[ARM] Enable spilling of the hGPR class in Thumb2

[petpav01]

Summary

Note: This patch solves a similar problem as D49364. The difference is that this change addresses the problem in Thumb2 while D49364 is for Thumb1.

LLVM fails with an assertion error or generates incorrect code (if assertions are disabled) when a register of the hGPR class needs to be spilled in Thumb2.

Example:

$ cat test.c
void constraint_h(void) {
  int i;
  asm volatile("@ %0" : : "h" (i) : "r12");
}
$ clang -target arm-none-eabi -march=armv7-m -c test.c
Unsupported addressing mode!
UNREACHABLE executed at /work/llvm/lib/Target/ARM/Thumb2InstrInfo.cpp:630!
[...]

The code was compiled at -O0 and so Fast Register Allocator is used. The allocator ends up calling Thumb2InstrInfo::storeRegToStackSlot() to store r12 (hGPR) in a stack slot but the method does not know how to do it. The request then gets deferred to ARMBaseInstrInfo::storeRegToStackSlot() which incorrectly generates ARM instruction STRi12 for the store. If assertions are enabled, this gets caught later in rewriteT2FrameIndex() because the instruction uses ARMII::AddrMode_i12 and the rewrite function does not know how to handle it.

The patch generalizes code in Thumb2InstrInfo::storeRegToStackSlot() and loadRegToStackSlot() to allow the GPR class or any of its sub-classes (including hGPR) to be stored/loaded by ARM::t2STRi12/ARM::t2LDRi12.

[petpav01]

Ping.

[efriedma]

LGTM

[petpav01]

Thanks.

llvm/trunk/lib/Target/ARM/Thumb2InstrInfo.cpp

storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
   if (I != MBB.end()) DL = I->getDebugLoc();
 
   MachineFunction &MF = *MBB.getParent();
   MachineFrameInfo &MFI = MF.getFrameInfo();
   MachineMemOperand *MMO = MF.getMachineMemOperand(
       MachinePointerInfo::getFixedStack(MF, FI), MachineMemOperand::MOStore,
       MFI.getObjectSize(FI), MFI.getObjectAlignment(FI));
 
-  if (RC == &ARM::GPRRegClass   || RC == &ARM::tGPRRegClass ||
-      RC == &ARM::tcGPRRegClass || RC == &ARM::rGPRRegClass ||
-      RC == &ARM::GPRnopcRegClass) {
+  if (ARM::GPRRegClass.hasSubClassEq(RC)) {
     BuildMI(MBB, I, DL, get(ARM::t2STRi12))
         .addReg(SrcReg, getKillRegState(isKill))
         .addFrameIndex(FI)
         .addImm(0)
         .addMemOperand(MMO)
         .add(predOps(ARMCC::AL));
     return;
   }
loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
   MachineFunction &MF = *MBB.getParent();
   MachineFrameInfo &MFI = MF.getFrameInfo();
   MachineMemOperand *MMO = MF.getMachineMemOperand(
       MachinePointerInfo::getFixedStack(MF, FI), MachineMemOperand::MOLoad,
       MFI.getObjectSize(FI), MFI.getObjectAlignment(FI));
   DebugLoc DL;
   if (I != MBB.end()) DL = I->getDebugLoc();
 
-  if (RC == &ARM::GPRRegClass   || RC == &ARM::tGPRRegClass ||
-      RC == &ARM::tcGPRRegClass || RC == &ARM::rGPRRegClass ||
-      RC == &ARM::GPRnopcRegClass) {
+  if (ARM::GPRRegClass.hasSubClassEq(RC)) {
     BuildMI(MBB, I, DL, get(ARM::t2LDRi12), DestReg)
         .addFrameIndex(FI)
         .addImm(0)
         .addMemOperand(MMO)
         .add(predOps(ARMCC::AL));
     return;
   }
 

llvm/trunk/test/CodeGen/Thumb2/high-reg-spill.mir

+# RUN: llc -run-pass regallocfast %s -o - | FileCheck %s
+
+# This test examines register allocation and spilling with Fast Register
+# Allocator. The test uses inline assembler that requests an input variable to
+# be loaded in a high register but at the same time has r12 marked as clobbered.
+# The allocator initially satisfies the load request by selecting r12 but then
+# needs to spill this register when it reaches the INLINEASM instruction and
+# notices its clobber definition.
+#
+# The test checks that the compiler is able to spill a register from the hGPR
+# class in Thumb2 by inserting the t2STRi12/t2LDRi12 instructions.
+
+--- |
+  ; ModuleID = 'test.ll'
+  source_filename = "test.c"
+  target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
+  target triple = "thumbv7m-none-unknown-eabi"
+
+  define dso_local void @constraint_h() {
+  entry:
+    %i = alloca i32, align 4
+    %0 = load i32, i32* %i, align 4
+    call void asm sideeffect "@ $0", "h,~{r12}"(i32 %0)
+    ret void
+  }
+
+...
+---
+name:            constraint_h
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: hgpr }
+  - { id: 1, class: tgpr }
+stack:
+  - { id: 0, name: i, size: 4, alignment: 4, stack-id: 0, local-offset: -4 }
+body:             |
+  bb.0.entry:
+    %1:tgpr = tLDRspi %stack.0.i, 0, 14, $noreg :: (dereferenceable load 4 from %ir.i)
+    %0:hgpr = COPY %1
+    INLINEASM &"@ $0", 1, 589833, %0, 12, implicit-def early-clobber $r12
+    tBX_RET 14, $noreg
+
+...
+# CHECK: bb.0.entry:
+# CHECK-NEXT: renamable $r0 = tLDRspi %stack.0.i, 0, 14, $noreg :: (dereferenceable load 4 from %ir.i)
+# CHECK-NEXT: renamable $r12 = COPY killed renamable $r0
+# CHECK-NEXT: t2STRi12 killed $r12, %stack.1, 0, 14, $noreg :: (store 4 into %stack.1)
+# CHECK-NEXT: $r8 = t2LDRi12 %stack.1, 0, 14, $noreg :: (load 4 from %stack.1)
+# CHECK-NEXT: INLINEASM &"@ $0", 1, 589833, killed renamable $r8, 12, implicit-def early-clobber $r12
+# CHECK-NEXT: tBX_RET 14, $noreg