CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-01 09:01:57 +00:00
Code Issues Packages Projects Releases Wiki Activity

vulkan: initial support for IQ4_XS quantization (#11501)

Browse Source
This commit is contained in:
Rémy O
2025-02-06 07:09:59 +01:00
committed by GitHub
parent 1b598b3058
commit 8a7e3bf17a
13 changed files with 169 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp
View File

@@ -95,7 +95,7 @@ shared ACC_TYPE coopmat_stage[TM * TN * NUM_WARPS];
#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)
#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_XS) || defined(DATA_A_IQ4_NL)
init_iq_shmem(gl_WorkGroupSize);
#endif
@@ -547,6 +547,25 @@ void main() {
const uint32_t grid = iq3s_grid[qs | ((qh << (8 - (iqs % 8))) & 256)] >> (16 * (idx % 2));
const vec2 v = db * vec2(sign01) * vec2(unpack8(grid).xy);
buf_a[buf_idx ] = FLOAT_TYPE(v.x);
buf_a[buf_idx + 1] = FLOAT_TYPE(v.y);
#elif defined(DATA_A_IQ4_XS)
const uint idx = pos_a + (loadc_a + l) * p.stride_a / LOAD_VEC_A + loadr_a;
const uint buf_idx = (loadc_a + l) * SHMEM_STRIDE + loadr_a * LOAD_VEC_A;
const uint ib = idx / 128; // 2 values per idx
const uint ib32 = (idx % 128) / 16; // 0..7
const uint iq = 16 * ib32 + 2 * (idx % 8);
const uint sl = (data_a[ib].scales_l[ib32/2] >> (4 * (ib32 & 1))) & 0xF;
const uint sh = ((data_a[ib].scales_h) >> (2 * ib32)) & 3;
const uint qshift = (idx & 8) >> 1;
u8vec2 qs = u8vec2(data_a[ib].qs[iq], data_a[ib].qs[iq + 1]);
qs = (qs >> qshift) & uint8_t(0xF);
const float d = float(data_a[ib].d);
const vec2 v = d * float(int(sl | (sh << 4)) - 32) * vec2(kvalues_iq4nl[qs.x], kvalues_iq4nl[qs.y]);
buf_a[buf_idx ] = FLOAT_TYPE(v.x);
buf_a[buf_idx + 1] = FLOAT_TYPE(v.y);
#elif defined(DATA_A_IQ4_NL)