runtime: floating point number conversion to int different on silicon

[asdine]

What version of Go are you using (go version)?

$ go version
go version go1.16.6 darwin/arm64

Does this issue reproduce with the latest release?

Yes

What operating system and processor architecture are you using (go env)?

go env Output

$ go env
GO111MODULE=""
GOARCH="arm64"
GOBIN=""
GOCACHE="/Users/asdine/Library/Caches/go-build"
GOENV="/Users/asdine/Library/Application Support/go/env"
GOEXE=""
GOFLAGS=""
GOHOSTARCH="arm64"
GOHOSTOS="darwin"
GOINSECURE=""
GOMODCACHE="/Users/asdine/go/pkg/mod"
GONOPROXY=""
GONOSUMDB=""
GOOS="darwin"
GOPATH="/Users/asdine/go"
GOPRIVATE=""
GOPROXY="https://proxy.golang.org,direct"
GOROOT="/opt/homebrew/Cellar/go/1.16.6/libexec"
GOSUMDB="sum.golang.org"
GOTMPDIR=""
GOTOOLDIR="/opt/homebrew/Cellar/go/1.16.6/libexec/pkg/tool/darwin_arm64"
GOVCS=""
GOVERSION="go1.16.6"
GCCGO="gccgo"
AR="ar"
CC="clang"
CXX="clang++"
CGO_ENABLED="1"
GOMOD="/Users/asdine/code/github.com/genjidb/genji/go.mod"
CGO_CFLAGS="-g -O2"
CGO_CPPFLAGS=""
CGO_CXXFLAGS="-g -O2"
CGO_FFLAGS="-g -O2"
CGO_LDFLAGS="-g -O2"
PKG_CONFIG="pkg-config"
GOGCCFLAGS="-fPIC -arch arm64 -pthread -fno-caret-diagnostics -Qunused-arguments -fmessage-length=0 -fdebug-prefix-map=/var/folders/z3/f6p4jtl563s0z7gth5c_ydlc0000gn/T/go-build1754120130=/tmp/go-build -gno-record-gcc-switches -fno-common"

What did you do?

Calling math.Abs(math.MinInt64) then converting to an int64 doesn't overflow on mac silicon.

package main

import (
	"fmt"
	"math"
)

func main() {
	a := int64(math.MinInt64)

	b := math.Abs(math.MinInt64)

	c := math.Float64bits(b)

	d := int64(b)

	fmt.Println(a, b, c, d)
	// on x86 cpus:
	// -9223372036854775808 9.223372036854776e+18 4890909195324358656 -9223372036854775808

	// on silicon:
	// -9223372036854775808 9.223372036854776e+18 4890909195324358656 9223372036854775807
}

https://play.golang.org/p/xAYEJonoDAW

I'm not sure if this is a bug or something specific to the Silicon architecture.

What did you expect to see?

I was expecting the int64 to overflow and be negative. This is important to detect overflows and return a proper error when that happens.

What did you see instead?

The number doesn't overflow and stays positive

[randall77]

From the Go spec:

In all non-constant conversions involving floating-point or complex values, if the result type cannot represent the value the conversion succeeds but the result value is implementation-dependent.

That's the case here. You're converting 1<<63 to int64 and it doesn't quite fit.

[ianlancetaylor]

Closing, as this is not a bug.

To detect a representation failure when converting from float64 to in64, convert to int64 and then back again, and see if you get the same number.