mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2025-11-09 10:17:06 +00:00
e29acf74fe
* vulkan (DRAFT): split shader generation by GLSL source file, to improve incremental build times * support dep-files so shaders are recompiled if their included files change * rename shader files which are used as "headers" to use .glsl extension * move glslc extension detection shaders to separate folders * the above is to prevent them from getting glob'd with the actual compute shaders that need to be compiled * vulkan : only write embedded shader .hpp/.cpp when they change * avoid recompiling ggml-vulkan.cpp when editing shaders * pass single --source argument instead of --input-dir & --filter to shader gen * check for source file match earlier * fix hang in vulkan-shaders-gen when there are compilation errors * early out did not decrement compile_count * clean up * fix glslc integer dot product test * unconditionally write the embedded shader cpp output * replace output filepath in generated dep-files to match output in CMakeLists --------- Co-authored-by: Jeff Bolz <jbolz@nvidia.com>
52 lines
1.4 KiB
Plaintext
52 lines
1.4 KiB
Plaintext
#version 450
|
|
|
|
#include "types.glsl"
|
|
#include "generic_unary_head.glsl"
|
|
#include "dequant_funcs.glsl"
|
|
|
|
#if defined(DATA_A_IQ4_NL) || defined(DATA_A_MXFP4)
|
|
// 16 invocations needed for init_iq_shmem
|
|
layout(local_size_x = 16, local_size_y = 1, local_size_z = 1) in;
|
|
#else
|
|
layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
|
|
#endif
|
|
|
|
void main() {
|
|
#ifdef NEEDS_INIT_IQ_SHMEM
|
|
init_iq_shmem(gl_WorkGroupSize);
|
|
if (gl_LocalInvocationIndex.x != 0) {
|
|
return;
|
|
}
|
|
#endif
|
|
|
|
const uint idx = gl_WorkGroupID.z * 262144 + gl_WorkGroupID.y * 512 + gl_WorkGroupID.x * QUANT_K;
|
|
|
|
if (idx >= p.ne) {
|
|
return;
|
|
}
|
|
|
|
uint dst_idx = get_doffset() + dst_idx(idx);
|
|
uint src_idx = src0_idx_quant(idx, QUANT_K);
|
|
|
|
const uint a_offset = 0;
|
|
const uint ib = src_idx;
|
|
const vec2 dm = get_dm(ib, a_offset);
|
|
|
|
[[unroll]] for (int j = 0; j < QUANT_K; j += 4) {
|
|
vec4 v = dequantize4(ib, j / QUANT_R, a_offset);
|
|
v = v * dm.x + vec4(dm.y);
|
|
|
|
#if QUANT_R == 2
|
|
data_d[dst_idx + j/2 + 0] = v[0];
|
|
data_d[dst_idx + j/2 + QUANT_K/2 + 0] = v[1];
|
|
data_d[dst_idx + j/2 + 1] = v[2];
|
|
data_d[dst_idx + j/2 + QUANT_K/2 + 1] = v[3];
|
|
#else
|
|
data_d[dst_idx + j + 0] = v[0];
|
|
data_d[dst_idx + j + 1] = v[1];
|
|
data_d[dst_idx + j + 2] = v[2];
|
|
data_d[dst_idx + j + 3] = v[3];
|
|
#endif
|
|
}
|
|
}
|