[AArch64] Add SK_Splice fixed-width costs

[dmgreen]

Summary

A fixed length SK_Splice shuffle vector is lowered to a Ext undef AArch64, which should have a cost of 1.

[fhahn]

LGTM, thanks!

llvm/test/Analysis/CostModel/AArch64/splice.ll

32

might be good to also have a test where the cost is != 0.

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

InstructionCost AArch64TTIImpl::getShuffleCost(TTI::ShuffleKind Kind,
   // from the perfect shuffle tables.
   if (Mask.size() == 4 && Tp->getElementCount() == ElementCount::getFixed(4) &&
       (Tp->getScalarSizeInBits() == 16 || Tp->getScalarSizeInBits() == 32) &&
       all_of(Mask, [](int E) { return E < 8; }))
     return getPerfectShuffleCost(Mask);
 
   if (Kind == TTI::SK_Broadcast || Kind == TTI::SK_Transpose ||
       Kind == TTI::SK_Select || Kind == TTI::SK_PermuteSingleSrc ||
-      Kind == TTI::SK_Reverse) {
-
+      Kind == TTI::SK_Reverse || Kind == TTI::SK_Splice) {
     static const CostTblEntry ShuffleTbl[] = {
       // Broadcast shuffle kinds can be performed with 'dup'.
       { TTI::SK_Broadcast, MVT::v8i8,  1 },
       { TTI::SK_Broadcast, MVT::v16i8, 1 },
       { TTI::SK_Broadcast, MVT::v4i16, 1 },
       { TTI::SK_Broadcast, MVT::v8i16, 1 },
       { TTI::SK_Broadcast, MVT::v2i32, 1 },
       { TTI::SK_Broadcast, MVT::v4i32, 1 },
static const CostTblEntry ShuffleTbl[] = {
       { TTI::SK_Reverse, MVT::v4f32, 2 }, // REV64; EXT
       { TTI::SK_Reverse, MVT::v2f64, 1 }, // mov.
       { TTI::SK_Reverse, MVT::v8f16, 2 }, // REV64; EXT
       { TTI::SK_Reverse, MVT::v8i16, 2 }, // REV64; EXT
       { TTI::SK_Reverse, MVT::v16i8, 2 }, // REV64; EXT
       { TTI::SK_Reverse, MVT::v4f16, 1 }, // REV64
       { TTI::SK_Reverse, MVT::v4i16, 1 }, // REV64
       { TTI::SK_Reverse, MVT::v8i8, 1 }, // REV64
+      // Splice can all be lowered as `ext`.
+      { TTI::SK_Splice, MVT::v2i32, 1 },
+      { TTI::SK_Splice, MVT::v4i32, 1 },
+      { TTI::SK_Splice, MVT::v2i64, 1 },
+      { TTI::SK_Splice, MVT::v2f32, 1 },
+      { TTI::SK_Splice, MVT::v4f32, 1 },
+      { TTI::SK_Splice, MVT::v2f64, 1 },
+      { TTI::SK_Splice, MVT::v8f16, 1 },
+      { TTI::SK_Splice, MVT::v8bf16, 1 },
+      { TTI::SK_Splice, MVT::v8i16, 1 },
+      { TTI::SK_Splice, MVT::v16i8, 1 },
+      { TTI::SK_Splice, MVT::v4bf16, 1 },
+      { TTI::SK_Splice, MVT::v4f16, 1 },
+      { TTI::SK_Splice, MVT::v4i16, 1 },
+      { TTI::SK_Splice, MVT::v8i8, 1 },
       // Broadcast shuffle kinds for scalable vectors
       { TTI::SK_Broadcast, MVT::nxv16i8,  1 },
       { TTI::SK_Broadcast, MVT::nxv8i16,  1 },
       { TTI::SK_Broadcast, MVT::nxv4i32,  1 },
       { TTI::SK_Broadcast, MVT::nxv2i64,  1 },
       { TTI::SK_Broadcast, MVT::nxv2f16,  1 },
       { TTI::SK_Broadcast, MVT::nxv4f16,  1 },
       { TTI::SK_Broadcast, MVT::nxv8f16,  1 },

llvm/test/Analysis/CostModel/AArch64/splice.ll

 ; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
 ; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -S -mtriple=aarch64--linux-gnu | FileCheck %s
 
