tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-28 12:25:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

metal : add special-case mat-vec mul for ne00 == 4

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-06-26 12:05:43 +03:00
parent 716301d1b0
commit 291f1e2810
3 changed files with 121 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

64
tests/test-backend-ops.cpp
View File

@ -4252,39 +4252,45 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
#if 1
for (ggml_type type_a : base_types) {
for (ggml_type type_b : {GGML_TYPE_F32, GGML_TYPE_F16}) {
// test cases without permutation
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {1, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {1, 1}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {3, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {3, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {3, 2}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {3, 2}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {3, 2}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {3, 2}, {2, 2}));
std::vector<int> ks = { 256 };
if (ggml_blck_size(type_a) == 1) {
ks.push_back(4);
}
for (auto k : ks) {
// test cases without permutation
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {1, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {1, 1}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {3, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {3, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {3, 2}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {3, 2}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {3, 2}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {3, 2}, {2, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {1, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {1, 1}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {3, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {3, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {3, 2}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {3, 2}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {3, 2}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {3, 2}, {2, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {1, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {1, 1}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {3, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {3, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {3, 2}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {3, 2}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {3, 2}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {3, 2}, {2, 2}));
// test cases with permutation
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {2, 3}, {1, 1}, {0, 2, 1, 3}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {2, 3}, {1, 1}, {0, 1, 3, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {2, 3}, {1, 1}, {0, 3, 2, 1}));
// test cases with permutation
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {2, 3}, {1, 1}, {0, 2, 1, 3}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {2, 3}, {1, 1}, {0, 1, 3, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {2, 3}, {1, 1}, {0, 3, 2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 8, 256, {2, 3}, {1, 1}, {0, 2, 1, 3}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 8, 256, {2, 3}, {1, 1}, {0, 1, 3, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 8, 256, {2, 3}, {1, 1}, {0, 3, 2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 8, k, {2, 3}, {1, 1}, {0, 2, 1, 3}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 8, k, {2, 3}, {1, 1}, {0, 1, 3, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 8, k, {2, 3}, {1, 1}, {0, 3, 2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {2, 3}, {1, 1}, {0, 2, 1, 3}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {2, 3}, {1, 1}, {0, 1, 3, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {2, 3}, {1, 1}, {0, 3, 2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {2, 3}, {1, 1}, {0, 2, 1, 3}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {2, 3}, {1, 1}, {0, 1, 3, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {2, 3}, {1, 1}, {0, 3, 2, 1}));
}
// test cases with large ne00/ne10 to cover stream-k fixup
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 1024, {3, 2}, {1, 1}));