VCVTR_LT_S32_F32:  "VCVTR.LT.S32.F32",
	VCVTR_GT_S32_F32:  "VCVTR.GT.S32.F32",
	VCVTR_LE_S32_F32:  "VCVTR.LE.S32.F32",
	VCVTR_S32_F32:     "VCVTR.S32.F32",
	VCVTR_ZZ_S32_F32:  "VCVTR.ZZ.S32.F32",
	VCVTR_EQ_S32_F64:  "VCVTR.EQ.S32.F64",
	VCVTR_NE_S32_F64:  "VCVTR.NE.S32.F64",
	VCVTR_CS_S32_F64:  "VCVTR.CS.S32.F64",
	VCVTR_CC_S32_F64:  "VCVTR.CC.S32.F64",
	VCVTR_MI_S32_F64:  "VCVTR.MI.S32.F64",

const CacheLinePadSize = 128

func doinit() {
	options = []option{
		{Name: "aes", Feature: &ARM64.HasAES},
		{Name: "pmull", Feature: &ARM64.HasPMULL},
		{Name: "sha1", Feature: &ARM64.HasSHA1},
		{Name: "sha2", Feature: &ARM64.HasSHA2},
		{Name: "sha512", Feature: &ARM64.HasSHA512},
		{Name: "crc32", Feature: &ARM64.HasCRC32},
		{Name: "atomics", Feature: &ARM64.HasATOMICS},
		{Name: "cpuid", Feature: &ARM64.HasCPUID},

  CHECK_OK(pjrt_client.status());
  auto c_api_client = down_cast<xla::PjRtCApiClient*>(pjrt_client->get());
  std::vector<int32_t> data(1, 0);
  xla::Shape shape = xla::ShapeUtil::MakeShape(xla::S32, {1});

  auto buffer = c_api_client->pjrt_c_client()->client->BufferFromHostBuffer(
      data.data(), shape.element_type(), shape.dimensions(),
      /*byte_strides=*/std::nullopt,

	u16, v16, w16 uint16  = 0, 0, 1
	u32, v32, w32 uint32  = 0, 0, 1
	u64, v64, w64 uint64  = 0, 0, 1
	up, vp, wp    uintptr = 0, 0, 1

	f, g, h                         float64 = 0, 0, 1
	f32, g32, h32                   float32 = 0, 0, 1
	f64, g64, h64, inf, negInf, nan float64 = 0, 0, 1, math.Inf(1), math.Inf(-1), math.NaN()

	c, d, e          complex128 = 0 + 0i, 0 + 0i, 1 + 1i

	case ACRC32B:
		return S32 | OPDP2(16)

	case ACRC32H:
		return S32 | OPDP2(17)

	case ACRC32W:
		return S32 | OPDP2(18)

	case ACRC32X:
		return S64 | OPDP2(19)

	case ACRC32CB:
		return S32 | OPDP2(20)

	case ACRC32CH:
		return S32 | OPDP2(21)

	case ACRC32CW:
		return S32 | OPDP2(22)

	case ACRC32CX:
		return S64 | OPDP2(23)

	hwcap2_SHA1  = 1 << 2
	hwcap2_SHA2  = 1 << 3
	hwcap2_CRC32 = 1 << 4
)

func initOptions() {
	options = []option{
		{Name: "pmull", Feature: &ARM.HasPMULL},
		{Name: "sha1", Feature: &ARM.HasSHA1},
		{Name: "sha2", Feature: &ARM.HasSHA2},
		{Name: "swp", Feature: &ARM.HasSWP},
		{Name: "thumb", Feature: &ARM.HasTHUMB},
		{Name: "thumbee", Feature: &ARM.HasTHUMBEE},
		{Name: "tls", Feature: &ARM.HasTLS},

	v := true
	s := strconv.FormatBool(v)
	fmt.Printf("%T, %v\n", s, s)

	// Output:
	// string, true
}

func ExampleFormatFloat() {
	v := 3.1415926535

	s32 := strconv.FormatFloat(v, 'E', -1, 32)
	fmt.Printf("%T, %v\n", s32, s32)

	s64 := strconv.FormatFloat(v, 'E', -1, 64)
	fmt.Printf("%T, %v\n", s64, s64)

	// fmt.Println uses these arguments to print floats

  TF_ASSERT_OK_AND_ASSIGN(auto pjrt_client, GetPjRtClient(DEVICE_GPU));
  std::vector<int32_t> data{1, 2, 3, 4, 5, 6};
  xla::Shape xla_shape = xla::ShapeUtil::MakeShape(xla::S32, dimensions);
  TF_ASSERT_OK_AND_ASSIGN(
      auto pjrt_buffer,
      pjrt_client->BufferFromHostBuffer(
          data.data(), xla_shape.element_type(), xla_shape.dimensions(),
          /*byte_strides=*/std::nullopt,

		{Name: "jscvt", Feature: &ARM64.HasJSCVT},
		{Name: "lrcpc", Feature: &ARM64.HasLRCPC},
		{Name: "pmull", Feature: &ARM64.HasPMULL},
		{Name: "sha1", Feature: &ARM64.HasSHA1},
		{Name: "sha2", Feature: &ARM64.HasSHA2},
		{Name: "sha3", Feature: &ARM64.HasSHA3},
		{Name: "sha512", Feature: &ARM64.HasSHA512},
		{Name: "sm3", Feature: &ARM64.HasSM3},
		{Name: "sm4", Feature: &ARM64.HasSM4},

	MOVL	AX, ret+12(FP)
	RET

TEXT runtime·usleep(SB),NOSPLIT,$24
	MOVL	$0, DX
	MOVL	usec+0(FP), AX
	MOVL	$1000000, CX
	DIVL	CX
	MOVL	AX, 12(SP)		// tv_sec - l32
	MOVL	$0, 16(SP)		// tv_sec - h32
	MOVL	$1000, AX
	MULL	DX
	MOVL	AX, 20(SP)		// tv_nsec

	MOVL	$0, 0(SP)
	LEAL	12(SP), AX
	MOVL	AX, 4(SP)		// arg 1 - rqtp
	MOVL	$0, 8(SP)		// arg 2 - rmtp
	MOVL	$SYS___nanosleep50, AX
	INT	$0x80