[AArch64][GlobalISel] Fix an assert fail/miscompile when fp16 types are copied

to gpr register banks.

PR36345.

rdar://36478867

Differential Revision: https://reviews.llvm.org/D43310

llvm-svn: 325463

diff --git a/llvm/lib/Target/AArch64/AArch64InstructionSelector.cpp b/llvm/lib/Target/AArch64/AArch64InstructionSelector.cpp
index f1dfe00..64531d7 100644 (file)
--- a/llvm/lib/Target/AArch64/AArch64InstructionSelector.cpp
+++ b/llvm/lib/Target/AArch64/AArch64InstructionSelector.cpp
@@ -356,6 +356,31 @@ static bool selectCopy(MachineInstr &I, const TargetInstrInfo &TII,
     return false;
   }
 
+  if (!TargetRegisterInfo::isPhysicalRegister(SrcReg)) {
+    const RegClassOrRegBank &RegClassOrBank =
+      MRI.getRegClassOrRegBank(SrcReg);
+
+    const TargetRegisterClass *SrcRC =
+        RegClassOrBank.dyn_cast<const TargetRegisterClass *>();
+    const RegisterBank *RB = nullptr;
+    if (!SrcRC) {
+      RB = RegClassOrBank.get<const RegisterBank *>();
+      SrcRC = getRegClassForTypeOnBank(MRI.getType(SrcReg), *RB, RBI, true);
+    }
+    // Copies from fpr16 to gpr32 need to use SUBREG_TO_REG.
+    if (RC == &AArch64::GPR32allRegClass && SrcRC == &AArch64::FPR16RegClass) {
+      unsigned PromoteReg = MRI.createVirtualRegister(&AArch64::FPR32RegClass);
+      BuildMI(*I.getParent(), I, I.getDebugLoc(),
+              TII.get(AArch64::SUBREG_TO_REG))
+          .addDef(PromoteReg)
+          .addImm(0)
+          .addUse(SrcReg)
+          .addImm(AArch64::hsub);
+      MachineOperand &RegOp = I.getOperand(1);
+      RegOp.setReg(PromoteReg);
+    }
+  }
+
   // No need to constrain SrcReg. It will get constrained when
   // we hit another of its use or its defs.
   // Copies do not have constraints.

diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/fp16-copy-gpr.mir b/llvm/test/CodeGen/AArch64/GlobalISel/fp16-copy-gpr.mir
new file mode 100644 (file)
index 0000000..3a718cf
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/fp16-copy-gpr.mir
@@ -0,0 +1,69 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=aarch64-unknown-unknown -o - -global-isel -verify-machineinstrs -run-pass=instruction-select %s | FileCheck %s
+
+# PR36345
+--- |
+  target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+  target triple = "aarch64-arm-none-eabi"
+
+  %struct.struct2 = type { [2 x half] }
+
+  @global_arg0 = common dso_local global %struct.struct2 zeroinitializer, align 2
+
+  ; Function Attrs: noinline nounwind optnone
+  define dso_local void @c_test([2 x half], [2 x half]* %addr) {
+    store [2 x half] %0, [2 x half]* %addr, align 2
+    ret void
+  }
+...
+---
+name:            c_test
+alignment:       2
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: gpr }
+  - { id: 1, class: fpr }
+  - { id: 2, class: fpr }
+  - { id: 3, class: gpr }
+  - { id: 4, class: gpr }
+  - { id: 5, class: gpr }
+  - { id: 6, class: gpr }
+  - { id: 7, class: gpr }
+  - { id: 8, class: gpr }
+  - { id: 9, class: gpr }
+  - { id: 10, class: gpr }
+  - { id: 11, class: gpr }
+  - { id: 12, class: gpr }
+body:             |
+  bb.1 (%ir-block.1):
+    liveins: $h0, $h1, $x0
+
+    ; CHECK-LABEL: name: c_test
+    ; CHECK: liveins: $h0, $h1, $x0
+    ; CHECK: [[COPY:%[0-9]+]]:fpr16 = COPY $h0
+    ; CHECK: [[COPY1:%[0-9]+]]:fpr16 = COPY $h1
+    ; CHECK: [[DEF:%[0-9]+]]:gpr32 = IMPLICIT_DEF
+    ; CHECK: [[SUBREG_TO_REG:%[0-9]+]]:fpr32 = SUBREG_TO_REG 0, [[COPY]], %subreg.hsub
+    ; CHECK: [[COPY2:%[0-9]+]]:gpr32 = COPY [[SUBREG_TO_REG]]
+    ; CHECK: [[BFMWri:%[0-9]+]]:gpr32 = BFMWri [[DEF]], [[COPY2]], 0, 15
+    ; CHECK: [[SUBREG_TO_REG1:%[0-9]+]]:fpr32 = SUBREG_TO_REG 0, [[COPY1]], %subreg.hsub
+    ; CHECK: [[COPY3:%[0-9]+]]:gpr32 = COPY [[SUBREG_TO_REG1]]
+    ; CHECK: [[BFMWri1:%[0-9]+]]:gpr32 = BFMWri [[BFMWri]], [[COPY3]], 16, 15
+    ; CHECK: [[COPY4:%[0-9]+]]:gpr64sp = COPY $x0
+    ; CHECK: STRWui [[BFMWri1]], [[COPY4]], 0 :: (store 4 into %ir.addr, align 2)
+    ; CHECK: RET_ReallyLR
+    %1:fpr(s16) = COPY $h0
+    %2:fpr(s16) = COPY $h1
+    %3:gpr(s32) = G_IMPLICIT_DEF
+    %11:gpr(s16) = COPY %1(s16)
+    %4:gpr(s32) = G_INSERT %3, %11(s16), 0
+    %12:gpr(s16) = COPY %2(s16)
+    %5:gpr(s32) = G_INSERT %4, %12(s16), 16
+    %0:gpr(s32) = COPY %5(s32)
+    %6:gpr(p0) = COPY $x0
+    G_STORE %0(s32), %6(p0) :: (store 4 into %ir.addr, align 2)
+    RET_ReallyLR
+
+...