Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

vulkan: Support Q1_0 (#21539)

Browse Source 
* vulkan: Support Q1_0

* use get_dm
This commit is contained in:
Jeff Bolz
2026-04-10 08:35:27 +02:00
committed by GitHub
parent e095a482a0
commit 7b69125331
9 changed files with 161 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl
+14
View File
@@ -130,6 +130,20 @@ void load_a_to_shmem(const uint pos_a, const uint row, const uint col, const uin
buf_a[buf_idx ] = FLOAT_TYPEV2(v.xy);
buf_a[buf_idx + 1] = FLOAT_TYPEV2(v.zw);
#elif defined(DATA_A_Q1_0)
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 / 16;
const uint iqs = idx & 0xfu;
const float d = float(data_a[ib].d);
const uint bits = uint(data_a[ib].qs[iqs]);
buf_a[buf_idx ] = FLOAT_TYPEV2((bits & 0x01u) != 0u ? d : -d, (bits & 0x02u) != 0u ? d : -d);
buf_a[buf_idx + 1] = FLOAT_TYPEV2((bits & 0x04u) != 0u ? d : -d, (bits & 0x08u) != 0u ? d : -d);
buf_a[buf_idx + 2] = FLOAT_TYPEV2((bits & 0x10u) != 0u ? d : -d, (bits & 0x20u) != 0u ? d : -d);
buf_a[buf_idx + 3] = FLOAT_TYPEV2((bits & 0x40u) != 0u ? d : -d, (bits & 0x80u) != 0u ? d : -d);
#elif defined(DATA_A_Q2_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;