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

vulkan: support copy from f32 to q4_0/q4_1/q5_0/q5_1/q8_0/iq4_nl (#11166)

Browse source 
* vulkan: support copy from f32 to q4_0/q4_1/q5_0/q5_1/q8_0/iq4_nl

Shaders are based on cpy.cu.

* vulkan: support copy from q4_0/q4_1/q5_0/q5_1/q8_0/iq4_nl to f32

* ggml: copy q->f32 assumes some contiguity in the destination
This commit is contained in:
Jeff Bolz 2025-01-16 15:47:10 -06:00 • committed by GitHub
parent 466300fe14
commit bd38ddea01
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
7 changed files with 446 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

6
tests/test-backend-ops.cpp
View file

@ -3808,6 +3808,12 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_cpy(type_src, type_dst, {256, 2, 3, 4}, {0, 2, 1, 3})); // cpy by rows
}
}
for (ggml_type type_dst : {GGML_TYPE_F32}) {
for (ggml_type type_src : all_types) {
test_cases.emplace_back(new test_cpy(type_src, type_dst, {256, 4, 4, 4}));
test_cases.emplace_back(new test_cpy(type_src, type_dst, {256, 2, 3, 4}, {0, 2, 1, 3})); // cpy by rows
}
}
for (ggml_type type_src : {GGML_TYPE_F16, GGML_TYPE_F32}) {
for (ggml_type type_dst : {GGML_TYPE_F16, GGML_TYPE_F32}) {
test_cases.emplace_back(new test_cpy(type_src, type_dst, {256, 2, 3, 4}, {1, 0, 2, 3})); // cpy not-contiguous