[Attributor] Remove function pointer restriction for AAAlign

[nikic]

Summary

This check is not compatible with opaque pointers. We can avoid it by adjusting the getPointerAlignment() implementation to avoid creating unnecessary ptrtoint expressions for bitcasted pointers. The code already uses OnlyIfReduced to not create an expression if it does not simplify, and this makes sure that folding a bitcast + ptrtoint into a ptrtoint doesn't count as a simplification.

[jdoerfert]

Removing this "hack" is a good idea. From the TODO I'd assumed we simply ignore function pointer i2p uses that have no uses themselves. But this seems to work as well I guess, or at least there is no test (yet) to expose it doesn't. LG

llvm/lib/IR/Value.cpp

if (!Alignment && Call->getCalledFunction())
       Alignment = Call->getCalledFunction()->getAttributes().getRetAlignment();
     return Alignment.valueOrOne();
   } else if (const LoadInst *LI = dyn_cast<LoadInst>(this)) {
     if (MDNode *MD = LI->getMetadata(LLVMContext::MD_align)) {
       ConstantInt *CI = mdconst::extract<ConstantInt>(MD->getOperand(0));
       return Align(CI->getLimitedValue());
     }
   } else if (auto *CstPtr = dyn_cast<Constant>(this)) {
+    // Strip pointer casts to avoid creating unnecessary ptrtoint expression
+    // if the only "reduction" is combining a bitcast + ptrtoint.
+    CstPtr = CstPtr->stripPointerCasts();
     if (auto *CstInt = dyn_cast_or_null<ConstantInt>(ConstantExpr::getPtrToInt(
             const_cast<Constant *>(CstPtr), DL.getIntPtrType(getType()),
             /*OnlyIfReduced=*/true))) {
       size_t TrailingZeros = CstInt->getValue().countTrailingZeros();
       // While the actual alignment may be large, elsewhere we have
       // an arbitrary upper alignmet limit, so let's clamp to it.
       return Align(TrailingZeros < Value::MaxAlignmentExponent
                        ? uint64_t(1) << TrailingZeros

llvm/lib/Transforms/IPO/AttributorAttributes.cpp

struct AAAlignImpl : AAAlign {
   /// See AbstractAttribute::initialize(...).
   void initialize(Attributor &A) override {
     SmallVector<Attribute, 4> Attrs;
     getAttrs({Attribute::Alignment}, Attrs);
     for (const Attribute &Attr : Attrs)
       takeKnownMaximum(Attr.getValueAsInt());
 
     Value &V = getAssociatedValue();
-    // TODO: This is a HACK to avoid getPointerAlignment to introduce a ptr2int
-    //       use of the function pointer. This was caused by D73131. We want to
-    //       avoid this for function pointers especially because we iterate
-    //       their uses and int2ptr is not handled. It is not a correctness
-    //       problem though!
-    if (!V.getType()->getPointerElementType()->isFunctionTy())
-      takeKnownMaximum(V.getPointerAlignment(A.getDataLayout()).value());
 
     if (getIRPosition().isFnInterfaceKind() &&
         (!getAnchorScope() ||
          !A.isFunctionIPOAmendable(*getAssociatedFunction()))) {
       indicatePessimisticFixpoint();
       return;
     }
 

llvm/test/Transforms/OpenMP/spmdization.ll

user_code.entry:                                  ; preds = %entry
   call void @__kmpc_parallel_51(%struct.ident_t* @1, i32 %1, i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__9 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__9_wrapper to i8*), i8** %5, i64 0)
   call void @__kmpc_target_deinit(%struct.ident_t* @1, i8 1, i1 true)
   br label %common.ret
 }
 
 ; Function Attrs: alwaysinline convergent nounwind
 define internal void @.omp_outlined.(i32 %.global_tid., i32* noalias %.part_id., i8* noalias %.privates., void (i8*, ...)* noalias %.copy_fn., i8* %.task_t., %struct.anon* noalias %__context) #9 {
 ; AMDGPU-LABEL: define {{[^@]+}}@.omp_outlined.
-; AMDGPU-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
+; AMDGPU-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone align 4294967296 [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
 ; AMDGPU-NEXT:  entry:
 ; AMDGPU-NEXT:    call void @spmd_amenable() #[[ATTR5]]
 ; AMDGPU-NEXT:    ret void
 ;
 ; NVPTX-LABEL: define {{[^@]+}}@.omp_outlined.
-; NVPTX-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
+; NVPTX-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone align 4294967296 [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
 ; NVPTX-NEXT:  entry:
 ; NVPTX-NEXT:    call void @spmd_amenable() #[[ATTR5]]
 ; NVPTX-NEXT:    ret void
 ;
 ; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@.omp_outlined.
-; AMDGPU-DISABLED-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone align 4294967296 [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
 ; AMDGPU-DISABLED-NEXT:  entry:
 ; AMDGPU-DISABLED-NEXT:    call void @spmd_amenable() #[[ATTR5]]
 ; AMDGPU-DISABLED-NEXT:    ret void
 ;
 ; NVPTX-DISABLED-LABEL: define {{[^@]+}}@.omp_outlined.
-; NVPTX-DISABLED-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone align 4294967296 [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
 ; NVPTX-DISABLED-NEXT:  entry:
 ; NVPTX-DISABLED-NEXT:    call void @spmd_amenable() #[[ATTR5]]
 ; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   call void @spmd_amenable() #10
   ret void
 }