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

vulkan: Implement mmvq for iq1_s/iq1_m (#18450)

Browse Source
This commit is contained in:
Jeff Bolz
2026-01-02 20:19:04 +01:00
committed by GitHub
parent 5755e52d15
commit 706e3f93a6
5 changed files with 433 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq.comp
+11
View File
@@ -14,6 +14,8 @@ layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
#define K_PER_ITER 8
#elif defined(DATA_A_QUANT_K)
#define K_PER_ITER 16
#elif defined(DATA_A_IQ1_S) || defined(DATA_A_IQ1_M)
#define K_PER_ITER 32
#else
#error unimplemented
#endif
@@ -49,6 +51,15 @@ void iter(inout FLOAT_TYPE temp[NUM_COLS][NUM_ROWS], const uint first_row, const
cache_b_qs[1] = data_b[b_block_idx_outer].qs[b_block_idx_inner * 8 + b_qs_idx * 4 + 1];
cache_b_qs[2] = data_b[b_block_idx_outer].qs[b_block_idx_inner * 8 + b_qs_idx * 4 + 2];
cache_b_qs[3] = data_b[b_block_idx_outer].qs[b_block_idx_inner * 8 + b_qs_idx * 4 + 3];
#elif K_PER_ITER == 32
cache_b_qs[0] = data_b[b_block_idx_outer].qs[b_block_idx_inner * 8 ];
cache_b_qs[1] = data_b[b_block_idx_outer].qs[b_block_idx_inner * 8 + 1];
cache_b_qs[2] = data_b[b_block_idx_outer].qs[b_block_idx_inner * 8 + 2];
cache_b_qs[3] = data_b[b_block_idx_outer].qs[b_block_idx_inner * 8 + 3];
cache_b_qs[4] = data_b[b_block_idx_outer].qs[b_block_idx_inner * 8 + 4];
cache_b_qs[5] = data_b[b_block_idx_outer].qs[b_block_idx_inner * 8 + 5];
cache_b_qs[6] = data_b[b_block_idx_outer].qs[b_block_idx_inner * 8 + 6];
cache_b_qs[7] = data_b[b_block_idx_outer].qs[b_block_idx_inner * 8 + 7];
#else
#error unimplemented
#endif