[PATCH] D153611: [AArch64] Treat the icmp in icmp(and(..), 0) as free

[Dave Green via Phabricator via llvm-commits]

dmgreen created this revision.
dmgreen added reviewers: SjoerdMeijer, jaykang10, samtebbs.
Herald added subscribers: hiraditya, kristof.beyls.
Herald added a project: All.
dmgreen requested review of this revision.
Herald added a project: LLVM.

As in https://godbolt.org/z/4dafd9Geq, the icmp from an `and` may use an `ands` to set flags, meaning the icmp is free.

This could also be done for add/sub, but those patterns often happen in the induction variable of a loop, making them quite performance sensitive.


https://reviews.llvm.org/D153611

Files:
  llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
  llvm/test/Analysis/CostModel/AArch64/cmp.ll


Index: llvm/test/Analysis/CostModel/AArch64/cmp.ll
===================================================================
--- llvm/test/Analysis/CostModel/AArch64/cmp.ll
+++ llvm/test/Analysis/CostModel/AArch64/cmp.ll
@@ -64,9 +64,9 @@
 ; CHECK-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %a16 = and i16 undef, undef
 ; CHECK-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c16 = icmp ne i16 %a16, 0
 ; CHECK-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %a32 = and i32 undef, undef
-; CHECK-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c32 = icmp eq i32 %a32, 0
+; CHECK-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %c32 = icmp eq i32 %a32, 0
 ; CHECK-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %a64 = and i64 undef, undef
-; CHECK-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c64 = icmp ne i64 %a64, 0
+; CHECK-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %c64 = icmp ne i64 %a64, 0
 ; CHECK-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %a128 = and i128 undef, undef
 ; CHECK-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %c128 = icmp eq i128 %a128, 0
 ; CHECK-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %av16i8 = and <16 x i8> undef, undef
Index: llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
===================================================================
--- llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
+++ llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
@@ -2815,6 +2815,16 @@
       return LT.first * 4; // fcvtl + fcvtl + fcmp + xtn
   }
 
+  // Treat the icmp in icmp(and, 0) as free, as we can make use of ands.
+  // FIXME: This can apply to more conditions and add/sub if it can be shown to
+  // be profitable.
+  if (ValTy->isIntegerTy() && ISD == ISD::SETCC && I &&
+      ICmpInst::isEquality(VecPred) &&
+      TLI->isTypeLegal(TLI->getValueType(DL, ValTy)) &&
+      match(I->getOperand(1), m_Zero()) &&
+      match(I->getOperand(0), m_And(m_Value(), m_Value())))
+    return 0;
+
   // The base case handles scalable vectors fine for now, since it treats the
   // cost as 1 * legalization cost.
   return BaseT::getCmpSelInstrCost(Opcode, ValTy, CondTy, VecPred, CostKind, I);


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D153611.533869.patch
Type: text/x-patch
Size: 2470 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230623/337f66ed/attachment.bin>