package main

import "fmt"

func f0() string {
	const f = 3.141592;
	return fmt.Sprintf("%v", float64(f));
}


func f1() string {
	const f = 3.141592;
	x := float64(float32(f));  // appears to change the precision of f
	_ = x;
	return fmt.Sprintf("%v", float64(f));
}


func main() {
	r0 := f0();
	r1 := f1();
	if r0 != r1 {
		println("r0 =", r0);
		println("r1 =", r1);

	T12 any
	T13 interface {
		m1()
		m2(int) float32
	}
	T14 interface {
		T12
		T13
		m3(x ...struct{}) []T9
	}
	T15 func()
	T16 func(int)
	T17 func(x int)
	T18 func() float32
	T19 func() (x float32)
	T20 func(...any)
	T21 struct{ next *T21 }
	T22 struct{ link *T23 }
	T23 struct{ link *T22 }
	T24 *T24
	T25 *T26
	T26 *T27
	T27 *T25
	T28 func(T28) T28
)

var (
	V0 int

	// Use exact fmt formatting if in float64 range (common case):
	// proceed if f doesn't underflow to 0 or overflow to inf.
	if x, _ := f.Float64(); f.Sign() == 0 == (x == 0) && !math.IsInf(x, 0) {
		return fmt.Sprintf("%.6g", x)
	}

	return "OOPS"
}

func (x complexVal) String() string { return fmt.Sprintf("(%s + %si)", x.re, x.im) }

func newFloat() *big.Float { return new(big.Float).SetPrec(prec) }

//go:noinline

(Load <t> ptr mem) && t.IsComplex() && t.Size() == 8 =>
  (ComplexMake
    (Load <typ.Float32> ptr mem)
    (Load <typ.Float32>
      (OffPtr <typ.Float32Ptr> [4] ptr)
      mem)
    )
(Store {t} dst (ComplexMake real imag) mem) && t.Size() == 8 =>
  (Store {typ.Float32}
    (OffPtr <typ.Float32Ptr> [4] dst)
    imag
    (Store {typ.Float32} dst real mem))
(Load <t> ptr mem) && t.IsComplex() && t.Size() == 16 =>
  (ComplexMake

		case reflect.Float32, reflect.Float64:
			return i.equalsNumber(reflect.ValueOf(expected).Float(), path)
		case reflect.Int, reflect.Int8, reflect.Int32, reflect.Int64:
			return i.equalsNumber(float64(reflect.ValueOf(expected).Int()), path)
		case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64:
			return i.equalsNumber(float64(reflect.ValueOf(expected).Uint()), path)

pkg log/slog, func SetLogLoggerLevel(Level) Level #62418
pkg math/big, method (*Rat) FloatPrec() (int, bool) #50489
pkg math/rand/v2, func ExpFloat64() float64 #61716
pkg math/rand/v2, func Float32() float32 #61716
pkg math/rand/v2, func Float64() float64 #61716
pkg math/rand/v2, func Int() int #61716
pkg math/rand/v2, func Int32() int32 #61716
pkg math/rand/v2, func Int32N(int32) int32 #61716

				vars[idx] = escaper + "<binary>" + escaper
			}
		case int, int8, int16, int32, int64, uint, uint8, uint16, uint32, uint64:
			vars[idx] = utils.ToString(v)
		case float32:
			vars[idx] = strconv.FormatFloat(float64(v), 'f', -1, 32)
		case float64:
			vars[idx] = strconv.FormatFloat(v, 'f', -1, 64)
		case string:
			vars[idx] = escaper + strings.ReplaceAll(v, escaper, escaper+escaper) + escaper
		default:

//
// The `input_type`, `output_type` and `inference_type` can be float32 / qint8 /
// int8 / int16.
//
// Returns a partially quantized model if `fully_quantize` is false. Returns a
// non-OK status if the quantization fails.
//
// When `verify_numeric` is true, the model will have it's original float ops
// and NumericVerify ops to compare output values from the quantized and float
// ops.
//

// It would likely be in a constraints package in the standard library.
type Numeric interface {
	~int|~int8|~int16|~int32|~int64|
		~uint|~uint8|~uint16|~uint32|~uint64|~uintptr|
		~float32|~float64|
		~complex64|~complex128
}

func DotProduct[T Numeric](s1, s2 []T) T {
	if len(s1) != len(s2) {
		panic("DotProduct: slices of unequal length")
	}
	var r T
	for i := range s1 {
		r += s1[i] * s2[i]

  with strategy.scope():
    # Create an mnist float model with the specified float state.
    model = FloatModel()
    optimizer = tf.keras.optimizers.Adam(learning_rate=learning_rate)

  def train_step(features):
    inputs = tf.image.convert_image_dtype(
        features['image'], dtype=tf.float32, saturate=False)
    labels = tf.one_hot(features['label'], num_classes)