Skip to content

[ROCM][KERNEL] Paged attention for V1 #15720

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 10 commits into from
Apr 3, 2025
Merged

[ROCM][KERNEL] Paged attention for V1 #15720

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
d2fd7c1
initial working commit with C++ kernel
Mar 28, 2025
d8f1598
crash fix for absent element in query
Mar 31, 2025
a053de0
Merge branch 'upstream/main' into v1_rocm_paged_attention_integration
Mar 31, 2025
e7a77aa
removing duplicated code
Mar 31, 2025
5e102cb
removing duplicated code
Mar 31, 2025
2f1ad6c
acting on comment about optional tensor
Apr 1, 2025
f7e8e7d
moved get_device_properties into function to let GPUs initialized
Apr 2, 2025
ee68e0e
borken kernel test fix plus comments application
Apr 2, 2025
233454b
Merge branch 'upstream/main' into v1_rocm_paged_attention_integration
Apr 2, 2025
67d07de
borken kernel test fix plus build fix
Apr 2, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
107 changes: 75 additions & 32 deletions csrc/rocm/attention.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,8 @@
#include <hip/hip_bf16.h>
#include "cuda_compat.h"

#include <inttypes.h>

#include <algorithm>
#include "../attention/dtype_fp8.cuh"
#include "../quantization/fp8/amd/quant_utils.cuh"
Expand Down Expand Up @@ -272,6 +274,7 @@ __launch_bounds__(NUM_THREADS, 5) void paged_attention_ll4mi_QKV_mfma16_kernel(
const float scale,
const int* __restrict__ block_tables, // [num_seqs, max_num_blocks_per_seq]
const int* __restrict__ context_lens, // [num_seqs]
const int* __restrict__ query_start_loc_ptr, // [num_seqs]
const int max_num_blocks_per_seq,
const float* __restrict__ alibi_slopes, // [num_heads]
const int q_stride,
Expand All @@ -291,6 +294,13 @@ __launch_bounds__(NUM_THREADS, 5) void paged_attention_ll4mi_QKV_mfma16_kernel(
const int rowid = laneid / 16;

const auto seq_idx = blockIdx.x;
// NOTE queries with sequence len > 1 are prefills and taken care by another
// kernel.
if (query_start_loc_ptr != nullptr &&
(query_start_loc_ptr[seq_idx + 1] - query_start_loc_ptr[seq_idx]) != 1) {
return;
}

const auto partition_idx = blockIdx.y;

constexpr int T_PAR_SIZE = 256; // token partition size set to 256
Expand Down Expand Up @@ -377,9 +387,10 @@ __launch_bounds__(NUM_THREADS, 5) void paged_attention_ll4mi_QKV_mfma16_kernel(
// fetch Q in shared across warps and then write to registers
const int local_qhead_idx = 4 * warpid + rowid;
const int global_qhead_idx = wg_start_head_idx + local_qhead_idx;
const int64_t seq_idx64 = static_cast<int64_t>(seq_idx);
const int64_t query_start_off =
query_start_loc_ptr ? query_start_loc_ptr[seq_idx] : seq_idx;
Copy link
Member

@mgoin mgoin Apr 2, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

keep the static cast?

Copy link
Contributor Author

@maleksan85 maleksan85 Apr 2, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

should this be conversion similar to what static cast provides? I mean when assignment happens to int64_t from int?

const scalar_t* q_ptr =
q + seq_idx64 * q_stride + global_qhead_idx * HEAD_SIZE;
q + query_start_off * q_stride + global_qhead_idx * HEAD_SIZE;

const int qhead_element = lane16id * CONTIGUOUS_SCALAR_ELEMS_16B;
if ((local_qhead_idx < GQA_RATIO) && (qhead_element < HEAD_SIZE)) {
Expand Down Expand Up @@ -777,6 +788,7 @@ __launch_bounds__(NUM_THREADS) void paged_attention_ll4mi_QKV_mfma4_kernel(
const float scale,
const int* __restrict__ block_tables, // [num_seqs, max_num_blocks_per_seq]
const int* __restrict__ context_lens, // [num_seqs]
const int* __restrict__ query_start_loc_ptr, // [num_seqs]
const int max_num_blocks_per_seq,
const float* __restrict__ alibi_slopes, // [num_heads]
const int q_stride,
Expand All @@ -794,6 +806,12 @@ __launch_bounds__(NUM_THREADS) void paged_attention_ll4mi_QKV_mfma4_kernel(
const int lane4id = laneid % 4;

const auto seq_idx = blockIdx.x;
// NOTE queries with sequence len > 1 are prefills and taken care by another
// kernel.
if (query_start_loc_ptr != nullptr &&
(query_start_loc_ptr[seq_idx + 1] - query_start_loc_ptr[seq_idx] != 1)) {
return;
}
const auto partition_idx = blockIdx.y;
const auto partition_size = blockDim.x;
const auto max_num_partitions = gridDim.y;
Expand Down Expand Up @@ -882,9 +900,11 @@ __launch_bounds__(NUM_THREADS) void paged_attention_ll4mi_QKV_mfma4_kernel(
}

