Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

opencl: add neg, exp and diag (#20127)

Browse Source 
* opencl: add `neg`

* opencl: add `exp`

* opencl: add `diag`
This commit is contained in:
lhez
2026-03-05 21:16:39 -08:00
committed by GitHub
parent 2b10b62677
commit 6c97bffd65
5 changed files with 573 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-opencl/kernels/diag.cl
+27
View File
@@ -0,0 +1,27 @@
kernel void kernel_diag_f32(
global const char * src0,
ulong offset0,
global char * dst,
ulong offsetd,
ulong nb01,
ulong nb02,
ulong nb03,
int ne0,
ulong nb0,
ulong nb2,
ulong nb3
) {
src0 = src0 + offset0;
dst = dst + offsetd;
int i3 = get_group_id(2);
int i2 = get_group_id(1);
int i1 = get_group_id(0);
global const float * src0_ptr = (global const float *)(src0 + i2*nb02 + i3*nb03);
global float * dst_ptr = (global float *)(dst + i1*nb01 + i2*nb2 + i3*nb3);
for (int i0 = get_local_id(0); i0 < ne0; i0 += get_local_size(0)) {
dst_ptr[i0] = i0 == i1 ? src0_ptr[i0] : 0.0f;
}
}
ggml/src/ggml-opencl/kernels/exp.cl
+125
View File
@@ -0,0 +1,125 @@
#pragma OPENCL EXTENSION cl_khr_fp16 : enable
kernel void kernel_exp_f32(
global const float * src0,
ulong offset0,
global float * dst,
ulong offsetd,
int n
) {
if (get_global_id(0) >= n) {
return;
}
src0 = (global float*)((global char*)src0 + offset0);
dst = (global float*)((global char*)dst + offsetd);
dst[get_global_id(0)] = exp(src0[get_global_id(0)]);
}
kernel void kernel_exp_f32_4(
global const float4 * src0,
ulong offset0,
global float4 * dst,
ulong offsetd,
int n
) {
if (get_global_id(0) >= n) {
return;
}
src0 = (global float4*)((global char*)src0 + offset0);
dst = (global float4*)((global char*)dst + offsetd);
dst[get_global_id(0)] = exp(src0[get_global_id(0)]);
}
kernel void kernel_exp_f16(
global const half * src0,
ulong offset0,
global half * dst,
ulong offsetd,
int n
) {
if (get_global_id(0) >= n) {
return;
}
src0 = (global half*)((global char*)src0 + offset0);
dst = (global half*)((global char*)dst + offsetd);
dst[get_global_id(0)] = exp(src0[get_global_id(0)]);
}
kernel void kernel_exp_f16_4(
global const half4 * src0,
ulong offset0,
global half4 * dst,
ulong offsetd,
int n
) {
if (get_global_id(0) >= n) {
return;
}
src0 = (global half4*)((global char*)src0 + offset0);
dst = (global half4*)((global char*)dst + offsetd);
dst[get_global_id(0)] = exp(src0[get_global_id(0)]);
}
kernel void kernel_exp_f32_nc(
global const char * src0,
ulong offset0,
global char * dst,
ulong offsetd,
int ne00,
ulong nb00,
ulong nb01,
ulong nb02,
ulong nb03,
ulong nb0,
ulong nb1,
ulong nb2,
ulong nb3
) {
src0 = src0 + offset0;
dst = dst + offsetd;
const int i3 = get_group_id(2);
const int i2 = get_group_id(1);
const int i1 = get_group_id(0);
for (int i0 = get_local_id(0); i0 < ne00; i0 += get_local_size(0)) {
global const float * x = (global const float *)(src0 + i3*nb03 + i2*nb02 + i1*nb01 + i0*nb00);
global float * y = (global float *)(dst + i3*nb3 + i2*nb2 + i1*nb1 + i0*nb0);
*y = exp(*x);
}
}
kernel void kernel_exp_f16_nc(
global const char * src0,
ulong offset0,
global char * dst,
