happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

fix softmax for iGPU (#17838)

This commit is contained in:
Neo Zhang Jianyu 2025-12-10 16:59:57 +08:00 committed by GitHub
parent 2fbe3b7bb7
commit 2e9eab80c2
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
ggml/src/ggml-sycl/softmax.cpp
View File

@ -277,7 +277,7 @@ static void soft_max_f32_sycl(const float *x, const T *mask,
const int id = get_current_device_id(); const int id = get_current_device_id();
const size_t smpbo = ggml_sycl_info().devices[id].smpbo; const size_t smpbo = ggml_sycl_info().devices[id].smpbo;
if (nbytes_shared <= smpbo) { if (nbytes_shared <= smpbo && ncols_x <= max_block_size) {
launch_soft_max_kernels<32, 64, 128, 256, 512, 1024, 2048, 4096>( launch_soft_max_kernels<32, 64, 128, 256, 512, 1024, 2048, 4096>(
x, mask, sinks, dst, params, stream, block_dims, block_nums, x, mask, sinks, dst, params, stream, block_dims, block_nums,
nbytes_shared); nbytes_shared);