[PATCH] D133154: [CostModel][AArch64] Fix ctpop intrinsic cost when NEON is disabled.

[Eli Friedman via Phabricator via llvm-commits]

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

If we don't have NEON, we use the generic fallback, which takes 12 instructions. Make sure the costs reflect that.

(On a related note, we could optimize the generic fallback a bit. It currently uses sequences like lsr+and+add; if we use and+lsr+add instead, we can fold the lsr into the add.  But I'm not planning to look into that at the moment.)


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D133154

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


Index: llvm/test/Analysis/CostModel/AArch64/ctpop.ll
===================================================================
--- llvm/test/Analysis/CostModel/AArch64/ctpop.ll
+++ llvm/test/Analysis/CostModel/AArch64/ctpop.ll
@@ -156,6 +156,22 @@
   ret <32 x i8> %ctpop
 }
 
+define i64 @test_ctpop_noneon_i64(i64 %a) "target-features"="-fp-armv8,-neon" {
+; CHECK-LABEL: 'test_ctpop_noneon_i64'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %ctpop = call i64 @llvm.ctpop.i64(i64 %a)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i64 %ctpop
+  %ctpop = call i64 @llvm.ctpop.i64(i64 %a)
+  ret i64 %ctpop
+}
+
+define <2 x i64> @test_ctpop_noneon_v2i64(<2 x i64> %a) "target-features"="-fp-armv8,-neon" {
+; CHECK-LABEL: 'test_ctpop_noneon_v2i64'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %ctpop = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %ctpop
+  %ctpop = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a)
+  ret <2 x i64> %ctpop
+}
+
 declare <2 x i64> @llvm.ctpop.v2i64(<2 x i64>)
 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32>)
 declare <4 x i32> @llvm.ctpop.v4i32(<4 x i32>)
Index: llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
===================================================================
--- llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
+++ llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
@@ -383,6 +383,10 @@
     break;
   }
   case Intrinsic::ctpop: {
+    if (!ST->hasNEON()) {
+      // 32-bit or 64-bit ctpop without NEON is 12 instructions.
+      return getTypeLegalizationCost(RetTy).first * 12;
+    }
     static const CostTblEntry CtpopCostTbl[] = {
         {ISD::CTPOP, MVT::v2i64, 4},
         {ISD::CTPOP, MVT::v4i32, 3},


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D133154.457395.patch
Type: text/x-patch
Size: 1858 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220901/2d294bb1/attachment-0001.bin>