[SystemZ] Copy Access registers and CC with the correct register class.

On SystemZ there are a set of "access registers" that can be copied in and
out of 32-bit GPRs with special instructions. These instructions can only
perform the copy using low 32-bit parts of the 64-bit GPRs. However, the
default register class for 32-bit integers is GRX32, which also contains the
high 32-bit part registers.

In order to never end up with a case of such a COPY into a high reg, this
patch adds a new simple pre-RA pass that selects such COPYs into target
instructions.

This pass also handles COPYs from CC (Condition Code register), and COPYs to
CC can now also be emitted from a high reg in copyPhysReg().

Fixes: https://bugs.llvm.org/show_bug.cgi?id=44254

Review: Ulrich Weigand.

Differential Revision: https://reviews.llvm.org/D75014

llvm/lib/Target/SystemZ/CMakeLists.txt

@@ -16,6 +16,7 @@ add_llvm_target(SystemZCodeGen
   SystemZAsmPrinter.cpp
   SystemZCallingConv.cpp
   SystemZConstantPoolValue.cpp
+  SystemZCopyPhysRegs.cpp
   SystemZElimCompare.cpp
   SystemZFrameLowering.cpp
   SystemZHazardRecognizer.cpp

llvm/lib/Target/SystemZ/SystemZ.h

@@ -193,6 +193,7 @@ FunctionPass *createSystemZElimComparePass(SystemZTargetMachine &TM);
 FunctionPass *createSystemZShortenInstPass(SystemZTargetMachine &TM);
 FunctionPass *createSystemZLongBranchPass(SystemZTargetMachine &TM);
 FunctionPass *createSystemZLDCleanupPass(SystemZTargetMachine &TM);
+FunctionPass *createSystemZCopyPhysRegsPass(SystemZTargetMachine &TM);
 FunctionPass *createSystemZPostRewritePass(SystemZTargetMachine &TM);
 FunctionPass *createSystemZTDCPass();
 } // end namespace llvm

llvm/lib/Target/SystemZ/SystemZCopyPhysRegs.cpp

@@ -0,0 +1,120 @@
+//===---------- SystemZPhysRegCopy.cpp - Handle phys reg copies -----------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This pass makes sure that a COPY of a physical register will be
+// implementable after register allocation in copyPhysReg() (this could be
+// done in EmitInstrWithCustomInserter() instead if COPY instructions would
+// be passed to it).
+//
+//===----------------------------------------------------------------------===//
+
+#include "SystemZMachineFunctionInfo.h"
+#include "SystemZTargetMachine.h"
+#include "llvm/CodeGen/MachineDominators.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/TargetInstrInfo.h"
+#include "llvm/CodeGen/TargetRegisterInfo.h"
+#include "llvm/Target/TargetMachine.h"
+
+using namespace llvm;
+
+#define SYSTEMZ_COPYPHYSREGS_NAME "SystemZ Copy Physregs"
+
+namespace llvm {
+  void initializeSystemZCopyPhysRegsPass(PassRegistry&);
+}
+
+namespace {
+
+class SystemZCopyPhysRegs : public MachineFunctionPass {
+public:
+  static char ID;
+  SystemZCopyPhysRegs()
+    : MachineFunctionPass(ID), TII(nullptr), MRI(nullptr) {
+    initializeSystemZCopyPhysRegsPass(*PassRegistry::getPassRegistry());
+  }
+
+  StringRef getPassName() const override { return SYSTEMZ_COPYPHYSREGS_NAME; }
+
+  bool runOnMachineFunction(MachineFunction &MF) override;
+  void getAnalysisUsage(AnalysisUsage &AU) const override;
+
+private:
+
+  bool visitMBB(MachineBasicBlock &MBB);
+
+  const SystemZInstrInfo *TII;
+  MachineRegisterInfo *MRI;
+};
+
+char SystemZCopyPhysRegs::ID = 0;
+
+} // end anonymous namespace
+
+INITIALIZE_PASS(SystemZCopyPhysRegs, "systemz-copy-physregs",
+                SYSTEMZ_COPYPHYSREGS_NAME, false, false)
+
+FunctionPass *llvm::createSystemZCopyPhysRegsPass(SystemZTargetMachine &TM) {
+  return new SystemZCopyPhysRegs();
+}
+
+void SystemZCopyPhysRegs::getAnalysisUsage(AnalysisUsage &AU) const {
+  AU.setPreservesCFG();
+  MachineFunctionPass::getAnalysisUsage(AU);
+}
+
+bool SystemZCopyPhysRegs::visitMBB(MachineBasicBlock &MBB) {
+  bool Modified = false;
+
+  // Certain special registers can only be copied from a subset of the
+  // default register class of the type. It is therefore necessary to create
+  // the target copy instructions before regalloc instead of in copyPhysReg().
+  for (MachineBasicBlock::iterator MBBI = MBB.begin(), E = MBB.end();
+       MBBI != E; ) {
+    MachineInstr *MI = &*MBBI++;
+    if (!MI->isCopy())
+      continue;
+
+    DebugLoc DL = MI->getDebugLoc();
+    Register SrcReg = MI->getOperand(1).getReg();
+    Register DstReg = MI->getOperand(0).getReg();
+    if (DstReg.isVirtual() &&
+        (SrcReg == SystemZ::CC || SystemZ::AR32BitRegClass.contains(SrcReg))) {
+      Register Tmp = MRI->createVirtualRegister(&SystemZ::GR32BitRegClass);
+      if (SrcReg == SystemZ::CC)
+        BuildMI(MBB, MI, DL, TII->get(SystemZ::IPM), Tmp);
+      else
+        BuildMI(MBB, MI, DL, TII->get(SystemZ::EAR), Tmp).addReg(SrcReg);
+      MI->getOperand(1).setReg(Tmp);
+      Modified = true;
+    }
+    else if (SrcReg.isVirtual() &&
+             SystemZ::AR32BitRegClass.contains(DstReg)) {
+      Register Tmp = MRI->createVirtualRegister(&SystemZ::GR32BitRegClass);
+      MI->getOperand(0).setReg(Tmp);
+      BuildMI(MBB, MBBI, DL, TII->get(SystemZ::SAR), DstReg).addReg(Tmp);
+      Modified = true;
+    }
+  }
+
+  return Modified;
+}
+
+bool SystemZCopyPhysRegs::runOnMachineFunction(MachineFunction &F) {
+  TII = static_cast<const SystemZInstrInfo *>(F.getSubtarget().getInstrInfo());
+  MRI = &F.getRegInfo();
+
+  bool Modified = false;
+  for (auto &MBB : F)
+    Modified |= visitMBB(MBB);
+
+  return Modified;
+}
+

