Implemented Neon scalar vdup_lane intrinsics.

Fixed scalar dup alias and added test case.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195330 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 129 insertions, 2 deletions

diff --git a/lib/Target/AArch64/AArch64InstrNEON.td b/lib/Target/AArch64/AArch64InstrNEON.td
index 5b6168eb08..c0c572a62e 100644
--- a/lib/Target/AArch64/AArch64InstrNEON.td
+++ b/lib/Target/AArch64/AArch64InstrNEON.td
@@ -5883,16 +5883,37 @@ defm : NeonI_SDUP<Neon_low2D, Neon_High2D, v1i64, v2i64>;
 defm : NeonI_SDUP<Neon_low4f, Neon_High4f, v2f32, v4f32>;
 defm : NeonI_SDUP<Neon_low2d, Neon_High2d, v1f64, v2f64>;
 
-// Patterns  for vector extract of FP data using scalar DUP instructions
+// Patterns for vector extract of FP data using scalar DUP instructions
 defm : NeonI_Scalar_DUP_Elt_pattern<DUPsv_S, f32,
   v4f32, neon_uimm2_bare, v2f32, v4f32, neon_uimm1_bare>;
 defm : NeonI_Scalar_DUP_Elt_pattern<DUPdv_D, f64,
   v2f64, neon_uimm1_bare, v1f64, v2f64, neon_uimm0_bare>;
 
+multiclass NeonI_Scalar_DUP_Vec_pattern<Instruction DUPI,
+  ValueType ResTy, ValueType OpTy,Operand OpLImm,
+  ValueType NOpTy, ValueType ExTy, Operand OpNImm> {
+
+  def : Pat<(ResTy (extract_subvector (OpTy VPR128:$Rn), OpLImm:$Imm)),
+            (ResTy (DUPI VPR128:$Rn, OpLImm:$Imm))>;
+
+  def : Pat<(ResTy (extract_subvector (NOpTy VPR64:$Rn), OpNImm:$Imm)),
+            (ResTy (DUPI
+              (ExTy (SUBREG_TO_REG (i64 0), VPR64:$Rn, sub_64)),
+                OpNImm:$Imm))>;
+}
+// Patterns for extract subvectors of v1ix data using scalar DUP instructions
+defm : NeonI_Scalar_DUP_Vec_pattern<DUPbv_B,
+  v1i8, v16i8, neon_uimm4_bare, v8i8, v16i8, neon_uimm3_bare>;
+defm : NeonI_Scalar_DUP_Vec_pattern<DUPhv_H,
+  v1i16, v8i16, neon_uimm3_bare, v4i16, v8i16, neon_uimm2_bare>;
+defm : NeonI_Scalar_DUP_Vec_pattern<DUPsv_S,
+  v1i32, v4i32, neon_uimm2_bare, v2i32, v4i32, neon_uimm1_bare>;
+
+
 multiclass NeonI_Scalar_DUP_alias<string asmop, string asmlane,
                                   Instruction DUPI, Operand OpImm,
                                   RegisterClass ResRC> {
-  def : NeonInstAlias<!strconcat(asmop, "$Rd, $Rn." # asmlane # "[$Imm]"),
+  def : NeonInstAlias<!strconcat(asmop, "$Rd, $Rn" # asmlane # "[$Imm]"),
           (DUPI ResRC:$Rd, VPR128:$Rn, OpImm:$Imm), 0b0>;
 }
 
