[InstCombine] ptrmask of gep for dynamic pointer aligment

Author: JonChesterfield

llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp

@@ -2040,6 +2040,40 @@ Instruction *InstCombinerImpl::visitCallInst(CallInst &CI) {
       Changed = true;
     }
 
+    // Combine:
+    // (ptrmask (getelementptr i8, ptr p, imm i), imm mask)
+    //   -> (ptrmask (getelementptr i8, ptr p, imm (i & mask)), imm mask)
+    // where only the low bits known to be zero in the pointer are changed
+    uint64_t GEPIndex;
+    uint64_t PtrMaskImmediate;
+    if (match(&CI, m_Intrinsic<Intrinsic::ptrmask>(
+                       m_GEP(m_Value(InnerPtr), m_ConstantInt(GEPIndex)),
+                       m_ConstantInt(PtrMaskImmediate)))) {
+      auto *GEP = cast<GetElementPtrInst>(II->getArgOperand(0));
+
+      if (GEP->getSourceElementType() == Type::getInt8Ty(CI.getContext())) {
+        unsigned Log2Align = llvm::Log2(InnerPtr->getPointerAlignment(DL));
+        uint64_t PointerAlignBits = (uint64_t(1) << Log2Align) - 1;
+
+        uint64_t HighBitsGEPIndex = GEPIndex & ~PointerAlignBits;
+        uint64_t MaskedLowBitsGEPIndex =
+            GEPIndex & PointerAlignBits & PtrMaskImmediate;
+
+        uint64_t MaskedGEPIndex = HighBitsGEPIndex | MaskedLowBitsGEPIndex;
+
+        if (MaskedGEPIndex != GEPIndex) {
+          Type *I64 = Type::getInt64Ty(CI.getContext());
+          auto MaskedGEP =
+              Builder.CreateGEP(GEP->getSourceElementType(), InnerPtr,
+                                ConstantInt::get(I64, MaskedGEPIndex),
+                                GEP->getName(), GEP->isInBounds());
+
+          replaceOperand(CI, 0, MaskedGEP);
+          Changed = true;
+        }
+      }
+    }
+
     // See if we can deduce non-null.
     if (!CI.hasRetAttr(Attribute::NonNull) &&
         (Known.isNonZero() ||

llvm/test/Transforms/InstCombine/ptrmask.ll

@@ -80,12 +80,12 @@ define ptr addrspace(1) @ptrmask_combine_consecutive_preserve_attrs_todo2(ptr ad
 define ptr @ptrmask_combine_add_nonnull(ptr %p) {
 ; CHECK-LABEL: define ptr @ptrmask_combine_add_nonnull
 ; CHECK-SAME: (ptr [[P:%.*]]) {
-; CHECK-NEXT:    [[PM0:%.*]] = call align 64 ptr @llvm.ptrmask.p0.i64(ptr [[P]], i64 -64)
-; CHECK-NEXT:    [[PGEP:%.*]] = getelementptr i8, ptr [[PM0]], i64 33
-; CHECK-NEXT:    [[R:%.*]] = call nonnull align 32 ptr @llvm.ptrmask.p0.i64(ptr [[PGEP]], i64 -32)
+; CHECK-NEXT:    [[PM0:%.*]] = call align 4 ptr @llvm.ptrmask.p0.i64(ptr [[P]], i64 -60)
+; CHECK-NEXT:    [[PGEP1:%.*]] = getelementptr i8, ptr [[PM0]], i64 32
+; CHECK-NEXT:    [[R:%.*]] = call nonnull align 32 ptr @llvm.ptrmask.p0.i64(ptr [[PGEP1]], i64 -32)
 ; CHECK-NEXT:    ret ptr [[R]]
 ;
-  %pm0 = call ptr @llvm.ptrmask.p0.i64(ptr %p, i64 -64)
+  %pm0 = call ptr @llvm.ptrmask.p0.i64(ptr %p, i64 -60)
   %pgep = getelementptr i8, ptr %pm0, i64 33
   %r = call ptr @llvm.ptrmask.p0.i64(ptr %pgep, i64 -16)
   ret ptr %r
@@ -287,6 +287,90 @@ define ptr addrspace(1) @ptrmask_maintain_provenance_i32(ptr addrspace(1) %p0) {
   ret ptr addrspace(1) %r
 }
 