llvm/lib/Target/SystemZ/SystemZInstrInfo.cpp

@@ -820,18 +820,11 @@ void SystemZInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
     return;
   }
 
-  // Move CC value from/to a GR32.
-  if (SrcReg == SystemZ::CC) {
-    auto MIB = BuildMI(MBB, MBBI, DL, get(SystemZ::IPM), DestReg);
-    if (KillSrc) {
-      const MachineFunction *MF = MBB.getParent();
-      const TargetRegisterInfo *TRI = MF->getSubtarget().getRegisterInfo();
-      MIB->addRegisterKilled(SrcReg, TRI);
-    }
-    return;
-  }
+  // Move CC value from a GR32.
   if (DestReg == SystemZ::CC) {
-    BuildMI(MBB, MBBI, DL, get(SystemZ::TMLH))
+    unsigned Opcode =
+      SystemZ::GR32BitRegClass.contains(SrcReg) ? SystemZ::TMLH : SystemZ::TMHH;
+    BuildMI(MBB, MBBI, DL, get(Opcode))
       .addReg(SrcReg, getKillRegState(KillSrc))
       .addImm(3 << (SystemZ::IPM_CC - 16));
     return;
@@ -856,12 +849,6 @@ void SystemZInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
     Opcode = SystemZ::VLR;
   else if (SystemZ::AR32BitRegClass.contains(DestReg, SrcReg))
     Opcode = SystemZ::CPYA;
-  else if (SystemZ::AR32BitRegClass.contains(DestReg) &&
-           SystemZ::GR32BitRegClass.contains(SrcReg))
-    Opcode = SystemZ::SAR;
-  else if (SystemZ::GR32BitRegClass.contains(DestReg) &&
-           SystemZ::AR32BitRegClass.contains(SrcReg))
-    Opcode = SystemZ::EAR;
   else
     llvm_unreachable("Impossible reg-to-reg copy");
 

llvm/lib/Target/SystemZ/SystemZTargetMachine.cpp

@@ -222,6 +222,7 @@ public:
   void addIRPasses() override;
   bool addInstSelector() override;
   bool addILPOpts() override;
+  void addPreRegAlloc() override;
   void addPostRewrite() override;
   void addPostRegAlloc() override;
   void addPreSched2() override;
@@ -253,6 +254,10 @@ bool SystemZPassConfig::addILPOpts() {
   return true;
 }
 
