- [x] DS memory operations, direct and transposed `ds_load_b32/b64/b12`, `_tr8/16` @martin-luecke #4420 and #4450 - [x] MUBUF operations (`buffer_load/store`) with tests for OOB handling, SRD decode @martin-luecke - [x] wide SMEM (b256, b512) - https://github.com/ROCm/llvm-project/pull/4306 - [ ] SetPC analysis and PC-relative SMEM loads - @ftynse - [ ] WMMA/MFMA remapping (this may benefit from an asm-to-asm test or perhaps an integration test) - @tgymnich - [ ] cross-lane VALU (`v_readfirstlane` and friends) - @tgymnich - [x] MSB support (`s_set_vgpr_msb`) - @tgymnich - #4383 - [ ] VOP3 packed ops (`v_pl_add/mul`) - @tgymnich - #4362 - [ ] VALU math ops (`v_exp/rcp/cvt/cndmask` and perhaps more) - @tgymnich - #4359 - [ ] https://github.com/ROCm/llvm-project/issues/4506
ds_load_b32/b64/b12,_tr8/16@martin-luecke [Comgr][hotswap] Raise direct LDS loads #4420 and [Comgr][transpiler] Raise transposed LDS loads #4450buffer_load/store) with tests for OOB handling, SRD decode @martin-lueckev_readfirstlaneand friends) - @tgymnichs_set_vgpr_msb) - @tgymnich - [Comgr][hotswap] Raise s_set_vgpr_msb #4383v_pl_add/mul) - @tgymnich - [Comgr][transpiler] Raise packed VOP3P add and multiply #4362v_exp/rcp/cvt/cndmaskand perhaps more) - @tgymnich - AMDGPU: Port hotswap VALU math operations #4359