[AArch64] Fix miscompile of sdiv-by-power-of-2.

When the constant divisor was larger than 32bits, then the optimized code
generated for the AArch64 backend would emit the wrong code, because the shift
was defined as a shift of a 32bit constant '(1<<Lg2(divisor))' and we would
loose the upper 32bits.

This fixes rdar://problem/18678801.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219934 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AArch64/AArch64FastISel.cpp b/lib/Target/AArch64/AArch64FastISel.cpp
index 4626302..5972d4a 100644 (file)
--- a/lib/Target/AArch64/AArch64FastISel.cpp
+++ b/lib/Target/AArch64/AArch64FastISel.cpp
@@ -4515,9 +4515,8 @@ bool AArch64FastISel::selectSDiv(const Instruction *I) {
     return true;
   }
 
-  unsigned Pow2MinusOne = (1 << Lg2) - 1;
-  unsigned AddReg = emitAddSub_ri(/*UseAdd=*/true, VT, Src0Reg,
-                                  /*IsKill=*/false, Pow2MinusOne);
+  int64_t Pow2MinusOne = (1ULL << Lg2) - 1;
+  unsigned AddReg = emitAdd_ri_(VT, Src0Reg, /*IsKill=*/false, Pow2MinusOne);
   if (!AddReg)
     return false;
 

diff --git a/lib/Target/AArch64/AArch64ISelLowering.cpp b/lib/Target/AArch64/AArch64ISelLowering.cpp
index 5cfd664..4d51c4f 100644 (file)
--- a/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -6822,7 +6822,7 @@ AArch64TargetLowering::BuildSDIVPow2(SDNode *N, const APInt &Divisor,
   SDValue N0 = N->getOperand(0);
   unsigned Lg2 = Divisor.countTrailingZeros();
   SDValue Zero = DAG.getConstant(0, VT);
-  SDValue Pow2MinusOne = DAG.getConstant((1 << Lg2) - 1, VT);
+  SDValue Pow2MinusOne = DAG.getConstant((1ULL << Lg2) - 1, VT);
 
   // Add (N0 < 0) ? Pow2 - 1 : 0;
   SDValue CCVal;

diff --git a/test/CodeGen/AArch64/sdivpow2.ll b/test/CodeGen/AArch64/sdivpow2.ll
index 6dee06d..6c02ea9 100644 (file)
--- a/test/CodeGen/AArch64/sdivpow2.ll
+++ b/test/CodeGen/AArch64/sdivpow2.ll
@@ -1,4 +1,5 @@
-; RUN: llc -mtriple=arm64-linux-gnu -o - %s | FileCheck %s
+; RUN: llc -mtriple=arm64-linux-gnu -fast-isel=0 -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=arm64-linux-gnu -fast-isel=1 -verify-machineinstrs < %s | FileCheck %s
 
 define i32 @test1(i32 %x) {
 ; CHECK-LABEL: test1
@@ -59,3 +60,15 @@ define i64 @test6(i64 %x) {
   %div = sdiv i64 %x, 64
   ret i64 %div
 }
+
+define i64 @test7(i64 %x) {
+; CHECK-LABEL: test7
+; CHECK: orr [[REG:x[0-9]+]], xzr, #0xffffffffffff
+; CHECK: add x8, x0, [[REG]]
+; CHECK: cmp x0, #0
+; CHECK: csel x8, x8, x0, lt
+; CHECK: asr x0, x8, #48
+  %div = sdiv i64 %x, 281474976710656
+  ret i64 %div
+}
+