 define void @vector_splice() #0 {
 ; CHECK-LABEL: 'vector_splice'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %splice.v16i8 = call <16 x i8> @llvm.experimental.vector.splice.v16i8(<16 x i8> zeroinitializer, <16 x i8> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %splice.v32i8 = call <32 x i8> @llvm.experimental.vector.splice.v32i8(<32 x i8> zeroinitializer, <32 x i8> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %splice.v2i16 = call <2 x i16> @llvm.experimental.vector.splice.v2i16(<2 x i16> zeroinitializer, <2 x i16> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %splice.v4i16 = call <4 x i16> @llvm.experimental.vector.splice.v4i16(<4 x i16> zeroinitializer, <4 x i16> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %splice.v8i16 = call <8 x i16> @llvm.experimental.vector.splice.v8i16(<8 x i16> zeroinitializer, <8 x i16> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %splice.v16i16 = call <16 x i16> @llvm.experimental.vector.splice.v16i16(<16 x i16> zeroinitializer, <16 x i16> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %splice.v4i32 = call <4 x i32> @llvm.experimental.vector.splice.v4i32(<4 x i32> zeroinitializer, <4 x i32> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %splice.v8i32 = call <8 x i32> @llvm.experimental.vector.splice.v8i32(<8 x i32> zeroinitializer, <8 x i32> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %splice.v2i64 = call <2 x i64> @llvm.experimental.vector.splice.v2i64(<2 x i64> zeroinitializer, <2 x i64> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %splice.v4i64 = call <4 x i64> @llvm.experimental.vector.splice.v4i64(<4 x i64> zeroinitializer, <4 x i64> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %splice.v2f16 = call <2 x half> @llvm.experimental.vector.splice.v2f16(<2 x half> zeroinitializer, <2 x half> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %splice.v4f16 = call <4 x half> @llvm.experimental.vector.splice.v4f16(<4 x half> zeroinitializer, <4 x half> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %splice.v8f16 = call <8 x half> @llvm.experimental.vector.splice.v8f16(<8 x half> zeroinitializer, <8 x half> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %splice.v16f16 = call <16 x half> @llvm.experimental.vector.splice.v16f16(<16 x half> zeroinitializer, <16 x half> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %splice.v2f32 = call <2 x float> @llvm.experimental.vector.splice.v2f32(<2 x float> zeroinitializer, <2 x float> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %splice.v4f32 = call <4 x float> @llvm.experimental.vector.splice.v4f32(<4 x float> zeroinitializer, <4 x float> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %splice.v8f32 = call <8 x float> @llvm.experimental.vector.splice.v8f32(<8 x float> zeroinitializer, <8 x float> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %splice.v2f64 = call <2 x double> @llvm.experimental.vector.splice.v2f64(<2 x double> zeroinitializer, <2 x double> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %splice.v4f64 = call <4 x double> @llvm.experimental.vector.splice.v4f64(<4 x double> zeroinitializer, <4 x double> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %splice.v2bf16 = call <2 x bfloat> @llvm.experimental.vector.splice.v2bf16(<2 x bfloat> zeroinitializer, <2 x bfloat> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %splice.v4bf16 = call <4 x bfloat> @llvm.experimental.vector.splice.v4bf16(<4 x bfloat> zeroinitializer, <4 x bfloat> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %splice.v8bf16 = call <8 x bfloat> @llvm.experimental.vector.splice.v8bf16(<8 x bfloat> zeroinitializer, <8 x bfloat> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %splice.v16bf16 = call <16 x bfloat> @llvm.experimental.vector.splice.v16bf16(<16 x bfloat> zeroinitializer, <16 x bfloat> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %splice.v16i1 = call <16 x i1> @llvm.experimental.vector.splice.v16i1(<16 x i1> zeroinitializer, <16 x i1> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %splice.v8i1 = call <8 x i1> @llvm.experimental.vector.splice.v8i1(<8 x i1> zeroinitializer, <8 x i1> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %splice.v4i1 = call <4 x i1> @llvm.experimental.vector.splice.v4i1(<4 x i1> zeroinitializer, <4 x i1> zeroinitializer, i32 1)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %splice.v2i1 = call <2 x i1> @llvm.experimental.vector.splice.v2i1(<2 x i1> zeroinitializer, <2 x i1> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v16i8 = call <16 x i8> @llvm.experimental.vector.splice.v16i8(<16 x i8> zeroinitializer, <16 x i8> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %splice.v32i8 = call <32 x i8> @llvm.experimental.vector.splice.v32i8(<32 x i8> zeroinitializer, <32 x i8> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v2i16 = call <2 x i16> @llvm.experimental.vector.splice.v2i16(<2 x i16> zeroinitializer, <2 x i16> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v4i16 = call <4 x i16> @llvm.experimental.vector.splice.v4i16(<4 x i16> zeroinitializer, <4 x i16> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v8i16 = call <8 x i16> @llvm.experimental.vector.splice.v8i16(<8 x i16> zeroinitializer, <8 x i16> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %splice.v16i16 = call <16 x i16> @llvm.experimental.vector.splice.v16i16(<16 x i16> zeroinitializer, <16 x i16> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v4i32 = call <4 x i32> @llvm.experimental.vector.splice.v4i32(<4 x i32> zeroinitializer, <4 x