if ENABLE_DROPOUT: + batch_philox_offset = ( + philox_offset_base + bid * stride_dropoutb + hqid * stride_dropouth + ) + dropout_offset = ( + Dropout_mask + bid * stride_dropoutb + hqid * stride_dropouth + ) + + if IS_FP8: + descale_q = tl.load(Descale_q + bid * stride_descale_q_z + hqid) + descale_k = tl.load(Descale_k + bid * stride_descale_k_z + hkid) + descale_v = tl.load(Descale_v + bid * stride_descale_v_z + hkid) + descale_do = tl.load(Descale_do + bid * stride_descale_do_z + hqid) + else: + descale_q, descale_k, descale_v, descale_do = 1.0, 1.0, 1.0, 1.0 + + MASK_BLOCK_M1: tl.constexpr = BLOCK_M1 // BLK_SLICE_FACTOR + # bound the masked operation to q len so it does not have to wast cycles + len_m = min(len_m, seqlen_q) + num_steps = tl.cdiv(len_m, MASK_BLOCK_M1) + # when q < k, we may skip the initial masked op + if pid < num_blocks_skip: + num_steps = 0 + + # if start_m is negative, the current N-tile has no block on the + # diagonal of causal mask, so everything have no causal mask + if DEBUG_TRITON: + print( + f"Masked: start_n: {start_n}; start_m: {start_m}, num_steps: {num_steps}" + ) # noqa: E701 + dk, dv = _bwd_dkdv_inner( + dk, + dv, # output tensors + Q_ptr, + k, + v, + DO_ptr, + M_ptr, + Delta_ptr, + sm_scale, # input tensors + stride_qm, + stride_qd, # strides for q + stride_dom, + stride_dod, # strides for o + stride_dropoutm, + stride_dropoutn, # strides for dropout + stride_lse_m, + stride_delta_m, + MASK_BLOCK_M1, + BLOCK_N1, # block dim + HEAD_DIM_QK, + HEAD_DIM_V, + ACTUAL_HEAD_DIM_QK, + ACTUAL_HEAD_DIM_V, # head dim + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, + alibi_slope, + seqlen_q, + seqlen_k, # max sequence length for q and k + start_n, + start_m, + num_steps, # iteration numbers + descale_q, + descale_k, + descale_v, + descale_do, + MASK=True, # causal masking + ENABLE_DROPOUT=ENABLE_DROPOUT, # activate dropout + USE_ALIBI=USE_ALIBI, + USE_EXP2=USE_EXP2, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + DEBUG_TRITON=DEBUG_TRITON, + DEBUG_TRITON_DETAIL=DEBUG_TRITON_DETAIL, + ) + start_m += num_steps * MASK_BLOCK_M1 + num_steps = tl.cdiv(seqlen_q - start_m, BLOCK_M1) + end_m = start_m + num_steps * BLOCK_M1 + + if DEBUG_TRITON: + print( + f"start_m after Masked step: {start_m}; num_steps: {num_steps}" + ) # noqa: E701 + if DEBUG_TRITON: + print( + f"unMasked: start_n: {start_n}, start_m: {start_m}, end_m: {end_m}, num_steps: {num_steps}" + ) # noqa: E701 + if DEBUG_TRITON: + print("unMasked") # noqa: E701 + dk, dv = _bwd_dkdv_inner( + dk, + dv, # output tensors + Q_ptr, + k, + v, + DO_ptr, + M_ptr, + Delta_ptr, + sm_scale, # input tensors + stride_qm, + stride_qd, # strides for q + stride_dom, + stride_dod, # strides for o + stride_dropoutm, + stride_dropoutn, # strides for dropout + stride_lse_m, + stride_delta_m, + BLOCK_M1, + BLOCK_N1, # block dim + HEAD_DIM_QK, + HEAD_DIM_V, + ACTUAL_HEAD_DIM_QK, + ACTUAL_HEAD_DIM_V, # head dim + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, + alibi_slope, + seqlen_q, + seqlen_k, # max sequence length for q and k + start_n, + start_m, + num_steps, # iteration numbers + descale_q, + descale_k, + descale_v, + descale_do, + MASK=False, # causal masking + ENABLE_DROPOUT=ENABLE_DROPOUT, # activate dropout + USE_ALIBI=USE_ALIBI, + USE_EXP2=USE_EXP2, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + DEBUG_TRITON=DEBUG_TRITON, + DEBUG_TRITON_DETAIL=DEBUG_TRITON_DETAIL, + ) + # end of GQA/MQA of dkdv + # Write back dV + adj_dv = bid * stride_dvb + hkid * stride_dvh + k_start * stride_dvn + offs_dv = offs_n[:, None] * stride_dvn + offs_d_v[None, :] * stride_dvd + tl.store(DV + adj_dv + offs_dv, dv, mask=mask_v) + # write back dk + adj_dk = bid * stride_dkb + hkid * stride_dkh + k_start * stride_dkn + offs_dk = offs_n[:, None] * stride_dkn + offs_d_qk[None, :] * stride_dkd + dk *= sm_scale + tl.store(DK + adj_dk + offs_dk, dk, mask=mask_k) + + # This part does dq + start_m = pid * BLOCK_M2 + if start_m < seqlen_q: + # seqlen_q > seqlen_k, no need to process these tile for dq + if DEBUG_TRITON: + print( + f"end_n = start_m + BLOCK_M = {start_m} + {BLOCK_M2} = {start_m + BLOCK_M2}" + ) # noqa: E701 + if start_m + BLOCK_M2 < delta_qk: + if DEBUG_TRITON: + print( + f"start_m + BLOCK_M2 = {start_m} + {BLOCK_M2} = {start_m + BLOCK_M2} < delta_qk of {delta_qk}" + ) # noqa: E701 + return + + offs_m = start_m + tl.arange(0, BLOCK_M2) + # Mask for loading K and V + mask_q = offs_m[:, None] < seqlen_q + mask_do = offs_m[:, None] < seqlen_q + if PADDED_HEAD_QK: + mask_d_qk = offs_d_qk < ACTUAL_HEAD_DIM_QK + mask_q &= mask_d_qk[None, :] + if PADDED_HEAD_V: + mask_d_v = offs_d_v < ACTUAL_HEAD_DIM_V + mask_do &= mask_d_v[None, :] + offs_q = offs_m[:, None] * stride_qm + offs_d_qk[None, :] * stride_qd + offs_do = offs_m[:, None] * stride_dom + offs_d_v[None, :] * stride_dod + # NOTE: don't assume that the strides for k and v are the same! + K += bid * stride_kb + hkid * stride_kh + k_start * stride_kn + V += bid * stride_vb + hkid * stride_vh + k_start * stride_vn + + # If MQA / GQA, set the K and V head offsets appropriately. + for hqid in range(hkid * GROUP_SIZE, hkid * GROUP_SIZE + GROUP_SIZE): + # seqlen_q < seqlen_k: delta_qk more kv tokens are added at the front + # for every M-tile + end_n = start_m + BLOCK_M2 - delta_qk + # clamp end_n at [0, seqlen_k] + end_n = max(min(end_n, seqlen_k), 0) + if DEBUG_TRITON: + print(f"delta_qk: {delta_qk}; end_n: {end_n}") # noqa: E701 + # offset input and output tensor by batch and Q/K heads + adj_q = bid * stride_qb + hqid * stride_qh + q_start * stride_qm + adj_do = bid * stride_dob + hqid * stride_doh + q_start * stride_dom + adj_delta = ( + bid * stride_delta_b + hqid * stride_delta_h + q_start * stride_delta_m + ) + Delta_ptr = Delta + adj_delta + adj_m = bid * stride_lse_b + hqid * stride_lse_h + q_start * stride_lse_m + M_ptr = M + adj_m + + if USE_ALIBI: + alibi_offset = bid * stride_az + hqid * stride_ah + alibi_slope = tl.load(Alibi_slopes + alibi_offset) + else: + alibi_slope = None + + # batch_philox_offset is the ACTUALLY dropout offset + # dropout_offset is for debug purpose and will be removed later + batch_philox_offset = 0 + dropout_offset = 0 + if ENABLE_DROPOUT: + batch_philox_offset = ( + philox_offset_base + bid * stride_dropoutb + hqid * stride_dropouth + ) + dropout_offset = ( + Dropout_mask + bid * stride_dropoutb + hqid * stride_dropouth + ) + q = tl.load(Q + adj_q + offs_q, mask=mask_q, other=0.0) + do = tl.load(DO + adj_do + offs_do, mask=mask_do, other=0.0) + m = tl.load(M + adj_m + offs_m * stride_lse_m, mask=offs_m < seqlen_q) + m = m[:, None] + + MASK_BLOCK_N2: tl.constexpr = BLOCK_N2 // BLK_SLICE_FACTOR + # start can only be 0 at minimum + start_n = max(end_n - BLOCK_M2, 0) + num_steps = tl.cdiv(end_n - start_n, MASK_BLOCK_N2) + + if IS_FP8: + descale_q = tl.load(Descale_q + bid * stride_descale_q_z + hqid) + descale_k = tl.load(Descale_k + bid * stride_descale_k_z + hkid) + descale_v = tl.load(Descale_v + bid * stride_descale_v_z + hkid) + descale_do = tl.load(Descale_do + bid * stride_descale_do_z + hqid) + else: + descale_q, descale_k, descale_v, descale_do = 1.0, 1.0, 1.0, 1.0 + + dq = tl.zeros([BLOCK_M2, HEAD_DIM_QK], dtype=tl.float32) + dq = _bwd_dq_inner( + dq, + q, + K, + V, + do, + m, + Delta_ptr, + sm_scale, + stride_qm, + stride_qd, + stride_kn, + stride_kd, + stride_vn, + stride_vd, + stride_dropoutm, + stride_dropoutn, + stride_lse_m, + stride_delta_m, + seqlen_q, + seqlen_k, + BLOCK_M2, + MASK_BLOCK_N2, + HEAD_DIM_QK, + HEAD_DIM_V, + ACTUAL_HEAD_DIM_QK, + ACTUAL_HEAD_DIM_V, + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, + alibi_slope, + start_m, + start_n, + end_n, + num_steps, + descale_q, + descale_k, + descale_v, + descale_do, + MASK=True, # + ENABLE_DROPOUT=ENABLE_DROPOUT, + USE_ALIBI=USE_ALIBI, + USE_EXP2=USE_EXP2, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + DEBUG_TRITON=DEBUG_TRITON, + DEBUG_TRITON_DETAIL=DEBUG_TRITON_DETAIL, + ) + end_n -= num_steps * MASK_BLOCK_N2 + num_steps = tl.cdiv(end_n, BLOCK_N2) + start_n = max(end_n - num_steps * BLOCK_N2, 0) + if DEBUG_TRITON: + print( + f"unMasked: start_m: {start_m}, start_n: {start_n}, end_n: {end_n}, num_steps: {num_steps}" + ) # noqa: E701 + dq = _bwd_dq_inner( + dq, + q, + K, + V, + do, + m, + Delta_ptr, + sm_scale, + stride_qm, + stride_qd, + stride_kn, + stride_kd, + stride_vn, + stride_vd, + stride_dropoutm, + stride_dropoutn, + stride_lse_m, + stride_delta_m, + seqlen_q, + seqlen_k, + BLOCK_M2, + BLOCK_N2, + HEAD_DIM_QK, + HEAD_DIM_V, + ACTUAL_HEAD_DIM_QK, + ACTUAL_HEAD_DIM_V, + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, + alibi_slope, + start_m, + start_n, + end_n, + num_steps, + descale_q, + descale_k, + descale_v, + descale_do, + MASK=False, + ENABLE_DROPOUT=ENABLE_DROPOUT, + USE_ALIBI=USE_ALIBI, + USE_EXP2=USE_EXP2, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + DEBUG_TRITON=DEBUG_TRITON, + DEBUG_TRITON_DETAIL=DEBUG_TRITON_DETAIL, + ) + # Write back dQ. + adj_dq = bid * stride_dqb + hqid * stride_dqh + q_start * stride_dqm + offs_dq = offs_m[:, None] * stride_dqm + offs_d_qk[None, :] * stride_dqd + dq *= sm_scale + tl.store(DQ + adj_dq + offs_dq, dq, mask=mask_q) + # end of GQA/MQA of dq + + +@triton.autotune( + configs=noncausal_autotune_configs, + key=noncausal_autotune_keys, + use_cuda_graph=True, +) +@triton.jit +def bwd_kernel_noncausal( + Q, + K, + V, + sm_scale, + DO, + DQ, + DK, + DV, + M, + Delta, + stride_qb, + stride_qh, + stride_qm, + stride_qd, + stride_kb, + stride_kh, + stride_kn, + stride_kd, + stride_vb, + stride_vh, + stride_vn, + stride_vd, + stride_dqb, + stride_dqh, + stride_dqm, + stride_dqd, + stride_dkb, + stride_dkh, + stride_dkn, + stride_dkd, + stride_dvb, + stride_dvh, + stride_dvn, + stride_dvd, + stride_lse_b, + stride_lse_h, + stride_lse_m, + stride_delta_b, + stride_delta_h, + stride_delta_m, + stride_dob, + stride_doh, + stride_dom, + stride_dod, + stride_dropoutb, + stride_dropouth, + stride_dropoutm, + stride_dropoutn, + stride_descale_q_z, + stride_descale_k_z, + stride_descale_v_z, + stride_descale_do_z, + stride_az, + stride_ah, + HQ, + HK, + cu_seqlens_q, + cu_seqlens_k, + seqused_q, + seqused_k, # Add seqused parameters + max_seqlen_q, + max_seqlen_k, + Dropout_mask, + dropout_p, + philox_seed, + philox_offset_base, + Alibi_slopes, + Descale_q, + Descale_k, + Descale_v, + Descale_do, + BLOCK_M1: tl.constexpr, # 32 + BLOCK_N1: tl.constexpr, # 128 + BLOCK_M2: tl.constexpr, # 128 + BLOCK_N2: tl.constexpr, # 32 + BLK_SLICE_FACTOR: tl.constexpr, + HEAD_DIM_QK: tl.constexpr, + HEAD_DIM_V: tl.constexpr, + ACTUAL_HEAD_DIM_QK: tl.constexpr, + ACTUAL_HEAD_DIM_V: tl.constexpr, + ENABLE_DROPOUT: tl.constexpr, + IS_VARLEN: tl.constexpr, + USE_ALIBI: tl.constexpr, + USE_EXP2: tl.constexpr, + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, + FP8_OUTPUT: tl.constexpr, + USE_SEQUSED: tl.constexpr, # Add flag for seqused + DEBUG_TRITON: tl.constexpr, + DEBUG_TRITON_DETAIL: tl.constexpr, +): + # program ids + hkid = tl.program_id(0) + pid = tl.program_id(1) + bid = tl.program_id(2) + if DEBUG_TRITON: + print(f"\npid: {pid}, bid: {bid}, hkid: {hkid}") # noqa: E701 + # figure out varlen start and end + q_start = 0 + k_start = 0 + seqlen_q = max_seqlen_q + seqlen_k = max_seqlen_k + if IS_VARLEN: + # Compute actual sequence lengths + q_start = tl.load(cu_seqlens_q + bid) + q_end = tl.load(cu_seqlens_q + bid + 1) + k_start = tl.load(cu_seqlens_k + bid) + k_end = tl.load(cu_seqlens_k + bid + 1) + + # If seqused is provided, use it to limit the actual sequence length + if USE_SEQUSED: + actual_seqlen_q = ( + tl.load(seqused_q + bid) if seqused_q is not None else q_end - q_start + ) + seqlen_q = tl.minimum(actual_seqlen_q, q_end - q_start) + actual_seqlen_k = ( + tl.load(seqused_k + bid) if seqused_k is not None else k_end - k_start + ) + seqlen_k = tl.minimum(actual_seqlen_k, k_end - k_start) + else: + seqlen_q = q_end - q_start + seqlen_k = k_end - k_start + + PADDED_HEAD_QK: tl.constexpr = ACTUAL_HEAD_DIM_QK != HEAD_DIM_QK + PADDED_HEAD_V: tl.constexpr = ACTUAL_HEAD_DIM_V != HEAD_DIM_V + offs_d_qk = tl.arange(0, HEAD_DIM_QK) + offs_d_v = tl.arange(0, HEAD_DIM_V) + GROUP_SIZE: tl.constexpr = HQ // HK + + start_n = pid * BLOCK_N1 + if start_n < seqlen_k: + dk = tl.zeros([BLOCK_N1, HEAD_DIM_QK], dtype=tl.float32) + dv = tl.zeros([BLOCK_N1, HEAD_DIM_V], dtype=tl.float32) + + offs_n = start_n + tl.arange(0, BLOCK_N1) + # Mask for loading K and V + mask_k = offs_n[:, None] < seqlen_k + mask_v = offs_n[:, None] < seqlen_k + if PADDED_HEAD_QK: + mask_d_qk = offs_d_qk < ACTUAL_HEAD_DIM_QK + mask_k &= mask_d_qk[None, :] + if PADDED_HEAD_V: + mask_d_v = offs_d_v < ACTUAL_HEAD_DIM_V + mask_v &= mask_d_v[None, :] + # NOTE: don't assume that the strides for k and v are the same! + # K/V tensors not changed for the group + adj_k = ( + bid * stride_kb + + hkid * stride_kh + + k_start * stride_kn + + offs_n[:, None] * stride_kn + + offs_d_qk[None, :] * stride_kd + ) + adj_v = ( + bid * stride_vb + + hkid * stride_vh + + k_start * stride_vn + + offs_n[:, None] * stride_vn + + offs_d_v[None, :] * stride_vd + ) + # load K and V: they stay in SRAM throughout the inner loop. + k = tl.load(K + adj_k, mask=mask_k, other=0.0) + v = tl.load(V + adj_v, mask=mask_v, other=0.0) + # If MQA / GQA, set the K and V head offsets appropriately. + for hqid in range(hkid * GROUP_SIZE, hkid * GROUP_SIZE + GROUP_SIZE): + # offset input and output tensor by batch and Q/K heads + adj_q = bid * stride_qb + hqid * stride_qh + q_start * stride_qm + Q_ptr = Q + adj_q + adj_do = bid * stride_dob + hqid * stride_doh + q_start * stride_dom + DO_ptr = DO + adj_do + adj_delta = ( + bid * stride_delta_b + hqid * stride_delta_h + q_start * stride_delta_m + ) + Delta_ptr = Delta + adj_delta + adj_m = bid * stride_lse_b + hqid * stride_lse_h + q_start * stride_lse_m + M_ptr = M + adj_m + + if USE_ALIBI: + alibi_offset = bid * stride_az + hqid * stride_ah + alibi_slope = tl.load(Alibi_slopes + alibi_offset) + else: + alibi_slope = None + + # batch_philox_offset is the ACTUALLY dropout offset + # dropout_offset is for debug purpose and will be removed later + batch_philox_offset = 0 + dropout_offset = 0 + if ENABLE_DROPOUT: + batch_philox_offset = ( + philox_offset_base + bid * stride_dropoutb + hqid * stride_dropouth + ) + dropout_offset = ( + Dropout_mask + bid * stride_dropoutb + hqid * stride_dropouth + ) + + if IS_FP8: + descale_q = tl.load(Descale_q + bid * stride_descale_q_z + hqid) + descale_k = tl.load(Descale_k + bid * stride_descale_k_z + hkid) + descale_v = tl.load(Descale_v + bid * stride_descale_v_z + hkid) + descale_do = tl.load(Descale_do + bid * stride_descale_do_z + hqid) + else: + descale_q, descale_k, descale_v, descale_do = 1.0, 1.0, 1.0, 1.0 + + # because there is no causal, we always start from the beginning + start_m = 0 + num_steps = tl.cdiv(seqlen_q, BLOCK_M1) + dk, dv = _bwd_dkdv_inner( + dk, + dv, # output tensors + Q_ptr, + k, + v, + DO_ptr, + M_ptr, + Delta_ptr, + sm_scale, # input tensors + stride_qm, + stride_qd, # strides for q + stride_dom, + stride_dod, # strides for o + stride_dropoutm, + stride_dropoutn, # strides for dropout + stride_lse_m, + stride_delta_m, + BLOCK_M1, + BLOCK_N1, # block dim + HEAD_DIM_QK, + HEAD_DIM_V, + ACTUAL_HEAD_DIM_QK, + ACTUAL_HEAD_DIM_V, # head dim + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, # + alibi_slope, + seqlen_q, + seqlen_k, # max sequence length for q and k + start_n, + start_m, + num_steps, # iteration numbers + descale_q, + descale_k, + descale_v, + descale_do, # fp8 descale factors from user + MASK=False, # causal masking + ENABLE_DROPOUT=ENABLE_DROPOUT, # activate dropout + USE_ALIBI=USE_ALIBI, + USE_EXP2=USE_EXP2, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + DEBUG_TRITON=DEBUG_TRITON, + DEBUG_TRITON_DETAIL=DEBUG_TRITON_DETAIL, + ) + + # Write back dV + adj_dv = bid * stride_dvb + hkid * stride_dvh + k_start * stride_dvn + offs_dv = offs_n[:, None] * stride_dvn + offs_d_v[None, :] * stride_dvd + tl.store(DV + adj_dv + offs_dv, dv, mask=mask_v) + # write back dk + adj_dk = bid * stride_dkb + hkid * stride_dkh + k_start * stride_dkn + offs_dk = offs_n[:, None] * stride_dkn + offs_d_qk[None, :] * stride_dkd + dk *= sm_scale + tl.store(DK + adj_dk + offs_dk, dk, mask=mask_k) + + # THIS PART DOES DQ + start_m = pid * BLOCK_M2 + if start_m < seqlen_q: + offs_m = start_m + tl.arange(0, BLOCK_M2) + # Mask for loading K and V + mask_q = offs_m[:, None] < seqlen_q + mask_do = offs_m[:, None] < seqlen_q + if PADDED_HEAD_QK: + mask_d_qk = offs_d_qk < ACTUAL_HEAD_DIM_QK + mask_q &= mask_d_qk[None, :] + if PADDED_HEAD_V: + mask_d_v = offs_d_v < ACTUAL_HEAD_DIM_V + mask_do &= mask_d_v[None, :] + offs_q = offs_m[:, None] * stride_qm + offs_d_qk[None, :] * stride_qd + offs_do = offs_m[:, None] * stride_dom + offs_d_v[None, :] * stride_dod + K += bid * stride_kb + hkid * stride_kh + k_start * stride_kn + V += bid * stride_vb + hkid * stride_vh + k_start * stride_vn + # If MQA / GQA, set the K and V head offsets appropriately. + for hqid in range(hkid * GROUP_SIZE, hkid * GROUP_SIZE + GROUP_SIZE): + # offset input and output tensor by batch and Q/K heads + adj_q = bid * stride_qb + hqid * stride_qh + q_start * stride_qm + adj_do = bid * stride_dob + hqid * stride_doh + q_start * stride_dom + adj_delta = ( + bid * stride_delta_b + hqid * stride_delta_h + q_start * stride_delta_m + ) + Delta_ptr = Delta + adj_delta + adj_m = bid * stride_lse_b + hqid * stride_lse_h + q_start * stride_lse_m + M_ptr = M + adj_m + + if USE_ALIBI: + alibi_offset = bid * stride_az + hqid * stride_ah + alibi_slope = tl.load(Alibi_slopes + alibi_offset) + else: + alibi_slope = None + + # batch_philox_offset is the ACTUALLY dropout offset + # dropout_offset is for debug purpose and will be removed later + batch_philox_offset = 0 + dropout_offset = 0 + if ENABLE_DROPOUT: + batch_philox_offset = ( + philox_offset_base + bid * stride_dropoutb + hqid * stride_dropouth + ) + dropout_offset = ( + Dropout_mask + bid * stride_dropoutb + hqid * stride_dropouth + ) + + q = tl.load(Q + adj_q + offs_q, mask=mask_q, other=0.0) + do = tl.load(DO + adj_do + offs_do, mask=mask_do, other=0.0) + m = tl.load(M + adj_m + offs_m * stride_lse_m, mask=offs_m < seqlen_q) + m = m[:, None] + + if IS_FP8: + descale_q = tl.load(Descale_q + bid * stride_descale_q_z + hqid) + descale_k = tl.load(Descale_k + bid * stride_descale_k_z + hkid) + descale_v = tl.load(Descale_v + bid * stride_descale_v_z + hkid) + descale_do = tl.load(Descale_do + bid * stride_descale_do_z + hqid) + else: + descale_q, descale_k, descale_v, descale_do = 1.0, 1.0, 1.0, 1.0 + + # start can only be 0 at minimum + start_n = 0 + end_n = seqlen_k + num_steps = tl.cdiv(seqlen_k, BLOCK_N2) + + dq = tl.zeros([BLOCK_M2, HEAD_DIM_QK], dtype=tl.float32) + dq = _bwd_dq_inner( + dq, + q, + K, + V, + do, + m, + Delta_ptr, + sm_scale, + stride_qm, + stride_qd, + stride_kn, + stride_kd, + stride_vn, + stride_vd, + stride_dropoutm, + stride_dropoutn, + stride_lse_m, + stride_delta_m, + seqlen_q, + seqlen_k, + BLOCK_M2, + BLOCK_N2, + HEAD_DIM_QK, + HEAD_DIM_V, + ACTUAL_HEAD_DIM_QK, + ACTUAL_HEAD_DIM_V, + dropout_p, + philox_seed, + batch_philox_offset, + dropout_offset, + alibi_slope, + start_m, + start_n, + end_n, + num_steps, + descale_q, + descale_k, + descale_v, + descale_do, + MASK=False, + ENABLE_DROPOUT=ENABLE_DROPOUT, + USE_ALIBI=USE_ALIBI, + USE_EXP2=USE_EXP2, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + DEBUG_TRITON=DEBUG_TRITON, + DEBUG_TRITON_DETAIL=DEBUG_TRITON_DETAIL, + ) + # Write back dQ. + adj_dq = bid * stride_dqb + hqid * stride_dqh + q_start * stride_dqm + offs_dq = offs_m[:, None] * stride_dqm + offs_d_qk[None, :] * stride_dqd + dq *= sm_scale + tl.store(DQ + adj_dq + offs_dq, dq, mask=mask_q) + + +def is_contiguous(x, name): + if x.is_contiguous(): + return x + else: + print(f"{name} is not contiguous") + return x.contiguous() + + +OLD_LSE: bool = False +DEBUG_TRITON: bool = False +DEBUG_TRITON_DETAIL: bool = False + + +def attention_prefill_backward_triton_split_fused_no_atomics_impl( + do: torch.Tensor, + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + o: torch.Tensor, + softmax_lse: torch.Tensor, + dq: torch.Tensor, + dk: torch.Tensor, + dv: torch.Tensor, + sm_scale: float, + alibi_slopes: Optional[torch.Tensor], + causal: bool, + layout: Literal["bshd", "bhsd", "thd"], + cu_seqlens_q: Optional[torch.Tensor], + cu_seqlens_k: Optional[torch.Tensor], + max_seqlen_q: Optional[int], + max_seqlen_k: Optional[int], + dropout_p: float, + philox_seed: Optional[int], + philox_offset: Optional[int], + use_exp2: bool, + # fp8 + descale_q: Optional[torch.Tensor], + descale_k: Optional[torch.Tensor], + descale_v: Optional[torch.Tensor], + descale_o: Optional[torch.Tensor], + descale_do: Optional[torch.Tensor], + descale_dq: Optional[torch.Tensor], + descale_dk: Optional[torch.Tensor], + descale_dv: Optional[torch.Tensor], + # seqused for FA v3 + seqused_q: Optional[torch.Tensor] = None, + seqused_k: Optional[torch.Tensor] = None, +): + # get params, strides and shape + IS_VARLEN = layout == "thd" + use_dropout = dropout_p > 0.0 + + # common assertions + assert ( + 0.0 <= dropout_p <= 1.0 + ), f"dropout_p must be between 0 and 1, got {dropout_p}" + assert ( + q.device == k.device == v.device == o.device == do.device == softmax_lse.device + ), f"All tensors must be on the same device. Got: q={q.device}, k={k.device}, v={v.device}, o={o.device}, do={do.device}, softmax_lse={softmax_lse.device}" + assert ( + q.dtype == k.dtype == v.dtype == do.dtype + ), "q, k, v, do must have the same dtype" + current_device = torch.cuda.current_device() + assert ( + q.is_cuda and q.device.index == current_device + ), f"Device mismatch: Kernel will launch on cuda:{current_device}, but tensors are on {q.device}" + + # get shapes and strides + if IS_VARLEN: + # shape + total_seqlen_q, nheads_q, head_size_q = q.shape + total_seqlen_k, nheads_k, head_size_k = k.shape + total_seqlen_v, nheads_v, head_size_v = v.shape + nheads_lse, total_seqlen_lse = softmax_lse.shape + + # assert shapes + assert ( + total_seqlen_lse == total_seqlen_q + ), f"softmax_lse seqlen {total_seqlen_lse} != q seqlen {total_seqlen_q}" + assert ( + cu_seqlens_q is not None + ), "cu_seqlens_q must be provided for varlen layout" + assert ( + cu_seqlens_k is not None + ), "cu_seqlens_k must be provided for varlen layout" + assert ( + max_seqlen_q is not None + ), "max_seqlen_q must be provided for varlen layout" + assert ( + max_seqlen_k is not None + ), "max_seqlen_k must be provided for varlen layout" + + # assert head dimensions + assert ( + head_size_q == head_size_k + ), f"head sizes must match: q={head_size_q}, k={head_size_k}" + assert ( + nheads_k == nheads_v + ), f"k and v must have same number of heads: k={nheads_k}, v={nheads_v}" + assert ( + nheads_q % nheads_k == 0 + ), f"nheads_q {nheads_q} must be divisible by nheads_k {nheads_k} for GQA/MQA" + assert ( + nheads_lse == nheads_q + ), f"softmax_lse heads {nheads_lse} != q heads {nheads_q}" + + # assert output shapes + assert o.shape == ( + total_seqlen_q, + nheads_q, + head_size_v, + ), f"o shape {o.shape} != expected {(total_seqlen_q, nheads_q, head_size_v)}" + assert do.shape == o.shape, f"do shape {do.shape} != o shape {o.shape}" + assert dq.shape == q.shape, f"dq shape {dq.shape} != q shape {q.shape}" + assert dk.shape == k.shape, f"dk shape {dk.shape} != k shape {k.shape}" + assert dv.shape == v.shape, f"dv shape {dv.shape} != v shape {v.shape}" + + # assert cu_seqlens + assert ( + cu_seqlens_q.dtype == torch.int32 + ), f"cu_seqlens_q must be int32, got {cu_seqlens_q.dtype}" + assert ( + cu_seqlens_k.dtype == torch.int32 + ), f"cu_seqlens_k must be int32, got {cu_seqlens_k.dtype}" + assert cu_seqlens_q[0] == 0, "cu_seqlens_q must start with 0" + assert cu_seqlens_k[0] == 0, "cu_seqlens_k must start with 0" + assert ( + cu_seqlens_q[-1] == total_seqlen_q + ), f"cu_seqlens_q[-1] {cu_seqlens_q[-1]} != total_seqlen_q {total_seqlen_q}" + assert ( + cu_seqlens_k[-1] == total_seqlen_k + ), f"cu_seqlens_k[-1] {cu_seqlens_k[-1]} != total_seqlen_k {total_seqlen_k}" + + # set vars + batch = len(cu_seqlens_q) - 1 + head_size_qk = head_size_q + + # strides + stride_qb, stride_qm, stride_qh, stride_qd = ( + 0, + q.stride(0), + q.stride(1), + q.stride(2), + ) + stride_kb, stride_kn, stride_kh, stride_kd = ( + 0, + k.stride(0), + k.stride(1), + k.stride(2), + ) + stride_vb, stride_vn, stride_vh, stride_vd = ( + 0, + v.stride(0), + v.stride(1), + v.stride(2), + ) + stride_ob, stride_om, stride_oh, stride_od = ( + 0, + o.stride(0), + o.stride(1), + o.stride(2), + ) + stride_dqb, stride_dqm, stride_dqh, stride_dqd = ( + 0, + dq.stride(0), + dq.stride(1), + dq.stride(2), + ) + stride_dkb, stride_dkn, stride_dkh, stride_dkd = ( + 0, + dk.stride(0), + dk.stride(1), + dk.stride(2), + ) + stride_dvb, stride_dvn, stride_dvh, stride_dvd = ( + 0, + dv.stride(0), + dv.stride(1), + dv.stride(2), + ) + stride_dob, stride_dom, stride_doh, stride_dod = ( + 0, + do.stride(0), + do.stride(1), + do.stride(2), + ) + stride_lse_b, stride_lse_h, stride_lse_m = ( + 0, + softmax_lse.stride(0), + softmax_lse.stride(1), + ) + else: + # shapes + batch_q, seqlen_q, nheads_q, head_size_q = q.shape + batch_k, seqlen_k, nheads_k, head_size_k = k.shape + batch_v, seqlen_v, nheads_v, head_size_v = v.shape + batch_lse, nheads_lse, seqlen_lse = softmax_lse.shape + + # assert batch dimensions + assert ( + batch_q == batch_k == batch_v + ), f"batch sizes must match: q={batch_q}, k={batch_k}, v={batch_v}" + + # assert head dimensions + assert ( + head_size_q == head_size_k + ), f"head sizes must match: q={head_size_q}, k={head_size_k}" + assert ( + nheads_k == nheads_v + ), f"k and v must have same number of heads: k={nheads_k}, v={nheads_v}" + assert ( + nheads_q % nheads_k == 0 + ), f"nheads_q {nheads_q} must be divisible by nheads_k {nheads_k} for GQA/MQA" + + # assert sequence lengths + assert ( + seqlen_k == seqlen_v + ), f"k and v sequence lengths must match: k={seqlen_k}, v={seqlen_v}" + + # assert output shapes + assert o.shape == ( + batch_q, + seqlen_q, + nheads_q, + head_size_v, + ), f"o shape {o.shape} != expected" + assert do.shape == o.shape, f"do shape {do.shape} != o shape {o.shape}" + assert dq.shape == q.shape, f"dq shape {dq.shape} != q shape {q.shape}" + assert dk.shape == k.shape, f"dk shape {dk.shape} != k shape {k.shape}" + assert dv.shape == v.shape, f"dv shape {dv.shape} != v shape {v.shape}" + + # assert softmax_lse shape + assert softmax_lse.shape == ( + batch_q, + nheads_q, + seqlen_q, + ), f"softmax_lse shape {softmax_lse.shape} != expected" + + # set vars + batch = batch_q + head_size_qk = head_size_q + max_seqlen_q = seqlen_q + max_seqlen_k = seqlen_k + + # strides + stride_qb, stride_qm, stride_qh, stride_qd = q.stride() + stride_kb, stride_kn, stride_kh, stride_kd = k.stride() + stride_vb, stride_vn, stride_vh, stride_vd = v.stride() + stride_ob, stride_om, stride_oh, stride_od = o.stride() + stride_dqb, stride_dqm, stride_dqh, stride_dqd = dq.stride() + stride_dkb, stride_dkn, stride_dkh, stride_dkd = dk.stride() + stride_dvb, stride_dvn, stride_dvh, stride_dvd = dv.stride() + stride_dob, stride_dom, stride_doh, stride_dod = do.stride() + stride_lse_b, stride_lse_h, stride_lse_m = softmax_lse.stride() + + # fp8 setup - moved after all assertions + IS_FP8 = is_fp8(q) + if IS_FP8: + FP8_MAX = torch.finfo(q.dtype).max + # we already asserted that do, q, k, v all have the same dtype, so no need to check each one + if is_fp8(o): + FP8_OUTPUT = True + assert ( + descale_o is not None + ), f"descale_o is None. In fp8, you need to pass a tensor for descale_o along with a tensor o." + assert ( + descale_dq is not None + ), f"descale_dq is None. In fp8, you need to pass a tensor for descale_dq along with a tensor dq." + assert ( + descale_dk is not None + ), f"descale_dk is None. In fp8, you need to pass a tensor for descale_dk along with a tensor dk." + assert ( + descale_dv is not None + ), f"descale_dv is None. In fp8, you need to pass a tensor for descale_dv along with a tensor dv." + else: + FP8_OUTPUT = False + + stride_descale_q_z = descale_q.stride(0) if descale_q is not None else None + stride_descale_k_z = descale_k.stride(0) if descale_k is not None else None + stride_descale_v_z = descale_v.stride(0) if descale_v is not None else None + stride_descale_o_z = descale_o.stride(0) if descale_o is not None else None + stride_descale_do_z = descale_do.stride(0) if descale_do is not None else None + + if DEBUG: + print( + f"FP8 path triggered in bwd_prefill_fused_no_atomics.py (FP8_OUTPUT={FP8_OUTPUT})" + ) + else: + FP8_MAX = None + FP8_OUTPUT = False + stride_descale_q_z = stride_descale_k_z = stride_descale_v_z = ( + stride_descale_o_z + ) = stride_descale_do_z = None + + # alibi setup + use_alibi, (stride_az, stride_ah) = ( + (True, alibi_slopes.stride()) if alibi_slopes is not None else (False, (0, 0)) + ) + + # get closest power of 2 over or equal to 32. + padded_d_model_qk = 1 << (head_size_qk - 1).bit_length() + padded_d_model_qk = max(padded_d_model_qk, 32) + padded_d_model_v = 1 << (head_size_v - 1).bit_length() + padded_d_model_v = max(padded_d_model_v, 32) + HEAD_DIM_QK = padded_d_model_qk + HEAD_DIM_V = padded_d_model_v + ACTUAL_HEAD_DIM_QK = head_size_qk + ACTUAL_HEAD_DIM_V = head_size_v + + # init delta + if OLD_LSE: + delta = torch.empty_like(softmax_lse) + if IS_VARLEN: + stride_delta_b, stride_delta_h, stride_delta_m = ( + 0, + delta.stride(0), + delta.stride(1), + ) + else: + stride_delta_b, stride_delta_h, stride_delta_m = delta.stride() + else: + if IS_VARLEN: + # interface expects the varlen sequence dims to rounded like this. Not sure why. + total_q, num_heads, _ = q.shape + total_q_rounded = total_q + 128 * batch + delta_padded = torch.zeros( + (nheads_q, total_q_rounded), device=q.device, dtype=torch.float32 + ) + delta = delta_padded[:, :total_q] + stride_delta_b, stride_delta_h, stride_delta_m = ( + 0, + delta.stride(0), + delta.stride(1), + ) + else: + # the interface expects the sequence dimension to be rounded to 128 + max_seqlen_q_rounded = round_multiple(max_seqlen_q, 128) + delta_padded = torch.zeros( + (batch, nheads_q, max_seqlen_q_rounded), + device=q.device, + dtype=torch.float32, + ) + delta = delta_padded[:, :, :max_seqlen_q] + stride_delta_b, stride_delta_h, stride_delta_m = delta.stride() + + pre_grid = lambda META: ( + triton.cdiv(max_seqlen_q, META["PRE_BLOCK"]), + batch, + nheads_q, + ) + _bwd_preprocess[pre_grid]( + o, + do, + delta, + stride_ob, + stride_oh, + stride_om, + stride_od, + stride_dob, + stride_doh, + stride_dom, + stride_dod, + stride_delta_b, + stride_delta_h, + stride_delta_m, + stride_descale_do_z, + cu_seqlens_q, + max_seqlen_q, + descale_do, + HEAD_DIM_V=HEAD_DIM_V, + ACTUAL_HEAD_DIM_V=ACTUAL_HEAD_DIM_V, + IS_VARLEN=IS_VARLEN, + IS_FP8=IS_FP8, + ) + + if False: + print("delta:", delta, delta.shape) + + # dropout mask tensor for debugging. We dump the dropout mask created in + # the kernel for testing + dropout_mask = None + stride_dropoutb, stride_dropouth, stride_dropoutm, stride_dropoutn = (0, 0, 0, 0) + if use_dropout: + dropout_mask = torch.zeros( + (batch, nheads_q, max_seqlen_q, max_seqlen_k), + device=q.device, + dtype=torch.float32, + ) + + if DROPOUT_USE_PYTORCH: + if not IS_VARLEN: + dropout_mask = create_dropout_mask( + dropout_p, + (batch, nheads_q, max_seqlen_q, max_seqlen_k), + seed=philox_seed, + ) + else: + dropout_mask = create_dropout_mask_varlen( + dropout_p, batch, nheads_q, cu_seqlens_q, cu_seqlens_k, philox_seed + ) + stride_dropoutb, stride_dropouth, stride_dropoutm, stride_dropoutn = ( + dropout_mask.stride() + ) + + seqlen = max(max_seqlen_q, max_seqlen_k) + grid = lambda META: ( + nheads_k, + (seqlen + META["BLOCK_N1"] - 1) // META["BLOCK_N1"], + batch, + ) + if causal: + if DEBUG_TRITON: + print(f"bwd_kernel: grid = {grid}") # noqa: E701 + bwd_kernel_causal[grid]( + q, + k, + v, + sm_scale, + do, + dq, + dk, + dv, + softmax_lse, + delta, + stride_qb, + stride_qh, + stride_qm, + stride_qd, + stride_kb, + stride_kh, + stride_kn, + stride_kd, + stride_vb, + stride_vh, + stride_vn, + stride_vd, + stride_dqb, + stride_dqh, + stride_dqm, + stride_dqd, + stride_dkb, + stride_dkh, + stride_dkn, + stride_dkd, + stride_dvb, + stride_dvh, + stride_dvn, + stride_dvd, + stride_lse_b, + stride_lse_h, + stride_lse_m, + stride_delta_b, + stride_delta_h, + stride_delta_m, + stride_dob, + stride_doh, + stride_dom, + stride_dod, + stride_dropoutb, + stride_dropouth, + stride_dropoutm, + stride_dropoutn, + stride_descale_q_z, + stride_descale_k_z, + stride_descale_v_z, + stride_descale_do_z, + stride_az, + stride_ah, + nheads_q, + nheads_k, + cu_seqlens_q, + cu_seqlens_k, + seqused_q, + seqused_k, # Pass seqused tensors + max_seqlen_q, + max_seqlen_k, + dropout_mask, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes, + descale_q, + descale_k, + descale_v, + descale_do, + HEAD_DIM_QK=HEAD_DIM_QK, + HEAD_DIM_V=HEAD_DIM_V, + ACTUAL_HEAD_DIM_QK=ACTUAL_HEAD_DIM_QK, + ACTUAL_HEAD_DIM_V=ACTUAL_HEAD_DIM_V, + ENABLE_DROPOUT=use_dropout, + IS_VARLEN=IS_VARLEN, + USE_ALIBI=use_alibi, + USE_EXP2=use_exp2, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + FP8_OUTPUT=FP8_OUTPUT, + USE_SEQUSED=( + seqused_q is not None or seqused_k is not None + ), # Add flag for seqused + DEBUG_TRITON=DEBUG_TRITON, + DEBUG_TRITON_DETAIL=DEBUG_TRITON_DETAIL, + ) + else: + bwd_kernel_noncausal[grid]( + q, + k, + v, + sm_scale, + do, + dq, + dk, + dv, + softmax_lse, + delta, + stride_qb, + stride_qh, + stride_qm, + stride_qd, + stride_kb, + stride_kh, + stride_kn, + stride_kd, + stride_vb, + stride_vh, + stride_vn, + stride_vd, + stride_dqb, + stride_dqh, + stride_dqm, + stride_dqd, + stride_dkb, + stride_dkh, + stride_dkn, + stride_dkd, + stride_dvb, + stride_dvh, + stride_dvn, + stride_dvd, + stride_lse_b, + stride_lse_h, + stride_lse_m, + stride_delta_b, + stride_delta_h, + stride_delta_m, + stride_dob, + stride_doh, + stride_dom, + stride_dod, + stride_dropoutb, + stride_dropouth, + stride_dropoutm, + stride_dropoutn, + stride_descale_q_z, + stride_descale_k_z, + stride_descale_v_z, + stride_descale_do_z, + stride_az, + stride_ah, + nheads_q, + nheads_k, + cu_seqlens_q, + cu_seqlens_k, + seqused_q, + seqused_k, # Pass seqused tensors + max_seqlen_q, + max_seqlen_k, + dropout_mask, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes, + descale_q, + descale_k, + descale_v, + descale_do, + HEAD_DIM_QK=HEAD_DIM_QK, + HEAD_DIM_V=HEAD_DIM_V, + ACTUAL_HEAD_DIM_QK=ACTUAL_HEAD_DIM_QK, + ACTUAL_HEAD_DIM_V=ACTUAL_HEAD_DIM_V, + ENABLE_DROPOUT=use_dropout, + IS_VARLEN=IS_VARLEN, + USE_ALIBI=use_alibi, + USE_EXP2=use_exp2, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + FP8_OUTPUT=FP8_OUTPUT, + USE_SEQUSED=( + seqused_q is not None or seqused_k is not None + ), # Add flag for seqused + DEBUG_TRITON=DEBUG_TRITON, + DEBUG_TRITON_DETAIL=DEBUG_TRITON_DETAIL, + ) + + if OLD_LSE: + return delta + else: + return delta_padded + + +def attention_prefill_backward_triton_impl( + *, + do: torch.Tensor, + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + o: torch.Tensor, + softmax_lse: torch.Tensor, + dq: torch.Tensor, + dk: torch.Tensor, + dv: torch.Tensor, + sm_scale: float, + alibi_slopes: Optional[torch.Tensor], + causal: bool, + layout: str, + cu_seqlens_q: Optional[torch.Tensor], + cu_seqlens_k: Optional[torch.Tensor], + max_seqlen_q: Optional[int], + max_seqlen_k: Optional[int], + seqused_q: Optional[torch.Tensor] = None, + seqused_k: Optional[torch.Tensor] = None, + dropout_p: float = 0.0, + philox_seed: Optional[int] = None, + philox_offset: Optional[int] = None, + descale_q: Optional[torch.Tensor] = None, + descale_k: Optional[torch.Tensor] = None, + descale_v: Optional[torch.Tensor] = None, + descale_o: Optional[torch.Tensor] = None, + descale_do: Optional[torch.Tensor] = None, + descale_dq: Optional[torch.Tensor] = None, + descale_dk: Optional[torch.Tensor] = None, + descale_dv: Optional[torch.Tensor] = None, + use_exp2: bool = True, + mode: str = "fused_no_atomics", +) -> torch.Tensor: + """Unified backward interface dispatching to atomics or no-atomics implementation. + + Parameters mirror the superset of the two legacy interfaces. The public API should + call ONLY this function going forward. + mode: 'fused_atomics' or 'fused_no_atomics'; layout: 'bshd' or 'thd'; use_exp2 retained for parity. + """ + if mode == "fused_atomics": + # Atomics path ignores layout & use_exp2; pass varlen metadata directly. + return attention_prefill_backward_triton_fused_atomics_impl( + do, + q, + k, + v, + o, + softmax_lse, + dq, + dk, + dv, + sm_scale, + alibi_slopes, + causal, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q if max_seqlen_q is not None else q.shape[1], + max_seqlen_k if max_seqlen_k is not None else k.shape[1], + dropout_p, + philox_seed or 0, + philox_offset or 0, + descale_q, + descale_k, + descale_v, + descale_o, + True, # fused flag + None, + None, + ) + elif mode == "fused_no_atomics": + return attention_prefill_backward_triton_split_fused_no_atomics_impl( + do, + q, + k, + v, + o, + softmax_lse, + dq, + dk, + dv, + sm_scale, + alibi_slopes, + causal, + layout, # layout required here + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_p, + philox_seed, + philox_offset, + use_exp2, + descale_q, + descale_k, + descale_v, + descale_o, + descale_do, + descale_dq, + descale_dk, + descale_dv, + seqused_q, + seqused_k, + ) + else: + raise ValueError( + f"Unknown backward mode '{mode}'. Expected 'fused_atomics' or 'fused_no_atomics'." + ) diff --git a/aiter/ops/triton/flash_attn_triton_amd/bwd_ref.py b/aiter/ops/triton/flash_attn_triton_amd/bwd_ref.py new file mode 100644 index 0000000000..3e154a1c58 --- /dev/null +++ b/aiter/ops/triton/flash_attn_triton_amd/bwd_ref.py @@ -0,0 +1,594 @@ +import torch +import math +from typing import Literal, Optional +from .utils import compute_alibi_tensor_ref + +DEBUG = False +DEBUG_CORE = False + + +def attention_backward_core_ref_impl( + do, + q, + k, + v, + o, + softmax_lse, + sm_scale, + causal, + window_size_left, + window_size_right, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes, + use_exp2, +): + if DEBUG_CORE: + print() + print("attention_backward_core_ref_impl") + print("do:", do, do.shape) + print("q:", q, q.shape) + print("k:", k, k.shape) + print("v:", v, v.shape) + print("o:", o, o.shape) + print("softmax_lse:", softmax_lse, softmax_lse.shape) + print("sm_scale:", sm_scale) + print("causal:", causal) + print("window_size_left:", window_size_left) + print("window_size_right:", window_size_right) + print("dropout_p:", dropout_p) + print("philox_seed:", philox_seed) + print("philox_offset:", philox_offset) + print("use_exp2:", use_exp2) + + # cast to float32 + do = do.to(torch.float32) + q = q.to(torch.float32) + k = k.to(torch.float32) + v = v.to(torch.float32) + o = o.to(torch.float32) + softmax_lse = softmax_lse.to(torch.float32) + + # recompute attention_scores. Make sure it matches the forward impl. i.e. It use float32 + attention_scores = torch.matmul(q, k.transpose(-2, -1)) + if DEBUG_CORE: + print("attention_scores:", attention_scores, attention_scores.shape) + + # scale scores + attention_scaled_scores = sm_scale * attention_scores + if DEBUG_CORE: + print( + "attention_scaled_scores:", + attention_scaled_scores, + attention_scaled_scores.shape, + ) + + if alibi_slopes is not None: + L_q, L_k = q.shape[1], k.shape[1] + if DEBUG_CORE: + print("alibi_slopes:", alibi_slopes, alibi_slopes.shape) + alibi_bias = compute_alibi_tensor_ref(alibi_slopes, L_q, L_k) + alibi_bias = alibi_bias.reshape(-1, L_q, L_k) + if DEBUG_CORE: + print("alibi_bias:", alibi_bias, alibi_bias.shape) + attention_scaled_scores = attention_scaled_scores + alibi_bias + if DEBUG_CORE: + print( + "attention_scaled_scores after alibi:", + attention_scaled_scores, + attention_scaled_scores.shape, + ) + + # Apply masks + L_q, L_k = q.shape[1], k.shape[1] + row_idx = torch.arange(L_q, device=q.device).unsqueeze(1) + col_idx = torch.arange(L_k, device=q.device).unsqueeze(0) + col_offset = L_k - L_q + + mask_applied = False + if causal and (window_size_left, window_size_right) == (-1, -1): + # Pure causal: ensure query doesn't attend to future keys + mask = row_idx >= (col_idx - col_offset) + mask_applied = True + if DEBUG_CORE: + print("causal_mask:", mask) + elif (window_size_left, window_size_right) != (-1, -1): + # Handle the case where window sizes exceed sequence length + if window_size_left >= L_k: + window_size_left = -1 # No left limit + if window_size_right >= L_k: + window_size_right = -1 # No right limit + + if causal: + # Causal + sliding window: ensure we don't