[PATCH] D141665: [DAGCombine][AArch64][ARM] Combine abd(sub(x, y)) to abd if the sub is nsw

[Dave Green via Phabricator via llvm-commits]

dmgreen created this revision.
dmgreen added reviewers: RKSimon, spatel, efriedma.
Herald added subscribers: StephenFan, ecnelises, hiraditya, kristof.beyls.
Herald added a project: All.
dmgreen requested review of this revision.
Herald added a project: LLVM.

This implements the fold `(abs (sub nsw x, y))` -> `abds(x, y)`. Providing the sub is nsw this appears to be valid without the extensions that are usually used for `abds`. https://alive2.llvm.org/ce/z/XHVaB3. The equivalent abdu combine seems to not be valid.


https://reviews.llvm.org/D141665

Files:
  llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
  llvm/test/CodeGen/AArch64/neon-abd.ll


Index: llvm/test/CodeGen/AArch64/neon-abd.ll
===================================================================
--- llvm/test/CodeGen/AArch64/neon-abd.ll
+++ llvm/test/CodeGen/AArch64/neon-abd.ll
@@ -409,8 +409,7 @@
 define <16 x i8> @sabd_v16i8_nsw(<16 x i8> %a, <16 x i8> %b) #0 {
 ; CHECK-LABEL: sabd_v16i8_nsw:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    sub v0.16b, v0.16b, v1.16b
-; CHECK-NEXT:    abs v0.16b, v0.16b
+; CHECK-NEXT:    sabd v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
   %sub = sub nsw <16 x i8> %a, %b
   %abs = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %sub, i1 true)
@@ -420,8 +419,7 @@
 define <8 x i16> @sabd_v8i16_nsw(<8 x i16> %a, <8 x i16> %b) #0 {
 ; CHECK-LABEL: sabd_v8i16_nsw:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    sub v0.8h, v0.8h, v1.8h
-; CHECK-NEXT:    abs v0.8h, v0.8h
+; CHECK-NEXT:    sabd v0.8h, v0.8h, v1.8h
 ; CHECK-NEXT:    ret
   %sub = sub nsw <8 x i16> %a, %b
   %abs = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %sub, i1 true)
@@ -431,8 +429,7 @@
 define <4 x i32> @sabd_v4i32_nsw(<4 x i32> %a, <4 x i32> %b) #0 {
 ; CHECK-LABEL: sabd_v4i32_nsw:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    sub v0.4s, v0.4s, v1.4s
-; CHECK-NEXT:    abs v0.4s, v0.4s
+; CHECK-NEXT:    sabd v0.4s, v0.4s, v1.4s
 ; CHECK-NEXT:    ret
   %sub = sub nsw <4 x i32> %a, %b
   %abs = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %sub, i1 true)
Index: llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
===================================================================
--- llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -10148,6 +10148,13 @@
   if (SDValue ABD = combineABSToABD(N, DAG, TLI))
     return ABD;
 
+  // fold (abs (sub nsw x, y)) -> abds(x, y)
+  if (N->getOperand(0).getOpcode() == ISD::SUB &&
+      N->getOperand(0)->getFlags().hasNoSignedWrap() &&
+      TLI.isOperationLegalOrCustom(ISD::ABDS, VT))
+    return DAG.getNode(ISD::ABDS, SDLoc(N), VT, N->getOperand(0).getOperand(0),
+                       N->getOperand(0).getOperand(1));
+
   // fold (abs (sign_extend_inreg x)) -> (zero_extend (abs (truncate x)))
   // iff zero_extend/truncate are free.
   if (N0.getOpcode() == ISD::SIGN_EXTEND_INREG) {


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D141665.488889.patch
Type: text/x-patch
Size: 2198 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230113/522f825c/attachment.bin>