This is an automated email from the ASF dual-hosted git repository.
github-bot pushed a change to branch nightly
in repository https://gitbox.apache.org/repos/asf/tvm.git
from 18a2a250f8 [KVCache] Support KVCache decode from forked sequence and
pop more tokens (#16995)
add 209971a62e [DLIGHT][GPU] Improved gemv outer fallback schedule (#16973)
add 3b976585c7 [DLIGHT][GPU] Enhance opencl thread limit for schedules
(#16972)
add 2e56421dda [DLight] Update Adreno GEMV Rules (#17016)
add a5862a5c69 [SVE] Use only powers of two as possible vscale values
(#17001)
No new revisions were added by this update.
Summary of changes:
python/tvm/dlight/gpu/gemv.py | 41 ++++++++---
python/tvm/dlight/gpu/general_reduction.py | 3 +
python/tvm/dlight/gpu/rmsnorm.py | 2 +
python/tvm/dlight/gpu/transpose.py | 4 +
python/tvm/dlight/gpu/utils.py | 2 +
src/arith/const_int_bound.cc | 4 +-
src/arith/scalable_expression.h | 3 +-
src/target/llvm/codegen_aarch64.cc | 6 +-
tests/python/dlight/test_gpu_gemv.py | 113 ++++++++++++++++-------------
9 files changed, 111 insertions(+), 67 deletions(-)