attend to future + window_size_right = ( + min(window_size_right, 0) if window_size_right != -1 else 0 + ) + + # Create sliding window mask + # Each query at position i attends to keys in [i + offset - left, i + offset + right] + if window_size_left == -1 and window_size_right == -1: + # No window restriction + mask = torch.ones((L_q, L_k), dtype=torch.bool, device=q.device) + else: + mask = torch.ones((L_q, L_k), dtype=torch.bool, device=q.device) + if window_size_left != -1: + # Each query at position i attends to keys from position (i - left) accounting for offset + mask = mask & (col_idx >= (row_idx + col_offset - window_size_left)) + if window_size_right != -1: + # Each query at position i attends to keys up to position (i + right) accounting for offset + mask = mask & (col_idx <= (row_idx + col_offset + window_size_right)) + + # Apply causal constraint + if causal: + causal_mask = row_idx >= (col_idx - col_offset) + mask = mask & causal_mask + + mask_applied = True + if DEBUG_CORE: + print( + f"sliding_window_mask (left={window_size_left}, right={window_size_right}):", + mask, + ) + + # Apply the mask if created + if mask_applied: + attention_scaled_scores = attention_scaled_scores.masked_fill( + torch.logical_not(mask.unsqueeze(0)), float("-inf") + ) + if DEBUG_CORE: + print( + "attention_scaled_scores after masking:", + attention_scaled_scores, + attention_scaled_scores.shape, + ) + + # compute probabilities using softmax_lse + if use_exp2: + RCP_LN = 1 / math.log(2) + attention_scaled_scores_base2 = attention_scaled_scores * RCP_LN + softmax_lse_base2 = softmax_lse * RCP_LN + softmax_lse_3d = softmax_lse_base2.unsqueeze(-1) + p = torch.exp2(attention_scaled_scores_base2 - softmax_lse_3d) + else: + softmax_lse_3d = softmax_lse.unsqueeze(-1) + p = torch.exp(attention_scaled_scores - softmax_lse_3d) + + # Zero out positions outside the mask + if mask_applied: + p = p.masked_fill(torch.logical_not(mask.unsqueeze(0)), 0.0) + + if DEBUG_CORE: + print("softmax_lse_3d:", softmax_lse_3d, softmax_lse_3d.shape) + print("p:", p, p.shape) + + if dropout_p > 0.0: + rand_vals = torch.rand( + p.shape, + generator=torch.Generator(device=p.device).manual_seed(philox_seed), + device=p.device, + dtype=p.dtype, + ) + dropout_mask, dropout_scale = rand_vals > dropout_p, (1.0 / (1 - dropout_p)) + if DEBUG_CORE: + print("dropout_scale:", dropout_scale) + print("dropout_mask:", dropout_mask) + + p_drop = torch.where(dropout_mask, p, torch.zeros_like(p)) + p_drop_scaled = p_drop * dropout_scale + if DEBUG_CORE: + print("dropout_scale:", dropout_scale) + print("p_drop:", p_drop, p_drop.shape) + print("p_drop_scaled:", p_drop_scaled, p_drop_scaled.shape) + + # compute dv + dv = torch.matmul(p_drop_scaled.transpose(-2, -1), do) + if DEBUG_CORE: + print("dv:", dv, dv.shape) + + # compute dp + dp_dropout = torch.matmul(do, v.transpose(-2, -1)) + dp = ( + torch.where(dropout_mask, dp_dropout, torch.zeros_like(dp_dropout)) + * dropout_scale + ) + if DEBUG_CORE: + print("dp_dropout:", dp_dropout, dp_dropout.shape) + print("dp:", dp, dp.shape) + else: + # compute dv + dv = torch.matmul(p.transpose(-2, -1), do) + if DEBUG_CORE: + print("dv:", dv, dv.shape) + + # compute dp + dp = torch.matmul(do, v.transpose(-2, -1)) + if DEBUG_CORE: + print("dp:", dp, dp.shape) + + # calculate ds + if True: + delta = torch.sum(o * do, axis=-1).unsqueeze(-1) + else: + delta = torch.sum(p * dp, axis=-1).unsqueeze(-1) + if DEBUG_CORE: + print("delta:", delta, delta.shape) + dscores_scaled = p * (dp - delta) + + # Zero out gradients for positions outside the mask + if mask_applied: + dscores_scaled = dscores_scaled.masked_fill( + torch.logical_not(mask.unsqueeze(0)), 0.0 + ) + + ds = dscores_scaled * sm_scale + if DEBUG_CORE: + print("dscores_scaled:", dscores_scaled, dscores_scaled.shape) + print("ds:", ds, ds.shape) + + # compute gradient wrt k & q + dk = torch.matmul(ds.transpose(-2, -1), q) + dq = torch.matmul(ds, k) + if DEBUG_CORE: + print("dk:", dk, dk.shape) + print("dq:", dq, dq.shape) + + # cast back to original dtype + dq = dq.to(torch.float16) + dk = dk.to(torch.float16) + dv = dv.to(torch.float16) + # remove d dim with size 1 + delta = delta.squeeze(-1) + + if DEBUG_CORE: + print("attention_backward_core_ref_impl output") + print("delta:", delta, delta.shape) + print("dv:", dv, dv.shape) + print("dk:", dk, dk.shape) + print("dq:", dq, dq.shape) + + return dq, dk, dv, delta + + +def attention_varlen_backward_pytorch_ref_impl( + do, + q, + k, + v, + o, + softmax_lse, + sm_scale, + causal, + window_size_left, + window_size_right, + layout, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes, + use_exp2, +): + # Ensure the layout is 'thd' + if layout != "thd": + raise ValueError(f"Unsupported layout {layout}. Expected 'thd'.") + + batch_size = cu_seqlens_q.shape[0] - 1 + nheads_q, head_dim = q.shape[1], q.shape[2] + nheads_k = k.shape[1] + + group_size = nheads_q // nheads_k + if nheads_q % nheads_k != 0: + raise ValueError("nheads_q must be divisible by nheads_k") + + # Pre-allocate outputs + total_L_q = q.shape[0] + total_L_k = k.shape[0] + + dq = torch.zeros_like(q) + dk = torch.zeros_like(k) + dv = torch.zeros_like(v) + # delta has the same shape as softmax_lse + delta = torch.zeros_like(softmax_lse) + + for i in range(batch_size): + # Get the start and end indices for the current sequence + start_q = cu_seqlens_q[i].item() + end_q = cu_seqlens_q[i + 1].item() + start_k = cu_seqlens_k[i].item() + end_k = cu_seqlens_k[i + 1].item() + + # Extract q_i, k_i, v_i, do_i, o_i, softmax_lse_i + q_i = q[start_q:end_q, :, :] # [L_q_i, nheads_q, head_dim] + k_i = k[start_k:end_k, :, :] # [L_k_i, nheads_k, head_dim] + v_i = v[start_k:end_k, :, :] # [L_k_i, nheads_k, head_dim] + do_i = do[start_q:end_q, :, :] # [L_q_i, nheads_q, head_dim] + o_i = o[start_q:end_q, :, :] # [L_q_i, nheads_q, head_dim] + softmax_lse_i = softmax_lse[:, start_q:end_q] # [nheads_q, L_q_i] + + if group_size != 1: + # MQA or GQA case + # Reshape tensors to include group dimension + q_i = q_i.view(q_i.shape[0], nheads_k, group_size, head_dim) + do_i = do_i.view(do_i.shape[0], nheads_k, group_size, head_dim) + o_i = o_i.view(o_i.shape[0], nheads_k, group_size, head_dim) + softmax_lse_i = softmax_lse_i.view( + nheads_k, group_size, softmax_lse_i.shape[1] + ) + # Expand k_i and v_i to match group_size + k_i = k_i.unsqueeze(2).expand(-1, -1, group_size, -1) + v_i = v_i.unsqueeze(2).expand(-1, -1, group_size, -1) + # Flatten the nheads_k and group_size dimensions + q_i = q_i.reshape(q_i.shape[0], nheads_k * group_size, head_dim) + do_i = do_i.reshape(do_i.shape[0], nheads_k * group_size, head_dim) + o_i = o_i.reshape(o_i.shape[0], nheads_k * group_size, head_dim) + softmax_lse_i = softmax_lse_i.reshape( + nheads_k * group_size, softmax_lse_i.shape[2] + ) + k_i = k_i.reshape(k_i.shape[0], nheads_k * group_size, head_dim) + v_i = v_i.reshape(v_i.shape[0], nheads_k * group_size, head_dim) + + # Permute to [nheads_total, L, head_dim] + q_i = q_i.permute(1, 0, 2) + k_i = k_i.permute(1, 0, 2) + v_i = v_i.permute(1, 0, 2) + do_i = do_i.permute(1, 0, 2) + o_i = o_i.permute(1, 0, 2) + + if alibi_slopes is not None: + alibi_slopes_i = alibi_slopes[i] + else: + alibi_slopes_i = None + + # Call the core backward function for this sequence + dq_i, dk_i, dv_i, delta_i = attention_backward_core_ref_impl( + do_i, + q_i, + k_i, + v_i, + o_i, + softmax_lse_i, + sm_scale, + causal, + window_size_left, + window_size_right, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes_i, + use_exp2, + ) + + # Convert back to 'thd' layout + dq_i = dq_i.permute(1, 0, 2) # [L_q_i, nheads_total, head_dim] + dk_i = dk_i.permute(1, 0, 2) # [L_k_i, nheads_total, head_dim] + dv_i = dv_i.permute(1, 0, 2) # [L_k_i, nheads_total, head_dim] + + if group_size != 1: + # Reshape dq_i and delta_i back to original shape + dq_i = dq_i.view(dq_i.shape[0], nheads_k, group_size, head_dim) + L_q_i = delta_i.shape[1] + delta_i = delta_i.view(nheads_k, group_size, L_q_i) + # Sum dk_i and dv_i over group dimension + dk_i = dk_i.view(dk_i.shape[0], nheads_k, group_size, head_dim) + dv_i = dv_i.view(dv_i.shape[0], nheads_k, group_size, head_dim) + dk_i = dk_i.sum(dim=2) + dv_i = dv_i.sum(dim=2) + # Reshape dq_i back to [L_q_i, nheads_q, head_dim] + dq_i = dq_i.reshape(dq_i.shape[0], nheads_q, head_dim) + delta_i = delta_i.reshape(nheads_q, L_q_i) + else: + # No need to reshape + pass + + # Place outputs in pre-allocated tensors + dq[start_q:end_q, :, :] = dq_i + dk[start_k:end_k, :, :] += dk_i # Accumulate gradients for shared keys + dv[start_k:end_k, :, :] += dv_i # Accumulate gradients for shared values + delta[:, start_q:end_q] = delta_i + + return dq, dk, dv, delta + + +def attention_vanilla_backward_pytorch_ref_impl( + do, + q, + k, + v, + o, + softmax_lse, + sm_scale, + causal, + window_size_left, + window_size_right, + layout, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes, + use_exp2, +): + if layout == "bshd": + if DEBUG: + print() + print("Changing layout to bhsd!") + do = do.transpose(1, 2).contiguous() + q = q.transpose(1, 2).contiguous() + k = k.transpose(1, 2).contiguous() + v = v.transpose(1, 2).contiguous() + o = o.transpose(1, 2).contiguous() + elif layout == "bhsd": + pass + else: + raise ValueError(f"Unknown layout {layout}") + + # Prepare tensors + batch_size, nheads_q, seq_len_q, head_dim = q.shape + batch_size, nheads_k, seq_len_k, head_dim = k.shape + + group_size = nheads_q // nheads_k + if nheads_q % nheads_k != 0: + raise ValueError("nheads_q must be divisible by nheads_k") + + if group_size != 1: + # MQA or GQA case + # Reshape do, q, o to [batch_size, nheads_k, group_size, seq_len_q, head_dim] + do = do.reshape(batch_size, nheads_k, group_size, seq_len_q, head_dim) + q = q.reshape(batch_size, nheads_k, group_size, seq_len_q, head_dim) + o = o.reshape(batch_size, nheads_k, group_size, seq_len_q, head_dim) + # Reshape softmax_lse to [batch_size, nheads_k, group_size, seq_len_q] + softmax_lse = softmax_lse.reshape(batch_size, nheads_k, group_size, seq_len_q) + # Expand k and v to match group_size + k = k.unsqueeze(2).expand( + -1, -1, group_size, -1, -1 + ) # [batch_size, nheads_k, group_size, seq_len_k, head_dim] + v = v.unsqueeze(2).expand(-1, -1, group_size, -1, -1) + # Flatten the first three dimensions for computation + do = do.reshape(batch_size * nheads_k * group_size, seq_len_q, head_dim) + q = q.reshape(batch_size * nheads_k * group_size, seq_len_q, head_dim) + k = k.reshape(batch_size * nheads_k * group_size, seq_len_k, head_dim) + v = v.reshape(batch_size * nheads_k * group_size, seq_len_k, head_dim) + o = o.reshape(batch_size * nheads_k * group_size, seq_len_q, head_dim) + softmax_lse = softmax_lse.reshape(batch_size * nheads_k * group_size, seq_len_q) + else: + # Standard case + do = do.reshape(batch_size * nheads_q, seq_len_q, head_dim) + q = q.reshape(batch_size * nheads_q, seq_len_q, head_dim) + k = k.reshape(batch_size * nheads_k, seq_len_k, head_dim) + v = v.reshape(batch_size * nheads_k, seq_len_k, head_dim) + o = o.reshape(batch_size * nheads_q, seq_len_q, head_dim) + softmax_lse = softmax_lse.reshape(batch_size * nheads_q, seq_len_q) + + # Call the core backward function + dq, dk, dv, delta = attention_backward_core_ref_impl( + do, + q, + k, + v, + o, + softmax_lse, + sm_scale, + causal, + window_size_left, + window_size_right, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes, + use_exp2, + ) + + if group_size != 1: + # Reshape dq back to [batch_size, nheads_k, group_size, seq_len_q, head_dim] + dq = dq.reshape(batch_size, nheads_k, group_size, seq_len_q, head_dim) + # Reshape delta back to [batch_size, nheads_k, group_size, seq_len_q] + delta = delta.reshape(batch_size, nheads_k, group_size, seq_len_q) + # Sum dk and dv over group_size dimension, since k and v are shared across groups + dk = dk.reshape(batch_size, nheads_k, group_size, seq_len_k, head_dim) + dk = dk.sum(dim=2) # Sum over group_size dimension + dv = dv.reshape(batch_size, nheads_k, group_size, seq_len_k, head_dim) + dv = dv.sum(dim=2) + # Reshape dq to [batch_size, nheads_q, seq_len_q, head_dim] + dq = dq.reshape(batch_size, nheads_k * group_size, seq_len_q, head_dim) + delta = delta.reshape(batch_size, nheads_k * group_size, seq_len_q) + else: + # Standard case + dq = dq.reshape(batch_size, nheads_q, seq_len_q, head_dim) + dk = dk.reshape(batch_size, nheads_k, seq_len_k, head_dim) + dv = dv.reshape(batch_size, nheads_k, seq_len_k, head_dim) + delta = delta.reshape(batch_size, nheads_q, seq_len_q) + + # Go back to original layout + if layout == "bshd": + if DEBUG: + print() + print("Changing back to bshd!") + dq = dq.transpose(1, 2) + dk = dk.transpose(1, 2) + dv = dv.transpose(1, 2) + elif layout == "bhsd": + pass + else: + raise ValueError(f"Unknown layout {layout}") + + return dq, dk, dv, delta + + +def attention_backward_pytorch_ref_impl( + do: torch.Tensor, + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + o: torch.Tensor, + softmax_lse: torch.Tensor, + dq: torch.Tensor, + dk: torch.Tensor, + dv: torch.Tensor, + sm_scale: float, + alibi_slopes: Optional[torch.Tensor], + causal: bool, + window_size_left: int, + window_size_right: int, + layout: Literal["bshd", "bhsd", "thd"], + cu_seqlens_q: Optional[torch.Tensor], + cu_seqlens_k: Optional[torch.Tensor], + max_seqlen_q: Optional[int], + max_seqlen_k: Optional[int], + dropout_p: float, + philox_seed: Optional[int], + philox_offset: Optional[int], + use_exp2: bool, +): + if layout == "thd": + dq_ref, dk_ref, dv_ref, delta = attention_varlen_backward_pytorch_ref_impl( + do, + q, + k, + v, + o, + softmax_lse, + sm_scale, + causal, + window_size_left, + window_size_right, + layout, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes, + use_exp2, + ) + else: + dq_ref, dk_ref, dv_ref, delta = attention_vanilla_backward_pytorch_ref_impl( + do, + q, + k, + v, + o, + softmax_lse, + sm_scale, + causal, + window_size_left, + window_size_right, + layout, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes, + use_exp2, + ) + + # copy into output tensor + dv.copy_(dv_ref.to(dv.dtype)) + dk.copy_(dk_ref.to(dk.dtype)) + dq.copy_(dq_ref.to(dq.dtype)) + + return delta diff --git a/aiter/ops/triton/flash_attn_triton_amd/fwd_decode.py b/aiter/ops/triton/flash_attn_triton_amd/fwd_decode.py new file mode 100755 index 0000000000..6dd8cc0297 --- /dev/null +++ b/aiter/ops/triton/flash_attn_triton_amd/fwd_decode.py @@ -0,0 +1,1360 @@ +import torch +import triton +import triton.language as tl +from typing import Literal, Optional +from .rotary import apply_rotary +from .utils import ( + AUTOTUNE, + get_padded_headsize, + get_shape_and_strides_from_layout, + is_cdna, + is_fp8, +) + +DEBUG = False + + +def get_cdna_autotune_configs(): + return [ + triton.Config( + {"BLOCK_M": 128, "BLOCK_N": 128, "waves_per_eu": 2, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ), + triton.Config( + {"BLOCK_M": 128, "BLOCK_N": 64, "waves_per_eu": 2, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ), + triton.Config( + {"BLOCK_M": 128, "BLOCK_N": 64, "waves_per_eu": 3, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ), + triton.Config( + {"BLOCK_M": 128, "BLOCK_N": 64, "waves_per_eu": 1, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ), + triton.Config( + {"BLOCK_M": 128, "BLOCK_N": 32, "waves_per_eu": 2, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ), + triton.Config( + {"BLOCK_M": 64, "BLOCK_N": 64, "waves_per_eu": 1, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ), + # Fall-back config. + triton.Config( + {"BLOCK_M": 16, "BLOCK_N": 16, "waves_per_eu": 1, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ), + ], [ + "IS_CAUSAL", + "dropout_p", + "MAX_SEQLENS_Q", + "MAX_SEQLENS_K", + "ACTUAL_BLOCK_DMODEL", + "VARLEN", + "HQ", + "HK", + ] + + +def get_autotune_configs(): + if AUTOTUNE: + if is_cdna(): + autotune_configs, autotune_keys = get_cdna_autotune_configs() + fwd_auto_tune_configs, fwd_autotune_keys = autotune_configs, autotune_keys + reduce_auto_tune_configs, reduce_autotune_keys = ( + autotune_configs, + autotune_keys, + ) + return (fwd_auto_tune_configs, fwd_autotune_keys), ( + reduce_auto_tune_configs, + reduce_autotune_keys, + ) + else: + raise ValueError("Unknown Device Type") + else: + autotune_configs, autotune_keys = [ + triton.Config( + {"BLOCK_M": 64, "BLOCK_N": 64, "waves_per_eu": 1, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ), + ], [ + "IS_CAUSAL", + "dropout_p", + "MAX_SEQLENS_Q", + "MAX_SEQLENS_K", + "ACTUAL_BLOCK_DMODEL", + "VARLEN", + "HQ", + "HK", + ] + + fwd_auto_tune_configs, fwd_autotune_keys = autotune_configs, autotune_keys + reduce_auto_tune_configs, reduce_autotune_keys = autotune_configs, autotune_keys + return (fwd_auto_tune_configs, fwd_autotune_keys), ( + reduce_auto_tune_configs, + reduce_autotune_keys, + ) + + +(fwd_auto_tune_configs, fwd_autotune_keys), ( + reduce_auto_tune_configs, + reduce_autotune_keys, +) = get_autotune_configs() + + +@triton.jit +def _attn_fwd_inner( + q, + kT, + v, + pos, + col_mask, + m_i, + l_i, + acc, + pid_m, + q_descale, + k_descale, + v_descale, # FP8 scaling factors + IS_FP8: tl.constexpr, # FP8 flag + BLOCK_M: tl.constexpr, + BLOCK_N: tl.constexpr, + N_CTX_Q: tl.constexpr, + N_CTX_K_FINAL: tl.constexpr, + USE_ALIBI: tl.constexpr, + alibi_slope, + USE_SLIDING_WINDOW: tl.constexpr, + IS_CAUSAL: tl.constexpr, + WINDOW_SIZE_LEFT: tl.constexpr, + WINDOW_SIZE_RIGHT: tl.constexpr, + APPLY_COL_MASK: tl.constexpr, # apply provided col_mask when True +): + # -- compute qk --- + qk = tl.zeros([BLOCK_M, BLOCK_N], dtype=tl.float32) + if IS_FP8: + qk += tl.dot(q, kT) * q_descale * k_descale # Apply FP8 scaling + else: + qk += tl.dot(q, kT) # noqa: F821 + + if USE_ALIBI: + row_idx = pid_m * BLOCK_M + tl.arange(0, BLOCK_M) + col_idx = pos + tl.arange(0, BLOCK_N) + + # Compute relative positions + relative_pos = row_idx[:, None] + N_CTX_K_FINAL - (N_CTX_Q + col_idx[None, :]) + relative_pos = tl.abs(relative_pos) + + # Compute ALiBi bias + alibi_bias = -1 * alibi_slope * relative_pos + qk += alibi_bias * 1.44269504 + + # ------------------------------------------------------------------ + # masking + # ------------------------------------------------------------------ + if USE_SLIDING_WINDOW: + row_idx = pid_m * BLOCK_M + tl.arange(0, BLOCK_M) # q positions + col_idx = pos + tl.arange(0, BLOCK_N) # k positions + row = row_idx[:, None] # [M,1] + col = col_idx[None, :] # [1,N] + + if IS_CAUSAL: + # -------- causal + window -------- + diag = N_CTX_K_FINAL - N_CTX_Q # sk-sq + causal_ok = col <= row + diag + if WINDOW_SIZE_LEFT < 0: # only right window + win_ok = col <= row + diag + WINDOW_SIZE_RIGHT + else: # both sides + win_ok = (col >= row + diag - WINDOW_SIZE_LEFT) & ( + col <= row + diag + WINDOW_SIZE_RIGHT + ) + mask = ~(causal_ok & win_ok) # True ⇒ -inf + else: + # -------- non-causal window -------- + sk, sq = N_CTX_K_FINAL, N_CTX_Q + if WINDOW_SIZE_LEFT < 0: + mask = col > row + (sk - sq) + WINDOW_SIZE_RIGHT + else: + right = tl.minimum(row + (sk - sq) + WINDOW_SIZE_RIGHT, sk) + left = row + (sk - sq) - WINDOW_SIZE_LEFT + mask = (col > right) | (col < left) + qk = tl.where(mask, float("-inf"), qk) + else: + if IS_CAUSAL: + row_idx = pid_m * BLOCK_M + tl.arange(0, BLOCK_M) + col_idx = pos + tl.arange(0, BLOCK_N) + + # create a N_CTX_Q x kv_len causal mask + col_offset = N_CTX_K_FINAL - N_CTX_Q + causal_mask = row_idx[:, None] >= (col_idx[None, :] - col_offset) + + # Apply the mask + qk = tl.where(causal_mask, qk, float("-inf")) + + # Column mask (tail / variable-length). Instead of recomputing an arange each time, + # we accept a precomputed mask from the caller (col_valid_mask). + if APPLY_COL_MASK: + # Expect col_mask shape: [BLOCK_N]. True where column is within sequence. + qk = tl.where(col_mask[None, :], qk, float("-inf")) + + m_i_new = tl.maximum(m_i, tl.max(qk, 1)) # per-row max so far + + # rows that are *all* -inf after masking + valid = m_i_new > float("-inf") + + # scale previous partial sums safely + alpha = tl.where(valid, tl.math.exp2(m_i - m_i_new), 0.0) + + # subtract the row max only on valid rows + qk = tl.where(valid[:, None], qk - m_i_new[:, None], float("-inf")) + p = tl.math.exp2(qk) + + # -- update m_i and l_i -- + l_i = l_i * alpha + tl.sum(p, 1) + m_i = m_i_new + p = p.to(q.dtype) + + # -- scale and update acc -- + acc *= alpha[:, None] + if IS_FP8: + acc += tl.dot(p.to(v.dtype), v) * v_descale # Apply FP8 scaling for V + else: + acc += tl.dot(p.to(v.dtype), v) + + return m_i, l_i, acc + + +# @triton.autotune( +# configs=fwd_auto_tune_configs, +# key=fwd_autotune_keys, +# use_cuda_graph=True, +# ) +@triton.jit +def _fwd_kernel_splitK( + Q, + K, + V, + Q_Descale, # FP8 descale factors for Q + K_Descale, # FP8 descale factors for K + V_Descale, # FP8 descale factors for V + sm_scale, + Out_splitK, # [B*H*G, split_k, Mq, K] + Metadata, # [B*H*G, 2, split_k, M_ceil] contains [mi, li] + K_new, + V_new, + Cache_seqlens, + Cache_batch_idx, + Block_table, + Alibi_slopes, + stride_qz, + stride_qm, + stride_qg, + stride_qh, + stride_qd, + stride_kz, + stride_kn, + stride_kg, + stride_kh, + stride_kd, + stride_vz, + stride_vn, + stride_vg, + stride_vh, + stride_vd, + stride_osk_zhg, + stride_osk_s, + stride_osk_m, + stride_osk_d, + stride_mzhg, + stride_m2, + stride_ms, + stride_mm, + stride_kn_z, + stride_kn_n, + stride_kn_g, + stride_kn_h, + stride_kn_d, + stride_vn_z, + stride_vn_n, + stride_vn_g, + stride_vn_h, + stride_vn_d, + stride_bt_b, + stride_bt_s, + stride_az, + stride_ah, + stride_q_descale_z, # FP8 descale strides + stride_q_descale_h, + stride_k_descale_z, + stride_k_descale_h, + stride_v_descale_z, + stride_v_descale_h, + Z, + N_CTX_Q, + N_CTX_K, + N_CTX_NEW, + BLOCK_N_PER_SPLIT, + BLOCK_SIZE_K: tl.constexpr, + H_q: tl.constexpr, + H_kv: tl.constexpr, + G_q: tl.constexpr, + BLOCK_M: tl.constexpr, + BLOCK_DMODEL: tl.constexpr, + ACTUAL_BLOCK_DMODEL: tl.constexpr, + BLOCK_N: tl.constexpr, + BOUNDS_CHECKS_N: tl.constexpr, + USE_CACHE_SEQLENs: tl.constexpr, + USE_CACHE_BATCH_IDX: tl.constexpr, + NEW_KV: tl.constexpr, + IS_GQA: tl.constexpr, + IS_CAUSAL: tl.constexpr, + USE_ALIBI: tl.constexpr, + PADDED_HEAD: tl.constexpr, + GROUP_SIZE: tl.constexpr, + USE_SLIDING_WINDOW: tl.constexpr, + WINDOW_SIZE_LEFT: tl.constexpr, + WINDOW_SIZE_RIGHT: tl.constexpr, + USE_BLOCK_TABLE: tl.constexpr, + IS_FP8: tl.constexpr, # FP8 flag +): + # get program ids + pid_m = tl.program_id(0) + pid_zhg = tl.program_id(1) + pid_splitk = tl.program_id(2) + + # compute z, h and g ids + z_id = pid_zhg // (H_q * G_q) + hq_id = (pid_zhg // G_q) % H_q + g_id = pid_zhg % G_q + + # is gqa + if IS_GQA: + hk_id = hq_id // GROUP_SIZE + hv_id = hk_id + else: + hk_id = hq_id + hv_id = hq_id + + # Load FP8 descale factors if needed + if IS_FP8: + if IS_GQA: + # For MQA/GQA, q_descale uses the same indexing as k/v (hk_id) + q_descale = tl.load( + Q_Descale + z_id * stride_q_descale_z + hk_id * stride_q_descale_h + ) + else: + # For MHA, q_descale uses hq_id + q_descale = tl.load( + Q_Descale + z_id * stride_q_descale_z + hq_id * stride_q_descale_h + ) + k_descale = tl.load( + K_Descale + z_id * stride_k_descale_z + hk_id * stride_k_descale_h + ) + v_descale = tl.load( + V_Descale + z_id * stride_v_descale_z + hv_id * stride_v_descale_h + ) + else: + q_descale, k_descale, v_descale = 1.0, 1.0, 1.0 + + # figure out seqlens + lo = pid_splitk * BLOCK_N_PER_SPLIT + if USE_CACHE_SEQLENs: + cache_seqlen_last_idx = tl.load(Cache_seqlens + z_id) + N_CTX_K_FINAL = cache_seqlen_last_idx + else: + N_CTX_K_FINAL = N_CTX_K + hi = tl.minimum((pid_splitk + 1) * BLOCK_N_PER_SPLIT, N_CTX_K_FINAL) + + # pick batch index + if USE_CACHE_BATCH_IDX: + cache_batch_idx = tl.load(Cache_batch_idx + z_id) + else: + cache_batch_idx = z_id + + # compute offsets + offs_m = pid_m * BLOCK_M + tl.arange(0, BLOCK_M) + offs_n = tl.arange(0, BLOCK_N) + offs_d = tl.arange(0, BLOCK_DMODEL) + + # compute ptrs + q_offset = Q + hq_id * stride_qh + z_id * stride_qz + g_id * stride_qg + q_ptrs = q_offset + offs_m[:, None] * stride_qm + offs_d[None, :] * stride_qd + + # Handle block table for paged attention + if USE_BLOCK_TABLE: + # K and V now point to paged cache + # Each batch has its own block table row + block_table_ptr = Block_table + z_id * stride_bt_b + else: + k_offset = ( + K + hk_id * stride_kh + cache_batch_idx * stride_kz + g_id * stride_kg + ) + v_offset = ( + V + hv_id * stride_vh + cache_batch_idx * stride_vz + g_id * stride_vg + ) + + # compute masks + if PADDED_HEAD: + q_mask = (offs_m < N_CTX_Q)[:, None] & (offs_d < ACTUAL_BLOCK_DMODEL)[None, :] + kT_mask = (offs_d < ACTUAL_BLOCK_DMODEL)[:, None] & (offs_n < N_CTX_K_FINAL)[ + None, : + ] + v_mask = (offs_n < N_CTX_K_FINAL)[:, None] & (offs_d < ACTUAL_BLOCK_DMODEL)[ + None, : + ] + osk_mask = (offs_m < N_CTX_Q)[:, None] & (offs_d < ACTUAL_BLOCK_DMODEL)[None, :] + else: + q_mask = (offs_m < N_CTX_Q)[:, None] + kT_mask = (offs_n < N_CTX_K_FINAL)[None, :] + v_mask = (offs_n < N_CTX_K_FINAL)[:, None] + osk_mask = (offs_m < N_CTX_Q)[:, None] + + # scale sm_scale by log_2(e) and use + # 2^x instead of exp in the loop because CSE and LICM + # don't work as expected with `exp` in the loop + qk_scale = sm_scale * 1.44269504 + + # load q: it will stay in SRAM throughout + q = tl.load(q_ptrs, mask=q_mask, other=0.0) + q = (q * qk_scale).to(q.dtype) + + # load ALiBi slope if enabled + if USE_ALIBI: + a_offset = z_id * stride_az + hq_id * stride_ah + alibi_slope = tl.load(Alibi_slopes + a_offset) + else: + alibi_slope = None + + # initialize pointer to m and l + m_i = tl.full([BLOCK_M], float("-inf"), dtype=tl.float32) + l_i = tl.zeros([BLOCK_M], dtype=tl.float32) + acc = tl.zeros([BLOCK_M, BLOCK_DMODEL], dtype=tl.float32) # noqa: F821 + + # loop over k, v and update accumulator + if USE_BLOCK_TABLE: + # Paged attention: process all KV blocks from cache + # Note: Cache should be updated externally before calling this kernel + num_kv_blocks = (N_CTX_K_FINAL + BLOCK_SIZE_K - 1) // BLOCK_SIZE_K + + for block_idx in range(num_kv_blocks): + # Calculate sequence range for this block + block_start = block_idx * BLOCK_SIZE_K + block_end = tl.minimum(block_start + BLOCK_SIZE_K, N_CTX_K_FINAL) + + # Check if block overlaps with our split-k range [lo, hi) + if block_end > lo and block_start < hi: + # Load physical block number + physical_block = tl.load(block_table_ptr + block_idx * stride_bt_s) + + # Calculate the range within this block that overlaps with [lo, hi) + process_start = tl.maximum(lo - block_start, 0) + process_end = tl.minimum(hi - block_start, BLOCK_SIZE_K) + process_end = tl.minimum(process_end, block_end - block_start) + + # Instead of forcing a floor alignment to BLOCK_N (which can still skip + # part of the intended range if start falls mid-tile for small splits), + # start from the raw (possibly unaligned) process_start rounded *down* but + # allow the loop to begin earlier (at most BLOCK_N before) so that any + # partial tile overlapping [lo, hi) is covered. Masking below will remove + # columns < lo or >= hi ensuring numerically identical coverage without + # duplication. + aligned_start = (process_start // BLOCK_N) * BLOCK_N + if aligned_start > 0 and aligned_start + BLOCK_N > process_start: + # ensure we include the tile that contains process_start + process_start = aligned_start + else: + process_start = aligned_start + + for offset in range(process_start, process_end, BLOCK_N): + # Current position (may begin slightly before logical split range; masking fixes it) + pos = block_start + offset + # Proceed unconditionally; masking below enforces [lo, hi) + # Calculate base addresses for K and V in this physical block + k_base = ( + K + + physical_block * BLOCK_SIZE_K * stride_kn + + hk_id * stride_kh + + g_id * stride_kg + ) + v_base = ( + V + + physical_block * BLOCK_SIZE_K * stride_vn + + hv_id * stride_vh + + g_id * stride_vg + ) + + # Offsets within the current block + block_offs = offset + offs_n + + # Masks for valid data respecting: + # (1) global key length (seq_mask) + # (2) block bounds (block_mask) + # (3) current split range [lo, hi) + seq_mask = (pos + offs_n) < N_CTX_K_FINAL + block_mask = block_offs < BLOCK_SIZE_K + end_mask = block_offs < process_end + split_mask = ((pos + offs_n) >= lo) & ((pos + offs_n) < hi) + col_mask = seq_mask & block_mask & end_mask & split_mask + + # Apply masks + kT_mask_final = kT_mask & col_mask[None, :] + v_mask_final = v_mask & col_mask[:, None] + + # Load K and V + kT_ptrs = ( + k_base + + offs_d[:, None] * stride_kd + + block_offs[None, :] * stride_kn + ) + v_ptrs = ( + v_base + + block_offs[:, None] * stride_vn + + offs_d[None, :] * stride_vd + ) + + kT = tl.load(kT_ptrs, mask=kT_mask_final, other=0.0) + v = tl.load(v_ptrs, mask=v_mask_final, other=0.0) + + # Unified inner function handles both paged and contiguous + m_i, l_i, acc = _attn_fwd_inner( + q, + kT, + v, + pos, + col_mask, + m_i, + l_i, + acc, + pid_m, + q_descale, + k_descale, + v_descale, + IS_FP8, + BLOCK_M, + BLOCK_N, + N_CTX_Q, + N_CTX_K_FINAL, + USE_ALIBI, + alibi_slope, + USE_SLIDING_WINDOW, + IS_CAUSAL, + WINDOW_SIZE_LEFT, + WINDOW_SIZE_RIGHT, + True, + ) + else: + # Non-paged attention: process KV from cache + # Note: Cache should be updated externally before calling this kernel + # loop over k, v and update accumulator + for start_n in range(lo, hi, BLOCK_N): + kT_ptrs = ( + k_offset + + offs_d[:, None] * stride_kd + + (start_n + offs_n)[None, :] * stride_kn + ) + V_ptrs = ( + v_offset + + (start_n + offs_n)[:, None] * stride_vn + + offs_d[None, :] * stride_vd + ) + + # load k + kT = tl.load(kT_ptrs, mask=kT_mask, other=0.0) + v = tl.load(V_ptrs, mask=v_mask, other=0.0) + + # Use the same inner loop logic + # Precompute column validity mask for this tile (all True for full tiles). + # hi is the upper bound of the overall split range; start_n marks this tile's base. + col_valid_mask = offs_n < (hi - start_n) + + m_i, l_i, acc = _attn_fwd_inner( + q, + kT, + v, + start_n, + col_valid_mask, + m_i, + l_i, + acc, + pid_m, + q_descale, + k_descale, + v_descale, + IS_FP8, + BLOCK_M, + BLOCK_N, + N_CTX_Q, + N_CTX_K_FINAL, + USE_ALIBI, + alibi_slope, + USE_SLIDING_WINDOW, + IS_CAUSAL, + WINDOW_SIZE_LEFT, + WINDOW_SIZE_RIGHT, + BOUNDS_CHECKS_N, + ) + + # write back O + osk_offset = Out_splitK + pid_zhg * stride_osk_zhg + pid_splitk * stride_osk_s + osk_ptrs = ( + osk_offset + offs_m[:, None] * stride_osk_m + offs_d[None, :] * stride_osk_d + ) + tl.store( + osk_ptrs, + acc, + mask=osk_mask, + ) + + # write metadata for split-K reduction + metadata_offset = Metadata + pid_zhg * stride_mzhg + pid_splitk * stride_ms + metadata_ptr = metadata_offset + offs_m + tl.store(metadata_ptr, m_i) + tl.store(metadata_ptr + stride_m2, l_i) + + +# @triton.autotune( +# configs=reduce_auto_tune_configs, +# key=reduce_autotune_keys, +# use_cuda_graph=True, +# ) +@triton.jit +def _splitK_reduce( + Out_splitK, # [B*H*G, split_k, Mq, K] + Metadata, # [B*H*G, 2, split_k, M_ceil] contains [mi, li] + Out, # [B, H, G, M, K] + LSE, # [B*H*G, M] + stride_osk_zhg, + stride_osk_s, + stride_osk_m, + stride_osk_k, + stride_mzhg, + stride_m2, + stride_ms, + stride_mm, + stride_oz, + stride_oh, + stride_og, + stride_om, + stride_ok, + stride_lse_zhg, + stride_lse_m, + K_BLOCK_SIZE: tl.constexpr, + BLOCK_DMODEL: tl.constexpr, + ACTUAL_BLOCK_DMODEL: tl.constexpr, + H: tl.constexpr, + G: tl.constexpr, + split_k: tl.constexpr, + splitK_pow2: tl.constexpr, + MASK_SPLITK: tl.constexpr, + PADDED_HEAD: tl.constexpr, +): + # get pids + pid_zhg = tl.program_id(0) + pid_m = tl.program_id(1) + pid_k = tl.program_id(2) + + # compute offsets + offs_splitK = tl.arange(0, splitK_pow2) + offs_k = pid_k * K_BLOCK_SIZE + tl.arange(0, K_BLOCK_SIZE) + + # compute masks + if PADDED_HEAD: + o_mask = offs_k < ACTUAL_BLOCK_DMODEL + else: + o_mask = None + + # compute ptrs + metadata_offset = Metadata + pid_zhg * stride_mzhg + metadata_ptr = metadata_offset + offs_splitK * stride_ms + pid_m * stride_mm + + osk_offset = Out_splitK + pid_zhg * stride_osk_zhg + pid_m * stride_osk_m + osk_ptr = ( + osk_offset + + offs_splitK[:, None] * stride_osk_s + + offs_k[None, :] * stride_osk_k + ) + + # read max values of each splitK + if MASK_SPLITK: + splitK_mask = offs_splitK < split_k + l_m = tl.load(metadata_ptr, mask=splitK_mask, other=float("-inf")) + l_sum = tl.load(metadata_ptr + stride_m2, mask=splitK_mask, other=0.0) + acc = tl.load(osk_ptr, mask=splitK_mask[:, None], other=0.0) + else: + l_m = tl.load(metadata_ptr) + l_sum = tl.load(metadata_ptr + stride_m2) + acc = tl.load(osk_ptr) + + g_m = tl.max(l_m, axis=0) + + alpha = tl.where(l_m > float("-inf"), tl.math.exp2(l_m - g_m), 0.0) + + # read sum + l_sum *= alpha + g_sum = tl.sum(l_sum, axis=0) + acc = acc * alpha[:, None] + + g_sum_safe = tl.where(g_sum > 0, g_sum, 1.0) + acc_out = tl.sum(acc, axis=0) / g_sum_safe + + # Store output + z_id = pid_zhg // (H * G) + h_id = (pid_zhg // G) % H + g_id = pid_zhg % G + out_offset = Out + z_id * stride_oz + h_id * stride_oh + g_id * stride_og + out_ptr = out_offset + pid_m * stride_om + offs_k + tl.store(out_ptr, acc_out, mask=o_mask) + + # Store lse + l_ptrs = LSE + pid_zhg * stride_lse_zhg + pid_m + lse_val = tl.where(g_sum > 0, (g_m + tl.math.log2(g_sum)) / 1.44269504, g_m) + tl.store(l_ptrs, lse_val) + + +@triton.jit +def cast_uint32_to_half2(scale_shift): + # Extract two float16 packed into one int32 + scale = scale_shift & 0xFFFF + shift = scale_shift >> 16 + scale = scale.to(tl.uint16).to(tl.float16, bitcast=True) + shift = shift.to(tl.uint16).to(tl.float16, bitcast=True) + return scale, shift + + +@triton.jit +def dequantize( + x_, + scale, + shift, + PACKED_PER_VAL: tl.constexpr = 8, +): + # PACKED_PER_VAL is the number of values packed into + # each element x_. For example, for int4 quantization + # and x_ of type int32, PACKED_PER_VAL is 8. + + BLOCK_N: tl.constexpr = x_.shape[0] + BLOCK_DMODEL_PACKED: tl.constexpr = x_.shape[1] + offsets = tl.arange(0, PACKED_PER_VAL) * 4 + quant_offset = ( + x_[:, None, :] >> offsets[None, :, None] + ) # (BLOCK_N, PACKED_PER_VAL, D // PACKED_PER_VAL) + + quant_offset = tl.view( + quant_offset, (BLOCK_N, BLOCK_DMODEL_PACKED * PACKED_PER_VAL) + ) + # Trick - instead of converting int4 to float16 we view it as float16 + # and then multiply by 32768 * 512 == 2**24 + quant_offset = (quant_offset & 0xF).to(tl.uint16).to(tl.float16, bitcast=True) + quant_offset = (quant_offset * 32768.0).to(tl.float16) + scale_512 = scale * 512 + + dequant = quant_offset * scale_512 + shift + return dequant + + +def quantize_kv_int4(k: torch.Tensor, num_groups: int = 1) -> torch.Tensor: + # Scale and shift are such that quantization linearly maps + # int4 values range [0..15] to input values range min(k)..max(k) + # individually for every row + k = k.reshape(*k.shape[:-1], num_groups, k.shape[-1] // num_groups) + max_vals = torch.max(k, dim=-1, keepdim=True).values + min_vals = torch.min(k, dim=-1, keepdim=True).values + scale_k: torch.Tensor = (max_vals - min_vals) / 15 + + shift_k = torch.min(k, dim=-1, keepdim=True).values + scale_k = scale_k.to(torch.float16) + shift_k = shift_k.to(torch.float16) + + in_bytes = ((k - shift_k.expand(k.shape)) / scale_k.expand(k.shape)) + 0.5 + in_bytes = in_bytes.to(torch.uint8) + in_int4 = in_bytes & 0xF + in_int4_packed = in_int4[..., ::2] + (in_int4[..., 1::2] << 4) + scale_shift = torch.concat( + [scale_k.view(torch.uint8), shift_k.view(torch.uint8)], dim=-1 + ) + k_quant = torch.concat( + [ + scale_shift.flatten(start_dim=-2), + in_int4_packed.flatten(start_dim=-2), + ], + dim=-1, + ).view(torch.int16) + return k_quant + + +def dequantize_kv_fp16(quant_k: torch.Tensor, num_groups: int = 1) -> torch.Tensor: + k_i16 = quant_k.view(torch.int16) + k_ui8 = k_i16.view(torch.uint8) + + ss_size = num_groups * 4 + scale_shift_ui8 = k_ui8[..., 0:ss_size] + scale_shift_ui8 = scale_shift_ui8.reshape( + *scale_shift_ui8.shape[:-1], num_groups, 4 + ) + scale = scale_shift_ui8[..., 0:2].view(torch.float16) + shift = scale_shift_ui8[..., 2:4].view(torch.float16) + + kv_ui8 = k_ui8[..., ss_size:] + k_ui8 = kv_ui8.reshape(*kv_ui8.shape[:-1], num_groups, -1) + k1_i4 = k_ui8 & 0xF + k2_i4 = (k_ui8 & 0xF0) >> 4 + k_shape = k1_i4.shape + k1_f16 = k1_i4.to(torch.float16) * scale.expand(k_shape) + shift.expand(k_shape) + k2_f16 = k2_i4.to(torch.float16) * scale.expand(k_shape) + shift.expand(k_shape) + + out = torch.empty( + (*k1_f16.shape[:-1], k1_f16.shape[-1] * 2), + dtype=torch.float16, + device=quant_k.device, + ) + out[..., ::2] = k1_f16 + out[..., 1::2] = k2_f16 + out = out.reshape(*k_shape[:-2], -1) + + return out + + +def get_split_k(B: int, G: int, H: int, Mk: int) -> int: + """Heuristic for the number of splits""" + bh = max(B * H, 1) # NOTE: Handle B*h=0 case + split_k = max(Mk, 1024) // bh + max_chunk_size = 64 + while split_k > 0 and Mk / split_k < max_chunk_size: + split_k = split_k // 2 + while B * H * G * split_k >= 1024: + split_k = split_k // 2 + split_k = min(split_k, 512) + split_k = max(split_k, 1) + return split_k + + +def attention_decode_forward_triton_impl( + q: torch.Tensor, + k_cache: torch.Tensor, + v_cache: torch.Tensor, + k_new: Optional[torch.Tensor], + v_new: Optional[torch.Tensor], + out: torch.Tensor, + sm_scale: float, + causal: bool, + window_size_left: int, + window_size_right: int, + alibi_slopes: Optional[torch.Tensor], + layout: Literal["bshd"], + cache_seqlens: Optional[torch.Tensor], + cache_batch_idx: Optional[torch.Tensor], + block_table: Optional[torch.Tensor] = None, + q_descale: Optional[torch.Tensor] = None, + k_descale: Optional[torch.Tensor] = None, + v_descale: Optional[torch.Tensor] = None, + # rotary (optional) + rotary_cos: Optional[torch.Tensor] = None, + rotary_sin: Optional[torch.Tensor] = None, + rotary_interleaved: bool = False, + seqlens_rotary: Optional[torch.Tensor] = None, +): + # apply rotary embedding + if rotary_cos is not None and rotary_sin is not None: + # Prefer explicitly provided rotary sequence start offsets if given; fall back to cache_seqlens. + seqlen_offsets = ( + seqlens_rotary + if seqlens_rotary is not None + else (cache_seqlens if cache_seqlens is not None else 0) + ) + local = (window_size_left != -1) or (window_size_right != -1) + q, k_new = apply_rotary( + q, + k_new, + rotary_cos, + rotary_sin, + causal=causal, + local=local, + interleaved=rotary_interleaved, + seqlen_offsets=seqlen_offsets, + ) + + # handle cache updates + if k_new is not None and v_new is not None: + # Update cache with new KV values + if block_table is None: + # Non-paged attention: update cache directly + batch_size = k_new.shape[0] + seqlen_new = k_new.shape[1] + + if cache_seqlens is not None: + # Use cache_seqlens to determine where to insert new KV + for b in range(batch_size): + start_idx = int(cache_seqlens[b].item()) + end_idx = start_idx + seqlen_new + k_cache[b, start_idx:end_idx] = k_new[b] + v_cache[b, start_idx:end_idx] = v_new[b] + cache_seqlens[b] = end_idx + else: + # Append at the end of existing cache + seqlen_cache = k_cache.shape[1] + k_cache[:, seqlen_cache - seqlen_new :] = k_new + v_cache[:, seqlen_cache - seqlen_new :] = v_new + else: + # Paged attention: update cache using block table + batch_size = k_new.shape[0] + seqlen_new = k_new.shape[1] + block_size = k_cache.shape[ + 1 + ] # k_cache shape: [num_blocks, block_size, nheads, head_dim] + + # Update cache for each batch element + for b in range(batch_size): + if cache_seqlens is not None: + start_idx = int(cache_seqlens[b].item()) + else: + # If no cache_seqlens, assume we're appending at the end + # Find the last used position from block table + start_idx = 0 + for block_idx in range(block_table.shape[1]): + if block_table[b, block_idx] >= 0: + start_idx = (block_idx + 1) * block_size + else: + start_idx = block_idx * block_size + break + + # Copy new KV values into the paged cache + for i in range(seqlen_new): + pos = start_idx + i + block_idx = pos // block_size + within_block_idx = pos % block_size + + # Get the physical block number from block table + if block_idx < block_table.shape[1]: + physical_block = int(block_table[b, block_idx].item()) + + # Update k_cache and v_cache at the physical block location + k_cache[physical_block, within_block_idx] = k_new[b, i] + v_cache[physical_block, within_block_idx] = v_new[b, i] + + # Update cache_seqlens if provided + if cache_seqlens is not None: + cache_seqlens[b] = start_idx + seqlen_new + + # triton configs + BLOCK_M = 16 + BLOCK_N = 64 + num_stages = 1 + num_warps_fwd = 1 + num_warps_reduce = 4 + + # kernel_configs + is_new_kv = False # Cache has been updated, so no new KV in kernel + use_alibi, (stride_az, stride_ah) = True if alibi_slopes is not None else False, ( + alibi_slopes.stride() if