vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-11 01:24:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

vulkan: fuse adds (#15252)

Browse source 
* vulkan: fuse adds

Fuse adds that have the same shape, which are common in MoE models.
It will currently fuse up to 6 adds, because we assume no more than
8 descriptors per dispatch. But this could be changed.

* check runtimeDescriptorArray feature

* disable multi_add for Intel due to likely driver bug
This commit is contained in:
Jeff Bolz 2025-08-16 11:48:22 -05:00 • committed by GitHub
parent de2192794f
commit 1fe00296f5
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
6 changed files with 301 additions and 25 deletions
Download patch file Download diff file Expand all files Collapse all files

5
tests/test-backend-ops.cpp
View file

@ -2491,12 +2491,12 @@ struct test_bin_bcast : public test_case {
: op(op), type(type), ne(ne), nr(nr), nf(nf) {}
ggml_tensor * build_graph(ggml_context * ctx) override {
GGML_ASSERT(nf <= 8);
GGML_ASSERT(nf <= 16);
ggml_tensor * a = ggml_new_tensor_4d(ctx, type, ne[0]*nr[0], ne[1]*nr[1], ne[2]*nr[2], ne[3]*nr[3]);
ggml_set_name(a, "a");
ggml_tensor * b[8];
ggml_tensor * b[16];
for (int i = 0; i < nf; ++i) {
b[i] = ggml_new_tensor(ctx, type, 4, ne.data());
ggml_set_name(b[i], (std::string("b") + std::to_string(i)).c_str());
@ -5658,6 +5658,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_bin_bcast(ggml_add, GGML_TYPE_F32, {10, 5, 4, 3}, {1, 1, 2, 2}, 6));
test_cases.emplace_back(new test_bin_bcast(ggml_add, GGML_TYPE_F32, {10, 5, 4, 3}, {1, 2, 2, 2}, 7));
test_cases.emplace_back(new test_bin_bcast(ggml_add, GGML_TYPE_F32, {16, 5, 4, 3}, {2, 2, 2, 2}, 8));
test_cases.emplace_back(new test_bin_bcast(ggml_add, GGML_TYPE_F32, {16, 5, 4, 3}, {1, 1, 1, 1}, 16));
test_cases.emplace_back(new test_add1());
test_cases.emplace_back(new test_scale());