mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2026-11-11 22:32:07 +01:00
vulkan: make FA mask/softcap enables spec constants (#19309)
* vulkan: make FA mask/softcap enables spec constants * don't specialize for sinks * bump timeout a little bit
This commit is contained in:
@@ -127,7 +127,7 @@ void main() {
|
||||
continue;
|
||||
}
|
||||
// Only load if the block is not all zeros
|
||||
if ((p.mask_n_head_log2 & MASK_ENABLE_BIT) != 0 && mask_opt_bits != MASK_OPT_ALL_ZERO) {
|
||||
if (MASK_ENABLE && mask_opt_bits != MASK_OPT_ALL_ZERO) {
|
||||
bool nem1_bounds_check = !(p.gqa_ratio > 1) && (p.nem1 % Br) != 0;
|
||||
|
||||
[[unroll]] for (uint32_t idx = 0; idx < Bc * Br; idx += gl_WorkGroupSize.x) {
|
||||
@@ -181,7 +181,7 @@ void main() {
|
||||
}
|
||||
}
|
||||
|
||||
if (p.logit_softcap != 0.0f) {
|
||||
if (LOGIT_SOFTCAP) {
|
||||
[[unroll]] for (uint32_t r = 0; r < Br; ++r) {
|
||||
[[unroll]] for (uint32_t c = 0; c < cols_per_thread; ++c) {
|
||||
Sf[r][c] = p.logit_softcap * tanh(Sf[r][c]);
|
||||
@@ -189,7 +189,7 @@ void main() {
|
||||
}
|
||||
}
|
||||
|
||||
if ((p.mask_n_head_log2 & MASK_ENABLE_BIT) != 0 && mask_opt_bits != MASK_OPT_ALL_ZERO) {
|
||||
if (MASK_ENABLE && mask_opt_bits != MASK_OPT_ALL_ZERO) {
|
||||
[[unroll]] for (uint32_t c = 0; c < cols_per_thread; ++c) {
|
||||
[[unroll]] for (uint32_t r = 0; r < Br; ++r) {
|
||||
float mvf = masksh[c * cols_per_iter + col_tid][r];
|
||||
|
||||
Reference in New Issue
Block a user