From 7a753d9bec529698c16475c72cf1de386b63381f Mon Sep 17 00:00:00 2001
From: Tim Northover <tnorthover@apple.com>
Date: Fri, 26 Aug 2016 17:46:06 +0000
Subject: [PATCH] GlobalISel: legalize under-width divisions.

llvm-svn: 279841
---
 .../CodeGen/GlobalISel/MachineLegalizeHelper.cpp   | 27 ++++++++++++++
 .../lib/Target/AArch64/AArch64MachineLegalizer.cpp |  8 +++--
 .../CodeGen/AArch64/GlobalISel/legalize-div.mir    | 42 ++++++++++++++++++++++
 3 files changed, 75 insertions(+), 2 deletions(-)
 create mode 100644 llvm/test/CodeGen/AArch64/GlobalISel/legalize-div.mir

diff --git a/llvm/lib/CodeGen/GlobalISel/MachineLegalizeHelper.cpp b/llvm/lib/CodeGen/GlobalISel/MachineLegalizeHelper.cpp
index 66b7166..401f29a 100644
--- a/llvm/lib/CodeGen/GlobalISel/MachineLegalizeHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/MachineLegalizeHelper.cpp
@@ -161,6 +161,33 @@ MachineLegalizeHelper::widenScalar(MachineInstr &MI, unsigned TypeIdx,
     MI.eraseFromParent();
     return Legalized;
   }
+  case TargetOpcode::G_SDIV:
+  case TargetOpcode::G_UDIV: {
+    unsigned ExtOp = MI.getOpcode() == TargetOpcode::G_SDIV
+                          ? TargetOpcode::G_SEXT
+                          : TargetOpcode::G_ZEXT;
+
+    unsigned LHSExt = MRI.createGenericVirtualRegister(WideSize);
+    MIRBuilder.buildInstr(ExtOp, {WideTy, MI.getType()})
+        .addDef(LHSExt)
+        .addUse(MI.getOperand(1).getReg());
+
+    unsigned RHSExt = MRI.createGenericVirtualRegister(WideSize);
+    MIRBuilder.buildInstr(ExtOp, {WideTy, MI.getType()})
+        .addDef(RHSExt)
+        .addUse(MI.getOperand(2).getReg());
+
+    unsigned ResExt = MRI.createGenericVirtualRegister(WideSize);
+    MIRBuilder.buildInstr(MI.getOpcode(), WideTy)
+        .addDef(ResExt)
+        .addUse(LHSExt)
+        .addUse(RHSExt);
+
+    MIRBuilder.buildTrunc({MI.getType(), WideTy}, MI.getOperand(0).getReg(),
+                          ResExt);
+    MI.eraseFromParent();
+    return Legalized;
+  }
   case TargetOpcode::G_LOAD: {
     assert(alignTo(Ty.getSizeInBits(), 8) == WideSize &&
            "illegal to increase number of bytes loaded");
diff --git a/llvm/lib/Target/AArch64/AArch64MachineLegalizer.cpp b/llvm/lib/Target/AArch64/AArch64MachineLegalizer.cpp
index 1a329361..0c9addf 100644
--- a/llvm/lib/Target/AArch64/AArch64MachineLegalizer.cpp
+++ b/llvm/lib/Target/AArch64/AArch64MachineLegalizer.cpp
@@ -36,17 +36,21 @@ AArch64MachineLegalizer::AArch64MachineLegalizer() {
   const LLT v4s32 = LLT::vector(4, 32);
   const LLT v2s64 = LLT::vector(2, 64);
 
-  for (auto BinOp : {G_ADD, G_SUB, G_MUL, G_AND, G_OR, G_XOR}) {
+  for (auto BinOp : {G_ADD, G_SUB, G_MUL, G_AND, G_OR, G_XOR, G_SHL}) {
     // These operations naturally get the right answer when used on
     // GPR32, even if the actual type is narrower.
     for (auto Ty : {s1, s8, s16, s32, s64, v2s32, v4s32, v2s64})
       setAction({BinOp, Ty}, Legal);
   }
 
-  for (auto BinOp : {G_SHL, G_LSHR, G_ASHR, G_SDIV, G_UDIV})
+  for (auto BinOp : {G_LSHR, G_ASHR, G_SDIV, G_UDIV}) {
     for (auto Ty : {s32, s64})
       setAction({BinOp, Ty}, Legal);
 
+    for (auto Ty : {s1, s8, s16})
+      setAction({BinOp, Ty}, WidenScalar);
+  }
+
   for (auto Op : { G_UADDE, G_USUBE, G_SADDO, G_SSUBO, G_SMULO, G_UMULO }) {
     for (auto Ty : { s32, s64 })
       setAction({Op, Ty}, Legal);
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/legalize-div.mir b/llvm/test/CodeGen/AArch64/GlobalISel/legalize-div.mir
new file mode 100644
index 0000000..ff8b5cb
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/legalize-div.mir
@@ -0,0 +1,42 @@
+# RUN: llc -O0 -run-pass=legalize-mir -global-isel %s -o - 2>&1 | FileCheck %s
+
+--- |
+  target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
+  target triple = "aarch64-apple-ios"
+  define void @test_div() {
+  entry:
+    ret void
+  }
+...
+
+---
+name:            test_div
+registers:
+  - { id: 0, class: _ }
+  - { id: 1, class: _ }
+  - { id: 2, class: _ }
+  - { id: 3, class: _ }
+  - { id: 4, class: _ }
+  - { id: 5, class: _ }
+body: |
+  bb.0.entry:
+    liveins: %x0, %x1, %x2, %x3
+    %0(64) = COPY %x0
+    %1(64) = COPY %x1
+    %2(32) = G_TRUNC { s8, s64 } %0
+    %3(32) = G_TRUNC { s8, s64 } %1
+
+
+    ; CHECK: [[LHS32:%[0-9]+]](32) = G_SEXT { s32, s8 } %2
+    ; CHECK: [[RHS32:%[0-9]+]](32) = G_SEXT { s32, s8 } %3
+    ; CHECK: [[QUOT32:%[0-9]+]](32) = G_SDIV s32 [[LHS32]], [[RHS32]]
+    ; CHECK: [[RES:%[0-9]+]](8) = G_TRUNC { s8, s32 } [[QUOT32]]
+    %4(8) = G_SDIV s8 %2, %3
+
+    ; CHECK: [[LHS32:%[0-9]+]](32) = G_ZEXT { s32, s8 } %2
+    ; CHECK: [[RHS32:%[0-9]+]](32) = G_ZEXT { s32, s8 } %3
+    ; CHECK: [[QUOT32:%[0-9]+]](32) = G_UDIV s32 [[LHS32]], [[RHS32]]
+    ; CHECK: [[RES:%[0-9]+]](8) = G_TRUNC { s8, s32 } [[QUOT32]]
+    %5(8) = G_UDIV s8 %2, %3
+
+...
-- 
2.7.4