alibi_slopes is not None else (None, None) + ) + use_cache_seqlens = cache_seqlens is not None + use_sliding_window = window_size_left != -1 or window_size_right != -1 + use_block_table = block_table is not None + SPLIT_K = None + NUM_QUANT_GROUPS = 1 + + # get shapes and strides + (batch_size, seqlen_q, nheads_q, dim_q), ( + stride_qz, + stride_qh, + stride_qm, + stride_qd, + ) = get_shape_and_strides_from_layout(q, layout) + + # Handle paged KV cache layout + if use_block_table: + # For paged attention, k_cache and v_cache have shape [num_blocks, block_size, nheads, head_dim] + num_blocks_kc, block_size_k, nheads_kc, dim_kc = k_cache.shape + num_blocks_vc, block_size_v, nheads_vc, dim_vc = v_cache.shape + # Get the actual sequence length from cache_seqlens or block_table + if cache_seqlens is not None: + seqlen_kc = int(cache_seqlens.max().item()) + else: + # Infer from block_table shape [batch_size, num_blocks_per_seq] + num_blocks_per_seq = block_table.shape[1] + seqlen_kc = num_blocks_per_seq * block_size_k + seqlen_vc = seqlen_kc + + # Strides for paged layout + stride_kc_z = 0 # No batch dimension in paged cache + stride_kc_n = k_cache.stride(1) # Sequence stride + stride_kc_h = k_cache.stride(2) # Head stride + stride_kc_d = k_cache.stride(3) # Dim stride + + stride_vc_z = 0 + stride_vc_n = v_cache.stride(1) + stride_vc_h = v_cache.stride(2) + stride_vc_d = v_cache.stride(3) + else: + (_, seqlen_kc, nheads_kc, dim_kc), ( + stride_kc_z, + stride_kc_h, + stride_kc_n, + stride_kc_d, + ) = get_shape_and_strides_from_layout(k_cache, layout) + (_, seqlen_vc, nheads_vc, dim_vc), ( + stride_vc_z, + stride_vc_h, + stride_vc_n, + stride_vc_d, + ) = get_shape_and_strides_from_layout(v_cache, layout) + block_size_k = 0 # Not used + if is_new_kv: + (_, seqlen_kn, nheads_kn, dim_kn), ( + stride_kn_z, + stride_kn_h, + stride_kn_n, + stride_kn_d, + ) = get_shape_and_strides_from_layout(k_new, layout) + (_, seqlen_vn, nheads_vn, dim_vn), ( + stride_vn_z, + stride_vn_h, + stride_vn_n, + stride_vn_d, + ) = get_shape_and_strides_from_layout(v_new, layout) + else: + (_, seqlen_kn, nheads_kn, dim_kn), ( + stride_kn_z, + stride_kn_h, + stride_kn_n, + stride_kn_d, + ) = (None, None, None, None), (None, None, None, None) + (_, seqlen_vn, nheads_vn, dim_vn), ( + stride_vn_z, + stride_vn_h, + stride_vn_n, + stride_vn_d, + ) = (None, None, None, None), (None, None, None, None) + (_, seqlen_o, nheads_o, dim_o), (stride_oz, stride_oh, stride_om, stride_od) = ( + get_shape_and_strides_from_layout(out, layout) + ) + assert ( + dim_q == dim_kc == dim_vc + ), f"Dimensions must match: {dim_q}, {dim_kc}, {dim_vc}" + + # add extra information needed by the kernels + if layout == "bshd": + (n_group_q, heads_per_group_q), stride_qg = (1, nheads_q), stride_qm + (n_group_k, heads_per_group_k), stride_kc_g = (1, nheads_kc), stride_kc_n + (n_group_v, heads_per_group_v), stride_vc_g = (1, nheads_vc), stride_vc_n + if is_new_kv: + (n_group_kn, heads_per_group_kn), stride_kn_g = (1, nheads_kn), stride_kn_n + (n_group_vn, heads_per_group_vn), stride_vn_g = (1, nheads_vn), stride_vn_n + else: + (n_group_kn, heads_per_group_kn), stride_kn_g = (None, None), None + (n_group_vn, heads_per_group_vn), stride_vn_g = (None, None), None + (n_group_o, heads_per_group_o), stride_og = (1, nheads_o), stride_om + else: + raise ValueError(f"{layout} layout is not supported") + + # get padded size + dim_padded = get_padded_headsize(dim_kc) + is_padded_head = dim_padded != dim_kc + + # Handle MQA/GQA case + group_size = nheads_q // nheads_kc + if group_size > 1: + is_gqa = True + else: + is_gqa = False + + if SPLIT_K is not None: + split_k = SPLIT_K + else: + # Use heuristics + if use_block_table: + # For paged attention, use the actual sequence length from cache_seqlens + max_seqlen = ( + int(cache_seqlens.max().item()) + if cache_seqlens is not None + else block_size_k + ) + split_k = get_split_k(batch_size, n_group_q, heads_per_group_q, max_seqlen) + else: + split_k = get_split_k(batch_size, n_group_q, heads_per_group_q, seqlen_kc) + split_size = (seqlen_kc + split_k - 1) // split_k + + # setup grid + seqlen_q_ceil = (seqlen_q + BLOCK_M - 1) // BLOCK_M * BLOCK_M + grid = lambda META: ( + triton.cdiv(seqlen_q, META["BLOCK_M"]), + batch_size * n_group_q * heads_per_group_q, + split_k, + ) + + # create intermediate tensors + out_splitk = torch.empty( + [batch_size * n_group_q * heads_per_group_q, split_k, seqlen_q_ceil, dim_kc], + dtype=torch.float32, + device=q.device, + ) + metadata = torch.empty( + [batch_size * n_group_q * heads_per_group_q, 2, split_k, seqlen_q_ceil], + dtype=torch.float32, + device=q.device, + ) + lse = torch.empty( + (batch_size * n_group_q * heads_per_group_q, seqlen_q), + dtype=torch.float32, + device=q.device, + ) + + # get intermediate tensor strides + stride_osk_zhg, stride_osk_s, stride_osk_m, stride_osk_d = out_splitk.stride() + stride_mzhg, stride_m2, stride_ms, stride_mm = metadata.stride() + stride_lse_zhg, stride_lse_m = lse.stride() + + # Block table strides + if use_block_table: + stride_bt_b, stride_bt_s = block_table.stride() + else: + stride_bt_b, stride_bt_s = 0, 0 + + # FP8 support + IS_FP8 = is_fp8(q) + if IS_FP8: + if (q_descale is None) or (k_descale is None) or (v_descale is None): + import warnings + + warnings.warn( + "FP8 tensors detected but descale factors not provided. Using default scale of 1.0", + UserWarning, + ) + # Create default descale tensors if not provided + if q_descale is None: + q_descale = torch.ones( + batch_size, nheads_q, dtype=torch.float32, device=q.device + ) + if k_descale is None: + k_descale = torch.ones( + batch_size, nheads_kc, dtype=torch.float32, device=q.device + ) + if v_descale is None: + v_descale = torch.ones( + batch_size, nheads_vc, dtype=torch.float32, device=q.device + ) + stride_q_descale_z, stride_q_descale_h = q_descale.stride() + stride_k_descale_z, stride_k_descale_h = k_descale.stride() + stride_v_descale_z, stride_v_descale_h = v_descale.stride() + else: + q_descale = None + k_descale = None + v_descale = None + stride_q_descale_z = 0 + stride_q_descale_h = 0 + stride_k_descale_z = 0 + stride_k_descale_h = 0 + stride_v_descale_z = 0 + stride_v_descale_h = 0 + + if DEBUG: + print( + "batch_size, seqlen_q, nheads_q, dim_q", + (batch_size, seqlen_q, nheads_q, dim_q), + ) + print("_, seqlen_kc, nheads_kc, dim_kc", (_, seqlen_kc, nheads_kc, dim_kc)) + print("dim_padded:", dim_padded) + print( + "stride_qz, stride_qm, stride_qg, stride_qh, stride_qd", + (stride_qz, stride_qm, stride_qg, stride_qh, stride_qd), + ) + print( + "stride_kc_z, stride_kc_n, stride_kc_g, stride_kc_h, stride_kc_d", + (stride_kc_z, stride_kc_n, stride_kc_g, stride_kc_h, stride_kc_d), + ) + print( + "stride_vc_z, stride_vc_n, stride_vc_g, stride_vc_h, stride_vc_d", + (stride_vc_z, stride_vc_n, stride_vc_g, stride_vc_h, stride_vc_d), + ) + if is_new_kv: + print( + "stride_kn_z, stride_kn_n, stride_kn_g, stride_kn_h, stride_kn_d", + (stride_kn_z, stride_kn_n, stride_kn_g, stride_kn_h, stride_kn_d), + ) + print( + "stride_vn_z, stride_vn_n, stride_vn_g, stride_vn_h, stride_vn_d", + (stride_vn_z, stride_vn_n, stride_vn_g, stride_vn_h, stride_vn_d), + ) + print( + "stride_oz, stride_om, stride_og, stride_oh, stride_od", + (stride_oz, stride_om, stride_og, stride_oh, stride_od), + ) + print( + "stride_osk_zhg, stride_osk_s, stride_osk_m, stride_osk_d", + (stride_osk_zhg, stride_osk_s, stride_osk_m, stride_osk_d), + ) + print( + "stride_mzhg, stride_m2, stride_ms, stride_mm", + (stride_mzhg, stride_m2, stride_ms, stride_mm), + ) + print("stride_lse_zhg, stride_lse_m", (stride_lse_zhg, stride_lse_m)) + + _fwd_kernel_splitK[grid]( + Q=q, + K=k_cache, + V=v_cache, + Q_Descale=q_descale, + K_Descale=k_descale, + V_Descale=v_descale, + sm_scale=sm_scale, + Out_splitK=out_splitk, + Metadata=metadata, + K_new=None, + V_new=None, + Cache_seqlens=cache_seqlens, + Cache_batch_idx=cache_batch_idx, + Block_table=block_table, + Alibi_slopes=alibi_slopes, + # q strides + stride_qz=stride_qz, + stride_qm=stride_qm, + stride_qg=stride_qg, + stride_qh=stride_qh, + stride_qd=stride_qd, + # k strides + stride_kz=stride_kc_z, + stride_kn=stride_kc_n, + stride_kg=stride_kc_g, + stride_kh=stride_kc_h, + stride_kd=stride_kc_d, + # v strides + stride_vz=stride_vc_z, + stride_vn=stride_vc_n, + stride_vg=stride_vc_g, + stride_vh=stride_vc_h, + stride_vd=stride_vc_d, + # out_splitk strides + stride_osk_zhg=stride_osk_zhg, + stride_osk_s=stride_osk_s, + stride_osk_m=stride_osk_m, + stride_osk_d=stride_osk_d, + # metadata strides + stride_mzhg=stride_mzhg, + stride_m2=stride_m2, + stride_ms=stride_ms, + stride_mm=stride_mm, + # k_new strides + stride_kn_z=stride_kn_z, + stride_kn_n=stride_kn_n, + stride_kn_g=stride_kn_g, + stride_kn_h=stride_kn_h, + stride_kn_d=stride_kn_d, + # v_new strides + stride_vn_z=stride_vn_z, + stride_vn_n=stride_vn_n, + stride_vn_g=stride_vn_g, + stride_vn_h=stride_vn_h, + stride_vn_d=stride_vn_d, + # block table strides + stride_bt_b=stride_bt_b, + stride_bt_s=stride_bt_s, + # alibi strides + stride_az=stride_az, + stride_ah=stride_ah, + # FP8 descale strides + stride_q_descale_z=stride_q_descale_z, + stride_q_descale_h=stride_q_descale_h, + stride_k_descale_z=stride_k_descale_z, + stride_k_descale_h=stride_k_descale_h, + stride_v_descale_z=stride_v_descale_z, + stride_v_descale_h=stride_v_descale_h, + Z=batch_size, + H_q=heads_per_group_q, + H_kv=heads_per_group_k, + G_q=n_group_q, + N_CTX_Q=seqlen_q, + N_CTX_K=seqlen_kc, + N_CTX_NEW=0, # No new KV, cache already updated + BLOCK_N_PER_SPLIT=split_size, + BLOCK_SIZE_K=block_size_k if use_block_table else 256, + BLOCK_M=BLOCK_M, + BLOCK_N=BLOCK_N, + BLOCK_DMODEL=dim_padded, + ACTUAL_BLOCK_DMODEL=dim_kc, + BOUNDS_CHECKS_N=(split_size % BLOCK_N) > 0 or use_cache_seqlens, + USE_CACHE_SEQLENs=use_cache_seqlens, + USE_CACHE_BATCH_IDX=cache_batch_idx is not None, + NEW_KV=False, # Cache already updated + IS_GQA=is_gqa, + IS_CAUSAL=causal, + USE_ALIBI=use_alibi, + PADDED_HEAD=is_padded_head, + GROUP_SIZE=group_size, + USE_SLIDING_WINDOW=use_sliding_window, + WINDOW_SIZE_LEFT=window_size_left, + WINDOW_SIZE_RIGHT=window_size_right, + USE_BLOCK_TABLE=use_block_table, + IS_FP8=IS_FP8, + num_warps=num_warps_fwd, + num_stages=num_stages, + ) + + if DEBUG: + print("Out_splitK:", out_splitk, out_splitk.shape) + print("metadata:", metadata, metadata.shape) + print("lse:", lse, lse.shape) + print("Out:", out, out.shape) + + # Merge together + splitK_pow2 = triton.next_power_of_2(split_k) + mask_split_k = splitK_pow2 > split_k + if batch_size * n_group_q * heads_per_group_q * seqlen_q >= 512: + k_block_num = 1 + else: + k_block_num = 2 + assert dim_padded % k_block_num == 0 + k_block_size = dim_padded // k_block_num + grid = (batch_size * n_group_q * heads_per_group_q, seqlen_q, k_block_num) + + if DEBUG: + print("splitK_pow2:", splitK_pow2) + print("k_block_num:", k_block_num) + print("k_block_size:", k_block_size) + print("grid:", grid) + + _splitK_reduce[grid]( + out_splitk, + metadata, + out, + lse, + # Split-K output strides + stride_osk_zhg=stride_osk_zhg, + stride_osk_s=stride_osk_s, + stride_osk_m=stride_osk_m, + stride_osk_k=stride_osk_d, + # Metadata strides + stride_mzhg=stride_mzhg, + stride_m2=stride_m2, + stride_ms=stride_ms, + stride_mm=stride_mm, + # Output tensor strides + stride_oz=stride_oz, + stride_oh=stride_oh, + stride_og=stride_og, + stride_om=stride_om, + stride_ok=stride_od, + # LSE strides + stride_lse_zhg=stride_lse_zhg, + stride_lse_m=stride_lse_m, + K_BLOCK_SIZE=k_block_size, + BLOCK_DMODEL=dim_padded, + ACTUAL_BLOCK_DMODEL=dim_kc, + G=n_group_q, + H=heads_per_group_q, + # TODO: Tune num_warps + split_k=split_k, + splitK_pow2=splitK_pow2, + MASK_SPLITK=mask_split_k, + PADDED_HEAD=is_padded_head, + num_warps=num_warps_reduce, + ) + + return lse diff --git a/aiter/ops/triton/flash_attn_triton_amd/fwd_prefill.py b/aiter/ops/triton/flash_attn_triton_amd/fwd_prefill.py new file mode 100755 index 0000000000..fc7398734b --- /dev/null +++ b/aiter/ops/triton/flash_attn_triton_amd/fwd_prefill.py @@ -0,0 +1,2002 @@ +import os +import torch +import triton +import triton.language as tl +from typing import Literal, Optional +from .rotary import apply_rotary +from .utils import ( + DROPOUT_USE_PYTORCH, + DROPOUT_DUMP, + AUTOTUNE, + compute_alibi_block, + compute_fp8_scaling_factors, + get_arch, + is_cdna, + is_fp8, + is_rdna, + create_dropout_mask, + DEBUG, +) + +# NOTE: triton fails to import tl.constexprs so create them here for the file +tl_DROPOUT_USE_PYTORCH: tl.constexpr = triton.language.constexpr(DROPOUT_USE_PYTORCH) +tl_DROPOUT_DUMP: tl.constexpr = triton.language.constexpr(DROPOUT_DUMP) + + +# ------------------------------- +# Autotune +# ------------------------------- +def get_fwd_prefill_cdna_autotune_configs(): + return [ + triton.Config( + {"BLOCK_M": 128, "BLOCK_N": 128, "waves_per_eu": 2, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ), + triton.Config( + {"BLOCK_M": 128, "BLOCK_N": 64, "waves_per_eu": 2, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ), + triton.Config( + {"BLOCK_M": 128, "BLOCK_N": 64, "waves_per_eu": 3, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ), + triton.Config( + {"BLOCK_M": 128, "BLOCK_N": 64, "waves_per_eu": 1, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ), + triton.Config( + {"BLOCK_M": 128, "BLOCK_N": 32, "waves_per_eu": 2, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ), + triton.Config( + {"BLOCK_M": 64, "BLOCK_N": 64, "waves_per_eu": 1, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ), + # Fall-back config. + triton.Config( + {"BLOCK_M": 16, "BLOCK_N": 16, "waves_per_eu": 1, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ), + ], [ + "IS_CAUSAL", + "dropout_p", + "MAX_SEQLENS_Q", + "MAX_SEQLENS_K", + "ACTUAL_BLOCK_DMODEL_QK", + "ACTUAL_BLOCK_DMODEL_V", + "IS_VARLEN", + "HQ", + "HK", + ] + + +def get_fwd_prefill_rdna_autotune_configs(): + return [ + triton.Config( + {"BLOCK_M": 32, "BLOCK_N": 32, "waves_per_eu": 4, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=2, + ), + triton.Config( + {"BLOCK_M": 32, "BLOCK_N": 32, "waves_per_eu": 2, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=2, + ), + triton.Config( + {"BLOCK_M": 32, "BLOCK_N": 16, "waves_per_eu": 4, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=2, + ), + triton.Config( + {"BLOCK_M": 32, "BLOCK_N": 16, "waves_per_eu": 2, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=2, + ), + triton.Config( + {"BLOCK_M": 16, "BLOCK_N": 16, "waves_per_eu": 4, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=2, + ), + triton.Config( + {"BLOCK_M": 16, "BLOCK_N": 16, "waves_per_eu": 2, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=2, + ), + # Fall-back config. + triton.Config( + {"BLOCK_M": 16, "BLOCK_N": 16, "waves_per_eu": 1, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=2, + ), + ], [ + "IS_CAUSAL", + "dropout_p", + "MAX_SEQLENS_Q", + "MAX_SEQLENS_K", + "ACTUAL_BLOCK_DMODEL_QK", + "ACTUAL_BLOCK_DMODEL_V", + "IS_VARLEN", + "HQ", + "HK", + ] + + +def get_fwd_prefill_autotune_configs(): + if AUTOTUNE: + if is_rdna(): + return get_fwd_prefill_rdna_autotune_configs() + elif is_cdna(): + return get_fwd_prefill_cdna_autotune_configs() + else: + raise ValueError("Unknown Device Type") + else: + arch = get_arch() + if arch == "gfx950": + default_config = triton.Config( + { + "BLOCK_M": 128, + "BLOCK_N": 128, + "waves_per_eu": 2, + "PRE_LOAD_V": False, + }, + num_stages=1, + num_warps=4, + ) + elif ( + arch == "gfx942" and False + ): # Disabled due shared mem oom in CI when using triton==3.3.0 when using top of tree everything seems fine. + default_config = triton.Config( + {"BLOCK_M": 128, "BLOCK_N": 64, "waves_per_eu": 2, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ) + else: + default_config = triton.Config( + {"BLOCK_M": 64, "BLOCK_N": 64, "waves_per_eu": 2, "PRE_LOAD_V": False}, + num_stages=1, + num_warps=4, + ) + + return [default_config], [ + "IS_CAUSAL", + "dropout_p", + "MAX_SEQLENS_Q", + "MAX_SEQLENS_K", + "ACTUAL_BLOCK_DMODEL_QK", + "ACTUAL_BLOCK_DMODEL_V", + "IS_VARLEN", + "HQ", + "HK", + ] + + +fwd_prefill_autotune_configs, fwd_prefill_autotune_keys = ( + get_fwd_prefill_autotune_configs() +) + + +@triton.jit +def _attn_fwd_no_mask( + acc, + l_i, + m_i, + q, + k_base_ptrs, + v_base_ptrs, + bias_base_ptrs, + stride_kn, + stride_vk, + stride_bn, + stride_sn, + start_m, + seqlen_k, + seqlen_q, + dropout_p, + philox_seed, + philox_base_ptrs, + sd_mask_base_ptrs, + dropout_mask_base_ptrs, + offs_m, + offs_n, + offs_d_qk, + offs_d_v, + block_min, + block_max, + alibi_slope, + q_descale, + k_descale, + v_descale, + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, + FP8_P_DESCALE: tl.constexpr, + BLOCK_M: tl.constexpr, + BLOCK_DMODEL_QK: tl.constexpr, + BLOCK_DMODEL_V: tl.constexpr, + BLOCK_N: tl.constexpr, + PRE_LOAD_V: tl.constexpr, + ENABLE_DROPOUT: tl.constexpr, + PADDED_HEAD_QK: tl.constexpr, + PADDED_HEAD_V: tl.constexpr, + ACTUAL_BLOCK_DMODEL_QK: tl.constexpr, + ACTUAL_BLOCK_DMODEL_V: tl.constexpr, + SM_SCALE: tl.constexpr, + USE_ALIBI: tl.constexpr, + USE_EXP2: tl.constexpr, + RETURN_SCORES: tl.constexpr, + ACCUMULATOR_TYPE, +): + if USE_EXP2: + RCP_LN2: tl.constexpr = 1.4426950408889634 + + # loop over k, v, and update accumulator + for start_n in range(block_min, block_max, BLOCK_N): + # get ptrs + k_ptrs = k_base_ptrs + start_n * stride_kn + v_ptrs = v_base_ptrs + start_n * stride_vk + + kv_offs_n = start_n + tl.arange(0, BLOCK_N) + if PADDED_HEAD_QK: + k_mask, k_mask_other = (offs_d_qk[:, None] < ACTUAL_BLOCK_DMODEL_QK), 0.0 + else: + k_mask, k_mask_other = None, None + + if PADDED_HEAD_V: + v_mask, v_mask_other = (offs_d_v[None, :] < ACTUAL_BLOCK_DMODEL_V), 0.0 + else: + v_mask, v_mask_other = None, None + + # load k and if preload_v then v + k = ( + tl.load(k_ptrs, mask=k_mask, other=k_mask_other) + if PADDED_HEAD_QK + else tl.load(k_ptrs) + ) + if PRE_LOAD_V: + v = ( + tl.load(v_ptrs, mask=v_mask, other=v_mask_other) + if PADDED_HEAD_V + else tl.load(v_ptrs) + ) + + # setup qk accumlator + qk = tl.zeros([BLOCK_M, BLOCK_N], dtype=ACCUMULATOR_TYPE) + + # -- compute qk ---- + if IS_FP8: + qk += tl.dot(q, k) * q_descale * k_descale + else: + qk += tl.dot(q, k) + qk_scaled = qk * SM_SCALE + + if USE_ALIBI: + # compute the global position of each token within the sequence + q_offs_m = start_m * BLOCK_M + tl.arange(0, BLOCK_M) + alibi_block = compute_alibi_block( + alibi_slope, seqlen_q, seqlen_k, q_offs_m, kv_offs_n + ) + qk_scaled += alibi_block + + # compute qk mask + qk_mask = (offs_m[:, None] < seqlen_q) & (kv_offs_n[None, :] < seqlen_k) + + # compute bias + if bias_base_ptrs is not None: + bias_ptrs = bias_base_ptrs + start_n * stride_bn + bias = tl.load(bias_ptrs, mask=qk_mask, other=0.0) + qk_scaled += bias + + # get max scores so far + m_ij = tl.maximum(m_i, tl.max(qk_scaled, 1)) + + # scale and subtract max + q_shifted = tl.where( + m_ij[:, None] == float("-inf"), float("-inf"), qk_scaled - m_ij[:, None] + ) + + # Compute scaled QK and softmax probabilities + if USE_EXP2: + p = tl.math.exp2(q_shifted * RCP_LN2) + else: + p = tl.math.exp(q_shifted) + + # CAVEAT: Must update l_ij before applying dropout + l_ij = tl.sum(p, 1) + if ENABLE_DROPOUT: + dropout_mask_ptrs = dropout_mask_base_ptrs + start_n * stride_sn + sd_mask_ptrs = sd_mask_base_ptrs + start_n * stride_sn + philox_ptrs = philox_base_ptrs + start_n * stride_sn + if tl_DROPOUT_USE_PYTORCH: + dropout_mask = tl.load(dropout_mask_ptrs, mask=qk_mask) + else: + rng_output = tl.rand( + philox_seed, philox_ptrs + ) # TODO: use tl.randint for better performance + dropout_mask = rng_output > dropout_p + if tl_DROPOUT_DUMP: + tl.store(dropout_mask_ptrs, dropout_mask, mask=qk_mask) + + # return scores with negative values for dropped vals + sd_mask = tl.where(dropout_mask, p, -p) + tl.store(sd_mask_ptrs, sd_mask, mask=qk_mask) + + # apply dropout mask in place + p = tl.where(dropout_mask, p, 0.0) + elif RETURN_SCORES: + # NOTE: the returned score is not the same as the reference because we need to adjust as we find new maxes per block. We are not doing that + sd_mask_ptrs = sd_mask_base_ptrs + start_n * stride_sn + tl.store(sd_mask_ptrs, p, mask=qk_mask) + + # -- update output accumulator -- + # alpha is an adjustment factor for acc and li as we loop and find new maxes + # store the diff in maxes to adjust acc and li as we discover new maxes + m_diff = tl.where(m_ij == float("-inf"), float("-inf"), m_i - m_ij) + if USE_EXP2: + alpha = tl.math.exp2(m_diff * RCP_LN2) + else: + alpha = tl.math.exp(m_diff) + acc = acc * alpha[:, None] + if not PRE_LOAD_V: + v = ( + tl.load(v_ptrs, mask=v_mask, other=v_mask_other) + if PADDED_HEAD_V + else tl.load(v_ptrs) + ) + + # -- update m_i and l_i + l_i = l_i * alpha + l_ij + m_i = m_ij + + if IS_FP8: + if FP8_P_DESCALE: + scale_p, descale_p = compute_fp8_scaling_factors(p, FP8_MAX) + acc += ( + tl.dot((p * scale_p).to(v.type.element_ty), v) + * descale_p + * v_descale + ) + else: + acc += tl.dot(p.to(v.type.element_ty), v) * v_descale + else: + acc += tl.dot(p.to(v.type.element_ty), v) + + return acc, l_i, m_i + + +@triton.jit +def _attn_fwd_mask( + acc, + l_i, + m_i, + q, + k_base_ptrs, + v_base_ptrs, + bias_base_ptrs, + stride_kn, + stride_vk, + stride_bn, + stride_sn, + start_m, + seqlen_k, + seqlen_q, + dropout_p, + philox_seed, + philox_base_ptrs, + sd_mask_base_ptrs, + dropout_mask_base_ptrs, + offs_m, + offs_n, + offs_d_qk, + offs_d_v, + block_min, + block_max, + n_extra_tokens, + alibi_slope, + q_descale, + k_descale, + v_descale, + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, + FP8_P_DESCALE: tl.constexpr, + IS_CAUSAL: tl.constexpr, + BLOCK_M: tl.constexpr, + BLOCK_DMODEL_QK: tl.constexpr, + BLOCK_DMODEL_V: tl.constexpr, + BLOCK_N: tl.constexpr, + PRE_LOAD_V: tl.constexpr, + ENABLE_DROPOUT: tl.constexpr, + PADDED_HEAD_QK: tl.constexpr, + PADDED_HEAD_V: tl.constexpr, + ACTUAL_BLOCK_DMODEL_QK: tl.constexpr, + ACTUAL_BLOCK_DMODEL_V: tl.constexpr, + SM_SCALE: tl.constexpr, + USE_ALIBI: tl.constexpr, + USE_EXP2: tl.constexpr, + RETURN_SCORES: tl.constexpr, + USE_SLIDING_WINDOW: tl.constexpr, + WINDOW_SIZE_LEFT: tl.constexpr, + WINDOW_SIZE_RIGHT: tl.constexpr, + ACCUMULATOR_TYPE, +): + if USE_EXP2: + RCP_LN2: tl.constexpr = 1.4426950408889634 + + # seqlen diff + seqlen_delta_qk = seqlen_k - seqlen_q + + # loop over k, v, and update accumulator + for start_n in range(block_min, block_max, BLOCK_N): + # get ptrs + k_ptrs = k_base_ptrs + start_n * stride_kn + v_ptrs = v_base_ptrs + start_n * stride_vk + + # For padded blocks, we will overrun the tensor size if + # we load all BLOCK_N. For others, the blocks are all within range. + kv_offs_n = start_n + tl.arange(0, BLOCK_N) + k_mask = kv_offs_n[None, :] < seqlen_k + v_mask = kv_offs_n[:, None] < seqlen_k + if PADDED_HEAD_QK: + k_mask = k_mask & (offs_d_qk[:, None] < ACTUAL_BLOCK_DMODEL_QK) + if PADDED_HEAD_V: + v_mask = v_mask & (offs_d_v[None, :] < ACTUAL_BLOCK_DMODEL_V) + + # load k and if preload_v then v + k = tl.load(k_ptrs, mask=k_mask, other=0.0) + if PRE_LOAD_V: + v = tl.load(v_ptrs, mask=v_mask, other=0.0) + + # setup qk accumlator + qk = tl.zeros([BLOCK_M, BLOCK_N], dtype=ACCUMULATOR_TYPE) + + # We start from end of seqlen_k so only the first iteration would need + # to be checked for padding if it is not a multiple of block_n + # TODO: This can be optimized to only be true for the padded block. + # If this is the last block / iteration, we want to + # mask if the sequence length is not a multiple of block size + # a solution is to always do BLOCK_M // BLOCK_N + 1 steps if not is_modulo_mn. + # last step might get wasted but that is okay. check if this masking works For + # that case. + if (n_extra_tokens != 0) and (start_n + BLOCK_N == block_max): + boundary_m = tl.full([BLOCK_M], seqlen_k, dtype=tl.int32) + size_n = start_n + offs_n[None, :] + mask = size_n < boundary_m[:, None] + qk = tl.where(mask, qk, float("-inf")) + + # -- compute qk ---- + if IS_FP8: + qk += tl.dot(q, k) * q_descale * k_descale + else: + qk += tl.dot(q, k) + qk_scaled = qk * SM_SCALE + + if USE_ALIBI: + # compute the global position of each token within the sequence + q_offs_m = start_m * BLOCK_M + tl.arange(0, BLOCK_M) + alibi_block = compute_alibi_block( + alibi_slope, seqlen_q, seqlen_k, q_offs_m, kv_offs_n + ) + qk_scaled += alibi_block + + if USE_SLIDING_WINDOW: + if IS_CAUSAL: + # ========== CAUSAL SLIDING WINDOW MASKING ========== + # For causal sliding window, we need to apply both constraints: + # 1. Causal: col_idx <= row_idx + (seqlen_k - seqlen_q) + # 2. Sliding window: row_idx - window_left <= col_idx <= row_idx + window_right + + # Get positions + row_idx = offs_m # Query positions + col_idx = kv_offs_n # Key positions + + # Expand for broadcasting + row_idx_expanded = row_idx[:, None] # [BLOCK_M, 1] + col_idx_expanded = col_idx[None, :] # [1, BLOCK_N] + + # Apply causal constraint: can only attend to positions before or at the diagonal + causal_offset = seqlen_k - seqlen_q + causal_mask = col_idx_expanded > (row_idx_expanded + causal_offset) + + # Apply sliding window constraint + if WINDOW_SIZE_LEFT < 0: + # Only right window constraint + window_mask = col_idx_expanded > ( + row_idx_expanded + causal_offset + WINDOW_SIZE_RIGHT + ) + else: + # Both left and right window constraints + # Adjust window bounds by causal offset + left_bound = row_idx_expanded + causal_offset - WINDOW_SIZE_LEFT + right_bound = row_idx_expanded + causal_offset + WINDOW_SIZE_RIGHT + + # Can't attend to positions outside the window + window_mask = (col_idx_expanded < left_bound) | ( + col_idx_expanded > right_bound + ) + + # Final mask is the union of both constraints (True = cannot attend) + mask = causal_mask | window_mask + + # Apply mask + qk_scaled = tl.where(mask, float("-inf"), qk_scaled) + else: + # ========== NON-CAUSAL SLIDING WINDOW MASKING ========== + # Exactly matching reference construct_local_mask: + # row_idx = query positions, col_idx = key positions + # sk = seqlen_k, sq = seqlen_q + + # Get positions + row_idx = offs_m # Query positions + col_idx = kv_offs_n # Key positions + + # sk and sq from reference (no padding masks in this test) + sk = seqlen_k + sq = seqlen_q + + # Expand for broadcasting + row_idx_expanded = row_idx[:, None] # [BLOCK_M, 1] + col_idx_expanded = col_idx[None, :] # [1, BLOCK_N] + + # Reference logic for mask computation + if WINDOW_SIZE_LEFT < 0: + # Reference: return col_idx > row_idx + sk - sq + window_size[1] + mask = col_idx_expanded > ( + row_idx_expanded + sk - sq + WINDOW_SIZE_RIGHT + ) + else: + # Reference: + # sk = torch.full_like(col_idx, seqlen_k) if key_padding_mask is None else sk + # return torch.logical_or( + # col_idx > torch.minimum(row_idx + sk - sq + window_size[1], sk), + # col_idx < row_idx + sk - sq - window_size[0], + # ) + # Create sk tensor with proper shape for broadcasting + # sk represents the key sequence length, which should be compared per column + sk_full = tl.full((1, BLOCK_N), sk, dtype=tl.int32) + + # Compute boundaries + right_bound_val = row_idx_expanded + sk - sq + WINDOW_SIZE_RIGHT + right_bound = tl.minimum(right_bound_val, sk_full) + left_bound = row_idx_expanded + sk - sq - WINDOW_SIZE_LEFT + + # Mask where True = cannot attend (matching reference) + mask = (col_idx_expanded > right_bound) | ( + col_idx_expanded < left_bound + ) + + # Apply mask (set to -inf where mask is True) + qk_scaled = tl.where(mask, float("-inf"), qk_scaled) + else: + if IS_CAUSAL: + causal_boundary = start_n + offs_n - seqlen_delta_qk + causal_mask = offs_m[:, None] >= causal_boundary[None, :] + qk_scaled = tl.where(causal_mask, qk_scaled, float("-inf")) + + # compute qk mask + qk_mask = (offs_m[:, None] < seqlen_q) & (kv_offs_n[None, :] < seqlen_k) + + # compute bias + if bias_base_ptrs is not None: + bias_ptrs = bias_base_ptrs + start_n * stride_bn + bias = tl.load(bias_ptrs, mask=qk_mask, other=0.0) + qk_scaled += bias + + # get max scores so far + m_ij = tl.maximum(m_i, tl.max(qk_scaled, 1)) + + # scale and subtract max + # IMPORTANT: Handle the case where all values are -inf + # When m_ij = -inf and qk_scaled = -inf, subtraction gives NaN + # We need to handle this explicitly + if USE_SLIDING_WINDOW: + # Check if this block has any valid values (m_ij != -inf) + # For rows where everything is -inf, set q_shifted to -inf (not NaN) + q_shifted = tl.where( + m_ij[:, None] == float("-inf"), float("-inf"), qk_scaled - m_ij[:, None] + ) + else: + q_shifted = qk_scaled - m_ij[:, None] + + # Compute scaled QK and softmax probabilities + if USE_EXP2: + p = tl.math.exp2(q_shifted * RCP_LN2) + else: + p = tl.math.exp(q_shifted) + + # CAVEAT: Must update l_ij before applying dropout + l_ij = tl.sum(p, 1) + if ENABLE_DROPOUT: + dropout_mask_ptrs = dropout_mask_base_ptrs + start_n * stride_sn + sd_mask_ptrs = sd_mask_base_ptrs + start_n * stride_sn + philox_ptrs = philox_base_ptrs + start_n * stride_sn + if tl_DROPOUT_USE_PYTORCH: + dropout_mask = tl.load(dropout_mask_ptrs, mask=qk_mask) + else: + rng_output = tl.rand( + philox_seed, philox_ptrs + ) # TODO: use tl.randint for better performance + dropout_mask = rng_output > dropout_p + if tl_DROPOUT_DUMP: + tl.store(dropout_mask_ptrs, dropout_mask, mask=qk_mask) + + # return scores with negative values for dropped vals + sd_mask = tl.where(dropout_mask, p, -p) + tl.store(sd_mask_ptrs, sd_mask, mask=qk_mask) + + # apply dropout mask in place + p = tl.where(dropout_mask, p, 0.0) + elif RETURN_SCORES: + # NOTE: the returned score is not the same as the reference because we need to adjust as we find new maxes per block. We are not doing that + sd_mask_ptrs = sd_mask_base_ptrs + start_n * stride_sn + tl.store(sd_mask_ptrs, p, mask=qk_mask) + + # -- update output accumulator -- + # alpha is an adjustment factor for acc and li as we loop and find new maxes + # store the diff in maxes to adjust acc and li as we discover new maxes + m_diff = tl.where(m_ij == float("-inf"), float("-inf"), m_i - m_ij) + if USE_EXP2: + alpha = tl.math.exp2(m_diff * RCP_LN2) + else: + alpha = tl.math.exp(m_diff) + acc = acc * alpha[:, None] + if not PRE_LOAD_V: + v = tl.load(v_ptrs, mask=v_mask, other=0.0) + + # -- update m_i and l_i + l_i = l_i * alpha + l_ij + m_i = m_ij + + if IS_FP8: + if FP8_P_DESCALE: + scale_p, descale_p = compute_fp8_scaling_factors(p, FP8_MAX) + acc += ( + tl.dot((p * scale_p).to(v.type.element_ty), v) + * descale_p + * v_descale + ) + else: + acc += tl.dot(p.to(v.type.element_ty), v) * v_descale + else: + acc += tl.dot(p.to(v.type.element_ty), v) + + return acc, l_i, m_i + + +@triton.jit +def compute_window_bounds( + q_start, + q_end, + diag, + seqlen_k, + WINDOW_SIZE_LEFT: tl.constexpr, + WINDOW_SIZE_RIGHT: tl.constexpr, + IS_CAUSAL: tl.constexpr, +): + """Calculate the window boundaries for a query block.""" + # Left boundary + if WINDOW_SIZE_LEFT < 0: + left_min = 0 + left_max = 0 + else: + left_min = tl.maximum(0, q_start + diag - WINDOW_SIZE_LEFT) + left_max = tl.maximum(0, q_end + diag - WINDOW_SIZE_LEFT) + + # Right boundary + if IS_CAUSAL: + # Causal cap: col ≤ row + diag + right_min = tl.minimum(seqlen_k - 1, q_start + diag) + right_max = tl.minimum(seqlen_k - 1, q_end + diag) + else: + if WINDOW_SIZE_RIGHT < 0: + right_min = tl.minimum(seqlen_k - 1, q_start + diag + WINDOW_SIZE_RIGHT) + right_max = tl.minimum(seqlen_k - 1, q_end + diag + WINDOW_SIZE_RIGHT) + else: + # Non-causal doesn't have the diagonal constraint + right_min = tl.minimum(seqlen_k - 1, q_start + diag + WINDOW_SIZE_RIGHT) + right_max = tl.minimum(seqlen_k - 1, q_end + diag + WINDOW_SIZE_RIGHT) + + return left_min, left_max, right_min, right_max + + +@triton.jit +def classify_window_blocks( + left_min, left_max, right_min, right_max, BLOCK_N: tl.constexpr +): + """Classify blocks based on window boundaries.""" + # First and last blocks that have ANY overlap with window + first_block = left_min // BLOCK_N + last_block = right_max // BLOCK_N + + # First block that is FULLY visible for all rows in Q block + full_left_block = left_max // BLOCK_N + (left_max % BLOCK_N != 0) + clipped_left = tl.minimum(full_left_block, last_block + 1) + + # Last block that is FULLY visible for all rows in Q block + last_full_block_candidate = right_min // BLOCK_N + if (last_full_block_candidate + 1) * BLOCK_N - 1 > right_min: + last_full_block_candidate -= 1 + full_right_block = tl.maximum(last_full_block_candidate, clipped_left - 1) + + # Calculate counts + n_front_skip_blocks = first_block + n_front_masked_blocks = tl.maximum(0, clipped_left - first_block) + n_full_blocks = tl.maximum(0, full_right_block - clipped_left + 1) + n_back_masked_blocks = tl.maximum(0, last_block - full_right_block) + + return ( + n_front_skip_blocks, + n_front_masked_blocks, + n_full_blocks, + n_back_masked_blocks, + clipped_left, + ) # Return clipped_left for padded block handling + + +@triton.jit +def handle_padded_last_block( + n_extra_tokens, + last_block, + total_k_blocks, + clipped_left, + n_front_masked_blocks, + n_full_blocks, + n_back_masked_blocks, +): + """Ensure a padded last K-block is never classified as 'full'. + + We move the padded last block (if visible) into the back-masked bucket. + If it's already back-masked, we do nothing. If it was counted in the + front-masked range, we decrement front-masked; if it was counted as full, + we decrement full. Then we increment back-masked. + """ + padded_last_k = (n_extra_tokens != 0) & (last_block == total_k_blocks - 1) + + if padded_last_k: + # current 'full' range right edge + full_right_block = clipped_left + n_full_blocks - 1 + + # If last_block is already beyond full_right_block, it's already in back-masked → nothing to do + last_already_back_masked = last_block > full_right_block + if not last_already_back_masked: + # If the window starts past last_block, it was counted in front-masked + if clipped_left > last_block: + n_front_masked_blocks = tl.maximum(0, n_front_masked_blocks - 1) + else: + # Otherwise it was counted 'full' → move it out of full + n_full_blocks = tl.maximum(0, n_full_blocks - 1) + # In both cases we need one more back-masked block + n_back_masked_blocks = n_back_masked_blocks + 1 + + return n_front_masked_blocks, n_full_blocks, n_back_masked_blocks + + +@triton.jit +def compute_padding_info(seqlen_k, BLOCK_N: tl.constexpr): + """Calculate padding information for the last K block.""" + # check if we will need to do masking due either BLOCK_N being bigger than seqlen_k or seqlen_k not being a factor of BLOCK_N + # n_extra_tokens = 10 % 4 = 2 + # This means the last K block has 2 valid tokens and 2 padding positions + # K blocks visualization: + # Block 0 Block 1 Block 2 (last) + # K0 K1 K2 K3 K4 K5 K6 K7 K8 K9 ?? ?? + # ↑---------↑ ↑---------↑ ↑---↑ ↑---↑ + # full block full block valid pad + if seqlen_k < BLOCK_N: + n_extra_tokens = BLOCK_N - seqlen_k + elif seqlen_k % BLOCK_N: + n_extra_tokens = seqlen_k % BLOCK_N + else: + n_extra_tokens = 0 + return n_extra_tokens + + +@triton.jit +def compute_block_masking( + seqlen_k, + seqlen_q, + start_m, + IS_CAUSAL: tl.constexpr, + USE_SLIDING_WINDOW: tl.constexpr, + WINDOW_SIZE_LEFT: tl.constexpr, + WINDOW_SIZE_RIGHT: tl.constexpr, + BLOCK_M: tl.constexpr, + BLOCK_N: tl.constexpr, +): + """ + Classify K blocks for attention computation with sliding window support. + + Returns: + - n_front_skip_blocks: Blocks completely before the window + - n_front_masked_blocks: Blocks partially overlapping window front + - n_full_blocks: Blocks completely inside the window + - n_back_masked_blocks: Blocks partially overlapping window back + - n_extra_tokens: Padding tokens in last K block + """ + + # common + q_start = start_m * BLOCK_M + q_end = tl.minimum((start_m + 1) * BLOCK_M - 1, seqlen_q - 1) + diag = seqlen_k - seqlen_q + total_k_blocks = tl.cdiv(seqlen_k, BLOCK_N) + n_extra_tokens = compute_padding_info(seqlen_k, BLOCK_N) + + if USE_SLIDING_WINDOW: + # get window bounds + left_min, left_max, right_min, right_max = compute_window_bounds( + q_start, + q_end, + diag, + seqlen_k, + WINDOW_SIZE_LEFT, + WINDOW_SIZE_RIGHT, + IS_CAUSAL, + ) + + # window vanishes → early exit + if right_max < left_min: + return 0, 0, 0, 0, n_extra_tokens + + # classify blocks + ( + n_front_skip_blocks, + n_front_masked_blocks, + n_full_blocks, + n_back_masked_blocks, + clipped_left, + ) = classify_window_blocks(left_min, left_max, right_min, right_max, BLOCK_N) + + # handle padded last block if needed + if n_extra_tokens != 0: + last_block = right_max // BLOCK_N + n_front_masked_blocks, n_full_blocks, n_back_masked_blocks = ( + handle_padded_last_block( + n_extra_tokens, + last_block, + total_k_blocks, + clipped_left, + n_front_masked_blocks, + n_full_blocks, + n_back_masked_blocks, + ) + ) + return ( + n_front_skip_blocks, + n_front_masked_blocks, + n_full_blocks, + n_back_masked_blocks, + n_extra_tokens, + ) + else: + if IS_CAUSAL: + # ========== CAUSAL MODE: Classify K Blocks ========== + # Calculate causal boundary for this Q block + # [K0 K1 K2 K3] [K4 K5 K6 K7] [K8 K9 ?? ??] + # Q0-Q3: [ 1 0 0 0] [ 0 0 0 0] [ 0 0 -- --] ← Q0 + # [ 1 1 0 0] [ 0 0 0 0] [ 0 0 -- --] ← Q1 + # [ 1 1 1 0] [ 0 0 0 0] [ 0 0 -- --] ← Q2 + # [ 1 1 1 1] [ 1 1 0 0] [ 0 0 -- --] ← Q3 + # ↑ can see up to K5 + # + # Q4-Q7: [ 1 1 1 1] [ 1 1 1 0] [ 0 0 -- --] ← Q4 + # [ 1 1 1 1] [ 1 1 1 1] [ 0 0 -- --] ← Q5 + # [ 1 1 1 1] [ 1 1 1 1] [ 1 0 -- --] ← Q6 + # [ 1 1 1 1] [ 1 1 1 1] [ 1 1 -- --] ← Q7 + + # ------------------------------------------------------------ + # 1. figure out, in tokens, the right-most K position + # this Q-block may attend to + # ------------------------------------------------------------ + k_max_token = q_end + diag # last visible K index + + # this Q-block is entirely above the diagonal ⇒ nothing to do + if k_max_token < 0: + return 0, 0, 0, 0, n_extra_tokens + + k_max_token = tl.minimum(k_max_token, seqlen_k - 1) + + # ------------------------------------------------------------ + # 2. translate token indices into K-block indices + # ------------------------------------------------------------ + last_visible_k_block = k_max_token // BLOCK_N + n_visible_k_blocks = tl.minimum(last_visible_k_block + 1, total_k_blocks) + + # ------------------------------------------------------------ + # 3. classify those visible blocks + # – we *never* skip or mask blocks in front, because causal + # attention always starts at K0 + # – the back side can require several masked blocks: + # • intersection of the causal diagonal with K-grid + # (at most ⌈BLOCK_M / BLOCK_N⌉ blocks) + # • plus one extra block if this Q-block stops in the + # middle of a K-block or the last K-block is padded + # ------------------------------------------------------------ + padded_last_k = n_extra_tokens != 0 + is_modulo_mn = (not padded_last_k) & (seqlen_q % BLOCK_M == 0) + + n_back_masked_blocks = BLOCK_M // BLOCK_N + tl.where(is_modulo_mn, 0, 1) + n_back_masked_blocks = tl.minimum(n_back_masked_blocks, n_visible_k_blocks) + + n_front_skip_blocks = 0 # causal never skips the left side + n_front_masked_blocks = 0 # ditto + n_full_blocks = n_visible_k_blocks - n_back_masked_blocks + else: + # ========== NON-CAUSAL MODE ========== + # Without causal mask, all positions can attend to all positions + # Only need to handle the padding in the last block + # [K0 K1 K2 K3] [K4 K5 K6 K7] [K8 K9 ?? ??] + # Q0-Q3: [ 1 1 1 1] [ 1 1 1 1] [ 1 1 -∞ -∞] + # [ 1 1 1 1] [ 1 1 1 1] [ 1 1 -∞ -∞] + # [ 1 1 1 1] [ 1 1 1 1] [ 1 1 -∞ -∞] + # [ 1 1 1 1] [ 1 1 1 1] [ 1 1 -∞ -∞] + # + # Q4-Q7: [ 1 1 1 1] [ 1 1 1 1] [ 1 1 -∞ -∞] + # [ 1 1 1 1] [ 1 1 1 1] [ 1 1 -∞ -∞] + # [ 1 1 1 1] [ 1 1 1 1] [ 1 1 -∞ -∞] + # [ 1 1 1 1] [ 1 1 1 1] [ 1 1 -∞ -∞] + + n_front_skip_blocks = 0 # never skips the left side + n_front_masked_blocks = 0 # ditto + if n_extra_tokens != 0: + n_back_masked_blocks = 1 # Last block needs padding mask + n_full_blocks = total_k_blocks - 1 + else: + n_back_masked_blocks = 0 # All blocks are aligned + n_full_blocks = total_k_blocks + + return ( + n_front_skip_blocks, + n_front_masked_blocks, + n_full_blocks, + n_back_masked_blocks, + n_extra_tokens, + ) + + +@triton.autotune( + configs=fwd_prefill_autotune_configs, + key=fwd_prefill_autotune_keys, + use_cuda_graph=True, +) +@triton.jit +def attn_fwd( + Q, + K, + V, + bias, + Q_Descale, + K_Descale, + V_Descale, + stride_q_descale_z, + stride_k_descale_z, + stride_v_descale_z, + SM_SCALE: tl.constexpr, + LSE, + Out, + stride_qz, + stride_qh, + stride_qm, + stride_qk, + stride_kz, + stride_kh, + stride_kn, + stride_kk, + stride_vz, + stride_vh, + stride_vk, + stride_vn, + stride_oz, + stride_oh, + stride_om, + stride_on, + stride_bz, + stride_bh, + stride_bm, + stride_bn, + stride_az, + stride_ah, + stride_sz, + stride_sh, + stride_sm, + stride_sn, + stride_lse_z, + stride_lse_h, + stride_lse_m, + cu_seqlens_q, + cu_seqlens_k, + seqused_q, + seqused_k, # Add