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

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.)

Differential Revision: https://reviews.llvm.org/D133154

diff --git a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
index ce3c34b..6bdd89d 100644 (file)
--- a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
@@ -383,6 +383,10 @@ AArch64TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
     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},

diff --git a/llvm/test/Analysis/CostModel/AArch64/ctpop.ll b/llvm/test/Analysis/CostModel/AArch64/ctpop.ll
index e34ce1f..03e719f 100644 (file)
--- a/llvm/test/Analysis/CostModel/AArch64/ctpop.ll
+++ b/llvm/test/Analysis/CostModel/AArch64/ctpop.ll
@@ -156,6 +156,22 @@ define <32 x i8> @test_ctpop_v32i8(<32 x i8> %a) {
   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>)