tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-15 04:33:06 -04:00
Code Issues Packages Projects Releases Wiki Activity

Vulkan Optimizations and Fixes (#8959)

Browse Source 
* Optimize Vulkan REPEAT performance

* Use Vulkan GLSL fused multiply-add instruction where possible

* Add GGML_VULKAN_PERF option to output performance data per operator

* Rework and fix Vulkan descriptor set and descriptor pool handling

* Fix float32 concat f16 shader validation error

* Add Vulkan GROUP_NORM eps parameter

* Fix validation error with transfer queue memory barrier flags

* Remove trailing whitespaces
This commit is contained in:
0cc4m
2024-08-14 18:32:53 +02:00
committed by GitHub
parent 98a532d474
commit 5fd89a70ea
16 changed files with 781 additions and 851 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
ggml/src/vulkan-shaders/mul_mat_vec.comp
View File

@@ -39,8 +39,7 @@ void main() {
vec2 v = dequantize(ib, iqs, a_offset / QUANT_K);
// matrix multiplication
tmp[tid] += FLOAT_TYPE(v.x) * FLOAT_TYPE(data_b[b_offset + iybs + iqs]) +
FLOAT_TYPE(v.y) * FLOAT_TYPE(data_b[b_offset + iybs + iqs + y_offset]);
tmp[tid] = fma(FLOAT_TYPE(v.x), FLOAT_TYPE(data_b[b_offset + iybs + iqs]), fma(FLOAT_TYPE(v.y), FLOAT_TYPE(data_b[b_offset + iybs + iqs + y_offset]), tmp[tid]));
}
// sum up partial sums and write back result