[ARM] GlobalISel: Legalize s8 and s16 G_(S|U)DIV

We have to widen the operands to 32 bits and then we can either use
hardware division if it is available or lower to a libcall otherwise.

At the moment it is not enough to set the Legalizer action to
WidenScalar, since for libcalls it won't know what to do (it won't be
able to find what size to widen to, because it will find Libcall and not
Legal for 32 bits). To hack around this limitation, we request Custom
lowering, and as part of that we widen first and then we run another
legalizeInstrStep on the widened DIV.

llvm-svn: 301166

2 files changed, 55 insertions, 0 deletions

diff --git a/llvm/lib/Target/ARM/ARMLegalizerInfo.cpp b/llvm/lib/Target/ARM/ARMLegalizerInfo.cpp
index d1c5d964d95..9b86030fdd2 100644
--- a/llvm/lib/Target/ARM/ARMLegalizerInfo.cpp
+++ b/llvm/lib/Target/ARM/ARMLegalizerInfo.cpp
@@ -13,6 +13,8 @@
 
 #include "ARMLegalizerInfo.h"
 #include "ARMSubtarget.h"
+#include "llvm/CodeGen/GlobalISel/LegalizerHelper.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/ValueTypes.h"
 #include "llvm/IR/DerivedTypes.h"
 #include "llvm/IR/Type.h"
@@ -48,6 +50,11 @@ ARMLegalizerInfo::ARMLegalizerInfo(const ARMSubtarget &ST) {
       setAction({Op, Ty}, Legal);
 
   for (unsigned Op : {G_SDIV, G_UDIV}) {
+    for (auto Ty : {s8, s16})
+      // FIXME: We need WidenScalar here, but in the case of targets with
+      // software division we'll also need Libcall afterwards. Treat as Custom
+      // until we have better support for chaining legalization actions.
+      setAction({Op, Ty}, Custom);
     if (ST.hasDivideInARMMode())
       setAction({Op, s32}, Legal);
     else
@@ -82,3 +89,48 @@ ARMLegalizerInfo::ARMLegalizerInfo(const ARMSubtarget &ST) {
 
   computeTables();
 }
+
+bool ARMLegalizerInfo::legalizeCustom(MachineInstr &MI,
+                                      MachineRegisterInfo &MRI,
+                                      MachineIRBuilder &MIRBuilder) const {
+  using namespace TargetOpcode;
+
+  switch (MI.getOpcode()) {
+  default:
+    return false;
+  case G_SDIV:
+  case G_UDIV: {
+    LLT Ty = MRI.getType(MI.getOperand(0).getReg());
+    if (Ty != LLT::scalar(16) && Ty != LLT::scalar(8))
+      return false;
+
+    // We need to widen to 32 bits and then maybe, if the target requires,
+    // transform into a libcall.
+    LegalizerHelper Helper(MIRBuilder.getMF());
+
+    MachineInstr *NewMI = nullptr;
+    Helper.MIRBuilder.recordInsertions([&](MachineInstr *MI) {
+      // Store the new, 32-bit div instruction.
+      if (MI->getOpcode() == G_SDIV || MI->getOpcode() == G_UDIV)
+        NewMI = MI;
+    });
+
+    auto Result = Helper.widenScalar(MI, 0, LLT::scalar(32));
+    Helper.MIRBuilder.stopRecordingInsertions();
+    if (Result == LegalizerHelper::UnableToLegalize) {
+      return false;
+    }
+    assert(NewMI && "Couldn't find widened instruction");
+    assert((NewMI->getOpcode() == G_SDIV || NewMI->getOpcode() == G_UDIV) &&
+           "Unexpected widened instruction");
+    assert(MRI.getType(NewMI->getOperand(0).getReg()).getSizeInBits() == 32 &&
+           "Unexpected type for the widened instruction");
+
+    Result = Helper.legalizeInstrStep(*NewMI);
+    if (Result == LegalizerHelper::UnableToLegalize) {
+      return false;
+    }
+    return true;
+  }
+  }
+}
diff --git a/llvm/lib/Target/ARM/ARMLegalizerInfo.h b/llvm/lib/Target/ARM/ARMLegalizerInfo.h
index 0b8a608a6bd..a9bdd367737 100644
--- a/llvm/lib/Target/ARM/ARMLegalizerInfo.h
+++ b/llvm/lib/Target/ARM/ARMLegalizerInfo.h
@@ -24,6 +24,9 @@ class ARMSubtarget;
 class ARMLegalizerInfo : public LegalizerInfo {
 public:
   ARMLegalizerInfo(const ARMSubtarget &ST);
+
+  bool legalizeCustom(MachineInstr &MI, MachineRegisterInfo &MRI,
+                      MachineIRBuilder &MIRBuilder) const override;
 };
 } // End llvm namespace.
 #endif