All four i386 BSD ports have been failing ACATS test C460007, case 9 and 10, but only under optimization level 2 and higher. The x86_64 BSD ports did not fail the floating-point/integer conversion test cases, and neither did the i386 port of OpenSolaris SXCE. For DragonFlyBSD and FreeBSD, it was the only failure in the ACATS testsuite, so I submitted GCC bug report 46490 last night.

After some initial assumptions were dispelled, Eric Botcazou realized that it was a legitimate optimization regression, and that it wasn't a bug within GNAT. The bug got fixed about 12 hours after the first response. I was worried that it would be ignored because it appeared to be a BSD-only issue, but apparently it could be reproduced on Linux using the -mtune=i386 switch. The next published version of DragonFly32 GNAT and FreeBSD32 GNAT are expected to pass all ACATS tests.