[PATCH] D12560: AMDGPU: Reduce number of copies emitted

[Matt Arsenault via llvm-commits]

arsenm created this revision.
arsenm added a reviewer: tstellarAMD.
arsenm added a subscriber: llvm-commits.
Herald added a subscriber: arsenm.

Instead of always inserting a copy in case
the super register is itself a subregister,
only extract to the super reg class if this is
actually the case.
    
This shouldn't really change codegen, but
makes looking at the output of SIFixSGPRCopies
easier to read.


http://reviews.llvm.org/D12560

Files:
  lib/Target/AMDGPU/SIInstrInfo.cpp

Index: lib/Target/AMDGPU/SIInstrInfo.cpp
===================================================================
--- lib/Target/AMDGPU/SIInstrInfo.cpp
+++ lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -1628,17 +1628,21 @@
                                          unsigned SubIdx,
                                          const TargetRegisterClass *SubRC)
                                          const {
-  assert(SuperReg.isReg());
-
-  unsigned NewSuperReg = MRI.createVirtualRegister(SuperRC);
+  MachineBasicBlock *MBB = MI->getParent();
+  DebugLoc DL = MI->getDebugLoc();
   unsigned SubReg = MRI.createVirtualRegister(SubRC);
 
+  if (SuperReg.getSubReg() == AMDGPU::NoSubRegister) {
+    BuildMI(*MBB, MI, DL, get(TargetOpcode::COPY), SubReg)
+      .addReg(SuperReg.getReg(), 0, SubIdx);
+    return SubReg;
+  }
+
   // Just in case the super register is itself a sub-register, copy it to a new
   // value so we don't need to worry about merging its subreg index with the
   // SubIdx passed to this function. The register coalescer should be able to
   // eliminate this extra copy.
-  MachineBasicBlock *MBB = MI->getParent();
-  DebugLoc DL = MI->getDebugLoc();
+  unsigned NewSuperReg = MRI.createVirtualRegister(SuperRC);
 
   BuildMI(*MBB, MI, DL, get(TargetOpcode::COPY), NewSuperReg)
     .addReg(SuperReg.getReg(), 0, SuperReg.getSubReg());


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D12560.33819.patch
Type: text/x-patch
Size: 1351 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150902/47380c74/attachment.bin>