i32> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %splice.v8i32 = call <8 x i32> @llvm.experimental.vector.splice.v8i32(<8 x i32> zeroinitializer, <8 x i32> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v2i64 = call <2 x i64> @llvm.experimental.vector.splice.v2i64(<2 x i64> zeroinitializer, <2 x i64> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %splice.v4i64 = call <4 x i64> @llvm.experimental.vector.splice.v4i64(<4 x i64> zeroinitializer, <4 x i64> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v2f16 = call <2 x half> @llvm.experimental.vector.splice.v2f16(<2 x half> zeroinitializer, <2 x half> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v4f16 = call <4 x half> @llvm.experimental.vector.splice.v4f16(<4 x half> zeroinitializer, <4 x half> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v8f16 = call <8 x half> @llvm.experimental.vector.splice.v8f16(<8 x half> zeroinitializer, <8 x half> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %splice.v16f16 = call <16 x half> @llvm.experimental.vector.splice.v16f16(<16 x half> zeroinitializer, <16 x half> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v2f32 = call <2 x float> @llvm.experimental.vector.splice.v2f32(<2 x float> zeroinitializer, <2 x float> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v4f32 = call <4 x float> @llvm.experimental.vector.splice.v4f32(<4 x float> zeroinitializer, <4 x float> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %splice.v8f32 = call <8 x float> @llvm.experimental.vector.splice.v8f32(<8 x float> zeroinitializer, <8 x float> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v2f64 = call <2 x double> @llvm.experimental.vector.splice.v2f64(<2 x double> zeroinitializer, <2 x double> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %splice.v4f64 = call <4 x double> @llvm.experimental.vector.splice.v4f64(<4 x double> zeroinitializer, <4 x double> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v2bf16 = call <2 x bfloat> @llvm.experimental.vector.splice.v2bf16(<2 x bfloat> zeroinitializer, <2 x bfloat> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v4bf16 = call <4 x bfloat> @llvm.experimental.vector.splice.v4bf16(<4 x bfloat> zeroinitializer, <4 x bfloat> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v8bf16 = call <8 x bfloat> @llvm.experimental.vector.splice.v8bf16(<8 x bfloat> zeroinitializer, <8 x bfloat> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %splice.v16bf16 = call <16 x bfloat> @llvm.experimental.vector.splice.v16bf16(<16 x bfloat> zeroinitializer, <16 x bfloat> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v16i1 = call <16 x i1> @llvm.experimental.vector.splice.v16i1(<16 x i1> zeroinitializer, <16 x i1> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v8i1 = call <8 x i1> @llvm.experimental.vector.splice.v8i1(<8 x i1> zeroinitializer, <8 x i1> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v4i1 = call <4 x i1> @llvm.experimental.vector.splice.v4i1(<4 x i1> zeroinitializer, <4 x i1> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %splice.v2i1 = call <2 x i1> @llvm.experimental.vector.splice.v2i1(<2 x i1> zeroinitializer, <2 x i1> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %splice.v2i128 = call <2 x i128> @llvm.experimental.vector.splice.v2i128(<2 x i128> zeroinitializer, <2 x i128> zeroinitializer, i32 1)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %splice.v16i8 = call <16 x i8> @llvm.experimental.vector.splice.v16i8(<16 x i8> zeroinitializer, <16 x i8> zeroinitializer, i32 1)
   %splice.v32i8 = call <32 x i8> @llvm.experimental.vector.splice.v32i8(<32 x i8> zeroinitializer, <32 x i8> zeroinitializer, i32 1)
   %splice.v2i16 = call <2 x i16> @llvm.experimental.vector.splice.v2i16(<2 x i16> zeroinitializer, <2 x i16> zeroinitializer, i32 1)
   %splice.v4i16 = call <4 x i16> @llvm.experimental.vector.splice.v4i16(<4 x i16> zeroinitializer, <4 x i16> zeroinitializer, i32 1)
   %splice.v8i16 = call <8 x i16> @llvm.experimental.vector.splice.v8i16(<8 x i16> zeroinitializer, <8 x i16> zeroinitializer, i32 1)
   %splice.v16i16 = call <16 x i16> @llvm.experimental.vector.splice.v16i16(<16 x i16> zeroinitializer, <16 x i16> zeroinitializer, i32 1)
   %splice.v4i32 = call <4 x i32> @llvm.experimental.vector.splice.v4i32(<4 x i32> zeroinitializer, <4 x i32> zeroinitializer, i32 1)
;
   %splice.v2bf16 = call <2 x bfloat> @llvm.experimental.vector.splice.v2bf16(<2 x bfloat> zeroinitializer, <2 x bfloat> zeroinitializer, i32 1)
   %splice.v4bf16 = call <4 x bfloat> @llvm.experimental.vector.splice.v4bf16(<4 x bfloat> zeroinitializer, <4 x bfloat> zeroinitializer, i32 1)
   %splice.v8bf16 = call <8 x bfloat> @llvm.experimental.vector.splice.v8bf16(<8 x bfloat> zeroinitializer, <8 x bfloat> zeroinitializer, i32 1)
   %splice.v16bf16 = call <16 x bfloat> @llvm.experimental.vector.splice.v16bf16(<16 x bfloat> zeroinitializer, <16 x bfloat> zeroinitializer, i32 1)
   %splice.v16i1 = call <16 x i1> @llvm.experimental.vector.splice.v16i1(<16 x i1> zeroinitializer, <16 x i1> zeroinitializer, i32 1)
   %splice.v8i1 =  call <8 x i1> @llvm.experimental.vector.splice.v8i1(<8 x i1> zeroinitializer, <8 x i1> zeroinitializer, i32 1)
   %splice.v4i1 = call <4 x i1> @llvm.experimental.vector.splice.v4i1(<4 x i1> zeroinitializer, <4 x i1> zeroinitializer, i32 1)
   %splice.v2i1 = call <2 x i1> @llvm.experimental.vector.splice.v2i1(<2 x i1> zeroinitializer, <2 x i1> zeroinitializer, i32 1)
+  %splice.v2i128 = call <2 x i128> @llvm.experimental.vector.splice.v2i128(<2 x i128> zeroinitializer, <2 x i128> zeroinitializer, i32 1)
   ret void
 }
 
