PR #17156 HIP: WMMA-MMQ kernels for RDNA 4

first commit naive test to enable mmq for RDNA4

jiachengjason committed 207 days ago

adding appropriate WMMA instructions

jiachengjason committed 207 days ago

git rebase on top of master: fixing the correctness of the mat mul operations, updating layout mappings for RDNA4

jiachengjason committed 207 days ago

clean up merge conflicts

jiachengjason committed 207 days ago

add comments and code clean up

jiachengjason committed 207 days ago

PR clean up, addressed comments

jiachengjason committed 207 days ago

enable MMQ fallback on RDNA4

jiachengjason committed 207 days ago

addressed comments: add guards in load generic, separate wmma branch for use_mmq function

jiachengjason committed 207 days ago

Revert build-xcframework.sh

jiachengjason committed 207 days ago

Formating: remove trailing whitespace

jiachengjason committed 207 days ago

revert CMake files

jiachengjason committed 207 days ago

clean up after rebase: remove duplicated change, revert cmake files

jiachengjason committed 207 days ago

clean up after rebase: revert changes from build-xcframework.sh

jiachengjason committed 207 days ago

clean up: remove extra space line in mma.cuh

jiachengjason committed 207 days ago

Revert "clean up: remove extra space line in mma.cuh"

jiachengjason committed 207 days ago

llama.cpp HIP: WMMA-MMQ kernels for RDNA 4 #17156 Merged