AArch64: improve handling & modelling of FP_TO_XINT nodes.

There's probably no acatual change in behaviour here, just updating
the LowerFP_TO_INT function to be more similar to the reverse
implementation and updating costs to current CodeGen.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@210985 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AArch64/AArch64ISelLowering.cpp b/lib/Target/AArch64/AArch64ISelLowering.cpp
index fd9e69d32319c88f44dcc5dd98c4b26154bd7168..c2cf5b7f17f0e7a8a9318a0b4228eba7d1e91dc5 100644 (file)
--- a/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -1387,24 +1387,22 @@ static SDValue LowerVectorFP_TO_INT(SDValue Op, SelectionDAG &DAG) {
   EVT InVT = Op.getOperand(0).getValueType();
   EVT VT = Op.getValueType();
 
-  // FP_TO_XINT conversion from the same type are legal.
-  if (VT.getSizeInBits() == InVT.getSizeInBits())
-    return Op;
-
-  if (InVT == MVT::v2f64 || InVT == MVT::v4f32) {
+  if (VT.getSizeInBits() < InVT.getSizeInBits()) {
     SDLoc dl(Op);
     SDValue Cv =
         DAG.getNode(Op.getOpcode(), dl, InVT.changeVectorElementTypeToInteger(),
                     Op.getOperand(0));
     return DAG.getNode(ISD::TRUNCATE, dl, VT, Cv);
-  } else if (InVT == MVT::v2f32) {
+  }
+
+  if (VT.getSizeInBits() > InVT.getSizeInBits()) {
     SDLoc dl(Op);
     SDValue Ext = DAG.getNode(ISD::FP_EXTEND, dl, MVT::v2f64, Op.getOperand(0));
     return DAG.getNode(Op.getOpcode(), dl, VT, Ext);
   }
 
   // Type changing conversions are illegal.
-  return SDValue();
+  return Op;
 }
 
 SDValue AArch64TargetLowering::LowerFP_TO_INT(SDValue Op,

diff --git a/lib/Target/AArch64/AArch64TargetTransformInfo.cpp b/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
index 4fae0a53b325d404104819dac48822f94f34999a..1dac14b96af9119e2f6ecf069c19e767d1f83251 100644 (file)
--- a/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
+++ b/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
@@ -315,8 +315,10 @@ unsigned AArch64TTI::getCastInstrCost(unsigned Opcode, Type *Dst,
     // Complex: to v2f32
     { ISD::SINT_TO_FP, MVT::v2f32, MVT::v2i8,  3 },
     { ISD::SINT_TO_FP, MVT::v2f32, MVT::v2i16, 3 },
+    { ISD::SINT_TO_FP, MVT::v2f32, MVT::v2i64, 2 },
     { ISD::UINT_TO_FP, MVT::v2f32, MVT::v2i8,  3 },
     { ISD::UINT_TO_FP, MVT::v2f32, MVT::v2i16, 3 },
+    { ISD::UINT_TO_FP, MVT::v2f32, MVT::v2i64, 2 },
 
     // Complex: to v4f32
     { ISD::SINT_TO_FP, MVT::v4f32, MVT::v4i8,  4 },
@@ -341,12 +343,27 @@ unsigned AArch64TTI::getCastInstrCost(unsigned Opcode, Type *Dst,
     { ISD::FP_TO_UINT, MVT::v4i32, MVT::v4f32, 1 },
     { ISD::FP_TO_UINT, MVT::v2i64, MVT::v2f64, 1 },
 
-    { ISD::FP_TO_SINT, MVT::v2i32, MVT::v2f64, 2 },
+    // Complex, from v2f32: legal type is v2i32 (no cost) or v2i64 (1 ext).
     { ISD::FP_TO_SINT, MVT::v2i64, MVT::v2f32, 2 },
-    { ISD::FP_TO_SINT, MVT::v4i16, MVT::v4f32, 2 },
-    { ISD::FP_TO_UINT, MVT::v2i32, MVT::v2f64, 2 },
+    { ISD::FP_TO_SINT, MVT::v2i16, MVT::v2f32, 1 },
+    { ISD::FP_TO_SINT, MVT::v2i8,  MVT::v2f32, 1 },
     { ISD::FP_TO_UINT, MVT::v2i64, MVT::v2f32, 2 },
+    { ISD::FP_TO_UINT, MVT::v2i16, MVT::v2f32, 1 },
+    { ISD::FP_TO_UINT, MVT::v2i8,  MVT::v2f32, 1 },
+
+    // Complex, from v4f32: legal type is v4i16, 1 narrowing => ~2
+    { ISD::FP_TO_SINT, MVT::v4i16, MVT::v4f32, 2 },
+    { ISD::FP_TO_SINT, MVT::v4i8,  MVT::v4f32, 2 },
     { ISD::FP_TO_UINT, MVT::v4i16, MVT::v4f32, 2 },
+    { ISD::FP_TO_UINT, MVT::v4i8,  MVT::v4f32, 2 },
+
+    // Complex, from v2f64: legal type is v2i32, 1 narrowing => ~2.
+    { ISD::FP_TO_SINT, MVT::v2i32, MVT::v2f64, 2 },
+    { ISD::FP_TO_SINT, MVT::v2i16, MVT::v2f64, 2 },
+    { ISD::FP_TO_SINT, MVT::v2i8,  MVT::v2f64, 2 },
+    { ISD::FP_TO_UINT, MVT::v2i32, MVT::v2f64, 2 },
+    { ISD::FP_TO_UINT, MVT::v2i16, MVT::v2f64, 2 },
+    { ISD::FP_TO_UINT, MVT::v2i8,  MVT::v2f64, 2 },
   };
 
