[PATCH] D54714: [AMDGPU] Add and update scalar instructions

Matt Arsenault via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Mon Nov 26 08:47:18 PST 2018


arsenm added inline comments.


================
Comment at: lib/Target/AMDGPU/SIInstrInfo.cpp:4538-4539
+    MachineInstr *Xor = nullptr;
+    unsigned Temp = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+    unsigned NewDest = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+
----------------
grahamsellers wrote:
> arsenm wrote:
> > Should probably use SReg_32_XEXEC
> I couldn't find a SReg_32_XEXEC class, only SReg_32_XEXEC_HI, which seems to only exclude EXEC_LO, or SReg_64_XEXEC, which is 64-bit. Is there any particular reason to avoid EXEC? I can't get the compiler to hit that anyway.
I meant  SReg_32_XM0 so vcc_lo/vcc_hi will be allowed, but not m0. Exec doesn't matter much. We reserve exec, so it can never be allocated.


================
Comment at: lib/Target/AMDGPU/SIInstrInfo.cpp:4614-4615
+
+  unsigned NewDest = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+  unsigned Interm = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+
----------------
SReg_32_XM0


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D54714/new/

https://reviews.llvm.org/D54714





More information about the llvm-commits mailing list