ulong offsetd,
int ne00,
ulong nb00,
ulong nb01,
ulong nb02,
ulong nb03,
ulong nb0,
ulong nb1,
ulong nb2,
ulong nb3
) {
src0 = src0 + offset0;
dst = dst + offsetd;
const int i3 = get_group_id(2);
const int i2 = get_group_id(1);
const int i1 = get_group_id(0);
for (int i0 = get_local_id(0); i0 < ne00; i0 += get_local_size(0)) {
global const half * x = (global const half *)(src0 + i3*nb03 + i2*nb02 + i1*nb01 + i0*nb00);
global half * y = (global half *)(dst + i3*nb3 + i2*nb2 + i1*nb1 + i0*nb0);
*y = exp(*x);
}
}
ggml/src/ggml-opencl/kernels/neg.cl
+125
View File
@@ -0,0 +1,125 @@
#pragma OPENCL EXTENSION cl_khr_fp16 : enable
kernel void kernel_neg_f32(
global const float * src0,
ulong offset0,
global float * dst,
ulong offsetd,
int n
) {
if (get_global_id(0) >= n) {
return;
}
src0 = (global float*)((global char*)src0 + offset0);
dst = (global float*)((global char*)dst + offsetd);
dst[get_global_id(0)] = -src0[get_global_id(0)];
}
kernel void kernel_neg_f32_4(
global const float4 * src0,
ulong offset0,
global float4 * dst,
ulong offsetd,
int n
) {
if (get_global_id(0) >= n) {
return;
}
src0 = (global float4*)((global char*)src0 + offset0);
dst = (global float4*)((global char*)dst + offsetd);
dst[get_global_id(0)] = -src0[get_global_id(0)];
}
kernel void kernel_neg_f16(
global const half * src0,
ulong offset0,
global half * dst,
ulong offsetd,
int n
) {
if (get_global_id(0) >= n) {
return;
}
src0 = (global half*)((global char*)src0 + offset0);
dst = (global half*)((global char*)dst + offsetd);
dst[get_global_id(0)] = -src0[get_global_id(0)];
}
kernel void kernel_neg_f16_4(
global const half4 * src0,
ulong offset0,
global half4 * dst,
ulong offsetd,
int n
) {
if (get_global_id(0) >= n) {
return;
}
src0 = (global half4*)((global char*)src0 + offset0);
dst = (global half4*)((global char*)dst + offsetd);
dst[get_global_id(0)] = -src0[get_global_id(0)];
}
kernel void kernel_neg_f32_nc(
global const char * src0,
ulong offset0,
global char * dst,
ulong offsetd,
int ne00,
ulong nb00,
ulong nb01,
ulong nb02,
ulong nb03,
ulong nb0,
ulong nb1,
ulong nb2,
ulong nb3
) {
src0 = src0 + offset0;
dst = dst + offsetd;
const int i3 = get_group_id(2);
const int i2 = get_group_id(1);
const int i1 = get_group_id(0);
for (int i0 = get_local_id(0); i0 < ne00; i0 += get_local_size(0)) {
global const float * x = (global const float *)(src0 + i3*nb03 + i2*nb02 + i1*nb01 + i0*nb00);
global float * y = (global float *)(dst + i3*nb3 + i2*nb2 + i1*nb1 + i0*nb0);
*y = -*x;
}
}
kernel void kernel_neg_f16_nc(
global const char * src0,
ulong offset0,
global char * dst,
ulong offsetd,
int ne00,
ulong nb00,
ulong nb01,
ulong nb02,
ulong nb03,
ulong nb0,
ulong nb1,
ulong nb2,
ulong nb3
) {
src0 = src0 + offset0;
dst = dst + offsetd;
const int i3 = get_group_id(2);
const int i2 = get_group_id(1);
const int i1 = get_group_id(0);
for (int i0 = get_local_id(0); i0 < ne00; i0 += get_local_size(0)) {
global const half * x = (global const half *)(src0 + i3*nb03 + i2*nb02 + i1*nb01 + i0*nb00);
global half * y = (global half *)(dst + i3*nb3 + i2*nb2 + i1*nb1 + i0*nb0);
*y = -*x;
}
}