seqused parameters + dropout_p, + philox_seed, + philox_offset_base, + sd_mask, + dropout_mask, + alibi_slopes, + HQ: tl.constexpr, + HK: tl.constexpr, + ACTUAL_BLOCK_DMODEL_QK: tl.constexpr, + ACTUAL_BLOCK_DMODEL_V: tl.constexpr, + MAX_SEQLENS_Q: tl.constexpr, + MAX_SEQLENS_K: tl.constexpr, + IS_VARLEN: tl.constexpr, + IS_CAUSAL: tl.constexpr, + USE_SLIDING_WINDOW: tl.constexpr, + WINDOW_SIZE_LEFT: tl.constexpr, + WINDOW_SIZE_RIGHT: tl.constexpr, + BLOCK_M: tl.constexpr, + BLOCK_DMODEL_QK: tl.constexpr, + BLOCK_DMODEL_V: tl.constexpr, + BLOCK_N: tl.constexpr, + PRE_LOAD_V: tl.constexpr, + USE_BIAS: tl.constexpr, + ENABLE_DROPOUT: tl.constexpr, + RETURN_SCORES: tl.constexpr, + NEEDS_SDMASK: tl.constexpr, + USE_ALIBI: tl.constexpr, + USE_EXP2: tl.constexpr, + IS_FP8: tl.constexpr, + FP8_MAX: tl.constexpr, + FP8_P_DESCALE: tl.constexpr, + USE_SEQUSED: tl.constexpr, +): + # set params + ACCUMULATOR_TYPE = tl.float32 + + # compute offsets + off_z = tl.program_id(0) + off_h_q = tl.program_id(1) + start_m = tl.program_id(2) + # If MQA / GQA, set the K and V head offsets appropriately. + GROUP_SIZE: tl.constexpr = HQ // HK + if GROUP_SIZE != 1: + off_h_k = off_h_q // GROUP_SIZE + else: + off_h_k = off_h_q + # Determine if we need to mask the heads + PADDED_HEAD_QK: tl.constexpr = ACTUAL_BLOCK_DMODEL_QK != BLOCK_DMODEL_QK + PADDED_HEAD_V: tl.constexpr = ACTUAL_BLOCK_DMODEL_V != BLOCK_DMODEL_V + + offs_m = start_m * BLOCK_M + tl.arange(0, BLOCK_M) + offs_n = tl.arange(0, BLOCK_N) + offs_d_qk = tl.arange(0, BLOCK_DMODEL_QK) + offs_d_v = tl.arange(0, BLOCK_DMODEL_V) + + # handle seqlen + if IS_VARLEN: + cu_seqlens_q_start = tl.load(cu_seqlens_q + off_z) + cu_seqlens_q_end = tl.load(cu_seqlens_q + off_z + 1) + + # If seqused is provided, use it to limit the actual sequence length + if USE_SEQUSED: + actual_seqlen_q = ( + tl.load(seqused_q + off_z) + if seqused_q is not None + else cu_seqlens_q_end - cu_seqlens_q_start + ) + seqlen_q = tl.minimum( + actual_seqlen_q, cu_seqlens_q_end - cu_seqlens_q_start + ) + else: + seqlen_q = cu_seqlens_q_end - cu_seqlens_q_start + + # we have a one-size-fits-all grid in id(0). Some seqlens might be too small for all start_m so for those we return early. + if start_m * BLOCK_M > seqlen_q: + return + cu_seqlens_k_start = tl.load(cu_seqlens_k + off_z) + cu_seqlens_k_end = tl.load(cu_seqlens_k + off_z + 1) + + # If seqused is provided, use it to limit the actual sequence length for keys + if USE_SEQUSED: + actual_seqlen_k = ( + tl.load(seqused_k + off_z) + if seqused_k is not None + else cu_seqlens_k_end - cu_seqlens_k_start + ) + seqlen_k = tl.minimum( + actual_seqlen_k, cu_seqlens_k_end - cu_seqlens_k_start + ) + else: + seqlen_k = cu_seqlens_k_end - cu_seqlens_k_start + else: + cu_seqlens_q_start = 0 + cu_seqlens_k_start = 0 + seqlen_q = MAX_SEQLENS_Q + seqlen_k = MAX_SEQLENS_K + + # Load scale factors if IS_FP8. + if IS_FP8: + # For MQA/GQA (GROUP_SIZE != 1), q_descale uses the same indexing as k/v (off_h_k) + # For MHA (GROUP_SIZE == 1), q_descale uses off_h_q (same as off_h_k) + if GROUP_SIZE != 1: + q_descale = tl.load( + Q_Descale + off_z * stride_q_descale_z + off_h_k + ) # MQA/GQA: broadcast using k/v head index + else: + q_descale = tl.load( + Q_Descale + off_z * stride_q_descale_z + off_h_q + ) # MHA: use q head index + k_descale = tl.load(K_Descale + off_z * stride_k_descale_z + off_h_k) + v_descale = tl.load(V_Descale + off_z * stride_v_descale_z + off_h_k) + else: + q_descale, k_descale, v_descale = 1.0, 1.0, 1.0 + + # figure out masking pattern + ( + n_front_skip_blocks, + n_front_masked_blocks, + n_full_blocks, + n_back_masked_blocks, + n_extra_tokens, + ) = compute_block_masking( + seqlen_k, + seqlen_q, + start_m, + IS_CAUSAL, + USE_SLIDING_WINDOW, + WINDOW_SIZE_LEFT, + WINDOW_SIZE_RIGHT, + BLOCK_M, + BLOCK_N, + ) + + # ============================================================ + # PROGRAM EARLY EXIT (All K Blocks Skipped) + # ============================================================ + total_visible_blocks = n_front_masked_blocks + n_full_blocks + n_back_masked_blocks + if total_visible_blocks == 0: + """ + No K blocks visible - write zeros and exit. + """ + # Write zeros to output + o_offset = ( + Out + + off_z * stride_oz + + off_h_q * stride_oh + + cu_seqlens_q_start * stride_om + ) + o_ptrs = o_offset + offs_m[:, None] * stride_om + offs_d_v[None, :] * stride_on + o_mask = offs_m[:, None] < seqlen_q + if PADDED_HEAD_V: + o_mask = o_mask & (offs_d_v[None, :] < ACTUAL_BLOCK_DMODEL_V) + tl.store( + o_ptrs, + tl.zeros([BLOCK_M, BLOCK_DMODEL_V], dtype=Out.type.element_ty), + mask=o_mask, + ) + + # Write zeros to LSE + l_ptrs = ( + LSE + + off_z * stride_lse_z + + off_h_q * stride_lse_h + + cu_seqlens_q_start * stride_lse_m + + offs_m * stride_lse_m + ) + tl.store(l_ptrs, tl.zeros([BLOCK_M], dtype=tl.float32), mask=offs_m < seqlen_q) + return + + # ============================================================ + # NORMAL PROCESSING (Some K Blocks Visible) + # ============================================================ + """ + This program has visible K blocks to process. + We'll use two calls to handle different block types efficiently. + """ + + # Initialize for processing + # Compute pointers for all the tensors used in this kernel. + q_offset = ( + Q + off_z * stride_qz + off_h_q * stride_qh + cu_seqlens_q_start * stride_qm + ) + q_ptrs = q_offset + offs_m[:, None] * stride_qm + offs_d_qk[None, :] * stride_qk + k_offset = ( + K + off_z * stride_kz + off_h_k * stride_kh + cu_seqlens_k_start * stride_kn + ) + k_ptrs = k_offset + offs_d_qk[:, None] * stride_kk + offs_n[None, :] * stride_kn + v_offset = ( + V + off_z * stride_vz + off_h_k * stride_vh + cu_seqlens_k_start * stride_vk + ) + v_ptrs = v_offset + offs_n[:, None] * stride_vk + offs_d_v[None, :] * stride_vn + if USE_BIAS: + # Note: this might get large enough to overflow on some configs + bias_offset = off_h_q * stride_bh + bias_ptrs = ( + bias + + bias_offset + + offs_m[:, None] * stride_bm + + offs_n[None, :] * stride_bn + ) + else: + bias_ptrs = None + + if USE_ALIBI: + a_offset = off_z * stride_az + off_h_q * stride_ah + alibi_slope = tl.load(alibi_slopes + a_offset) + else: + alibi_slope = None + + if NEEDS_SDMASK: + sd_mask_offset = ( + sd_mask + off_z * stride_sz + off_h_q * stride_sh + ) # + cu_seqlens_q_start * stride_sm + sd_mask_ptrs = ( + sd_mask_offset + offs_m[:, None] * stride_sm + offs_n[None, :] * stride_sn + ) + else: + sd_mask_ptrs = None + + if ENABLE_DROPOUT: + dropout_mask_offset = ( + dropout_mask + off_z * stride_sz + off_h_q * stride_sh + ) # + cu_seqlens_q_start * stride_sm + dropout_mask_ptrs = ( + dropout_mask_offset + + offs_m[:, None] * stride_sm + + offs_n[None, :] * stride_sn + ) + batch_philox_offset = ( + philox_offset_base + off_z * stride_sz + off_h_q * stride_sh + ) # + cu_seqlens_q_start * stride_sm + philox_ptrs = ( + batch_philox_offset + + offs_m[:, None] * stride_sm + + offs_n[None, :] * stride_sn + ) + else: + dropout_mask_ptrs = None + philox_ptrs = 0 + + # initialize pointer to m and l + m_i = tl.full([BLOCK_M], float("-inf"), dtype=ACCUMULATOR_TYPE) + l_i = tl.full([BLOCK_M], 1.0, dtype=ACCUMULATOR_TYPE) + acc = tl.zeros([BLOCK_M, BLOCK_DMODEL_V], dtype=ACCUMULATOR_TYPE) + + # Q is loaded once at the beginning and shared by all N blocks. + q_ptrs_mask = offs_m[:, None] < seqlen_q + if PADDED_HEAD_QK: + q_ptrs_mask = q_ptrs_mask & (offs_d_qk[None, :] < ACTUAL_BLOCK_DMODEL_QK) + q = tl.load(q_ptrs, mask=q_ptrs_mask, other=0.0) + + # ========== Process MASKED K Blocks in the front ========== + # NOTE: we use USE_SLIDING_WINDOW as guard because the compiler will crash other wise. front masking is only for sliding window so that is fine. + if n_front_masked_blocks > 0 and USE_SLIDING_WINDOW: + block_min = n_front_skip_blocks * BLOCK_N + block_max = (n_front_skip_blocks + n_front_masked_blocks) * BLOCK_N + + acc, l_i, m_i = _attn_fwd_mask( + acc, + l_i, + m_i, + q, + k_ptrs, + v_ptrs, + bias_ptrs, + stride_kn, + stride_vk, + stride_bn, + stride_sn, + start_m, + seqlen_k, + seqlen_q, + dropout_p, + philox_seed, + philox_ptrs, + sd_mask_ptrs, + dropout_mask_ptrs, + offs_m, + offs_n, + offs_d_qk, + offs_d_v, + block_min, # Start of front masked blocks + block_max, # End of front masked blocks + 0, # n_extra_tokens (0 for front blocks, only relevant for last block) + alibi_slope, + q_descale, + k_descale, + v_descale, + IS_FP8, + FP8_MAX, + FP8_P_DESCALE, + IS_CAUSAL, + BLOCK_M, + BLOCK_DMODEL_QK, + BLOCK_DMODEL_V, + BLOCK_N, + PRE_LOAD_V, + ENABLE_DROPOUT, + PADDED_HEAD_QK, + PADDED_HEAD_V, + ACTUAL_BLOCK_DMODEL_QK, + ACTUAL_BLOCK_DMODEL_V, + SM_SCALE, + USE_ALIBI=USE_ALIBI, + USE_EXP2=USE_EXP2, + RETURN_SCORES=RETURN_SCORES, + USE_SLIDING_WINDOW=USE_SLIDING_WINDOW, + WINDOW_SIZE_LEFT=WINDOW_SIZE_LEFT, + WINDOW_SIZE_RIGHT=WINDOW_SIZE_RIGHT, + ACCUMULATOR_TYPE=ACCUMULATOR_TYPE, + ) + + # ========== Process FULL K Blocks (Fast Path) ========== + if n_full_blocks > 0: + block_min = (n_front_skip_blocks + n_front_masked_blocks) * BLOCK_N + block_max = ( + n_front_skip_blocks + n_front_masked_blocks + n_full_blocks + ) * BLOCK_N + + acc, l_i, m_i = _attn_fwd_no_mask( + acc, + l_i, + m_i, + q, + k_ptrs, + v_ptrs, + bias_ptrs, + stride_kn, + stride_vk, + stride_bn, + stride_sn, + start_m, + seqlen_k, + seqlen_q, + dropout_p, + philox_seed, + philox_ptrs, + sd_mask_ptrs, + dropout_mask_ptrs, + offs_m, + offs_n, + offs_d_qk, + offs_d_v, + block_min, # Start of range: 0 + block_max, # End of range: n_full_blocks * BLOCK_N + alibi_slope, + q_descale, + k_descale, + v_descale, + IS_FP8, + FP8_MAX, + FP8_P_DESCALE, + BLOCK_M, + BLOCK_DMODEL_QK, + BLOCK_DMODEL_V, + BLOCK_N, + PRE_LOAD_V, + ENABLE_DROPOUT, + PADDED_HEAD_QK, + PADDED_HEAD_V, + ACTUAL_BLOCK_DMODEL_QK, + ACTUAL_BLOCK_DMODEL_V, + SM_SCALE, + USE_ALIBI=USE_ALIBI, + USE_EXP2=USE_EXP2, + RETURN_SCORES=RETURN_SCORES, + ACCUMULATOR_TYPE=ACCUMULATOR_TYPE, + ) + + # ========== Process MASKED K Blocks in the back ========== + if n_back_masked_blocks > 0: + block_min = ( + n_front_skip_blocks + n_front_masked_blocks + n_full_blocks + ) * BLOCK_N + block_max = ( + n_front_skip_blocks + + n_front_masked_blocks + + n_full_blocks + + n_back_masked_blocks + ) * BLOCK_N + + acc, l_i, m_i = _attn_fwd_mask( + acc, + l_i, + m_i, + q, + k_ptrs, + v_ptrs, + bias_ptrs, + stride_kn, + stride_vk, + stride_bn, + stride_sn, + start_m, + seqlen_k, + seqlen_q, + dropout_p, + philox_seed, + philox_ptrs, + sd_mask_ptrs, + dropout_mask_ptrs, + offs_m, + offs_n, + offs_d_qk, + offs_d_v, + block_min, # Start of range: n_full_blocks * BLOCK_N + block_max, # End of range: n_visible_k_blocks * BLOCK_N + n_extra_tokens, # Padding tokens in last block + alibi_slope, + q_descale, + k_descale, + v_descale, + IS_FP8, + FP8_MAX, + FP8_P_DESCALE, + IS_CAUSAL, # Use actual causal flag + BLOCK_M, + BLOCK_DMODEL_QK, + BLOCK_DMODEL_V, + BLOCK_N, + PRE_LOAD_V, + ENABLE_DROPOUT, + PADDED_HEAD_QK, + PADDED_HEAD_V, + ACTUAL_BLOCK_DMODEL_QK, + ACTUAL_BLOCK_DMODEL_V, + SM_SCALE, + USE_ALIBI=USE_ALIBI, + USE_EXP2=USE_EXP2, + RETURN_SCORES=RETURN_SCORES, + USE_SLIDING_WINDOW=USE_SLIDING_WINDOW, + WINDOW_SIZE_LEFT=WINDOW_SIZE_LEFT, + WINDOW_SIZE_RIGHT=WINDOW_SIZE_RIGHT, + ACCUMULATOR_TYPE=ACCUMULATOR_TYPE, + ) + + # ============================================================ + # EPILOGUE + # ============================================================ + # This helps the compiler do Newton Raphson on l_i vs on acc which is much larger. + # Instead of directly computing 1/l_i which can be inf, + # we check for the invalid case first + if USE_SLIDING_WINDOW: + # For rows where m_i is still -inf, no keys were valid + # Set l_i to 1.0 to avoid division by zero (acc is already 0) + invalid_mask = m_i == float("-inf") + l_i_safe = tl.where(invalid_mask, 1.0, l_i) + l_recip = 1 / l_i_safe[:, None] + else: + invalid_mask = None + l_recip = 1 / l_i[:, None] + acc = acc * l_recip + if ENABLE_DROPOUT: + dropout_scale = 1 / (1 - dropout_p) + acc = acc * dropout_scale + + # compute log-sum-exp + if USE_EXP2: + RCP_LN2: tl.constexpr = 1.4426950408889634 + LN2: tl.constexpr = 0.6931471824645996 + # compute log-sum-exp in base 2 units + mi_base2 = m_i * RCP_LN2 + # For invalid rows, log(l_i) would be -inf, but we want LSE to be -inf + # So we handle this case explicitly + if USE_SLIDING_WINDOW: + log_l_i = tl.where(invalid_mask, 0.0, tl.math.log2(l_i)) + softmax_lse = mi_base2 + log_l_i + # Ensure invalid rows have LSE = -inf + softmax_lse = tl.where(invalid_mask, float("-inf"), softmax_lse) + else: + softmax_lse = mi_base2 + tl.math.log2(l_i) + # convert back to natural units + softmax_lse *= LN2 + else: + if USE_SLIDING_WINDOW: + log_l_i = tl.where(invalid_mask, 0.0, tl.math.log(l_i)) + softmax_lse = m_i + log_l_i + softmax_lse = tl.where(invalid_mask, float("-inf"), softmax_lse) + else: + softmax_lse = m_i + tl.math.log(l_i) + + # handle masking edge cases + if USE_SLIDING_WINDOW: + if IS_CAUSAL: + pass + else: + pass + else: + if IS_CAUSAL: + # When seqlen_q > seqlen_k, some rows are completely above the causal diagonal + # These rows have all -inf attention scores, resulting in NaN after softmax + # e.g. + # Q length: 6, K length: 4 + # Causal mask (X = can attend, . = cannot): + # K0 K1 K2 K3 + # Q0 . . . . <- All masked, would give NaN + # Q1 . . . . <- All masked, would give NaN + # Q2 X . . . <- First valid row + # Q3 X X . . + # Q4 X X X . + # Q5 X X X X + causal_start_idx = seqlen_q - seqlen_k + start_m_idx = start_m * BLOCK_M + + # Create mask for rows that need zeroing + row_indices = start_m_idx + tl.arange(0, BLOCK_M) + causal_mask = row_indices < causal_start_idx + + # Zero out both acc and LSE for these rows + if causal_start_idx > start_m_idx: + end_m_idx = (start_m + 1) * BLOCK_M + if causal_start_idx < end_m_idx: + # This block contains the boundary - need to mask acc + out_mask_boundary = tl.full( + (BLOCK_DMODEL_V,), causal_start_idx, dtype=tl.int32 + ) + out_ptrs_mask = row_indices[:, None] >= out_mask_boundary[None, :] + z = 0.0 + acc = tl.where(out_ptrs_mask, acc, z.to(acc.type.element_ty)) + + # Zero out LSE for rows above diagonal + softmax_lse = tl.where(causal_mask, 0.0, softmax_lse) + + # write back LSE(Log Sum Exponents), the log of the normalization constant + l_offset = ( + LSE + + off_z * stride_lse_z + + off_h_q * stride_lse_h + + cu_seqlens_q_start * stride_lse_m + ) + l_ptrs = l_offset + offs_m * stride_lse_m + + # If seqlen_q not multiple of BLOCK_M, we need to mask out the last few rows. + # This is only true for the last Q block. For others, overflow_size will be -ve + end_m_idx = (start_m + 1) * BLOCK_M + overflow_size = end_m_idx - seqlen_q + if overflow_size > 0: + boundary = tl.full((BLOCK_M,), BLOCK_M - overflow_size, dtype=tl.int32) + l_ptrs_mask = tl.arange(0, BLOCK_M) < boundary + tl.store(l_ptrs, softmax_lse, mask=l_ptrs_mask) + else: + tl.store(l_ptrs, softmax_lse) + + # write back O + o_offset = ( + Out + off_z * stride_oz + off_h_q * stride_oh + cu_seqlens_q_start * stride_om + ) + o_ptrs = o_offset + offs_m[:, None] * stride_om + offs_d_v[None, :] * stride_on + o_ptrs_mask = tl.full([BLOCK_M, BLOCK_DMODEL_V], 1, dtype=tl.int1) + if overflow_size > 0: + o_ptrs_mask = o_ptrs_mask & (offs_m[:, None] < seqlen_q) + if PADDED_HEAD_V: + o_ptrs_mask = o_ptrs_mask & (offs_d_v[None, :] < ACTUAL_BLOCK_DMODEL_V) + + tl.store(o_ptrs, acc.to(Out.dtype.element_ty), mask=o_ptrs_mask) + + +def attention_prefill_forward_triton_impl( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + o: torch.Tensor, + sm_scale: float, + alibi_slopes: Optional[torch.Tensor], + causal: bool, + window_size_left: int, + window_size_right: int, + bias: Optional[torch.Tensor], + layout: Literal["bshd", "bhsd", "thd"], + # varlen + cu_seqlens_q: Optional[torch.Tensor], + cu_seqlens_k: Optional[torch.Tensor], + max_seqlens_q: int, + max_seqlens_k: int, + # dropout + dropout_p: float, + philox_seed: Optional[int], + philox_offset: Optional[int], + # misc + return_softmax: bool, + use_exp2: bool, + # fp8 + q_descale: Optional[torch.Tensor], + k_descale: Optional[torch.Tensor], + v_descale: Optional[torch.Tensor], + # seqused for FA v3 + seqused_q: Optional[torch.Tensor] = None, + seqused_k: Optional[torch.Tensor] = None, + # rotary (optional) + rotary_cos: Optional[torch.Tensor] = None, + rotary_sin: Optional[torch.Tensor] = None, + rotary_interleaved: bool = False, + seqlens_rotary: Optional[torch.Tensor] = None, +): + # get params, strides and shape + IS_VARLEN = layout == "thd" + + # common assertions + assert ( + 0.0 <= dropout_p <= 1.0 + ), f"dropout_p must be between 0 and 1, got {dropout_p}" + assert ( + q.device == k.device == v.device == o.device + ), f"All tensors must be on the same device. Got: q={q.device}, k={k.device}, v={v.device}, o={o.device}" + assert q.dtype == k.dtype == v.dtype, "q, k, v must have the same dtype" + current_device = torch.cuda.current_device() + assert ( + q.is_cuda and q.device.index == current_device + ), f"Device mismatch: Kernel will launch on cuda:{current_device}, but tensors are on {q.device}" + + # get shapes and strides + if IS_VARLEN: + # shape + total_seqlen_q, nheads_q, head_size_q = q.shape + total_seqlen_k, nheads_k, head_size_k = k.shape + total_seqlen_v, nheads_v, head_size_v = v.shape + + # assert shapes + assert ( + cu_seqlens_q is not None + ), "cu_seqlens_q must be provided for varlen layout" + assert ( + cu_seqlens_k is not None + ), "cu_seqlens_k must be provided for varlen layout" + assert ( + max_seqlens_q is not None and max_seqlens_q > 0 + ), "max_seqlens_q must be provided and positive for varlen layout" + assert ( + max_seqlens_k is not None and max_seqlens_k > 0 + ), "max_seqlens_k must be provided and positive for varlen layout" + + # assert head dimensions + assert ( + head_size_q == head_size_k + ), f"head sizes must match: q={head_size_q}, k={head_size_k}" + assert ( + nheads_k == nheads_v + ), f"k and v must have same number of heads: k={nheads_k}, v={nheads_v}" + assert ( + nheads_q % nheads_k == 0 + ), f"nheads_q {nheads_q} must be divisible by nheads_k {nheads_k} for GQA/MQA" + + # assert output shapes + assert o.shape == ( + total_seqlen_q, + nheads_q, + head_size_v, + ), f"o shape {o.shape} != expected {(total_seqlen_q, nheads_q, head_size_v)}" + + # assert cu_seqlens + assert ( + cu_seqlens_q.dtype == torch.int32 + ), f"cu_seqlens_q must be int32, got {cu_seqlens_q.dtype}" + assert ( + cu_seqlens_k.dtype == torch.int32 + ), f"cu_seqlens_k must be int32, got {cu_seqlens_k.dtype}" + assert cu_seqlens_q[0] == 0, "cu_seqlens_q must start with 0" + assert cu_seqlens_k[0] == 0, "cu_seqlens_k must start with 0" + assert ( + cu_seqlens_q[-1] == total_seqlen_q + ), f"cu_seqlens_q[-1] {cu_seqlens_q[-1]} != total_seqlen_q {total_seqlen_q}" + assert ( + cu_seqlens_k[-1] == total_seqlen_k + ), f"cu_seqlens_k[-1] {cu_seqlens_k[-1]} != total_seqlen_k {total_seqlen_k}" + + # set vars + batch = len(cu_seqlens_q) - 1 + head_size_qk = head_size_q + + # softmax_lse shape + softmax_lse = torch.zeros( + (nheads_q, total_seqlen_q), device=q.device, dtype=torch.float32 + ) + + # strides + stride_qb, stride_qh, stride_qm, stride_qd = ( + 0, + q.stride(1), + q.stride(0), + q.stride(2), + ) + stride_kb, stride_kh, stride_kn, stride_kd = ( + 0, + k.stride(1), + k.stride(0), + k.stride(2), + ) + stride_vb, stride_vh, stride_vn, stride_vd = ( + 0, + v.stride(1), + v.stride(0), + v.stride(2), + ) + stride_ob, stride_oh, stride_om, stride_od = ( + 0, + o.stride(1), + o.stride(0), + o.stride(2), + ) + stride_lse_z, stride_lse_h, stride_lse_m = ( + 0, + softmax_lse.stride(0), + softmax_lse.stride(1), + ) + else: + # shapes + batch_q, seqlen_q, nheads_q, head_size_q = q.shape + batch_k, seqlen_k, nheads_k, head_size_k = k.shape + batch_v, seqlen_v, nheads_v, head_size_v = v.shape + + # assert batch dimensions + assert ( + batch_q == batch_k == batch_v + ), f"batch sizes must match: q={batch_q}, k={batch_k}, v={batch_v}" + + # assert head dimensions + assert ( + head_size_q == head_size_k + ), f"head sizes must match: q={head_size_q}, k={head_size_k}" + assert ( + nheads_k == nheads_v + ), f"k and v must have same number of heads: k={nheads_k}, v={nheads_v}" + assert ( + nheads_q % nheads_k == 0 + ), f"nheads_q {nheads_q} must be divisible by nheads_k {nheads_k} for GQA/MQA" + + # assert sequence lengths + assert ( + seqlen_k == seqlen_v + ), f"k and v sequence lengths must match: k={seqlen_k}, v={seqlen_v}" + + # assert output shapes + assert o.shape == ( + batch_q, + seqlen_q, + nheads_q, + head_size_v, + ), f"o shape {o.shape} != expected {(batch_q, seqlen_q, nheads_q, head_size_v)}" + + # set vars + batch = batch_q + head_size_qk = head_size_q + max_seqlens_q = seqlen_q + max_seqlens_k = seqlen_k + + # softmax_lse shape + softmax_lse = torch.zeros( + (batch, nheads_q, seqlen_q), device=q.device, dtype=torch.float32 + ) + + # strides + stride_qb, stride_qh, stride_qm, stride_qd = ( + q.stride(0), + q.stride(2), + q.stride(1), + q.stride(3), + ) + stride_kb, stride_kh, stride_kn, stride_kd = ( + k.stride(0), + k.stride(2), + k.stride(1), + k.stride(3), + ) + stride_vb, stride_vh, stride_vn, stride_vd = ( + v.stride(0), + v.stride(2), + v.stride(1), + v.stride(3), + ) + stride_ob, stride_oh, stride_om, stride_od = ( + o.stride(0), + o.stride(2), + o.stride(1), + o.stride(3), + ) + stride_lse_z, stride_lse_h, stride_lse_m = softmax_lse.stride() + + # apply rotary embeddings + if rotary_cos is not None and rotary_sin is not None: + if IS_VARLEN: + raise NotImplementedError( + "Rotary embeddings with varlen (thd layout) prefill are not implemented yet." + ) + seqlen_offsets = seqlens_rotary if seqlens_rotary is not None else 0 + local = (window_size_left != -1) or (window_size_right != -1) + q, _ = apply_rotary( + q, + None, + rotary_cos, + rotary_sin, + causal=causal, + local=local, + interleaved=rotary_interleaved, + seqlen_offsets=seqlen_offsets, + ) + + # fp8 setup and assertions + IS_FP8 = is_fp8(q) + if IS_FP8: + # we already asserted that q, k, v all have the same dtype, so no need to check each one + + FP8_MAX = torch.finfo(q.dtype).max + + # Check and create default descale tensors if not provided + if (q_descale is None) or (k_descale is None) or (v_descale is None): + import warnings + + warnings.warn( + "FP8 tensors detected but descale factors not provided. Using default scale of 1.0", + UserWarning, + ) + # Create default descale tensors if not provided + if q_descale is None: + q_descale = torch.ones( + batch, nheads_q, dtype=torch.float32, device=q.device + ) + if k_descale is None: + k_descale = torch.ones( + batch, nheads_k, dtype=torch.float32, device=q.device + ) + if v_descale is None: + v_descale = torch.ones( + batch, nheads_k, dtype=torch.float32, device=q.device + ) + + # o should be fp32 or fp16/bf16 + assert o.dtype in [ + torch.float16, + torch.bfloat16, + torch.float32, + ], f"Output tensor o must be fp16, bf16, or fp32 when using fp8, got {o.dtype}" + + stride_q_descale_z = q_descale.stride(0) if q_descale is not None else 0 + stride_k_descale_z = k_descale.stride(0) if k_descale is not None else 0 + stride_v_descale_z = v_descale.stride(0) if v_descale is not None else 0 + + if DEBUG: + print(f"FP8 path triggered in fwd_prefill.py") + else: + FP8_MAX = None + q_descale = k_descale = v_descale = None + stride_q_descale_z = stride_k_descale_z = stride_v_descale_z = None + + # check output dtype matches input dtype when not using fp8 + assert ( + o.dtype == q.dtype + ), f"Output dtype {o.dtype} must match input dtype {q.dtype} when not using fp8" + + # check features + use_sliding_window = window_size_left != -1 or window_size_right != -1 + use_alibi, (stride_az, stride_ah) = ( + (True, alibi_slopes.stride()) if alibi_slopes is not None else (False, (0, 0)) + ) + # NOTE: a large bias tensor leads to overflow during pointer arithmetic + if bias is not None: + assert bias.numel() < 2**31 + + # Get closest power of 2 over or equal to 32 for both QK and V dimensions + padded_d_model_qk = 1 << (head_size_qk - 1).bit_length() + padded_d_model_v = 1 << (head_size_v - 1).bit_length() + # Smallest head_dim supported is 16. If smaller, the tile in the + # kernel is padded - there is no padding in memory for any dims. + padded_d_model_qk = max(padded_d_model_qk, 16) + padded_d_model_v = max(padded_d_model_v, 16) + + # sd_mask is used to validate dropout behavior vs the PyTorch SDPA math backend reference. We zero this out + # to give a consistent starting point and then populate it with the output of softmax with the sign bit set according + # to the dropout mask. The resulting return allows this mask to be fed into the reference implementation for testing + # only. This return holds no useful output aside from debugging. + NEEDS_SDMASK = (dropout_p > 0.0) or return_softmax + if NEEDS_SDMASK: + sd_mask = torch.zeros( + (batch, nheads_q, max_seqlens_q, max_seqlens_k), + device=q.device, + dtype=torch.float32, + ) + if DROPOUT_USE_PYTORCH: + dropout_mask = create_dropout_mask( + dropout_p, + (batch, nheads_q, max_seqlens_q, max_seqlens_k), + seed=philox_seed, + ) + else: + dropout_mask = torch.zeros( + (batch, nheads_q, max_seqlens_q, max_seqlens_k), + device=q.device, + dtype=torch.float32, + ) + stride_sz, stride_sh, stride_sm, stride_sn = ( + sd_mask.stride(0), + sd_mask.stride(1), + sd_mask.stride(2), + sd_mask.stride(3), + ) + else: + sd_mask = None + dropout_mask = None + stride_sz, stride_sh, stride_sm, stride_sn = (0, 0, 0, 0) + + if bias is not None: + stride_bz, stride_bh, stride_bm, stride_bn = ( + bias.stride(0), + bias.stride(1), + bias.stride(2), + bias.stride(3), + ) + else: + stride_bz, stride_bh, stride_bm, stride_bn = (0, 0, 0, 0) + + # launch kernel + grid = lambda META: (batch, nheads_q, triton.cdiv(max_seqlens_q, META["BLOCK_M"])) + attn_fwd[grid]( + q, + k, + v, + bias, + q_descale, + k_descale, + v_descale, + stride_q_descale_z, + stride_k_descale_z, + stride_v_descale_z, + sm_scale, + softmax_lse, + o, + stride_qb, + stride_qh, + stride_qm, + stride_qd, + stride_kb, + stride_kh, + stride_kn, + stride_kd, + stride_vb, + stride_vh, + stride_vn, + stride_vd, + stride_ob, + stride_oh, + stride_om, + stride_od, + stride_bz, + stride_bh, + stride_bm, + stride_bn, + stride_az, + stride_ah, + stride_sz, + stride_sh, + stride_sm, + stride_sn, + stride_lse_z, + stride_lse_h, + stride_lse_m, + cu_seqlens_q, + cu_seqlens_k, + seqused_q, + seqused_k, # Pass seqused tensors + dropout_p=dropout_p, + philox_seed=philox_seed, + philox_offset_base=philox_offset, + sd_mask=sd_mask, + dropout_mask=dropout_mask, + alibi_slopes=alibi_slopes, + HQ=nheads_q, + HK=nheads_k, + ACTUAL_BLOCK_DMODEL_QK=head_size_qk, + ACTUAL_BLOCK_DMODEL_V=head_size_v, + MAX_SEQLENS_Q=max_seqlens_q, + MAX_SEQLENS_K=max_seqlens_k, + IS_CAUSAL=causal, + USE_SLIDING_WINDOW=use_sliding_window, + WINDOW_SIZE_LEFT=window_size_left, + WINDOW_SIZE_RIGHT=window_size_right, + IS_VARLEN=IS_VARLEN, + BLOCK_DMODEL_QK=padded_d_model_qk, + BLOCK_DMODEL_V=padded_d_model_v, + USE_BIAS=False if bias is None else True, + USE_ALIBI=use_alibi, + ENABLE_DROPOUT=dropout_p > 0.0, + USE_EXP2=use_exp2, + RETURN_SCORES=return_softmax, + NEEDS_SDMASK=NEEDS_SDMASK, + IS_FP8=IS_FP8, + FP8_MAX=FP8_MAX, + FP8_P_DESCALE=False, + USE_SEQUSED=(seqused_q is not None or seqused_k is not None), + ) # Add flag for seqused + + return softmax_lse, sd_mask if return_softmax else None diff --git a/aiter/ops/triton/flash_attn_triton_amd/fwd_ref.py b/aiter/ops/triton/flash_attn_triton_amd/fwd_ref.py new file mode 100755 index 0000000000..8d9f6a2604 --- /dev/null +++ b/aiter/ops/triton/flash_attn_triton_amd/fwd_ref.py @@ -0,0 +1,1057 @@ +import torch +import math +from typing import Literal, Optional, Union +from .utils import compute_alibi_tensor_ref + +DEBUG = False +DEBUG_CORE = False + + +def attention_forward_core_ref_impl( + q, + k, + v, + sm_scale, + causal, + window_size_left, + window_size_right, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes, + use_exp2, + cache_seqlens=None, + block_table=None, + paged_kv_block_size=None, +): + if DEBUG_CORE: + print() + print("attention_forward_core_ref_impl") + print("q:", q, q.shape) + print("k:", k, k.shape) + print("v:", v, v.shape) + print("sm_scale:", sm_scale) + print("causal:", causal) + print("window_size_left:", window_size_left) + print("window_size_right:", window_size_right) + print("dropout_p:", dropout_p) + print("philox_seed:", philox_seed) + print("philox_offset:", philox_offset) + print("use_exp2:", use_exp2) + print("cache_seqlens:", cache_seqlens) + print("block_table:", block_table) + print("paged_kv_block_size:", paged_kv_block_size) + + # cast to float32 + q = q.to(torch.float32) + + # Check if we're in paged KV mode + is_paged = block_table is not None and paged_kv_block_size is not None + + if False: # Debug paged attention (disabled for production) + print(f"\n=== attention_forward_core_ref_impl DEBUG ===") + print(f"is_paged: {is_paged}") + print(f"block_table: {block_table.shape if block_table is not None else None}") + print(f"paged_kv_block_size: {paged_kv_block_size}") + if is_paged: + print(f"k shape (paged): {k.shape}") + print(f"v shape (paged): {v.shape}") + print(f"cache_seqlens: {cache_seqlens}") + + if is_paged: + # In paged mode, k and v are [num_blocks, block_size, nheads_k, head_dim] + # We'll compute attention on-the-fly without reconstructing + nheads_q = q.shape[0] + L_q = q.shape[1] + head_dim = q.shape[2] + + # Get number of KV heads from the cache + nheads_k = k.shape[2] # k shape: [num_blocks, block_size, nheads_k, head_dim] + + # Handle MQA/GQA + assert ( + nheads_q % nheads_k == 0 + ), f"nheads_q ({nheads_q}) must be divisible by nheads_k ({nheads_k})" + group_size = nheads_q // nheads_k + + # Determine the actual KV sequence length from cache_seqlens + L_k = cache_seqlens if isinstance(cache_seqlens, int) else cache_seqlens.item() + + if False: # Debug disabled + print( + f"L_q: {L_q}, L_k: {L_k}, nheads_q: {nheads_q}, nheads_k: {nheads_k}, group_size: {group_size}, head_dim: {head_dim}" + ) + print( + f"block_table contents: {block_table if block_table is not None else 'None'}" + ) + + # Initialize attention scores + attention_scores = torch.zeros( + (nheads_q, L_q, L_k), dtype=torch.float32, device=q.device + ) + + # Compute attention scores on-the-fly by accessing blocks directly + for kv_pos in range(L_k): + # Calculate which block and position within block + block_idx = kv_pos // paged_kv_block_size + within_block_idx = kv_pos % paged_kv_block_size + + # Get the physical block number from block_table + # block_table is [1, num_blocks] for single batch in core function + if block_table.dim() == 2: + physical_block = block_table[0, block_idx].item() + else: + physical_block = block_table[block_idx].item() + + # Debug output disabled + # if kv_pos == 0: + # print(f"First KV access: block_idx={block_idx}, within_block={within_block_idx}, physical_block={physical_block}") + # print(f"k_vec shape will be: {k[physical_block, within_block_idx, :, :].shape}") + + # Access k values directly from paged cache + # k shape: [num_blocks, block_size, nheads_k, head_dim] + k_vec = k[physical_block, within_block_idx, :, :].to( + torch.float32 + ) # [nheads_k, head_dim] + + # For GQA/MQA, we need to repeat k_vec for each group + if group_size > 1: + # Expand k_vec to match query heads + # k_vec: [nheads_k, head_dim] -> [nheads_q, head_dim] + k_vec = k_vec.repeat_interleave(group_size, dim=0) + + # Compute dot product with all query positions + # q is [nheads_q, L_q, head_dim], k_vec is [nheads_q, head_dim] + # Result should be [nheads_q, L_q] for this kv_pos + attention_scores[:, :, kv_pos] = torch.sum(q * k_vec.unsqueeze(1), dim=-1) + + # Keep k and v in original format for later v computation + k_paged = k + v_paged = v + + # Debug output disabled + # print(f"attention_scores computed shape: {attention_scores.shape}") + # print(f"attention_scores sample values: {attention_scores[0, 0, :5]}") + else: + # Standard non-paged mode + k = k.to(torch.float32) + v = v.to(torch.float32) + + # get seqlens + L_q, L_k = q.shape[1], k.shape[1] + + # Compute attention scores + attention_scores = torch.matmul(q, k.transpose(-2, -1)) + if DEBUG_CORE: + print("attention_scores:", attention_scores, attention_scores.shape) + + # Scale scores + attention_scaled_scores = sm_scale * attention_scores + if DEBUG_CORE: + print( + "attention_scaled_scores:", + attention_scaled_scores, + attention_scaled_scores.shape, + ) + + # Apply ALiBi if slopes are provided + if alibi_slopes is not None: + if cache_seqlens is not None: + # DECODE MODE: Special ALiBi handling + # In decode mode, k has shape [nheads, max_cache_len, head_dim] + # but only cache_seqlens positions are valid + + # The test's attn_bias_from_alibi_slopes uses this formula: + # relative_pos = torch.abs(row_idx + sk - sq - col_idx) + # where sk = actual valid key length, sq = query length + + row_idx = torch.arange(L_q, device=q.device, dtype=torch.float32).unsqueeze( + 1 + ) + col_idx = torch.arange(L_k, device=q.device, dtype=torch.float32).unsqueeze( + 0 + ) + + # Compute relative positions + # cache_seqlens is the actual number of valid keys (sk in the test) + # L_q is the query sequence length (sq in the test) + relative_pos = torch.abs(row_idx + cache_seqlens - L_q - col_idx) + + # Apply slopes + if alibi_slopes.dim() == 1: + # Shape: [nheads] -> [nheads, 1, 1] + alibi_slopes_expanded = alibi_slopes.view(-1, 1, 1) + else: + # Already has batch dimension + alibi_slopes_expanded = alibi_slopes + + alibi_bias = -alibi_slopes_expanded * relative_pos + + if DEBUG_CORE: + print( + f"Decode ALiBi: cache_seqlens={cache_seqlens}, L_q={L_q}, L_k={L_k}" + ) + print(f"relative_pos shape: {relative_pos.shape}") + print(f"alibi_bias shape: {alibi_bias.shape}") + else: + if DEBUG_CORE: + print("alibi_slopes:", alibi_slopes, alibi_slopes.shape) + alibi_bias = compute_alibi_tensor_ref(alibi_slopes, L_q, L_k) + if DEBUG_CORE: + print("alibi_bias:", alibi_bias, alibi_bias.shape) + alibi_bias = alibi_bias.reshape(-1, L_q, L_k) + if DEBUG_CORE: + print("alibi_bias_flat:", alibi_bias, alibi_bias.shape) + + attention_scaled_scores = attention_scaled_scores + alibi_bias + if DEBUG_CORE: + print( + "attention_scaled_scores after alibi:", + attention_scaled_scores, + attention_scaled_scores.shape, + ) + + # Apply masks + row_idx = torch.arange(L_q, device=q.device).unsqueeze(1) + col_idx = torch.arange(L_k, device=q.device).unsqueeze(0) + + if cache_seqlens is not None: + # We're in decode mode with a KV cache + # k and v are full allocated size, but only cache_seqlens positions are valid + + # Create a mask for valid cache positions + cache_mask = col_idx < cache_seqlens + + # Use cache_seqlens for offset calculation to match test's construct_local_mask + # which uses key_padding_mask.sum() as the sequence length + col_offset = cache_seqlens - L_q + + if DEBUG_CORE: + print(f"Cache mode: valid_len={cache_seqlens}, L_k={L_k}") + print(f"Using col_offset={col_offset} based on valid cache length") + else: + # Calculate offset for when seqlen_q != seqlen_k + # This offset aligns query positions to key positions + # When L_q < L_k, offset is positive, meaning query i maps to key position (i + offset) + # This is consistent with construct_local_mask in the tests which uses (sk - sq) + col_offset = L_k - L_q + cache_mask = None + + mask_applied = False + if causal and (window_size_left, window_size_right) == (-1, -1): + # Pure causal: ensure query doesn't attend to future keys + # With offset, query i can attend to keys up to position (i + col_offset) + mask = row_idx >= (col_idx - col_offset) + mask_applied = True + if DEBUG_CORE: + print("causal_mask:", mask) + elif (window_size_left, window_size_right) != (-1, -1): + # Handle the case where window sizes exceed sequence length + if window_size_left >= L_k: + window_size_left = -1 # No left limit + if window_size_right >= L_k: + window_size_right = -1 # No right limit + + if causal: + # Causal + sliding window: ensure we don't attend to future + window_size_right = ( + min(window_size_right, 0) if window_size_right != -1 else 0 + ) + + # Create sliding window mask + # Each query at position i attends to keys in [i + offset - left, i + offset + right] + if window_size_left == -1 and window_size_right == -1: + # No window restriction + mask = torch.ones((L_q, L_k), dtype=torch.bool, device=q.device) + else: + mask = torch.ones((L_q, L_k), dtype=torch.bool, device=q.device) + if window_size_left != -1: + # Each query at position i attends to keys from position (i - left) accounting for offset + mask = mask & (col_idx >= (row_idx + col_offset - window_size_left)) + if window_size_right != -1: + # Each query at position i attends to keys up to position (i + right) accounting for offset + mask = mask & (col_idx <= (row_idx + col_offset + window_size_right)) + + # Apply causal constraint + if causal: + causal_mask = row_idx >= (col_idx - col_offset) + mask = mask & causal_mask + + mask_applied = True + if DEBUG_CORE: + print( + f"sliding_window_mask (left={window_size_left}, right={window_size_right}):", + mask, + ) + + # Apply cache mask if needed + if cache_mask is not None: + if mask_applied: + mask = mask & cache_mask + else: + mask = cache_mask + mask_applied = True + + # Apply the mask if created + if mask_applied: + attention_scaled_scores = attention_scaled_scores.masked_fill( + torch.logical_not(mask.unsqueeze(0)), float("-inf") + ) + if DEBUG_CORE: + print( + "attention_scaled_scores after masking:", + attention_scaled_scores, + attention_scaled_scores.shape, + ) + + # Compute max for numerical stability + max_scores = torch.max(attention_scaled_scores, dim=-1, keepdim=True)[0] + if DEBUG_CORE: + print("max_scores:", max_scores, max_scores.shape) + if mask_applied: + # Replace -inf in max_scores with zeros to avoid NaN in subtraction + max_scores = torch.where( + torch.isinf(max_scores), torch.zeros_like(max_scores), max_scores + ) + if DEBUG_CORE: + print("max_scores after mask handling:", max_scores, max_scores.shape) + + # Shift scores + attention_shifted_scaled_scores = attention_scaled_scores - max_scores + if DEBUG_CORE: + print( + "attention_shifted_scaled_scores:", + attention_shifted_scaled_scores, + attention_shifted_scaled_scores.shape, + ) + + # Exponentiate + if use_exp2: + RCP_LN = 1 / math.log(2) + exp_scores = torch.exp2(RCP_LN * attention_shifted_scaled_scores) + else: + exp_scores = torch.exp(attention_shifted_scaled_scores) + + if DEBUG_CORE: + print("exp_scores:", exp_scores, exp_scores.shape) + + # Sum of exponentials + sum_exp_scores = torch.sum(exp_scores, dim=-1, keepdim=True) + if DEBUG_CORE: + print("sum_exp_scores:", sum_exp_scores, sum_exp_scores.shape) + if mask_applied: + # if sum of exp scores is 0.0 it means scores where -inf, we cannot compute softmax and softmax_lse. Setting to 1 deals with -inf case cleanly + sum_exp_scores = torch.where( + sum_exp_scores == 0, torch.ones_like(sum_exp_scores), sum_exp_scores + ) + if DEBUG_CORE: + print("sum_exp_scores:", sum_exp_scores, sum_exp_scores.shape) + + # Compute softmax probabilities + p = exp_scores / sum_exp_scores + + if DEBUG_CORE: + print("softmax:", p, p.shape) + + # apply dropout if specified + if dropout_p > 0.0: + rand_vals = torch.rand( + p.shape, + generator=torch.Generator(device=p.device).manual_seed(philox_seed), + device=p.device, + dtype=p.dtype, + ) + dropout_mask, dropout_scale = rand_vals > dropout_p, (1.0 / (1 - dropout_p)) + if DEBUG_CORE: + print("dropout_scale:", dropout_scale) + print("dropout_mask:", dropout_mask) + # Apply dropout mask and scale + # Set -1 for dropped positions and 1 for kept positions in exp_scores + sd_mask = torch.where(dropout_mask, exp_scores, -exp_scores) + p = torch.where(dropout_mask, p, torch.zeros_like(p)) * dropout_scale + if DEBUG_CORE: + print("softmax after dropout:", p) + print("sd_mask:", sd_mask) + else: + sd_mask = exp_scores + + # Compute log-sum-exp + if use_exp2: + LN2 = math.log(2) + RCP_LN = 1 / math.log(2) + max_scores_base2 = max_scores * RCP_LN + softmax_lse_base2 = max_scores_base2 + torch.log2(sum_exp_scores) + softmax_lse = softmax_lse_base2 * LN2 + softmax_lse.squeeze_(-1) + else: + softmax_lse = max_scores + torch.log(sum_exp_scores) + softmax_lse = softmax_lse.squeeze(-1) + + if DEBUG_CORE: + print("softmax_lse:", softmax_lse, softmax_lse.shape) + + # Compute output + if is_paged: + # Compute output on-the-fly using paged v cache + nheads_q = p.shape[0] + L_q = p.shape[1] + nheads_v = v_paged.shape[2] # [num_blocks, block_size, nheads_v, head_dim] + head_dim = v_paged.shape[3] + + # Handle MQA/GQA for v + assert ( + nheads_q % nheads_v == 0 + ), f"nheads_q ({nheads_q}) must be divisible by nheads_v ({nheads_v})" + v_group_size = nheads_q // nheads_v + + o = torch.zeros((nheads_q, L_q, head_dim), dtype=torch.float32, device=p.device) + + # Accumulate weighted v values + for kv_pos in range(L_k): + # Calculate which block and position within block + block_idx = kv_pos // paged_kv_block_size + within_block_idx = kv_pos % paged_kv_block_size + + # Get the physical block number from block_table + if block_table.dim() == 2: + physical_block = block_table[0, block_idx].item() + else: + physical_block = block_table[block_idx].item() + + # Access v values directly from paged cache + # v_paged shape: [num_blocks, block_size, nheads_v, head_dim] + v_vec = v_paged[physical_block, within_block_idx, :, :].to( + torch.float32 + ) # [nheads_v, head_dim] + + # For GQA/MQA, we need to repeat v_vec for each group + if v_group_size > 1: + # Expand v_vec to match query heads + # v_vec: [nheads_v, head_dim] -> [nheads_q, head_dim] + v_vec = v_vec.repeat_interleave(v_group_size, dim=0) + + # Weight by attention probabilities + # p is [nheads_q, L_q, L_k], we need p[:, :, kv_pos] which is [nheads_q, L_q] + # v_vec is [nheads_q, head_dim] + # We want to add p[:, :, kv_pos] * v_vec to each query position + weights = p[:, :, kv_pos].unsqueeze(-1) # [nheads_q, L_q, 1] + o += weights * v_vec.unsqueeze(1) # [nheads_q, L_q, head_dim] + else: + o = torch.matmul(p, v) + + # Debug output disabled + # if False: + # print(f"Output o shape: {o.shape}") + # print(f"Output o sample values: {o[0, 0, :5]}") + + if DEBUG_CORE: + print("o:", o, o.shape) + + # cast back to original dtype + o = o.to(torch.float16) + # softmax_lse = softmax_lse.to(torch.float16) # NOTE: if you cast lse to fp16 it cause accuracy issues. keep fp32 + sd_mask = sd_mask.to(torch.float16) + + return o, softmax_lse, sd_mask + + +def attention_vanilla_forward_pytorch_ref_impl( + q, + k, + v, + sm_scale, + causal, + window_size_left, + window_size_right, + layout, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes, + use_exp2, +): + """Compute reference output and softmax_lse using PyTorch's built-in function""" + + # Ensure the layout is 'bhsd' + if layout == "bshd": + q = q.transpose(1, 2).contiguous() + k = k.transpose(1, 2).contiguous() + v = v.transpose(1, 2).contiguous() + elif layout != "bhsd": + raise ValueError(f"Unknown layout {layout}") + + # Prepare tensors + batch_size, nheads_q, seq_len_q, head_dim = q.shape + batch_size, nheads_k, seq_len_k, head_dim = k.shape + group_size = nheads_q // nheads_k + if nheads_q % nheads_k != 0: + raise ValueError("nheads_q must be divisible by nheads_k") + + if group_size != 1: + # MQA or GQA case + # Reshape q to [batch_size, nheads_k, group_size, seq_len_q, head_dim] + q = q.reshape(batch_size, nheads_k, group_size, seq_len_q, head_dim) + # Expand k and v to match group_size + k = k.unsqueeze(2).expand(-1, -1, group_size, -1, -1) + v = v.unsqueeze(2).expand(-1, -1, group_size, -1, -1) + # Flatten the first three dimensions for computation + q = q.reshape(batch_size * nheads_k * group_size, seq_len_q, head_dim) + k = k.reshape(batch_size * nheads_k * group_size, seq_len_k, head_dim) + v = v.reshape(batch_size * nheads_k * group_size, seq_len_k, head_dim) + else: + q = q.reshape(batch_size * nheads_q, seq_len_q, head_dim) + k = k.reshape(batch_size * nheads_k, seq_len_k, head_dim) + v = v.reshape(batch_size * nheads_k, seq_len_k, head_dim) + + # Call the core attention function + o, softmax_lse, sd_mask = attention_forward_core_ref_impl( + q, + k, + v, + sm_scale, + causal, + window_size_left, + window_size_right, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes, + use_exp2, + ) + + if group_size != 1: + # Reshape outputs back to original dimensions + o = o.reshape(batch_size, nheads_k, group_size, seq_len_q, head_dim) + o = o.reshape(batch_size, nheads_q, seq_len_q, head_dim) + softmax_lse = softmax_lse.reshape(batch_size, nheads_k, group_size, seq_len_q) + softmax_lse = softmax_lse.reshape(batch_size, nheads_q, seq_len_q) + sd_mask = sd_mask.reshape( + batch_size, nheads_k, group_size, seq_len_q, seq_len_k + ) + sd_mask = sd_mask.reshape(batch_size, nheads_q, seq_len_q, seq_len_k) + else: + # Standard case + o = o.reshape(batch_size, nheads_q, seq_len_q, head_dim) + softmax_lse = softmax_lse.reshape(batch_size, nheads_q, seq_len_q) + sd_mask = sd_mask.reshape(batch_size, nheads_q, seq_len_q, seq_len_k) + + # Restore original layout if necessary + if layout == "bshd": + o = o.transpose(1, 2) + + return o, softmax_lse, sd_mask + + +def attention_varlen_forward_pytorch_ref_impl( + q, + k, + v, + sm_scale, + causal, + window_size_left, + window_size_right, + layout, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes, + use_exp2, +): + # Ensure the layout is 'thd' + if layout != "thd": + raise ValueError(f"Unsupported layout {layout}. Expected 'thd'.") + + batch_size = cu_seqlens_q.shape[0] - 1 + nheads_q, nheads_k = q.shape[1], k.shape[1] + head_dim = q.shape[2] + + # Pre-allocate outputs + total_L_q = q.shape[0] + total_L_k = k.shape[0] + + o = torch.zeros((total_L_q, nheads_q, head_dim), dtype=q.dtype, device=q.device) + softmax_lse = torch.zeros( + (nheads_q, total_L_q), dtype=torch.float32, device=q.device + ) + sd_mask = torch.zeros( + (batch_size, nheads_q, max_seqlen_q, max_seqlen_k), + dtype=torch.float32, + device=q.device, + ) + + # Compute group_size for MQA/GQA handling + group_size = nheads_q // nheads_k + if nheads_q % nheads_k != 0: + raise ValueError("nheads_q must be divisible by nheads_k") + + for i in range(batch_size): + # Get the start and end indices for the current sequence + start_q = cu_seqlens_q[i].item() + end_q = cu_seqlens_q[i + 1].item() + start_k = cu_seqlens_k[i].item() + end_k = cu_seqlens_k[i + 1].item() + + seqlen_q = end_q - start_q + seqlen_k = end_k - start_k + + if DEBUG: + print( + f"Batch {i} with seqlen_q = {seqlen_q}, seqlen_k = {seqlen_k}, Hq= {nheads_q}, Hk = {nheads_k}" + ) + + # Extract q_i, k_i, v_i + q_i = q[start_q:end_q, :, :] # [L_q_i, nheads_q, head_dim] + k_i = k[start_k:end_k, :, :] # [L_k_i, nheads_k, head_dim] + v_i = v[start_k:end_k, :, :] # [L_k_i, nheads_k, head_dim] + + # Permute to [nheads, L_q_i, head_dim] + q_i = q_i.permute(1, 0, 2) + k_i = k_i.permute(1, 0, 2) + v_i = v_i.permute(1, 0, 2) + + # Handle MQA/GQA by adjusting shapes based on group_size + if group_size != 1: + # Reshape q_i to [nheads_k, group_size, L_q_i, head_dim] + q_i = q_i.reshape(nheads_k, group_size, seqlen_q, head_dim) + # Expand k_i and v_i to match group_size + k_i = k_i.unsqueeze(1).expand(-1, group_size, -1, -1) + v_i = v_i.unsqueeze(1).expand(-1, group_size, -1, -1) + # Flatten the first two dimensions for computation + q_i = q_i.reshape(nheads_k * group_size, seqlen_q, head_dim) + k_i = k_i.reshape(nheads_k * group_size, seqlen_k, head_dim) + v_i = v_i.reshape(nheads_k * group_size, seqlen_k, head_dim) + else: + # Standard case + q_i = q_i.reshape(nheads_q, seqlen_q, head_dim) + k_i = k_i.reshape(nheads_k, seqlen_k, head_dim) + v_i = v_i.reshape(nheads_k, seqlen_k, head_dim) + + if alibi_slopes is not None: + alibi_slopes_i = alibi_slopes[i] + else: + alibi_slopes_i = None + + # Call the core attention function for this sequence + o_i, softmax_lse_i, sd_mask_i = attention_forward_core_ref_impl( + q_i, + k_i, + v_i, + sm_scale, + causal, + window_size_left, + window_size_right, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes_i, + use_exp2, + ) + + # Reshape outputs back to original dimensions + if group_size != 1: + # Reshape outputs to [nheads_k, group_size, seqlen_q, head_dim] + o_i = o_i.reshape(nheads_k, group_size, seqlen_q, head_dim) + # Combine the first two dimensions back to nheads_q + o_i = o_i.reshape(nheads_q, seqlen_q, head_dim) + # Reshape softmax_lse_i similarly + softmax_lse_i = softmax_lse_i.reshape(nheads_k, group_size, seqlen_q) + softmax_lse_i = softmax_lse_i.reshape(nheads_q, seqlen_q) + else: + # Outputs are already in the correct shape + pass + + # Convert back to 'thd' layout + o_i = o_i.permute(1, 0, 2) # [L_q_i, nheads_q, head_dim] + sd_mask_i = sd_mask_i # [nheads_q, L_q_i, L_k_i] + + # Place outputs in pre-allocated tensors + o[start_q:end_q, :, :] = o_i + softmax_lse[:, start_q:end_q] = softmax_lse_i + sd_mask[i, :, :seqlen_q, :seqlen_k] = sd_mask_i + + return o, softmax_lse, sd_mask + + +def attention_prefill_forward_ref_impl( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + out: torch.Tensor, + sm_scale: float, + alibi_slopes: Optional[torch.Tensor], + causal: bool, + window_size_left: int, + window_size_right: int, + layout: Literal["bshd", "bhsd", "thd"], + cu_seqlens_q: torch.Tensor, + cu_seqlens_k: torch.Tensor, + max_seqlen_q: int, + max_seqlen_k: int, + dropout_p: float, + philox_seed: Optional[int], + philox_offset: Optional[int], + use_exp2: bool, + rotary_cos: Optional[torch.Tensor] = None, + rotary_sin: Optional[torch.Tensor] = None, + rotary_interleaved: bool = False, + rotary_seqlen_offsets: Optional[torch.Tensor] = None, +): + if (rotary_cos is not None) or (rotary_sin is not None): + raise NotImplementedError( + "Rotary embeddings not implemented in reference prefill path." + ) + # compute reference + if layout == "thd": + o_ref, softmax_lse_ref, sd_mask_ref = attention_varlen_forward_pytorch_ref_impl( + q.clone(), + k.clone(), + v.clone(), + sm_scale, + causal, + window_size_left, + window_size_right, + layout, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes, + use_exp2, + ) + else: + o_ref, softmax_lse_ref, sd_mask_ref = ( + attention_vanilla_forward_pytorch_ref_impl( + q.clone(), + k.clone(), + v.clone(), + sm_scale, + causal, + window_size_left, + window_size_right, + layout, + dropout_p, + philox_seed, + philox_offset, + alibi_slopes, + use_exp2, + ) + ) + + # copy back to ouput tensor + out.copy_(o_ref.to(out.dtype)) + + return softmax_lse_ref, sd_mask_ref + + +def attention_decode_forward_ref_impl( + q: torch.Tensor, + k_cache: torch.Tensor, + v_cache: torch.Tensor, + k_new: Optional[torch.Tensor], + v_new: Optional[torch.Tensor], + out: torch.Tensor, + sm_scale: float, + causal: bool, + window_size_left: int, + window_size_right: int, + alibi_slopes: Optional[torch.Tensor], + layout: Literal["bshd"], + cache_seqlens: Optional[torch.Tensor], + cache_batch_idx: Optional[torch.Tensor], + block_table: Optional[torch.Tensor] = None, + q_descale: Optional[torch.Tensor] = None, + k_descale: Optional[torch.Tensor] = None, + v_descale: Optional[torch.Tensor] = None, + rotary_cos: Optional[torch.Tensor] = None, + rotary_sin: Optional[torch.Tensor] = None, + rotary_interleaved: bool = False, +): + """Compute reference output for decode attention using PyTorch's built-in functions""" + + if False: # Permanently disabled old debug output + pass + # print("\n========== attention_decode_forward_ref_impl inputs ==========") + # print(f"q shape: {q.shape}, dtype: {q.dtype}, device: {q.device}") + print(f"q values:\n{q}") + print( + f"\nk_cache shape: {k_cache.shape}, dtype: {k_cache.dtype}, device: {k_cache.device}" + ) + print(f"k_cache values:\n{k_cache}") + print( + f"\nv_cache shape: {v_cache.shape}, dtype: {v_cache.dtype}, device: {v_cache.device}" + ) + print(f"v_cache values:\n{v_cache}") + print( + f"\nk_new: {k_new.shape if k_new is not None else None}, dtype: {k_new.dtype if k_new is not None else None}" + ) + if k_new is not None: + print(f"k_new values:\n{k_new}") + print( + f"\nv_new: {v_new.shape if v_new is not None else None}, dtype: {v_new.dtype if v_new is not None else None}" + ) + if v_new is not None: + print(f"v_new values:\n{v_new}") + print(f"\nout shape: {out.shape}, dtype: {out.dtype}, device: {out.device}") + print(f"out values:\n{out}") + print(f"\nsm_scale: {sm_scale}") + print(f"causal: {causal}") + print(f"window_size_left: {window_size_left}") + print(f"window_size_right: {window_size_right}") + print( + f"\nalibi_slopes: {alibi_slopes.shape if alibi_slopes is not None else None}, dtype: {alibi_slopes.dtype if alibi_slopes is not None else None}" + ) + if alibi_slopes is not None: + print(f"alibi_slopes values:\n{alibi_slopes}") + print(f"\nlayout: {layout}") + print(f"cache_seqlens: {cache_seqlens}") + if cache_seqlens is not None and torch.is_tensor(cache_seqlens): + print(f"cache_seqlens values: {cache_seqlens}") + print(f"cache_batch_idx: {cache_batch_idx}") + if cache_batch_idx is not None: + print(f"cache_batch_idx values: {cache_batch_idx}") + print( + f"\nblock_table: {block_table.shape if block_table is not None else None}, dtype: {block_table.dtype if block_table is not None else None}" + ) + if block_table is not None: + print(f"block_table values:\n{block_table}") + print("=" * 60) + + if (rotary_cos is not None) or (rotary_sin is not None): + raise NotImplementedError( + "Rotary embeddings not implemented in reference decode path." + ) + + # get batch size before any layout conversion + batch_size = q.shape[0] + + # Determine if we're in paged KV mode + is_paged = block_table is not None + if is_paged: + # Infer block size from cache shape + # k_cache shape for paged: [num_blocks, block_size, nheads, head_dim] + paged_kv_block_size = k_cache.shape[1] + else: + paged_kv_block_size = None + + # handle cache_batch_idx + if cache_batch_idx is not None: + # remap batch indices for cache access + batch_indices = cache_batch_idx + else: + batch_indices = torch.arange(batch_size, device=q.device) + + # copy new keys and values into cache if provided (before any layout conversion) + if k_new is not None and v_new is not None: + if is_paged: + # For paged KV cache, we need to update the blocks with new k/v values + _, seq_len_new, _, _ = ( + k_new.shape + ) # shape is [batch, seq_len, nheads, head_dim] for bshd layout + + for b in range(batch_size): + # Determine where to place new k/v in cache + if cache_seqlens is not None: + if torch.is_tensor(cache_seqlens): + start_pos = cache_seqlens[b].item() + else: + start_pos = cache_seqlens + else: + start_pos = 0 + + # For each new position, find the corresponding block and update it + for pos_offset in range(seq_len_new): + kv_pos = start_pos + pos_offset + + # Calculate which block and position within block + block_idx = kv_pos // paged_kv_block_size + within_block_idx = kv_pos % paged_kv_block_size + + # Get the physical block number from block_table + physical_block = block_table[b, block_idx].item() + + # Update the k and v values in the paged cache + # k_cache shape: [num_blocks, block_size, nheads, head_dim] + # k_new shape: [batch, seq_len, nheads, head_dim] + k_cache[physical_block, within_block_idx, :, :] = k_new[ + b, pos_offset, :, : + ] + v_cache[physical_block, within_block_idx, :, :] = v_new[ + b, pos_offset, :, : + ] + else: + _, seq_len_new, _, _ = ( + k_new.shape + ) # shape is [batch, seq_len, nheads, head_dim] for bshd layout + + for b in range(batch_size): + cache_idx = ( + batch_indices[b].item() + if torch.is_tensor(batch_indices) + else batch_indices + ) + + # determine where to place new k/v in cache + if cache_seqlens is not None: + if torch.is_tensor(cache_seqlens): + start_pos = cache_seqlens[b].item() + else: + start_pos = cache_seqlens + else: + # if no cache_seqlens, assume we're filling from the beginning + start_pos = 0 + + end_pos = start_pos + seq_len_new + + # copy new keys and values into cache (both are in bshd layout) + k_cache[cache_idx, start_pos:end_pos, :, :] = k_new[b, :, :, :] + v_cache[cache_idx, start_pos:end_pos, :, :] = v_new[b, :, :, :] + + # ensure the layout is 'bhsd' + if layout == "bshd": + q = q.transpose(1, 2).contiguous() + if not is_paged: + k_cache = k_cache.transpose(1, 2).contiguous() + v_cache = v_cache.transpose(1, 2).contiguous() + elif layout != "bhsd": + raise ValueError(f"Unknown layout {layout}") + + # prepare tensors + batch_size_q, nheads_q, seq_len_q, head_dim = q.shape + + if is_paged: + # For paged cache: [num_blocks, block_size, nheads, head_dim] + num_blocks, block_size, nheads_k, head_dim_k = k_cache.shape + _, _, nheads_v, head_dim_v = v_cache.shape + max_cache_len = None # Not directly available in paged mode + batch_size_cache = None # Not applicable in paged mode + else: + batch_size_cache, nheads_k, max_cache_len, head_dim_k = k_cache.shape + _, nheads_v, _, head_dim_v = v_cache.shape + + # validate dimensions + assert ( + head_dim == head_dim_k == head_dim_v + ), f"Head dimensions must match: {head_dim}, {head_dim_k}, {head_dim_v}" + + # handle MQA/GQA + group_size = nheads_q // nheads_k + if nheads_q % nheads_k != 0: + raise ValueError("nheads_q must be divisible by nheads_k") + + # handle cache_batch_idx + if cache_batch_idx is not None: + # remap batch indices for cache access + batch_indices = cache_batch_idx + else: + batch_indices = torch.arange(batch_size, device=q.device) + + # prepare outputs + o = torch.zeros_like(q) + softmax_lse = torch.zeros( + (batch_size, nheads_q, seq_len_q), dtype=torch.float32, device=q.device + ) + + # process each batch element + for b in range(batch_size): + if not is_paged: + cache_idx = ( + batch_indices[b].item() + if torch.is_tensor(batch_indices) + else batch_indices + ) + + # determine valid cache length for this batch element + if cache_seqlens is not None: + if torch.is_tensor(cache_seqlens): + cache_len = cache_seqlens[b].item() + if k_new is not None: + _, seq_len_new, _, _ = k_new.shape + cache_len += seq_len_new + else: + cache_len = cache_seqlens + if k_new is not None: + _, seq_len_new, _, _ = k_new.shape + cache_len += seq_len_new + else: + if is_paged: + # For paged mode, we need cache_seqlens to know the valid length + raise ValueError("cache_seqlens must be provided for paged KV cache") + else: + cache_len = max_cache_len + + if is_paged: + # For paged KV cache, pass the cache and block table directly + # Extract block table for this batch element + block_table_b = block_table[b : b + 1, :] # [1, num_blocks] + k_b = k_cache # Pass entire paged cache + v_b = v_cache # Pass entire paged cache + q_b = q[b : b + 1, :, :, :] # [1, nheads_q, seq_len_q, head_dim] + + # For paged mode with MQA/GQA, we handle expansion in the core function + # Just reshape q for now + q_b = q_b.reshape(nheads_q, seq_len_q, head_dim) + else: + # Standard non-paged mode + k_b = k_cache[cache_idx, :, :, :] # [nheads_k, max_cache_len, head_dim] + v_b = v_cache[cache_idx, :, :, :] # [nheads_v, max_cache_len, head_dim] + q_b = q[b : b + 1, :, :, :] # [1, nheads_q, seq_len_q, head_dim] + block_table_b = None + + # handle MQA/GQA by expanding k and v + if group_size != 1: + # expand k and v to match q's number of heads + k_b = k_b.unsqueeze(1).expand(-1, group_size, -1, -1) + k_b = k_b.reshape(nheads_q, max_cache_len, head_dim) + + v_b = v_b.unsqueeze(1).expand(-1, group_size, -1, -1) + v_b = v_b.reshape(nheads_q, max_cache_len, head_dim) + + # reshape for attention_forward_core_ref_impl + q_b = q_b.reshape(nheads_q, seq_len_q, head_dim) + + # handle alibi slopes for this batch + alibi_slopes_b = None + if alibi_slopes is not None: + if alibi_slopes.dim() == 2: + alibi_slopes_b = alibi_slopes[b] + else: + alibi_slopes_b = alibi_slopes + + # call core attention function with cache information + o_b, softmax_lse_b, _ = attention_forward_core_ref_impl( + q_b, + k_b, + v_b, + sm_scale, + causal, + window_size_left, + window_size_right, + dropout_p=0.0, + philox_seed=None, + philox_offset=None, + alibi_slopes=alibi_slopes_b, + use_exp2=True, + cache_seqlens=cache_len, # Pass valid cache length + block_table=block_table_b, # Pass block table for paged mode + paged_kv_block_size=paged_kv_block_size, # Pass block size for paged mode + ) + + # store outputs + o[b, :, :, :] = o_b.reshape(nheads_q, seq_len_q, head_dim) + softmax_lse[b, :, :] = softmax_lse_b.reshape(nheads_q, seq_len_q) + + # restore original layout if necessary + if layout == "bshd": + o = o.transpose(1, 2) + + # copy output to the provided tensor + out.copy_(o.to(out.dtype)) + + return softmax_lse diff --git a/aiter/ops/triton/flash_attn_triton_amd/interface_fa.py b/aiter/ops/triton/flash_attn_triton_amd/interface_fa.py new file mode 100644 index 0000000000..a2e0179f12 --- /dev/null +++ b/aiter/ops/triton/flash_attn_triton_amd/interface_fa.py @@ -0,0 +1,802 @@ +import torch +import os +from typing import Literal, Optional, Union +from .fwd_prefill import attention_prefill_forward_triton_impl +from .bwd_prefill_fused_no_atomics import attention_prefill_backward_triton_impl +from .fwd_decode import attention_decode_forward_triton_impl +from .fwd_ref import ( + attention_prefill_forward_ref_impl, + attention_decode_forward_ref_impl, +) +from .bwd_ref import attention_backward_pytorch_ref_impl +from .utils import DEBUG, USE_REF, MetaData + +USE_EXP2 = True +BWD_MODE = os.environ.get("BWD_MODE", "fused_no_atomics").lower() + + +def fwd( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + out: Optional[torch.Tensor], + alibi_slopes: Optional[torch.Tensor], + dropout_p: float, + softmax_scale: float, + causal: bool, + window_size_left: int, + window_size_right: int, + softcap: float, + return_softmax: bool, + gen_: Optional[torch.Tensor] = None, +): + + # Reject FP8 tensors (FA2 AMD path does not support FP8) + if str(q.dtype).startswith("torch.float8"): + raise NotImplementedError( + "FP8 tensors are not supported in the AMD Triton FA2 interface. Use the FA3 path instead." + ) + + # Unsupported features assertions (keep behavior explicit like v3 shim) + if softcap != 0.0: + raise NotImplementedError( + "softcap is not supported in the AMD Triton FA2 interface (expected 0.0)." + ) + + if DEBUG: + print() + print("flash_attn_triton_amd.py::fwd inputs") + print("q:", q, q.shape) + print("k:", k, k.shape) + print("v:", v, v.shape) + print("out:", out, out.shape if out is not None else None) + print("alibi_slopes:", alibi_slopes) + print("dropout_p:", dropout_p) + print("softmax_scale:", softmax_scale) + print("causal:", causal) + print("window_size_left:", window_size_left) + print("window_size_right:", window_size_right) + print("softcap:", softcap) + print("return_softmax:", return_softmax) + out = torch.zeros_like(q) if out is None else out.zero_() + + # Setup metadata + metadata = MetaData(sm_scale=softmax_scale) + metadata.max_seqlens_q = q.shape[1] + metadata.max_seqlens_k = k.shape[1] + metadata.layout = "bshd" + + # get shape + batch, _, nheads_q, _ = q.shape + + if causal: + metadata.need_causal(True) + + if alibi_slopes is not None: + if alibi_slopes.dim() == 2: + pass + elif alibi_slopes.dim() == 1: + alibi_slopes = alibi_slopes.unsqueeze(0).expand(batch, -1) + else: + raise ValueError( + f"Alibi can be (nheads,) or (batch_size, nheads). Given tensor with shape {alibi_slopes.shape}" + ) + metadata.need_alibi(alibi_slopes, batch, nheads_q) + + # store rng state + metadata.need_dropout(dropout_p, return_softmax) + rng_state = torch.as_tensor( + [metadata.philox_seed, metadata.philox_offset] + ) # as_tensors uses the underlying data and doesnot cast + + # check arguments + metadata.check_args(q, k, v, out) + + # call implementation + if USE_REF: + if DEBUG: + print("Using reference implementation") + softmax_lse_ref, sd_mask_ref = attention_prefill_forward_ref_impl( + q, + k, + v, + out, + metadata.sm_scale, + metadata.alibi_slopes, + metadata.causal, + window_size_left, + window_size_right, + metadata.layout, + metadata.cu_seqlens_q, + metadata.cu_seqlens_k, + metadata.max_seqlens_q, + metadata.max_seqlens_k, + metadata.dropout_p, + metadata.philox_seed, + metadata.philox_offset, + USE_EXP2, + rotary_cos=None, + rotary_sin=None, + rotary_interleaved=False, + rotary_seqlen_offsets=None, + ) + softmax_lse = softmax_lse_ref + sd_mask = sd_mask_ref + else: + if DEBUG: + print("Using Triton implementation") + softmax_lse_triton, sd_mask_triton = attention_prefill_forward_triton_impl( + q, + k, + v, + out, + metadata.sm_scale, + metadata.alibi_slopes, + metadata.causal, + window_size_left, + window_size_right, + None, + metadata.layout, + metadata.cu_seqlens_q, + metadata.cu_seqlens_k, + metadata.max_seqlens_q, + metadata.max_seqlens_k, + metadata.dropout_p, + metadata.philox_seed, + metadata.philox_offset, + metadata.return_softmax, + USE_EXP2, + None, + None, + None, + ) + softmax_lse = softmax_lse_triton + sd_mask = sd_mask_triton + + if DEBUG: + print("flash_attn_triton_amd.py::fwd outputs") + print("o:", out, out.shape) + print("softmax_lse:", softmax_lse, softmax_lse.shape) + print("sd_mask:", sd_mask, sd_mask.shape if sd_mask is not None else None) + print("rng_state:", rng_state) + + return out, softmax_lse, sd_mask, rng_state + + +def bwd( + dout: torch.Tensor, + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + out: torch.Tensor, + softmax_lse: torch.Tensor, + dq: Optional[torch.Tensor], + dk: Optional[torch.Tensor], + dv: Optional[torch.Tensor], + alibi_slopes: Optional[torch.Tensor], + dropout_p: float, + softmax_scale: float, + causal: bool, + window_size_left: int, + window_size_right: int, + softcap: float, + deterministic: bool, + gen_: Optional[torch.Tensor] = None, + rng_state: Optional[torch.Tensor] = None, +): + if softcap != 0.0: + raise NotImplementedError( + "softcap is not supported in the AMD Triton FA2 interface (expected 0.0)." + ) + + if DEBUG: + print() + print("flash_attn_triton_amd.py::bwd inputs") + print("dout:", dout, dout.shape) + print("q:", q, q.shape) + print("k:", k, k.shape) + print("v:", v, v.shape) + print("out:", out, out.shape) + print("softmax_lse:", softmax_lse, softmax_lse.shape) + print("dq:", dq, dq.shape if dq is not None else None) + print("dk:", dk, dk.shape if dk is not None else None) + print("dv:", dv, dv.shape if dv is not None else None) + print("alibi_slopes:", alibi_slopes) + print("dropout_p:", dropout_p) + print("out:", out) + print("softmax_scale:", softmax_scale) + print("causal:", causal) + print("window_size_left:", window_size_left) + print("window_size_right:", window_size_right) + print("deterministic:", deterministic) + print("gen_:", gen_) + print("rng_state:", rng_state) + + dq = torch.zeros_like(q) if dq is None else dq.zero_() + dk = torch.zeros_like(k) if dk is None else dk.zero_() + dv = torch.zeros_like(v) if dv is None else dv.zero_() + + # get shape + batch, _, nheads_q, _ = q.shape + + if dropout_p > 0.0: + assert rng_state is not None + philox_seed, philox_offset = rng_state[0].item(), rng_state[1].item() + else: + philox_seed, philox_offset = None, None + + if alibi_slopes is not None: + if alibi_slopes.dim() == 2: + pass + elif alibi_slopes.dim() == 1: + alibi_slopes = alibi_slopes.unsqueeze(0).expand(batch, -1) + else: + raise ValueError("Alibi can be (nheads,) or (batch_size, nheads).") + + # call implementation + if USE_REF: + if DEBUG: + print("Using reference implementation") + + delta_ref = attention_backward_pytorch_ref_impl( + dout, + q, + k, + v, + out, + softmax_lse, + dq, + dk, + dv, + softmax_scale, + alibi_slopes, + causal, + window_size_left, + window_size_right, + "bshd", + None, + None, + None, + None, + dropout_p, + philox_seed, + philox_offset, + USE_EXP2, + ) + delta = delta_ref + else: + if DEBUG: + print("Using Triton implementation") + delta = attention_prefill_backward_triton_impl( + do=dout, + q=q, + k=k, + v=v, + o=out, + softmax_lse=softmax_lse, + dq=dq, + dk=dk, + dv=dv, + sm_scale=softmax_scale, + alibi_slopes=alibi_slopes, + causal=causal, + layout="bshd", + cu_seqlens_q=None, + cu_seqlens_k=None, + max_seqlen_q=q.shape[1], + max_seqlen_k=k.shape[1], + seqused_q=None, + seqused_k=None, + dropout_p=dropout_p, + philox_seed=philox_seed, + philox_offset=philox_offset, + descale_q=None, + descale_k=None, + descale_v=None, + descale_o=None, + descale_do=None, + descale_dq=None, + descale_dk=None, + descale_dv=None, + use_exp2=USE_EXP2, + mode=BWD_MODE, + ) + + if DEBUG: + print("flash_attn_triton_amd.py::bwd outputs") + print("dv:", dv, dv.shape) + print("dk:", dk, dk.shape) + print("dq:", dq, dq.shape) + return dq, dk, dv, delta + + +def varlen_fwd( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + out: Optional[torch.Tensor], + cu_seqlens_q: torch.Tensor, + cu_seqlens_k: torch.Tensor, + seqused_k: Optional[torch.Tensor], + leftpad_k: Optional[torch.Tensor], + block_table_: Optional[torch.Tensor], + alibi_slopes: Optional[torch.Tensor], + max_seqlen_q: int, + max_seqlen_k: int, + dropout_p: float, + softmax_scale: float, + zero_tensors: bool, + causal: bool, + window_size_left: int, + window_size_right: int, + softcap: float, + return_softmax: bool, + gen_: Optional[torch.Tensor] = None, +): + + if str(q.dtype).startswith("torch.float8"): + raise NotImplementedError( + "FP8 tensors are not supported in the AMD Triton FA2 interface (varlen_fwd). Use the FA3 path instead." + ) + + if softcap != 0.0: + raise NotImplementedError( + "softcap is not supported in varlen_fwd (expected 0.0)." + ) + if leftpad_k is not None: + raise NotImplementedError( + "leftpad_k is not supported in AMD Triton FA2 varlen_fwd." + ) + if block_table_ is not None: + raise NotImplementedError( + "block_table / paged attention is not supported in AMD Triton FA2 varlen_fwd." + ) + if seqused_k is not None: + raise NotImplementedError( + "seqused_k is not supported in AMD Triton FA2 varlen_fwd." + ) + + if DEBUG: + print() + print("flash_attn_triton_amd.py::varlen_fwd") + print("q:", q, q.shape) + print("k:", k, k.shape) + print("v:", v, v.shape) + print("cu_seqlens_q:", cu_seqlens_q, cu_seqlens_q.shape) + print("cu_seqlens_k:", cu_seqlens_k, cu_seqlens_k.shape) + print("alibi_slopes:", alibi_slopes) + print("max_seqlen_q:", max_seqlen_q) + print("max_seqlen_k:", max_seqlen_k) + print("dropout_p:", dropout_p) + print("softmax_scale:", softmax_scale) + print("causal:", causal) + print("window_size_left:", window_size_left) + print("window_size_right:", window_size_right) + print("gen_:", gen_) + out = torch.zeros_like(q) if out is None else out.zero_() + + # Setup metadata + metadata = MetaData(sm_scale=softmax_scale) + metadata.set_varlen_params( + cu_seqlens_q, cu_seqlens_k, max_seqlen_q, max_seqlen_k + ) # set layout to "thd" and other metdata + assert metadata.layout is not None + + # get shape + batch = len(cu_seqlens_q) - 1 + _, nheads_q, _ = q.shape + + if causal: + metadata.need_causal(True) + + if alibi_slopes is not None: + if alibi_slopes.dim() == 2: + pass + elif alibi_slopes.dim() == 1: + alibi_slopes = alibi_slopes.unsqueeze(0).expand(batch, -1) + else: + raise ValueError("Alibi can be (nheads,) or (batch_size, nheads).") + metadata.need_alibi(alibi_slopes, batch, nheads_q) + + # store rng state + metadata.need_dropout(dropout_p, return_softmax) + rng_state = torch.as_tensor( + [metadata.philox_seed, metadata.philox_offset] + ) # as_tensors uses the underlying data and doesnot cast + + # Check arguments + metadata.check_args(q, k, v, out) + + # call implementation + if USE_REF: + if DEBUG: + print("Using reference implementation") + softmax_lse_ref, sd_mask_ref = attention_prefill_forward_ref_impl( + q, + k, + v, + out, + metadata.sm_scale, + metadata.alibi_slopes, + metadata.causal, + window_size_left, + window_size_right, + metadata.layout, + metadata.cu_seqlens_q, + metadata.cu_seqlens_k, + metadata.max_seqlens_q, + metadata.max_seqlens_k, + metadata.dropout_p, + metadata.philox_seed, + metadata.philox_offset, + USE_EXP2, + rotary_cos=None, + rotary_sin=None, + rotary_interleaved=False, + rotary_seqlen_offsets=None, + ) + softmax_lse = softmax_lse_ref + sd_mask = sd_mask_ref + else: + if DEBUG: + print("Using Triton implementation") + softmax_lse_triton, sd_mask_triton = attention_prefill_forward_triton_impl( + q, + k, + v, + out, + metadata.sm_scale, + metadata.alibi_slopes, + metadata.causal, + window_size_left, + window_size_right, + None, + metadata.layout, + metadata.cu_seqlens_q, + metadata.cu_seqlens_k, + metadata.max_seqlens_q, + metadata.max_seqlens_k, + metadata.dropout_p, + metadata.philox_seed, + metadata.philox_offset, + metadata.return_softmax, + USE_EXP2, + None, + None, + None, + ) + softmax_lse = softmax_lse_triton + sd_mask = sd_mask_triton + + if DEBUG: + print("varlen_fwd outputs") + print("out:", out, out.shape) + print("softmax_lse:", softmax_lse, softmax_lse.shape) + print("sd_mask:", sd_mask, sd_mask.shape if sd_mask is not None else None) + + return out, softmax_lse, sd_mask, rng_state + + +def varlen_bwd( + dout: torch.Tensor, + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + out: torch.Tensor, + softmax_lse: torch.Tensor, + dq: Optional[torch.Tensor], + dk: Optional[torch.Tensor], + dv: Optional[torch.Tensor], + cu_seqlens_q: torch.Tensor, + cu_seqlens_k: torch.Tensor, + alibi_slopes: Optional[torch.Tensor], + max_seqlen_q: int, + max_seqlen_k: int, + dropout_p: float, + softmax_scale: float, + zero_tensors: bool, + causal: bool, + window_size_left: int, + window_size_right: int, + softcap: float, + deterministic: bool, + gen_: Optional[torch.Tensor] = None, + rng_state: Optional[torch.Tensor] = None, +): + if str(q.dtype).startswith("torch.float8"): + raise NotImplementedError( + "FP8 tensors are not supported in the AMD Triton FA2 interface (varlen_bwd). Use the FA3 path instead." + ) + if softcap != 0.0: + raise NotImplementedError( + "softcap is not supported in varlen_bwd (expected 0.0)." + ) + + if DEBUG: + print() + print("varlen_bwd") + print("dout:", dout, dout.shape) + print("q:", q, q.shape) + print("k:", k, k.shape) + print("v:", v, v.shape) + print("out:", out) + print("softmax_lse:", softmax_lse, softmax_lse.shape) + print("dq:", dq, dq.shape if dq is not None else None) + print("dk:", dk, dk.shape if dk is not None else None) + print("dv:", dv, dv.shape if dv is not None else None) + print("cu_seqlens_q:", cu_seqlens_q, cu_seqlens_q.shape) + print("cu_seqlens_k:", cu_seqlens_k, cu_seqlens_k.shape) + print("alibi_slopes:", alibi_slopes) + print("max_seqlen_q:", max_seqlen_q) + print("max_seqlen_k:", max_seqlen_k) + print("dropout_p:", dropout_p) + print("softmax_scale:", softmax_scale) + print("causal:", causal) + print("window_size_left:", window_size_left) + print("window_size_right:", window_size_right) + print("deterministic:", deterministic) + print("gen_:", gen_) + print("rng_state:", rng_state) + + dq = torch.zeros_like(q) if dq is None else dq.zero_() + dk = torch.zeros_like(k) if dk is None else dk.zero_() + dv = torch.zeros_like(v) if dv is None else dv.zero_() + + # get shape + batch = len(cu_seqlens_q) - 1 + _, nheads_q, _ = q.shape + + if dropout_p > 0.0: + assert rng_state is not None + philox_seed, philox_offset = rng_state[0].item(), rng_state[1].item() + else: + philox_seed, philox_offset = None, None + + if alibi_slopes is not None: + if alibi_slopes.dim() == 2: + pass + elif alibi_slopes.dim() == 1: + alibi_slopes = alibi_slopes.unsqueeze(0).expand(batch, -1) + else: + raise ValueError("Alibi can be (nheads,) or (batch_size, nheads).") + + # call implementation + if USE_REF: + if DEBUG: + print("Using reference implementation") + delta_ref = attention_backward_pytorch_ref_impl( + dout, + q, + k, + v, + out, + softmax_lse, + dq, + dk, + dv, + softmax_scale, + alibi_slopes, + causal, + window_size_left, + window_size_right, + "thd", + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_p, + philox_seed, + philox_offset, + USE_EXP2, + ) + delta = delta_ref + else: + if DEBUG: + print("Using Triton implementation") + delta = attention_prefill_backward_triton_impl( + do=dout, + q=q, + k=k, + v=v, + o=out, + softmax_lse=softmax_lse, + dq=dq, + dk=dk, + dv=dv, + sm_scale=softmax_scale, + alibi_slopes=alibi_slopes, + causal=causal, + layout="thd", + cu_seqlens_q=cu_seqlens_q, + cu_seqlens_k=cu_seqlens_k, + max_seqlen_q=max_seqlen_q, + max_seqlen_k=max_seqlen_k, + seqused_q=None, + seqused_k=None, + dropout_p=dropout_p, + philox_seed=philox_seed, + philox_offset=philox_offset, + descale_q=None, + descale_k=None, + descale_v=None, + descale_o=None, + descale_do=None, + descale_dq=None, + descale_dk=None, + descale_dv=None, + use_exp2=USE_EXP2, + mode=BWD_MODE, + ) + + if DEBUG: + print("varlen_bwd outputs") + print("delta:", delta, delta.shape) + print("dv:", dv, dv.shape) + print("dk:", dk, dk.shape) + print("dq:", dq, dq.shape) + + return dq, dk, dv, delta + + +def fwd_kvcache( + q: torch.Tensor, + k_cache: torch.Tensor, + v_cache: torch.Tensor, + k: Optional[torch.Tensor], + v: Optional[torch.Tensor], + cache_seqlens: Optional[Union[(int, torch.Tensor)]], + rotary_cos: Optional[torch.Tensor], + rotary_sin: Optional[torch.Tensor], + cache_batch_idx: Optional[torch.Tensor], + cache_leftpad: Optional[torch.Tensor], + block_table: Optional[torch.Tensor], + alibi_slopes: Optional[torch.Tensor], + out: Optional[torch.Tensor], + softmax_scale: float, + causal: bool, + window_size_left: int, + window_size_right: int, + softcap: float, + rotary_interleaved: bool, + num_splits: int, +): + + if softcap != 0.0: + raise NotImplementedError( + "softcap is not supported in fwd_kvcache (expected 0.0)." + ) + if num_splits not in (0, 1): + raise NotImplementedError( + "num_splits > 1 not supported in AMD Triton FA2 fwd_kvcache." + ) + + if DEBUG: + print() + print("flash_attn_triton_amd.py::fwd_kvcache inputs") + print("q:", q, q.shape) + print("k_cache:", k_cache, k_cache.shape) + print("v_cache:", v_cache, v_cache.shape) + print("k:", k, k.shape if k is not None else None) + print("v:", v, v.shape if v is not None else None) + print("cache_seqlens:", cache_seqlens) + print("rotary_cos:", rotary_cos) + print("rotary_sin:", rotary_sin) + print("cache_batch_idx:", cache_batch_idx) + print("cache_leftpad:", cache_leftpad) + print("block_table:", block_table) + print("alibi_slopes:", alibi_slopes) + print("out:", out) + print("softmax_scale:", softmax_scale) + print("causal:", causal) + print("window_size_left:", window_size_left) + print("window_size_right:", window_size_right) + print("softcap:", softcap) + print("rotary_interleaved:", rotary_interleaved) + print("num_splits:", num_splits) + + # output + out = torch.zeros_like(q) if out is None else out.zero_() + + # fill metadata + metadata = MetaData(sm_scale=softmax_scale) + metadata.layout = "bshd" + metadata.max_seqlens_q = q.shape[1] + metadata.max_seqlens_k = k_cache.shape[1] + metadata.cache_batch_idx = cache_batch_idx + if isinstance(cache_seqlens, int): + metadata.cache_seqlens = torch.tensor(cache_seqlens, device=q.device) + else: + metadata.cache_seqlens = cache_seqlens + + # window_size can be a tensor sometimes + if isinstance(window_size_left, torch.Tensor): + metadata.window_size_left = int(window_size_left.item()) + else: + metadata.window_size_left = window_size_left + if isinstance(window_size_right, torch.Tensor): + metadata.window_size_right = int(window_size_right.item()) + else: + metadata.window_size_right = window_size_right + + k_new = k + v_new = v + + # get shape + batch, _, nheads_q, _ = q.shape + + if causal: + metadata.need_causal(True) + + if alibi_slopes is not None: + if alibi_slopes.dim() == 2: + pass + elif alibi_slopes.dim() == 1: + alibi_slopes = alibi_slopes.unsqueeze(0).expand(batch, -1) + else: + raise ValueError("Alibi can be (nheads,) or (batch_size, nheads).") + metadata.need_alibi(alibi_slopes, batch, nheads_q) + + # record rotary info in metadata + if torch.is_tensor(rotary_cos) and torch.is_tensor(rotary_sin): + metadata.need_rotary(rotary_sin, rotary_cos, rotary_interleaved) + + # launch kernel + if USE_REF: + if DEBUG: + print("Using reference implementation") + softmax_lse_ref = attention_decode_forward_ref_impl( + q, + k_cache, + v_cache, + k_new, + v_new, + out, + metadata.sm_scale, + metadata.causal, + metadata.window_size_left, + metadata.window_size_right, + metadata.alibi_slopes, + metadata.layout, + metadata.cache_seqlens, + metadata.cache_batch_idx, + block_table, + q_descale=None, + k_descale=None, + v_descale=None, + rotary_cos=rotary_cos, + rotary_sin=rotary_sin, + rotary_interleaved=rotary_interleaved, + ) + softmax_lse = softmax_lse_ref + else: + if DEBUG: + print("Using Triton implementation") + softmax_lse_triton = attention_decode_forward_triton_impl( + q, + k_cache, + v_cache, + k_new, + v_new, + out, + metadata.sm_scale, + metadata.causal, + metadata.window_size_left, + metadata.window_size_right, + metadata.alibi_slopes, + metadata.layout, + metadata.cache_seqlens, + metadata.cache_batch_idx, + block_table, + None, + None, + None, + rotary_cos=rotary_cos, + rotary_sin=rotary_sin, + rotary_interleaved=rotary_interleaved, + ) + softmax_lse = softmax_lse_triton + + if DEBUG: + print("out:", out, out.shape) + print("softmax_lse:", softmax_lse, softmax_lse.shape) + return out, softmax_lse