vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

tests : fix UB in test-quantize-perf

Browse source
This commit is contained in:
Georgi Gerganov 2023-10-04 14:39:19 +03:00
parent 43bbfb7a47
commit 5b8d50574b
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
1 changed files with 14 additions and 15 deletions
Download patch file Download diff file Expand all files Collapse all files

17
tests/test-quantize-perf.cpp
View file

@ -76,22 +76,21 @@ static void * align_with_offset(void * ptr, int offset) {
return (char *) std::align(MAX_ALIGNMENT, MAX_ALIGNMENT, ptr, dummy_size) + offset;
}
static void benchmark_function(size_t size, size_t q_size, int64_t iterations, const std::function<size_t(void)> & function) {
static void benchmark_function(size_t size, size_t q_size, int64_t iterations, const std::function<float(void)> & func) {
int64_t min_time_us = INT64_MAX;
int64_t total_time_us = 0;
int64_t min_time_cycles = INT64_MAX;
int64_t total_time_cycles = 0;
for (int i = 0; i < WARMUP; i++) {
function();
func();
}
for (int i = 0; i < iterations; i++) {
const int64_t start_time = ggml_time_us();
const int64_t start_cycles = cpu_cycles();
function();
func();
const int64_t end_cycles = cpu_cycles();
const int64_t end_time = ggml_time_us();
@ -283,7 +282,7 @@ int main(int argc, char * argv[]) {
printf(" quantize_row_q_reference\n");
for (size_t size : params.test_sizes) {
printf(" %zu values (%.2f MB)\n", size, 4*size/(float)(1024*1024));
auto quantize_fn = [&](void ) {
auto quantize_fn = [&](void) -> float {
qfns.from_float_reference(test_data1, test_q1, size);
return test_q1[0];
};
@ -297,7 +296,7 @@ int main(int argc, char * argv[]) {
printf(" quantize_row_q\n");
for (size_t size : params.test_sizes) {
printf(" %zu values (%.2f MB)\n", size, 4*size/(float)(1024*1024));
auto quantize_fn = [&](void ) {
auto quantize_fn = [&](void) -> float {
qfns.from_float(test_data1, test_q1, size);
return test_q1[0];
};
@ -312,7 +311,7 @@ int main(int argc, char * argv[]) {
qfns.from_float(test_data1, test_q1, largest);
for (size_t size : params.test_sizes) {
printf(" %zu values (%.2f MB)\n", size, 4*size/(float)(1024*1024));
auto quantize_fn = [&](void ) {
auto quantize_fn = [&](void) -> float {
qfns.to_float(test_q1, test_out, size);
return test_out[0];
};
@ -326,7 +325,7 @@ int main(int argc, char * argv[]) {
printf(" quantize_row_q_dot\n");
for (size_t size : params.test_sizes) {
printf(" %zu values (%.2f MB)\n", size, 4*size/(float)(1024*1024));
auto quantize_fn = [&](void ) {
auto quantize_fn = [&](void) -> float {
auto vdot = ggml_internal_get_type_traits(qfns.vec_dot_type);
vdot.from_float(test_data1, test_q1, size);
return test_q1[0];
@ -343,7 +342,7 @@ int main(int argc, char * argv[]) {
qfns.from_float(test_data2, test_q2, largest);
for (size_t size : params.test_sizes) {
printf(" %zu values (%.2f MB)\n", size, 4*size/(float)(1024*1024));
auto quantize_fn = [&](void ) {
auto quantize_fn = [&](void) -> float {
float result;
qfns.vec_dot(size, &result, test_q1, test_q2);
return result;