[Doc] Optimize the quickstart guide for clarity and not just for CUDA

README.md

@@ -123,35 +123,24 @@ Below is an example that demonstrates more advanced features: layout annotation,
 ```python
 import tilelang
 import tilelang.language as T
-# `make_mma_swizzle_layout` is a python defined layout function
-# specifically designed for for MMA operations
-# which ensures the consistency with the nvidia CUTLASS Library.
-# to avoid bank conflicts and maximize the performance.
-from tilelang.intrinsics import (
-    make_mma_swizzle_layout as make_swizzle_layout,)
-
-# add decorator @tilelang.jit if you want to return a torch function
-# @tilelang.jit
+
+# @tilelang.jit(target="cuda")
+# target currently can be "cuda" or "hip" or "cpu".
+# if not specified, it will be inferred from the input tensors during compile time
+@tilelang.jit
 def matmul(M, N, K, block_M, block_N, block_K, dtype="float16", accum_dtype="float"):
 
     @T.prim_func
-    def main(
-        A: T.Tensor((M, K), dtype),
-        B: T.Tensor((K, N), dtype),
-        C: T.Tensor((M, N), dtype),
+    def matmul_relu_kernel(
+            A: T.Tensor((M, K), dtype),
+            B: T.Tensor((K, N), dtype),
+            C: T.Tensor((M, N), dtype),
     ):
         # Initialize Kernel Context
         with T.Kernel(T.ceildiv(N, block_N), T.ceildiv(M, block_M), threads=128) as (bx, by):
             A_shared = T.alloc_shared((block_M, block_K), dtype)
             B_shared = T.alloc_shared((block_K, block_N), dtype)
-            C_local  = T.alloc_fragment((block_M, block_N), accum_dtype)
-
-            # Apply layout optimizations or define your own layout (Optional)
-            # If not specified, we will deduce the layout automatically
-            # T.annotate_layout({
-            #     A_shared: make_swizzle_layout(A_shared),
-            #     B_shared: make_swizzle_layout(B_shared),
-            # })
+            C_local = T.alloc_fragment((block_M, block_N), accum_dtype)
 
             # Enable rasterization for better L2 cache locality (Optional)
             # T.use_swizzle(panel_size=10, enable=True)
@@ -164,53 +153,58 @@ def matmul(M, N, K, block_M, block_N, block_K, dtype="float16", accum_dtype="flo
                 # This is a sugar syntax for parallelized copy
                 T.copy(A[by * block_M, ko * block_K], A_shared)
 
-                # Demonstrate parallelized copy from global to shared for B
-                for k, j in T.Parallel(block_K, block_N):
-                    B_shared[k, j] = B[ko * block_K + k, bx * block_N + j]
+                # Copy tile of B
+                T.copy(B[ko * block_K, bx * block_N], B_shared)
 
                 # Perform a tile-level GEMM on the shared buffers
                 # Currently we dispatch to the cute/hip on Nvidia/AMD GPUs
                 T.gemm(A_shared, B_shared, C_local)
+
+            # relu
+            for i, j in T.Parallel(block_M, block_N):
+                C_local[i, j] = T.max(C_local[i, j], 0)
 
             # Copy result back to global memory
             T.copy(C_local, C[by * block_M, bx * block_N])
 
-    return main
+    return matmul_relu_kernel
 
 
-# 1. Define the kernel (matmul) with the desired dimensions
-func = matmul(1024, 1024, 1024, 128, 128, 32)
+M = 1024  # M = T.symbolic("m") if you want to use dynamic shape
+N = 1024
+K = 1024
+block_M = 128
+block_N = 128
+block_K = 32
 
-# 2. Compile the kernel into a torch function
-# out_idx specifies the index of the output buffer in the argument list
-# if out_idx is specified, the tensor will be created during runtime
-# target currently can be "cuda" or "hip" or "cpu".
-jit_kernel = tilelang.compile(func, out_idx=[2], target="cuda")
+# 1. Define the kernel (matmul) and compile/lower it into an executable module
+matmul_relu_kernel = matmul(M, N, K, block_M, block_N, block_K)
 
 # 3. Test the kernel in Python with PyTorch data
 import torch
 
 # Create random input tensors on the GPU
-a = torch.randn(1024, 1024, device="cuda", dtype=torch.float16)
-b = torch.randn(1024, 1024, device="cuda", dtype=torch.float16)
-
+a = torch.randn(M, K, device="cuda", dtype=torch.float16)
+b = torch.randn(K, N, device="cuda", dtype=torch.float16)
+c = torch.empty(M, N, device="cuda", dtype=torch.float16)
 
-# Run the kernel through the JIT-compiled function
-c = jit_kernel(a, b)
+# Run the kernel through the Profiler
+matmul_relu_kernel(a, b, c)
 
+print(c)
 # Reference multiplication using PyTorch
-ref_c = a @ b
+ref_c = torch.relu(a @ b)
 
 # Validate correctness
 torch.testing.assert_close(c, ref_c, rtol=1e-2, atol=1e-2)
 print("Kernel output matches PyTorch reference.")
