[llvm-branch-commits] [cfe-branch] r322555 - Merging r321771:

Hans Wennborg via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Tue Jan 16 06:49:22 PST 2018


Author: hans
Date: Tue Jan 16 06:49:22 2018
New Revision: 322555

URL: http://llvm.org/viewvc/llvm-project?rev=322555&view=rev
Log:
Merging r321771:
------------------------------------------------------------------------
r321771 | vedantk | 2018-01-03 15:11:32 -0800 (Wed, 03 Jan 2018) | 21 lines

[CGBuiltin] Handle unsigned mul overflow properly (PR35750)

r320902 fixed the IRGen for some types of checked multiplications. It
did not handle unsigned overflow correctly in the case where the signed
operand is negative (PR35750).

Eli pointed out that on overflow, the result must be equal to the unique
value that is equivalent to the mathematically-correct result modulo two
raised to the k power, where k is the number of bits in the result type.

This patch fixes the specialized IRGen from r320902 accordingly.

Testing: Apart from check-clang, I modified the test harness from
r320902 to validate the results of all multiplications -- not just the
ones which don't overflow:

  https://gist.github.com/vedantk/3eb9c88f82e5c32f2e590555b4af5081

llvm.org/PR35750, rdar://34963321

Differential Revision: https://reviews.llvm.org/D41717
------------------------------------------------------------------------

Modified:
    cfe/branches/release_60/   (props changed)
    cfe/branches/release_60/lib/CodeGen/CGBuiltin.cpp
    cfe/branches/release_60/test/CodeGen/builtins-overflow.c

Propchange: cfe/branches/release_60/
------------------------------------------------------------------------------
--- svn:mergeinfo (original)
+++ svn:mergeinfo Tue Jan 16 06:49:22 2018
@@ -1,4 +1,4 @@
 /cfe/branches/type-system-rewrite:134693-134817
-/cfe/trunk:321933,322518
+/cfe/trunk:321771,321933,322518
 /cfe/trunk/test:170344
 /cfe/trunk/test/SemaTemplate:126920

Modified: cfe/branches/release_60/lib/CodeGen/CGBuiltin.cpp
URL: http://llvm.org/viewvc/llvm-project/cfe/branches/release_60/lib/CodeGen/CGBuiltin.cpp?rev=322555&r1=322554&r2=322555&view=diff
==============================================================================
--- cfe/branches/release_60/lib/CodeGen/CGBuiltin.cpp (original)
+++ cfe/branches/release_60/lib/CodeGen/CGBuiltin.cpp Tue Jan 16 06:49:22 2018
@@ -915,7 +915,11 @@ EmitCheckedMixedSignMultiply(CodeGenFunc
       Overflow = CGF.Builder.CreateOr(Overflow, TruncOverflow);
     }
 
-    Result = CGF.Builder.CreateTrunc(UnsignedResult, ResTy);
+    // Negate the product if it would be negative in infinite precision.
+    Result = CGF.Builder.CreateSelect(
+        IsNegative, CGF.Builder.CreateNeg(UnsignedResult), UnsignedResult);
+
+    Result = CGF.Builder.CreateTrunc(Result, ResTy);
   }
   assert(Overflow && Result && "Missing overflow or result");
 

Modified: cfe/branches/release_60/test/CodeGen/builtins-overflow.c
URL: http://llvm.org/viewvc/llvm-project/cfe/branches/release_60/test/CodeGen/builtins-overflow.c?rev=322555&r1=322554&r2=322555&view=diff
==============================================================================
--- cfe/branches/release_60/test/CodeGen/builtins-overflow.c (original)
+++ cfe/branches/release_60/test/CodeGen/builtins-overflow.c Tue Jan 16 06:49:22 2018
@@ -373,7 +373,9 @@ int test_mixed_sign_mull_overflow_unsign
 // CHECK-NEXT:  [[NotNull:%.*]] = icmp ne i32 [[UnsignedResult]], 0
 // CHECK-NEXT:  [[Underflow:%.*]] = and i1 [[IsNeg]], [[NotNull]]
 // CHECK-NEXT:  [[OFlow:%.*]] = or i1 [[UnsignedOFlow]], [[Underflow]]
-// CHECK-NEXT:  store i32 [[UnsignedResult]], i32* %{{.*}}, align 4
+// CHECK-NEXT:  [[NegatedResult:%.*]] = sub i32 0, [[UnsignedResult]]
+// CHECK-NEXT:  [[Result:%.*]] = select i1 [[IsNeg]], i32 [[NegatedResult]], i32 [[UnsignedResult]]
+// CHECK-NEXT:  store i32 [[Result]], i32* %{{.*}}, align 4
 // CHECK:       br i1 [[OFlow]]
 
   unsigned result;
@@ -432,7 +434,9 @@ long long test_mixed_sign_mulll_overflow
 // CHECK-NEXT:  [[OVERFLOW_PRE_TRUNC:%.*]] = or i1 {{.*}}, [[UNDERFLOW]]
 // CHECK-NEXT:  [[TRUNC_OVERFLOW:%.*]] = icmp ugt i64 [[UNSIGNED_RESULT]], 4294967295
 // CHECK-NEXT:  [[OVERFLOW:%.*]] = or i1 [[OVERFLOW_PRE_TRUNC]], [[TRUNC_OVERFLOW]]
-// CHECK-NEXT:  trunc i64 [[UNSIGNED_RESULT]] to i32
+// CHECK-NEXT:  [[NEGATED:%.*]] = sub i64 0, [[UNSIGNED_RESULT]]
+// CHECK-NEXT:  [[RESULT:%.*]] = select i1 {{.*}}, i64 [[NEGATED]], i64 [[UNSIGNED_RESULT]]
+// CHECK-NEXT:  trunc i64 [[RESULT]] to i32
 // CHECK-NEXT:  store
   unsigned result;
   if (__builtin_mul_overflow(y, x, &result))




More information about the llvm-branch-commits mailing list