[PATCH] D143283: [AArch64][SVE]: custom lower AVGFloor/AVGCeil.

Mon Feb 20 05:11:44 PST 2023

hassnaa-arm added inline comments.

================
Comment at: llvm/test/CodeGen/AArch64/sve-avg_floor_ceil.ll:62
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    and z0.d, z0.d, #0xffffffff
-; CHECK-NEXT:    adr z0.d, [z0.d, z1.d, uxtw]
-; CHECK-NEXT:    lsr z0.d, z0.d, #1
+; CHECK-NEXT:    lsr z2.d, z1.d, #1
+; CHECK-NEXT:    lsr z3.d, z0.d, #1
----------------
sdesmalen wrote:
> This doesn't look like an improvement, we don't really want to do this transform if it makes the resulting code worse. Do you know why this results in worse code?
In the new changes, the generated instructions are exactly the equivalent for AVGFloor, no additional instructions.
I think nothing can be done for this case.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D143283/new/

https://reviews.llvm.org/D143283