Inconsistent overflow behavior between amd64 and arm64 on a float64->int64 conversion following math.Ceil

conversion.go

package main

import (
	"fmt"
	"math"
)

func main() {
	var f float64 = math.MaxInt64 + 1
	c := math.Ceil(f)
	fInt64 := int64(f)
	cInt64 := int64(c)
	fmt.Printf("f=%.1f math.Ceil(f)=%.1f int64(f)=%d int64(math.Ceil(f))=%d\n", f, c, fInt64, cInt64)
}

run.txt

❯ GOARCH=arm64 go run . && GOARCH=amd64 go run .
f=9223372036854775808.0 math.Ceil(f)=9223372036854775808.0 int64(f)=9223372036854775807 int64(math.Ceil(f))=9223372036854775807
f=9223372036854775808.0 math.Ceil(f)=9223372036854775808.0 int64(f)=9223372036854775807 int64(math.Ceil(f))=-9223372036854775808