[PATCH] D129751: [AArch64][SVE] Add patterns to select masked add/sub instructions

[Paul Walker via Phabricator via llvm-commits]

paulwalker-arm added a comment.

Perhaps worth having a less descriptive name for the test (e.g. sve-masked-int-arith.ll or even less descriptive) given we're likely to add many more cases.



================
Comment at: llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td:340-341
+                              (sub node:$op2, node:$op1)>;
+def AArch64subr : PatFrag<(ops node:$pred, node:$op1, node:$op2),
+                          (int_aarch64_sve_subr node:$pred, node:$op1, node:$op2)>;
+def AArch64add : PatFrags<(ops node:$pred, node:$op1, node:$op2),
----------------
Is this strictly required? There's only a single pattern and no operand shuffling going on so I think not.


================
Comment at: llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td:342-347
+def AArch64add : PatFrags<(ops node:$pred, node:$op1, node:$op2),
+                          [(int_aarch64_sve_add node:$pred, node:$op1, node:$op2),
+                           (add node:$op1, (vselect node:$pred, node:$op2, (SVEDup0)))]>;
+def AArch64sub : PatFrags<(ops node:$pred, node:$op1, node:$op2),
+                          [(int_aarch64_sve_sub node:$pred, node:$op1, node:$op2),
+                           (sub node:$op1, (vselect node:$pred, node:$op2, (SVEDup0)))]>;
----------------
We have a naming scheme documented above which says these should be called `AArch64add_m1` and `AArch64sub_m1` which means you don't need to rename the original `AArch64subr` def.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D129751/new/

https://reviews.llvm.org/D129751