From 8b62d78689bfdfa72f7ded3ad0bd0de412a55831 Mon Sep 17 00:00:00 2001
From: Juergen Ributzka <juergen@apple.com>
Date: Tue, 18 Nov 2014 19:58:59 +0000
Subject: [PATCH] [FastISel][AArch64] Fix shift-immediate emission for "zero"
 shifts.

This change emits a COPY for a shift-immediate with a "zero" shift value.
This fixes PR21594 where we emitted a shift instruction with an incorrect
immediate operand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@222247 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/AArch64/AArch64FastISel.cpp  | 39 +++++++++++++++++++++----
 test/CodeGen/AArch64/fast-isel-shift.ll | 25 +++++++++++++++-
 2 files changed, 57 insertions(+), 7 deletions(-)

diff --git a/lib/Target/AArch64/AArch64FastISel.cpp b/lib/Target/AArch64/AArch64FastISel.cpp
index 4d18a4e0b0f..eaa89b22628 100644
--- a/lib/Target/AArch64/AArch64FastISel.cpp
+++ b/lib/Target/AArch64/AArch64FastISel.cpp
@@ -3899,6 +3899,17 @@ unsigned AArch64FastISel::emitLSL_ri(MVT RetVT, MVT SrcVT, unsigned Op0,
   unsigned RegSize = Is64Bit ? 64 : 32;
   unsigned DstBits = RetVT.getSizeInBits();
   unsigned SrcBits = SrcVT.getSizeInBits();
+  const TargetRegisterClass *RC =
+      Is64Bit ? &AArch64::GPR64RegClass : &AArch64::GPR32RegClass;
+
+  // Just emit a copy for "zero" shifts.
+  if (Shift == 0) {
+    unsigned ResultReg = createResultReg(RC);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+            TII.get(TargetOpcode::COPY), ResultReg)
+        .addReg(Op0, getKillRegState(Op0IsKill));
+    return ResultReg;
+  }
 
   // Don't deal with undefined shifts.
   if (Shift >= DstBits)
@@ -3937,8 +3948,6 @@ unsigned AArch64FastISel::emitLSL_ri(MVT RetVT, MVT SrcVT, unsigned Op0,
     {AArch64::UBFMWri, AArch64::UBFMXri}
   };
   unsigned Opc = OpcTable[IsZext][Is64Bit];
-  const TargetRegisterClass *RC =
-      Is64Bit ? &AArch64::GPR64RegClass : &AArch64::GPR32RegClass;
   if (SrcVT.SimpleTy <= MVT::i32 && RetVT == MVT::i64) {
     unsigned TmpReg = MRI.createVirtualRegister(RC);
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
@@ -3993,6 +4002,17 @@ unsigned AArch64FastISel::emitLSR_ri(MVT RetVT, MVT SrcVT, unsigned Op0,
   unsigned RegSize = Is64Bit ? 64 : 32;
   unsigned DstBits = RetVT.getSizeInBits();
   unsigned SrcBits = SrcVT.getSizeInBits();
+  const TargetRegisterClass *RC =
+      Is64Bit ? &AArch64::GPR64RegClass : &AArch64::GPR32RegClass;
+
+  // Just emit a copy for "zero" shifts.
+  if (Shift == 0) {
+    unsigned ResultReg = createResultReg(RC);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+            TII.get(TargetOpcode::COPY), ResultReg)
+        .addReg(Op0, getKillRegState(Op0IsKill));
+    return ResultReg;
+  }
 
   // Don't deal with undefined shifts.
   if (Shift >= DstBits)
@@ -4045,8 +4065,6 @@ unsigned AArch64FastISel::emitLSR_ri(MVT RetVT, MVT SrcVT, unsigned Op0,
     {AArch64::UBFMWri, AArch64::UBFMXri}
   };
   unsigned Opc = OpcTable[IsZExt][Is64Bit];
-  const TargetRegisterClass *RC =
-      Is64Bit ? &AArch64::GPR64RegClass : &AArch64::GPR32RegClass;
   if (SrcVT.SimpleTy <= MVT::i32 && RetVT == MVT::i64) {
     unsigned TmpReg = MRI.createVirtualRegister(RC);
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
@@ -4101,6 +4119,17 @@ unsigned AArch64FastISel::emitASR_ri(MVT RetVT, MVT SrcVT, unsigned Op0,
   unsigned RegSize = Is64Bit ? 64 : 32;
   unsigned DstBits = RetVT.getSizeInBits();
   unsigned SrcBits = SrcVT.getSizeInBits();
+  const TargetRegisterClass *RC =
+      Is64Bit ? &AArch64::GPR64RegClass : &AArch64::GPR32RegClass;
+
+  // Just emit a copy for "zero" shifts.
+  if (Shift == 0) {
+    unsigned ResultReg = createResultReg(RC);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+            TII.get(TargetOpcode::COPY), ResultReg)
+        .addReg(Op0, getKillRegState(Op0IsKill));
+    return ResultReg;
+  }
 
   // Don't deal with undefined shifts.
   if (Shift >= DstBits)
@@ -4141,8 +4170,6 @@ unsigned AArch64FastISel::emitASR_ri(MVT RetVT, MVT SrcVT, unsigned Op0,
     {AArch64::UBFMWri, AArch64::UBFMXri}
   };
   unsigned Opc = OpcTable[IsZExt][Is64Bit];
-  const TargetRegisterClass *RC =
-      Is64Bit ? &AArch64::GPR64RegClass : &AArch64::GPR32RegClass;
   if (SrcVT.SimpleTy <= MVT::i32 && RetVT == MVT::i64) {
     unsigned TmpReg = MRI.createVirtualRegister(RC);
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
diff --git a/test/CodeGen/AArch64/fast-isel-shift.ll b/test/CodeGen/AArch64/fast-isel-shift.ll
index e4a3b860d2e..fd3ee27a4b6 100644
--- a/test/CodeGen/AArch64/fast-isel-shift.ll
+++ b/test/CodeGen/AArch64/fast-isel-shift.ll
@@ -1,4 +1,4 @@
-; RUN: llc -fast-isel -fast-isel-abort -mtriple=arm64-apple-darwin -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -fast-isel -fast-isel-abort -mtriple=aarch64-apple-darwin -verify-machineinstrs < %s | FileCheck %s
 
 ; CHECK-LABEL: lsl_zext_i1_i16
 ; CHECK:       ubfiz {{w[0-9]*}}, {{w[0-9]*}}, #4, #1
@@ -404,3 +404,26 @@ define i32 @shift_test1(i8 %a) {
   ret i32 %3
 }
 
+; Test zero shifts
+
+; CHECK-LABEL: shl_zero
+; CHECK-NOT:   lsl
+define i32 @shl_zero(i32 %a) {
+  %1 = shl i32 %a, 0
+  ret i32 %1
+}
+
+; CHECK-LABEL: lshr_zero
+; CHECK-NOT:   lsr
+define i32 @lshr_zero(i32 %a) {
+  %1 = lshr i32 %a, 0
+  ret i32 %1
+}
+
+; CHECK-LABEL: ashr_zero
+; CHECK-NOT:   asr
+define i32 @ashr_zero(i32 %a) {
+  %1 = ashr i32 %a, 0
+  ret i32 %1
+}
+