 declare <2 x i1> @llvm.experimental.vector.splice.v2i1(<2 x i1>, <2 x i1>, i32)
 declare <4 x i1> @llvm.experimental.vector.splice.v4i1(<4 x i1>, <4 x i1>, i32)
 declare <8 x i1> @llvm.experimental.vector.splice.v8i1(<8 x i1>, <8 x i1>, i32)
 declare <16 x i1> @llvm.experimental.vector.splice.v16i1(<16 x i1>, <16 x i1>, i32)
 declare <2 x i8> @llvm.experimental.vector.splice.v2i8(<2 x i8>, <2 x i8>, i32)

 declare <8 x bfloat> @llvm.experimental.vector.splice.v8bf16(<8 x bfloat>, <8 x bfloat>, i32)
 declare <16 x bfloat> @llvm.experimental.vector.splice.v16bf16(<16 x bfloat>, <16 x bfloat>, i32)
 declare <2 x float> @llvm.experimental.vector.splice.v2f32(<2 x float>, <2 x float>, i32)
 declare <4 x float> @llvm.experimental.vector.splice.v4f32(<4 x float>, <4 x float>, i32)
 declare <8 x float> @llvm.experimental.vector.splice.v8f32(<8 x float>, <8 x float>, i32)
 declare <16 x float> @llvm.experimental.vector.splice.v16f32(<16 x float>, <16 x float>, i32)
 declare <2 x double> @llvm.experimental.vector.splice.v2f64(<2 x double>, <2 x double>, i32)
 declare <4 x double> @llvm.experimental.vector.splice.v4f64(<4 x double>, <4 x double>, i32)
+declare <2 x i128> @llvm.experimental.vector.splice.v2i128(<2 x i128>, <2 x i128>, i32)
 
 attributes #0 = { "target-features"="+bf16" }