[InstCombine] Folding `@llvm.ptrmask` with itself

[goldstein.w.n]

Summary

@llvm.ptrmask is basically just and with a ptr operand. This is
a trivial combine to do with and (many others could also be added).

[nikic]

llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp

1917

Should capture the intrinsic operands via m_Value here.

[goldstein.w.n]

Caputer inner values directly

[nikic]

LGTM

[goldstein.w.n]

Rebase

llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp

case Intrinsic::fshr: {
     if (!isPowerOf2_32(BitWidth))
       break;
     APInt Op2Demanded = APInt::getLowBitsSet(BitWidth, Log2_32_Ceil(BitWidth));
     KnownBits Op2Known(BitWidth);
     if (SimplifyDemandedBits(II, 2, Op2Demanded, Op2Known))
       return &CI;
     break;
   }
+  case Intrinsic::ptrmask: {
+    if (match(II->getArgOperand(0), m_OneUse(m_Intrinsic<Intrinsic::ptrmask>(
+                                        m_Value(), m_Value())))) {
+      auto *InnerII = cast<IntrinsicInst>(II->getArgOperand(0));
+      if (II->getArgOperand(1)->getType() ==
+          InnerII->getArgOperand(1)->getType()) {
+        Value *NewMask =
+            Builder.CreateAnd(II->getArgOperand(1), InnerII->getArgOperand(1));
+        return replaceInstUsesWith(
+            *II, Builder.CreateIntrinsic(InnerII->getArgOperand(0)->getType(),
+                                         Intrinsic::ptrmask,
+                                         {InnerII->getArgOperand(0), NewMask}));
+      }
+    }
+    break;
+  }
   case Intrinsic::uadd_with_overflow:
   case Intrinsic::sadd_with_overflow: {
     if (Instruction *I = foldIntrinsicWithOverflowCommon(II))
       return I;
 
     // Given 2 constant operands whose sum does not overflow:
     // uaddo (X +nuw C0), C1 -> uaddo X, C0 + C1
     // saddo (X +nsw C0), C1 -> saddo X, C0 + C1

llvm/test/Transforms/InstCombine/consecutive-ptrmask.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
 ; RUN: opt < %s -passes=instcombine -S | FileCheck %s
 
 declare ptr @llvm.ptrmask(ptr, i64)
 declare ptr @llvm.ptrmask.p0.i32(ptr, i32)
 declare void @use.ptr(ptr)
 define ptr @fold_2x(ptr %p, i64 %m0, i64 %m1) {
 ; CHECK-LABEL: define ptr @fold_2x
 ; CHECK-SAME: (ptr [[P:%.*]], i64 [[M0:%.*]], i64 [[M1:%.*]]) {
-; CHECK-NEXT:    [[P0:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[P]], i64 [[M0]])
-; CHECK-NEXT:    [[P1:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[P0]], i64 [[M1]])
+; CHECK-NEXT:    [[TMP1:%.*]] = and i64 [[M1]], [[M0]]
+; CHECK-NEXT:    [[P1:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[P]], i64 [[TMP1]])
 ; CHECK-NEXT:    ret ptr [[P1]]
 ;
   %p0 = call ptr @llvm.ptrmask(ptr %p, i64 %m0)
   %p1 = call ptr @llvm.ptrmask(ptr %p0, i64 %m1)
   ret ptr %p1
 }
 
 define ptr @fold_2x_i32(ptr %p, i32 %m0, i32 %m1) {
 ; CHECK-LABEL: define ptr @fold_2x_i32
 ; CHECK-SAME: (ptr [[P:%.*]], i32 [[M0:%.*]], i32 [[M1:%.*]]) {
-; CHECK-NEXT:    [[P0:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr [[P]], i32 [[M0]])
-; CHECK-NEXT:    [[P1:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr [[P0]], i32 [[M1]])
+; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[M1]], [[M0]]
+; CHECK-NEXT:    [[P1:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr [[P]], i32 [[TMP1]])
 ; CHECK-NEXT:    ret ptr [[P1]]
 ;
   %p0 = call ptr @llvm.ptrmask.p0.i32(ptr %p, i32 %m0)
   %p1 = call ptr @llvm.ptrmask.p0.i32(ptr %p0, i32 %m1)
   ret ptr %p1
 }
 
 define ptr @fold_2x_fail_multiuse(ptr %p, i64 %m0, i64 %m1) {