Good catch!
On 1/21/22, fft <f...@feedforward.com.cn> wrote: > Hi team's > I was troubled by a strange problem whole day,and at last,i found is about > the nuttx math lib function float frexpf(float x, int *exponent),when the > input x is postive, there is no problem,but when the input x is negative,the > result is wrong! I tested with inpu x=60 and x=-60: > x = 60; > float sig = frexpf(x, &e); > sig = 0.9375 > e = 6; > > that's right! > x = -60; > float sig = frexpf(x, &e); > sig = -0 > e = 2147483647; > that's wrong! > > I check online in stackoverflow,there's a recommended > design:https://stackoverflow.com/questions/40416682/portable-way-to-serialize-float-as-32-bit-integer > According to the recommended design, i change float frexpf(float x, int > *exponent) > > from: > float frexpf(float x, int *exponent) > { > *exponent = (int)ceilf(log2f(x)); > return x / ldexpf(1.0F, *exponent); > } > to: > float frexpf(float x, int *exponent) > { > *exponent = (int)ceilf(log2f(fabsf(x))); > return x / ldexpf(1.0F, *exponent); > } > and the problem is solved,i commit a pull request to solve the problem > ,please help me to check is right or not, thanks! > https://github.com/apache/incubator-nuttx/pull/5307 > > > Best regards, > Zou
