ggml-metal: FWHT kernel for metal backend (#25924)

* metal fwht wip

* shape guard and formatting

* formatting

* Formatting and typos

Co-authored-by: YiChen Lv <63285796+forforever73@users.noreply.github.com>

* fix narrowing issue

Co-authored-by: YiChen Lv <63285796+forforever73@users.noreply.github.com>

* cont : minor style

---------

Co-authored-by: YiChen Lv <63285796+forforever73@users.noreply.github.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Nick Lafleur
2026-07-28 10:44:06 +03:00
committed by GitHub
co-authored by YiChen Lv Georgi Gerganov
parent f87067841b
commit f95de9776b
7 changed files with 143 additions and 1 deletions
+7
View File
@@ -8793,6 +8793,9 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_mul_mat_hadamard(GGML_TYPE_F32, GGML_TYPE_F32, 512, 1, 512));
test_cases.emplace_back(new test_mul_mat_hadamard(GGML_TYPE_F32, GGML_TYPE_F32, 128, 32, 128));
test_cases.emplace_back(new test_mul_mat_hadamard(GGML_TYPE_F32, GGML_TYPE_F32, 128, 4, 128, {2, 3}));
test_cases.emplace_back(new test_mul_mat_hadamard(GGML_TYPE_F32, GGML_TYPE_F32, 256, 512, 256)); // many rows
test_cases.emplace_back(new test_mul_mat_hadamard(GGML_TYPE_F32, GGML_TYPE_F32, 32, 1, 32)); // too small (N<64)
test_cases.emplace_back(new test_mul_mat_hadamard(GGML_TYPE_F32, GGML_TYPE_F32, 1024, 1, 1024)); // too big (N>512)
#if 0
// > 4GB A matrix. Too slow to be enabled by default.
@@ -9803,6 +9806,10 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_perf() {
test_cases.emplace_back(new test_mul_mat_hadamard(GGML_TYPE_F32, GGML_TYPE_F32, 64, 1, 64));
test_cases.emplace_back(new test_mul_mat_hadamard(GGML_TYPE_F32, GGML_TYPE_F32, 256, 1, 256));
test_cases.emplace_back(new test_mul_mat_hadamard(GGML_TYPE_F32, GGML_TYPE_F32, 128, 32, 128));
test_cases.emplace_back(new test_mul_mat_hadamard(GGML_TYPE_F32, GGML_TYPE_F32, 64, 2048, 64));
test_cases.emplace_back(new test_mul_mat_hadamard(GGML_TYPE_F32, GGML_TYPE_F32, 128, 2048, 128));
test_cases.emplace_back(new test_mul_mat_hadamard(GGML_TYPE_F32, GGML_TYPE_F32, 256, 2048, 256));
test_cases.emplace_back(new test_mul_mat_hadamard(GGML_TYPE_F32, GGML_TYPE_F32, 512, 2048, 512));
test_cases.emplace_back(new test_solve_tri(GGML_TYPE_F32, { 64, 64, 4, 4 }, { 32, 64, 4, 4 }));
test_cases.emplace_back(new test_solve_tri(GGML_TYPE_F32, { 128, 128, 4, 2 }, { 32, 128, 4, 2 }));