From e22184940b843730b11d403bc2fb28f63f6aca4a Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Mon, 24 Apr 2017 21:08:32 +0000
Subject: [PATCH] AMDGPU: Slightly simplify prolog reserved register handling

Rely on MachineRegisterInfo's knowledge of used physical
registers.

Move flat_scratch initialization earlier, so the uses are visible
when making these decisions.

This will make it easier to add another reserved register
at the end for the stack pointer rather than handling another
special case.

llvm-svn: 301254
---
 llvm/lib/Target/AMDGPU/SIFrameLowering.cpp | 52 ++++++++++++++++--------------
 1 file changed, 27 insertions(+), 25 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/SIFrameLowering.cpp b/llvm/lib/Target/AMDGPU/SIFrameLowering.cpp
index 9c215188..e03a416 100644
--- a/llvm/lib/Target/AMDGPU/SIFrameLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIFrameLowering.cpp
@@ -101,10 +101,12 @@ unsigned SIFrameLowering::getReservedPrivateSegmentBufferReg(
   const SIRegisterInfo *TRI,
   SIMachineFunctionInfo *MFI,
   MachineFunction &MF) const {
+  MachineRegisterInfo &MRI = MF.getRegInfo();
 
   // We need to insert initialization of the scratch resource descriptor.
   unsigned ScratchRsrcReg = MFI->getScratchRSrcReg();
-  if (ScratchRsrcReg == AMDGPU::NoRegister)
+  if (ScratchRsrcReg == AMDGPU::NoRegister ||
+      !MRI.isPhysRegUsed(ScratchRsrcReg))
     return AMDGPU::NoRegister;
 
   if (ST.hasSGPRInitBug() ||
@@ -122,8 +124,6 @@ unsigned SIFrameLowering::getReservedPrivateSegmentBufferReg(
 
   // We find the resource first because it has an alignment requirement.
 
-  MachineRegisterInfo &MRI = MF.getRegInfo();
-
   unsigned NumPreloaded = (MFI->getNumPreloadedSGPRs() + 3) / 4;
   ArrayRef<MCPhysReg> AllSGPR128s = getAllSGPR128(ST, MF);
   AllSGPR128s = AllSGPR128s.slice(std::min(static_cast<unsigned>(AllSGPR128s.size()), NumPreloaded));
@@ -149,13 +149,18 @@ unsigned SIFrameLowering::getReservedPrivateSegmentWaveByteOffsetReg(
   const SIRegisterInfo *TRI,
   SIMachineFunctionInfo *MFI,
   MachineFunction &MF) const {
+  MachineRegisterInfo &MRI = MF.getRegInfo();
   unsigned ScratchWaveOffsetReg = MFI->getScratchWaveOffsetReg();
+
+  // No replacement necessary.
+  if (ScratchWaveOffsetReg == AMDGPU::NoRegister ||
+      !MRI.isPhysRegUsed(ScratchWaveOffsetReg))
+    return AMDGPU::NoRegister;
+
   if (ST.hasSGPRInitBug() ||
       ScratchWaveOffsetReg != TRI->reservedPrivateSegmentWaveByteOffsetReg(MF))
     return ScratchWaveOffsetReg;
 
-  unsigned ScratchRsrcReg = MFI->getScratchRSrcReg();
-  MachineRegisterInfo &MRI = MF.getRegInfo();
   unsigned NumPreloaded = MFI->getNumPreloadedSGPRs();
 
   ArrayRef<MCPhysReg> AllSGPRs = getAllSGPRs(ST, MF);
@@ -183,11 +188,7 @@ unsigned SIFrameLowering::getReservedPrivateSegmentWaveByteOffsetReg(
   for (MCPhysReg Reg : AllSGPRs.drop_back(13)) {
     // Pick the first unallocated SGPR. Be careful not to pick an alias of the
     // scratch descriptor, since we haven’t added its uses yet.
-    if (!MRI.isPhysRegUsed(Reg)) {
-      if (!MRI.isAllocatable(Reg) ||
-          TRI->isSubRegisterEq(ScratchRsrcReg, Reg))
-        continue;
-
+    if (!MRI.isPhysRegUsed(Reg) && MRI.isAllocatable(Reg)) {
       MRI.replaceRegWith(ScratchWaveOffsetReg, Reg);
       MFI->setScratchWaveOffsetReg(Reg);
       return Reg;
@@ -220,18 +221,6 @@ void SIFrameLowering::emitPrologue(MachineFunction &MF,
   const SIRegisterInfo *TRI = &TII->getRegisterInfo();
   MachineRegisterInfo &MRI = MF.getRegInfo();
 
-  unsigned ScratchRsrcReg
-    = getReservedPrivateSegmentBufferReg(ST, TII, TRI, MFI, MF);
-  unsigned ScratchWaveOffsetReg
-    = getReservedPrivateSegmentWaveByteOffsetReg(ST, TII, TRI, MFI, MF);
-
-  if (ScratchRsrcReg == AMDGPU::NoRegister) {
-    assert(ScratchWaveOffsetReg == AMDGPU::NoRegister);
-    return;
-  }
-
-  assert(!TRI->isSubRegister(ScratchRsrcReg, ScratchWaveOffsetReg));
-
   // We need to do the replacement of the private segment buffer and wave offset
   // register even if there are no stack objects. There could be stores to undef
   // or a constant without an associated object.
@@ -244,19 +233,32 @@ void SIFrameLowering::emitPrologue(MachineFunction &MF,
   if (MF.getFrameInfo().hasStackObjects() && MFI->hasFlatScratchInit())
     emitFlatScratchInit(ST, MF, MBB);
 
+  unsigned ScratchRsrcReg
+    = getReservedPrivateSegmentBufferReg(ST, TII, TRI, MFI, MF);
+  unsigned ScratchWaveOffsetReg
+    = getReservedPrivateSegmentWaveByteOffsetReg(ST, TII, TRI, MFI, MF);
+
+  // It's possible to have uses of only ScratchWaveOffsetReg without
+  // ScratchRsrcReg if it's only used for the initialization of flat_scratch,
+  // but the inverse is not true.
+  if (ScratchWaveOffsetReg == AMDGPU::NoRegister) {
+    assert(ScratchRsrcReg == AMDGPU::NoRegister);
+    return;
+  }
+
   // We need to insert initialization of the scratch resource descriptor.
   unsigned PreloadedScratchWaveOffsetReg = TRI->getPreloadedValue(
     MF, SIRegisterInfo::PRIVATE_SEGMENT_WAVE_BYTE_OFFSET);
 
-
   unsigned PreloadedPrivateBufferReg = AMDGPU::NoRegister;
   if (ST.isAmdCodeObjectV2(MF) || ST.isMesaGfxShader(MF)) {
     PreloadedPrivateBufferReg = TRI->getPreloadedValue(
       MF, SIRegisterInfo::PRIVATE_SEGMENT_BUFFER);
   }
 
-  bool OffsetRegUsed = !MRI.use_empty(ScratchWaveOffsetReg);
-  bool ResourceRegUsed = !MRI.use_empty(ScratchRsrcReg);
+  bool OffsetRegUsed = MRI.isPhysRegUsed(ScratchWaveOffsetReg);
+  bool ResourceRegUsed = ScratchRsrcReg != AMDGPU::NoRegister &&
+                         MRI.isPhysRegUsed(ScratchRsrcReg);
 
   // We added live-ins during argument lowering, but since they were not used
   // they were deleted. We're adding the uses now, so add them back.
-- 
2.7.4