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

metal : slight speed-up for add and mul kernels (#2917)

Browse source
This commit is contained in:
Georgi Gerganov 2023-09-01 13:42:41 +03:00 • committed by GitHub
parent 4dcd47d71d
commit 13268c5331
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 32 additions and 20 deletions
Download patch file Download diff file Expand all files Collapse all files

32
ggml-metal.metal
View file

@ -25,9 +25,9 @@ typedef struct {
} block_q8_0;
kernel void kernel_add(
device const float * src0,
device const float * src1,
device float * dst,
device const float4 * src0,
device const float4 * src1,
device float4 * dst,
uint tpig[[thread_position_in_grid]]) {
dst[tpig] = src0[tpig] + src1[tpig];
}
@ -35,18 +35,18 @@ kernel void kernel_add(
// assumption: src1 is a row
// broadcast src1 into src0
kernel void kernel_add_row(
device const float * src0,
device const float * src1,
device float * dst,
constant int64_t & ne00,
device const float4 * src0,
device const float4 * src1,
device float4 * dst,
constant int64_t & nb,
uint tpig[[thread_position_in_grid]]) {
dst[tpig] = src0[tpig] + src1[tpig % ne00];
dst[tpig] = src0[tpig] + src1[tpig % nb];
}
kernel void kernel_mul(
device const float * src0,
device const float * src1,
device float * dst,
device const float4 * src0,
device const float4 * src1,
device float4 * dst,
uint tpig[[thread_position_in_grid]]) {
dst[tpig] = src0[tpig] * src1[tpig];
}
@ -54,12 +54,12 @@ kernel void kernel_mul(
// assumption: src1 is a row
// broadcast src1 into src0
kernel void kernel_mul_row(
device const float * src0,
device const float * src1,
device float * dst,
constant int64_t & ne00,
device const float4 * src0,
device const float4 * src1,
device float4 * dst,
constant int64_t & nb,
uint tpig[[thread_position_in_grid]]) {
dst[tpig] = src0[tpig] * src1[tpig % ne00];
dst[tpig] = src0[tpig] * src1[tpig % nb];
}
kernel void kernel_scale(