vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

CUDA: fix FA out-of-bounds reads (#7479)

Browse source
This commit is contained in:
Johannes Gäßler 2024-05-23 00:31:20 +02:00 • committed by GitHub
parent 1e374365d1
commit cd93a28cb1
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 8 additions and 8 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml-cuda/fattn-tile-f32.cu
View file

@ -79,7 +79,7 @@ static __global__ void flash_attn_tile_ext_f32(
#pragma unroll
for (int i0 = 0; i0 < D; i0 += 2*WARP_SIZE) {
float2 tmp = Q_f2[j*(nb01/sizeof(float2)) + i0/2 + threadIdx.x];
float2 tmp = ic0 + j < ne01 ? Q_f2[j*(nb01/sizeof(float2)) + i0/2 + threadIdx.x] : make_float2(0.0f, 0.0f);
Q_f[j][i0 + 0*WARP_SIZE + threadIdx.x] = tmp.x * scale;
Q_f[j][i0 + 1*WARP_SIZE + threadIdx.x] = tmp.y * scale;
}