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

Another faster f16 x f32 matrix multiply kernel

Browse source
This commit is contained in:
Iwan Kawrakow 2023-09-08 16:34:07 +02:00
parent c9e38057f0
commit 9f353f0536
1 changed files with 1 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

1
ggml-metal.metal
View file

@ -682,6 +682,7 @@ kernel void kernel_mul_mat_f16_f32_l4(
device const half4 * x4 = (device const half4 *) (src0 + r0*nb01 + im/(ne12/ne02)*nb02); device const half4 * x4 = (device const half4 *) (src0 + r0*nb01 + im/(ne12/ne02)*nb02);
for (int r1 = 0; r1 < nrows; ++r1) { for (int r1 = 0; r1 < nrows; ++r1) {
device const float4 * y4 = (device const float4 *) (src1 + r1*nb11 + im*nb12); device const float4 * y4 = (device const float4 *) (src1 + r1*nb11 + im*nb12);
float sumf = 0; float sumf = 0;