Re: GNU Guile 2.1.6 released (beta) [numbers.c]

[Matt Wette <matt.wette@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2296 bytes --]


> On Jan 19, 2017, at 7:01 PM, Matt Wette <matt.wette@gmail.com> wrote:
> 
> 
>> On Jan 18, 2017, at 6:26 PM, Andy Wingo <wingo@pobox.com <mailto:wingo@pobox.com>> wrote:
>> 
>> We are pleased to announce GNU Guile release 2.1.6.
>> 
>> Guile 2.1.6 is the sixth pre-release in what will eventually become the
>> 2.2 release series.  We encourage you to test this release and provide
>> feedback to guile-devel@gnu.org <mailto:guile-devel@gnu.org>.
> 
> Saw this one last round.  Mac OS, now gcc-6.3.0:
> 
> ;;; ("#i1@-0" 1.0 -0.0)
> FAIL: numbers.test: string->number: valid complex number strings
> 
> I am going to see if I can generate the assembly.

Short story: scm_c_make_polar is broken for the Mac.  Guile needs to decide if it want to use __sincos() on Mac, or suppress optimization, or ...

I was able to get the above to work (guile-2.1.5) by using
 SCM
+#ifdef __APPLE__
+__attribute__((optimize("O0")))
+#endif
 scm_c_make_polar (double mag, double ang)


In scm_c_make_polar, “gcc -O2”  turns sin(), cos() into cexp(), since cexp(i*x) = cos(x) + i*sin(x):

gcc -O0 =>
LM4339:
	movq	-32(%rbp), %rax
	movd	%rax, %xmm0
	call	_sin
	movd	%xmm0, %rax
	movq	%rax, -8(%rbp)
LM4340:
	movq	-32(%rbp), %rax
	movd	%rax, %xmm0
	call	_cos
	movd	%xmm0, %rax
	movq	%rax, -16(%rbp)

gcc -O2 =>
	pxor	%xmm0, %xmm0
LVL2703:
	call	_cexp

I wrote a little C program to show that cexp() does not preserve the zero-signed-ness:

 cos,sin: +1.000000 -0.000000
__sincos: +1.000000 -0.000000
    cexp: +1.000000 +0.000000

The scm_c_make_polar will use sincos() if available, but macOS does not have sincos(), it has __sincos().

#include <stdio.h>
#include <stdint.h>
#include <math.h>
#include <complex.h>

extern double z, p, n;

int main() {
  double complex c;
  double d, sine, cosine;

  d = z*n;
  printf(" cos,sin: %+f %+f\n", cos(d), sin(d));
  __sincos(d, &sine, &cosine);
  printf("__sincos: %+f %+f\n", cosine, sine);
  c = cexp(CMPLX(0.0, d));
  printf("    cexp: %+f %+f\n", creal(c), cimag(c));
}

double z = 0.0, p = +1.0, n = -1.0;

Incidentally, the above program will not compile on my machine w/ gcc-6.3.0.  “gcc -std=c11” or “gcc -std=c99” will not recognize the standard macro CMPLX().




[-- Attachment #2: Type: text/html, Size: 13644 bytes --]