math/rand: spinning tests on mips64?

[bradfitz]

The math/rand tests are failing on mips64:

// Float32 returns, as a float32, a pseudo-random number in [0.0,1.0).                                                                   
func (r *Rand) Float32() float32 {
        // Same rationale as in Float64: we want to preserve the Go 1 value                                                              
        // stream except we want to fix it not to return 1.0                                                                             
        // This only happens 1/2²⁴ of the time (plus the 1/2⁵³ of the time in Float64).                                                  
again:  
        f := float32(r.Float64())
        if f == 1 {
                goto again // resample; this branch is taken O(very rarely)                                                              
        }
        return f
}

goroutine 35 [runnable]:
math/rand.(*Rand).Float32(0x1051d78c, 0x3f0ddf90)
	/data/mips/go/src/math/rand/rand.go:173 +0x1c
math/rand.TestFloat32(0x10572360)
	/data/mips/go/src/math/rand/rand_test.go:343 +0xf8
testing.tRunner(0x10572360, 0x192d24)
	/data/mips/go/src/testing/testing.go:679 +0x150
created by testing.(*T).Run
	/data/mips/go/src/testing/testing.go:716 +0x284
FAIL	math/rand	180.061s

Is the float64 to float32 conversion broken?

/cc @cherrymui @crawshaw

[cherrymui]

Is the float64 to float32 conversion broken?

I don't think it's broken.

It seems on the builder the tests timeout at different points across runs, e.g. TestReadUniformity in https://build.golang.org/log/cc903f92d6a4660454d77a5f2c1267c9dd258dbc, so it's probably just a timeout. Does the machine have hardware floating point? If not, FP is emulated with kernel traps, which is extremely slow.

[bradfitz]

Ah, I see. I sent you https://go-review.googlesource.com/33326

[gopherbot]

CL https://golang.org/cl/33326 mentions this issue.