66ee4f297c
* vulkan: initial support for IQ3_S * vulkan: initial support for IQ3_XXS * vulkan: initial support for IQ2_XXS * vulkan: initial support for IQ2_XS * vulkan: optimize Q3_K by removing branches * vulkan: implement dequantize variants for coopmat2 * vulkan: initial support for IQ2_S * vulkan: vertically realign code * port failing dequant callbacks from mul_mm * Fix array length mismatches * vulkan: avoid using workgroup size before it is referenced * tests: increase timeout for Vulkan llvmpipe backend --------- Co-authored-by: Jeff Bolz <jbolz@nvidia.com>
51 lines
1.5 KiB
Text
51 lines
1.5 KiB
Text
#version 450
|
|
|
|
#include "types.comp"
|
|
#include "generic_unary_head.comp"
|
|
#include "dequant_funcs.comp"
|
|
|
|
#if defined(DATA_A_IQ4_NL)
|
|
// 16 invocations needed for init_iq4nl_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() {
|
|
#if defined(DATA_A_IQ2_XXS) || defined(DATA_A_IQ2_XS) || defined(DATA_A_IQ2_S) || defined(DATA_A_IQ3_XXS) || defined(DATA_A_IQ3_S) || defined(DATA_A_IQ4_NL)
|
|
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
|
|
}
|
|
}
|