happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

updated test cases using tensor shapes from sd.cpp wan video generation

This commit is contained in:
bssrdf 2025-11-10 15:26:09 -05:00
parent 15daa5a6a8
commit 89103a856c
2 changed files with 17 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
tests/test-backend-ops.cpp
View File

@ -7375,10 +7375,11 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_perf() {
}
for (ggml_type kernel_type : {GGML_TYPE_F32, GGML_TYPE_F16}) {
// for (ggml_type kernel_type : {GGML_TYPE_F32, GGML_TYPE_F16}) {
for (ggml_type kernel_type : {GGML_TYPE_F16}) {
for (int N : {1}) {
for (int IC : {320, 640}) {
for (int OC : {320, 640}) {
for (int IC : {48, 320, 640, 1024}) {
for (int OC : {320, 640, 1024, 2048}) {
for (int s0 : {1}) {
for (int p1 : {1}) {
for (int d2 : {1}) {
@ -7404,12 +7405,12 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_perf() {
OC, KD, KH, KW,
s0, s1, s2, p0, p1, p2, d0, d1, d2,
kernel_type));
IW = 208; IH = 316;
test_cases.emplace_back(new test_conv_3d(
N, IC, ID, IH, IW,
OC, KD, KH, KW,
s0, s1, s2, p0, p1, p2, d0, d1, d2,
kernel_type));
// IW = 208; IH = 316;
// test_cases.emplace_back(new test_conv_3d(
// N, IC, ID, IH, IW,
// OC, KD, KH, KW,
// s0, s1, s2, p0, p1, p2, d0, d1, d2,
// kernel_type));
}
}
}

9
tests/test-conv3d.cpp
View File

@ -358,7 +358,12 @@ int main(void)
// std::make_tuple(320,1280,104,152,8,3,3,3),
// std::make_tuple(1280,1280,104,152,8,3,3,3),
// std::make_tuple(320,1280,208,304,4,3,3,3),
// std::make_tuple(640,1280,208,304,4,3,3,3),
std::make_tuple(1024,2048,30,52,3,3,3,3),
std::make_tuple(1024,2048,52,76,4,3,3,3),
std::make_tuple(1024,2048,52,76,6,3,3,3),
std::make_tuple(48,3072,64,64,9,2,2,1),
std::make_tuple(48,3072,64,64,17,2,2,1),
std::make_tuple(48,3072,64,64,33,2,2,1),
};
int k = 0;
@ -380,7 +385,7 @@ int main(void)
// fprintf(stderr, "%s: compute buffer size: %.2f MB\n", __func__, mem_size/1024.0f/1024.0f);
int iterations = 0;
int iterations = 20;
double run_time0;
std::vector<float> im2col_data = compute_graph(model, allocr, build_graph_0, iterations,