mirror of
https://github.com/RYDE-WORK/llama.cpp.git
synced 2026-01-24 08:33:17 +08:00
* llava: Add ACC OP for GPU acceleration to the Vulkan backend in the LLAVA CLIP model. - The CLIP model now prioritizes the Vulkan backend over the CPU when vulkan available. - A GGML_OP_ACC shader has been added. - The encoding performance of the CLIP model improved from 4.2s on the CPU to 0.9s on the GPU. Signed-off-by: Changyeon Kim <cyzero.kim@samsung.com> * fix-up coding style. Signed-off-by: Changyeon Kim <cyzero.kim@samsung.com> * Fix-up the missing initial parameter to resolve the compilation warning. Signed-off-by: Changyeon Kim <cyzero.kim@samsung.com> * [fix] Add missing parameters. Signed-off-by: Changyeon Kim <cyzero.kim@samsung.com> * [fix] Use nb1 and nb2 for dst. Signed-off-by: Changyeon Kim <cyzero.kim@samsung.com> * Fix check results ggml_acc call --------- Signed-off-by: Changyeon Kim <cyzero.kim@samsung.com> Co-authored-by: 0cc4m <picard12@live.de>
25 lines
692 B
Plaintext
25 lines
692 B
Plaintext
#version 450
|
|
|
|
#include "types.comp"
|
|
#include "generic_binary_head.comp"
|
|
|
|
void main() {
|
|
const uint idx = gl_GlobalInvocationID.x;
|
|
if (idx >= p.ne) {
|
|
return;
|
|
}
|
|
|
|
const uint offset = p.param3;
|
|
const uint src1_i = idx - offset;
|
|
const uint oz = src1_i / p.nb02;
|
|
const uint oy = (src1_i - (oz * p.nb02)) / p.nb01;
|
|
const uint ox = src1_i % p.nb01;
|
|
|
|
if (ox < p.ne10 && oy < p.ne11 && oz < p.ne12) {
|
|
data_d[p.d_offset + dst_idx(idx)] = D_TYPE(FLOAT_TYPE(data_a[src0_idx(idx)]) + FLOAT_TYPE(data_b[ox + oy * p.ne10 + oz * p.ne10 * p.ne11]));
|
|
} else {
|
|
data_d[p.d_offset + dst_idx(idx)] = D_TYPE(FLOAT_TYPE(data_a[src0_idx(idx)]));
|
|
}
|
|
}
|
|
|