From 0774fb325eb3867f039ba1e7513ce8125d36babf Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Mon, 15 Jul 2019 19:44:07 +0000
Subject: [PATCH] AMDGPU/GlobalISel: Fix handling of sgpr (not scc bank) s1 to
 VCC

This was emitting a copy from a 32-bit register to a 64-bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@366117 91177308-0d34-0410-b5e6-96231b3b80d8
---
 .../AMDGPU/AMDGPUInstructionSelector.cpp      | 40 ++++++++++-------
 .../AMDGPU/GlobalISel/inst-select-copy.mir    | 45 +++++++++++++++----
 2 files changed, 59 insertions(+), 26 deletions(-)

diff --git a/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp b/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
index 5a73b0d1139..0d02e738b4b 100644
--- a/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
+++ b/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
@@ -86,8 +86,9 @@ bool AMDGPUInstructionSelector::isVCC(Register Reg,
   const TargetRegisterClass *RC =
       RegClassOrBank.dyn_cast<const TargetRegisterClass*>();
   if (RC) {
+    const LLT Ty = MRI.getType(Reg);
     return RC->hasSuperClassEq(TRI.getBoolRC()) &&
-           MRI.getType(Reg).getSizeInBits() == 1;
+           Ty.isValid() && Ty.getSizeInBits() == 1;
   }
 
   const RegisterBank *RB = RegClassOrBank.get<const RegisterBank *>();
@@ -95,29 +96,34 @@ bool AMDGPUInstructionSelector::isVCC(Register Reg,
 }
 
 bool AMDGPUInstructionSelector::selectCOPY(MachineInstr &I) const {
+  const DebugLoc &DL = I.getDebugLoc();
   MachineBasicBlock *BB = I.getParent();
   MachineFunction *MF = BB->getParent();
   MachineRegisterInfo &MRI = MF->getRegInfo();
   I.setDesc(TII.get(TargetOpcode::COPY));
 
-  // Special case for COPY from the scc register bank.  The scc register bank
-  // is modeled using 32-bit sgprs.
   const MachineOperand &Src = I.getOperand(1);
-  unsigned SrcReg = Src.getReg();
-  if (!TargetRegisterInfo::isPhysicalRegister(SrcReg) && isSCC(SrcReg, MRI)) {
-    unsigned DstReg = I.getOperand(0).getReg();
-
-    // Specially handle scc->vcc copies.
-    if (isVCC(DstReg, MRI)) {
-      const DebugLoc &DL = I.getDebugLoc();
-      BuildMI(*BB, &I, DL, TII.get(AMDGPU::V_CMP_NE_U32_e64), DstReg)
-        .addImm(0)
-        .addReg(SrcReg);
-      if (!MRI.getRegClassOrNull(SrcReg))
-        MRI.setRegClass(SrcReg, TRI.getConstrainedRegClassForOperand(Src, MRI));
-      I.eraseFromParent();
-      return true;
+  MachineOperand &Dst = I.getOperand(0);
+  Register DstReg = Dst.getReg();
+  Register SrcReg = Src.getReg();
+
+  if (isVCC(DstReg, MRI)) {
+    if (SrcReg == AMDGPU::SCC) {
+      const TargetRegisterClass *RC
+        = TRI.getConstrainedRegClassForOperand(Dst, MRI);
+      if (!RC)
+        return true;
+      return RBI.constrainGenericRegister(DstReg, *RC, MRI);
     }
+
+    BuildMI(*BB, &I, DL, TII.get(AMDGPU::V_CMP_NE_U32_e64), DstReg)
+      .addImm(0)
+      .addReg(SrcReg);
+
+    if (!MRI.getRegClassOrNull(SrcReg))
+      MRI.setRegClass(SrcReg, TRI.getConstrainedRegClassForOperand(Src, MRI));
+    I.eraseFromParent();
+    return true;
   }
 
