[PATCH] D101593: [AArch64][SVE] Remove index_vector node.

[Paul Walker via Phabricator via llvm-commits]

paulwalker-arm accepted this revision.
paulwalker-arm added a comment.
This revision is now accepted and ready to land.

Please can you also remove INDEX_VECTOR from the AArch64ISD enum.



================
Comment at: llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td:1371-1374
+  defm INDEX_RR : sve_int_index_rr<"index", step_vector, step_vector_oneuse, AArch64mul_p_oneuse>;
+  defm INDEX_IR : sve_int_index_ir<"index", step_vector, step_vector_oneuse, AArch64mul_p, AArch64mul_p_oneuse>;
+  defm INDEX_RI : sve_int_index_ri<"index", step_vector, step_vector_oneuse>;
+  defm INDEX_II : sve_int_index_ii<"index", step_vector, step_vector_oneuse>;
----------------
As a tidy up I'd say these multiclasses are intrinsically linked to stepvector and related nodes.  As these nodes now have a unified form I personally don't see much value with the indirection and think they can use all the operations directly within the multiclass.


================
Comment at: llvm/lib/Target/AArch64/SVEInstrFormats.td:4875
 
-  // add(mul(index_vector(0, 1), dup(Y), dup(X)) -> index_vector(X, Y).
-  def : Pat<(add (muloneuseop (nxv16i1 (AArch64ptrue 31)), (nxv16i8 (oneuseop (i32 0), (i32 1))), (nxv16i8 (AArch64dup(i32 GPR32:$Rm)))), (nxv16i8 (AArch64dup(simm5_8b:$imm5)))),
+  // add(mul(step_vector(1), dup(Y), dup(X)) -> index(X, Y).
+  def : Pat<(add (muloneuseop (nxv16i1 (AArch64ptrue 31)), (nxv16i8 (oneuseop (i32 1))), (nxv16i8 (AArch64dup(i32 GPR32:$Rm)))), (nxv16i8 (AArch64dup(simm5_8b:$imm5)))),
----------------
The `add(mul(step_vector` comments in this class and sve_int_index_rr are missing a close `)` for the `mul`.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D101593/new/

https://reviews.llvm.org/D101593