{
    Output tr0 = ops::Transpose(on_gpu.WithOpName("tr0"), in0, perm);
    Output tr1 = ops::Transpose(on_gpu.WithOpName("tr1"), in1, perm);
  }

  std::unique_ptr<Graph> result;
  TF_ASSERT_OK(CloneConstantsForBetterClustering(root, &result));

  OutputTensor tr0_perm;
  TF_ASSERT_OK(FindNodeByName(result.get(), "tr0")->input_tensor(1, &tr0_perm));

  OutputTensor tr1_perm;

	// long branches inserted by the linker where supported.
	// Multiple .s files are generated instead of one.
	instOnArch := map[string]string{
		"ppc64":   "\tMOVD\tR0,R3\n",
		"ppc64le": "\tMOVD\tR0,R3\n",
		"arm":     "\tMOVW\tR0,R1\n",
	}
	inst := instOnArch[buildcfg.GOARCH]
	for j := 0; j < FN; j++ {
		testname := fmt.Sprintf("bigfn%d", j)
		fmt.Fprintf(&w, "TEXT ·%s(SB),$0\n", testname)

	// R0: zero
	// R1: ptr to memory to be zeroed
	// R1 is updated as a side effect.
	fmt.Fprintln(w, "TEXT runtime·duffzero(SB), NOSPLIT, $0-0")
	for i := 0; i < 128; i++ {
		fmt.Fprintln(w, "\tMOVW.P\tR0, 4(R1)")
	}
	fmt.Fprintln(w, "\tRET")
}

func copyARM(w io.Writer) {
	// R0: scratch space
	// R1: ptr to source memory
	// R2: ptr to destination memory
	// R1 and R2 are updated as a side effect

	DR4:  "DR4",
	DR5:  "DR5",
	DR6:  "DR6",
	DR7:  "DR7",
	DR8:  "DR8",
	DR9:  "DR9",
	DR10: "DR10",
	DR11: "DR11",
	DR12: "DR12",
	DR13: "DR13",
	DR14: "DR14",
	DR15: "DR15",
	TR0:  "TR0",
	TR1:  "TR1",
	TR2:  "TR2",
	TR3:  "TR3",
	TR4:  "TR4",
	TR5:  "TR5",
	TR6:  "TR6",
	TR7:  "TR7",

	"CR7",
	"CR8",
	"CR9",
	"CR10",
	"CR11",
	"CR12",
	"CR13",
	"CR14",
	"CR15",
	"DR0", // [D_DR]
	"DR1",
	"DR2",
	"DR3",
	"DR4",
	"DR5",
	"DR6",
	"DR7",
	"TR0", // [D_TR]
	"TR1",
	"TR2",
	"TR3",
	"TR4",
	"TR5",
	"TR6",
	"TR7",
	"TLS",    // [D_TLS]
	"MAXREG", // [MAXREG]
}

func init() {