public_git
/
whisper.cpp
mirror of https://github.com/ggml-org/whisper.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge 116669655c into 1fe009caed

This commit is contained in:
Jamaika1 2026-08-15 23:11:23 +08:00 committed by GitHub
parent 1fe009caed 116669655c
commit b7dac07a7b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
ggml/src/ggml-cuda/fattn-common.cuh
View File

@ -4,6 +4,7 @@
#include "convert.cuh" #include "convert.cuh"
#include "vecdotq.cuh" #include "vecdotq.cuh"
#include <cmath>
#include <cstdint> #include <cstdint>
#define FATTN_KQ_STRIDE 256 #define FATTN_KQ_STRIDE 256
@ -689,7 +690,7 @@ static __global__ void flash_attn_mask_to_KV_max(
#pragma unroll #pragma unroll
for (int j = 0; j < ncols1; ++j) { for (int j = 0; j < ncols1; ++j) {
const float2 tmp = __half22float2(mask[j*s31 + KV_max_sj/2 + tid]); const float2 tmp = __half22float2(mask[j*s31 + KV_max_sj/2 + tid]);
all_inf = all_inf && int(isinf(tmp.x)) && int(isinf(tmp.y)); all_inf = all_inf && int(std::isinf(tmp.x)) && int(std::isinf(tmp.y));
} }
all_inf = warp_reduce_all(all_inf); all_inf = warp_reduce_all(all_inf);