// Args
      {"x: float", "y: resource"},
      // Return values
      {"z: float"},
      // Attr def
      {},
      // Nodes
      {
          {{"y0"}, "ReadVariableOp", {"y"}, {{"dtype", DT_FLOAT}}},
          {{"z"}, "Mul", {"x", "y0"}, {{"T", DT_FLOAT}}},
      });
}

class XlaKernelCreatorTest : public ::testing::Test {
 protected:
  void Init(const std::vector<FunctionDef>& flib) {

				x0 := natFromString(tt.x)
				y0 := natFromString(tt.y)
				want := natFromString(tt.z)
				x := nat(nil).set(x0)
				y := nat(nil).set(y0)
				var z nat
				switch mode {
				case "aliasX":
					z = x
				case "aliasY":
					z = y
				}
				z = z.subMod2N(x, y, tt.n)
				if z.cmp(want) != 0 {
					t.Fatalf("subMod2N(%d, %d, %d) = %d, want %d", x0, y0, tt.n, z, want)
				}

#endif

	// Compare 64-bytes per loop iteration.
	// Loop is unrolled and uses AVX2.
big_loop_avx2:
	VMOVDQU	(SI), Y2
	VMOVDQU	(DI), Y3
	VMOVDQU	32(SI), Y4
	VMOVDQU	32(DI), Y5
	VPCMPEQB Y2, Y3, Y0
	VPMOVMSKB Y0, AX
	XORL	$0xffffffff, AX
	JNE	diff32_avx2
	VPCMPEQB Y4, Y5, Y6
	VPMOVMSKB Y6, AX
	XORL	$0xffffffff, AX
	JNE	diff64_avx2

	ADDQ	$64, SI
	ADDQ	$64, DI
	SUBQ	$64, R8
	CMPQ	R8, $64

_33_to_63:
	LEAQ 1(DI)(DX*1), DX
	SUBQ AX, DX
	VMOVDQU -32(R8)(AX*1), Y0
	VMOVDQU (R8), Y1
loop33to63:
	VMOVDQU (DI), Y2
	VPCMPEQB Y1, Y2, Y3
	VPMOVMSKB Y3, SI
	CMPL  SI, $0xffffffff
	JE   partial_success33to63
	ADDQ $1,DI
	CMPQ DI,DX
	JB loop33to63
	JMP fail_avx2
partial_success33to63:
	VMOVDQU -32(AX)(DI*1), Y3
	VPCMPEQB Y0, Y3, Y4
	VPMOVMSKB Y4, SI
	CMPL  SI, $0xffffffff
	JE success_avx2

          range(batch_dims_count, contracting_dims_count)),
      dot.getPrecisionConfigAttr());
  return success();
}

// Convert:
//   %y0 = dot_general(%x0, %w)
//   %y1 = dot_general(%x1, %w)
//   ...
//   concatenate(%y0, %y1, ...)
// To:
//   %x = concatenate(%x0, %x1, ...)
//   dot_general(%x, %w)
LogicalResult LiftDotConcatLHS(mhlo::ConcatenateOp concat,

	if err != nil {
		t.Fatalf("error %s output %s", err, objout)
	}

	return objout
}

func TestVexEvexPCrelative(t *testing.T) {
	testenv.MustHaveGoBuild(t)
LOOP:
	for _, reg := range []string{"Y0", "Y8", "Z0", "Z8", "Z16", "Z24"} {
		asm := fmt.Sprintf(asmData, reg)
		objout := objdumpOutput(t, "pcrelative", asm)
		data := bytes.Split(objout, []byte("\n"))
		for idx := len(data) - 1; idx >= 0; idx-- {

// Mod implements Euclidean modulus (unlike Go); see [Int.DivMod] for more details.
func (z *Int) Mod(x, y *Int) *Int {
	y0 := y // save y
	if z == y || alias(z.abs, y.abs) {
		y0 = new(Int).Set(y)
	}
	var q Int
	q.QuoRem(x, y, z)
	if z.neg {
		if y0.neg {
			z.Sub(z, y0)
		} else {
			z.Add(z, y0)
		}
	}
	return z
}

// DivMod sets z to the quotient x div y and m to the modulus x mod y

00000000  16 03 01 01 0a 01 00 01  06 03 03 2d b6 ca ea 39  |...........-...9|
00000010  59 17 86 df 90 2f 73 e0  a0 5c 6e 28 09 78 69 d6  |Y..../s..\n(.xi.|
00000020  30 06 b7 7b 17 a9 79 30  2a d8 57 20 c5 5c ed 86  |0..{..y0*.W .\..|
00000030  15 f4 3b c8 d2 5f 7a 80  2a 6a cd 40 c2 da 6f a8  |..;.._z.*j.@..o.|
00000040  cd d7 e7 bf 48 bd fb a1  e9 4b 9b a9 00 04 00 2f  |....H....K...../|

//
// This function's execution time does not depend on the inputs.
func Mul64(x, y uint64) (hi, lo uint64) {
	const mask32 = 1<<32 - 1
	x0 := x & mask32
	x1 := x >> 32
	y0 := y & mask32
	y1 := y >> 32
	w0 := x0 * y0
	t := x1*y0 + w0>>32
	w1 := t & mask32
	w2 := t >> 32
	w1 += x0 * y1
	hi = x1*y1 + w2 + w1>>32
	lo = x * y
	return
}

// --- Full-width divide ---

		printed string
		reg0    int16
		reg1    int16
	}{
		{"[R10-R13]", REG_R10, REG_R13},
		{"[X0-AX]", REG_X0, REG_AX},

		{"[X0-X3]", REG_X0, REG_X3},
		{"[X21-X24]", REG_X21, REG_X24},

		{"[Y0-Y3]", REG_Y0, REG_Y3},
		{"[Y21-Y24]", REG_Y21, REG_Y24},

		{"[Z0-Z3]", REG_Z0, REG_Z3},
		{"[Z21-Z24]", REG_Z21, REG_Z24},
	}

	for _, test := range tests {