 
 # 4. Retrieve and inspect the generated CUDA source (optional)
-cuda_source = jit_kernel.get_kernel_source()
-print("Generated CUDA kernel:\n", cuda_source)
+# cuda_source = jit_kernel.get_kernel_source()
+# print("Generated CUDA kernel:\n", cuda_source)
 
-# 5.Pofile latency with the profiler
-profiler = jit_kernel.get_profiler()
+# 5.Profile latency with kernel
+profiler = matmul_relu_kernel.get_profiler(tensor_supply_type=tilelang.TensorSupplyType.Normal)
 
 latency = profiler.do_bench()
 

examples/quickstart.py

@@ -1,20 +1,15 @@
 import tilelang
 import tilelang.language as T
 
-# `make_mma_swizzle_layout` is a python defined layout function
-# specifically designed for MMA operations
-# which ensures the consistency with the nvidia CUTLASS Library.
-# to avoid bank conflicts and maximize the performance.
-from tilelang.intrinsics import (
-    make_mma_swizzle_layout as make_swizzle_layout,)  # noqa: F401
 
-
-# add decorator @tilelang.jit if you want to return a torch function
-# @tilelang.jit
+# @tilelang.jit(target="cuda")
+# target currently can be "cuda" or "hip" or "cpu".
+# if not specified, it will be inferred from the input tensors during compile time
+@tilelang.jit
 def matmul(M, N, K, block_M, block_N, block_K, dtype="float16", accum_dtype="float"):
 
     @T.prim_func
-    def main(
+    def matmul_relu_kernel(
             A: T.Tensor((M, K), dtype),
             B: T.Tensor((K, N), dtype),
             C: T.Tensor((M, N), dtype),
@@ -25,13 +20,6 @@ def main(
             B_shared = T.alloc_shared((block_K, block_N), dtype)
             C_local = T.alloc_fragment((block_M, block_N), accum_dtype)
 
-            # Apply layout optimizations or define your own layout (Optional)
-            # If not specified, we will deduce the layout automatically
-            # T.annotate_layout({
-            #     A_shared: make_swizzle_layout(A_shared),
-            #     B_shared: make_swizzle_layout(B_shared),
-            # })
-
             # Enable rasterization for better L2 cache locality (Optional)
             # T.use_swizzle(panel_size=10, enable=True)
 
@@ -41,8 +29,6 @@ def main(
             for ko in T.Pipelined(T.ceildiv(K, block_K), num_stages=3):
                 # Copy tile of A
                 # This is a sugar syntax for parallelized copy
-                # for i, k in T.Parallel(M, block_K):
-                #     A_shared[i, k] = A[by * block_M + i, ko * block_K + k]
                 T.copy(A[by * block_M, ko * block_K], A_shared)
 
                 # Copy tile of B
@@ -52,10 +38,14 @@ def main(
                 # Currently we dispatch to the cute/hip on Nvidia/AMD GPUs
                 T.gemm(A_shared, B_shared, C_local)
 
+            # relu
+            for i, j in T.Parallel(block_M, block_N):
+                C_local[i, j] = T.max(C_local[i, j], 0)
+
             # Copy result back to global memory
             T.copy(C_local, C[by * block_M, bx * block_N])
 
-    return main
+    return matmul_relu_kernel
 
 
 M = 1024  # M = T.symbolic("m") if you want to use dynamic shape
@@ -66,27 +56,22 @@ def main(
 block_K = 32
 
 # 1. Define the kernel (matmul) and compile/lower it into an executable module
-func = matmul(M, N, K, block_M, block_N, block_K)
-
-# 2. Compile the kernel into a torch function
-# out_idx specifies the index of the output buffer in the argument list
-# if out_idx is specified, the tensor will be created during runtime
-# target currently can be "cuda" or "hip" or "cpu".
-jit_kernel = tilelang.compile(func, out_idx=[2], target="cuda")
+matmul_relu_kernel = matmul(M, N, K, block_M, block_N, block_K)
 
 # 3. Test the kernel in Python with PyTorch data
 import torch
 
 # Create random input tensors on the GPU
 a = torch.randn(M, K, device="cuda", dtype=torch.float16)
 b = torch.randn(K, N, device="cuda", dtype=torch.float16)
+c = torch.empty(M, N, device="cuda", dtype=torch.float16)
 
 # Run the kernel through the Profiler
-c = jit_kernel(a, b)
+matmul_relu_kernel(a, b, c)
 
 print(c)
 # Reference multiplication using PyTorch
-ref_c = a @ b
+ref_c = torch.relu(a @ b)
 
 # Validate correctness
 torch.testing.assert_close(c, ref_c, rtol=1e-2, atol=1e-2)
@@ -97,7 +82,7 @@ def main(
 # print("Generated CUDA kernel:\n", cuda_source)
 
 # 5.Profile latency with kernel
-profiler = jit_kernel.get_profiler(tensor_supply_type=tilelang.TensorSupplyType.Normal)
+profiler = matmul_relu_kernel.get_profiler(tensor_supply_type=tilelang.TensorSupplyType.Normal)
 
 latency = profiler.do_bench()