llama.cpp
cuda : CUDA GGML_OP_LIGHTNING_INDEXER implementation (generic vector kernel + wmma kernel)
#25545
Merged
Merged
cuda : CUDA GGML_OP_LIGHTNING_INDEXER implementation (generic vector kernel + wmma kernel) #25545
fairydreaming
merged
13 commits into
ggml-org:master
from
fairydreaming:deepseek-lid-cuda
cuda : CUDA GGML_OP_LIGHTNING_INDEXER implementation (generic vector …
50c49c83
Merge remote-tracking branch 'upstream/master' into deepseek-lid-cuda
e624d0fa
JohannesGaessler
commented
on 2026-07-11
chore : remove indentation of #pragma unroll
1ebcf222
cuda : remove unnecessary kernel template declarations
07bfc349
cuda : add WARPS_PER_BLOCK and K_VECS_PER_BLOCK template parameters i…
8d59c209
cuda : relax MMA architecture requirements to Turing in lightning ind…
d372562b
chore : renamed variables
9937427c
chore : rename ggml_cuda_op_lightning_indexer() to ggml_cuda_lightnin…
44b29b78
chore : TODO for AMD rocWMMA
8e683483
JohannesGaessler
approved these changes
on 2026-07-14
am17an
approved these changes
on 2026-07-14
chore : whitespace formatting
aaa8968f
chore : another variable rename to fix problems caused by shadowing
01878d93
chore : yet another rename, this time uppercased all constants
3ab3bb73
cuda : added alignment checks for Q and K tensors in lightning indexe…
37b6a65d
am17an
approved these changes
on 2026-07-15
Login to write a write a comment.
Login via GitHub