Intel icc compiler warnings

php.internals

Rasmus Lerdorf

21 years ago
Out of curiosity I built HEAD with Intel's icc 8.1 compiler on my Debian laptop today. The full set of warnings (425k worth) are at: http://lerdorf.com/icc_warnings.txt There are some fun ones in there, like these: /home/rasmus/php-src/Zend/zend_alloc.c(252): warning #186: pointless comparison of unsigned integer with zero && offset >= 0) { ^ Zend/zend_execute.h(83): warning #1572: floating-point equality and inequality comparisons are unreliable result = (op->value.dval ? 1 : 0); ^ /home/rasmus/php-src/sapi/cli/php_cli.c(773): warning #25: quoted string should contain at least one character case '': /* generate indented source mode*/ ^ /home/rasmus/php-src/ext/session/mod_files.c(332): warning #266: function declared implicitly n = pread(data->fd, *val, sbuf.st_size, 0); ^ /home/rasmus/php-src/ext/session/mod_files.c(368): warning #266: function declared implicitly n = pwrite(data->fd, val, vallen, 0); ^ The invalid attribute and unsigned vs. signed warnings really clutter things up making it hard to find the more interesting warnings. The icc compiled libphp5.so was 9,816,098 bytes vs. 11,804,220 for gcc. Other things I compiled also consistently showed icc creating much smaller binaries. But that appears to just be due to a lot more symbol info. If I strip both I get: gcc libphp5.so 2,590,412 icc libphp5.so 2,793,780 Performance-wise it seems the icc binary is slower, but that is only from a very unscientific benchmark and I wonder if loading an icc php binary into a gcc Apache binary might not slow things down a bit. I haven't looked into the gcc compatibility libraries icc ships with to understand that well enough. But on first glance icc appears to generate a larger and slower libphp5.so. -Rasmus

Meno Abels

21 years ago
Hello, my experiences with performance and icc are, it is only faster if you use the highest optimization flags and the exact target processor flags. If you have a amd based notebook you are not in business with icc. -meno

Andi Gutmans

21 years ago
I have done some playing around with Intel compiler. It usually tends to be faster 10-20% but it depends on the flags you use, inlining of the program (Intel compiler seems to optimize inlines much better). Andi At 10:34 PM 9/18/2004 -0700, Rasmus Lerdorf wrote: