mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2024-12-27 14:49:25 +01:00
7c7836d9d4
* Refactor shaders, extract GLSL code from ggml_vk_generate_shaders.py into vulkan-shaders directory * Improve debug log code * Add memory debug output option * Fix flake8 * Fix unnecessary high llama-3 VRAM use
35 lines
799 B
Plaintext
35 lines
799 B
Plaintext
#version 450
|
|
|
|
#extension GL_EXT_shader_16bit_storage : require
|
|
#extension GL_EXT_control_flow_attributes : enable
|
|
|
|
layout (push_constant) uniform parameter
|
|
{
|
|
uint ncols;
|
|
uint rows_per_channel;
|
|
uint n_past;
|
|
} p;
|
|
|
|
#include "types.comp"
|
|
|
|
layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
|
|
|
|
layout (binding = 0) readonly buffer X {A_TYPE data_a[];};
|
|
layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
|
|
|
|
void main() {
|
|
const uint col = gl_GlobalInvocationID.y;
|
|
const uint row = gl_GlobalInvocationID.x;
|
|
|
|
if (col >= p.ncols) {
|
|
return;
|
|
}
|
|
|
|
const uint i = row*p.ncols + col;
|
|
if (col > p.n_past + row % p.rows_per_channel) {
|
|
data_d[i] = D_TYPE(uintBitsToFloat(0xFF800000));
|
|
} else {
|
|
data_d[i] = D_TYPE(data_a[i]);
|
|
}
|
|
}
|