ReadMe.txt This directory contains results saved from running tests with Libm-93 with modifications for my current work. While it would be preferable to have results from a clean version of Libm, I have made a cursory review of the differences between these results and those I did previously from a version of Libm near Libm-90, and I believe these results can be used as a basis for comparison with future results. Any differences that turn up should be investigated to understand why they occur. (And, of course, the test programs should be improved not to report false positives. Or to run without failing due to bus errors or segmentation faults. Or to build without compiler errors.) The operating systems used were Tiger 8A410 and Karma 8V6. CVS information: $Revision: 1.2 $, $Date: 2005/04/15 18:05:00 $. Test Programs. The libm target was built, and then an attempt was made to build and run each of the targets below. The builds occurred in the main Libm directory, but the test programs appear to expect to be run from the build subdirectory. VectorTestLDBL64. Executable name is VectorTestLDBL64. Output file name is ../noship.subproj/VectorTestLDBL64.results. PowerPC: Runs to completion. Differences from Libm-90x: Four new errors in modff because it is run with test cases intended only for modf. The test program appears to be ignoring the "d" mode that says the test case is for double-precision and not for single-precision. Actually, the saved file reports only three errors in modff, but a later run shows four. Four is expected, matching the four test cases at issue. I am at a loss to explain the difference. A later run built from code only slightly changed showed the four expected differences. The total test counts suggest one case was not executed -- was something changed in the test vector data file? New totals for test cases, as new cases were added. IA-32: Does not build because -mlong-double-64 is not supported. VectorTestLDBL128. Executable name is VectorTest. Output file name is ../noship.subproj/VectorTestLDBL128.results. PowerPC: Runs to completion. Differences from Libm-90x: New errors in modff and modfl and new totals as above. New differences in output that at first appearance are due to test program correctly printing very large finite numbers, where as the saved results show "inf". Perhaps a bug in printf was fixed? IA-32: Does not build because -mlong-double-128 is not supported. VectorTest x86. Executable name is VectorTest. Output file name is ../noship.subproj/VectorTextx86.results. PowerPC: Fails to build: "cc1: error: invalid option '96bit-long-double'". (Obviously, this does not build for a PowerPC, but it also does not build on a PowerPC. The target appears not to contain explicit indication of its target that would build it for i386 on a PowerPC.) IA-32: Builds and gets "Bus error", after printing "../noship.subproj/vectors/TVB2D2B.TEXT". Writes some results. VectorTestLP64. Executable name is VectorTestLP64. Output file name would be ../noship.subproj/VectorTextLP64.results? PowerPC: Gets Segmentation fault. IA-32: Builds a PowerPC executable (not executed). LibmTestHarness. (With libm-test.c revision 1.3.) Executable name is LibmTestHarness. Writes results to standard output, redirected to LibmTestHarness.stdout. PowerPC: Reports 1139 test cases plus 953 for exception flags, 122 errors. IA-32: Reports 1139 test cases plus 953 for exception flags, 62 errors. VectorTestLDBL64++. Executable name is VectorTestLDBL64++. Output file name is ../noship.subproj/VectorTestLDBL64++.results. PowerPC: Runs to completion. Differences from Libm-90x: New errors in modff and new totals as above. IA-32: Does not build because -mlong-double-64 is not supported. VectorTestLDBL128++. Executable name is VectorTestLDBL128++. Output file name is ../noship.subproj/VectorTestLDBL128++.results. PowerPC: Runs to completion. Differences from Libm-90x: New errors in modff and modfl and new totals as above. IA-32: Does not build because -mlong-double-128 is not supported.