[IR][VP] Remove IntrArgMemOnly from vp.gather/scatter.

[craig.topper]

Summary

IntrArgMemOnly is only valid for intrinsics that use a scalar
pointer argument. These intrinsics use a vector of pointer.

Alias analysis will try to find a scalar pointer argument and
will return incorrect alias results when it doesn't find one.

[reames]

LGTM - though I see this mostly as a workaround. I think we should revise the argmemonly semantics to allow vector of pointers, but that's a broader topic. We should fix the bug in the meantime.

llvm/include/llvm/IR/Intrinsics.td

def int_vp_load  : DefaultAttrsIntrinsic<[ llvm_anyvector_ty],
                                LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
                                llvm_i32_ty],
                              [ NoCapture<ArgIndex<0>>, IntrNoSync, IntrReadMem, IntrWillReturn, IntrArgMemOnly ]>;
 
 def int_vp_gather: DefaultAttrsIntrinsic<[ llvm_anyvector_ty],
                              [ LLVMVectorOfAnyPointersToElt<0>,
                                LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
                                llvm_i32_ty],
-                             [ IntrReadMem, IntrNoSync, IntrWillReturn, IntrArgMemOnly ]>;
+                             [ IntrReadMem, IntrNoSync, IntrWillReturn]>;
 
 def int_vp_scatter: DefaultAttrsIntrinsic<[],
                              [ llvm_anyvector_ty,
                                LLVMVectorOfAnyPointersToElt<0>,
                                LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
                                llvm_i32_ty],
-                             [ IntrArgMemOnly, IntrNoSync, IntrWillReturn ]>; // TODO allow IntrNoCapture for vectors of pointers
+                             [ IntrNoSync, IntrWillReturn ]>; // TODO allow IntrNoCapture for vectors of pointers
 
 // Experimental strided memory accesses
 def int_experimental_vp_strided_store : DefaultAttrsIntrinsic<[],
                              [ llvm_anyvector_ty,
                                LLVMAnyPointerToElt<0>,
                                llvm_anyint_ty, // Stride in bytes
                                LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
                                llvm_i32_ty],

llvm/test/Transforms/GVN/vp_gather_scatter.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -gvn -S < %s | FileCheck %s
 
-; FIXME: We shouldn't delete the second vp.gather.
+; Make sure we don't delete the second vp.gather.
 define <vscale x 8 x i32> @foo(<vscale x 8 x ptr> %p, <vscale x 8 x i32> %x, <vscale x 8 x i1> %mask, i32 %vl) {
 ; CHECK-LABEL: @foo(
 ; CHECK-NEXT:    [[A:%.*]] = call <vscale x 8 x i32> @llvm.vp.gather.nxv8i32.nxv8p0(<vscale x 8 x ptr> [[P:%.*]], <vscale x 8 x i1> [[MASK:%.*]], i32 [[VL:%.*]])
 ; CHECK-NEXT:    [[B:%.*]] = call <vscale x 8 x i32> @llvm.vp.add.nxv8i32(<vscale x 8 x i32> [[A]], <vscale x 8 x i32> [[X:%.*]], <vscale x 8 x i1> [[MASK]], i32 [[VL]])
 ; CHECK-NEXT:    call void @llvm.vp.scatter.nxv8i32.nxv8p0(<vscale x 8 x i32> [[B]], <vscale x 8 x ptr> [[P]], <vscale x 8 x i1> [[MASK]], i32 [[VL]])
-; CHECK-NEXT:    ret <vscale x 8 x i32> [[A]]
+; CHECK-NEXT:    [[C:%.*]] = call <vscale x 8 x i32> @llvm.vp.gather.nxv8i32.nxv8p0(<vscale x 8 x ptr> [[P]], <vscale x 8 x i1> [[MASK]], i32 [[VL]])
+; CHECK-NEXT:    ret <vscale x 8 x i32> [[C]]
 ;
   %a = call <vscale x 8 x i32> @llvm.vp.gather.nxv8i32.nxv8p0(<vscale x 8 x ptr> %p, <vscale x 8 x i1> %mask, i32 %vl)
   %b = call <vscale x 8 x i32> @llvm.vp.add.nxv8i32(<vscale x 8 x i32> %a, <vscale x 8 x i32> %x, <vscale x 8 x i1> %mask, i32 %vl)
   call void @llvm.vp.scatter.nxv8i32.nxv8p0(<vscale x 8 x i32> %b, <vscale x 8 x ptr> %p, <vscale x 8 x i1> %mask, i32 %vl)
   %c = call <vscale x 8 x i32> @llvm.vp.gather.nxv8i32.nxv8p0(<vscale x 8 x ptr> %p, <vscale x 8 x i1> %mask, i32 %vl)
   ret <vscale x 8 x i32> %c
 }
 
 declare <vscale x 8 x i32> @llvm.vp.add.nxv8i32(<vscale x 8 x i32>, <vscale x 8 x i32>, <vscale x 8 x i1>, i32)
 declare <vscale x 8 x i32> @llvm.vp.gather.nxv8i32.nxv8p0(<vscale x 8 x ptr>, <vscale x 8 x i1>, i32)
 declare void @llvm.vp.scatter.nxv8i32.nxv8p0(<vscale x 8 x i32>, <vscale x 8 x ptr>, <vscale x 8 x i1>, i32)