mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2025-08-11 03:03:47 -04:00
Vulkan Shader Refactor, Memory Debugging Option (#7947)
* 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
This commit is contained in:
34
vulkan-shaders/diag_mask_inf.comp
Normal file
34
vulkan-shaders/diag_mask_inf.comp
Normal file
@@ -0,0 +1,34 @@
|
||||
#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]);
|
||||
}
|
||||
}
|
Reference in New Issue
Block a user