// fetch q elements
// every 4 lanes fetch 8 elems, so warp fetches 8*16 = 128 elems
// every 4 lanes fetch 8 elems, so warp fetches 8*16 = 128 elemsc
const int64_t query_start_off =
query_start_loc_ptr ? query_start_loc_ptr[seq_idx] : seq_idx;
const scalar_t* q_ptr =
q + seq_idx * q_stride + wg_start_head_idx * HEAD_SIZE;
q + query_start_off * q_stride + wg_start_head_idx * HEAD_SIZE;
const _B16x8* q_ptrh8 = reinterpret_cast<const _B16x8*>(q_ptr);
const int qhead_elemh8 = laneid / 4;

Expand Down Expand Up @@ -1267,10 +1287,19 @@ __launch_bounds__(NUM_THREADS) void paged_attention_ll4mi_reduce_kernel(
const scalar_t* __restrict__ tmp_out, // [num_seqs, num_heads,
// max_num_partitions, head_size]
const int* __restrict__ context_lens, // [num_seqs]
const int* __restrict__ query_start_loc_ptr, // [num_seqs]
const int max_num_partitions) {
const auto num_heads = gridDim.x;
const auto head_idx = blockIdx.x;
const auto seq_idx = blockIdx.y;

// NOTE queries with sequence len > 1 are prefills and taken care by another
// kernel.
if (query_start_loc_ptr != nullptr &&
(query_start_loc_ptr[seq_idx + 1] - query_start_loc_ptr[seq_idx] != 1)) {
return;
}

const int context_len = context_lens[seq_idx];
const int num_partitions = DIVIDE_ROUND_UP(context_len, PARTITION_SIZE);
[[maybe_unused]] constexpr int NUM_WARPS = NUM_THREADS / WARP_SIZE;
Expand Down Expand Up @@ -1439,7 +1468,9 @@ __launch_bounds__(NUM_THREADS) void paged_attention_ll4mi_reduce_kernel(
__fdividef(1.0f, shared_global_exp_sum + 1e-6f);
acc *= inv_global_exp_sum;

OUTT* out_ptr = out + static_cast<int64_t>(seq_idx) * num_heads * HEAD_SIZE +
const int64_t query_start_off =
query_start_loc_ptr ? query_start_loc_ptr[seq_idx] : seq_idx;
OUTT* out_ptr = out + query_start_off * num_heads * HEAD_SIZE +
static_cast<int64_t>(head_idx) * HEAD_SIZE;
if constexpr (std::is_same<OUTT, bit8_t>::value) {
out_ptr[threadIdx.x] =
Expand All @@ -1466,6 +1497,7 @@ __launch_bounds__(NUM_THREADS) void paged_attention_ll4mi_QKV_mfma16_kernel(
const float scale,
const int* __restrict__ block_tables, // [num_seqs, max_num_blocks_per_seq]
const int* __restrict__ context_lens, // [num_seqs]
const int* __restrict__ query_start_loc_ptr, // [num_seqs]
const int max_num_blocks_per_seq,
const float* __restrict__ alibi_slopes, // [num_heads]
const int q_stride,
Expand All @@ -1492,6 +1524,7 @@ __launch_bounds__(NUM_THREADS) void paged_attention_ll4mi_QKV_mfma4_kernel(
const float scale,
const int* __restrict__ block_tables, // [num_seqs, max_num_blocks_per_seq]
const int* __restrict__ context_lens, // [num_seqs]
const int* __restrict__ query_start_loc_ptr, // [num_seqs]
const int max_num_blocks_per_seq,
const float* __restrict__ alibi_slopes, // [num_heads]
const int q_stride,
Expand All @@ -1515,41 +1548,42 @@ __launch_bounds__(NUM_THREADS) void paged_attention_ll4mi_reduce_kernel(
const float* __restrict__ max_logits, // [num_seqs, num_heads, max_num_partitions]
const scalar_t* __restrict__ tmp_out, // [num_seqs, num_heads, max_num_partitions, head_size]
const int* __restrict__ context_lens, // [num_seqs]
const int* __restrict__ query_start_loc_ptr, // [num_seqs]
const int max_num_partitions) {
UNREACHABLE_CODE
}
// clang-format on

