Skip to content

Commit 04100e8

Browse files
committed
Optimize soft_max
1 parent 14ea4fa commit 04100e8

File tree

1 file changed

+3
-3
lines changed

1 file changed

+3
-3
lines changed

ggml/src/ggml-vulkan/ggml-vulkan.cpp

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -1428,12 +1428,12 @@ static bool ggml_vk_matmul_shmem_support(const vk_device& device, const std::vec
14281428
// Inner map: key is pipeline name; value is the subgroup size.
14291429
static std::unordered_map<std::string, std::unordered_map<std::string, uint32_t>> gpu_pipeline_config = {
14301430
{"RX 5700", {
1431-
{"im2col_f32", 64},
1432-
{"im2col_f32_f16", 64}
1431+
{"soft_max_f32", 64}, {"soft_max_f32_wg512", 64},
1432+
{"soft_max_f32_f16", 64}, {"soft_max_f32_f16_wg512", 64},
1433+
{"im2col_f32", 64}, {"im2col_f32_f16", 64},
14331434
}}
14341435
};
14351436

1436-
// Helper function defined at namespace scope.
14371437
static uint32_t get_subgroup_size(const std::string &pipeline_name, const std::string &device_name) {
14381438
std::string foundKey;
14391439
for (const auto &entry : gpu_pipeline_config) {

0 commit comments

Comments
 (0)