R600/SI: improve vector interpolation

Prevent loading M0 multiple times.

Signed-off-by: Christian König <christian.koenig@amd.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@178023 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/R600/SIInstrInfo.cpp b/lib/Target/R600/SIInstrInfo.cpp
index de2373b11a756670e15f8f5ad51a1dff2982ef3a..b9b25b5ecfb3c8ba26beb47c707b459fe766a1b5 100644 (file)
--- a/lib/Target/R600/SIInstrInfo.cpp
+++ b/lib/Target/R600/SIInstrInfo.cpp
@@ -65,6 +65,26 @@ SIInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
   unsigned Opcode;
   const int16_t *SubIndices;
 
+  if (AMDGPU::M0 == DestReg) {
+    // Check if M0 isn't already set to this value
+    for (MachineBasicBlock::reverse_iterator E = MBB.rend(),
+      I = MachineBasicBlock::reverse_iterator(MI); I != E; ++I) {
+
+      if (!I->definesRegister(AMDGPU::M0))
+        continue;
+
+      unsigned Opc = I->getOpcode();
+      if (Opc != TargetOpcode::COPY && Opc != AMDGPU::S_MOV_B32)
+        break;
+
+      if (!I->readsRegister(SrcReg))
+        break;
+
+      // The copy isn't necessary
+      return;
+    }
+  }
+
   if (AMDGPU::SReg_32RegClass.contains(DestReg)) {
     assert(AMDGPU::SReg_32RegClass.contains(SrcReg));
     BuildMI(MBB, MI, DL, get(AMDGPU::S_MOV_B32), DestReg)