https://bugs.kde.org/show_bug.cgi?id=339596
--- Comment #18 from Mark Wielaard <m...@redhat.com> --- Created attachment 100955 --> https://bugs.kde.org/attachment.cgi?id=100955&action=edit Testcases for fma4 instructions Here are some testcases for the FMA4 instructions. I haven't looked yet at the XOP instructions. Maybe it is an idea to do FMA4 and XOP as separate patches? The testcase is based on the idea from the avx-1.c testcase. It creates a a block of values loads some of the values into xmm/ymm registers and creates a memory reference to another. Then it prints the contents of the block before and after the various instructions. The expected output comes from running the program on an actual processor having fma4 instructions. This is then compared with the same program running under valgrind. With the current patch there are some differences. First the 256bit ymm operations aren't supported, so they have been disabled in the testcase for now. But I am not sure we really should enable the fma4 cpuid bit in valgrind before we really support them. Secondly some 128bit xmm operations should clear the upper 128 bits of the corresponding YMM register to zeros and don't do so at the moment. Lastly the "full 0xFF" testcases do show some differences (but the zeros, ones and random cases all look fine). I'll also attach the stdout.diff next. If you think doing full bit-indentical tests are not fair/achievable for these instructions then let me know and we can see if we can create some testcases that explicitly work on float/double results. -- You are receiving this mail because: You are watching all bug changes.
