InstCombine: Fix a crash in Descale for multiply-by-zero

Fix a crash in `InstCombiner::Descale()` when a multiply-by-zero gets
created as an argument to a GEP partway through an iteration, causing
-instcombine to optimize the GEP before the multiply.

rdar://problem/17615671

llvm-svn: 212742

diff --git a/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp b/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
index 46e3bfc..08e2446 100644 (file)
--- a/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
@@ -1120,6 +1120,12 @@ Value *InstCombiner::Descale(Value *Val, APInt Scale, bool &NoSignedWrap) {
     return nullptr;
   }
 
+  // If Op is zero then Val = Op * Scale.
+  if (match(Op, m_Zero())) {
+    NoSignedWrap = true;
+    return Op;
+  }
+
   // We know that we can successfully descale, so from here on we can safely
   // modify the IR.  Op holds the descaled version of the deepest term in the
   // expression.  NoSignedWrap is 'true' if multiplying Op by Scale is known

diff --git a/llvm/test/Transforms/InstCombine/descale-zero.ll b/llvm/test/Transforms/InstCombine/descale-zero.ll
new file mode 100644 (file)
index 0000000..7990fdb
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/descale-zero.ll
@@ -0,0 +1,21 @@
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.10.0"
+
+define internal i8* @descale_zero() {
+entry:
+; CHECK: load i16** inttoptr (i64 48 to i16**), align 16
+; CHECK-NEXT: bitcast i16*
+; CHECK-NEXT: ret i8*
+  %i16_ptr = load i16** inttoptr (i64 48 to i16**), align 16
+  %num = load i64* inttoptr (i64 64 to i64*), align 64
+  %num_times_2 = shl i64 %num, 1
+  %num_times_2_plus_4 = add i64 %num_times_2, 4
+  %i8_ptr = bitcast i16* %i16_ptr to i8*
+  %i8_ptr_num_times_2_plus_4 = getelementptr i8* %i8_ptr, i64 %num_times_2_plus_4
+  %num_times_neg2 = mul i64 %num, -2
+  %num_times_neg2_minus_4 = add i64 %num_times_neg2, -4
+  %addr = getelementptr i8* %i8_ptr_num_times_2_plus_4, i64 %num_times_neg2_minus_4
+  ret i8* %addr
+}