[Thanks to all those who responded to my previous problem-report wrt to the IEEE inexact-flag on the Alpha. We have filed a bug report for that one.]
Hi there,
we keep finding problems on the Alpha, and we are unsure about what is going on. I anticipate that the present problem does not seem to be a compiler issue (except that the -mieee-with-inexact option promises full compliance with the IEEE floating point standard). However, I hope some Alpha expert can advise about how to investigate the issue further.
Here are the details:
$ cat bug.c #include <stdio.h>
float f = 1.4e-45; double d = 1e-300;
int main() { if (1.4e-45f > 1e-300) printf("compile-time test says 1.4e-45f > 1e-300\n"); if (f < d) printf(" run-time test says 1.4e-45f < 1e-300\n"); return 0; }
$ gcc -mieee-with-inexact bug.c
$ ./a.out compile-time test says 1.4e-45f > 1e-300 run-time test says 1.4e-45f < 1e-300
$ gcc -v Using built-in specs. Target: alpha-linux-gnu Configured with: ../src/configure -v --enable-languages=c,c++,fortran,objc,obj-c++ --prefix=/usr --enable-shared --with-system-zlib --libexecdir=/usr/lib --without-included-gettext --enable-threads=posix --enable-nls --with-gxx-include-dir=/usr/include/c++/4.2 --program-suffix=-4.2 --enable-clocale=gnu --enable-libstdcxx-debug --enable-objc-gc --enable-mpfr --disable-libssp --with-long-double-128 --enable-checking=release --build=alpha-linux-gnu --host=alpha-linux-gnu --target=alpha-linux-gnu Thread model: posix gcc version 4.2.4 (Debian 4.2.4-3)
$ cat /proc/cpuinfo cpu : Alpha cpu model : EV56 cpu variation : 7 cpu revision : 0 cpu serial number : system type : Rawhide system variation : Tincup system revision : 0 system serial number : AY74642662 cycle frequency [Hz] : 399642346 est. timer frequency [Hz] : 1200.00 page size [bytes] : 8192 phys. address bits : 40 max. addr. space # : 127 BogoMIPS : 705.16 kernel unaligned acc : 0 (pc=0,va=0) user unaligned acc : 35 (pc=20000074c18,va=87) platform string : AlphaServer 1200 5/400 4MB cpus detected : 1 cpus active : 1 cpu active mask : 0000000000000001 L1 Icache : 8K, 1-way, 32b line L1 Dcache : 8K, 1-way, 32b line L2 cache : 96K, 3-way, 64b line L3 cache : 4096K, 1-way, 64b line
All the best,
Roberto
Roberto Bagnara ha scritto:
[Thanks to all those who responded to my previous problem-report wrt to the IEEE inexact-flag on the Alpha. We have filed a bug report for that one.]
Hi there,
we keep finding problems on the Alpha, and we are unsure about what is going on. I anticipate that the present problem does not seem to be a compiler issue (except that the -mieee-with-inexact option promises full compliance with the IEEE floating point standard). However, I hope some Alpha expert can advise about how to investigate the issue further.
Here are the details:
$ cat bug.c #include <stdio.h>
float f = 1.4e-45; double d = 1e-300;
int main() { if (1.4e-45f > 1e-300) printf("compile-time test says 1.4e-45f > 1e-300\n"); if (f < d) printf(" run-time test says 1.4e-45f < 1e-300\n"); return 0; }
The problem come from the fact that for astonishing reasons 1.4e-45f is not seen as 1.4e-45.
So, adding
printf("%g\n", f);
before
return 0;
we get
$ ./a.out compile-time test says 1.4e-45f > 1e-300 run-time test says 1.4e-45f < 1e-300 2.65249e-315
Note that 2.65249e-315 is not representable in ieee754/iec559 32 bit floating point.
32 bit floats are emulated on Alpha? There is an huge bug in FP emulation?
On Mon, Sep 22, 2008 at 09:56:30PM +0200, Abramo Bagnara wrote:
The problem come from the fact that for astonishing reasons 1.4e-45f is not seen as 1.4e-45.
Why is this astonishing? The smallest positive single-precision IEEE floating point number is roughly 1.175494e-38. Since user-specified numbers are rounded to the closest representable value, 1.4e-45f is a fancy way of typing zero.
Double-precision values have much greater dynamic range; DBL_MIN is roughly 2.225074e-308.
On Mon, 22 Sep 2008, Joe Buck wrote:
On Mon, Sep 22, 2008 at 09:56:30PM +0200, Abramo Bagnara wrote:
The problem come from the fact that for astonishing reasons 1.4e-45f is not seen as 1.4e-45.
Why is this astonishing? The smallest positive single-precision IEEE floating point number is roughly 1.175494e-38. Since user-specified numbers are rounded to the closest representable value, 1.4e-45f is a fancy way of typing zero.
The smallest subnormal value is approximately 1.40129846e-45F (__FLT_DENORM_MIN__). The obvious conclusion is that subnormal support is broken or disabled on Alpha by default.
On Mon, Sep 22, 2008 at 09:41:05PM +0000, Joseph S. Myers wrote:
On Mon, 22 Sep 2008, Joe Buck wrote:
On Mon, Sep 22, 2008 at 09:56:30PM +0200, Abramo Bagnara wrote:
The problem come from the fact that for astonishing reasons 1.4e-45f is not seen as 1.4e-45.
Why is this astonishing? The smallest positive single-precision IEEE floating point number is roughly 1.175494e-38. Since user-specified numbers are rounded to the closest representable value, 1.4e-45f is a fancy way of typing zero.
The smallest subnormal value is approximately 1.40129846e-45F (__FLT_DENORM_MIN__). The obvious conclusion is that subnormal support is broken or disabled on Alpha by default.
OK. I apologize for the noise.
participants (4)
-
Abramo Bagnara -
Joe Buck -
Joseph S. Myers -
Roberto Bagnara