mozempk/ggml-org-llama-cpp-mirror
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-11-11 22:32:07 +01:00
Code Issues Packages Projects Releases Wiki Activity

opencl: add optimized q4_1 mm kernel for adreno (#19840)

Browse Source 
* Add Q4_1 OpenCL Kernels

* opencl: refactor transpose

* opencl: format

* opencl: refactor q4_1 unpack

* opencl: move `ggml_cl_mul_mat_q4_1_f32_adreno`

* opencl: refactor `ggml_cl_mul_mat_q4_1_f32_adreno` and kernels

* opencl: rename kernel files and kernes

* opencl: fix build for non adreno

* opencl: move code around and format

---------

Co-authored-by: Li He <lih@qti.qualcomm.com>
This commit is contained in:
shaofeiqi
2026-03-02 19:49:41 -08:00
committed by GitHub
co-authored by Li He
parent 49a7564ac1
commit 24350fdf9b
7 changed files with 879 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-opencl/kernels/cvt.cl
+52
View File
@@ -199,6 +199,58 @@ kernel void kernel_restore_block_q4_1(
}
}
kernel void kernel_convert_block_q4_1_noshuffle(
global struct block_q4_1 * src0,
global uchar * dst_q,
global half * dst_d,
global half * dst_m
) {
global struct block_q4_1 * b = (global struct block_q4_1 *) src0 + get_global_id(0);
global uchar * q = (global uchar *) dst_q + QK4_1/2*get_global_id(0);
global half * d = (global half *) dst_d + get_global_id(0);
global half * m = (global half *) dst_m + get_global_id(0);
*d = b->d;
*m = b->m;
for (int i = 0; i < QK4_1/4; ++i) {
uchar x0 = b->qs[2*i + 0];
uchar x1 = b->qs[2*i + 1];
q[i + 0 ] = convert_uchar(x0 & 0x0F) | convert_uchar((x1 & 0x0F) << 4);
q[i + QK4_1/4] = convert_uchar((x0 & 0xF0) >> 4) | convert_uchar(x1 & 0xF0);
#ifdef ADRENO_GPU
if (get_global_id(0) == 65536*4096) {
printf("%04x - %02x\n", *(global ushort*)d, ((x0 & 0xF0) >> 4) | (x1 & 0xF0));
}
#endif
}
}
kernel void kernel_restore_block_q4_1_noshuffle(
global uchar * src_q,
global half * src_d,
global half * src_m,
global struct block_q4_1 * dst,
uchar mask_0F,
uchar mask_F0
) {
global struct block_q4_1 * b = (global struct block_q4_1 *) dst + get_global_id(0);
global uchar * q = (global uchar *) src_q + QK4_1/2*get_global_id(0);
global half * d = (global half *) src_d + get_global_id(0);
global half * m = (global half *) src_m + get_global_id(0);
b->d = *d;
b->m = *m;
for (int i = 0; i < QK4_1/4; ++i) {
uchar x0 = q[i + 0 ] ;
uchar x1 = q[i + QK4_1/4];
b->qs[2*i + 0] = convert_uchar((x0 & mask_0F) | ((x1 & mask_0F) << 4));
b->qs[2*i + 1] = convert_uchar(((x0 & mask_F0) >> 4) | (x1 & mask_F0));
}
}
//------------------------------------------------------------------------------
// block_mxfp4
//------------------------------------------------------------------------------