ExternalVendorCode/whisper.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2025-04-27 06:20:07 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Vulkan: Fix mmq int dot float cache size (llama/12722)

Browse Source
This commit is contained in:
0cc4m 2025-04-02 19:12:30 +02:00 committed by Georgi Gerganov
parent a71c64512a
commit f82622180f
2 changed files with 4 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq.comp
View File

@ -234,9 +234,9 @@ void main() {
#endif #endif
#if QUANT_AUXF == 1 #if QUANT_AUXF == 1
FLOAT_TYPE cache_a_dm[TM]; FLOAT_TYPE cache_a_dm[WMITER * TM];
#else #else
FLOAT_TYPE_VEC2 cache_a_dm[TM]; FLOAT_TYPE_VEC2 cache_a_dm[WMITER * TM];
#endif #endif
FLOAT_TYPE_VEC2 cache_b_ds[TN]; FLOAT_TYPE_VEC2 cache_b_ds[TN];
@ -247,7 +247,6 @@ void main() {
const uint iqs = loadr_a; const uint iqs = loadr_a;
const uint buf_ib = loadc_a + l; const uint buf_ib = loadc_a + l;
// Should ds be gated to a single thread?
if (iqs == 0) { if (iqs == 0) {
#if QUANT_AUXF == 1 #if QUANT_AUXF == 1
buf_a_dm[buf_ib] = get_d(ib); buf_a_dm[buf_ib] = get_d(ib);
@ -276,7 +275,6 @@ void main() {
const uint buf_ib = loadc_b + l; const uint buf_ib = loadc_b + l;
// Should ds be gated to a single thread?
if (iqs == 0) { if (iqs == 0) {
buf_b_ds[buf_ib] = FLOAT_TYPE_VEC2(data_b[ib].ds); buf_b_ds[buf_ib] = FLOAT_TYPE_VEC2(data_b[ib].ds);
} }

4
ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.comp
View File

@ -17,7 +17,7 @@ i32vec2 repack(uint ib, uint iqs) {
} }
ACC_TYPE mul_q8_1(int32_t q_sum, float da, vec2 dsb) { ACC_TYPE mul_q8_1(int32_t q_sum, float da, vec2 dsb) {
return ACC_TYPE(da * (float(q_sum) * dsb.x - 8.0 * dsb.y)); return ACC_TYPE(da * (float(q_sum) * dsb.x - 8.0f * dsb.y));
} }
#endif #endif
@ -51,7 +51,7 @@ i32vec2 repack(uint ib, uint iqs) {
} }
ACC_TYPE mul_q8_1(int32_t q_sum, float da, vec2 dsb) { ACC_TYPE mul_q8_1(int32_t q_sum, float da, vec2 dsb) {
return ACC_TYPE(da * (float(q_sum) * dsb.x - 16.0 * dsb.y)); return ACC_TYPE(da * (float(q_sum) * dsb.x - 16.0f * dsb.y));
} }
#endif #endif