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

Merge branch 'optimize_quants_upstream' into concedo_experimental

Browse source
This commit is contained in:
Concedo 2023-06-23 23:56:31 +08:00
commit 6d718525c4
1 changed files with 3 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

3
ggml-opencl.cpp
View file

@ -184,7 +184,9 @@ void convert_f16(__global half* x, const int ib, const int iqs, float* v0, float
*v0 = vload_half(0, &x[ib + 0]); *v0 = vload_half(0, &x[ib + 0]);
*v1 = vload_half(0, &x[ib + 1]); *v1 = vload_half(0, &x[ib + 1]);
} }
);
static std::string k_quants_source = MULTILINE_QUOTE(
inline void get_scale_min_k4(int j, const __global uint8_t *q, uint8_t *d, uint8_t *m) inline void get_scale_min_k4(int j, const __global uint8_t *q, uint8_t *d, uint8_t *m)
{ {
if (j < 4) if (j < 4)
@ -856,6 +858,7 @@ std::string& replace(std::string& s, const std::string& from, const std::string&
std::string generate_kernels() { std::string generate_kernels() {
std::stringstream src; std::stringstream src;
src << program_source << '\n'; src << program_source << '\n';
src << k_quants_source << '\n';
for (size_t i = 0; i < dequant_str_values.size(); i += dequant_str_keys.size()) { for (size_t i = 0; i < dequant_str_values.size(); i += dequant_str_keys.size()) {
std::string dequant_kernel = dequant_template; std::string dequant_kernel = dequant_template;
std::string dmmv_kernel = dequant_mul_mat_vec_template; std::string dmmv_kernel = dequant_mul_mat_vec_template;