[llvm] r212742 - InstCombine: Fix a crash in Descale for multiply-by-zero

[Duncan P. N. Exon Smith]

Author: dexonsmith
Date: Thu Jul 10 12:13:27 2014
New Revision: 212742

URL: http://llvm.org/viewvc/llvm-project?rev=212742&view=rev
Log:
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

Added:
    llvm/trunk/test/Transforms/InstCombine/descale-zero.ll
Modified:
    llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp

Modified: llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp?rev=212742&r1=212741&r2=212742&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp (original)
+++ llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp Thu Jul 10 12:13:27 2014
@@ -1120,6 +1120,12 @@ Value *InstCombiner::Descale(Value *Val,
     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

Added: llvm/trunk/test/Transforms/InstCombine/descale-zero.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/descale-zero.ll?rev=212742&view=auto
==============================================================================
--- llvm/trunk/test/Transforms/InstCombine/descale-zero.ll (added)
+++ llvm/trunk/test/Transforms/InstCombine/descale-zero.ll Thu Jul 10 12:13:27 2014
@@ -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
+}