happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

q2_k

This commit is contained in:
Eve 2025-12-31 17:03:24 -05:00
parent 722509e57f
commit 5f3886e615
2 changed files with 7 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl
View File

@ -131,20 +131,21 @@ void load_a_to_shmem(const uint pos_a, const uint row, const uint col, const uin
const uint idx = pos_a + col * p.stride_a / LOAD_VEC_A + row;
const uint buf_idx = col * SHMEM_STRIDE + row * LOAD_VEC_A / 2;
const uint ib = idx / 128; // 2 values per idx
const uint iqs = idx % 128; // 0..127
const uint ib = idx / 64; // 4 values per idx
const uint iqs = (idx % 64) * 2; // 0,2,4..126
const uint qsi = (iqs / 64) * 16 + (iqs % 16); // 0..15
const uint scalesi = iqs / 8; // 0..15
const uint qsshift = ((iqs % 64) / 16) * 2; // 0,2,4,6
const uvec2 qs = uvec2(unpack8(data_a_packed16[ib].qs[qsi]));
const vec4 qs = vec4(unpack8((data_a_packed32[ib].qs[qsi / 2] >> qsshift) & 0x03030303));
const uint scales = data_a[ib].scales[scalesi];
const vec2 dm = vec2(data_a[ib].dm);
const vec2 v = dm.x * float(scales & 0xF) * vec2((qs >> qsshift) & 3) - dm.y * float(scales >> 4);
const vec4 v = dm.x * float(scales & 0xF) * qs - dm.y * float(scales >> 4);
buf_a[buf_idx] = FLOAT_TYPE_VEC2(v.xy);
buf_a[buf_idx ] = FLOAT_TYPE_VEC2(v.xy);
buf_a[buf_idx + 1] = FLOAT_TYPE_VEC2(v.zw);
#elif defined(DATA_A_Q3_K)
const uint idx = pos_a + col * p.stride_a / LOAD_VEC_A + row;
const uint buf_idx = col * SHMEM_STRIDE + row * LOAD_VEC_A / 2;

2
ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
View File

@ -554,7 +554,7 @@ void matmul_shaders(bool fp16, MatMulIdType matmul_id_type, bool coopmat, bool c
std::string load_vec_quant = "2";
if ((tname == "q4_0") || (tname == "q4_1") || (tname == "iq1_s") || (tname == "iq1_m") || (tname == "iq2_xxs") || (tname == "iq2_xs") || (tname == "iq2_s"))
load_vec_quant = "8";
else if ((tname == "q5_0") || (tname == "q5_1") || (tname == "q8_0") || (tname == "q4_k") || (tname == "q5_k") || (tname == "iq3_xxs") || (tname == "iq3_s") || (tname == "iq4_nl") || (tname == "mxfp4"))
else if ((tname == "q5_0") || (tname == "q5_1") || (tname == "q8_0") || (tname == "q2_k") || (tname == "q4_k") || (tname == "q5_k") || (tname == "iq3_xxs") || (tname == "iq3_s") || (tname == "iq4_nl") || (tname == "mxfp4"))
load_vec_quant = "4";
if (tname == "bf16") {