first commit

csrc/flash_attn/src/flash_bwd_hdim160_bf16_causal_sm80.cu

+// Copyright (c) 2024, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::bfloat16_t, 160, true>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim160<cutlass::bfloat16_t, true>(params, stream);
+}

csrc/flash_attn/src/flash_bwd_hdim160_bf16_sm80.cu

+// Copyright (c) 2024, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::bfloat16_t, 160, false>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim160<cutlass::bfloat16_t, false>(params, stream);
+}

csrc/flash_attn/src/flash_bwd_hdim160_fp16_causal_sm80.cu

+// Copyright (c) 2024, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::half_t, 160, true>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim160<cutlass::half_t, true>(params, stream);
+}

csrc/flash_attn/src/flash_bwd_hdim160_fp16_sm80.cu

+// Copyright (c) 2024, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::half_t, 160, false>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim160<cutlass::half_t, false>(params, stream);
+}

csrc/flash_attn/src/flash_bwd_hdim192_bf16_causal_sm80.cu

+// Copyright (c) 2024, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::bfloat16_t, 192, true>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim192<cutlass::bfloat16_t, true>(params, stream);
+}

csrc/flash_attn/src/flash_bwd_hdim192_bf16_sm80.cu

+// Copyright (c) 2024, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::bfloat16_t, 192, false>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim192<cutlass::bfloat16_t, false>(params, stream);
+}

csrc/flash_attn/src/flash_bwd_hdim192_fp16_causal_sm80.cu

+// Copyright (c) 2024, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::half_t, 192, true>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim192<cutlass::half_t, true>(params, stream);
+}

csrc/flash_attn/src/flash_bwd_hdim192_fp16_sm80.cu

+// Copyright (c) 2024, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::half_t, 192, false>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim192<cutlass::half_t, false>(params, stream);
+}

csrc/flash_attn/src/flash_bwd_hdim192_hdim128_bf16_causal_sm80.cu

+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_mla_<cutlass::bfloat16_t, 192, 128, true>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim192_hdim128<cutlass::bfloat16_t, true>(params, stream);
+}
\ No newline at end of file

csrc/flash_attn/src/flash_bwd_hdim192_hdim128_bf16_sm80.cu

+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_mla_<cutlass::bfloat16_t, 192, 128, false>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim192_hdim128<cutlass::bfloat16_t, false>(params, stream);
+}
\ No newline at end of file

csrc/flash_attn/src/flash_bwd_hdim192_hdim128_fp16_causal_sm80.cu

+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_mla_<cutlass::half_t, 192, 128, true>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim192_hdim128<cutlass::half_t, true>(params, stream);
+}
\ No newline at end of file

csrc/flash_attn/src/flash_bwd_hdim192_hdim128_fp16_sm80.cu

+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_mla_<cutlass::half_t, 192, 128, false>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim192_hdim128<cutlass::half_t, false>(params, stream);
+}
\ No newline at end of file

csrc/flash_attn/src/flash_bwd_hdim224_bf16_causal_sm80.cu

+// Copyright (c) 2024, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::bfloat16_t, 224, true>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim224<cutlass::bfloat16_t, true>(params, stream);
+}

csrc/flash_attn/src/flash_bwd_hdim224_bf16_sm80.cu

+// Copyright (c) 2023, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::bfloat16_t, 224, false>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim224<cutlass::bfloat16_t, false>(params, stream);
+}

csrc/flash_attn/src/flash_bwd_hdim224_fp16_causal_sm80.cu

+// Copyright (c) 2024, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::half_t, 224, true>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim224<cutlass::half_t, true>(params, stream);
+}

csrc/flash_attn/src/flash_bwd_hdim224_fp16_sm80.cu

+// Copyright (c) 2023, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::half_t, 224, false>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim224<cutlass::half_t, false>(params, stream);
+}

csrc/flash_attn/src/flash_bwd_hdim256_bf16_causal_sm80.cu

+// Copyright (c) 2024, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::bfloat16_t, 256, true>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim256<cutlass::bfloat16_t, true>(params, stream);
+}

csrc/flash_attn/src/flash_bwd_hdim256_bf16_sm80.cu

+// Copyright (c) 2024, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::bfloat16_t, 256, false>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim256<cutlass::bfloat16_t, false>(params, stream);
+}

csrc/flash_attn/src/flash_bwd_hdim256_fp16_causal_sm80.cu

+// Copyright (c) 2024, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::half_t, 256, true>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim256<cutlass::half_t, true>(params, stream);
+}

csrc/flash_attn/src/flash_bwd_hdim256_fp16_sm80.cu

+// Copyright (c) 2024, Tri Dao.
+// Splitting the different head dimensions to different files to speed up compilation.
+// This file is auto-generated. See "generate_kernels.py"
+
+#include "flash_bwd_launch_template.h"
+
+template<>
+void run_mha_bwd_<cutlass::half_t, 256, false>(Flash_bwd_params &params, cudaStream_t stream) {
+    run_mha_bwd_hdim256<cutlass::half_t, false>(params, stream);
+}