EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases 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
43 changed files with 5021 additions and 4996 deletions
Download Patch File Download Diff File Expand all files Collapse all files

62
ggml/src/ggml-opencl/kernels/gelu.cl Normal file
View File

@@ -0,0 +1,62 @@
#pragma OPENCL EXTENSION cl_khr_fp16 : enable
//------------------------------------------------------------------------------
// gelu
//------------------------------------------------------------------------------
#define GELU_COEF_A 0.044715f
#define GELU_QUICK_COEF -1.702f
#define SQRT_2_OVER_PI 0.79788456080286535587989211986876f
kernel void kernel_gelu(
global float * src0,
ulong offset0,
global float * dst,
ulong offsetd
) {
src0 = (global float*)((global char*)src0 + offset0);
dst = (global float*)((global char*)dst + offsetd);
float x = src0[get_global_id(0)];
dst[get_global_id(0)] = 0.5f*x*(1.0f + tanh(SQRT_2_OVER_PI*x*(1.0f + GELU_COEF_A*x*x)));
}
kernel void kernel_gelu_4(
global float4 * src0,
ulong offset0,
global float4 * dst,
ulong offsetd
) {
src0 = (global float4*)((global char*)src0 + offset0);
dst = (global float4*)((global char*)dst + offsetd);
float4 x = src0[get_global_id(0)];
dst[get_global_id(0)] = 0.5f*x*(1.0f + tanh(SQRT_2_OVER_PI*x*(1.0f + GELU_COEF_A*x*x)));
}
kernel void kernel_gelu_quick(
global float * src0,
ulong offset0,
global float * dst,
ulong offsetd
) {
src0 = (global float*)((global char*)src0 + offset0);
dst = (global float*)((global char*)dst + offsetd);
float x = src0[get_global_id(0)];
dst[get_global_id(0)] = x*(1.0f/(1.0f+exp(GELU_QUICK_COEF*x)));
}
kernel void kernel_gelu_quick_4(
global float4 * src0,
ulong offset0,
global float4 * dst,
ulong offsetd
) {
src0 = (global float4*)((global char*)src0 + offset0);
dst = (global float4*)((global char*)dst + offsetd);
float4 x = src0[get_global_id(0)];
dst[get_global_id(0)] = x*(1.0f/(1.0f+exp(GELU_QUICK_COEF*x)));
}