feat(rocm): enable native GGUF kernels on RDNA4 - #136

Draft

zihaomu wants to merge 10 commits into

FlashML-org:mainfrom

zihaomu:feat/rocm-rdna4-gguf

Commits on Aug 24, 2026

feat: add simple ROCm GPU support for RDNA3 (gfx1100-1103)

bouclem
authored and
zihaomu
committed
7caa62d
fix(rocm): fall back for single-bank fast index copy

nekomario28
authored and
zihaomu
committed
af67560
test(rocm): cover fast index copy fallback

nekomario28
authored and
zihaomu
committed
d994b24
test(rocm): use explicit fast copy module import

nekomario28
authored and
zihaomu
committed
3753764
fix(rocm): complete RDNA3 runtime path

nekomario28
authored and
zihaomu
committed
c6c3639
fix(rocm): preserve CPU MoE implementation

nekomario28
authored and
zihaomu
committed
62bb962
build(rocm): support RDNA4 discovery and modular SDKs
zihaomu
committed
c7a95f7
fix(rocm): complete modular SDK runtime support
zihaomu
committed
4541756
fix(rocm): make TVM-FFI JIT kernels portable to HIP
zihaomu
committed
89ad7ab
feat(rocm): enable GGUF kernels on RDNA4
zihaomu
committed
4959197