[AMDGPU][GFX10] Many VOP3 opcodes do not support op_sel

llvm / llvm-project

The LLVM Project is a collection of modular and reusable compiler and toolchain technologies.

http://llvm.org

Other

28.05k stars 11.58k forks source link

[AMDGPU][GFX10] Many VOP3 opcodes do not support op_sel #54793

Open dpreobra opened 2 years ago

dpreobra commented 2 years ago

The following GFX10 opcodes do not currently support op_sel but they should:

v_ashrrev_i16
v_lshlrev_b16
v_lshrrev_b16
v_max_i16
v_max_u16
v_min_i16
v_min_u16
v_mul_lo_u16
v_add_nc_u16
v_sub_nc_u16
v_alignbit_b32
v_alignbyte_b32
v_interp_p2_f16

llvmbot commented 2 years ago

@llvm/issue-subscribers-backend-amdgpu

dpreobra commented 2 years ago

There is a similar issue with GFX9: https://github.com/llvm/llvm-project/issues/38650

yashssh commented 1 year ago

Hi @dpreobra! Are you working on this issue and #38650?

dpreobra commented 1 year ago

Hi @dpreobra! Are you working on this issue and #38650?

I’m not, feel free to take over.

Sisyph commented 1 year ago

FYI, I am working on this. Specifically in the context of true 16 bit instructions on GFX11.