feat: support sm90 cutlass group gemm (#509)

Co-authored-by: Zihao Ye <[email protected]>

Author: xslingcn

flashinfer-aot/csrc_aot/flashinfer_ops.cu

@@ -13,7 +13,6 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-#pragma once
 #include <torch/extension.h>
 
 void append_paged_kv_cache(torch::Tensor append_key, torch::Tensor append_value,

flashinfer-aot/csrc_aot/flashinfer_sm90_ops.cu

@@ -0,0 +1,26 @@
+/*
+ * Copyright (c) 2023 by FlashInfer team.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include <torch/extension.h>
+
+
+torch::Tensor CutlassSegmentGEMMSM90(torch::Tensor float_workspace_buffer, torch::Tensor int_workspace_buffer, torch::Tensor seg_indptr,
+                                 torch::Tensor weight_indices, torch::Tensor x,
+                                 torch::Tensor weight, unsigned int batch_size,
+                                 bool weight_column_major);
+
+PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
+  m.def("cutlass_segment_gemm_sm90", &CutlassSegmentGEMMSM90, "Cutlass Segment GEMM operator for SM90");
+}

flashinfer-aot/setup.py

@@ -355,6 +355,7 @@ def __init__(self, *args, **kwargs) -> None:
     include_dirs = [
         str(root.resolve() / "include"),
         str(root.resolve() / "3rdparty" / "cutlass" / "include"),  # for group gemm
+        str(root.resolve() / "3rdparty" / "cutlass" / "tools" / "util" / "include"),
     ]
     extra_compile_args = {
         "cxx": [
@@ -371,6 +372,10 @@ def __init__(self, *args, **kwargs) -> None:
             "-use_fast_math",
         ],
     }
+    extra_compile_args_sm90 = extra_compile_args.copy()
+    extra_compile_args_sm90["nvcc"].extend(
+        "-gencode arch=compute_90a,code=sm_90a".split()
+    )
     ext_modules = []
     ext_modules.append(
         torch_cpp_ext.CUDAExtension(
@@ -385,12 +390,23 @@ def __init__(self, *args, **kwargs) -> None:
                 "csrc/quantization.cu",
                 "csrc/group_gemm.cu",
                 "csrc/bmm_fp8.cu",
-                "csrc_aot/flashinfer_ops.cu",
+                "csrc_aot/flashinfer_ops.cu"
             ],
             include_dirs=include_dirs,
             extra_compile_args=extra_compile_args,
         )
     )
+    ext_modules.append(
+        torch_cpp_ext.CUDAExtension(
+            name="flashinfer._kernels_sm90",
+            sources=[
+                "csrc/group_gemm_sm90.cu",
+                "csrc_aot/flashinfer_sm90_ops.cu",
+            ],
+            include_dirs=include_dirs,
+            extra_compile_args=extra_compile_args_sm90,
+        )
+    )
     ext_modules.append(
         torch_cpp_ext.CUDAExtension(
             name="flashinfer._decode_kernels",

include/flashinfer/gemm/group_gemm.cuh

@@ -53,7 +53,7 @@ cudaError_t CutlassSegmentGEMMRun(void* workspace_buffer, size_t workspace_buffe
 
   // NOTE(Zihao): I didn't successfully launch the kernel with cudaLaunchKernel API,
   // so I just use the kernel function directly, need to investigate more.
-  auto compute_args_kernel = compute_cutlass_group_gemm_args<DType>;
+  auto compute_args_kernel = compute_sm80_cutlass_group_gemm_args<DType, DType>;
   compute_args_kernel<<<batch_size, 1, 0, stream>>>(
       problem_sizes_device, x_data, w_data, y_data, ld_x, ld_w, ld_y, (DType*)x, (DType*)w,
       (DType*)y, xy_indptr_d, w_indices_d, d_in, d_out, weight_column_major);
@@ -116,4 +116,4 @@ cudaError_t CutlassSegmentGEMMRun(void* workspace_buffer, size_t workspace_buffe
 