+define ptr @ptrmask_is_nop0(ptr align 8 %p) {
+; CHECK-LABEL: define ptr @ptrmask_is_nop0
+; CHECK-SAME: (ptr align 8 [[P:%.*]]) {
+; CHECK-NEXT:    ret ptr [[P]]
+;
+  %pm = call ptr @llvm.ptrmask.p0.i64(ptr %p, i64 -8)
+  ret ptr %pm
+}
+
+define ptr @ptrmask_is_nop1(ptr align 8 %p) {
+; CHECK-LABEL: define ptr @ptrmask_is_nop1
+; CHECK-SAME: (ptr align 8 [[P:%.*]]) {
+; CHECK-NEXT:    ret ptr [[P]]
+;
+  %pm = call ptr @llvm.ptrmask.p0.i64(ptr %p, i64 -4)
+  ret ptr %pm
+}
+
+define ptr @ptrmask_to_modified_gep0(ptr align 8 %p) {
+; CHECK-LABEL: define ptr @ptrmask_to_modified_gep0
+; CHECK-SAME: (ptr align 8 [[P:%.*]]) {
+; CHECK-NEXT:    [[PM:%.*]] = call align 16 ptr @llvm.ptrmask.p0.i64(ptr [[P]], i64 -16)
+; CHECK-NEXT:    ret ptr [[PM]]
+;
+  %gep = getelementptr i8, ptr %p, i32 5
+  %pm = call ptr @llvm.ptrmask.p0.i64(ptr %gep, i64 -16)
+  ret ptr %pm
+}
+
+define ptr @ptrmask_to_modified_gep1(ptr align 8 %p) {
+; CHECK-LABEL: define ptr @ptrmask_to_modified_gep1
+; CHECK-SAME: (ptr align 8 [[P:%.*]]) {
+; CHECK-NEXT:    ret ptr [[P]]
+;
+  %gep = getelementptr i8, ptr %p, i32 6
+  %pm = call ptr @llvm.ptrmask.p0.i64(ptr %gep, i64 -8)
+  ret ptr %pm
+}
+
+define ptr @ptrmask_to_modified_gep2(ptr align 16 %p) {
+; CHECK-LABEL: define ptr @ptrmask_to_modified_gep2
+; CHECK-SAME: (ptr align 16 [[P:%.*]]) {
+; CHECK-NEXT:    [[GEP1:%.*]] = getelementptr i8, ptr [[P]], i64 12
+; CHECK-NEXT:    ret ptr [[GEP1]]
+;
+  %gep = getelementptr i8, ptr %p, i32 15
+  %pm = call ptr @llvm.ptrmask.p0.i64(ptr %gep, i64 -4)
+  ret ptr %pm
+}
+
+define ptr @ptrmask_to_modified_gep4(ptr align 8 %p) {
+; CHECK-LABEL: define ptr @ptrmask_to_modified_gep4
+; CHECK-SAME: (ptr align 8 [[P:%.*]]) {
+; CHECK-NEXT:    [[GEP1:%.*]] = getelementptr i8, ptr [[P]], i64 24
+; CHECK-NEXT:    [[PM:%.*]] = call align 16 ptr @llvm.ptrmask.p0.i64(ptr [[GEP1]], i64 -16)
+; CHECK-NEXT:    ret ptr [[PM]]
+;
+  %gep = getelementptr i8, ptr %p, i32 29
+  %pm = call ptr @llvm.ptrmask.p0.i64(ptr %gep, i64 -16)
+  ret ptr %pm
+}
+
+define ptr @ptrmask_to_modified_gep5(ptr align 8 %p) {
+; CHECK-LABEL: define ptr @ptrmask_to_modified_gep5
+; CHECK-SAME: (ptr align 8 [[P:%.*]]) {
+; CHECK-NEXT:    [[GEP1:%.*]] = getelementptr i8, ptr [[P]], i64 24
+; CHECK-NEXT:    ret ptr [[GEP1]]
+;
+  %gep = getelementptr i8, ptr %p, i32 30
+  %pm = call ptr @llvm.ptrmask.p0.i64(ptr %gep, i64 -8)
+  ret ptr %pm
+}
+
+define ptr @ptrmask_to_modified_gep6(ptr align 16 %p) {
+; CHECK-LABEL: define ptr @ptrmask_to_modified_gep6
+; CHECK-SAME: (ptr align 16 [[P:%.*]]) {
+; CHECK-NEXT:    [[GEP1:%.*]] = getelementptr i8, ptr [[P]], i64 28
+; CHECK-NEXT:    ret ptr [[GEP1]]
+;
+  %gep = getelementptr i8, ptr %p, i32 31
+  %pm = call ptr @llvm.ptrmask.p0.i64(ptr %gep, i64 -4)
+  ret ptr %pm
+}
+
 define ptr @ptrmask_is_useless0(i64 %i, i64 %m) {
 ; CHECK-LABEL: define ptr @ptrmask_is_useless0
 ; CHECK-SAME: (i64 [[I:%.*]], i64 [[M:%.*]]) {