mirror of
				https://github.com/ggml-org/llama.cpp.git
				synced 2025-10-31 08:51:55 +00:00 
			
		
		
		
	 7c7836d9d4
			
		
	
	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
		
			
				
	
	
		
			38 lines
		
	
	
		
			887 B
		
	
	
	
		
			Plaintext
		
	
	
	
	
	
			
		
		
	
	
			38 lines
		
	
	
		
			887 B
		
	
	
	
		
			Plaintext
		
	
	
	
	
	
| #version 450
 | |
| 
 | |
| #include "generic_head.comp"
 | |
| #include "types.comp"
 | |
| 
 | |
| #extension GL_EXT_control_flow_attributes : enable
 | |
| layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
 | |
| 
 | |
| layout (binding = 0) readonly buffer A {A_TYPE data_a[];};
 | |
| layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
 | |
| 
 | |
| layout (constant_id = 0) const uint BLOCK_SIZE = 32;
 | |
| 
 | |
| shared FLOAT_TYPE tmp[BLOCK_SIZE];
 | |
| 
 | |
| void main() {
 | |
|     const uint row = gl_WorkGroupID.x;
 | |
|     const uint col = gl_LocalInvocationID.x;
 | |
| 
 | |
|     tmp[col] = FLOAT_TYPE(0.0f);
 | |
| 
 | |
|     for (uint i = col; i < p.KX; i += BLOCK_SIZE) {
 | |
|         tmp[col] += FLOAT_TYPE(data_a[row*p.KX + i]);
 | |
|     }
 | |
| 
 | |
|     barrier();
 | |
|     [[unroll]] for (int s = int(BLOCK_SIZE) / 2; s > 0; s >>= 1) {
 | |
|         if (col < s) {
 | |
|             tmp[col] += tmp[col + s];
 | |
|         }
 | |
|         barrier();
 | |
|     }
 | |
| 
 | |
|     if (col == 0) {
 | |
|         data_d[row] = D_TYPE(tmp[0]);
 | |
|     }
 | |
| }
 |