   int Idx = ConvertCostTableLookup<MVT>(

diff --git a/test/CodeGen/AArch64/complex-fp-to-int.ll b/test/CodeGen/AArch64/complex-fp-to-int.ll
index 1ea47ade81ead1b54e6a8025ee3253d209f1b14d..13cf762c3d2e2839da5443df4fdb5e187fb1f7d2 100644 (file)
--- a/test/CodeGen/AArch64/complex-fp-to-int.ll
+++ b/test/CodeGen/AArch64/complex-fp-to-int.ll
@@ -1,24 +1,141 @@
 ; RUN: llc < %s -march=arm64 -aarch64-neon-syntax=apple | FileCheck %s
 
-; CHECK: fptosi_1
-; CHECK: fcvtzs.2d
-; CHECK: xtn.2s
-; CHECK: ret
-define void @fptosi_1(<2 x double> %in, <2 x i32>* %addr) nounwind noinline ssp {
-entry:
-  %0 = fptosi <2 x double> %in to <2 x i32>
-  store <2 x i32> %0, <2 x i32>* %addr, align 8
-  ret void
-}
-
-; CHECK: fptoui_1
-; CHECK: fcvtzu.2d
-; CHECK: xtn.2s
-; CHECK: ret
-define void @fptoui_1(<2 x double> %in, <2 x i32>* %addr) nounwind noinline ssp {
-entry:
-  %0 = fptoui <2 x double> %in to <2 x i32>
-  store <2 x i32> %0, <2 x i32>* %addr, align 8
-  ret void
+define <2 x i64> @test_v2f32_to_signed_v2i64(<2 x float> %in) {
+; CHECK-LABEL: test_v2f32_to_signed_v2i64:
+; CHECK: fcvtl [[VAL64:v[0-9]+]].2d, v0.2s
+; CHECK: fcvtzs.2d v0, [[VAL64]]
+
+  %val = fptosi <2 x float> %in to <2 x i64>
+  ret <2 x i64> %val
+}
+
+define <2 x i64> @test_v2f32_to_unsigned_v2i64(<2 x float> %in) {
+; CHECK-LABEL: test_v2f32_to_unsigned_v2i64:
+; CHECK: fcvtl [[VAL64:v[0-9]+]].2d, v0.2s
+; CHECK: fcvtzu.2d v0, [[VAL64]]
+
+  %val = fptoui <2 x float> %in to <2 x i64>
+  ret <2 x i64> %val
+}
+
+define <2 x i16> @test_v2f32_to_signed_v2i16(<2 x float> %in) {
+; CHECK-LABEL: test_v2f32_to_signed_v2i16:
+; CHECK: fcvtzs.2s v0, v0
+
+  %val = fptosi <2 x float> %in to <2 x i16>
+  ret <2 x i16> %val
+}
+
+define <2 x i16> @test_v2f32_to_unsigned_v2i16(<2 x float> %in) {
+; CHECK-LABEL: test_v2f32_to_unsigned_v2i16:
+; CHECK: fcvtzs.2s v0, v0
+
+  %val = fptoui <2 x float> %in to <2 x i16>
+  ret <2 x i16> %val
 }
 