   for (const MachineOperand &MO : I.operands()) {
diff --git a/test/CodeGen/AMDGPU/GlobalISel/inst-select-copy.mir b/test/CodeGen/AMDGPU/GlobalISel/inst-select-copy.mir
index 6722f1608e2..1e20e6c76eb 100644
--- a/test/CodeGen/AMDGPU/GlobalISel/inst-select-copy.mir
+++ b/test/CodeGen/AMDGPU/GlobalISel/inst-select-copy.mir
@@ -148,9 +148,6 @@ regBankSelected: true
 body: |
   bb.0:
     liveins: $sgpr0
-    ; GCN-LABEL: name: copy_sgpr_no_type
-    ; GCN: [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
-    ; GCN: S_ENDPGM 0, implicit [[COPY]]
     ; WAVE64-LABEL: name: copy_sgpr_no_type
     ; WAVE64: [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
     ; WAVE64: S_ENDPGM 0, implicit [[COPY]]
@@ -174,9 +171,6 @@ regBankSelected: true
 body: |
   bb.0:
     liveins: $vgpr0
-    ; GCN-LABEL: name: copy_vgpr_no_type
-    ; GCN: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
-    ; GCN: S_ENDPGM 0, implicit [[COPY]]
     ; WAVE64-LABEL: name: copy_vgpr_no_type
     ; WAVE64: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
     ; WAVE64: S_ENDPGM 0, implicit [[COPY]]
@@ -200,9 +194,6 @@ regBankSelected: true
 body: |
   bb.0:
     liveins: $sgpr0_sgpr1
-    ; GCN-LABEL: name: copy_maybe_vcc
-    ; GCN: [[COPY:%[0-9]+]]:sreg_64_xexec = COPY $sgpr0_sgpr1
-    ; GCN: S_ENDPGM 0, implicit [[COPY]]
     ; WAVE64-LABEL: name: copy_maybe_vcc
     ; WAVE64: [[COPY:%[0-9]+]]:sreg_64_xexec = COPY $sgpr0_sgpr1
     ; WAVE64: S_ENDPGM 0, implicit [[COPY]]
@@ -215,3 +206,39 @@ body: |
     S_ENDPGM 0, implicit %1
 
 ...
+
+---
+
+name:            copy_sgpr_s1_to_vcc
+legalized:       true
+regBankSelected: true
+
+
+body: |
+  ; WAVE64-LABEL: name: copy_sgpr_s1_to_vcc
+  ; WAVE64: bb.0:
+  ; WAVE64:   successors: %bb.1(0x80000000)
+  ; WAVE64:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+  ; WAVE64:   [[V_CMP_NE_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_NE_U32_e64 0, [[COPY]], implicit $exec
+  ; WAVE64:   $vcc = COPY [[V_CMP_NE_U32_e64_]]
+  ; WAVE64:   S_CBRANCH_VCCNZ %bb.1, implicit $vcc
+  ; WAVE64: bb.1:
+  ; WAVE32-LABEL: name: copy_sgpr_s1_to_vcc
+  ; WAVE32: bb.0:
+  ; WAVE32:   successors: %bb.1(0x80000000)
+  ; WAVE32:   $vcc_hi = IMPLICIT_DEF
+  ; WAVE32:   [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+  ; WAVE32:   [[V_CMP_NE_U32_e64_:%[0-9]+]]:sreg_32_xm0 = V_CMP_NE_U32_e64 0, [[COPY]], implicit $exec
+  ; WAVE32:   $vcc_lo = COPY [[V_CMP_NE_U32_e64_]]
+  ; WAVE32:   S_CBRANCH_VCCNZ %bb.1, implicit $vcc_lo
+  ; WAVE32: bb.1:
+  bb.0:
+    liveins: $sgpr0_sgpr1
+    %0:sgpr(s32) = COPY $sgpr0
+    %1:sgpr(s1) = G_TRUNC %0
+    %2:vcc(s1) = COPY %1
+    G_BRCOND %2, %bb.1
+
+  bb.1:
+
+...
-- 
2.40.0