Skip to content

vulkan: faster q6_k matmul #17813

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
netrunnereve wants to merge 3 commits into
base: master
Choose a base branch
from
+18 −8
Open

vulkan: faster q6_k matmul #17813

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
b4bae3f
q6_k faster mul mat
netrunnereve Dec 5, 2025
d8c0d03
8 values
netrunnereve Dec 6, 2025
def0e1b
fix comment
netrunnereve Dec 6, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions .gitignore
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,6 +54,7 @@
/out/
/tmp/
/autogen-*.md
/common/build-info.cpp

# Deprecated

Expand Down
23 changes: 16 additions & 7 deletions ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -240,21 +240,30 @@ 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 / 32; // 8 values per idx
const uint iqs = (idx % 32) * 4; // 0,4,8..124

const uint n = iqs / 64; // 0,1
const uint b = (iqs % 64) / 32; // 0,1
const uint b = ((iqs % 64) / 32) * 4; // 0,4
const uint is_b = (iqs % 16) / 8; // 0,1
const uint qhshift = ((iqs % 64) / 16) * 2; // 0,2,4,6
const uint is = 8 * n + qhshift + is_b; // 0..15
const uint qsi = n * 64 + (iqs % 32) * 2; // 0,2,4..126
const uint qhi = n * 32 + (iqs % 16) * 2; // 0,2,4..62
const uint qsi = n * 32 + (iqs % 32); // 0,4,8..60
const uint qhi = n * 16 + (iqs % 16); // 0,4,8..28

const float dscale = float(data_a[ib].d) * float(data_a[ib].scales[is]);

buf_a[buf_idx] = FLOAT_TYPE_VEC2(dscale * float(int8_t(((data_a[ib].ql[qsi ] >> (b * 4)) & 0xF) | (((data_a[ib].qh[qhi ] >> qhshift) & 3) << 4)) - 32),
dscale * float(int8_t(((data_a[ib].ql[qsi + 1] >> (b * 4)) & 0xF) | (((data_a[ib].qh[qhi + 1] >> qhshift) & 3) << 4)) - 32));
const uint ql_0 = (bitfieldInsert(uint(data_a_packed16[ib].ql[qsi]), uint(data_a_packed16[ib].ql[qsi + 1]), 16, 16) >> b) & 0x0F0F0F0F;
const uint ql_1 = (bitfieldInsert(uint(data_a_packed16[ib].ql[qsi + 2]), uint(data_a_packed16[ib].ql[qsi + 3]), 16, 16) >> b) & 0x0F0F0F0F;
const uint qh_0 = (bitfieldInsert(uint(data_a_packed16[ib].qh[qhi]), uint(data_a_packed16[ib].qh[qhi + 1]), 16, 16) >> qhshift) & 0x03030303;
const uint qh_1 = (bitfieldInsert(uint(data_a_packed16[ib].qh[qhi + 2]), uint(data_a_packed16[ib].qh[qhi + 3]), 16, 16) >> qhshift) & 0x03030303;
const vec4 q_0 = (vec4(unpack8(ql_0 | (qh_0 << 4))) - 32) * dscale;
const vec4 q_1 = (vec4(unpack8(ql_1 | (qh_1 << 4))) - 32) * dscale;

buf_a[buf_idx] = FLOAT_TYPE_VEC2(q_0.x, q_0.y);
buf_a[buf_idx + 1] = FLOAT_TYPE_VEC2(q_0.z, q_0.w);
buf_a[buf_idx + 2] = FLOAT_TYPE_VEC2(q_1.x, q_1.y);
buf_a[buf_idx + 3] = FLOAT_TYPE_VEC2(q_1.z, q_1.w);
#elif defined(DATA_A_IQ1_S)
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;
Expand Down
2 changes: 1 addition & 1 deletion ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -552,7 +552,7 @@ void matmul_shaders(bool fp16, MatMulIdType matmul_id_type, bool coopmat, bool c

for (const auto& tname : type_names) {
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"))
if ((tname == "q4_0") || (tname == "q4_1") || (tname == "q6_k") || (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 == "iq3_xxs") || (tname == "iq3_s") || (tname == "iq4_nl") || (tname == "mxfp4"))
load_vec_quant = "4";
Expand Down
Loading