[AArch64] Add BF16 REV costs.

Same as FP16 costs, these full reverse shuffles can use REV or REV+EXT.

Author: davemgreen

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

@@ -5023,9 +5023,11 @@ InstructionCost AArch64TTIImpl::getShuffleCost(
         {TTI::SK_Reverse, MVT::v4f32, 2}, // REV64; EXT
         {TTI::SK_Reverse, MVT::v2f64, 1}, // EXT
         {TTI::SK_Reverse, MVT::v8f16, 2}, // REV64; EXT
+        {TTI::SK_Reverse, MVT::v8bf16, 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::v4bf16, 1}, // REV64
         {TTI::SK_Reverse, MVT::v4i16, 1}, // REV64
         {TTI::SK_Reverse, MVT::v8i8, 1},  // REV64
         // Splice can all be lowered as `ext`.
@@ -5039,8 +5041,8 @@ InstructionCost AArch64TTIImpl::getShuffleCost(
         {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::v4bf16, 1},
         {TTI::SK_Splice, MVT::v4i16, 1},
         {TTI::SK_Splice, MVT::v8i8, 1},
         // Broadcast shuffle kinds for scalable vectors

llvm/test/Analysis/CostModel/AArch64/shuffle-reverse.ll

@@ -27,10 +27,10 @@ define void @reverse() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4f16 = shufflevector <4 x half> undef, <4 x half> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8f16 = shufflevector <8 x half> undef, <8 x half> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v16f16 = shufflevector <16 x half> undef, <16 x half> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2bf16 = shufflevector <2 x bfloat> undef, <2 x bfloat> undef, <2 x i32> <i32 1, i32 0>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2bf16 = shufflevector <2 x bfloat> undef, <2 x bfloat> undef, <2 x i32> <i32 1, i32 0>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4bf16 = shufflevector <4 x bfloat> undef, <4 x bfloat> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v8bf16 = shufflevector <8 x bfloat> undef, <8 x bfloat> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v16bf16 = shufflevector <16 x bfloat> undef, <16 x bfloat> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8bf16 = shufflevector <8 x bfloat> undef, <8 x bfloat> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v16bf16 = shufflevector <16 x bfloat> undef, <16 x bfloat> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f32 = shufflevector <2 x float> undef, <2 x float> undef, <2 x i32> <i32 1, i32 0>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4f32 = shufflevector <4 x float> undef, <4 x float> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8f32 = shufflevector <8 x float> undef, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
@@ -94,7 +94,7 @@ define void @vrev64() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v16f16 = shufflevector <16 x half> undef, <16 x half> undef, <16 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4bf16 = shufflevector <4 x bfloat> undef, <4 x bfloat> undef, <4 x i32> <i32 1, i32 0, i32 3, i32 2>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8bf16 = shufflevector <8 x bfloat> undef, <8 x bfloat> undef, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v16bf16 = shufflevector <16 x bfloat> undef, <16 x bfloat> undef, <16 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v16bf16 = shufflevector <16 x bfloat> undef, <16 x bfloat> undef, <16 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4f32 = shufflevector <4 x float> undef, <4 x float> undef, <4 x i32> <i32 1, i32 0, i32 3, i32 2>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8f32 = shufflevector <8 x float> undef, <8 x float> undef, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16f32 = shufflevector <16 x float> undef, <16 x float> undef, <16 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8>

llvm/test/Analysis/CostModel/AArch64/vector-reverse.ll

@@ -21,8 +21,8 @@ define void @vector_reverse() #0{
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %12 = call <8 x float> @llvm.vector.reverse.v8f32(<8 x float> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %13 = call <2 x double> @llvm.vector.reverse.v2f64(<2 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %14 = call <4 x double> @llvm.vector.reverse.v4f64(<4 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %15 = call <8 x bfloat> @llvm.vector.reverse.v8bf16(<8 x bfloat> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %16 = call <16 x bfloat> @llvm.vector.reverse.v16bf16(<16 x bfloat> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %15 = call <8 x bfloat> @llvm.vector.reverse.v8bf16(<8 x bfloat> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %16 = call <16 x bfloat> @llvm.vector.reverse.v16bf16(<16 x bfloat> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;