#endif // defined(__HIP__MI300_MI250__) TODO: Add NAVI support

#define LAUNCH_CUSTOM_ATTENTION_MFMA16(GQA_RATIO) \
paged_attention_ll4mi_QKV_mfma16_kernel<T, KVT, KV_DTYPE, OUTT, BLOCK_SIZE, \
HEAD_SIZE, NTHR, ALIBI_ENABLED, \
GQA_RATIO> \
<<<grid, block, 0, stream>>>( \
query_ptr, key_cache_ptr, value_cache_ptr, num_kv_heads, scale, \
block_tables_ptr, context_lens_ptr, max_num_blocks_per_seq, \
alibi_slopes_ptr, q_stride, kv_block_stride, kv_head_stride, \
exp_sums_ptr, max_logits_ptr, tmp_out_ptr, out_ptr, max_ctx_blocks, \
k_scale_ptr, v_scale_ptr);

#define LAUNCH_CUSTOM_ATTENTION_MFMA4(GQA_RATIO) \
paged_attention_ll4mi_QKV_mfma4_kernel<T, KVT, KV_DTYPE, OUTT, BLOCK_SIZE, \
HEAD_SIZE, NTHR, ALIBI_ENABLED, \
GQA_RATIO> \
<<<grid, block, 0, stream>>>( \
query_ptr, key_cache_ptr, value_cache_ptr, num_kv_heads, scale, \
block_tables_ptr, context_lens_ptr, max_num_blocks_per_seq, \
alibi_slopes_ptr, q_stride, kv_block_stride, kv_head_stride, \
exp_sums_ptr, max_logits_ptr, tmp_out_ptr, out_ptr, max_ctx_blocks, \
k_scale_ptr, v_scale_ptr);
#define LAUNCH_CUSTOM_ATTENTION_MFMA16(GQA_RATIO) \
paged_attention_ll4mi_QKV_mfma16_kernel<T, KVT, KV_DTYPE, OUTT, BLOCK_SIZE, \
HEAD_SIZE, NTHR, ALIBI_ENABLED, \
GQA_RATIO> \
<<<grid, block, 0, stream>>>( \
query_ptr, key_cache_ptr, value_cache_ptr, num_kv_heads, scale, \
block_tables_ptr, context_lens_ptr, query_start_loc_ptr, \
max_num_blocks_per_seq, alibi_slopes_ptr, q_stride, kv_block_stride, \
kv_head_stride, exp_sums_ptr, max_logits_ptr, tmp_out_ptr, out_ptr, \
max_ctx_blocks, k_scale_ptr, v_scale_ptr);

#define LAUNCH_CUSTOM_ATTENTION_MFMA4(GQA_RATIO) \
paged_attention_ll4mi_QKV_mfma4_kernel<T, KVT, KV_DTYPE, OUTT, BLOCK_SIZE, \
HEAD_SIZE, NTHR, ALIBI_ENABLED, \
GQA_RATIO> \
<<<grid, block, 0, stream>>>( \
query_ptr, key_cache_ptr, value_cache_ptr, num_kv_heads, scale, \
block_tables_ptr, context_lens_ptr, query_start_loc_ptr, \
max_num_blocks_per_seq, alibi_slopes_ptr, q_stride, kv_block_stride, \
kv_head_stride, exp_sums_ptr, max_logits_ptr, tmp_out_ptr, out_ptr, \
max_ctx_blocks, k_scale_ptr, v_scale_ptr);

#define LAUNCH_CUSTOM_REDUCTION(NPAR_LOOPS) \
paged_attention_ll4mi_reduce_kernel<T, OUTT, HEAD_SIZE, HEAD_SIZE, \
PARTITION_SIZE, NPAR_LOOPS> \
<<<reduce_grid, reduce_block, 0, stream>>>( \
out_ptr, exp_sums_ptr, max_logits_ptr, tmp_out_ptr, \
context_lens_ptr, max_num_partitions);
context_lens_ptr, query_start_loc_ptr, max_num_partitions);

template <typename T, typename KVT, vllm::Fp8KVCacheDataType KV_DTYPE,
int BLOCK_SIZE, int HEAD_SIZE, typename OUTT, int PARTITION_SIZE_OLD,
Expand All @@ -1559,16 +1593,24 @@ void paged_attention_custom_launcher(
torch::Tensor& tmp_out, torch::Tensor& query, torch::Tensor& key_cache,
torch::Tensor& value_cache, const int num_kv_heads, float scale,
torch::Tensor& block_tables, torch::Tensor& context_lens,
int max_context_len, const std::optional<torch::Tensor>& alibi_slopes,
torch::Tensor& k_scale, torch::Tensor& v_scale) {
int num_seqs = query.size(0);
const std::optional<torch::Tensor>& query_start_loc, int max_context_len,
const std::optional<torch::Tensor>& alibi_slopes, torch::Tensor& k_scale,
torch::Tensor& v_scale) {
int num_seqs = block_tables.size(0);
int num_heads = query.size(1);
int head_size = query.size(2);
int max_num_blocks_per_seq = block_tables.size(1);
int q_stride = query.stride(0);
int kv_block_stride = key_cache.stride(0);
int kv_head_stride = key_cache.stride(1);

