mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2025-08-27 18:49:40 -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:
29
vulkan-shaders/mul_mat_split_k_reduce.comp
Normal file
29
vulkan-shaders/mul_mat_split_k_reduce.comp
Normal file
@@ -0,0 +1,29 @@
|
||||
#version 450
|
||||
|
||||
#extension GL_EXT_control_flow_attributes : enable
|
||||
|
||||
layout(local_size_x = 256, local_size_y = 1, local_size_z = 1) in;
|
||||
|
||||
layout (binding = 0) readonly buffer A {float data_a[];};
|
||||
layout (binding = 1) writeonly buffer D {float data_d[];};
|
||||
|
||||
layout (push_constant) uniform parameter {
|
||||
uint ne;
|
||||
uint k_num;
|
||||
} p;
|
||||
|
||||
void main() {
|
||||
const uint idx = gl_GlobalInvocationID.x;
|
||||
|
||||
if (idx >= p.ne) {
|
||||
return;
|
||||
}
|
||||
|
||||
float result = 0.0f;
|
||||
|
||||
[[unroll]] for (uint i = 0; i < p.k_num; i++) {
|
||||
result += data_a[i * p.ne + idx];
|
||||
}
|
||||
|
||||
data_d[idx] = result;
|
||||
}
|
Reference in New Issue
Block a user