Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

vulkan: make FA mask/softcap enables spec constants (#19309)

Browse Source 
* vulkan: make FA mask/softcap enables spec constants

* don't specialize for sinks

* bump timeout a little bit
This commit is contained in:
Jeff Bolz
2026-02-06 08:49:58 +01:00
committed by GitHub
parent 7fcf1ef45d
commit f9bd518a6b
6 changed files with 45 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp
+3 -3
View File
@@ -155,7 +155,7 @@ void main() {
for (uint32_t j = start_j; j < end_j; ++j) {
coopmat<float16_t, gl_ScopeWorkgroup, Br, Bc, gl_MatrixUseAccumulator> mv = coopmat<float16_t, gl_ScopeWorkgroup, Br, Bc, gl_MatrixUseAccumulator>(0);
if ((p.mask_n_head_log2 & MASK_ENABLE_BIT) != 0) {
if (MASK_ENABLE) {
if (USE_MASK_OPT && mask_opt_idx != j / 16) {
mask_opt_idx = j / 16;
@@ -197,14 +197,14 @@ void main() {
coopMatLoadTensorNV(K_T, data_k, k_offset, sliceTensorLayoutNV(tensorLayoutK, j * Bc, Bc, 0, HSK_pad), tensorViewTranspose DECODEFUNC);
S = coopMatMulAdd(Qf16, K_T, S);
if (p.logit_softcap != 0.0f) {
if (LOGIT_SOFTCAP) {
[[unroll]]
for (int k = 0; k < S.length(); ++k) {
S[k] = ACC_TYPE(p.logit_softcap)*tanh(S[k]);
}
}
if ((p.mask_n_head_log2 & MASK_ENABLE_BIT) != 0) {
if (MASK_ENABLE) {
S += slopeMat*coopmat<ACC_TYPE, gl_ScopeWorkgroup, Br, Bc, gl_MatrixUseAccumulator>(mv);
}