ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

opencl: split ggml-opencl.cl into multiple files and cleanup (#12886)

Browse source 
* opencl: refactor - split the kernel files

---------

Co-authored-by: Shangqing Gu <quic_shawngu@quicinc.com>

* opencl: split more kernels into separate files

* opencl: specify subgroup size instead of querying it

* opencl: refine Adreno cl compiler version parsing

* opencl: skip some kernels not used by Adreno on old compilers

* opencl: refine logic for selecting Adreno kernels

* opencl: refine Adreno cl compiler version

* opencl: cleanup preprocessor for kernels

* opencl: consider Adreno CL compiler on Windows

* opencl: add final newline for `mul_mv_f16_f16.cl`

---------

Co-authored-by: Shangqing Gu <quic_shawngu@quicinc.com>
This commit is contained in:
lhez 2025-04-15 12:26:00 -07:00 • committed by GitHub
parent f8f820cc4d
commit 80f19b4186
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
43 changed files with 5021 additions and 4996 deletions
Download patch file Download diff file Expand all files Collapse all files

184
ggml/src/ggml-opencl/kernels/cpy.cl Normal file
View file

@ -0,0 +1,184 @@
#pragma OPENCL EXTENSION cl_khr_fp16 : enable
//------------------------------------------------------------------------------
// cpy
//------------------------------------------------------------------------------
kernel void kernel_cpy_f16_f16(
global half * src0,
ulong offset0,
global half * dst,
ulong offsetd,
int ne00,
int ne01,
int ne02,
int ne03,
ulong nb00,
ulong nb01,
ulong nb02,
ulong nb03,
int ne0,
int ne1,
int ne2,
int ne3,
ulong nb0,
ulong nb1,
ulong nb2,
ulong nb3
) {
src0 = (global half*)((global char*)src0 + offset0);
dst = (global half*)((global char*)dst + offsetd);
int i03 = get_group_id(2);
int i02 = get_group_id(1);
int i01 = get_group_id(0);
int n = i03*ne02*ne01*ne00 + i02*ne01*ne00 + i01*ne00;
int i3 = n / (ne2*ne1*ne0);
int i2 = (n - i3*ne2*ne1*ne0) / (ne1*ne0);
int i1 = (n - i3*ne2*ne1*ne0 - i2*ne1*ne0) / ne0;
int i0 = (n - i3*ne2*ne1*ne0 - i2*ne1*ne0 - i1*ne0);
global half * dst_data = (global half *) ((global char *) dst + i3*nb3 + i2*nb2 + i1*nb1 + i0*nb0);
for (int i00 = get_local_id(0); i00 < ne00; i00 += get_local_size(0)) {
global const half * src = (global half *)((global char *) src0 + i03*nb03 + i02*nb02 + i01*nb01 + i00*nb00);
dst_data[i00] = src[0];
}
}
kernel void kernel_cpy_f16_f32(
global half * src0,
ulong offset0,
global float * dst,
ulong offsetd,
int ne00,
int ne01,
int ne02,
int ne03,
ulong nb00,
ulong nb01,
ulong nb02,
ulong nb03,
int ne0,
int ne1,
int ne2,
int ne3,
ulong nb0,
ulong nb1,
ulong nb2,
ulong nb3
) {
src0 = (global half*)((global char*)src0 + offset0);
dst = (global float*)((global char*)dst + offsetd);
int i03 = get_group_id(2);
int i02 = get_group_id(1);
int i01 = get_group_id(0);
int n = i03*ne02*ne01*ne00 + i02*ne01*ne00 + i01*ne00;
int i3 = n / (ne2*ne1*ne0);
int i2 = (n - i3*ne2*ne1*ne0) / (ne1*ne0);
int i1 = (n - i3*ne2*ne1*ne0 - i2*ne1*ne0) / ne0;
int i0 = (n - i3*ne2*ne1*ne0 - i2*ne1*ne0 - i1*ne0);
global float * dst_data = (global float *) ((global char *) dst + i3*nb3 + i2*nb2 + i1*nb1 + i0*nb0);
for (int i00 = get_local_id(0); i00 < ne00; i00 += get_local_size(0)) {
global half * src = (global half *)((global char *) src0 + i03*nb03 + i02*nb02 + i01*nb01 + i00*nb00);
dst_data[i00] = src[0];
}
}
kernel void kernel_cpy_f32_f16(
global float * src0,
ulong offset0,
global half * dst,
ulong offsetd,
int ne00,
int ne01,
int ne02,
int ne03,
ulong nb00,
ulong nb01,
ulong nb02,
ulong nb03,
int ne0,
int ne1,
int ne2,
int ne3,
ulong nb0,
ulong nb1,
ulong nb2,
ulong nb3
) {
src0 = (global float*)((global char*)src0 + offset0);
dst = (global half*)((global char*)dst + offsetd);
int i03 = get_group_id(2);
int i02 = get_group_id(1);
int i01 = get_group_id(0);
int n = i03*ne02*ne01*ne00 + i02*ne01*ne00 + i01*ne00;
int i3 = n / (ne2*ne1*ne0);
int i2 = (n - i3*ne2*ne1*ne0) / (ne1*ne0);
int i1 = (n - i3*ne2*ne1*ne0 - i2*ne1*ne0) / ne0;
int i0 = (n - i3*ne2*ne1*ne0 - i2*ne1*ne0 - i1*ne0);
global half * dst_data = (global half *) ((global char *) dst + i3*nb3 + i2*nb2 + i1*nb1 + i0*nb0);
for (int i00 = get_local_id(0); i00 < ne00; i00 += get_local_size(0)) {
global const float * src = (global float *)((global char *) src0 + i03*nb03 + i02*nb02 + i01*nb01 + i00*nb00);
dst_data[i00] = src[0];
}
}
kernel void kernel_cpy_f32_f32(
global float * src0,
ulong offset0,
global float * dst,
ulong offsetd,
int ne00,
int ne01,
int ne02,
int ne03,
ulong nb00,
ulong nb01,
ulong nb02,
ulong nb03,
int ne0,
int ne1,
int ne2,
int ne3,
ulong nb0,
ulong nb1,
ulong nb2,
ulong nb3
) {
src0 = (global float*)((global char*)src0 + offset0);
dst = (global float*)((global char*)dst + offsetd);
int i03 = get_group_id(2);
int i02 = get_group_id(1);
int i01 = get_group_id(0);
int n = i03*ne02*ne01*ne00 + i02*ne01*ne00 + i01*ne00;
int i3 = n / (ne2*ne1*ne0);
int i2 = (n - i3*ne2*ne1*ne0) / (ne1*ne0);
int i1 = (n - i3*ne2*ne1*ne0 - i2*ne1*ne0) / ne0;
int i0 = (n - i3*ne2*ne1*ne0 - i2*ne1*ne0 - i1*ne0);
global float * dst_data = (global float *) ((global char *) dst + i3*nb3 + i2*nb2 + i1*nb1 + i0*nb0);
for (int i00 = get_local_id(0); i00 < ne00; i00 += get_local_size(0)) {
global const float * src = (global float *)((global char *) src0 + i03*nb03 + i02*nb02 + i01*nb01 + i00*nb00);
dst_data[i00] = src[0];
}
}