[llvm] [InstCombine] Add combines/simplifications for `llvm.ptrmask` (PR #67166)

[via llvm-commits]

================
@@ -898,6 +910,53 @@ Value *InstCombinerImpl::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
         }
         break;
       }
+      case Intrinsic::ptrmask: {
+        // Fail loudly in case this is ever changed.
+        // Likely not much needs to be changed here to support vector types.
+        assert(!I->getOperand(0)->getType()->isVectorTy() &&
+               !I->getOperand(1)->getType()->isVectorTy() &&
+               "These simplifications where written at a time when ptrmask did "
+               "not support vector types and may not work for vectors");
+
+        unsigned MaskWidth = I->getOperand(1)->getType()->getScalarSizeInBits();
+        RHSKnown = KnownBits(MaskWidth);
+        // If either the LHS or the RHS are Zero, the result is zero.
+        if (SimplifyDemandedBits(I, 0, DemandedMask, LHSKnown, Depth + 1) ||
+            SimplifyDemandedBits(
+                I, 1, (DemandedMask & ~LHSKnown.Zero).zextOrTrunc(MaskWidth),
+                RHSKnown, Depth + 1))
+          return I;
+
+        RHSKnown = RHSKnown.zextOrTrunc(BitWidth);
+        assert(!RHSKnown.hasConflict() && "Bits known to be one AND zero?");
+        assert(!LHSKnown.hasConflict() && "Bits known to be one AND zero?");
+
+        Known = LHSKnown & RHSKnown;
+        KnownBitsComputed = DemandedMask.isAllOnes();
+
+        // If the client is only demanding bits we know to be zero, return
+        // `llvm.ptrmask(p, 0)`. We can't return `null` here due to pointer
+        // provenance, but making the mask zero will be easily optimizable in
+        // the backend.
+        if (DemandedMask.isSubsetOf(Known.Zero))
+          return replaceOperand(
+              *I, 1, Constant::getNullValue(I->getOperand(1)->getType()));
----------------
goldsteinn wrote:

Added proper tests.

https://github.com/llvm/llvm-project/pull/67166