Hi all: I'm wondered if somebody have been in troubles with UltraSparc IV proccessors with C code that use arithmetic / fpu instrucctions... Let me explain, we (my company) have a customer who uses Oracle 7.3.4 on both systems with Solaris 9 (09/04 and latests patches applied): 1) a sunblade 2000 (UltraSparc III) with GCC 3.4.2 and the oracle environment 2) a Sunfire V490 (UltraSparc IV) with GCC 3.4.2 and the oracle environment They have a wierd problem and it seems to be the generated code from the gcc on the USIV platform. They use a script which calls proc (C precompiler from Oracle) and make a .c code file. This is compiled with gcc and is used to make a X process which logs make some arithmetic computations with double, int and other data types. With the Blade 2000 thereis no compilations errors and the final binary runs fine against 90k records and do its job. But with the SFV490 thereis no complaints on the compilation phase the final binary is about 1KB bigger that the US3 counterpart and if falls at 5 records proccessed. The ORACLE errors tell us a problem with the precision on the number. Since we use the same .pc file and script to generate the object and final binary, use the same version on compiler we suspect a sligth diffrence on the FPU/aritmetic code generate on the USIV machine. By now my computer OS hard disk crashed, so, I can't post od/mdb and other files to shed some ligth on this issue.. But I want to know by know if somebody got this kind of problems with the USIV procs.... PD: We have made a lot of tests, downgraded the gcc on both machines and run the scripts to finally test the binaries... sames results... from USIII fine... from USIV if fails. The latest test involve some tunning (add -O2 and remove -lpthreads) to the compilation script and the binary run fine with 100record but after that if fails.. Thanks a lot, for any help.
