llama.cpp
opencl: add several gemm variants for small batches and some fixes
#30850
Merged

Commits
  • opencl: fix OOB read in `mul_mv_mxfp4_f32_flat`
    lhez committed 33 days ago
  • opencl: remove large-N guard that forces CPU weight placement
    lhez committed 32 days ago
  • opencl: cap num of subgroups by workgroup size for q4_0 mc3
    lhez committed 32 days ago
  • opencl: add small batch variant for q4_k dp4a GEMM
    lhez committed 32 days ago
  • opencl: add a narrow variant of mul_mm_f16_f32_l4_lm
    lhez committed 32 days ago
  • opencl: add a one subgroup variant for gemv_noshuffle_q1_0_f32 for big M
    lhez committed 32 days ago
Loading