Fix a bogus srem rule - a negative value srem'd by a power-of-2
can have a non-negative result; for example, -16%16 is 0. Also,
clarify the related comments. This fixes PR2670.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54767 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Analysis/ValueTracking.cpp b/lib/Analysis/ValueTracking.cpp
index e35f0d0..3a04f5e 100644
--- a/lib/Analysis/ValueTracking.cpp
+++ b/lib/Analysis/ValueTracking.cpp
@@ -369,15 +369,13 @@
         ComputeMaskedBits(I->getOperand(0), Mask2, KnownZero2, KnownOne2, TD, 
                           Depth+1);
 
-        // The sign of a remainder is equal to the sign of the first
-        // operand (zero being positive).
+        // If the sign bit of the first operand is zero, the sign bit of
+        // the result is zero. If the first operand has no one bits below
+        // the second operand's single 1 bit, its sign will be zero.
         if (KnownZero2[BitWidth-1] || ((KnownZero2 & LowBits) == LowBits))
           KnownZero2 |= ~LowBits;
-        else if (KnownOne2[BitWidth-1])
-          KnownOne2 |= ~LowBits;
 
         KnownZero |= KnownZero2 & Mask;
-        KnownOne |= KnownOne2 & Mask;
 
         assert((KnownZero & KnownOne) == 0&&"Bits known to be one AND zero?"); 
       }