[CGBuiltin] Handle unsigned mul overflow properly (PR35750)

[vsk]

Summary

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

[efriedma]

LGTM

lib/CodeGen/CGBuiltin.cpp

if (ResultInfo.Signed) {
     if (ResultInfo.Width < Op1Info.Width) {
       auto IntMax =
           llvm::APInt::getMaxValue(ResultInfo.Width).zext(Op1Info.Width);
       llvm::Value *TruncOverflow = CGF.Builder.CreateICmpUGT(
           UnsignedResult, llvm::ConstantInt::get(OpTy, IntMax));
       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");
 
   bool isVolatile =
       ResultArg->getType()->getPointeeType().isVolatileQualified();
   CGF.Builder.CreateStore(CGF.EmitToMemory(Result, ResultQTy), ResultPtr,
                           isVolatile);
   return RValue::get(Overflow);

test/CodeGen/builtins-overflow.c

 // CHECK-NEXT:  [[Signed:%.*]] = sub i32 0, [[Op1]]
 // CHECK-NEXT:  [[AbsSigned:%.*]] = select i1 [[IsNeg]], i32 [[Signed]], i32 [[Op1]]
 // CHECK-NEXT:  call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[AbsSigned]], i32 %{{.*}})
 // CHECK-NEXT:  [[UnsignedOFlow:%.*]] = extractvalue { i32, i1 } %{{.*}}, 1
 // CHECK-NEXT:  [[UnsignedResult:%.*]] = extractvalue { i32, i1 } %{{.*}}, 0
 // 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;
   if (__builtin_mul_overflow(x, y, &result))
     return LongErrorCode;
   return result;
 }
 

 long long test_mixed_sign_mulll_overflow_trunc_unsigned(long long x, unsigned long long y) {
 // CHECK: @test_mixed_sign_mulll_overflow_trunc_unsigned
 // CHECK:       call { i64, i1 } @llvm.umul.with.overflow.i64
 // CHECK:       [[NON_ZERO:%.*]] = icmp ne i64 [[UNSIGNED_RESULT:%.*]], 0
 // CHECK-NEXT:  [[UNDERFLOW:%.*]] = and i1 {{.*}}, [[NON_ZERO]]
 // 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))
     return LongLongErrorCode;
   return result;
 }
 
 long long test_mixed_sign_mul_overflow_extend_signed(int x, unsigned y) {