[PATCH] D156103: [AMDGPU][NFC] Add True16 operand definitions.

[Joe Nash via Phabricator via llvm-commits]

Joe_Nash added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/SIRegisterInfo.td:1237
+
+def VSrcT_b16 : RegOrB16T <"VS_16", "OPERAND_REG_IMM"> {
+  let DecoderMethod = "decodeOperand_VSrcT16";
----------------
Please add a comment saying when each of these 4 versions shall be used. I think the intention is:
VSrc_b16 - the current and temporary future default used case for VOP3
VSrcT_b16 - true16 VOP3
VSrcT_b16_Lo128 - true16 VOP1/2/C
VSrcFake16_b16_Lo128 - fake16 VOP1/2/C - the current and temporary future default used case for VOP1/2/C



================
Comment at: llvm/lib/Target/AMDGPU/VOP2Instructions.td:381
 def VOP_MADAK_F16 : VOP_MADAK <f16>;
 def VOP_MADAK_F16_t16 : VOP_MADAK <f16> {
   let IsTrue16 = 1;
----------------
Shall this be renamed to include fake16? I guess a distinct true16 version will be added at a later point.


================
Comment at: llvm/lib/Target/AMDGPU/VOP2Instructions.td:406
 def VOP_MADMK_F16 : VOP_MADMK <f16>;
 def VOP_MADMK_F16_t16 : VOP_MADMK <f16> {
   let IsTrue16 = 1;
----------------
Shall this be renamed to include fake16? I guess a distinct true16 version will be added at a later point.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D156103/new/

https://reviews.llvm.org/D156103