[PATCH] D91271: [AArch64] Attempt to sink mul operands

[Dave Green via Phabricator via llvm-commits]

dmgreen added inline comments.


================
Comment at: llvm/lib/Target/AArch64/AArch64ISelLowering.cpp:10967
+      ShuffleVectorInst *Shuffle = dyn_cast<ShuffleVectorInst>(Op);
+      if (!Shuffle)
+        continue;
----------------
Should this be checking for isZeroEltSplat too?


================
Comment at: llvm/lib/Target/AArch64/AArch64ISelLowering.cpp:10972
+      InsertElementInst *Insert = dyn_cast<InsertElementInst>(ShuffleOperand);
+      if (!Insert)
+        continue;
----------------
Check for insert into element 0 too?


================
Comment at: llvm/test/CodeGen/AArch64/aarch64-matrix-umull-smull.ll:169
+  %broadcast.splatinsert31 = insertelement <4 x i32> undef, i32 %splat.input.ext, i32 0
+  %broadcast.splat32 = shufflevector <4 x i32> %broadcast.splatinsert31, <4 x i32> %broadcast.splat, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
+  %3 = trunc i64 %index to i32
----------------
Try loading the value in the loop, to keep the mul loop-variant:
```
vector.body:                                      ; preds = %vector.header, %vector.body
  %index = phi i64 [ %index.next, %vector.body ], [ 0, %vector.header ]
  %g = getelementptr inbounds i16, i16* %A, i64 %index
  %val1 = load i16, i16* %g
  %splat.input.ext = zext i16 %val1 to i32
  %broadcast.splatinsert31 = insertelement <4 x i32> undef, i32 %splat.input.ext, i32 0
  %broadcast.splat32 = shufflevector <4 x i32> %broadcast.splatinsert31, <4 x i32> %broadcast.splat, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
```


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D91271/new/

https://reviews.llvm.org/D91271