+define <2 x i8> @test_v2f32_to_signed_v2i8(<2 x float> %in) {
+; CHECK-LABEL: test_v2f32_to_signed_v2i8:
+; CHECK: fcvtzs.2s v0, v0
+
+  %val = fptosi <2 x float> %in to <2 x i8>
+  ret <2 x i8> %val
+}
+
+define <2 x i8> @test_v2f32_to_unsigned_v2i8(<2 x float> %in) {
+; CHECK-LABEL: test_v2f32_to_unsigned_v2i8:
+; CHECK: fcvtzs.2s v0, v0
+
+  %val = fptoui <2 x float> %in to <2 x i8>
+  ret <2 x i8> %val
+}
+
+define <4 x i16> @test_v4f32_to_signed_v4i16(<4 x float> %in) {
+; CHECK-LABEL: test_v4f32_to_signed_v4i16:
+; CHECK: fcvtzs.4s [[VAL64:v[0-9]+]], v0
+; CHECK: xtn.4h v0, [[VAL64]]
+
+  %val = fptosi <4 x float> %in to <4 x i16>
+  ret <4 x i16> %val
+}
+
+define <4 x i16> @test_v4f32_to_unsigned_v4i16(<4 x float> %in) {
+; CHECK-LABEL: test_v4f32_to_unsigned_v4i16:
+; CHECK: fcvtzu.4s [[VAL64:v[0-9]+]], v0
+; CHECK: xtn.4h v0, [[VAL64]]
+
+  %val = fptoui <4 x float> %in to <4 x i16>
+  ret <4 x i16> %val
+}
+
+define <4 x i8> @test_v4f32_to_signed_v4i8(<4 x float> %in) {
+; CHECK-LABEL: test_v4f32_to_signed_v4i8:
+; CHECK: fcvtzs.4s [[VAL64:v[0-9]+]], v0
+; CHECK: xtn.4h v0, [[VAL64]]
+
+  %val = fptosi <4 x float> %in to <4 x i8>
+  ret <4 x i8> %val
+}
+
+define <4 x i8> @test_v4f32_to_unsigned_v4i8(<4 x float> %in) {
+; CHECK-LABEL: test_v4f32_to_unsigned_v4i8:
+; CHECK: fcvtzs.4s [[VAL64:v[0-9]+]], v0
+; CHECK: xtn.4h v0, [[VAL64]]
+
+  %val = fptoui <4 x float> %in to <4 x i8>
+  ret <4 x i8> %val
+}
+
+define <2 x i32> @test_v2f64_to_signed_v2i32(<2 x double> %in) {
+; CHECK-LABEL: test_v2f64_to_signed_v2i32:
+; CHECK: fcvtzs.2d [[VAL64:v[0-9]+]], v0
+; CHECK: xtn.2s v0, [[VAL64]]
+
+  %val = fptosi <2 x double> %in to <2 x i32>
+  ret <2 x i32> %val
+}
+
+define <2 x i32> @test_v2f64_to_unsigned_v2i32(<2 x double> %in) {
+; CHECK-LABEL: test_v2f64_to_unsigned_v2i32:
+; CHECK: fcvtzu.2d [[VAL64:v[0-9]+]], v0
+; CHECK: xtn.2s v0, [[VAL64]]
+
+  %val = fptoui <2 x double> %in to <2 x i32>
+  ret <2 x i32> %val
+}
+
+define <2 x i16> @test_v2f64_to_signed_v2i16(<2 x double> %in) {
+; CHECK-LABEL: test_v2f64_to_signed_v2i16:
+; CHECK: fcvtzs.2d [[VAL64:v[0-9]+]], v0
+; CHECK: xtn.2s v0, [[VAL64]]
+
+  %val = fptosi <2 x double> %in to <2 x i16>
+  ret <2 x i16> %val
+}
+
+define <2 x i16> @test_v2f64_to_unsigned_v2i16(<2 x double> %in) {
+; CHECK-LABEL: test_v2f64_to_unsigned_v2i16:
+; CHECK: fcvtzs.2d [[VAL64:v[0-9]+]], v0
+; CHECK: xtn.2s v0, [[VAL64]]
+
+  %val = fptoui <2 x double> %in to <2 x i16>
+  ret <2 x i16> %val
+}
+
+define <2 x i8> @test_v2f64_to_signed_v2i8(<2 x double> %in) {
+; CHECK-LABEL: test_v2f64_to_signed_v2i8:
+; CHECK: fcvtzs.2d [[VAL64:v[0-9]+]], v0
+; CHECK: xtn.2s v0, [[VAL64]]
+
+  %val = fptosi <2 x double> %in to <2 x i8>
+  ret <2 x i8> %val
+}
+
+define <2 x i8> @test_v2f64_to_unsigned_v2i8(<2 x double> %in) {
+; CHECK-LABEL: test_v2f64_to_unsigned_v2i8:
+; CHECK: fcvtzs.2d [[VAL64:v[0-9]+]], v0
+; CHECK: xtn.2s v0, [[VAL64]]
+
+  %val = fptoui <2 x double> %in to <2 x i8>
+  ret <2 x i8> %val
+}