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

vulkan: change an assertion

Browse source 
src1 can go down the first pipeline as nullptr and src0 only needs to be checked once

this means the assertion is only required to check if the type is GGML_TYPE_F16 and can usually be skipped
This commit is contained in:
FirstTimeEZ 2024-11-16 17:17:10 +13:00
parent dd3a6ce9f8
commit d283dbcd37
1 changed files with 8 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

15
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View file

@ -3929,13 +3929,14 @@ static vk_pipeline ggml_vk_op_get_pipeline(ggml_backend_vk_context * ctx, const
}
return nullptr;
case GGML_OP_SOFT_MAX:
GGML_ASSERT(!src1 || src1->type == GGML_TYPE_F32 || src1->type == GGML_TYPE_F16);
if (src0->type == GGML_TYPE_F32 && (src1 == nullptr || src1->type == GGML_TYPE_F32) && dst->type == GGML_TYPE_F32) {
return ctx->device->pipeline_soft_max_f32;
}
if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F16 && dst->type == GGML_TYPE_F32) {
return ctx->device->pipeline_soft_max_f32_f16;
if (src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32) {
if (src1 == nullptr || src1->type == GGML_TYPE_F32) {
return ctx->device->pipeline_soft_max_f32;
}
else {
GGML_ASSERT(src1->type == GGML_TYPE_F16);
return ctx->device->pipeline_soft_max_f32_f16;
}
}
return nullptr;
case GGML_OP_ROPE: