upstream-archive/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-07-23 11:10:55 -05:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: batch out_prod broadcast (dps2>1) path with cublasSgemmBatched (#24426)

Browse Source
This commit is contained in:
leonardHONG
2026-06-26 13:51:25 +08:00
committed by GitHub
parent 960d628f46
commit f818065d75
2 changed files with 61 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -8672,6 +8672,12 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
256, 16, 16, {ne2, 1}, {1, 1}));
}
// nr2 sweep to cover the cublasSgemmBatched pointer-array path (dps2 > 1)
for (int64_t nr2 : {8, 16, 32}) {
test_cases.emplace_back(new test_out_prod(GGML_TYPE_F32, GGML_TYPE_F32,
256, 16, 16, {1, 1}, {nr2, 1}));
}
// add_id
for (ggml_type type_a : {GGML_TYPE_F32}) {
for (ggml_type type_b : {GGML_TYPE_F32}) {