 }  // namespace flashinfer
 
-#endif  // FLASHINFER_GEMM_GROUP_GEMM_CUH_
+#endif  // FLASHINFER_GEMM_GROUP_GEMM_CUH_

include/flashinfer/gemm/group_gemm_cutlass.cuh

@@ -16,11 +16,16 @@
 #ifndef FLASHINFER_GROUP_GEMM_CUTLASS_CUH_
 #define FLASHINFER_GROUP_GEMM_CUTLASS_CUH_
 
+#include <cuda_bf16.h>
+#include <cuda_fp16.h>
+#include <cuda_fp8.h>
+
 #include "cutlass/cutlass.h"
 #include "cutlass/gemm/device/gemm_grouped.h"
 #include "cutlass/gemm/kernel/default_gemm_grouped.h"
 #include "cutlass/layout/matrix.h"
 #include "cutlass/numeric_types.h"
+#include "cutlass/util/packed_stride.hpp"
 
 namespace flashinfer {
 
@@ -41,21 +46,49 @@ struct cutlass_dtype<nv_bfloat16> {
   using type = cutlass::bfloat16_t;
 };
 
-template <typename T>
-__global__ void compute_cutlass_group_gemm_args(cutlass::gemm::GemmCoord* all_problems, T** ptr_x,
-                                                T** ptr_w, T** ptr_y, int64_t* ld_x, int64_t* ld_w,
-                                                int64_t* ld_y, T* x, T* w, T* y, int64_t* xy_indptr,
-                                                int64_t* w_indices, size_t d_in, size_t d_out,
-                                                bool w_column_major) {
+template <>
+struct cutlass_dtype<__nv_fp8_e4m3> {
+  using type = cutlass::float_e4m3_t;
+};
+
+template <>
+struct cutlass_dtype<__nv_fp8_e5m2> {
+  using type = cutlass::float_e5m2_t;
+};
+
+template <typename DTypeIn, typename DTypeOut>
+__global__ void compute_sm80_cutlass_group_gemm_args(
+    cutlass::gemm::GemmCoord* all_problems, DTypeIn** x_ptr, DTypeIn** w_ptr, DTypeOut** y_ptr,
+    int64_t* x_ld, int64_t* w_ld, int64_t* y_ld, DTypeIn* x, DTypeIn* w, DTypeOut* y,
+    int64_t* xy_indptr, int64_t* w_indices, size_t d_in, size_t d_out, bool w_column_major) {
   int i = blockIdx.x;
   int m = xy_indptr[i + 1] - xy_indptr[i], k = d_in, n = d_out;
   all_problems[i] = cutlass::gemm::GemmCoord(m, n, k);
-  ptr_w[i] = w + (w_indices == nullptr ? i : w_indices[i]) * d_in * d_out;
-  ptr_x[i] = x + xy_indptr[i] * d_in;
-  ptr_y[i] = y + xy_indptr[i] * d_out;
-  ld_x[i] = k;                       // m * k
-  ld_w[i] = w_column_major ? k : n;  // k * n if column major, n * k if row major
-  ld_y[i] = n;                       // m * n
+  w_ptr[i] = w + (w_indices == nullptr ? i : w_indices[i]) * k * n;
+  x_ptr[i] = x + xy_indptr[i] * k;
+  y_ptr[i] = y + xy_indptr[i] * n;
+  x_ld[i] = k;                       // m * k
+  w_ld[i] = w_column_major ? k : n;  // k * n if column major, n * k if row major
+  y_ld[i] = n;                       // m * n
+}
+
+template <typename DTypeIn, typename DTypeOut, typename ProblemShape, typename StrideA,
+          typename StrideB, typename StrideCD>
+__global__ void compute_sm90_cutlass_group_gemm_args(
+    ProblemShape* all_problems, DTypeIn** x_ptr, DTypeIn** w_ptr, DTypeOut** y_ptr,
+    StrideA* x_stride, StrideB* w_stride, StrideCD* y_stride, DTypeIn* x, DTypeIn* w, DTypeOut* y,
+    int64_t* xy_indptr, int64_t* w_indices, size_t d_in, size_t d_out, bool w_column_major) {
+  int i = blockIdx.x;
+  int m = xy_indptr[i + 1] - xy_indptr[i], k = d_in, n = d_out;
+  all_problems[i] = ProblemShape(m, n, k);
+  w_ptr[i] = w + (w_indices == nullptr ? i : w_indices[i]) * k * n;
+  x_ptr[i] = x + xy_indptr[i] * k;
+  y_ptr[i] = y + xy_indptr[i] * n;
+
+  x_stride[i] = cutlass::make_cute_packed_stride(StrideA{}, {m, k, 1});
+  w_stride[i] = w_column_major ? cutlass::make_cute_packed_stride(StrideB{}, {k, n, 1})
+                               : cutlass::make_cute_packed_stride(StrideB{}, {n, k, 1});
+  y_stride[i] = cutlass::make_cute_packed_stride(StrideCD{}, {m, n, 1});
 }
 
 }  // namespace group_gemm