diff --git a/aiter/ops/triton/flash_attn_triton_amd/interface_fa_v3.py b/aiter/ops/triton/flash_attn_triton_amd/interface_fa_v3.py new file mode 100755 index 0000000000..487ef0bb16 --- /dev/null +++ b/aiter/ops/triton/flash_attn_triton_amd/interface_fa_v3.py @@ -0,0 +1,777 @@ +import torch +import os +from typing import Optional, Union, Tuple +from .fwd_prefill import attention_prefill_forward_triton_impl +from .bwd_prefill_fused_no_atomics import attention_prefill_backward_triton_impl +from .fwd_decode import attention_decode_forward_triton_impl +from .fwd_ref import ( + attention_prefill_forward_ref_impl, + attention_decode_forward_ref_impl, +) +from .bwd_ref import attention_backward_pytorch_ref_impl +from .utils import DEBUG, USE_REF, MetaData, is_fp8 + +USE_EXP2 = True +BWD_MODE = os.environ.get("BWD_MODE", "fused_no_atomics").lower() +USE_DECODE_PATH = os.environ.get("FLASH_ATTENTION_V3_USE_DECODE", "0") == "1" + + +def fwd( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + k_new: Optional[torch.Tensor], + v_new: Optional[torch.Tensor], + qv: Optional[torch.Tensor], + out: Optional[torch.Tensor], + cu_seqlens_q: Optional[torch.Tensor], + cu_seqlens_k: Optional[torch.Tensor], + cu_seqlens_k_new: Optional[torch.Tensor], + seqused_q: Optional[torch.Tensor], + seqused_k: Optional[torch.Tensor], + max_seqlen_q: Optional[int], + max_seqlen_k: Optional[int], + page_table: Optional[torch.Tensor], + kv_batch_idx: Optional[torch.Tensor], + leftpad_k: Optional[torch.Tensor], + rotary_cos: Optional[torch.Tensor], + rotary_sin: Optional[torch.Tensor], + seqlens_rotary: Optional[torch.Tensor], + q_descale: Optional[torch.Tensor], + k_descale: Optional[torch.Tensor], + v_descale: Optional[torch.Tensor], + softmax_scale: float, + causal: bool, + window_size_left: int, + window_size_right: int, + attention_chunk: int, + softcap: float, + rotary_interleaved: bool, + scheduler_metadata=None, + num_splits: int = 1, + pack_gqa=None, + sm_margin: int = 0, +) -> Tuple[torch.Tensor, torch.Tensor]: + """ + Flash Attention v3 forward pass compatible interface for AMD Triton implementation. + + This function maps v3 parameters to the existing AMD Triton implementation. + """ + + if DEBUG: + print() + print("interface_fa_v3.py::fwd inputs") + print("q:", q, q.shape) + print("k:", k, k.shape) + print("v:", v, v.shape) + print("k_new:", k_new, k_new.shape if k_new is not None else None) + print("v_new:", v_new, v_new.shape if v_new is not None else None) + print("qv:", qv, qv.shape if qv is not None else None) + print("out:", out, out.shape if out is not None else None) + print( + "cu_seqlens_q:", + cu_seqlens_q, + cu_seqlens_q.shape if cu_seqlens_q is not None else None, + ) + print( + "cu_seqlens_k:", + cu_seqlens_k, + cu_seqlens_k.shape if cu_seqlens_k is not None else None, + ) + print( + "cu_seqlens_k_new:", + cu_seqlens_k_new, + cu_seqlens_k_new.shape if cu_seqlens_k_new is not None else None, + ) + print( + "seqused_q:", seqused_q, seqused_q.shape if seqused_q is not None else None + ) + print( + "seqused_k:", seqused_k, seqused_k.shape if seqused_k is not None else None + ) + print("max_seqlen_q:", max_seqlen_q) + print("max_seqlen_k:", max_seqlen_k) + print( + "page_table:", + page_table, + page_table.shape if page_table is not None else None, + ) + print( + "kv_batch_idx:", + kv_batch_idx, + kv_batch_idx.shape if kv_batch_idx is not None else None, + ) + print( + "leftpad_k:", leftpad_k, leftpad_k.shape if leftpad_k is not None else None + ) + print( + "rotary_cos:", + rotary_cos, + rotary_cos.shape if rotary_cos is not None else None, + ) + print( + "rotary_sin:", + rotary_sin, + rotary_sin.shape if rotary_sin is not None else None, + ) + print( + "seqlens_rotary:", + seqlens_rotary, + seqlens_rotary.shape if seqlens_rotary is not None else None, + ) + print( + "q_descale:", q_descale, q_descale.shape if q_descale is not None else None + ) + print( + "k_descale:", k_descale, k_descale.shape if k_descale is not None else None + ) + print( + "v_descale:", v_descale, v_descale.shape if v_descale is not None else None + ) + print("softmax_scale:", softmax_scale) + print("causal:", causal) + print("window_size_left:", window_size_left) + print("window_size_right:", window_size_right) + print("attention_chunk:", attention_chunk) + print("softcap:", softcap) + print("rotary_interleaved:", rotary_interleaved) + print("scheduler_metadata:", scheduler_metadata) + print("num_splits:", num_splits) + print("pack_gqa:", pack_gqa) + print("sm_margin:", sm_margin) + + # Handle qv packed input + if qv is not None: + raise NotImplementedError( + "QV packed input is not yet supported in the AMD Triton backend" + ) + + # Handle softcap + if softcap != 0.0: + raise NotImplementedError( + f"Softcap is not yet supported in the AMD Triton backend (got softcap={softcap}, expected 0.0)" + ) + + # Handle attention_chunk + if attention_chunk != 0 and attention_chunk != 1: + raise NotImplementedError( + f"attention_chunk is not yet supported in the AMD Triton backend (got attention_chunk={attention_chunk})" + ) + + # Handle scheduler metadata + if scheduler_metadata is not None: + raise NotImplementedError( + "Scheduler metadata is not yet supported in the AMD Triton backend" + ) + + # Handle pack_gqa + if pack_gqa is not None and pack_gqa is not False: + raise NotImplementedError( + f"pack_gqa is not yet supported in the AMD Triton backend (got pack_gqa={pack_gqa})" + ) + + # Handle num_splits + if num_splits != 1: + raise NotImplementedError( + f"Split attention (num_splits > 1) is not yet supported in the AMD Triton backend (got num_splits={num_splits})" + ) + + # Handle sm_margin + if sm_margin != 0: + raise NotImplementedError( + f"sm_margin is not yet supported in the AMD Triton backend (got sm_margin={sm_margin}, expected 0)" + ) + + # Handle leftpad_k + if leftpad_k is not None: + raise NotImplementedError( + "Left padding (leftpad_k) is not yet supported in the AMD Triton backend" + ) + + # Handle cu_seqlens_k_new + if cu_seqlens_k_new is not None: + raise NotImplementedError( + "cu_seqlens_k_new is not yet supported in the AMD Triton backend" + ) + + # if seqlens_rotary is not None: + # raise NotImplementedError("seqlens_rotary is not yet supported in the AMD Triton backend") + + # Setup metadata + metadata = MetaData(sm_scale=softmax_scale) + + # Handle variable length sequences first to determine layout + # Determine layout based on tensor dimensions and cu_seqlens presence + if cu_seqlens_q is not None: + # Q has variable length - check tensor dimensions to confirm + if len(q.shape) == 3: # [total_seqlen, nheads, head_dim] + metadata.layout = "thd" + metadata.varlen = True + metadata.cu_seqlens_q = cu_seqlens_q + metadata.max_seqlens_q = max_seqlen_q + + # K might be varlen or batch mode + if cu_seqlens_k is not None: + metadata.cu_seqlens_k = cu_seqlens_k + metadata.max_seqlens_k = max_seqlen_k + else: + # K is in batch mode while Q is varlen (KV cache scenario) + metadata.cu_seqlens_k = None + metadata.max_seqlens_k = ( + k.shape[1] if len(k.shape) == 4 else max_seqlen_k + ) + else: + raise ValueError( + f"cu_seqlens_q provided but q has shape {q.shape}, expected 3D tensor for varlen" + ) + else: + # Regular batch mode + metadata.layout = "bshd" + metadata.varlen = False + metadata.cu_seqlens_q = None + metadata.cu_seqlens_k = None + metadata.max_seqlens_q = q.shape[1] if max_seqlen_q is None else max_seqlen_q + metadata.max_seqlens_k = k.shape[1] if max_seqlen_k is None else max_seqlen_k + + # Now determine if we should use decode or prefill kernel + # Decode kernel should be used for KV cache scenarios where: + # 1. k_new/v_new are provided - incremental KV cache update (primary KV cache indicator) + # 2. kv_batch_idx is provided - KV cache batch indexing (primary KV cache indicator) + # 3. seqused_k without seqused_q - indicates KV cache fill levels (not varlen masking) + # Note: In varlen, both seqused_q and seqused_k are used for sequence masking + # In KV cache, only seqused_k is used to track cache fill levels + if USE_DECODE_PATH: + # Force decode path + use_decode = True + else: + # Detect KV cache scenarios: + # - Clear KV cache indicators (k_new, v_new, kv_batch_idx) + # - OR seqused_k without seqused_q (KV cache fill tracking, not varlen masking) + use_decode = ( + k_new is not None # Have new KV to append (KV cache indicator) + or v_new is not None # Have new KV to append (KV cache indicator) + or kv_batch_idx + is not None # Have KV cache batch indexing (KV cache indicator) + or ( + seqused_k is not None and seqused_q is None + ) # KV cache fill levels (not varlen) + ) + + # Check for unsupported features with decode kernel + if use_decode: + if metadata.layout == "thd": + raise NotImplementedError( + "Varlen is not yet supported with the decode kernel in the AMD Triton backend" + ) + if kv_batch_idx is not None: + raise NotImplementedError( + "kv_batch_idx is not yet supported with the decode kernel in the AMD Triton backend" + ) + + if out is None: + out_dtype = torch.float32 if is_fp8(q) else q.dtype + if metadata.layout == "bshd": + out = torch.zeros( + q.shape[0], + q.shape[1], + q.shape[2], + v.shape[-1], + dtype=out_dtype, + device=q.device, + ) + elif metadata.layout == "thd": + out = torch.zeros( + q.shape[0], q.shape[1], v.shape[-1], dtype=out_dtype, device=q.device + ) + else: + raise ValueError( + f"Unsupported layout: {metadata.layout}. Only 'bshd' and 'thd' layouts are supported." + ) + else: + out = out.zero_() + + if is_fp8(q): + if (q_descale is None) or (k_descale is None) or (v_descale is None): + import warnings + + warnings.warn( + "FP8 tensors detected but descale factors not provided. Using default scale of 1.0", + UserWarning, + ) + else: + # Enforce exact expected shapes; no reshaping or normalization. + if metadata.layout == "bshd": + expected_batch = q.shape[0] + expected_q_heads = q.shape[2] + expected_kv_heads = k.shape[2] + else: # thd layout + expected_batch = ( + (len(cu_seqlens_q) - 1) if cu_seqlens_q is not None else 1 + ) + expected_q_heads = q.shape[1] + expected_kv_heads = k.shape[1] + + assert ( + q_descale.dim() == 2 + and q_descale.shape[0] == expected_batch + and q_descale.shape[1] == expected_kv_heads + ), f"q_descale expected shape ({expected_batch}, {expected_kv_heads}) got {tuple(q_descale.shape)}" + assert ( + k_descale.dim() == 2 + and k_descale.shape[0] == expected_batch + and k_descale.shape[1] == expected_kv_heads + ), f"k_descale expected shape ({expected_batch}, {expected_kv_heads}) got {tuple(k_descale.shape)}" + assert ( + v_descale.dim() == 2 + and v_descale.shape[0] == expected_batch + and v_descale.shape[1] == expected_kv_heads + ), f"v_descale expected shape ({expected_batch}, {expected_kv_heads}) got {tuple(v_descale.shape)}" + + # Get shape + if metadata.layout == "bshd": + batch, _, nheads_q, _ = q.shape + else: # "thd" layout for varlen + _, nheads_q, _ = q.shape + batch = len(cu_seqlens_q) - 1 if cu_seqlens_q is not None else 1 + + # Handle causal mask + if causal: + metadata.need_causal(True) + + # Handle alibi slopes (not directly supported in v3 interface, but we'll keep the logic) + alibi_slopes = None # V3 doesn't have alibi_slopes in the signature + if alibi_slopes is not None: + if alibi_slopes.dim() == 2: + pass + elif alibi_slopes.dim() == 1: + alibi_slopes = alibi_slopes.unsqueeze(0).expand(batch, -1) + else: + raise ValueError( + f"Alibi can be (nheads,) or (batch_size, nheads). Given tensor with shape {alibi_slopes.shape}" + ) + metadata.need_alibi(alibi_slopes, batch, nheads_q) + + # Handle dropout (v3 doesn't have dropout in forward) + dropout_p = 0.0 + return_softmax = False + metadata.need_dropout(dropout_p, return_softmax) + + # rotary embeddings + if rotary_cos is not None and rotary_sin is not None: + metadata.need_rotary(rotary_sin, rotary_cos, rotary_interleaved) + + # Store RNG state + rng_state = torch.as_tensor([metadata.philox_seed, metadata.philox_offset]) + + # Call implementation + if USE_REF: + if DEBUG: + print("Using reference implementation") + + if use_decode: + if DEBUG: + print( + f"Using decode reference implementation ( layout={metadata.layout}, cache_seqlens={seqused_k is not None}, k_new={k_new is not None}, v_new={v_new is not None}, kv_batch_idx={kv_batch_idx is not None})" + ) + # Use decode reference implementation + softmax_lse = attention_decode_forward_ref_impl( + q, + k, # k_cache + v, # v_cache + k_new, + v_new, + out, + metadata.sm_scale, + metadata.causal, + window_size_left, + window_size_right, + metadata.alibi_slopes, + metadata.layout, + seqused_k, # cache_seqlens + kv_batch_idx, # cache_batch_idx + page_table, # block_table + q_descale, + k_descale, + v_descale, + rotary_cos=rotary_cos, + rotary_sin=rotary_sin, + rotary_interleaved=rotary_interleaved, + ) + else: + if DEBUG: + print("Using prefill reference implementation") + # Use prefill reference implementation + softmax_lse_ref, sd_mask_ref = attention_prefill_forward_ref_impl( + q, + k, + v, + out, + metadata.sm_scale, + metadata.alibi_slopes, + metadata.causal, + window_size_left, + window_size_right, + metadata.layout, + metadata.cu_seqlens_q, + metadata.cu_seqlens_k, + metadata.max_seqlens_q, + metadata.max_seqlens_k, + metadata.dropout_p, + metadata.philox_seed, + metadata.philox_offset, + USE_EXP2, + rotary_cos=rotary_cos, + rotary_sin=rotary_sin, + rotary_interleaved=rotary_interleaved, + rotary_seqlen_offsets=seqlens_rotary, + ) + softmax_lse = softmax_lse_ref + else: + if DEBUG: + print("Using Triton implementation") + + if use_decode: + if DEBUG: + print( + f"Using Decode Triton implementation (cache_seqlens={seqused_k is not None}, k_new={k_new is not None}, v_new={v_new is not None}, kv_batch_idx={kv_batch_idx is not None})" + ) + + # Use decode kernel for KV cache scenarios + # Note: seqused_k can serve as cache_seqlens in v3 + softmax_lse = attention_decode_forward_triton_impl( + q, + k, # k_cache in v2 terminology + v, # v_cache in v2 terminology + k_new, # New KV values to append to cache + v_new, # New KV values to append to cache + out, + metadata.sm_scale, + metadata.causal, + window_size_left, + window_size_right, + metadata.alibi_slopes, + metadata.layout, + seqused_k, # cache_seqlens + kv_batch_idx, # cache_batch_idx + page_table, # block_table for paged attention + q_descale, + k_descale, + v_descale, + rotary_cos=rotary_cos, + rotary_sin=rotary_sin, + rotary_interleaved=rotary_interleaved, + seqlens_rotary=seqlens_rotary, + ) + # Decode kernel returns only softmax_lse, not sd_mask + sd_mask_triton = None + else: + if DEBUG: + print("Using prefill Triton implementation") + # Use prefill kernel + softmax_lse_triton, sd_mask_triton = attention_prefill_forward_triton_impl( + q, + k, + v, + out, + metadata.sm_scale, + metadata.alibi_slopes, + metadata.causal, + window_size_left, + window_size_right, + None, # block_table + metadata.layout, + metadata.cu_seqlens_q, + metadata.cu_seqlens_k, + metadata.max_seqlens_q, + metadata.max_seqlens_k, + metadata.dropout_p, + metadata.philox_seed, + metadata.philox_offset, + metadata.return_softmax, + USE_EXP2, + q_descale, + k_descale, + v_descale, + seqused_q, + seqused_k, + rotary_cos=rotary_cos, + rotary_sin=rotary_sin, + rotary_interleaved=rotary_interleaved, + seqlens_rotary=seqlens_rotary, + ) + softmax_lse = softmax_lse_triton + + if DEBUG: + print("interface_fa_v3.py::fwd outputs") + print("out:", out, out.shape) + print("softmax_lse:", softmax_lse, softmax_lse.shape) + + # Return format compatible with v3 + # V3 returns (out, softmax_lse, *rest) where rest can be empty or contain additional outputs + return out, softmax_lse + + +def bwd( + dout: torch.Tensor, + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + out: torch.Tensor, + softmax_lse: torch.Tensor, + dq: Optional[torch.Tensor], + dk: Optional[torch.Tensor], + dv: Optional[torch.Tensor], + cu_seqlens_q: Optional[torch.Tensor], + cu_seqlens_k: Optional[torch.Tensor], + seqused_q: Optional[torch.Tensor], + seqused_k: Optional[torch.Tensor], + max_seqlen_q: Optional[int], + max_seqlen_k: Optional[int], + softmax_scale: float, + causal: bool, + window_size_left: int, + window_size_right: int, + softcap: float, + deterministic: bool, + sm_margin: int = 0, +) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]: + """ + Flash Attention v3 backward pass compatible interface for AMD Triton implementation. + + This function maps v3 parameters to the existing AMD Triton implementation. + """ + + if DEBUG: + print() + print("interface_fa_v3.py::bwd inputs") + print("dout:", dout, dout.shape) + print("q:", q, q.shape) + print("k:", k, k.shape) + print("v:", v, v.shape) + print("out:", out, out.shape) + print("softmax_lse:", softmax_lse, softmax_lse.shape) + print("dq:", dq, dq.shape if dq is not None else None) + print("dk:", dk, dk.shape if dk is not None else None) + print("dv:", dv, dv.shape if dv is not None else None) + print( + "cu_seqlens_q:", + cu_seqlens_q, + cu_seqlens_q.shape if cu_seqlens_q is not None else None, + ) + print( + "cu_seqlens_k:", + cu_seqlens_k, + cu_seqlens_k.shape if cu_seqlens_k is not None else None, + ) + print( + "seqused_q:", seqused_q, seqused_q.shape if seqused_q is not None else None + ) + print( + "seqused_k:", seqused_k, seqused_k.shape if seqused_k is not None else None + ) + print("max_seqlen_q:", max_seqlen_q) + print("max_seqlen_k:", max_seqlen_k) + print("softmax_scale:", softmax_scale) + print("causal:", causal) + print("window_size_left:", window_size_left) + print("window_size_right:", window_size_right) + print("softcap:", softcap) + print("deterministic:", deterministic) + print("sm_margin:", sm_margin) + + # Check for unsupported features in backward pass + + # Handle softcap + if softcap != 0.0: + raise NotImplementedError( + f"Softcap is not yet supported in the AMD Triton backend backward pass (got softcap={softcap}, expected 0.0)" + ) + + # Handle sm_margin + if sm_margin != 0: + raise NotImplementedError( + f"sm_margin is not yet supported in the AMD Triton backend backward pass (got sm_margin={sm_margin}, expected 0)" + ) + + # Initialize gradient tensors if not provided + dq = torch.zeros_like(q) if dq is None else dq.zero_() + dk = torch.zeros_like(k) if dk is None else dk.zero_() + dv = torch.zeros_like(v) if dv is None else dv.zero_() + + # Determine layout based on cu_seqlens + if cu_seqlens_q is not None and cu_seqlens_k is not None: + # Variable length sequence mode + layout = "thd" + batch = len(cu_seqlens_q) - 1 + _, nheads_q, _ = q.shape + else: + # Regular batch mode + layout = "bshd" + batch, _, nheads_q, _ = q.shape + max_seqlen_q = q.shape[1] if max_seqlen_q is None else max_seqlen_q + max_seqlen_k = k.shape[1] if max_seqlen_k is None else max_seqlen_k + + # V3 backward doesn't have dropout or alibi slopes + dropout_p = 0.0 + philox_seed, philox_offset = None, None + alibi_slopes = None + + # For fp8, we would need descale factors, but v3 interface doesn't expose them + # So we'll pass None for now + descale_q = None + descale_k = None + descale_v = None + descale_o = None + descale_do = None + descale_dq = None + descale_dk = None + descale_dv = None + + # Call implementation + if USE_REF: + if DEBUG: + print("Using reference implementation") + delta_ref = attention_backward_pytorch_ref_impl( + dout, + q, + k, + v, + out, + softmax_lse, + dq, + dk, + dv, + softmax_scale, + alibi_slopes, + causal, + window_size_left, + window_size_right, + layout, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_p, + philox_seed, + philox_offset, + USE_EXP2, + ) + delta = delta_ref + else: + if DEBUG: + print("Using Triton implementation (unified backward dispatcher)") + # Call unified backward implementation; it internally dispatches on mode. + delta = attention_prefill_backward_triton_impl( + do=dout, + q=q, + k=k, + v=v, + o=out, + softmax_lse=softmax_lse, + dq=dq, + dk=dk, + dv=dv, + sm_scale=softmax_scale, + alibi_slopes=alibi_slopes, + causal=causal, + layout=layout, + cu_seqlens_q=cu_seqlens_q, + cu_seqlens_k=cu_seqlens_k, + max_seqlen_q=max_seqlen_q, + max_seqlen_k=max_seqlen_k, + seqused_q=seqused_q, + seqused_k=seqused_k, + dropout_p=dropout_p, + philox_seed=philox_seed, + philox_offset=philox_offset, + descale_q=descale_q, + descale_k=descale_k, + descale_v=descale_v, + descale_o=descale_o, + descale_do=descale_do, + descale_dq=descale_dq, + descale_dk=descale_dk, + descale_dv=descale_dv, + use_exp2=USE_EXP2, + mode=BWD_MODE, + ) + + if DEBUG: + print("interface_fa_v3.py::bwd outputs") + print("dq:", dq, dq.shape) + print("dk:", dk, dk.shape) + print("dv:", dv, dv.shape) + print("delta:", delta, delta.shape if delta is not None else None) + + # V3 expects (dq, dk, dv, softmax_d, *rest) + # delta is the softmax_d in this case + return dq, dk, dv, delta + + +def fwd_combine( + out_partial: torch.Tensor, + lse_partial: torch.Tensor, + out: Optional[torch.Tensor] = None, + out_dtype: Optional[torch.dtype] = None, +) -> torch.Tensor: + """ + Combine partial outputs from split attention computation. + + This is used when num_splits > 1 to combine the partial results. + + Args: + out_partial: Partial output tensor from split computation + lse_partial: Partial log-sum-exp tensor + out: Optional output tensor to write to + out_dtype: Optional dtype for output + + Returns: + Combined output tensor + """ + raise NotImplementedError( + "fwd_combine is not yet implemented in the AMD Triton backend" + ) + + +def get_scheduler_metadata( + batch_size: int, + max_seqlen_q: int, + max_seqlen_k: int, + num_heads_q: int, + num_heads_kv: int, + headdim: int, + headdim_v: int, + qkv_dtype: torch.dtype, + cache_seqlens: torch.Tensor, + cu_seqlens_q: Optional[torch.Tensor] = None, + cu_seqlens_k: Optional[torch.Tensor] = None, + cu_seqlens_k_new: Optional[torch.Tensor] = None, + seqused_q: Optional[torch.Tensor] = None, + cache_leftpad: Optional[torch.Tensor] = None, + page_size: Optional[int] = None, + max_seqlen_k_new: int = 0, + causal: bool = False, + window_size_left: int = -1, + window_size_right: int = -1, + attention_chunk: int = 0, + has_softcap: bool = False, + num_splits: int = 0, + pack_gqa: Optional[bool] = None, + sm_margin: int = 0, +): + """ + Get scheduler metadata for optimized kernel selection. + + This function is used to precompute metadata for kernel scheduling in FA3. + The AMD Triton backend currently doesn't use scheduler metadata, so this + raises an error. + + Args: + Various attention parameters used for scheduling decisions + + Returns: + None - scheduler metadata is not used in AMD Triton backend + """ + raise NotImplementedError( + "get_scheduler_metadata is not supported in the AMD Triton backend yet." + ) diff --git a/aiter/ops/triton/flash_attn_triton_amd/rotary.py b/aiter/ops/triton/flash_attn_triton_amd/rotary.py new file mode 100644 index 0000000000..a7bf58c726 --- /dev/null +++ b/aiter/ops/triton/flash_attn_triton_amd/rotary.py @@ -0,0 +1,378 @@ +from __future__ import annotations + +from typing import Optional, Union, Tuple + +import torch +import triton +import triton.language as tl + +__all__ = ["apply_rotary"] + + +@triton.jit +def _rotary_kernel( + OUT, + X, + COS, + SIN, + CU_SEQLENS, + SEQLEN_OFFSETS, + seqlen, + nheads, + seqlen_ro, + stride_out_batch, + stride_out_seqlen, + stride_out_nheads, + stride_out_headdim, + stride_x_batch, + stride_x_seqlen, + stride_x_nheads, + stride_x_headdim, + ROTARY_DIM: tl.constexpr, + IS_SEQLEN_OFFSETS_TENSOR: tl.constexpr, + IS_VARLEN: tl.constexpr, + INTERLEAVED: tl.constexpr, + CONJUGATE: tl.constexpr, + BLOCK_H: tl.constexpr, + BLOCK_M: tl.constexpr, +): + BLOCK_K: tl.constexpr = triton.next_power_of_2(ROTARY_DIM) + ROTARY_DIM_HALF = ROTARY_DIM // 2 + pid_head = tl.program_id(axis=0) + pid_m = tl.program_id(axis=1) + pid_batch = tl.program_id(axis=2) + + if not IS_VARLEN: + X = X + pid_batch * stride_x_batch + OUT = OUT + pid_batch * stride_out_batch + else: + start_idx = tl.load(CU_SEQLENS + pid_batch) + seqlen = tl.load(CU_SEQLENS + pid_batch + 1) - start_idx + X = X + start_idx * stride_x_seqlen + OUT = OUT + start_idx * stride_out_seqlen + + if pid_m * BLOCK_M >= seqlen: + return + + rh = pid_head * BLOCK_H + tl.arange(0, BLOCK_H) + rm = pid_m * BLOCK_M + tl.arange(0, BLOCK_M) + if not IS_SEQLEN_OFFSETS_TENSOR: + rm_cs = rm + SEQLEN_OFFSETS + else: + rm_cs = rm + tl.load(SEQLEN_OFFSETS + pid_batch) + + rk_half = tl.arange(0, BLOCK_K // 2) + COS = COS + (rm_cs[:, None] * ROTARY_DIM_HALF + rk_half[None, :]) + SIN = SIN + (rm_cs[:, None] * ROTARY_DIM_HALF + rk_half[None, :]) + mask_cs = (rm_cs[:, None] < seqlen_ro) & (rk_half[None, :] < ROTARY_DIM_HALF) + cos = tl.load(COS, mask=mask_cs, other=1.0).to(tl.float32) + sin = tl.load(SIN, mask=mask_cs, other=0.0).to(tl.float32) + if CONJUGATE: + sin = -sin + + if not INTERLEAVED: + X = X + ( + rh[:, None, None] * stride_x_nheads + + rm[None, :, None] * stride_x_seqlen + + rk_half[None, None, :] * stride_x_headdim + ) + OUT = OUT + ( + rh[:, None, None] * stride_out_nheads + + rm[None, :, None] * stride_out_seqlen + + rk_half[None, None, :] * stride_out_headdim + ) + mask = ( + (rh[:, None, None] < nheads) + & (rm[None, :, None] < seqlen) + & (rk_half[None, None, :] < ROTARY_DIM_HALF) + ) + x0 = tl.load(X, mask=mask, other=0.0).to(tl.float32) + x1 = tl.load(X + ROTARY_DIM_HALF * stride_x_headdim, mask=mask, other=0.0).to( + tl.float32 + ) + o0 = x0 * cos - x1 * sin + o1 = x0 * sin + x1 * cos + tl.store(OUT, o0, mask=mask) + tl.store(OUT + ROTARY_DIM_HALF * stride_out_headdim, o1, mask=mask) + else: + rk = tl.arange(0, BLOCK_K) + X = X + ( + rh[:, None, None] * stride_x_nheads + + rm[None, :, None] * stride_x_seqlen + + rk[None, None, :] * stride_x_headdim + ) + OUT = OUT + ( + rh[:, None, None] * stride_out_nheads + + rm[None, :, None] * stride_out_seqlen + + rk[None, None, :] * stride_out_headdim + ) + mask = ( + (rh[:, None, None] < nheads) + & (rm[None, :, None] < seqlen) + & (rk[None, None, :] < ROTARY_DIM) + ) + x = tl.load(X, mask=mask, other=0.0).to(tl.float32) + x0, x1 = tl.split(tl.reshape(x, [BLOCK_H, BLOCK_M, BLOCK_K // 2, 2])) + o0 = x0 * cos - x1 * sin + o1 = x0 * sin + x1 * cos + o = tl.reshape(tl.join(o0, o1), [BLOCK_H, BLOCK_M, BLOCK_K]) + tl.store(OUT, o, mask=mask) + + +def _apply_rotary_kernel( + x: torch.Tensor, + cos: torch.Tensor, + sin: torch.Tensor, + seqlen_offsets: Union[int, torch.Tensor] = 0, + cu_seqlens: Optional[torch.Tensor] = None, + max_seqlen: Optional[int] = None, + interleaved: bool = False, + inplace: bool = False, + conjugate: bool = False, +) -> torch.Tensor: + is_varlen = cu_seqlens is not None + if not is_varlen: + batch, seqlen, nheads, headdim = x.shape + else: + assert ( + max_seqlen is not None + ), "If cu_seqlens is passed, max_seqlen must also be provided" + total_seqlen, nheads, headdim = x.shape + batch_p_1 = cu_seqlens.shape[0] + batch = batch_p_1 - 1 + seqlen = max_seqlen + seqlen_ro, rotary_dim_half = cos.shape + assert sin.shape == cos.shape + rotary_dim = 2 * rotary_dim_half + assert rotary_dim <= headdim + assert headdim <= 256 + assert seqlen_ro >= seqlen + + cos, sin = cos.contiguous(), sin.contiguous() + if isinstance(seqlen_offsets, torch.Tensor): + assert seqlen_offsets.shape == (batch,) + assert seqlen_offsets.dtype in (torch.int32, torch.int64) + seqlen_offsets = seqlen_offsets.contiguous() + else: + assert seqlen_offsets + seqlen <= seqlen_ro + + out = torch.empty_like(x) if not inplace else x + if rotary_dim < headdim and not inplace: + out[..., rotary_dim:].copy_(x[..., rotary_dim:]) + + # Block heuristics + BLOCK_M = 8 if rotary_dim <= 128 else 4 + grid = ( + triton.cdiv(nheads, 2), + triton.cdiv(seqlen, BLOCK_M), + batch, + ) + + # NOTE: We assume CUDA device indexing compatibility in upstream; adapt for ROCm by using device context. + # For ROCm, torch.cuda.device works if HIP_VISIBLE_DEVICES mapping is set. + with torch.cuda.device(x.device.index): # Works for ROCm as alias + torch.library.wrap_triton(_rotary_kernel)[grid]( + out, + x, + cos, + sin, + cu_seqlens, + seqlen_offsets, + seqlen, + nheads, + seqlen_ro, + out.stride(0) if not is_varlen else 0, + out.stride(-3), + out.stride(-2), + out.stride(-1), + x.stride(0) if not is_varlen else 0, + x.stride(-3), + x.stride(-2), + x.stride(-1), + rotary_dim, + isinstance(seqlen_offsets, torch.Tensor), + is_varlen, + interleaved, + conjugate, + BLOCK_M=BLOCK_M, + BLOCK_H=2, + ) + return out + + +class _ApplyRotary(torch.autograd.Function): + @staticmethod + def forward( + ctx, + x: torch.Tensor, + cos: torch.Tensor, + sin: torch.Tensor, + interleaved: bool, + inplace: bool, + seqlen_offsets: Union[int, torch.Tensor], + cu_seqlens: Optional[torch.Tensor], + max_seqlen: Optional[int], + ): + out = _apply_rotary_kernel( + x, + cos, + sin, + seqlen_offsets=seqlen_offsets, + cu_seqlens=cu_seqlens, + max_seqlen=max_seqlen, + interleaved=interleaved, + inplace=inplace, + conjugate=False, + ) + if isinstance(seqlen_offsets, int): + ctx.save_for_backward(cos, sin, cu_seqlens) + ctx.seqlen_offsets = seqlen_offsets + else: + ctx.save_for_backward(cos, sin, cu_seqlens, seqlen_offsets) + ctx.seqlen_offsets = None + ctx.interleaved = interleaved + ctx.inplace = inplace + ctx.max_seqlen = max_seqlen + return out if not inplace else x + + @staticmethod + def backward(ctx, do: torch.Tensor): + seqlen_offsets = ctx.seqlen_offsets + if seqlen_offsets is None: + cos, sin, cu_seqlens, seqlen_offsets = ctx.saved_tensors + else: + cos, sin, cu_seqlens = ctx.saved_tensors + dx = _apply_rotary_kernel( + do, + cos, + sin, + seqlen_offsets=seqlen_offsets, + cu_seqlens=cu_seqlens, + max_seqlen=ctx.max_seqlen, + interleaved=ctx.interleaved, + inplace=ctx.inplace, + conjugate=True, + ) + return dx, None, None, None, None, None, None, None + + +def apply_rotary_emb( + x: torch.Tensor, + cos: torch.Tensor, + sin: torch.Tensor, + interleaved: bool = False, + inplace: bool = False, + seqlen_offsets: Union[int, torch.Tensor] = 0, + cu_seqlens: Optional[torch.Tensor] = None, + max_seqlen: Optional[int] = None, +) -> torch.Tensor: + """Public API: apply rotary embeddings to tensor x. + + Args: + x: (B, S, H, D) if `cu_seqlens` is None else (total_S, H, D). + cos, sin: (S_rotary, rotary_dim/2) + interleaved: GPT-J style if True. + inplace: modify x in place (saves memory if rotary_dim == D). + seqlen_offsets: int or (B,) tensor of starting offsets per sequence (KV cache decode). + cu_seqlens: (B+1,) tensor enabling varlen mode. + max_seqlen: required when `cu_seqlens` is provided. + """ + # FP8 path: upcast to bfloat16 (preferred) or float16 for rotary math to avoid excessive error + original_dtype = x.dtype + is_fp8_input = original_dtype == getattr(torch, "float8_e4m3fn", None) + if is_fp8_input: + # Choose bf16 if available in cos.dtype path; otherwise fallback to float16 + target_dtype = ( + torch.bfloat16 + if cos.dtype == torch.bfloat16 or torch.cuda.is_bf16_supported() + else torch.float16 + ) + # Upcast x, cos, sin for computation (without modifying originals in-place) + x_up = x.to(target_dtype) + cos_up = cos.to(target_dtype) if cos.dtype != target_dtype else cos + sin_up = sin.to(target_dtype) if sin.dtype != target_dtype else sin + out_up = _ApplyRotary.apply( + x_up, + cos_up, + sin_up, + interleaved, + False, + seqlen_offsets, + cu_seqlens, + max_seqlen, + ) + # Cast result back to original fp8 dtype + if inplace: + x.copy_(out_up.to(original_dtype)) + return x + return out_up.to(original_dtype) + else: + return _ApplyRotary.apply( + x, cos, sin, interleaved, inplace, seqlen_offsets, cu_seqlens, max_seqlen + ) + + +def apply_rotary( + q: torch.Tensor, + k_new: Optional[torch.Tensor], + cos: torch.Tensor, + sin: torch.Tensor, + *, + causal: bool, + local: bool, + interleaved: bool = False, + seqlen_offsets: Union[int, torch.Tensor] = 0, +) -> Tuple[torch.Tensor, Optional[torch.Tensor]]: + """High-level rotary application used by AMD prefill & decode paths. + + Policy (matches test reference & legacy semantics): + - If causal OR local attention ⇒ apply rotary directly on (B, S, H, D). + - Else (non-causal global) ⇒ flatten heads into sequence: (B, 1, S*H, D), + apply rotary once, then unflatten back. + - k_new (incremental KV slice) is always rotated directly when provided. + + Args: + q: (B, S, H, D) + k_new: Optional (B, S_k, H_k, D) + cos, sin: rotary caches (S_rotary, rotary_dim/2) + causal: causal attention flag + local: sliding-window / local attention flag (pre-computed outside) + interleaved: GPT-J style rotary layout + seqlen_offsets: int or (B,) tensor of per-sequence start offsets + Returns: + (q_rot, k_new_rot) + """ + assert q.ndim == 4, f"Expected q shape (B,S,H,D), got {q.shape}" + B, S, H, D = q.shape + use_flatten = (not causal) and (not local) + + if use_flatten: + # Flatten (S,H) -> (S*H) with an added singleton dim to preserve expected 4D shape. + q_flat = q.reshape(B, S * H, D).unsqueeze(1) # (B, 1, S*H, D) + q_flat = apply_rotary_emb( + q_flat, + cos, + sin, + interleaved=interleaved, + seqlen_offsets=seqlen_offsets, + ) + # Restore shape back to (B, S, H, D) + q = q_flat.view(B, 1, S * H, D).reshape(B, S, H, D) + else: + q = apply_rotary_emb( + q, + cos, + sin, + interleaved=interleaved, + seqlen_offsets=seqlen_offsets, + ) + + if k_new is not None: + k_new = apply_rotary_emb( + k_new, + cos, + sin, + interleaved=interleaved, + seqlen_offsets=seqlen_offsets, + ) + return q, k_new diff --git a/aiter/ops/triton/flash_attn_triton_amd/utils.py b/aiter/ops/triton/flash_attn_triton_amd/utils.py new file mode 100644 index 0000000000..078b364bf4 --- /dev/null +++ b/aiter/ops/triton/flash_attn_triton_amd/utils.py @@ -0,0 +1,1552 @@ +import csv +import math +import torch +import os +import random +import functools +import triton +import triton.language as tl +import numpy as np +from typing import Literal, Optional + +# ------------------------------- +# Gloabl Variables +# ------------------------------- +AUTOTUNE = os.environ.get("FLASH_ATTENTION_TRITON_AMD_AUTOTUNE", "0").lower() in ( + "1", + "true", + "yes", +) +if AUTOTUNE: + os.environ["TRITON_PRINT_AUTOTUNING"] = "1" +DEBUG = os.environ.get("FLASH_ATTENTION_TRITON_AMD_DEBUG", "0").lower() in ( + "1", + "true", + "yes", +) +USE_REF = os.environ.get("FLASH_ATTENTION_TRITON_AMD_REF", "0").lower() in ( + "1", + "true", + "yes", +) +PERF = os.environ.get("FLASH_ATTENTION_TRITON_AMD_PERF", "0").lower() in ( + "1", + "true", + "yes", +) +USE_SINGLE_BWD_KERNEL = os.environ.get("USE_SINGLE_BWD_KERNEL", "0").lower() in ( + "1", + "true", + "yes", +) +USE_TRITON_ROCM = os.getenv("FLASH_ATTENTION_TRITON_AMD_ENABLE", "FALSE") == "TRUE" +USE_TRITON_INTERPRET = os.environ.get("TRITON_INTERPRET", "0").lower() in ( + "1", + "true", + "yes", +) +DEBUG_TRITON = ( + os.environ.get("DEBUG_TRITON", "0").lower() in ("1", "true", "yes") + and USE_TRITON_INTERPRET +) +DEBUG_TRITON_DETAIL = ( + os.environ.get("DEBUG_TRITON_DETAIL", "0").lower() in ("1", "true", "yes") + and USE_TRITON_INTERPRET +) +if USE_TRITON_ROCM: # TODO remove this + random.seed(42) +DROPOUT_USE_PYTORCH = False +DROPOUT_DUMP = False + + +# ------------------------------- +# Metadata +# ------------------------------- +class MetaData: + cu_seqlens_q: Optional[torch.Tensor] = None + cu_seqlens_k: Optional[torch.Tensor] = None + max_seqlens_q: int = 0 + max_seqlens_k: int = 0 + bias: Optional[torch.Tensor] = None + alibi_slopes: Optional[torch.Tensor] = None + causal: bool = False + num_contexts = 0 + varlen: bool = False + layout: Optional[Literal["bshd", "bhsd", "thd"]] = None + cache_seqlens: Optional[torch.Tensor] = None + cache_batch_idx = None + packing: Optional[bool] = None + return_softmax: bool = False + dropout_p: float = 0.0 + philox_seed: Optional[int] = None + philox_offset: Optional[int] = ( + None # if dropout_p > 0.0 seed the RNG so we get reproducible results for testing. + ) + # NOTE: scale sm_scale by log_2(e) and use 2^x in the loop as we do not have native e^x support in HW. + rotary_sin: Optional[torch.Tensor] = None + rotary_cos: Optional[torch.Tensor] = None + rotary_interleaved: bool = False + rotary_conjunction: bool = False + window_size_left: int = -1 + window_size_right: int = -1 + + def __repr__(self) -> str: + return ( + f"MetaData(\n" + f" sm_scale={self.sm_scale},\n" + f" cu_seqlens_q={self.cu_seqlens_q},\n" + f" cu_seqlens_k={self.cu_seqlens_k},\n" + f" max_seqlens_q={self.max_seqlens_q},\n" + f" max_seqlens_k={self.max_seqlens_k},\n" + f" bias={self.bias},\n" + f" alibi_slopes={self.alibi_slopes},\n" + f" causal={self.causal},\n" + f" num_contexts={self.num_contexts},\n" + f" varlen={self.varlen},\n" + f" layout={self.layout},\n" + f" cache_seqlens={self.cache_seqlens},\n" + f" cache_batch_idx={self.cache_batch_idx},\n" + f" dropout_p={self.dropout_p},\n" + f" return_softmax={self.return_softmax}\n" + f" window_size_left={self.window_size_left},\n" + f" window_size_right={self.window_size_right},\n" + f")" + ) + + def __init__(self, sm_scale=1.0): + self.sm_scale = sm_scale + + def set_varlen_params(self, cu_seqlens_q, cu_seqlens_k, max_seqlen_q, max_seqlen_k): + self.varlen = True + self.layout = "thd" + self.cu_seqlens_q = cu_seqlens_q + self.cu_seqlens_k = cu_seqlens_k + self.max_seqlens_q = max_seqlen_q + self.max_seqlens_k = max_seqlen_k + + # Without "varlen", there should still be one sequence. + assert len(cu_seqlens_q) >= 2 + assert len(cu_seqlens_q) == len(cu_seqlens_k) + + def need_bias(self, bias, batch, nheads, seqlen_q, seqlen_k): + assert bias.is_cuda + assert bias.dim() == 4 + assert bias.shape[0] == 1 + assert bias.shape[2:] == (seqlen_q, seqlen_k) + self.bias = bias + + def need_alibi(self, alibi_slopes, batch, nheads): + assert alibi_slopes.is_cuda + assert alibi_slopes.dim() == 2 + assert alibi_slopes.shape[0] == batch + assert alibi_slopes.shape[1] == nheads + self.alibi_slopes = alibi_slopes + + def need_causal(self, causal): + self.causal = causal + + def need_rotary(self, sin, cos, rotary_interleaved, rotary_conjunction=False): + self.rotary_sin = sin + self.rotary_cos = cos + self.rotary_interleaved = rotary_interleaved + self.rotary_conjunction = rotary_conjunction + + def need_dropout(self, dropout_p, return_softmax): + self.dropout_p = dropout_p + self.return_softmax = return_softmax + self.philox_seed, self.philox_offset = 0x1BF58, 0x1D4B49 + + def check_args(self, q, k, v, o): + assert q.dim() == k.dim() and q.dim() == v.dim() + + batch, nheads_q, nheads_k, head_size, _, _ = get_shapes_from_layout( + q, + k, + self.layout, + self.cu_seqlens_q, + self.cu_seqlens_k, + self.max_seqlens_q, + self.max_seqlens_k, + ) + if self.varlen: + assert q.dim() == 3 + assert self.cu_seqlens_q is not None + assert self.cu_seqlens_k is not None + assert len(self.cu_seqlens_q) == len(self.cu_seqlens_k) + # TODO: Remove once bias is supported with varlen + assert self.bias is None + # assert not self.return_softmax + else: + assert q.dim() == 4 + assert self.max_seqlens_q > 0 and self.max_seqlens_k > 0 + assert self.cu_seqlens_q is None and self.cu_seqlens_k is None + # assert k.shape == v.shape + assert q.shape[-1] == k.shape[-1] # and q.shape[-1] == v.shape[-1] + # TODO: Change assert if we support qkl f8 and v f16 + assert q.dtype == k.dtype and q.dtype == v.dtype + assert o.shape[:-1] == q.shape[:-1] and o.shape[-1] == v.shape[-1] + assert (nheads_q % nheads_k) == 0 + assert self.layout is not None + assert self.layout == "thd" or not self.varlen + + +# ------------------------------- +# Input Helper +# ------------------------------- +def random_seqlens_composition(SEQ_LEN, BATCH): + # generate a random composition of N into Z positive parts. + idx = torch.randperm(SEQ_LEN - 1)[: BATCH - 1] + 1 + idx, _ = torch.sort(idx) + breakpoints = torch.cat( + [ + torch.tensor([0], dtype=torch.long), + idx, + torch.tensor([SEQ_LEN], dtype=torch.long), + ] + ) + seqlens = (breakpoints[1:] - breakpoints[:-1]).to(torch.int32) + return seqlens + + +def generate_varlen_tensor( + total_seqlen: int, + num_heads: int, + head_size: int, + batch_size: Optional[int] = None, + equal_seqlens: bool = False, + device: str = "cuda", + dtype: torch.dtype = torch.float16, + mode: Literal["random", "ones", "incremental", "identity"] = "random", +): + if DEBUG: + print("total_seqlen", total_seqlen) + print("num_heads", num_heads) + print("head_size", head_size) + + # save fp8 type + is_fp8_dtype = is_dtype_fp8(dtype) + if is_fp8_dtype: + og_fp8_dtype = dtype + dtype = torch.float32 + + # get valid batch_size + if batch_size is None: + valid_batch_sizes = [ + bs for bs in [1, 2, 4, 8, 16, 32, 64] if bs <= total_seqlen + ] + batch_size = random.choice(valid_batch_sizes) + + # get seqlens + if equal_seqlens: + seqlens = torch.full( + (batch_size,), total_seqlen // batch_size, dtype=torch.int32, device=device + ) + seqlens[-1] += total_seqlen % batch_size + else: + seqlens = random_seqlens_composition(total_seqlen, batch_size).to(device=device) + + # create cumulative sequence lengths + cu_seqlens = ( + torch.cat( + [torch.tensor([0], dtype=torch.int32, device=device), seqlens.cumsum(dim=0)] + ) + .to(torch.int32) + .to(device=device) + ) + max_seqlen = torch.max(seqlens).to(torch.int32).item() + + # create varlen tensor based on mode + if mode == "incremental": + x = torch.zeros(total_seqlen, num_heads, head_size, dtype=dtype, device=device) + for i in range(batch_size): + start = cu_seqlens[i].item() + end = cu_seqlens[i + 1].item() + length = end - start + + x[start:end, :, :] = ( + torch.arange(length, dtype=dtype, device=device) + .view(length, 1, 1) + .expand(length, num_heads, head_size) + ) + elif mode == "identity": + x = torch.zeros(total_seqlen, num_heads, head_size, dtype=dtype, device=device) + # for each batch, create identity pattern within that batch's sequence + for i in range(batch_size): + start = cu_seqlens[i].item() + end = cu_seqlens[i + 1].item() + length = end - start + + # create identity pattern for positions within this batch + for pos in range(min(length, head_size)): + x[start + pos, :, pos] = 1.0 + elif mode == "random": + x = torch.randn( + (total_seqlen, num_heads, head_size), dtype=dtype, device=device + ) + elif mode == "ones": + x = torch.ones((total_seqlen, num_heads, head_size), dtype=dtype, device=device) + else: + raise ValueError(f"Unkown mode {mode}") + + if is_fp8_dtype: + # cast to fp8 + x, descale_x = cast_to_fp8( + x, og_fp8_dtype, "thd", cu_seqlens=cu_seqlens, max_seqlen=max_seqlen + ) + x.requires_grad_() + return x, cu_seqlens, max_seqlen, descale_x + else: + x.requires_grad_() + return x, cu_seqlens, max_seqlen + + +def generate_bshd_tensor( + BATCH, + SEQ_LEN, + NUM_HEADS, + D_HEAD, + dtype: torch.dtype = torch.float16, + device="cuda", + mode: Literal["random", "ones", "incremental", "identity"] = "random", +): + # save fp8 type + is_fp8_dtype = is_dtype_fp8(dtype) + if is_fp8_dtype: + og_fp8_dtype = dtype + dtype = torch.float32 + + # gen tensor based on mode + tensor_shape = (BATCH, SEQ_LEN, NUM_HEADS, D_HEAD) + if mode == "incremental": + x = ( + torch.arange(SEQ_LEN, dtype=dtype, device=device) + .view(1, SEQ_LEN, 1, 1) + .expand(*tensor_shape) + .contiguous() + ) + elif mode == "identity": + x = torch.zeros(tensor_shape, dtype=dtype, device=device) + # create identity pattern: position i has value 1 at dimension i + for i in range(min(SEQ_LEN, D_HEAD)): + x[:, i, :, i] = 1.0 + elif mode == "random": + x = torch.randn(tensor_shape, dtype=dtype, device=device) + elif mode == "ones": + x = torch.ones(tensor_shape, dtype=dtype, device=device) + else: + raise ValueError(f"Unkown mode {mode}") + + if is_fp8_dtype: + # cast to fp8 + x, descale_x = cast_to_fp8(x, og_fp8_dtype, "bshd") + x.requires_grad_() + return x, descale_x + else: + x.requires_grad_() + return x + + +def generate_bhsd_tensor( + BATCH, + NUM_HEADS, + SEQ_LEN, + D_HEAD, + dtype: torch.dtype = torch.float16, + device="cuda", + mode: Literal["random", "ones", "incremental", "identity"] = "random", +): + # save fp8 type + is_fp8_dtype = is_dtype_fp8(dtype) + if is_fp8_dtype: + og_fp8_dtype = dtype + dtype = torch.float32 + + # gen tensor based on mode + tensor_shape = (BATCH, NUM_HEADS, SEQ_LEN, D_HEAD) + if mode == "incremental": + x = ( + torch.arange(SEQ_LEN, dtype=dtype, device=device) + .view(1, 1, SEQ_LEN, 1) + .expand(*tensor_shape) + .contiguous() + ) + elif mode == "identity": + x = torch.zeros(tensor_shape, dtype=dtype, device=device) + # create identity pattern: position i has value 1 at dimension i + for i in range(min(SEQ_LEN, D_HEAD)): + x[:, :, i, i] = 1.0 + elif mode == "random": + x = torch.randn(tensor_shape, dtype=dtype, device=device) + elif mode == "ones": + x = torch.ones(tensor_shape, dtype=dtype, device=device) + else: + raise ValueError(f"Unkown mode {mode}") + + if is_fp8_dtype: + raise ValueError("fp8 not supported for bhsd yet") + else: + x.requires_grad_() + return x + + +def generate_bshd_qkv_packed( + BATCH, + SEQ_LEN, + NUM_HEADS, + D_HEAD, + dtype: torch.dtype = torch.float16, + device="cuda", + DEBUG_INPUT=False, +): + """Generate QKV packed tensor with shape (BATCH, SEQ_LEN, 3, NUM_HEADS, D_HEAD)""" + # save fp8 type + is_fp8_dtype = is_dtype_fp8(dtype) + if is_fp8_dtype: + og_fp8_dtype = dtype + dtype = torch.float32 + + # gen tensor + tensor_shape = (BATCH, SEQ_LEN, 3, NUM_HEADS, D_HEAD) + if DEBUG_INPUT: + x = ( + torch.arange(SEQ_LEN, dtype=dtype, device=device) + .view(1, SEQ_LEN, 1, 1, 1) + .expand(*tensor_shape) + .contiguous() + ) + else: + x = torch.randn(tensor_shape, dtype=dtype, device=device) + + if is_fp8_dtype: + # cast to fp8 - need to handle the packed dimension + raise NotImplementedError("FP8 not supported for QKV packing yet") + else: + x.requires_grad_() + return x + + +def generate_bshd_kv_packed( + BATCH, + SEQ_LEN, + NUM_HEADS, + D_HEAD, + dtype: torch.dtype = torch.float16, + device="cuda", + DEBUG_INPUT=False, +): + """Generate KV packed tensor with shape (BATCH, SEQ_LEN, 2, NUM_HEADS, D_HEAD)""" + # save fp8 type + is_fp8_dtype = is_dtype_fp8(dtype) + if is_fp8_dtype: + og_fp8_dtype = dtype + dtype = torch.float32 + + # gen tensor + tensor_shape = (BATCH, SEQ_LEN, 2, NUM_HEADS, D_HEAD) + if DEBUG_INPUT: + x = ( + torch.arange(SEQ_LEN, dtype=dtype, device=device) + .view(1, SEQ_LEN, 1, 1, 1) + .expand(*tensor_shape) + .contiguous() + ) + else: + x = torch.randn(tensor_shape, dtype=dtype, device=device) + + if is_fp8_dtype: + # cast to fp8 - need to handle the packed dimension + raise NotImplementedError("FP8 not supported for KV packing yet") + else: + x.requires_grad_() + return x + + +def generate_bhsd_qkv_packed( + BATCH, + NUM_HEADS, + SEQ_LEN, + D_HEAD, + dtype: torch.dtype = torch.float16, + device="cuda", + DEBUG_INPUT=False, +): + """Generate QKV packed tensor with shape (BATCH, 3, NUM_HEADS, SEQ_LEN, D_HEAD)""" + # save fp8 type + is_fp8_dtype = is_dtype_fp8(dtype) + if is_fp8_dtype: + og_fp8_dtype = dtype + dtype = torch.float32 + + # gen tensor + tensor_shape = (BATCH, 3, NUM_HEADS, SEQ_LEN, D_HEAD) + if DEBUG_INPUT: + x = ( + torch.arange(SEQ_LEN, dtype=dtype, device=device) + .view(1, 1, 1, SEQ_LEN, 1) + .expand(*tensor_shape) + .contiguous() + ) + else: + x = torch.randn(tensor_shape, dtype=dtype, device=device) + + if is_fp8_dtype: + # cast to fp8 - need to handle the packed dimension + raise NotImplementedError("FP8 not supported for QKV packing yet") + else: + x.requires_grad_() + return x + + +def generate_bhsd_kv_packed( + BATCH, + NUM_HEADS, + SEQ_LEN, + D_HEAD, + dtype: torch.dtype = torch.float16, + device="cuda", + DEBUG_INPUT=False, +): + """Generate KV packed tensor with shape (BATCH, 2, NUM_HEADS, SEQ_LEN, D_HEAD)""" + # save fp8 type + is_fp8_dtype = is_dtype_fp8(dtype) + if is_fp8_dtype: + og_fp8_dtype = dtype + dtype = torch.float32 + + # gen tensor + tensor_shape = (BATCH, 2, NUM_HEADS, SEQ_LEN, D_HEAD) + if DEBUG_INPUT: + x = ( + torch.arange(SEQ_LEN, dtype=dtype, device=device) + .view(1, 1, 1, SEQ_LEN, 1) + .expand(*tensor_shape) + .contiguous() + ) + else: + x = torch.randn(tensor_shape, dtype=dtype, device=device) + + if is_fp8_dtype: + # cast to fp8 - need to handle the packed dimension + raise NotImplementedError("FP8 not supported for KV packing yet") + else: + x.requires_grad_() + return x + + +def generate_varlen_qkv_packed( + total_seqlen: int, + num_heads: int, + head_size: int, + batch_size: Optional[int] = None, + equal_seqlens: bool = False, + device: str = "cuda", + dtype: torch.dtype = torch.float16, + DEBUG_INPUT: bool = False, +): + """Generate varlen QKV packed tensor with shape (total_seqlen, 3, num_heads, head_size)""" + if DEBUG: + print("generate_varlen_qkv_packed") + print("total_seqlen", total_seqlen) + print("num_heads", num_heads) + print("head_size", head_size) + + # save fp8 type + is_fp8_dtype = is_dtype_fp8(dtype) + if is_fp8_dtype: + og_fp8_dtype = dtype + dtype = torch.float32 + + # get valid batch_size + if batch_size is None: + valid_batch_sizes = [ + bs for bs in [1, 2, 4, 8, 16, 32, 64] if bs <= total_seqlen + ] + batch_size = random.choice(valid_batch_sizes) + + # get seqlens + if equal_seqlens: + seqlens = torch.full( + (batch_size,), total_seqlen // batch_size, dtype=torch.int32, device=device + ) + seqlens[-1] += total_seqlen % batch_size + else: + seqlens = random_seqlens_composition(total_seqlen, batch_size).to(device=device) + + # create cumulative sequence lengths + cu_seqlens = ( + torch.cat( + [torch.tensor([0], dtype=torch.int32, device=device), seqlens.cumsum(dim=0)] + ) + .to(torch.int32) + .to(device=device) + ) + max_seqlen = torch.max(seqlens).to(torch.int32).item() + + # create varlen qkv packed tensor + if DEBUG_INPUT: + x = torch.zeros( + total_seqlen, 3, num_heads, head_size, dtype=dtype, device=device + ) + for i in range(batch_size): + start = cu_seqlens[i].item() + end = cu_seqlens[i + 1].item() + length = end - start + + x[start:end, :, :, :] = ( + torch.arange(length, dtype=dtype, device=device) + .view(length, 1, 1, 1) + .expand(length, 3, num_heads, head_size) + ) + else: + x = torch.randn( + (total_seqlen, 3, num_heads, head_size), dtype=dtype, device=device + ) + + if is_fp8_dtype: + # cast to fp8 - need to handle the packed dimension + raise NotImplementedError("FP8 not supported for QKV packing yet") + else: + x.requires_grad_() + return x, cu_seqlens, max_seqlen + + +def generate_varlen_kv_packed( + total_seqlen: int, + num_heads: int, + head_size: int, + batch_size: Optional[int] = None, + equal_seqlens: bool = False, + device: str = "cuda", + dtype: torch.dtype = torch.float16, + DEBUG_INPUT: bool = False, +): + """Generate varlen KV packed tensor with shape (total_seqlen, 2, num_heads, head_size)""" + if DEBUG: + print("generate_varlen_kv_packed") + print("total_seqlen", total_seqlen) + print("num_heads", num_heads) + print("head_size", head_size) + + # save fp8 type + is_fp8_dtype = is_dtype_fp8(dtype) + if is_fp8_dtype: + og_fp8_dtype = dtype + dtype = torch.float32 + + # get valid batch_size + if batch_size is None: + valid_batch_sizes = [ + bs for bs in [1, 2, 4, 8, 16, 32, 64] if bs <= total_seqlen + ] + batch_size = random.choice(valid_batch_sizes) + + # get seqlens + if equal_seqlens: + seqlens = torch.full( + (batch_size,), total_seqlen // batch_size, dtype=torch.int32, device=device + ) + seqlens[-1] += total_seqlen % batch_size + else: + seqlens = random_seqlens_composition(total_seqlen, batch_size).to(device=device) + + # create cumulative sequence lengths + cu_seqlens = ( + torch.cat( + [torch.tensor([0], dtype=torch.int32, device=device), seqlens.cumsum(dim=0)] + ) + .to(torch.int32) + .to(device=device) + ) + max_seqlen = torch.max(seqlens).to(torch.int32).item() + + # create varlen kv packed tensor + if DEBUG_INPUT: + x = torch.zeros( + total_seqlen, 2, num_heads, head_size, dtype=dtype, device=device + ) + for i in range(batch_size): + start = cu_seqlens[i].item() + end = cu_seqlens[i + 1].item() + length = end - start + + x[start:end, :, :, :] = ( + torch.arange(length, dtype=dtype, device=device) + .view(length, 1, 1, 1) + .expand(length, 2, num_heads, head_size) + ) + else: + x = torch.randn( + (total_seqlen, 2, num_heads, head_size), dtype=dtype, device=device + ) + + if is_fp8_dtype: + # cast to fp8 - need to handle the packed dimension + raise NotImplementedError("FP8 not supported for KV packing yet") + else: + x.requires_grad_() + return x, cu_seqlens, max_seqlen + + +# Replace the existing input_helper function in utils.py with this updated version + + +def input_helper( + BATCH: int, + HQ: int, + HK: int, + N_CTX_Q: int, + N_CTX_K: int, + D_HEAD: int, + CAUSAL: bool, + DROPOUT_P: float, + dtype: torch.dtype, + layout: Literal["bshd", "bhsd", "thd"], + packing: Optional[Literal["kv", "qkv"]] = None, + device: Literal["cpu", "cuda"] = "cuda", +): + torch.manual_seed(20) + is_fp8_dtype = is_dtype_fp8(dtype) + + if layout == "thd": + # set params + TOTAL_SEQLENS_Q = BATCH * N_CTX_Q + TOTAL_SEQLENS_K = BATCH * N_CTX_K + equal_seqlens = False + + # deal with packing + if packing is None: + # gen tensors + if is_fp8_dtype: + q, cu_seqlens_q, max_seqlen_q, descale_q = generate_varlen_tensor( + TOTAL_SEQLENS_Q, + HQ, + D_HEAD, + batch_size=BATCH, + dtype=dtype, + device=device, + equal_seqlens=equal_seqlens, + ) + k, cu_seqlens_k, max_seqlen_k, descale_k = generate_varlen_tensor( + TOTAL_SEQLENS_K, + HK, + D_HEAD, + batch_size=BATCH, + dtype=dtype, + device=device, + equal_seqlens=equal_seqlens, + ) + v, _, _, descale_v = generate_varlen_tensor( + TOTAL_SEQLENS_K, + HK, + D_HEAD, + batch_size=BATCH, + dtype=dtype, + device=device, + equal_seqlens=equal_seqlens, + ) + do, _, _, descale_do = generate_varlen_tensor( + TOTAL_SEQLENS_Q, + HQ, + D_HEAD, + batch_size=BATCH, + dtype=dtype, + device=device, + equal_seqlens=equal_seqlens, + ) + else: + q, cu_seqlens_q, max_seqlen_q = generate_varlen_tensor( + TOTAL_SEQLENS_Q, + HQ, + D_HEAD, + batch_size=BATCH, + dtype=dtype, + device=device, + equal_seqlens=equal_seqlens, + ) + k, cu_seqlens_k, max_seqlen_k = generate_varlen_tensor( + TOTAL_SEQLENS_K, + HK, + D_HEAD, + batch_size=BATCH, + dtype=dtype, + device=device, + equal_seqlens=equal_seqlens, + ) + v, _, _ = generate_varlen_tensor( + TOTAL_SEQLENS_K, + HK, + D_HEAD, + batch_size=BATCH, + dtype=dtype, + device=device, + equal_seqlens=equal_seqlens, + ) + do, _, _ = generate_varlen_tensor( + TOTAL_SEQLENS_Q, + HQ, + D_HEAD, + batch_size=BATCH, + dtype=dtype, + device=device, + equal_seqlens=equal_seqlens, + ) + elif packing == "kv": + # gen tensors with kv packing + if is_fp8_dtype: + raise ValueError("FP8 not supported for KV packing yet") + else: + q, cu_seqlens_q, max_seqlen_q = generate_varlen_tensor( + TOTAL_SEQLENS_Q, + HQ, + D_HEAD, + batch_size=BATCH, + dtype=dtype, + device=device, + equal_seqlens=equal_seqlens, + ) + kv, cu_seqlens_k, max_seqlen_k = generate_varlen_kv_packed( + TOTAL_SEQLENS_K, + HK, + D_HEAD, + batch_size=BATCH, + dtype=dtype, + device=device, + equal_seqlens=equal_seqlens, + ) + do, _, _ = generate_varlen_tensor( + TOTAL_SEQLENS_Q, + HQ, + D_HEAD, + batch_size=BATCH, + dtype=dtype, + device=device, + equal_seqlens=equal_seqlens, + ) + elif packing == "qkv": + # qkv packing - requires same sequence length for q and k + assert ( + N_CTX_Q == N_CTX_K + ), "For QKV packing, Q and K must have same sequence length" + assert HQ == HK, "For QKV packing, Q and K must have same number of heads" + + if is_fp8_dtype: + raise ValueError("FP8 not supported for QKV packing yet") + else: + qkv, cu_seqlens_q, max_seqlen_q = generate_varlen_qkv_packed( + TOTAL_SEQLENS_Q, + HQ, + D_HEAD, + batch_size=BATCH, + dtype=dtype, + device=device, + equal_seqlens=equal_seqlens, + ) + cu_seqlens_k = cu_seqlens_q + max_seqlen_k = max_seqlen_q + do, _, _ = generate_varlen_tensor( + TOTAL_SEQLENS_Q, + HQ, + D_HEAD, + batch_size=BATCH, + dtype=dtype, + device=device, + equal_seqlens=equal_seqlens, + ) + + # setup metadata + sm_scale = D_HEAD**-0.5 + metadata = MetaData(sm_scale=sm_scale) + metadata.set_varlen_params( + cu_seqlens_q, cu_seqlens_k, max_seqlen_q, max_seqlen_k + ) + metadata.need_causal(CAUSAL) + metadata.need_dropout(DROPOUT_P, True) + + elif layout == "bshd" or layout == "bhsd": + # deal with packing + if packing is None: + # gen tensors + if layout == "bshd": + if is_fp8_dtype: + q, descale_q = generate_bshd_tensor( + BATCH, N_CTX_Q, HQ, D_HEAD, dtype=dtype, device=device + ) + k, descale_k = generate_bshd_tensor( + BATCH, N_CTX_K, HK, D_HEAD, dtype=dtype, device=device + ) + v, descale_v = generate_bshd_tensor( + BATCH, N_CTX_K, HK, D_HEAD, dtype=dtype, device=device + ) + do, descale_do = generate_bshd_tensor( + BATCH, N_CTX_Q, HQ, D_HEAD, dtype=dtype, device=device + ) + else: + q = generate_bshd_tensor( + BATCH, N_CTX_Q, HQ, D_HEAD, dtype=dtype, device=device + ) + k = generate_bshd_tensor( + BATCH, N_CTX_K, HK, D_HEAD, dtype=dtype, device=device + ) + v = generate_bshd_tensor( + BATCH, N_CTX_K, HK, D_HEAD, dtype=dtype, device=device + ) + do = generate_bshd_tensor( + BATCH, N_CTX_Q, HQ, D_HEAD, dtype=dtype, device=device + ) + elif layout == "bhsd": + q, descale_q = generate_bhsd_tensor( + BATCH, HQ, N_CTX_Q, D_HEAD, dtype=dtype, device=device + ) + k, descale_k = generate_bhsd_tensor( + BATCH, HK, N_CTX_K, D_HEAD, dtype=dtype, device=device + ) + v, descale_v = generate_bhsd_tensor( + BATCH, HK, N_CTX_K, D_HEAD, dtype=dtype, device=device + ) + do, descale_do = generate_bhsd_tensor( + BATCH, HQ, N_CTX_Q, D_HEAD, dtype=dtype, device=device + ) + else: + q = generate_bhsd_tensor( + BATCH, HQ, N_CTX_Q, D_HEAD, dtype=dtype, device=device + ) + k = generate_bhsd_tensor( + BATCH, HK, N_CTX_K, D_HEAD, dtype=dtype, device=device + ) + v = generate_bhsd_tensor( + BATCH, HK, N_CTX_K, D_HEAD, dtype=dtype, device=device + ) + do = generate_bhsd_tensor( + BATCH, HQ, N_CTX_Q, D_HEAD, dtype=dtype, device=device + ) + elif packing == "kv": + # gen tensors with kv packing + if is_fp8_dtype: + raise ValueError("FP8 not supported for KV packing yet") + else: + if layout == "bshd": + q = generate_bshd_tensor( + BATCH, N_CTX_Q, HQ, D_HEAD, dtype=dtype, device=device + ) + kv = generate_bshd_kv_packed( + BATCH, N_CTX_K, HK, D_HEAD, dtype=dtype, device=device + ) + do = generate_bshd_tensor( + BATCH, N_CTX_Q, HQ, D_HEAD, dtype=dtype, device=device + ) + elif layout == "bhsd": + q = generate_bhsd_tensor( + BATCH, HQ, N_CTX_Q, D_HEAD, dtype=dtype, device=device + ) + kv = generate_bhsd_kv_packed( + BATCH, HK, N_CTX_K, D_HEAD, dtype=dtype, device=device + ) + do = generate_bhsd_tensor( + BATCH, HQ, N_CTX_Q, D_HEAD, dtype=dtype, device=device + ) + elif packing == "qkv": + # qkv packing - requires same sequence length for q and k + assert ( + N_CTX_Q == N_CTX_K + ), "For QKV packing, Q and K must have same sequence length" + assert HQ == HK, "For QKV packing, Q and K must have same number of heads" + + if is_fp8_dtype: + raise ValueError("FP8 not supported for QKV packing yet") + else: + if layout == "bshd": + qkv = generate_bshd_qkv_packed( + BATCH, N_CTX_Q, HQ, D_HEAD, dtype=dtype, device=device + ) + do = generate_bshd_tensor( + BATCH, N_CTX_Q, HQ, D_HEAD, dtype=dtype, device=device + ) + elif layout == "bhsd": + qkv = generate_bhsd_qkv_packed( + BATCH, HQ, N_CTX_Q, D_HEAD, dtype=dtype, device=device + ) + do = generate_bhsd_tensor( + BATCH, HQ, N_CTX_Q, D_HEAD, dtype=dtype, device=device + ) + + # setup metadata + sm_scale = D_HEAD**-0.5 + metadata = MetaData(sm_scale=sm_scale) + metadata.max_seqlens_q = N_CTX_Q + metadata.max_seqlens_k = N_CTX_K + metadata.layout = layout + metadata.need_causal(CAUSAL) + metadata.need_dropout(DROPOUT_P, True) + else: + raise ValueError(f"Unknown layout: {layout}") + + # return based on packing + if packing is None: + if is_fp8_dtype: + return ( + (q, descale_q), + (k, descale_k), + (v, descale_v), + (do, descale_do), + metadata, + ) + else: + return q, k, v, do, metadata + elif packing == "kv": + if is_fp8_dtype: + raise ValueError("FP8 not supported kv packing yet") + else: + return q, kv, do, metadata + elif packing == "qkv": + if is_fp8_dtype: + raise ValueError("FP8 not supported qkv packing yet") + else: + return qkv, do, metadata + else: + assert False, f"Unsupported packing mode: {packing}" + + +# ------------------------------- +# Alibi +# ------------------------------- +@triton.jit +def compute_alibi_block( + alibi_slope, seqlen_q, seqlen_k, offs_m, offs_n, transpose=False +): + # when seqlen_k and seqlen_q are different we want the diagonal to stick to the bottom right of the attention matrix + # for casual mask we want something like this where (1 is kept and 0 is masked) + # seqlen_q = 2 and seqlen_k = 5 + # 1 1 1 1 0 + # 1 1 1 1 1 + # seqlen_q = 5 and seqlen_k = 2 + # 0 0 + # 0 0 + # 0 0 + # 1 0 + # 1 1 + # for alibi the diagonal is 0 indicating no penalty for attending to that spot and increasing penalty for attending further from the diagonal + # e.g. alibi_slope = 1, seqlen_q = 2, seqlen_k = 5, offs_m = [0, 1, 2, 3], offs_n = [0, 1, 2, 3, 4], transpose = False + # 1. offs_m[:,None] = [[0], + # [1], + # 2. offs_m[:,None] + seqlen_k = [[5], + # [6], + # 3. offs_m[:,None] + seqlen_k - seqlen_q = [[3], + # [4], + # 4. offs_m[:,None] + seqlen_k - seqlen_q - offs_n[None,:] = [[3], - [[0, 1, 2, 3, 4]] = [[ 3, 2, 1, 0,-1], + # [4], [ 4, 3, 2, 1, 0]] + # 5. -1 * alibi_slope * tl.abs(relative_pos_block) = [[ -3, -2, -1, 0,-1], + # [ -4, -3, -2, -1, 0]], + relative_pos_block = offs_m[:, None] + seqlen_k - seqlen_q - offs_n[None, :] + alibi_block = -1 * alibi_slope * tl.abs(relative_pos_block) + if transpose: + return alibi_block.T + else: + return alibi_block + + +# ------------------------------- +# FP8 +# ------------------------------- +def is_dtype_fp8(dtype): + if dtype in { + torch.float8_e4m3fnuz, + torch.float8_e4m3fn, + torch.float8_e5m2, + torch.float8_e5m2fnuz, + }: + if arch_supports_fp8(): + return True + else: + raise RuntimeError("This device doesnot support fp8") + else: + return False + + +def is_fp8(x): + return is_dtype_fp8(x.dtype) + + +@triton.jit +def compute_fp8_scaling_factors(x, fp8_max: tl.constexpr): + # compute fp8 scaling and descaling factor for a block + x_amax = tl.max(tl.abs(x)) # NOTE: abs deals with negative values + x_amax = tl.where(x_amax <= 1e-9, 1e-9, x_amax) + scale_x = fp8_max / x_amax + descale_x = x_amax / fp8_max + return scale_x, descale_x + + +@triton.jit +def _cast_varlen_to_fp8_kernel_2d( + X, + X_fp8, + Descale, + cu_seqlens, + H, + MAX_SEQLEN, + stride_batch, + stride_seq, + stride_head, + stride_dim, + stride_out_batch, + stride_out_seq, + stride_out_head, + stride_out_dim, + stride_desc_batch, + stride_desc_head, + FP8_CLAMP_VAL, + FP8_MAX, + BLOCK_SIZE: tl.constexpr, + HEAD_DIM: tl.constexpr, + ACTUAL_HEAD_DIM: tl.constexpr, + IS_VARLEN: tl.constexpr, +): + # Process one (batch, head) pair per kernel + b_id = tl.program_id(0) + h_id = tl.program_id(1) + + # Get sequence bounds for this batch + if IS_VARLEN: + seq_start = tl.load(cu_seqlens + b_id) + seq_end = tl.load(cu_seqlens + b_id + 1) + seqlen = seq_end - seq_start + else: + seq_start = 0 + seqlen = MAX_SEQLEN + + # initialize max value tracker + x_max_val = 0.0 + + # STEP 1: Find max absolute value across the entire sequence + num_of_blocks = tl.cdiv(seqlen, BLOCK_SIZE) + for blk_idx in range(0, num_of_blocks): + # print("blk_idx:", blk_idx) + # offsets + offs_seq = blk_idx * BLOCK_SIZE + tl.arange(0, BLOCK_SIZE) + offs_dim = tl.arange(0, HEAD_DIM) + + # Create mask for valid elements + mask_seq = offs_seq[:, None] < seqlen + if ACTUAL_HEAD_DIM != HEAD_DIM: + mask_dim = offs_dim[None, :] < ACTUAL_HEAD_DIM + mask_seq = mask_seq & mask_dim + + # Load block + adj_x = ( + b_id * stride_batch + + h_id * stride_head + + seq_start * stride_seq + + offs_seq[:, None] * stride_seq + + offs_dim[None, :] * stride_dim + ) + x_block = tl.load(X + adj_x, mask=mask_seq, other=0.0) + # print("x_block:", x_block) + + # Find max absolute value in this block + block_max = tl.max(tl.abs(x_block)) + # print("block_max:", block_max) + + # Update overall max + x_max_val = tl.maximum(x_max_val, block_max) + # print("x_max_val:", x_max_val) + + # clamp to avoid division by zero issues + x_max_val = tl.maximum(x_max_val, FP8_CLAMP_VAL) + + # compute scale and descale factors for the entire sequence + scale = FP8_MAX / x_max_val + descale = x_max_val / FP8_MAX + + # store descale factor for this (batch, head) pair + desc_ptr = Descale + b_id * stride_desc_batch + h_id # * stride_desc_head + tl.store(desc_ptr, descale) + + # STEP 2: Apply scaling to the entire sequence and convert to FP8 + for blk_idx in range(0, num_of_blocks): + # offsets + offs_seq = blk_idx * BLOCK_SIZE + tl.arange(0, BLOCK_SIZE) + offs_dim = tl.arange(0, HEAD_DIM) + + # Create mask for valid elements + mask_seq = offs_seq[:, None] < seqlen + if ACTUAL_HEAD_DIM != HEAD_DIM: + mask_dim = offs_dim[None, :] < ACTUAL_HEAD_DIM + mask_seq = mask_seq & mask_dim + + # Load block - Using the fixed addressing + addr = ( + b_id * stride_batch + + h_id * stride_head + + seq_start * stride_seq + + offs_seq[:, None] * stride_seq + + offs_dim[None, :] * stride_dim + ) + x_block = tl.load(X + addr, mask=mask_seq, other=0.0) + + # Apply scale and convert to FP8 + x_fp8_block = (x_block * scale).to(X_fp8.type.element_ty) + + # Store results + addr_out = ( + b_id * stride_out_batch + + h_id * stride_out_head + + seq_start * stride_out_seq + + offs_seq[:, None] * stride_out_seq + + offs_dim[None, :] * stride_out_dim + ) + tl.store(X_fp8 + addr_out, x_fp8_block, mask=mask_seq) + + +def cast_to_fp8( + x: torch.Tensor, + fp8_dtype: torch.dtype, + layout: Literal["bshd", "thd"], + clamp_val: float = 1e-9, + cu_seqlens: Optional[torch.Tensor] = None, + max_seqlen: Optional[int] = None, +) -> tuple[torch.Tensor, torch.Tensor]: + if False: + print() + print("cast_to_fp8") + print("x:", x, x.shape) + print("fp8_dtype:", fp8_dtype) + print("cu_seqlens:", cu_seqlens) + print("max_seqlen:", max_seqlen) + print("clamp_val:", clamp_val) + + # check types are valid + assert x.dtype in { + torch.float16, + torch.float32, + torch.float64, + torch.bfloat16, + } and is_dtype_fp8(fp8_dtype), f"Cannot cast {x.dtype} to {fp8_dtype}" + + # extract dimensions + batch, max_seqlen_final, num_heads, head_dim = get_shape_from_layout( + x, layout, cu_seqlens, max_seqlen + ) + is_varlen = layout == "thd" + fp8_max = torch.finfo(fp8_dtype).max + if False: + print("batch:", batch) + print("max_seqlen_final:", max_seqlen_final) + print("num_heads:", num_heads) + print("head_dim:", head_dim) + + # get closest power of 2 for head_dim + padded_head_dim = 1 << (head_dim - 1).bit_length() + padded_head_dim = max(padded_head_dim, 32) + + # kernel params + x_fp8 = torch.zeros_like(x, dtype=fp8_dtype) + descale_factors = torch.zeros( + (batch, num_heads), device=x.device, dtype=torch.float32 + ) + BLOCK_SIZE = 128 + + # calculate strides + stride_batch, stride_head, stride_seq, stride_dim = get_stride_from_layout( + x, layout + ) + stride_out_batch, stride_out_head, stride_out_seq, stride_out_dim = ( + get_stride_from_layout(x_fp8, layout) + ) + stride_desc_batch, stride_desc_head = descale_factors.stride() + + if False: + print("stride_batch", stride_batch) + print("stride_head", stride_head) + print("stride_seq", stride_seq) + print("stride_dim", stride_dim) + print("stride_out_batch", stride_out_batch) + print("stride_out_head", stride_out_head) + print("stride_out_seq", stride_out_seq) + print("stride_out_dim", stride_out_dim) + print("stride_desc_batch", stride_desc_batch) + print("stride_desc_head", stride_desc_head) + + grid = (batch, num_heads) + _cast_varlen_to_fp8_kernel_2d[grid]( + x, + x_fp8, + descale_factors, + cu_seqlens, + num_heads, + max_seqlen_final, + stride_batch, + stride_seq, + stride_head, + stride_dim, + stride_out_batch, + stride_out_seq, + stride_out_head, + stride_out_dim, + stride_desc_batch, + stride_desc_head, + clamp_val, + fp8_max, + BLOCK_SIZE=BLOCK_SIZE, + HEAD_DIM=padded_head_dim, + ACTUAL_HEAD_DIM=head_dim, + IS_VARLEN=is_varlen, + ) + + if False: + print("x_fp8:", x_fp8, x_fp8.shape) + print("descale_factors:", descale_factors, descale_factors.shape) + return x_fp8, descale_factors + + +# ------------------------------- +# Misc +# ------------------------------- +def get_shape_from_layout( + x: torch.Tensor, + layout: Literal["bshd", "bhsd", "thd"], + cu_seqlens: Optional[torch.Tensor] = None, + max_seqlen: Optional[int] = None, +) -> tuple[int, int, int, int]: + if layout == "bhsd": + batch, num_heads, max_seqlen_final, head_dim = x.shape + elif layout == "bshd": + batch, max_seqlen_final, num_heads, head_dim = x.shape + elif layout == "thd": + total_seqlen, num_heads, head_dim = x.shape + if cu_seqlens is None: + raise ValueError("cu_seqlens must be provided for varlen (thd) layout") + if max_seqlen is None: + raise ValueError("max_seqlen must be provided for varlen (thd) layout") + + batch, max_seqlen_final, num_heads, head_dim = ( + len(cu_seqlens) - 1, + max_seqlen, + num_heads, + head_dim, + ) + else: + assert False, "Got unsupported layout." + + return batch, max_seqlen_final, num_heads, head_dim + + +def get_shapes_from_layout( + q, + k, + layout, + cu_seqlens_q=None, + cu_seqlens_k=None, + max_seqlen_q=None, + max_seqlen_k=None, +): + batch_q, seqlen_q, nheads_q, head_size_q = get_shape_from_layout( + q, layout, cu_seqlens_q, max_seqlen_q + ) + batch_k, seqlen_k, nheads_k, head_size_k = get_shape_from_layout( + k, layout, cu_seqlens_k, max_seqlen_k + ) + + # assert + assert batch_q == batch_k + assert head_size_q == head_size_k + + return batch_q, nheads_q, nheads_k, head_size_q, seqlen_q, seqlen_k + + +def get_stride_from_layout(x: torch.Tensor, layout: Literal["bshd", "bhsd", "thd"]): + if layout == "thd": + strides = (0, x.stride(1), x.stride(0), x.stride(2)) + elif layout == "bhsd": + strides = (x.stride(0), x.stride(1), x.stride(2), x.stride(3)) + elif layout == "bshd": + strides = (x.stride(0), x.stride(2), x.stride(1), x.stride(3)) + else: + assert False, "Got unsupported layout." + return strides + + +def get_shape_and_strides_from_layout( + x: torch.Tensor, + layout: Literal["bshd", "bhsd", "thd"], + cu_seqlens: Optional[torch.Tensor] = None, + max_seqlen: Optional[int] = None, +): + return get_shape_from_layout( + x, layout, cu_seqlens, max_seqlen + ), get_stride_from_layout(x, layout) + + +def get_strides_from_layout(q, k, v, o, layout): + q_strides = get_stride_from_layout(q, layout) + k_strides = get_stride_from_layout(k, layout) + v_strides = get_stride_from_layout(v, layout) + o_strides = get_stride_from_layout(o, layout) + return q_strides, k_strides, v_strides, o_strides + + +def get_padded_headsize(size): + # Get closest power of 2 over or equal to 32. + padded_d_model = 1 << (size - 1).bit_length() + # Smallest head_dim supported is 16. If smaller, the tile in the + # kernel is padded - there is no padding in memory for any dims. + padded_d_model = max(padded_d_model, 16) + return padded_d_model + + +def compute_alibi_tensor_ref(alibi_slopes, seqlen_q, seqlen_k): + q_idx = torch.arange(seqlen_q, dtype=torch.int32, device="cuda").unsqueeze( + -1 + ) # (N_CTX_Q, 1) + k_idx = torch.arange(seqlen_k, dtype=torch.int32, device="cuda").unsqueeze( + 0 + ) # (1, N_CTX_K) + relative_pos = torch.abs(q_idx + seqlen_k - seqlen_q - k_idx) # (N_CTX_Q, N_CTX_K) + return ( + -1 * alibi_slopes.unsqueeze(-1).unsqueeze(-1) * relative_pos + ) # (Z, H, N_CTX_Q, N_CTX_K) + + +def round_multiple(x, m): + return (x + m - 1) // m * m + + +def save_tensor_to_csv(tensor, filename, decimal_places=2): + """ + save a 2d tensor to csv file + + args: + tensor: torch tensor of shape [rows, cols] + filename: output csv filename + decimal_places: number of decimal places (default: 2) + """ + # ensure tensor is 2d + if tensor.ndim != 2: + raise ValueError(f"tensor must be 2d, got shape {tensor.shape}") + + # ensure filename ends with .csv + if not filename.endswith(".csv"): + filename = filename + ".csv" + + # save to csv using numpy + np.savetxt( + filename, + tensor.detach().cpu().numpy(), + delimiter=",", + fmt=f"%.{decimal_places}f", + ) + + +# ------------------------------- +# Dropouts +# ------------------------------- +def create_dropout_mask(dropout_p, shape, seed): + device = "cuda" + rand_vals = torch.rand( + shape, + generator=torch.Generator(device=device).manual_seed(seed), + device=device, + dtype=torch.float32, + ) + return rand_vals > dropout_p + + +def create_dropout_mask_varlen( + dropout_p, batch, nheads_q, cu_seqlens_q, cu_seqlens_k, philox_seed +): + device = "cuda" + qlens = cu_seqlens_q[1:] - cu_seqlens_q[:-1] + klens = cu_seqlens_k[1:] - cu_seqlens_k[:-1] + max_qlen = qlens.max() + max_klen = klens.max() + dropout_mask = torch.zeros((batch, nheads_q, max_qlen, max_klen), device=device) + for b in range(batch): + qlen = qlens[b] + klen = klens[b] + rand_vals = torch.rand( + (nheads_q, qlen, klen), + generator=torch.Generator(device=device).manual_seed(philox_seed), + device=device, + dtype=torch.float32, + ) + submask = rand_vals > dropout_p + dropout_mask[b, :, :qlen, :klen] = submask + + return dropout_mask + + +def write_dropout_mask(x, tensor_name="tensor"): + batch, head, seqlen_m, seqlen_n = x.shape + x = x.tolist() + + with open(f"{tensor_name}.csv", "w") as f: + writer = csv.writer(f) + for b in range(batch): + for h in range(head): + dropout_mask = x[b][h] + if True: + BLOCK_M = 64 + BLOCK_N = 64 + + # Calculate number of blocks in each dimension + m_blocks = math.ceil(seqlen_m / BLOCK_M) + n_blocks = math.ceil(seqlen_n / BLOCK_N) + + # Process each block + for m_block in range(m_blocks): + # Calculate row range for current block + row_start = m_block * BLOCK_M + row_end = min(row_start + BLOCK_M, seqlen_m) + + for n_block in range(n_blocks): + # Calculate column range for current block + col_start = n_block * BLOCK_N + col_end = min(col_start + BLOCK_N, seqlen_n) + + # Extract and write the current block + for row_idx in range(row_start, row_end): + row_data = dropout_mask[row_idx][col_start:col_end] + writer.writerow(row_data) + else: + writer.writerows(dropout_mask) + + +# ------------------------------- +# Runtime info +# ------------------------------- +@functools.cache +def is_hip(): + return triton.runtime.driver.active.get_current_target().backend == "hip" + + +@functools.cache +def get_arch(): + return triton.runtime.driver.active.get_current_target().arch + + +@functools.cache +def is_cdna(): + return is_hip() and get_arch() in ( + "gfx908", + "gfx90a", + "gfx940", + "gfx941", + "gfx942", + "gfx950", + ) + + +@functools.cache +def is_rdna(): + return is_hip() and get_arch() in ( + "gfx1030", + "gfx1100", + "gfx1101", + "gfx1102", + "gfx1200", + "gfx1201", + ) + + +@functools.cache +def arch_supports_fp8(): + return is_hip() and get_arch() in ("gfx942") diff --git a/aiter/ops/triton/mha.py b/aiter/ops/triton/mha.py old mode 100644 new mode 100755 index c9c8ed79cc..3587fbf521 --- a/aiter/ops/triton/mha.py +++ b/aiter/ops/triton/mha.py @@ -25,12 +25,24 @@ global _USE_FUSED_BWD_KERNEL _USE_FUSED_BWD_KERNEL = False +global _USE_FP8_P_DESCALE +_USE_FP8_P_DESCALE = False # Default to direct cast without descale factors + def mha_set_use_fused_bwd_kernel(value: bool): global _USE_FUSED_BWD_KERNEL _USE_FUSED_BWD_KERNEL = value +def mha_set_use_fp8_p_descale(value: bool): + """Set whether to compute descale factors for softmax probabilities in FP8. + False (default): Direct cast to FP8 without descaling + True: Compute and apply descale factors for p + """ + global _USE_FP8_P_DESCALE + _USE_FP8_P_DESCALE = value + + _USE_INT64_STRIDES = True @@ -45,6 +57,7 @@ def _cast_to_fp8( fp8_dtype, layout, clamp_val=1e-9, + group_size: int | None = None, ) -> Tuple[torch.Tensor, torch.Tensor]: """ Convert a tensor to FP8 format, returning an FP8 tensor and a descale factor. @@ -52,31 +65,36 @@ def _cast_to_fp8( - x (torch.Tensor): shape [batch, seq_len, heads, dim] Returns: - x_fp8 (torch.Tensor): FP8 tensor with the same shape as x - - descale_factor (torch.Tensor): tensor of shape [batch, 1, heads, 1] + - descale_factor (torch.Tensor): tensor of shape [batch, heads] """ if len(x.shape) != 4: raise ValueError( f"'bshd' tensor should have shape [batch, seqlen, heads, dim], got {x.shape}" ) - reduce_dims = (1, 3) # seq_len and dim dimensions - - # Compute the absolute max along reduce_dims, clamped to avoid 0-scale - x_abs_max = x.abs().amax(dim=reduce_dims) - x_abs_max = torch.maximum(x_abs_max, x.new_tensor(clamp_val)) - - # Unsqueeze back to a shape suitable for broadcast - unsqueeze_dims = sorted(reduce_dims) - for d in unsqueeze_dims: - x_abs_max = x_abs_max.unsqueeze(d) - - # compute scale and descale + batch, seqlen, nheads, dim = x.shape + if group_size is None: + # Standard per-head + x_abs_max = x.abs().amax(dim=(1, 3)) # (batch, heads) + x_abs_max = torch.maximum(x_abs_max, x.new_tensor(clamp_val)) + fp8_max = torch.finfo(fp8_dtype).max + scale = (fp8_max / x_abs_max).view(batch, 1, nheads, 1) + x_fp8 = (x * scale).to(fp8_dtype) + descale_factor = x_abs_max / fp8_max + return x_fp8, descale_factor + # Grouped path + if nheads % group_size != 0: + raise ValueError( + f"group_size {group_size} must divide number of heads {nheads} in _cast_to_fp8" + ) + ngroups = nheads // group_size + # reshape to (B,S,ngroups,group_size,D) + xg = x.view(batch, seqlen, ngroups, group_size, dim) + x_abs_max_group = xg.abs().amax(dim=(1, 3, 4)) # (B, ngroups) + x_abs_max_group = torch.maximum(x_abs_max_group, x.new_tensor(clamp_val)) fp8_max = torch.finfo(fp8_dtype).max - scale = fp8_max / x_abs_max - descale_factor = x_abs_max / fp8_max - - # cast to FP8, optionally setting requires_grad - x_fp8 = (x * scale).to(fp8_dtype) - + scale_group = (fp8_max / x_abs_max_group).view(batch, 1, ngroups, 1, 1) + x_fp8 = (xg * scale_group).to(fp8_dtype).view(batch, seqlen, nheads, dim) + descale_factor = x_abs_max_group / fp8_max # (B, ngroups) return x_fp8, descale_factor @@ -85,6 +103,7 @@ def _cast_varlen_to_fp8( fp8_dtype: torch.dtype, cu_seqlens, clamp_val: float = 1e-9, + group_size: int | None = None, ) -> tuple[torch.Tensor, torch.Tensor]: """ Convert a tensor of sequences with variable seq_len into fp8. @@ -107,8 +126,13 @@ def _cast_varlen_to_fp8( # Compute scale and descale factors per sequence x_fp8 = torch.zeros_like(x, dtype=fp8_dtype) + if group_size is not None and num_heads % group_size != 0: + raise ValueError( + f"group_size {group_size} must divide number of heads {num_heads} in _cast_varlen_to_fp8" + ) + out_heads = num_heads if group_size is None else num_heads // group_size descale_factors = torch.zeros( - (batch, num_heads), device=x.device, dtype=torch.float32 + (batch, out_heads), device=x.device, dtype=torch.float32 ) for i in range(batch): @@ -116,23 +140,28 @@ def _cast_varlen_to_fp8( end = cu_seqlens[i + 1] x_slice = x[start:end] # Slice for current sequence - # Standard tensor (0: seq_len, 2: head_dim) - x_abs_max = x_slice.abs().amax(dim=(0, 2)) # [heads] - - # apply minimum clamping - x_abs_max = torch.maximum(x_abs_max, x.new_tensor(clamp_val)) - - # compute scale and descale factors - scale_i = fp8_max / x_abs_max - descale_i = x_abs_max / fp8_max - - # store descale factors - descale_factors[i, :] = descale_i - - scale_reshape = scale_i.reshape(1, num_heads, 1) - - # scale and cast to FP8 - x_fp8[start:end] = (x_slice * scale_reshape).to(fp8_dtype) + if group_size is None: + x_abs_max = x_slice.abs().amax(dim=(0, 2)) # (heads) + x_abs_max = torch.maximum(x_abs_max, x.new_tensor(clamp_val)) + scale_i = fp8_max / x_abs_max + descale_i = x_abs_max / fp8_max + descale_factors[i, :] = descale_i + scale_reshape = scale_i.view(1, num_heads, 1) + x_fp8[start:end] = (x_slice * scale_reshape).to(fp8_dtype) + else: + ngroups = num_heads // group_size + xg = x_slice.view(end - start, ngroups, group_size, x_slice.shape[2]) + x_abs_max_group = xg.abs().amax(dim=(0, 2, 3)) # (ngroups) + x_abs_max_group = torch.maximum(x_abs_max_group, x.new_tensor(clamp_val)) + scale_group = fp8_max / x_abs_max_group + descale_group = x_abs_max_group / fp8_max + descale_factors[i, :] = descale_group + scale_group_reshape = scale_group.view(1, ngroups, 1, 1) + x_fp8[start:end] = ( + (xg * scale_group_reshape) + .to(fp8_dtype) + .view(end - start, num_heads, x_slice.shape[2]) + ) return x_fp8, descale_factors @@ -237,6 +266,7 @@ def _attn_fwd_inner( PADDED_HEAD: tl.constexpr, IS_FP8: tl.constexpr, FP8_MAX: tl.constexpr, + FP8_P_DESCALE: tl.constexpr = False, ): RCP_LN2: tl.constexpr = 1.4426950408889634 @@ -289,7 +319,7 @@ def _attn_fwd_inner( if IS_CAUSAL: causal_boundary = start_n + offs_n_causal causal_mask = OFFS_M[:, None] >= causal_boundary[None, :] - mask = mask and causal_mask + mask = mask & causal_mask qk = tl.where(mask, qk, float("-inf")) @@ -343,10 +373,18 @@ def _attn_fwd_inner( m_i = m_ij if IS_FP8: - scale_p, descale_p = _compute_fp8_scaling_factors(p, FP8_MAX) - acc += ( - tl.dot((p * scale_p).to(v.type.element_ty), v) * descale_p * descale_v - ) + if FP8_P_DESCALE: + # compute and use descale factors for p + scale_p, descale_p = _compute_fp8_scaling_factors(p, FP8_MAX) + acc += ( + tl.dot((p * scale_p).to(v.type.element_ty), v) + * descale_p + * descale_v + ) + else: + # direct cast p to fp8 without computing descale factors + p_fp8 = p.to(v.type.element_ty) + acc += tl.dot(p_fp8, v) * descale_v else: acc += tl.dot(p.to(v.type.element_ty), v) @@ -422,6 +460,7 @@ def _attn_fwd( ENABLE_DROPOUT: tl.constexpr, IS_FP8: tl.constexpr, FP8_MAX: tl.constexpr, + FP8_P_DESCALE: tl.constexpr, VARLEN: tl.constexpr, BATCH, NUM_XCD: