Many aliases are defined incorrectly which affects identification of unsupported opcodes.
See this issue for details.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Overall looks good. Did you observe any instructions that assembled when they shouldn't? It looks like only differences in error message from the tests.
llvm/lib/Target/AMDGPU/VOP2Instructions.td | ||
---|---|---|
272 | Shouldn't it be set on inst.AsmVariantName? I believe by luck they are both Default in all uses. |
Did you observe any instructions that assembled when they shouldn't? It looks like only differences in error message from the tests.
The script I used detected some opcodes which are accepted by the assembler, but I'm not sure if they are part of GFX11 ISA:
- buffer_load_lds_* opcodes (AFAIK support of these depends on specific HW features and I cannot verify if GFX1100 has these features);
- aliases for *_nc_* opcodes like v_add_i32, v_add_u16, v_add_u32;
- buffer_wbinvl1 alias.
llvm/lib/Target/AMDGPU/VOP2Instructions.td | ||
---|---|---|
272 | Using inst is more logical of course. Thanks! |
buffer_load_lds_* opcodes (AFAIK support of these depends on specific HW features and I cannot verify if GFX1100 has these features);
These opcodes are new in GFX11, so yes gfx1100 supports them.
Shouldn't it be set on inst.AsmVariantName? I believe by luck they are both Default in all uses.