Skip to content

CUDA: fix pointer incrementation in FA #14916

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 1 commit into from
Jul 28, 2025
Merged

CUDA: fix pointer incrementation in FA #14916

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
9 changes: 4 additions & 5 deletions ggml/src/ggml-cuda/fattn-vec-f16.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -174,7 +174,10 @@ static __global__ void flash_attn_vec_ext_f16(
K += blockIdx.y*D * nb11;
V += blockIdx.y*D * nb21;
maskh += blockIdx.y*D;
for (int k_VKQ_0 = blockIdx.y*D; k_VKQ_0 < ne11; k_VKQ_0 += gridDim.y*D) {
for (int k_VKQ_0 = blockIdx.y*D; k_VKQ_0 < ne11; k_VKQ_0 += gridDim.y*D,
// Increment pointers after each loop:
K += gridDim.y*D*nb11, V += gridDim.y*D*nb21, maskh += gridDim.y*D) {

// Calculate KQ tile and keep track of new maximum KQ values:

if (mask) {
Expand Down Expand Up @@ -291,10 +294,6 @@ static __global__ void flash_attn_vec_ext_f16(
}
}

K += gridDim.y*D * nb11;
V += gridDim.y*D * nb21;
maskh += gridDim.y*D;

__syncthreads();
}

Expand Down
9 changes: 4 additions & 5 deletions ggml/src/ggml-cuda/fattn-vec-f32.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -180,7 +180,10 @@ static __global__ void flash_attn_vec_ext_f32(
K += blockIdx.y*D * nb11;
V += blockIdx.y*D * nb21;
maskh += blockIdx.y*D;
for (int k_VKQ_0 = blockIdx.y*D; k_VKQ_0 < ne11; k_VKQ_0 += gridDim.y*D) {
for (int k_VKQ_0 = blockIdx.y*D; k_VKQ_0 < ne11; k_VKQ_0 += gridDim.y*D,
// Increment pointers after each loop:
K += gridDim.y*D*nb11, V += gridDim.y*D*nb21, maskh += gridDim.y*D) {

// Calculate KQ tile and keep track of new maximum KQ values:

if (mask) {
Expand Down Expand Up @@ -286,10 +289,6 @@ static __global__ void flash_attn_vec_ext_f32(
}
}

K += gridDim.y*D * nb11;
V += gridDim.y*D * nb21;
maskh += gridDim.y*D;

__syncthreads();
}

Expand Down
Loading