// CHECK-NEXT:    %[[ARG0]] = f32[] parameter(0)
// CHECK-NEXT:    %[[ARG1]] = f32[] parameter(1)
// CHECK-NEXT:    [[ADD:%.*]] = f32[] add(f32[] %[[ARG0]], f32[] %[[ARG1]])
// CHECK-NEXT:    ROOT %tuple.{{[0-9]+}} = (f32[]) tuple(f32[] [[ADD]])
// CHECK-NEXT:  }

// CHECK:       // InputMapping {0, 1}
// CHECK-NEXT:  // XlaInputShape f32[]
// CHECK-NEXT:  // XlaInputShape f32[]

# RUN: tf_tfl_translate -tf-input-arrays=input0,input1 -tf-input-shapes=4:4 -tf-input-data-types=DT_INT32,DT_INT32 -tf-output-arrays=Add %s -o - | flatbuffer_to_string - | FileCheck %s

# Add two tensor<4xi32> inputs and return the result

node {
  name: "Add"
  op: "Add"
  input: "input0"
  input: "input1"
  attr {
    key: "T"
    value {
      type: DT_INT32
    }
  }
}
node {
  name: "input0"
  op: "Placeholder"
  attr {

	"ROSBG",
	"RNSBGT",
	"RXSBGT",
	"ROSBGT",
	"RISBG",
	"RISBGN",
	"RISBGZ",
	"RISBGNZ",
	"RISBHG",
	"RISBLG",
	"RISBHGZ",
	"RISBLGZ",
	"FABS",
	"FADD",
	"FADDS",
	"FCMPO",
	"FCMPU",
	"CEBR",
	"FDIV",
	"FDIVS",
	"FMADD",
	"FMADDS",
	"FMOVD",
	"FMOVS",
	"FMSUB",
	"FMSUBS",
	"FMUL",
	"FMULS",
	"FNABS",
	"FNEG",
	"FNEGS",
	"LEDBR",
	"LDEBR",
	"LPDFR",

		return true
	case DIVD, DIVDCC, DIVDU, DIVDUCC, DIVDE, DIVDECC, DIVDEU, DIVDEUCC, DIVDO, DIVDOCC, DIVDUO, DIVDUOCC:
		return true
	case MODUD, MODSD, MODUW, MODSW:
		return true
	case FADD, FADDS, FSUB, FSUBS, FMUL, FMULS, FDIV, FDIVS, FMADD, FMADDS, FMSUB, FMSUBS, FNMADD, FNMADDS, FNMSUB, FNMSUBS, FMULSCC:
		return true
	case FADDCC, FADDSCC, FSUBCC, FMULCC, FDIVCC, FDIVSCC:
		return true

	C_ZOREG
	C_SOREG
	C_LOREG
	C_GOK
	C_ADDR
	C_TLS
	C_TEXTSIZE

	C_NCLASS /* must be the last */
)

const (
	AABSD = obj.ABaseMIPS + obj.A_ARCHSPECIFIC + iota
	AABSF
	AABSW
	AADD
	AADDD
	AADDF
	AADDU
	AADDW
	AAND
	ABEQ
	ABFPF
	ABFPT
	ABGEZ
	ABGEZAL
	ABGTZ
	ABLEZ
	ABLTZ
	ABLTZAL
	ABNE
	ABREAK
	ACLO
	ACLZ
	ACMOVF
	ACMOVN

	BYTE	$0x19
	BGE	L16
	BVS	L16
	WFCEDBS	V2, V2, V2
	BVS	LEXITTAGexp
	MOVD	$·expxaddexp<>+0(SB), R1
	FMOVD	72(R5), F6
	FMOVD	0(R1), F2
	WFMSDB	V0, V6, V2, V6
	FMOVD	64(R5), F4
	FADD	F6, F2
	FMOVD	56(R5), F1
	FMADD	F4, F2, F0
	FMOVD	48(R5), F3
	WFMADB	V2, V1, V0, V2
	FMOVD	40(R5), F1
	FMOVD	32(R5), F4
	FMUL	F0, F0
	WFMADB	V2, V4, V1, V4
	LGDR	F6, R1
	FMOVD	24(R5), F1
	WFMADB	V2, V3, V1, V3

	WFMDB   V4, V4, V6
	WFMADB  V4, V1, V7, V1
	WFMADB  V2, V3, V4, V2
	WFMADB  V1, V6, V5, V1
	FMOVD   56(R5), F3
	FMOVD   48(R5), F5
	WFMADB  V4, V5, V3, V4
	FMOVD   40(R5), F3
	FMADD   F1, F6, F4
	FMOVD   32(R5), F1
	FMADD   F3, F2, F1
	ANDW    $0xFFFFFF00, R1
	WFMADB  V6, V4, V1, V6
	FMOVD   24(R5), F3
	ORW $0x45000000, R1
	WFMADB  V2, V6, V3, V6
	VLVGF   $0, R1, V4
	LDEBR   F4, F4

	FMOVD	0(R1), F0
	FMOVD	F0, ret+8(FP)
	RET

L8:
	LDGR	R2, F2
	FSUB	F4, F3
	FMADD	F2, F4, F1
	MOVD	$·log1pc4<>+0(SB), R2
	WORD	$0xB3130041	// lcdbr %f4,%f1
	FMOVD	0(R2), F7
	FSUB	F3, F0
	MOVD	$·log1pc3<>+0(SB), R2
	FMOVD	0(R2), F3
	MOVD	$·log1pc2<>+0(SB), R2
	WFMDB	V1, V1, V6
	FMADD	F7, F4, F3
	WFMSDB	V0, V2, V1, V0
	FMOVD	0(R2), F7
	WFMADB	V4, V5, V20, V5
	MOVD	$·log1pc1<>+0(SB), R2

	// need to create a copy of the stack pointer that we can adjust.
	// We also need to do this if we are going to loop.
	if off < 0 || off > 4096-clearLoopCutoff || cnt > clearLoopCutoff {
		p = pp.Append(p, s390x.AADD, obj.TYPE_CONST, 0, off, obj.TYPE_REG, s390x.REGRT1, 0)
		p.Reg = int16(s390x.REGSP)
		reg = s390x.REGRT1
		off = 0
	}

	// Generate a loop of large clears.
	if cnt > clearLoopCutoff {

	FMOVD	24(R9), F1
	FMOVD	16(R9), F6
	MOVD	$·acoshtab2068<>+0(SB), R1
	WFMADB	V2, V1, V6, V2
	FMOVD	0(R3)(R1*1), F3
	WFMADB	V0, V4, V3, V0
	FMOVD	8(R9), F4
	FMADD	F4, F2, F0
	FMOVD	F0, ret+8(FP)
	RET
L10:
	FMOVD	F0, F8
	FMOVD	0(R9), F0
	FMADD	F8, F8, F0
	LTDBR	F0, F0
	FSQRT	F0, F10
L4:
	WFADB	V10, V8, V0
	WORD	$0xC0398006	//iilf	%r3,2147909631
	BYTE	$0x7F
	BYTE	$0xFF
	LGDR	F0, R5