// NOTE: query start location is optional for V0 decode should not be used.
// If batch contains mix of prefills and decode, prefills should be skipped.
const int* query_start_loc_ptr =
query_start_loc
? reinterpret_cast<const int*>(query_start_loc.value().data_ptr())
: nullptr;

// NOTE: alibi_slopes is optional.
const float* alibi_slopes_ptr =
alibi_slopes
Expand Down Expand Up @@ -1700,8 +1742,8 @@ void paged_attention_custom_launcher(
paged_attention_custom_launcher<T, KVT, KV_DTYPE, BLK_SIZE, HEAD_SIZE, T, \
PSIZE, ALIBI_ENABLED>( \
out, exp_sums, max_logits, tmp_out, query, key_cache, value_cache, \
num_kv_heads, scale, block_tables, context_lens, max_context_len, \
alibi_slopes, k_scale, v_scale);
num_kv_heads, scale, block_tables, context_lens, query_start_loc, \
max_context_len, alibi_slopes, k_scale, v_scale);

#define CALL_CUSTOM_LAUNCHER_ALIBI(T, KVT, KV_DTYPE, BLK_SIZE, HEAD_SIZE, \
PSIZE) \
Expand Down Expand Up @@ -1750,6 +1792,7 @@ void paged_attention(
double scale,
torch::Tensor& block_tables, // [num_seqs, max_num_blocks_per_seq]
torch::Tensor& context_lens, // [num_seqs]
const std::optional<torch::Tensor>& query_start_loc, // [num_seqs]
int64_t block_size, int64_t max_context_len,
const std::optional<torch::Tensor>& alibi_slopes,
const std::string& kv_cache_dtype, torch::Tensor& k_scale,
Expand Down
5 changes: 3 additions & 2 deletions csrc/rocm/ops.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,8 +7,9 @@ void paged_attention(torch::Tensor& out, torch::Tensor& exp_sums,
torch::Tensor& query, torch::Tensor& key_cache,
torch::Tensor& value_cache, int64_t num_kv_heads,
double scale, torch::Tensor& block_tables,
torch::Tensor& context_lens, int64_t block_size,
int64_t max_context_len,
torch::Tensor& context_lens,
const std::optional<torch::Tensor>& query_start_loc,
int64_t block_size, int64_t max_context_len,
const std::optional<torch::Tensor>& alibi_slopes,
const std::string& kv_cache_dtype, torch::Tensor& k_scale,
torch::Tensor& v_scale);
4 changes: 3 additions & 1 deletion csrc/rocm/torch_bindings.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,9 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, rocm_ops) {
" Tensor query, Tensor key_cache,"
" Tensor value_cache, int num_kv_heads,"
" float scale, Tensor block_tables,"
" Tensor context_lens, int block_size,"
" Tensor context_lens,"
" Tensor? query_start_loc,"
" int block_size,"
" int max_context_len,"
" Tensor? alibi_slopes,"
" str kv_cache_dtype,"
Expand Down
6 changes: 4 additions & 2 deletions vllm/_custom_ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -110,6 +110,7 @@ def paged_attention_rocm(
scale: float,
block_tables: torch.Tensor,
seq_lens: torch.Tensor,
query_start_loc: torch.Tensor,
block_size: int,
max_seq_len: int,
alibi_slopes: Optional[torch.Tensor],
Expand All @@ -120,8 +121,9 @@ def paged_attention_rocm(
torch.ops._rocm_C.paged_attention(out, exp_sum, max_logits, tmp_out, query,
key_cache, value_cache, num_kv_heads,
scale, block_tables, seq_lens,
block_size, max_seq_len, alibi_slopes,
kv_cache_dtype, k_scale, v_scale)
query_start_loc, block_size, max_seq_len,
alibi_slopes, kv_cache_dtype, k_scale,
v_scale)


def mla_decode_kvcache_cpu(
Expand Down
1 change: 1 addition & 0 deletions vllm/attention/backends/rocm_flash_attn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -833,6 +833,7 @@ def forward(
decode_meta.seq_lens_tensor
if self.attn_type != AttentionType.ENCODER_DECODER else
decode_meta.encoder_seq_lens_tensor,
None,
block_size,
max_seq_len,
self.alibi_slopes,
Expand Down
Loading