mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2024-11-11 21:39:52 +00:00
24 lines
528 B
Plaintext
24 lines
528 B
Plaintext
|
#version 450
|
||
|
|
|
||
|
|
#include "common.comp"
|
||
|
|
|
||
|
|
layout(local_size_x = 1) in;
|
||
|
|
|
||
|
|
layout(binding = 0) buffer restrict readonly tensorIn { float in_[]; };
|
||
|
|
layout(binding = 1) buffer restrict writeonly tensorOut { float out_[]; };
|
||
|
|
|
||
|
|
layout(push_constant) uniform PushConstants {
|
||
|
|
uint inOff;
|
||
|
|
uint outOff;
|
||
|
|
float scale;
|
||
|
|
} pcs;
|
||
|
|
|
||
|
|
void main() {
|
||
|
|
const uint baseIndex = gl_WorkGroupID.x * 8;
|
||
|
|
|
||
|
|
for (uint x = 0; x < 8; x++) {
|
||
|
|
const uint i = baseIndex + x;
|
||
|
|
out_[i + pcs.outOff] = in_[i + pcs.inOff] * pcs.scale;
|
||
|
|
}
|
||
|
|
}
|