diff --git a/test/CodeGen/AArch64/neon-scalar-copy.ll b/test/CodeGen/AArch64/neon-scalar-copy.ll
new file mode 100644
index 0000000000..59f62374d4
--- /dev/null
+++ b/test/CodeGen/AArch64/neon-scalar-copy.ll
@@ -0,0 +1,80 @@
+; RUN: llc -mtriple=aarch64-none-linux-gnu -mattr=+neon < %s | FileCheck %s
+
+define float @test_dup_sv2S(<2 x float> %v) {
+ ;CHECK: test_dup_sv2S
+ ;CHECK: dup {{s[0-31]+}}, {{v[0-31]+}}.s[1]
+ %tmp1 = extractelement <2 x float> %v, i32 1
+ ret float  %tmp1
+}
+
+define float @test_dup_sv4S(<4 x float> %v) {
+ ;CHECK: test_dup_sv4S
+ ;CHECK: dup {{s[0-31]+}}, {{v[0-31]+}}.s[0]
+ %tmp1 = extractelement <4 x float> %v, i32 0
+ ret float  %tmp1
+}
+
+define double @test_dup_dvD(<1 x double> %v) {
+ ;CHECK: test_dup_dvD
+ ;CHECK-NOT: dup {{d[0-31]+}}, {{v[0-31]+}}.d[0]
+ ;CHECK: ret
+ %tmp1 = extractelement <1 x double> %v, i32 0
+ ret double  %tmp1
+}
+
+define double @test_dup_dv2D(<2 x double> %v) {
+ ;CHECK: test_dup_dv2D
+ ;CHECK: dup {{d[0-31]+}}, {{v[0-31]+}}.d[1]
+ %tmp1 = extractelement <2 x double> %v, i32 1
+ ret double  %tmp1
+}
+
+define <1 x i8> @test_vector_dup_bv16B(<16 x i8> %v1) {
+ ;CHECK: test_vector_dup_bv16B
+ ;CHECK: dup {{b[0-31]+}}, {{v[0-31]+}}.b[14]
+ %shuffle.i = shufflevector <16 x i8> %v1, <16 x i8> undef, <1 x i32> <i32 14> 
+ ret <1 x i8> %shuffle.i
+}
+
+define <1 x i8> @test_vector_dup_bv8B(<8 x i8> %v1) {
+ ;CHECK: test_vector_dup_bv8B
+ ;CHECK: dup {{b[0-31]+}}, {{v[0-31]+}}.b[7]
+ %shuffle.i = shufflevector <8 x i8> %v1, <8 x i8> undef, <1 x i32> <i32 7> 
+ ret <1 x i8> %shuffle.i
+}
+
+define <1 x i16> @test_vector_dup_hv8H(<8 x i16> %v1) {
+ ;CHECK: test_vector_dup_hv8H
+ ;CHECK: dup {{h[0-31]+}}, {{v[0-31]+}}.h[7]
+ %shuffle.i = shufflevector <8 x i16> %v1, <8 x i16> undef, <1 x i32> <i32 7> 
+ ret <1 x i16> %shuffle.i
+}
+
+define <1 x i16> @test_vector_dup_hv4H(<4 x i16> %v1) {
+ ;CHECK: test_vector_dup_hv4H
+ ;CHECK: dup {{h[0-31]+}}, {{v[0-31]+}}.h[3]
+ %shuffle.i = shufflevector <4 x i16> %v1, <4 x i16> undef, <1 x i32> <i32 3> 
+ ret <1 x i16> %shuffle.i
+}
+
+define <1 x i32> @test_vector_dup_sv4S(<4 x i32> %v1) {
+ ;CHECK: test_vector_dup_sv4S
+ ;CHECK: dup {{s[0-31]+}}, {{v[0-31]+}}.s[3]
+ %shuffle = shufflevector <4 x i32> %v1, <4 x i32> undef, <1 x i32> <i32 3> 
+ ret <1 x i32> %shuffle
+}
+
+define <1 x i32> @test_vector_dup_sv2S(<2 x i32> %v1) {
+ ;CHECK: test_vector_dup_sv2S
+ ;CHECK: dup {{s[0-31]+}}, {{v[0-31]+}}.s[1]
+ %shuffle = shufflevector <2 x i32> %v1, <2 x i32> undef, <1 x i32> <i32 1> 
+ ret <1 x i32> %shuffle
+}
+
+define <1 x i64> @test_vector_dup_dv2D(<2 x i64> %v1) {
+ ;CHECK: test_vector_dup_dv2D
+ ;CHECK: dup {{d[0-31]+}}, {{v[0-31]+}}.d[1]
+ %shuffle.i = shufflevector <2 x i64> %v1, <2 x i64> undef, <1 x i32> <i32 1> 
+ ret <1 x i64> %shuffle.i
+}
+
diff --git a/test/MC/AArch64/neon-scalar-dup.s b/test/MC/AArch64/neon-scalar-dup.s
index 64366f2edc..77c638df09 100644
--- a/test/MC/AArch64/neon-scalar-dup.s
+++ b/test/MC/AArch64/neon-scalar-dup.s
@@ -27,3 +27,29 @@
 // CHECK: dup d3, v5.d[0]       // encoding: [0xa3,0x04,0x08,0x5e]
 // CHECK: dup d6, v5.d[1]       // encoding: [0xa6,0x04,0x18,0x5e]
 
+//------------------------------------------------------------------------------
+// Aliases for Duplicate element (scalar)
+//------------------------------------------------------------------------------
+         mov b0, v0.b[15]
+         mov b1, v0.b[7]
+         mov b17, v0.b[0]
+         mov h5, v31.h[7]
+         mov h9, v1.h[4]
+         mov h11, v17.h[0]
+         mov s2, v2.s[3]
+         mov s4, v21.s[0]
+         mov s31, v21.s[2]
+         mov d3, v5.d[0]
+         mov d6, v5.d[1]
+
+// CHECK: dup b0, v0.b[15]      // encoding: [0x00,0x04,0x1f,0x5e]
+// CHECK: dup b1, v0.b[7]       // encoding: [0x01,0x04,0x0f,0x5e]
+// CHECK: dup b17, v0.b[0]      // encoding: [0x11,0x04,0x01,0x5e]
+// CHECK: dup h5, v31.h[7]      // encoding: [0xe5,0x07,0x1e,0x5e]
+// CHECK: dup h9, v1.h[4]       // encoding: [0x29,0x04,0x12,0x5e]
+// CHECK: dup h11, v17.h[0]     // encoding: [0x2b,0x06,0x02,0x5e]
+// CHECK: dup s2, v2.s[3]       // encoding: [0x42,0x04,0x1c,0x5e]
+// CHECK: dup s4, v21.s[0]      // encoding: [0xa4,0x06,0x04,0x5e]
+// CHECK: dup s31, v21.s[2]     // encoding: [0xbf,0x06,0x14,0x5e]
+// CHECK: dup d3, v5.d[0]       // encoding: [0xa3,0x04,0x08,0x5e]
+// CHECK: dup d6, v5.d[1]       // encoding: [0xa6,0x04,0x18,0x5e]