mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2024-12-30 21:34:36 +00:00
54 lines
1.4 KiB
Plaintext
54 lines
1.4 KiB
Plaintext
|
/**
|
||
|
|
* Copyright (c) 2023 Nomic, Inc. All rights reserved.
|
||
|
|
*
|
||
|
|
* This software is licensed under the terms of the Software for Open Models
|
||
|
|
* License (SOM), version 1.0, as detailed in the LICENSE_SOM.txt file. A copy
|
||
|
|
* of this license should accompany this software. Except as expressly granted
|
||
|
|
* in the SOM license, all rights are reserved by Nomic, Inc.
|
||
|
|
*/
|
||
|
|
|
||
|
|
#version 450
|
||
|
|
|
||
|
|
#include "common.comp"
|
||
|
|
|
||
|
|
#extension GL_KHR_shader_subgroup_arithmetic : require
|
||
|
|
#extension GL_EXT_debug_printf : enable
|
||
|
|
|
||
|
|
// layout(local_size_x = 8) in;
|
||
|
|
|
||
|
|
layout(binding = 0) readonly buffer tensorInA { float inA[]; };
|
||
|
|
layout(binding = 1) readonly buffer tensorInB { float inB[]; };
|
||
|
|
layout(binding = 2) writeonly buffer tensorOut { float out_[]; };
|
||
|
|
|
||
|
|
layout(push_constant) uniform parameter {
|
||
|
|
uint inAOff;
|
||
|
|
uint inBOff;
|
||
|
|
uint outOff;
|
||
|
|
int ne00;
|
||
|
|
int ne01;
|
||
|
|
int ne02;
|
||
|
|
int ne11;
|
||
|
|
int ne12;
|
||
|
|
uint nb01;
|
||
|
|
uint nb02;
|
||
|
|
uint nb11;
|
||
|
|
uint nb12;
|
||
|
|
uint nb1;
|
||
|
|
uint nb2;
|
||
|
|
}
|
||
|
|
pcs;
|
||
|
|
|
||
|
|
|
||
|
|
void main() {
|
||
|
|
uvec3 gid = gl_GlobalInvocationID;
|
||
|
|
|
||
|
|
const uint x = (gid.x*pcs.nb01 + gid.z/(pcs.ne12/pcs.ne02)*pcs.nb02) / 4 + pcs.inAOff; // Based from inA
|
||
|
|
const uint y = (gid.y*pcs.nb11 + gid.z/(pcs.ne02/pcs.ne12)*pcs.nb12) / 4 + pcs.inBOff; // based from inB
|
||
|
|
float sum = 0.0f;
|
||
|
|
for (uint i = 0; i < pcs.ne00; i ++) {
|
||
|
|
sum += float(inA[x+i]) * float(inB[y+i]);
|
||
|
|
}
|
||
|
|
|
||
|
|
out_[gid.z*(pcs.nb2/4) + gid.y*(pcs.nb1/4) + gid.x + pcs.outOff] = sum;
|
||
|
|
}
|