EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases Wiki Activity

Vulkan MMQ Integer Dot Refactor and K-Quant support (#16536)

Browse Source 
* vulkan: add mmq q2_k integer dot support

* Refactor mmq caching

* Reduce mmq register use

* Load 4 quant blocks into shared memory in one step

* Pack q2_k blocks into caches of 32

* Use 32-bit accumulators for integer dot matmul

* Add q4_k mmq

* Add q3_k mmq

* Add q5_k mmq

* Add q6_k mmq

* Add mxfp4 mmq, enable MMQ MUL_MAT_ID

* Fix mmv dm loads
This commit is contained in:
Ruben Ortlam
2025-10-29 14:39:03 +01:00
committed by GitHub
parent 3eb2be1ca5
commit bcf5bda6f5
18 changed files with 941 additions and 418 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q2_k.comp
View File

@@ -41,9 +41,7 @@ void calc_superblock(const uint a_offset, const uint b_offset, const uint itid,
const vec4 qs_u32_4 = vec4(unpack8((qs_u32 >> 4) & 0x03030303));
const vec4 qs_u32_6 = vec4(unpack8((qs_u32 >> 6) & 0x03030303));
vec2 d = vec2(data_a[ib0 + i].d);
const FLOAT_TYPE dall = FLOAT_TYPE(d.x);
const FLOAT_TYPE dmin = FLOAT_TYPE(d.y);
const FLOAT_TYPE_VEC2 dm = vec2(data_a[ib0 + i].dm);
[[unroll]] for (uint j = 0; j < NUM_COLS; ++j) {
vec2 b0 = vec2(data_b_v2[(j*p.batch_stride_b + b_offset + y_idx) / 2 + 0]);
@@ -75,7 +73,7 @@ void calc_superblock(const uint a_offset, const uint b_offset, const uint itid,
fma(FLOAT_TYPE(b96[l]), sccache2[csel][ix][6 + 8*v_im],
fma(FLOAT_TYPE(b112[l]), sccache2[csel][ix][7 + 8*v_im], sum2))))))));
}
temp[j][n] = fma(dall, sum1, fma(-dmin, sum2, temp[j][n]));
temp[j][n] = fma(dm.x, sum1, fma(-dm.y, sum2, temp[j][n]));
}
}
}