}{
	{"0s", 0},
	{"1ns", 1 * Nanosecond},
	{"1.1µs", 1100 * Nanosecond},
	{"2.2ms", 2200 * Microsecond},
	{"3.3s", 3300 * Millisecond},
	{"4m5s", 4*Minute + 5*Second},
	{"4m5.001s", 4*Minute + 5001*Millisecond},
	{"5h6m7.001s", 5*Hour + 6*Minute + 7001*Millisecond},
	{"8m0.000000001s", 8*Minute + 1*Nanosecond},
	{"2562047h47m16.854775807s", 1<<63 - 1},
	{"-2562047h47m16.854775808s", -1 << 63},
}

		{"panicCustomBool", `panic: main.MyBool(true)`},
		{"panicCustomComplex128", `panic: main.MyComplex128(+3.210000e+001+1.000000e+001i)`},
		{"panicCustomComplex64", `panic: main.MyComplex64(+1.100000e-001+3.000000e+000i)`},
		{"panicCustomFloat32", `panic: main.MyFloat32(-9.370000e+001)`},
		{"panicCustomFloat64", `panic: main.MyFloat64(-9.370000e+001)`},
		{"panicCustomInt", `panic: main.MyInt(93)`},
		{"panicCustomInt8", `panic: main.MyInt8(93)`},

      -> tensor<1x64x114x114xf32> {
  %tspos = stablehlo.transpose %arg0, dims = [0, 3, 1, 2]
      : (tensor<1x112x112x64xf32>) -> tensor<1x64x112x112xf32>
  %ret = stablehlo.pad %tspos, %padding_val,
      low = [0, 0, 1, 1], high = [0, 0, 1, 1], interior = [0, 0, 0, 0]
      : (tensor<1x64x112x112xf32>, tensor<f32>) -> tensor<1x64x114x114xf32>
  return %ret :tensor<1x64x114x114xf32>
}

// -----

  %0 = "tfl.pseudo_qconst"() {qtype = tensor<128x512x!quant.uniform<i8<-127:127>:f32, 0.01>>, value = dense<0> : tensor<128x512xi8>} : () -> tensor<128x512x!quant.uniform<i8<-127:127>:f32, 0.01>>
  %1 = "tfl.pseudo_qconst"() {qtype = tensor<128x!quant.uniform<i32:f32, 0.7>>, value = dense<0> : tensor<128xi32>} : () -> tensor<128x!quant.uniform<i32:f32, 0.7>>

	ct(reflect.TypeOf(int32(0)), -1, 0, 1),
	ct(reflect.TypeOf(int64(0)), -1, 0, 1),
	ct(reflect.TypeOf(uint(0)), 0, 1, 5),
	ct(reflect.TypeOf(uint8(0)), 0, 1, 5),
	ct(reflect.TypeOf(uint16(0)), 0, 1, 5),
	ct(reflect.TypeOf(uint32(0)), 0, 1, 5),
	ct(reflect.TypeOf(uint64(0)), 0, 1, 5),
	ct(reflect.TypeOf(uintptr(0)), 0, 1, 5),
	ct(reflect.TypeOf(string("")), "", "a", "ab"),

  %recurrent_stats = "quantfork.stats"(%recurrent_input) {layerStats = dense<[0.0, 1.0]> : tensor<2xf32>} : (tensor<1x3xf32>) -> tensor<1x3xf32>
  %cell_input = "tfl.pseudo_const"() {value = dense<1.000000e+00> : tensor<1x3xf32>} : () -> tensor<1x3xf32>
  %cell_stats = "quantfork.stats"(%cell_input) {layerStats = dense<[0.0, 1.0]> : tensor<2xf32>} : (tensor<1x3xf32>) -> tensor<1x3xf32>
  %16 = "tfl.unidirectional_sequence_lstm"(

    %1:2 = func.call @while_body_add2(%arg0, %arg1) : (tensor<i32>, tensor<i32>) -> (tensor<i32>, tensor<i32>)
    func.return %0#0, %1#0 : tensor<i32>, tensor<i32>
  }

  // CHECK:     func @"[[while_func_prefix]]/tfrt_body_1"
  // CHECK:     "tf.AddV2"
  // CHECK-NOT: "tf.Div"

video-001.restart2.jpeg...

00000080  01 00 00 7b 00 0b 00 02  01 00 ff 01 00 01 00 00  |...{............|
00000090  17 00 00 00 12 00 00 00  05 00 05 01 00 00 00 00  |................|
000000a0  00 0a 00 06 00 04 00 1d  00 17 00 0d 00 1a 00 18  |................|
000000b0  08 04 04 03 08 07 08 05  08 06 04 01 05 01 06 01  |................|
000000c0  05 03 06 03 02 01 02 03  00 2b 00 09 08 03 04 03  |.........+......|

  func.return %0#0, %1#0 : tensor<8x8x8x8xf32>, tensor<8x8x8x8xf32>
}
// CHECK: func @customAggregator