ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

vulkan: implement several ops relevant for ggml_opt (#11769)

Browse source 
* vulkan: support memset_tensor

* vulkan: support GGML_OP_SUM

* vulkan: implement GGML_OP_ARGMAX

* vulkan: implement GGML_OP_SUB

* vulkan: implement GGML_OP_COUNT_EQUAL

* vulkan: implement GGML_OP_OPT_STEP_ADAMW

* vulkan: fix check_results RWKV_WKV6 crash and memory leaks

* vulkan: implement GGML_OP_REPEAT_BACK

* tests: remove invalid test-backend-ops REPEAT_BACK tests

* vulkan: fix COUNT_EQUAL memset using a fillBuffer command
This commit is contained in:
Rémy O 2025-02-17 07:55:57 +01:00 • committed by GitHub
parent 0f2bbe6564
commit 2eea03d86a
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
8 changed files with 569 additions and 223 deletions
Download patch file Download diff file Expand all files Collapse all files

29
ggml/src/ggml-vulkan/vulkan-shaders/sub.comp Normal file
View file

@ -0,0 +1,29 @@
#version 450
#extension GL_EXT_shader_16bit_storage : require
#include "types.comp"
#include "generic_binary_head.comp"
const uint num_threads = 256;
layout(local_size_x = num_threads, local_size_y = 1, local_size_z = 1) in;
void main() {
uint idx = get_idx();
// num_threads * num_iter must equal 512, to match the wg_denoms and get_idx calculation
const uint num_iter = 2;
[[unroll]] for (uint i = 0; i < num_iter; ++i) {
if (idx >= p.ne) {
continue;
}
uint i00, i01, i02, i03;
get_indices(idx, i00, i01, i02, i03);
data_d[get_doffset() + dst_idx(i00, i01, i02, i03)] = D_TYPE(FLOAT_TYPE(data_a[get_aoffset() + src0_idx(i00, i01, i02, i03)]) - FLOAT_TYPE(data_b[get_boffset() + src1_idx(i00, i01, i02, i03)]));
idx += num_threads;
}
}