+void SystemZPassConfig::addPreRegAlloc() {
+  addPass(createSystemZCopyPhysRegsPass(getSystemZTargetMachine()));
+}
+
 void SystemZPassConfig::addPostRewrite() {
   addPass(createSystemZPostRewritePass(getSystemZTargetMachine()));
 }

llvm/test/CodeGen/SystemZ/tls-08.ll

@@ -0,0 +1,24 @@
+; RUN: llc < %s -mcpu=z196 -mtriple=s390x-linux-gnu -O0 \
+; RUN:   -stop-before=regallocfast 2>&1 | FileCheck %s
+; RUN: llc < %s -mcpu=z196 -mtriple=s390x-linux-gnu -O3 \
+; RUN:   -stop-before=livevars 2>&1 | FileCheck %s
+;
+; Test that copies to/from access registers are handled before regalloc with
+; GR32 regs.
+
+@x = dso_local thread_local global i32 0, align 4
+define weak_odr hidden i32* @fun0() {
+; CHECK: name: fun0
+; CHECK: {{%[0-9]+}}:gr32bit = EAR $a0
+; CHECK: {{%[0-9]+}}:gr32bit = EAR $a1
+  ret i32* @x
+}
+
+define i32 @fun1() {
+; CHECK: name: fun1
+; CHECK: [[VREG0:%[0-9]+]]:gr32bit = COPY %0
+; CHECK-NEXT: $a1 = SAR [[VREG0]]
+; CHECK: {{%[0-9]+}}:gr32bit = EAR $a0
+  %val = call i32 asm "blah", "={a0}, {a1}" (i32 0)
+  ret i32 %val
+}

llvm/test/CodeGen/SystemZ/tls-09.ll

@@ -0,0 +1,37 @@
+; RUN: llc < %s -mcpu=z196 -mtriple=s390x-linux-gnu -O0
+;
+; Test that a0 and a1 are copied successfully into GR32 registers.
+
+@x = dso_local thread_local global i32 0, align 4
+define i32 @fun0(i32 signext, i32 signext, i32 signext, i32 signext, i32 signext, i32 signext, i32 signext)  {
+  %8 = alloca i32, align 4
+  %9 = alloca i32, align 4
+  %10 = alloca i32, align 4
+  %11 = alloca i32, align 4
+  %12 = alloca i32, align 4
+  %13 = alloca i32, align 4
+  %14 = alloca i32, align 4
+  %15 = load i32, i32* @x, align 4
+  store i32 %0, i32* %8, align 4
+  store i32 %1, i32* %9, align 4
+  store i32 %2, i32* %10, align 4
+  store i32 %3, i32* %11, align 4
+  store i32 %4, i32* %12, align 4
+  store i32 %5, i32* %13, align 4
+  store i32 %6, i32* %14, align 4
+  %16 = load i32, i32* %8, align 4
+  %17 = add nsw i32 %15, %16
+  %18 = load i32, i32* %9, align 4
+  %19 = add nsw i32 %17, %18
+  %20 = load i32, i32* %10, align 4
+  %21 = add nsw i32 %19, %20
+  %22 = load i32, i32* %11, align 4
+  %23 = add nsw i32 %21, %22
+  %24 = load i32, i32* %12, align 4
+  %25 = add nsw i32 %23, %24
+  %26 = load i32, i32* %13, align 4
+  %27 = add nsw i32 %25, %26
+  %28 = load i32, i32* %14, align 4
+  %29 = add nsw i32 %27, %28
+  ret i32 %29
+}

llvm/test/CodeGen/SystemZ/tls-10.mir

@@ -0,0 +1,24 @@
+# RUN: llc -mtriple=s390x-linux-gnu -mcpu=z196 -O0 -start-after=finalize-isel \
+# RUN:   -stop-before=regallocfast -o - %s | FileCheck %s
+# RUN: llc -mtriple=s390x-linux-gnu -mcpu=z196 -O3 -start-after=finalize-isel \
+# RUN:   -stop-before=livevars -o - %s | FileCheck %s
+#
+# Test that a COPY from CC gets implemented with an IPM to a GR32 reg.
+
+---
+name:            fun0
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: grx32bit }
+body:             |
+  bb.0:
+  liveins: $cc
+    ; CHECK-LABEL: name: fun0
+    ; CHECK: %1:gr32bit = IPM implicit $cc
+    ; CHECK-NEXT: %0:grx32bit = COPY %1
+    ; CHECK-NEXT: $r2l = COPY %0
+    ; CHECK-NEXT: Return implicit $r2l
+    %0:grx32bit = COPY $cc
+    $r2l = COPY %0
+    Return implicit $r2l
+...

llvm/test/CodeGen/SystemZ/tls-11.mir

@@ -0,0 +1,18 @@
+# RUN: llc -mtriple=s390x-linux-gnu -mcpu=z196 -O0 -start-before=prologepilog \
+# RUN:   -o - %s | FileCheck %s
+#
+# Test that a COPY to CC gets implemented with a tmlh or tmhh depending on
+# the source register.
+
+---
+name:            fun0
+tracksRegLiveness: true
+body:             |
+  bb.0:
+  liveins: $r3l, $r4h
+    ; CHECK-LABEL: fun0
+    ; CHECK: tmlh %r3, 12288
+    ; CHECK: tmhh %r4, 12288
+    $cc = COPY $r3l
+    $cc = COPY $r4h
+...