tl.constexpr, @@ -742,6 +781,7 @@ def _attn_fwd( PADDED_HEAD=BLOCK_DMODEL != BLOCK_DMODEL_POW2, IS_FP8=IS_FP8, FP8_MAX=FP8_MAX, + FP8_P_DESCALE=FP8_P_DESCALE, ) block_min = block_max block_max = n_blocks * BLOCK_N @@ -758,6 +798,7 @@ def _attn_fwd( s_dmask_ptrs += n_full_blocks * BLOCK_N * stride_sd_n if ENABLE_DROPOUT: dropout_mask_ptrs += n_full_blocks * BLOCK_N * stride_sd_n + philox_ptrs += n_full_blocks * BLOCK_N * stride_sd_n acc, l_i, m_i = _attn_fwd_inner( acc, l_i, @@ -799,6 +840,7 @@ def _attn_fwd( PADDED_HEAD=BLOCK_DMODEL != BLOCK_DMODEL_POW2, IS_FP8=IS_FP8, FP8_MAX=FP8_MAX, + FP8_P_DESCALE=FP8_P_DESCALE, ) # epilogue # This helps the compiler do Newton Raphson on l_i vs on acc which is much larger. @@ -913,9 +955,9 @@ def _flash_attn_forward( max_seqlen_k: int, cu_seqlens_q: Optional[torch.Tensor] = None, cu_seqlens_k: Optional[torch.Tensor] = None, - descale_q: Optional[torch.Tensor] = None, - descale_k: Optional[torch.Tensor] = None, - descale_v: Optional[torch.Tensor] = None, + q_descale: Optional[torch.Tensor] = None, + k_descale: Optional[torch.Tensor] = None, + v_descale: Optional[torch.Tensor] = None, config: Optional[dict[str, any]] = None, ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]: @@ -1036,9 +1078,9 @@ def _flash_attn_forward( q, k, v, - descale_q, - descale_k, - descale_v, + q_descale, + k_descale, + v_descale, o, alibi_slopes, s_dmask, @@ -1047,9 +1089,9 @@ def _flash_attn_forward( *q_strides, *k_strides, *v_strides, - descale_q.stride(0) if descale_q is not None else 0, - descale_k.stride(0) if descale_k is not None else 0, - descale_v.stride(0) if descale_v is not None else 0, + q_descale.stride(0) if q_descale is not None else 0, + k_descale.stride(0) if k_descale is not None else 0, + v_descale.stride(0) if v_descale is not None else 0, *o_strides, alibi_slopes.stride(0) if alibi_slopes is not None else 0, alibi_slopes.stride(1) if alibi_slopes is not None else 0, @@ -1077,6 +1119,7 @@ def _flash_attn_forward( ENABLE_DROPOUT=enable_dropout, IS_FP8=IS_FP8, FP8_MAX=FP8_MAX, + FP8_P_DESCALE=_USE_FP8_P_DESCALE, VARLEN=is_varlen, BATCH=batch, NUM_XCD=get_num_xcds(), @@ -1321,221 +1364,6 @@ def flash_attn_func( ) -class _FlashAttnFP8Func(torch.autograd.Function): - @staticmethod - def forward( - ctx, - q, - k, - v, - dropout_p, - softmax_scale, - causal, - window_size, - alibi_slopes, - deterministic, - return_lse, - return_softmax, - is_grad_enabled, - config=None, - ): - is_grad = is_grad_enabled and any(x.requires_grad for x in [q, k, v]) - if softmax_scale is None: - softmax_scale = q.shape[-1] ** (-0.5) - head_size_og = q.size(3) - if head_size_og % 8 != 0: - q = torch.nn.functional.pad(q, [0, 8 - head_size_og % 8]) - k = torch.nn.functional.pad(k, [0, 8 - head_size_og % 8]) - v = torch.nn.functional.pad(v, [0, 8 - head_size_og % 8]) - - # cast input to fp8 - fp8_dtype = arch_info.get_fp8_e4m3_dtype() - q_fp8, descale_q = _cast_to_fp8(q, fp8_dtype, "bshd") - k_fp8, descale_k = _cast_to_fp8(k, fp8_dtype, "bshd") - v_fp8, descale_v = _cast_to_fp8(v, fp8_dtype, "bshd") - - out_padded, softmax_lse, S_dmask, philox_seed, philox_offset = ( - _flash_attn_forward( - q_fp8, - k_fp8, - v_fp8, - dropout_p, - softmax_scale, - causal=causal, - window_size_left=int(window_size[0]), - window_size_right=int(window_size[1]), - bias=None, - alibi_slopes=alibi_slopes, - return_lse=return_lse, - return_softmax=return_softmax and dropout_p > 0, - max_seqlen_q=q.shape[1], - max_seqlen_k=k.shape[1], - cu_seqlens_q=None, - cu_seqlens_k=None, - descale_q=descale_q, - descale_k=descale_k, - descale_v=descale_v, - config=config, - ) - ) - - if is_grad: - ctx.save_for_backward( - q_fp8, - k_fp8, - v_fp8, - out_padded, - softmax_lse, - descale_q, - descale_k, - descale_v, - ) - ctx.philox_seed = philox_seed - ctx.philox_offset = philox_offset - ctx.dropout_p = dropout_p - ctx.softmax_scale = softmax_scale - ctx.causal = causal - ctx.window_size = window_size - ctx.alibi_slopes = alibi_slopes - - out = out_padded[..., :head_size_og] - result = [out] - if return_lse: - result.append(softmax_lse) - if return_softmax: - result.append(S_dmask) - - return result[0] if len(result) == 1 else tuple(result) - - @staticmethod - def backward(ctx, do, *args): - q_fp8, k_fp8, v_fp8, out, softmax_lse, descale_q, descale_k, descale_v = ( - ctx.saved_tensors - ) - dq, dk, dv = ( - torch.zeros_like(q_fp8, dtype=torch.float32), - torch.zeros_like(k_fp8, dtype=torch.float32), - torch.zeros_like(v_fp8, dtype=torch.float32), - ) - head_size_v_og = do.size(3) - do_padded = do - if head_size_v_og % 8 != 0: - do_padded = torch.nn.functional.pad(do, [0, 8 - head_size_v_og % 8]) - - fp8_dtype = arch_info.get_fp8_e4m3_dtype() - do_padded_fp8, descale_do = _cast_to_fp8(do_padded, fp8_dtype, "bshd") - if _USE_FUSED_BWD_KERNEL: - flash_attn_fused_backward( - do_padded_fp8, - q_fp8, - k_fp8, - v_fp8, - out, - softmax_lse, - dq, - dk, - dv, - None, - ctx.softmax_scale, - ctx.alibi_slopes, - ctx.causal, - None, - None, - max_seqlen_q=q_fp8.shape[1], - max_seqlen_k=k_fp8.shape[1], - dropout_p=ctx.dropout_p, - philox_seed=ctx.philox_seed, - philox_offset=ctx.philox_offset, - descale_q=descale_q, - descale_k=descale_k, - descale_v=descale_v, - descale_do=descale_do, - USE_INT64_STRIDES=_USE_INT64_STRIDES, - ) - else: - flash_attn_onekernel_backward( - do_padded_fp8, - q_fp8, - k_fp8, - v_fp8, - out, - softmax_lse, - dq, - dk, - dv, - None, - ctx.softmax_scale, - ctx.alibi_slopes, - ctx.causal, - None, - None, - max_seqlen_q=q_fp8.shape[1], - max_seqlen_k=k_fp8.shape[1], - dropout_p=ctx.dropout_p, - philox_seed=ctx.philox_seed, - philox_offset=ctx.philox_offset, - descale_q=descale_q, - descale_k=descale_k, - descale_v=descale_v, - descale_do=descale_do, - USE_INT64_STRIDES=_USE_INT64_STRIDES, - ) - - # dq = dq[..., : q_fp8.shape[-1]] # We could have padded the head dimension - # dk = dk[..., : k_fp8.shape[-1]] - # dv = dv[..., : v_fp8.shape[-1]] - return ( - dq, - dk, - dv, - None, - None, - None, - None, - None, - None, - None, - None, - None, - None, - None, - ) - - -def flash_attn_fp8_func( - q, - k, - v, - dropout_p=0.0, - softmax_scale=None, - causal=False, - window_size=(-1, -1), # -1 means infinite context window - alibi_slopes=None, - deterministic=False, - return_lse=False, - return_attn_probs=False, - config: Optional[dict[str, any]] = None, -): - _LOGGER.info( - f"FLASH_ATTN_FP8: q={tuple(q.shape)} k={tuple(k.shape)} v={tuple(v.shape)}" - ) - return _FlashAttnFP8Func.apply( - q, - k, - v, - dropout_p, - softmax_scale, - causal, - window_size, - alibi_slopes, - deterministic, - return_lse, - return_attn_probs, - torch.is_grad_enabled(), - config, - ) - - class _FlashAttnVarlenFunc(torch.autograd.Function): @staticmethod def forward( @@ -1804,231 +1632,3 @@ def flash_attn_varlen_func( torch.is_grad_enabled(), config, ) - - -class _FlashAttnVarlenFP8Func(torch.autograd.Function): - @staticmethod - def forward( - ctx, - q, - k, - v, - cu_seqlens_q, - cu_seqlens_k, - max_seqlen_q, - max_seqlen_k, - dropout_p, - softmax_scale, - causal, - window_size, - alibi_slopes, - deterministic, - return_lse, - return_softmax, - block_table, - is_grad_enabled, - config=None, - ): - is_grad = is_grad_enabled and any(x.requires_grad for x in [q, k, v]) - if softmax_scale is None: - softmax_scale = q.shape[-1] ** (-0.5) - head_size_og = q.size(2) - if head_size_og % 8 != 0: - q = torch.nn.functional.pad(q, [0, 8 - head_size_og % 8]) - k = torch.nn.functional.pad(k, [0, 8 - head_size_og % 8]) - v = torch.nn.functional.pad(v, [0, 8 - head_size_og % 8]) - - # cast input to fp8 - fp8_dtype = arch_info.get_fp8_e4m3_dtype() - q_fp8, descale_q = _cast_varlen_to_fp8(q, fp8_dtype, cu_seqlens=cu_seqlens_q) - k_fp8, descale_k = _cast_varlen_to_fp8(k, fp8_dtype, cu_seqlens=cu_seqlens_k) - v_fp8, descale_v = _cast_varlen_to_fp8(v, fp8_dtype, cu_seqlens=cu_seqlens_k) - - out_padded, softmax_lse, S_dmask, philox_seed, philox_offset = ( - _flash_attn_forward( - q_fp8, - k_fp8, - v_fp8, - dropout_p, - softmax_scale, - causal=causal, - window_size_left=int(window_size[0]), - window_size_right=int(window_size[1]), - bias=None, - alibi_slopes=alibi_slopes, - return_lse=return_lse, - return_softmax=return_softmax and dropout_p > 0, - max_seqlen_q=max_seqlen_q, - max_seqlen_k=max_seqlen_k, - cu_seqlens_q=cu_seqlens_q, - cu_seqlens_k=cu_seqlens_k, - descale_q=descale_q, - descale_k=descale_k, - descale_v=descale_v, - config=config, - ) - ) - if is_grad: - ctx.save_for_backward( - q_fp8, - k_fp8, - v_fp8, - out_padded, - softmax_lse, - cu_seqlens_q, - cu_seqlens_k, - descale_q, - descale_k, - descale_v, - ) - ctx.max_seqlen_q = max_seqlen_q - ctx.max_seqlen_k = max_seqlen_k - ctx.philox_seed = philox_seed - ctx.philox_offset = philox_offset - ctx.dropout_p = dropout_p - ctx.softmax_scale = softmax_scale - ctx.causal = causal - ctx.window_size = window_size - ctx.alibi_slopes = alibi_slopes - - out = out_padded[..., :head_size_og] - result = [out] - if return_lse: - result.append(softmax_lse) - if return_softmax: - result.append(S_dmask) - - return result[0] if len(result) == 1 else tuple(result) - - @staticmethod - def backward(ctx, do, *args): - ( - q_fp8, - k_fp8, - v_fp8, - out, - softmax_lse, - cu_seqlens_q, - cu_seqlens_k, - descale_q, - descale_k, - descale_v, - ) = ctx.saved_tensors - dq, dk, dv = ( - torch.zeros_like(q_fp8, dtype=torch.float32), - torch.zeros_like(k_fp8, dtype=torch.float32), - torch.zeros_like(v_fp8, dtype=torch.float32), - ) - head_size_v_og = do.size(3) - do_padded = do - if head_size_v_og % 8 != 0: - do_padded = torch.nn.functional.pad(do, [0, 8 - head_size_v_og % 8]) - - fp8_dtype = arch_info.get_fp8_e4m3_dtype() - do_padded_fp8, descale_do = _cast_varlen_to_fp8( - do_padded, fp8_dtype, "thd", cu_seqlens_q - ) - if _USE_FUSED_BWD_KERNEL: - flash_attn_fused_backward( - do_padded_fp8, - q_fp8, - k_fp8, - v_fp8, - out, - softmax_lse, - dq, - dk, - dv, - None, - ctx.softmax_scale, - ctx.alibi_slopes, - ctx.causal, - cu_seqlens_q, - cu_seqlens_k, - max_seqlen_q=ctx.max_seqlen_q, - max_seqlen_k=ctx.max_seqlen_k, - dropout_p=ctx.dropout_p, - philox_seed=ctx.philox_seed, - philox_offset=ctx.philox_offset, - descale_q=descale_q, - descale_k=descale_k, - descale_v=descale_v, - descale_do=descale_do, - USE_INT64_STRIDES=_USE_INT64_STRIDES, - ) - else: - flash_attn_onekernel_backward( - do_padded_fp8, - q_fp8, - k_fp8, - v_fp8, - out, - softmax_lse, - dq, - dk, - dv, - None, - ctx.softmax_scale, - ctx.alibi_slopes, - ctx.causal, - cu_seqlens_q, - cu_seqlens_k, - max_seqlen_q=ctx.max_seqlen_q, - max_seqlen_k=ctx.max_seqlen_k, - dropout_p=ctx.dropout_p, - philox_seed=ctx.philox_seed, - philox_offset=ctx.philox_offset, - descale_q=descale_q, - descale_k=descale_k, - descale_v=descale_v, - descale_do=descale_do, - USE_INT64_STRIDES=_USE_INT64_STRIDES, - ) - dq = dq[..., : q_fp8.shape[-1]] # We could have padded the head dimension - dk = dk[..., : k_fp8.shape[-1]] - dv = dv[..., : v_fp8.shape[-1]] - return dq, dk, dv, None, None, None, None, None, None, None, None, None, None - - -def flash_attn_varlen_fp8_func( - q, - k, - v, - cu_seqlens_q, - cu_seqlens_k, - max_seqlen_q, - max_seqlen_k, - dropout_p=0.0, - softmax_scale=None, - causal=False, - window_size=(-1, -1), # -1 means infinite context window - alibi_slopes=None, - deterministic=False, - return_lse=False, - return_attn_probs=False, - block_table=None, - config: Optional[dict[str, any]] = None, -): - _LOGGER.info( - f"FLASH_ATTN_VARLEN_FP8: q={tuple(q.shape)} k={tuple(k.shape)} v={tuple(v.shape)}" - ) - return _FlashAttnVarlenFP8Func.apply( - q, - k, - v, - cu_seqlens_q, - cu_seqlens_k, - max_seqlen_q, - max_seqlen_k, - dropout_p, - softmax_scale, - causal, - window_size, - alibi_slopes, - deterministic, - return_lse, - return_attn_probs, - block_table, - torch.is_grad_enabled(), - config, - ) diff --git a/aiter/ops/triton/mha_v3.py b/aiter/ops/triton/mha_v3.py new file mode 100644 index 0000000000..b6b250aa19 --- /dev/null +++ b/aiter/ops/triton/mha_v3.py @@ -0,0 +1,484 @@ +# SPDX-License-Identifier: MIT +# Copyright (C) 2024-2025, Advanced Micro Devices, Inc. All rights reserved. + +from __future__ import annotations +from typing import Optional, Tuple +import torch + +from aiter.ops.triton.flash_attn_triton_amd import flash_attn_3 + + +class _FlashAttnV3Func(torch.autograd.Function): # Backend (native) path + @staticmethod + def forward( + ctx, + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + softmax_scale: float | None, + causal: bool, + qv: Optional[torch.Tensor], + q_descale: Optional[torch.Tensor], + k_descale: Optional[torch.Tensor], + v_descale: Optional[torch.Tensor], + window_size: Tuple[int, int], + attention_chunk: int, + softcap: float, + num_splits: int, + pack_gqa: Optional[bool], + deterministic: bool, + sm_margin: int, + ): + # Derive softmax scale if not provided (include qv width like Hopper v3) + if softmax_scale is None: + q_extra = qv.shape[-1] if qv is not None else 0 + softmax_scale = (q.shape[-1] + q_extra) ** (-0.5) + + # Fast validation of unsupported features + if qv is not None: + raise NotImplementedError("qv is not supported in AMD Triton v3 yet") + if attention_chunk not in (0, 1): + raise NotImplementedError("attention_chunk > 1 not supported (0 or 1 only)") + if softcap != 0.0: + raise NotImplementedError("softcap not implemented in AMD Triton v3") + if num_splits != 1: + raise NotImplementedError("num_splits != 1 not supported in AMD Triton v3") + if pack_gqa is not None: + raise NotImplementedError("pack_gqa not implemented in AMD Triton v3") + if sm_margin != 0: + raise NotImplementedError("sm_margin != 0 not supported in AMD Triton v3") + + requires_grad = ( + any(t.requires_grad for t in (q, k, v)) and torch.is_grad_enabled() + ) + + # Forward: basic path (no varlen / kv-cache parameters used here) + out, softmax_lse = flash_attn_3.fwd( + q, + k, + v, + None, # k_new + None, # v_new + None, # qv + None, # out tensor (allocate inside) + None, # cu_seqlens_q + None, # cu_seqlens_k + None, # cu_seqlens_k_new + None, # seqused_q + None, # seqused_k + None, # max_seqlen_q + None, # max_seqlen_k + None, # page_table + None, # kv_batch_idx + None, # leftpad_k + None, # rotary_cos + None, # rotary_sin + None, # seqlens_rotary + q_descale, + k_descale, + v_descale, + softmax_scale, + causal, + int(window_size[0]), + int(window_size[1]), + attention_chunk, + softcap, + False, # rotary_interleaved + None, # scheduler_metadata + num_splits, + pack_gqa, + sm_margin, + ) + + if requires_grad: + ctx.save_for_backward(q, k, v, out, softmax_lse) + ctx.softmax_scale = softmax_scale + ctx.causal = causal + ctx.window_size = window_size + ctx.softcap = softcap + ctx.deterministic = deterministic + ctx.sm_margin = sm_margin + return out + + @staticmethod + def backward(ctx, dout: torch.Tensor): + q, k, v, out, softmax_lse = ctx.saved_tensors + dq, dk, dv, _delta = flash_attn_3.bwd( + dout, + q, + k, + v, + out, + softmax_lse, + None, # dq + None, # dk + None, # dv + None, # cu_seqlens_q + None, # cu_seqlens_k + None, # seqused_q + None, # seqused_k + None, # max_seqlen_q + None, # max_seqlen_k + ctx.softmax_scale, + ctx.causal, + int(ctx.window_size[0]), + int(ctx.window_size[1]), + ctx.softcap, + ctx.deterministic, + ctx.sm_margin, + ) + return ( + dq, # q + dk, # k + dv, # v + None, # softmax_scale + None, # causal + None, # qv + None, # q_descale + None, # k_descale + None, # v_descale + None, # window_size + None, # attention_chunk + None, # softcap + None, # num_splits + None, # pack_gqa + None, # deterministic + None, # sm_margin + ) + + +def flash_attn_func( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + softmax_scale: float | None = None, + causal: bool = False, + qv: torch.Tensor | None = None, + q_descale: torch.Tensor | None = None, + k_descale: torch.Tensor | None = None, + v_descale: torch.Tensor | None = None, + window_size: tuple[int, int] = (-1, -1), + attention_chunk: int = 0, + softcap: float = 0.0, + num_splits: int = 1, + pack_gqa: bool | None = None, + deterministic: bool = False, + sm_margin: int = 0, +): + """FlashAttention v3 entry point.""" + return _FlashAttnV3Func.apply( + q, + k, + v, + softmax_scale, + causal, + qv, + q_descale, + k_descale, + v_descale, + window_size, + attention_chunk, + softcap, + num_splits, + pack_gqa, + deterministic, + sm_margin, + ) + + +class _FlashAttnVarlenV3Func(torch.autograd.Function): + @staticmethod + def forward( + ctx, + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + cu_seqlens_q: torch.Tensor, + cu_seqlens_k: torch.Tensor, + max_seqlen_q: int, + max_seqlen_k: int, + softmax_scale: float | None, + causal: bool, + q_descale: torch.Tensor | None, + k_descale: torch.Tensor | None, + v_descale: torch.Tensor | None, + window_size: tuple[int, int], + attention_chunk: int, + softcap: float, + deterministic: bool, + sm_margin: int, + ): + if softmax_scale is None: + softmax_scale = q.shape[-1] ** (-0.5) + if attention_chunk != 0: + raise NotImplementedError( + "attention_chunk != 0 not supported in varlen v3 yet" + ) + if softcap != 0.0: + raise NotImplementedError("softcap not implemented in varlen v3 yet") + if sm_margin != 0: + raise NotImplementedError("sm_margin != 0 not supported in varlen v3 yet") + + requires_grad = ( + any(t.requires_grad for t in (q, k, v)) and torch.is_grad_enabled() + ) + + out, softmax_lse = flash_attn_3.fwd( + q, + k, + v, + None, # k_new + None, # v_new + None, # qv + None, # out tensor + cu_seqlens_q, + cu_seqlens_k, + None, # cu_seqlens_k_new + None, # seqused_q + None, # seqused_k + max_seqlen_q, + max_seqlen_k, + None, # page_table + None, # kv_batch_idx + None, # leftpad_k + None, # rotary_cos + None, # rotary_sin + None, # seqlens_rotary + q_descale, + k_descale, + v_descale, + softmax_scale, + causal, + int(window_size[0]), + int(window_size[1]), + attention_chunk, + softcap, + False, # rotary_interleaved + None, # scheduler_metadata + 1, # num_splits + None, # pack_gqa + sm_margin, + ) + + if requires_grad: + ctx.save_for_backward(q, k, v, out, softmax_lse) + ctx.softmax_scale = softmax_scale + ctx.causal = causal + ctx.window_size = window_size + ctx.softcap = softcap + ctx.deterministic = deterministic + ctx.sm_margin = sm_margin + ctx.cu_seqlens_q = cu_seqlens_q + ctx.cu_seqlens_k = cu_seqlens_k + ctx.max_seqlen_q = max_seqlen_q + ctx.max_seqlen_k = max_seqlen_k + return out + + @staticmethod + def backward(ctx, dout: torch.Tensor): + q, k, v, out, softmax_lse = ctx.saved_tensors + dq, dk, dv, _delta = flash_attn_3.bwd( + dout, + q, + k, + v, + out, + softmax_lse, + None, # dq + None, # dk + None, # dv + ctx.cu_seqlens_q, + ctx.cu_seqlens_k, + None, # seqused_q + None, # seqused_k + ctx.max_seqlen_q, + ctx.max_seqlen_k, + ctx.softmax_scale, + ctx.causal, + int(ctx.window_size[0]), + int(ctx.window_size[1]), + ctx.softcap, + ctx.deterministic, + ctx.sm_margin, + ) + return ( + dq, + dk, + dv, + None, # cu_seqlens_q + None, # cu_seqlens_k + None, # max_seqlen_q + None, # max_seqlen_k + None, # softmax_scale + None, # causal + None, # q_descale + None, # k_descale + None, # v_descale + None, # window_size + None, # attention_chunk + None, # softcap + None, # deterministic + None, # sm_margin + ) + + +def flash_attn_varlen_func( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + cu_seqlens_q: torch.Tensor, + cu_seqlens_k: torch.Tensor, + max_seqlen_q: int, + max_seqlen_k: int, + softmax_scale: float | None = None, + causal: bool = False, + q_descale: torch.Tensor | None = None, + k_descale: torch.Tensor | None = None, + v_descale: torch.Tensor | None = None, + window_size: tuple[int, int] = (-1, -1), + attention_chunk: int = 0, + softcap: float = 0.0, + deterministic: bool = False, + sm_margin: int = 0, +): + """FlashAttention v3 varlen path.""" + return _FlashAttnVarlenV3Func.apply( + q, + k, + v, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + softmax_scale, + causal, + q_descale, + k_descale, + v_descale, + window_size, + attention_chunk, + softcap, + deterministic, + sm_margin, + ) + + +def flash_attn_with_kvcache( + q: torch.Tensor, + k_cache: torch.Tensor, + v_cache: torch.Tensor, + k: torch.Tensor | None = None, + v: torch.Tensor | None = None, + qv: torch.Tensor | None = None, + cache_seqlens: torch.Tensor | int | None = None, + softmax_scale: float | None = None, + causal: bool = True, + window_size: tuple[int, int] = (-1, -1), + attention_chunk: int = 0, + softcap: float = 0.0, + num_splits: int = 0, + pack_gqa: bool | None = None, + sm_margin: int = 0, + q_descale: torch.Tensor | None = None, + k_descale: torch.Tensor | None = None, + v_descale: torch.Tensor | None = None, + max_seqlen_q: int | None = None, + return_softmax_lse: bool = False, + page_table: torch.Tensor | None = None, + cache_batch_idx: torch.Tensor | None = None, + cache_leftpad: torch.Tensor | None = None, + rotary_cos: torch.Tensor | None = None, + rotary_sin: torch.Tensor | None = None, + rotary_seqlens: torch.Tensor | None = None, + cu_seqlens_q: torch.Tensor | None = None, + cu_seqlens_k_new: torch.Tensor | None = None, +): + """FlashAttention v3 decode + KV cache update helper (forward only, no autograd wrapper). + + Arguments mirror Hopper's `flash_attn_with_kvcache` with current backend limitations. + Unsupported: backward, qv, softcap!=0, pack_gqa, sm_margin!=0, attention_chunk>1, num_splits>1, + simultaneous varlen (cu_seqlens_q) + cache_seqlens tensor, and partial rotary inputs. + """ + # Scale + if softmax_scale is None: + q_extra = qv.shape[-1] if qv is not None else 0 + softmax_scale = (q.shape[-1] + q_extra) ** (-0.5) + + # Feature guards + if qv is not None: + raise NotImplementedError("qv not supported in KV cache path yet") + if softcap != 0.0: + raise NotImplementedError("softcap not implemented in KV cache path") + if pack_gqa is not None: + raise NotImplementedError("pack_gqa not implemented in KV cache path") + if sm_margin != 0: + raise NotImplementedError("sm_margin != 0 not supported in KV cache path") + if attention_chunk not in (0, 1): + raise NotImplementedError("attention_chunk > 1 not supported (0 or 1 only)") + if num_splits not in (0, 1): + raise NotImplementedError("num_splits > 1 not supported in KV cache path") + + if cache_seqlens is not None and isinstance(cache_seqlens, int): + cache_seqlens = torch.full( + (k_cache.shape[0],), cache_seqlens, dtype=torch.int32, device=k_cache.device + ) + + if cu_seqlens_q is not None and cache_seqlens is not None: + raise NotImplementedError( + "Varlen decode with cache_seqlens tensor not supported yet" + ) + if (rotary_cos is None) ^ (rotary_sin is None): + raise ValueError( + "Both rotary_cos and rotary_sin must be provided together or neither" + ) + if ( + (rotary_cos is not None) + and rotary_seqlens is not None + and cu_seqlens_q is None + and cache_seqlens is None + ): + raise ValueError( + "rotary_seqlens provided without cu_seqlens_q or cache_seqlens context" + ) + + kv_batch_idx = cache_batch_idx + leftpad_k = cache_leftpad + seqlens_rotary = rotary_seqlens + + out, softmax_lse = flash_attn_3.fwd( + q, + k_cache, + v_cache, + k, + v, + None, # qv + None, # out allocate + cu_seqlens_q, + None, # cu_seqlens_k + cu_seqlens_k_new, + None, # seqused_q + cache_seqlens if isinstance(cache_seqlens, torch.Tensor) else None, # seqused_k + max_seqlen_q, + None, # max_seqlen_k + page_table, + kv_batch_idx, + leftpad_k, + rotary_cos, + rotary_sin, + seqlens_rotary, + q_descale, + k_descale, + v_descale, + softmax_scale, + causal, + int(window_size[0]), + int(window_size[1]), + attention_chunk, + softcap, + False, # rotary_interleaved + None, # scheduler_metadata + num_splits if num_splits != 0 else 1, + pack_gqa, + sm_margin, + ) + return (out, softmax_lse) if return_softmax_lse else out diff --git a/op_tests/op_benchmarks/triton/bench_mha.py b/op_tests/op_benchmarks/triton/bench_mha.py index e661d0119f..e8612c50a1 100644 --- a/op_tests/op_benchmarks/triton/bench_mha.py +++ b/op_tests/op_benchmarks/triton/bench_mha.py @@ -6,11 +6,16 @@ import triton from aiter.ops.triton.mha import ( flash_attn_func, - flash_attn_fp8_func, flash_attn_varlen_func, - flash_attn_varlen_fp8_func, mha_set_use_fused_bwd_kernel, + _cast_to_fp8, + _cast_varlen_to_fp8, ) +from aiter.ops.triton.mha_v3 import ( + flash_attn_func as flash_attn_func_v3, + flash_attn_varlen_func as flash_attn_varlen_func_v3, +) +from aiter.ops.triton.utils.arch_info import get_fp8_e4m3_dtype from aiter.test_mha_common import ( generate_random_padding_mask, generate_qkv, @@ -363,19 +368,35 @@ def bench_mha( if args.fp8: def fn(): - return flash_attn_varlen_fp8_func( - q_input, - k_input, - v_input, + if dropout > 0.0 or return_lse or return_attn_probs: + raise NotImplementedError( + "flash_attn_varlen_func_v3 doesn't support dropout, return_lse, or return_attn_probs" + ) + + fp8_dtype = get_fp8_e4m3_dtype() + q_fp8, q_descale = _cast_varlen_to_fp8( + q_input, fp8_dtype, cu_seqlens_q + ) + k_fp8, k_descale = _cast_varlen_to_fp8( + k_input, fp8_dtype, cu_seqlens_k + ) + v_fp8, v_descale = _cast_varlen_to_fp8( + v_input, fp8_dtype, cu_seqlens_k + ) + + return flash_attn_varlen_func_v3( + q_fp8, + k_fp8, + v_fp8, cu_seqlens_q, cu_seqlens_k, max_seqlen_q, max_seqlen_k, - dropout_p=dropout, softmax_scale=sm_scale, causal=causal, - return_lse=return_lse, - return_attn_probs=return_attn_probs, + q_descale=q_descale, + k_descale=k_descale, + v_descale=v_descale, ) else: @@ -400,15 +421,25 @@ def fn(): if args.fp8: def fn(): - return flash_attn_fp8_func( - q_input, - k_input, - v_input, - dropout_p=dropout, + if dropout > 0.0 or return_lse or return_attn_probs: + raise NotImplementedError( + "flash_attn_func (v3) doesn't support dropout, return_lse, or return_attn_probs" + ) + + fp8_dtype = get_fp8_e4m3_dtype() + q_fp8, q_descale = _cast_to_fp8(q_input, fp8_dtype, "bshd") + k_fp8, k_descale = _cast_to_fp8(k_input, fp8_dtype, "bshd") + v_fp8, v_descale = _cast_to_fp8(v_input, fp8_dtype, "bshd") + + return flash_attn_func_v3( + q_fp8, + k_fp8, + v_fp8, softmax_scale=sm_scale, causal=causal, - return_lse=return_lse, - return_attn_probs=return_attn_probs, + q_descale=q_descale, + k_descale=k_descale, + v_descale=v_descale, ) else: diff --git a/op_tests/triton_tests/test_mha.py b/op_tests/triton_tests/test_mha.py old mode 100644 new mode 100755 index 4e6635915f..ffaf4bc597 --- a/op_tests/triton_tests/test_mha.py +++ b/op_tests/triton_tests/test_mha.py @@ -7,12 +7,18 @@ import numpy as np from aiter.ops.triton.mha import ( flash_attn_func, - flash_attn_fp8_func, flash_attn_varlen_func, - flash_attn_varlen_fp8_func, mha_set_use_fused_bwd_kernel, mha_set_use_int64_strides, + _cast_to_fp8, + _cast_varlen_to_fp8, ) +from aiter.ops.triton.mha_v3 import ( + flash_attn_func as flash_attn_func_v3, + flash_attn_varlen_func as flash_attn_varlen_func_v3, + flash_attn_with_kvcache as flash_attn_with_kvcache_v3, +) +from aiter.ops.triton.utils.arch_info import get_fp8_e4m3_dtype from aiter.test_mha_common import ( attention_ref, generate_random_padding_mask, @@ -131,14 +137,26 @@ def test_mha( dropout_mask = None if FP8: - triton_out = flash_attn_fp8_func( - q, - k, - v, - dropout_p=DROPOUT, + if DROPOUT > 0.0 or RETURN_LSE or RETURN_SOFTMAX: + pytest.skip( + "flash_attn_func (v3) doesn't support dropout, return_lse, or return_attn_probs" + ) + fp8_dtype = get_fp8_e4m3_dtype() + group_size = ( + NUM_Q_HEADS // NUM_K_HEADS if NUM_Q_HEADS % NUM_K_HEADS == 0 else None + ) + k_fp8, k_descale = _cast_to_fp8(k, fp8_dtype, "bshd") + v_fp8, v_descale = _cast_to_fp8(v, fp8_dtype, "bshd") + q_fp8, q_descale = _cast_to_fp8(q, fp8_dtype, "bshd", group_size=group_size) + triton_out = flash_attn_func_v3( + q_fp8, + k_fp8, + v_fp8, + softmax_scale=None, causal=CAUSAL, - return_lse=RETURN_LSE, - return_attn_probs=RETURN_SOFTMAX, + q_descale=q_descale, + k_descale=k_descale, + v_descale=v_descale, ) else: triton_out = flash_attn_func( @@ -369,18 +387,33 @@ def test_mha_varlen( print(f"cu_seqlens_q={cu_seqlens_q }") print(f"cu_seqlens_k={cu_seqlens_k }") if FP8: - triton_out = flash_attn_varlen_fp8_func( - q_unpad, - k_unpad, - v_unpad, + if DROPOUT > 0.0 or RETURN_LSE or RETURN_SOFTMAX: + pytest.skip( + "flash_attn_varlen_func_v3 doesn't support dropout, return_lse, or return_attn_probs" + ) + + fp8_dtype = get_fp8_e4m3_dtype() + group_size = ( + NUM_Q_HEADS // NUM_K_HEADS if NUM_Q_HEADS % NUM_K_HEADS == 0 else None + ) + k_fp8, k_descale = _cast_varlen_to_fp8(k_unpad, fp8_dtype, cu_seqlens_k) + v_fp8, v_descale = _cast_varlen_to_fp8(v_unpad, fp8_dtype, cu_seqlens_k) + q_fp8, q_descale = _cast_varlen_to_fp8( + q_unpad, fp8_dtype, cu_seqlens_q, group_size=group_size + ) + triton_out = flash_attn_varlen_func_v3( + q_fp8, + k_fp8, + v_fp8, cu_seqlens_q, cu_seqlens_k, max_seqlen_q, max_seqlen_k, - dropout_p=DROPOUT, + softmax_scale=None, causal=CAUSAL, - return_lse=RETURN_LSE, - return_attn_probs=RETURN_SOFTMAX, + q_descale=q_descale, + k_descale=k_descale, + v_descale=v_descale, ) else: triton_out = flash_attn_varlen_func( @@ -474,9 +507,8 @@ def test_mha_varlen( "NUM_Q_HEADS, NUM_K_HEADS", [(1, 1), (16, 16), (2, 1), (48, 8)] ) @pytest.mark.parametrize("HEAD_SZ", [8, 32, 128]) -@pytest.mark.parametrize("FP8", [False]) +@pytest.mark.parametrize("FP8", [False, True]) @pytest.mark.parametrize("FUSED", [False, True]) -# @pytest.mark.parametrize('FP8',[(False), (True)]) #TODO Debug FP8 def test_mha_backward( BATCH: int, SEQLEN_Q: int, @@ -492,10 +524,11 @@ def test_mha_backward( ): torch.cuda.empty_cache() torch.manual_seed(20) - pytest.skip("Backward accuracy issues due to Triton compiler") if FUSED and CAUSAL: pytest.skip("FUSED+CAUSAL results in NaNs") + if FP8 and DROPOUT > 0.0: + pytest.skip("Flash Attention v3 doesn't support dropout") mha_set_use_fused_bwd_kernel(FUSED) q = torch.randn((BATCH, SEQLEN_Q, NUM_Q_HEADS, HEAD_SZ), device="cuda", dtype=dtype) k = torch.randn((BATCH, SEQLEN_K, NUM_K_HEADS, HEAD_SZ), device="cuda", dtype=dtype) @@ -515,15 +548,28 @@ def test_mha_backward( with torch.enable_grad(): if FP8: - triton_out = flash_attn_fp8_func( - q, - k, - v, - dropout_p=DROPOUT, + # Use Flash Attention v3 for FP8 + fp8_dtype = get_fp8_e4m3_dtype() + group_size = ( + NUM_Q_HEADS // NUM_K_HEADS if NUM_Q_HEADS % NUM_K_HEADS == 0 else None + ) + k_fp8, k_descale = _cast_to_fp8(k, fp8_dtype, "bshd") + v_fp8, v_descale = _cast_to_fp8(v, fp8_dtype, "bshd") + q_fp8, q_descale = _cast_to_fp8(q, fp8_dtype, "bshd", group_size=group_size) + triton_out = flash_attn_func_v3( + q_fp8, + k_fp8, + v_fp8, + softmax_scale=None, causal=CAUSAL, - return_lse=True, - return_attn_probs=True, + q_descale=q_descale, + k_descale=k_descale, + v_descale=v_descale, ) + # For v3, we don't have LSE and attention probs + lse = None + sd_mask = None + dropout_mask = None else: triton_out = flash_attn_func( q, @@ -534,14 +580,13 @@ def test_mha_backward( return_lse=True, return_attn_probs=True, ) + assert len(triton_out) == 3 + triton_out, lse, sd_mask = triton_out[0], triton_out[1], triton_out[2] - assert len(triton_out) == 3 - triton_out, lse, sd_mask = triton_out[0], triton_out[1], triton_out[2] - - if DROPOUT > 0.0: - dropout_mask = sd_mask >= 0 - else: - dropout_mask = None + if DROPOUT > 0.0: + dropout_mask = sd_mask >= 0 + else: + dropout_mask = None triton_dq, triton_dk, triton_dv = torch.autograd.grad( triton_out, (q, k, v), do.clone() @@ -549,12 +594,15 @@ def test_mha_backward( if DEBUG_MODE: print(f"triton_out={triton_out}") - print(f"triton_lse={lse}") - print(f"sd_mask={sd_mask}") + if lse is not None: + print(f"triton_lse={lse}") + if sd_mask is not None: + print(f"sd_mask={sd_mask}") print(f"triton_dq.shape={triton_dq.shape} triton_dq={triton_dq}") print(f"triton_dk.shape={triton_dk.shape} triton_dk={triton_dk}") print(f"triton_dv.shape={triton_dv.shape} triton_dv={triton_dv}") - print(f"dropout_mask={dropout_mask}") + if dropout_mask is not None: + print(f"dropout_mask={dropout_mask}") if DEBUG_MODE: print("--------------Torch----------------") @@ -614,9 +662,8 @@ def test_mha_backward( "NUM_Q_HEADS, NUM_K_HEADS", [(1, 1), (16, 16), (2, 1), (48, 8)] ) @pytest.mark.parametrize("HEAD_SZ", [8, 32, 128]) -@pytest.mark.parametrize("FP8", [False]) +@pytest.mark.parametrize("FP8", [False, True]) @pytest.mark.parametrize("FUSED", [False, True]) -# @pytest.mark.parametrize('FP8',[(False), (True)]) #TODO Debug FP8 def test_mha_backward_varlen( BATCH: int, SEQLEN_Q: int, @@ -635,6 +682,8 @@ def test_mha_backward_varlen( pytest.skip("Backward accuracy issues due to Triton compiler") if FUSED and CAUSAL: pytest.skip("FUSED+CAUSAL results in NaNs") + if FP8 and DROPOUT > 0.0: + pytest.skip("Flash Attention v3 doesn't support dropout") mha_set_use_fused_bwd_kernel(FUSED) q = torch.randn((BATCH, SEQLEN_Q, NUM_Q_HEADS, HEAD_SZ), device="cuda", dtype=dtype) @@ -694,38 +743,67 @@ def test_mha_backward_varlen( print(f"do.shape={do.shape} do={do}") with torch.enable_grad(): - triton_out = flash_attn_varlen_func( - q_unpad, - k_unpad, - v_unpad, - cu_seqlens_q, - cu_seqlens_k, - max_seqlen_q, - max_seqlen_k, - dropout_p=DROPOUT, - causal=CAUSAL, - return_lse=True, - return_attn_probs=True, - ) - - assert len(triton_out) == 3 - triton_out, lse, sd_mask = triton_out[0], triton_out[1], triton_out[2] - - if DROPOUT > 0.0: - dropout_mask = sd_mask >= 0 - dropout_mask = pad_rearrange_dropout_mask( - dropout_mask, - cu_seqlens_q, - cu_seqlens_k, - max_seqlen_q, - max_seqlen_k, - SEQLEN_Q, - SEQLEN_K, - NUM_Q_HEADS, - ) - dropout_mask = dropout_mask > 0 - else: - dropout_mask = None + if FP8: + # Use Flash Attention v3 for FP8 + fp8_dtype = get_fp8_e4m3_dtype() + group_size = ( + NUM_Q_HEADS // NUM_K_HEADS if NUM_Q_HEADS % NUM_K_HEADS == 0 else None + ) + k_fp8, k_descale = _cast_varlen_to_fp8(k_unpad, fp8_dtype, cu_seqlens_k) + v_fp8, v_descale = _cast_varlen_to_fp8(v_unpad, fp8_dtype, cu_seqlens_k) + q_fp8, q_descale = _cast_varlen_to_fp8( + q_unpad, fp8_dtype, cu_seqlens_q, group_size=group_size + ) + triton_out = flash_attn_varlen_func_v3( + q_fp8, + k_fp8, + v_fp8, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + softmax_scale=None, + causal=CAUSAL, + q_descale=q_descale, + k_descale=k_descale, + v_descale=v_descale, + ) + # For v3, we don't have LSE and attention probs + lse = None + sd_mask = None + dropout_mask = None + else: + triton_out = flash_attn_varlen_func( + q_unpad, + k_unpad, + v_unpad, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + dropout_p=DROPOUT, + causal=CAUSAL, + return_lse=True, + return_attn_probs=True, + ) + assert len(triton_out) == 3 + triton_out, lse, sd_mask = triton_out[0], triton_out[1], triton_out[2] + + if DROPOUT > 0.0: + dropout_mask = sd_mask >= 0 + dropout_mask = pad_rearrange_dropout_mask( + dropout_mask, + cu_seqlens_q, + cu_seqlens_k, + max_seqlen_q, + max_seqlen_k, + SEQLEN_Q, + SEQLEN_K, + NUM_Q_HEADS, + ) + dropout_mask = dropout_mask > 0 + else: + dropout_mask = None triton_out = output_pad_fn(triton_out) triton_dq, triton_dk, triton_dv = torch.autograd.grad( @@ -737,11 +815,13 @@ def test_mha_backward_varlen( triton_dv = dk_pad_fn(triton_dv) if DEBUG_MODE: print(f"triton_out={triton_out}") - print(f"triton_lse.shape={lse.shape} triton_lse={lse}") + if lse is not None: + print(f"triton_lse.shape={lse.shape} triton_lse={lse}") print(f"triton_dq.shape={triton_dq.shape} triton_dq={triton_dq}") print(f"triton_dk.shape={triton_dk.shape} triton_dk={triton_dk}") print(f"triton_dv.shape={triton_dv.shape} triton_dv={triton_dv}") - print(f"dropout_mask={dropout_mask}") + if dropout_mask is not None: + print(f"dropout_mask={dropout_mask}") if DEBUG_MODE: print("--------------Torch----------------") @@ -772,12 +852,174 @@ def test_mha_backward_varlen( print(f"torch_dk.shape={torch_dk.shape} torch_dk={torch_dk}") print(f"torch_dv.shape={torch_dv.shape} torch_dv={torch_dv}") - torch.testing.assert_close( - triton_dq, torch_dq.to(triton_out.dtype), atol=1e-2, rtol=1e-2 + if FP8: + fp8_assert_close( + triton_dq, torch_dq.to(triton_dq.dtype), atol=ATOL_fp8, rtol=RTOL_fp8 + ) + fp8_assert_close( + triton_dk, torch_dk.to(triton_dk.dtype), atol=ATOL_fp8, rtol=RTOL_fp8 + ) + fp8_assert_close( + triton_dv, torch_dv.to(triton_dv.dtype), atol=ATOL_fp8, rtol=RTOL_fp8 + ) + else: + torch.testing.assert_close( + triton_dq, torch_dq.to(triton_out.dtype), atol=1e-2, rtol=1e-2 + ) + torch.testing.assert_close( + triton_dk, torch_dk.to(triton_out.dtype), atol=1e-2, rtol=1e-2 + ) + torch.testing.assert_close( + triton_dv, torch_dv.to(triton_out.dtype), atol=1e-2, rtol=1e-2 + ) + + +@pytest.mark.parametrize("FP8", [False, True]) +@pytest.mark.parametrize("causal", [False, True]) +@pytest.mark.parametrize("new_kv", [False]) +@pytest.mark.parametrize( + "seqlen_q, seqlen_k, d", + [ + (1, 64, 64), + (8, 128, 64), + (16, 256, 64), + (32, 256, 128), + ], +) +def test_mha_kvcache( + seqlen_q: int, + seqlen_k: int, + d: int, + causal: bool, + new_kv: bool, + FP8: bool, +): + device = "cuda" + torch.manual_seed(0) + + batch_size = 3 + nheads = 4 + nheads_k = nheads # Only MHA path exercised + + base_dtype = torch.float16 # allocate in fp16 first then cast if FP8 + k_cache = torch.randn( + batch_size, seqlen_k, nheads_k, d, device=device, dtype=base_dtype ) - torch.testing.assert_close( - triton_dk, torch_dk.to(triton_out.dtype), atol=1e-2, rtol=1e-2 + v_cache = torch.randn( + batch_size, seqlen_k, nheads_k, d, device=device, dtype=base_dtype ) - torch.testing.assert_close( - triton_dv, torch_dv.to(triton_out.dtype), atol=1e-2, rtol=1e-2 + + if new_kv: + seqlen_new = seqlen_q # Append a block same length as query + max_prev = seqlen_k - seqlen_new + assert max_prev >= 0 + cache_seqlens = torch.randint( + low=0, + high=max_prev + 1, + size=(batch_size,), + device=device, + dtype=torch.int32, + ) + k_new = torch.randn( + batch_size, seqlen_new, nheads_k, d, device=device, dtype=base_dtype + ) + v_new = torch.randn( + batch_size, seqlen_new, nheads_k, d, device=device, dtype=base_dtype + ) + else: + cache_seqlens = torch.randint( + low=1, + high=seqlen_k + 1, + size=(batch_size,), + device=device, + dtype=torch.int32, + ) + k_new = None + v_new = None + + q = torch.randn(batch_size, seqlen_q, nheads, d, device=device, dtype=base_dtype) + + # Cast to FP8 using helper (keep original fp16 copies for reference math) + if FP8: + fp8_dtype = get_fp8_e4m3_dtype() + group_size = None + k_cache_fp8, k_descale = _cast_to_fp8(k_cache, fp8_dtype, "bshd", group_size=group_size) + v_cache_fp8, v_descale = _cast_to_fp8(v_cache, fp8_dtype, "bshd", group_size=group_size) + q_fp8, q_descale = _cast_to_fp8(q, fp8_dtype, "bshd", group_size=group_size) + if new_kv: + k_new_fp8, _ = _cast_to_fp8(k_new, fp8_dtype, "bshd", group_size=group_size) + v_new_fp8, _ = _cast_to_fp8(v_new, fp8_dtype, "bshd", group_size=group_size) + else: + k_new_fp8 = v_new_fp8 = None + else: + k_cache_fp8 = k_cache + v_cache_fp8 = v_cache + q_fp8 = q + k_new_fp8 = k_new + v_new_fp8 = v_new + q_descale = k_descale = v_descale = None + + # Reference: clone cache & apply append if needed + k_cache_ref = k_cache.clone() + v_cache_ref = v_cache.clone() + if new_kv: + for b in range(batch_size): + start = cache_seqlens[b].item() + end = start + seqlen_q + k_cache_ref[b, start:end] = k_new[b] + v_cache_ref[b, start:end] = v_new[b] + final_used = cache_seqlens + seqlen_q + else: + final_used = cache_seqlens + + arange = torch.arange(seqlen_k, device=device).unsqueeze(0) + key_padding_mask = arange < final_used.unsqueeze(1) + + max_used = final_used.max().item() + k_eff = k_cache_ref[:, :max_used] + v_eff = v_cache_ref[:, :max_used] + key_mask_eff = key_padding_mask[:, :max_used] + + # Reference runs entirely on original fp16 tensors + out_ref, _ = attention_ref( + q, + k_eff, + v_eff, + query_padding_mask=None, + key_padding_mask=key_mask_eff, + causal=causal, + window_size=(-1, -1), + ) + + out_kernel = flash_attn_with_kvcache_v3( + q_fp8, + k_cache_fp8, + v_cache_fp8, + k=k_new_fp8, + v=v_new_fp8, + cache_seqlens=cache_seqlens, + causal=causal, + window_size=(-1, -1), + return_softmax_lse=False, + q_descale=q_descale, + k_descale=k_descale, + v_descale=v_descale, ) + + assert out_kernel.shape == out_ref.shape + if FP8: + # Reuse fp8 tolerance constants from earlier (slightly relaxed) + torch.testing.assert_close( + out_kernel, out_ref.to(out_kernel.dtype), atol=2.5e-1, rtol=2.5e-1 + ) + else: + torch.testing.assert_close( + out_kernel, out_ref.to(out_kernel.dtype), atol=1e-2, rtol=1e-2 + ) + + if new_kv and not FP8: # Only verify in-place cache contents directly for non-FP8 for now + for b in range(batch_size): + start = cache_seqlens[b].item() + end = start + seqlen_q + torch.testing.assert_close(k_cache[b, start:end], k_new[b], atol=1e-5, rtol=1e-5) + torch.testing.assert_close(v_cache[b, start:end], v_new[b], atol=1e-5, rtol=1e-5)