Skip to content
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

fix: add "typename" prior to dependent type name #54

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix: add "typename" prior to dependent type name #54

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
8 changes: 4 additions & 4 deletions csrc/flash_attn/src/flash_fwd_kernel.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -362,7 +362,7 @@ inline __device__ void compute_attn_1rowblock(const Params &params, const int bi

// Reshape rP from (MMA=4, MMA_M, MMA_N) to ((4, 2), MMA_M, MMA_N / 2)
// if using m16n8k16 or (4, MMA_M, MMA_N) if using m16n8k8.
Tensor tOrP = make_tensor(rP.data(), FLASH_NAMESPACE::convert_layout_acc_Aregs<Kernel_traits::TiledMma>(rP.layout()));
Tensor tOrP = make_tensor(rP.data(), FLASH_NAMESPACE::convert_layout_acc_Aregs<typename Kernel_traits::TiledMma>(rP.layout()));
// if (cute::thread0()) { print(tOrP); }
FLASH_NAMESPACE::gemm_rs(acc_o, tOrP, tOrVt, tOsVt, tiled_mma, smem_tiled_copy_V, smem_thr_copy_V);
// if (cute::thread0()) { print(scores); }
Expand Down Expand Up @@ -424,7 +424,7 @@ inline __device__ void compute_attn_1rowblock(const Params &params, const int bi

// Reshape rP from (MMA=4, MMA_M, MMA_N) to ((4, 2), MMA_M, MMA_N / 2)
// if using m16n8k16 or (4, MMA_M, MMA_N) if using m16n8k8.
Tensor tOrP = make_tensor(rP.data(), FLASH_NAMESPACE::convert_layout_acc_Aregs<Kernel_traits::TiledMma>(rP.layout()));
Tensor tOrP = make_tensor(rP.data(), FLASH_NAMESPACE::convert_layout_acc_Aregs<typename Kernel_traits::TiledMma>(rP.layout()));
FLASH_NAMESPACE::gemm_rs(acc_o, tOrP, tOrVt, tOsVt, tiled_mma, smem_tiled_copy_V, smem_thr_copy_V);
}

Expand Down Expand Up @@ -942,7 +942,7 @@ inline __device__ void compute_attn_1rowblock_splitkv(const Params &params, cons
Tensor rP = FLASH_NAMESPACE::convert_type<Element>(acc_s);
// Reshape rP from (MMA=4, MMA_M, MMA_N) to ((4, 2), MMA_M, MMA_N / 2)
// if using m16n8k16 or (4, MMA_M, MMA_N) if using m16n8k8.
Tensor tOrP = make_tensor(rP.data(), FLASH_NAMESPACE::convert_layout_acc_Aregs<Kernel_traits::TiledMma>(rP.layout()));
Tensor tOrP = make_tensor(rP.data(), FLASH_NAMESPACE::convert_layout_acc_Aregs<typename Kernel_traits::TiledMma>(rP.layout()));

FLASH_NAMESPACE::gemm_rs(acc_o, tOrP, tOrVt, tOsVt, tiled_mma, smem_tiled_copy_V, smem_thr_copy_V);

Expand Down Expand Up @@ -1002,7 +1002,7 @@ inline __device__ void compute_attn_1rowblock_splitkv(const Params &params, cons
Tensor rP = FLASH_NAMESPACE::convert_type<Element>(acc_s);
// Reshape rP from (MMA=4, MMA_M, MMA_N) to ((4, 2), MMA_M, MMA_N / 2)
// if using m16n8k16 or (4, MMA_M, MMA_N) if using m16n8k8.
Tensor tOrP = make_tensor(rP.data(), FLASH_NAMESPACE::convert_layout_acc_Aregs<Kernel_traits::TiledMma>(rP.layout()));
Tensor tOrP = make_tensor(rP.data(), FLASH_NAMESPACE::convert_layout_acc_Aregs<typename Kernel_traits::TiledMma>(rP.layout()));

FLASH_NAMESPACE::gemm_rs(acc_o, tOrP, tOrVt, tOsVt, tiled_mma, smem_tiled_copy_V, smem_thr_copy_V);
}
Expand Down
6 changes: 3 additions & 3 deletions csrc/flash_attn/src/flash_fwd_sparse_kernel.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -363,7 +363,7 @@ inline __device__ void sparse_attn_1rowblock(const Params &params, const int bid

// Reshape rP from (MMA=4, MMA_M, MMA_N) to ((4, 2), MMA_M, MMA_N / 2)
// if using m16n8k16 or (4, MMA_M, MMA_N) if using m16n8k8.
Tensor tOrP = make_tensor(rP.data(), flash::convert_layout_acc_Aregs<Kernel_traits::TiledMma>(rP.layout()));
Tensor tOrP = make_tensor(rP.data(), flash::convert_layout_acc_Aregs<typename Kernel_traits::TiledMma>(rP.layout()));
// if (cute::thread0()) { print(tOrP); }
flash::gemm_rs(acc_o, tOrP, tOrVt, tOsVt, tiled_mma, smem_tiled_copy_V, smem_thr_copy_V);
// if (cute::thread0()) { print(scores); }
Expand Down Expand Up @@ -422,7 +422,7 @@ inline __device__ void sparse_attn_1rowblock(const Params &params, const int bid

// Reshape rP from (MMA=4, MMA_M, MMA_N) to ((4, 2), MMA_M, MMA_N / 2)
// if using m16n8k16 or (4, MMA_M, MMA_N) if using m16n8k8.
Tensor tOrP = make_tensor(rP.data(), flash::convert_layout_acc_Aregs<Kernel_traits::TiledMma>(rP.layout()));
Tensor tOrP = make_tensor(rP.data(), flash::convert_layout_acc_Aregs<typename Kernel_traits::TiledMma>(rP.layout()));
// if (cute::thread0()) { print(tOrP); }
flash::gemm_rs(acc_o, tOrP, tOrVt, tOsVt, tiled_mma, smem_tiled_copy_V, smem_thr_copy_V);
// if (cute::thread0()) { print(scores); }
Expand Down Expand Up @@ -591,7 +591,7 @@ inline __device__ void sparse_attn_1rowblock(const Params &params, const int bid

// Reshape rP from (MMA=4, MMA_M, MMA_N) to ((4, 2), MMA_M, MMA_N / 2)
// if using m16n8k16 or (4, MMA_M, MMA_N) if using m16n8k8.
Tensor tOrP = make_tensor(rP.data(), flash::convert_layout_acc_Aregs<Kernel_traits::TiledMma>(rP.layout()));
Tensor tOrP = make_tensor(rP.data(), flash::convert_layout_acc_Aregs<typename Kernel_traits::TiledMma>(rP.layout()));
// if (cute::thread0()) { print(tOrP); }
flash::gemm_rs(acc_o, tOrP, tOrVt, tOsVt, tiled_mma, smem_tiled_copy_V, smem_thr_copy_V);
// if (cute::thread0()) { print(scores); }
Expand Down