ggml : build backends as libraries (#10256)

* ggml : build backends as libraries

---------

Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: R0CKSTAR <xiaodong.ye@mthreads.com>
This commit is contained in:
Diego Devesa
2024-11-14 18:04:35 +01:00
committed by GitHub
parent 4a8ccb37ad
commit ae8de6d50a
191 changed files with 17541 additions and 16879 deletions

View File

@@ -0,0 +1,30 @@
#version 450
#include "dequant_head.comp"
layout(local_size_x = 256, local_size_y = 1, local_size_z = 1) in;
layout (binding = 0) readonly buffer A {block_q4_0 data_a[];};
layout (binding = 1) writeonly buffer D {D_TYPE data_b[];};
void main() {
const uint i = gl_WorkGroupID.x * 4 + gl_LocalInvocationID.x / 64;
const uint tid = gl_LocalInvocationID.x % 64;
const uint il = tid/32;
const uint ir = tid%32;
const uint ib = 32*i + ir;
if (ib >= p.nel / 32) {
return;
}
const uint q_idx = 8*il;
const uint b_idx = 1024*i + 32*ir + q_idx;
const float d = float(data_a[ib].d);
[[unroll]] for (uint l = 0; l < 8; ++l) {
data_b[b_idx + l + 0] = D_TYPE(d * ((data_a[ib].qs[q_idx + l] & 0xF) - 8.0f));
data_b[b_idx + l + 16] = D_TYPE(d * ((data_a[ib].qs[q_idx + l] >> 4) - 8.0f));
}
}