[PATCH] D75580: [llvm][CodeGen][SVE] Implement IR intrinsics for gather prefetch.

[Sander de Smalen via Phabricator via llvm-commits]

sdesmalen added inline comments.


================
Comment at: llvm/include/llvm/IR/IntrinsicsAArch64.td:1263
+                  llvm_anyvector_ty, // Offsets
+                  LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>, // Predicate
+                  llvm_i32_ty // Prfop
----------------
fpetrogalli wrote:
> sdesmalen wrote:
> > I realise this only now, but having the predicate as the third operand is different from the gather loads. Can you make this to be the first operand, rather than the third, to streamline their definition with the gather loads?
> Good point. I have redefined the class as 
> 
> ```
> class SVE_gather_prf_scalar_base_vector_offset_scaled
>     : Intrinsic<[],
>                 [
>                   LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>, // Predicate
>                   llvm_ptr_ty, // Base address
>                   llvm_anyvector_ty, // Offsets
>                   llvm_i32_ty // Prfop
> 		],
>                 [IntrInaccessibleMemOrArgMemOnly, NoCapture<0>, ImmArg<3>]>;
> ```
> 
> and consequently redefined the intrinsics from, say, `declare void @llvm.aarch64.sve.gather.prfb.scaled.uxtw.nx4vi32(i8* %base, <vscale x 4 x i32> %offset, <vscale x 4 x i1> %Pg, i32 %prfop)`, to `declare void @llvm.aarch64.sve.gather.prfb.scaled.uxtw.nx4vi32(<vscale x 4 x i1> %Pg, i8* %base, <vscale x 4 x i32> %offset, i32 %prfop)`, but I get `llc` runtime failures as the following:
> 
> ```
> Wrong types for attribute: inalloca nest noalias nocapture nonnull readnone readonly signext sret zeroext byval dereferenceable(1) dereferenceable_or_null(1)
> void (<vscale x 4 x i1>, i8*, <vscale x 4 x i32>, i32)* @llvm.aarch64.sve.gather.prfb.scaled.uxtw.nxv4i32
> ```
> 
> I cannot figure out what I am doing wrong though... Am I right assuming that `llvm_anyvec_ty` is still the overloaded type that needs to be mentioned in the name of the intrinsic in IR? (`nxv4i32` for the example reported here).
The pointer is now the second argument, and I suspect you forgot to update `NoCapture<0>` to `NoCapture<1>`.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D75580/new/

https://reviews.llvm.org/D75580