sleepy/llama.cpp
1
0
Fork 0
Code Issues 8 Pull Requests 1 Actions 177 Packages Projects Releases Wiki Activity

vulkan: handle quantize_q8_1 overflowing the max workgroup count (#18515)

Browse Source 
* vulkan: handle quantize_q8_1 overflowing the max workgroup count

* vulkan: Fix small tile size matmul on lavapipe

* fix mul_mat_id failures
This commit is contained in:
Jeff Bolz
2026-01-05 04:30:14 -06:00
committed by GitHub
parent eadc4184ca
commit b37124d2d2
3 changed files with 44 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/test-backend-ops.cpp
+4
View File
@@ -7625,6 +7625,10 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F32, GGML_TYPE_F32, 64, 77, 77, {12,1}, {1,1}));
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q4_0, GGML_TYPE_F32, 576, 512, 576, {1,1}, {1,1}));
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q4_0, GGML_TYPE_F32, 1, 2048, 8192, {1, 1}, {1, 1}));
for (ggml_type type_a : all_types) {
test_cases.emplace_back(new test_mul_mat(type_a, GGML_TYPE_F32, 1, 64, 256, {1, 1}, {1, 1}));
}
#if 0
// test the mat-mat path for Metal