tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-29 12:35:16 +00:00
Code Issues Packages Projects Releases Wiki Activity

vulkan: fix mul_mat_vec failure in backend tests (#12529)

Browse Source 
The OOB calculation could be wrong if the last iteration was during one of
the unrolled loops. Adjust the unrolling counts to avoid this. Add a couple
new backend tests that hit this failure on NVIDIA GPUs.
This commit is contained in:
Jeff Bolz
2025-03-24 01:56:17 -05:00
committed by GitHub
parent 77f9c6bbe5
commit 9b169a4d4e
2 changed files with 22 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec.comp
View File

@ -105,6 +105,16 @@ void compute_outputs(const uint32_t first_row, const uint32_t num_rows) {
int unroll_count = 4; int unroll_count = 4;
uint unrolled_iters = num_iters & ~(unroll_count - 1); uint unrolled_iters = num_iters & ~(unroll_count - 1);
#if K_PER_ITER == 2
// If the K dimension is odd, we need lastiter==true on the last iteration
// so OOB is computed correctly. Skip some unrolling to make that happen.
if ((p.ncols & 1) != 0 &&
unrolled_iters == num_iters &&
unrolled_iters > 0) {
unrolled_iters -= unroll_count;
}
#endif
uint i = 0; uint i = 0;
while (i < unrolled_iters) { while (i < unrolled_iters) {
// Manually partially unroll the loop // Manually partially unroll the loop
@ -113,8 +123,18 @@ void compute_outputs(const uint32_t first_row, const uint32_t num_rows) {
i++; i++;
} }
} }
unroll_count = 2; unroll_count = 2;
unrolled_iters = num_iters & ~(unroll_count - 1); unrolled_iters = num_iters & ~(unroll_count - 1);
#if K_PER_ITER == 2
if ((p.ncols & 1) != 0 &&
unrolled_iters == num_iters &&
unrolled_iters > 0) {
unrolled_iters -= unroll_count;
}
#endif
while (i < unrolled_iters) { while (i < unrolled_iters) {
// Manually partially unroll the loop // Manually partially unroll the loop
[[unroll]] for (uint k = 0; k < unroll_count; ++k) { [[unroll]] for (uint k = 0; k < unroll_count; ++k) {

2
tests/test-backend-ops.cpp
View File

@ -4204,6 +4204,8 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32, 83, 2, 64, { 8, 1}, {4, 1})); test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32, 83, 2, 64, { 8, 1}, {4, 1}));
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32, 64, 45, 128, { 8, 1}, {4, 1})); test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32, 64, 45, 128, { 8, 1}, {4, 1}));
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32, 128, 45, 64, { 8, 1}, {4, 1})); test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32, 128, 45, 64, { 8, 1}, {4, 1}));
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32, 1056, 1, 193, {1, 1}, {4, 1}, {0, 2, 1, 3}));
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32, 1056, 1, 67, {1, 1}, {4, 1}, {0, 2, 1, 3}));
for (auto bs : {1,2,4,8}) { for (auto bs : {1,2,4,8}) {
for (auto nr : {1,4}) { for (auto nr : {1,4}) {