mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2024-11-14 14:59:52 +00:00
26 lines
640 B
Plaintext
26 lines
640 B
Plaintext
|
#version 450
|
||
|
|
||
|
#include "common.comp"
|
||
|
|
||
|
layout(local_size_x = 1) in;
|
||
|
|
||
|
layout(binding = 0) buffer restrict readonly tensorInA { float inA[]; };
|
||
|
layout(binding = 1) buffer restrict readonly tensorInB { float inB[]; };
|
||
|
layout(binding = 2) buffer restrict writeonly tensorOut { float out_[]; };
|
||
|
|
||
|
layout(push_constant) uniform PushConstants {
|
||
|
uint inAOff;
|
||
|
uint inBOff;
|
||
|
uint outOff;
|
||
|
uint row;
|
||
|
} pcs;
|
||
|
|
||
|
void main() {
|
||
|
const uint baseIndex = gl_WorkGroupID.x * 4;
|
||
|
|
||
|
for (uint x = 0; x < 4; x++) {
|
||
|
const uint i = baseIndex + x;
|
||
|
out_[i + pcs.outOff] = inA[i + pcs.inAOff] + inB[(i % pcs.row) + pcs.inBOff];
|
||
|
}
|
||
|
}
|