Bug 16679 – prefetch on old pentium d results in an illegal instruction

change 09e3fdd31d53fcb96a787a6a9de466a7c1f72ac1, pull request 6188 vendor_id : GenuineIntel cpu family : 15 model : 4 model name : Intel(R) Pentium(R) D CPU 3.00GHz stepping : 4 microcode : 0x6 Before that commit all tests pass. With that change, testxmm fails. ... runnable/testxmm.d (-inline -release -g -O -fPIC) Test failed. The logged output: ../src/dmd -conf= -m64 -Irunnable -odgenerated/runnable -ofgenerated/runnable/testxmm_0 runnable/testxmm.d runnable/testxmm.d(1328): Deprecation: instead of C-style syntax, use D-style syntax 'ubyte16[1] table' generated/runnable/testxmm_0 Illegal instruction (core dumped) ============================== Test failed: expected rc == 0, exited with rc == 132 (gdb) run Starting program: /home/braddr/sandbox/at-client/master-test-Linux_64_64/dmd/test/generated/runnable/testxmm_0 [Thread debugging using libthread_db enabled] Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1". [2, 0, 0, 0, 4, 0, 0, 0, 6, 0, 0, 0, 8, 0, 0, 0] Program received signal SIGILL, Illegal instruction. 0x000000000041d454 in core.simd.prefetch!(true, 0u).prefetch(const(void)*) () (gdb) disass Dump of assembler code for function _D4core4simd21__T8prefetchVbi1Vhi0Z8prefetchFNaNbNiNfPxvZv: 0x000000000041d448 <+0>: push %rbp 0x000000000041d449 <+1>: mov %rsp,%rbp 0x000000000041d44c <+4>: sub $0x10,%rsp 0x000000000041d450 <+8>: mov %rdi,-0x8(%rbp) => 0x000000000041d454 <+12>: rex.W prefetchw -0x8(%rbp) 0x000000000041d459 <+17>: leaveq 0x000000000041d45a <+18>: retq End of assembler dump.

The exact commits across the three repositories I've used, the druntime and phobos commits could likely be a range of valid enough points, but these are what I used. dmd passes: 9a935f68feb360f78ef32913053ce05c838fcafe dmd fails: 0fc8be57bb0107a718ac2dabd1b5482c1ebe89ef druntime: 565b5c3163ae93858702daab9bccfcd7a4d3f748 phobos: 30aca423cdd6d7943bb1dde45f526a6078a024ae Chances are that this is largely due to the age of the cpu given that all the other systems running test builds are passing. This machine is old-farm-1, part of the fleet of auto test machines, but I've had it offline for a while for unrelated reasons. Unfortunate timing. I can't add it back with the test failing.

Commits pushed to master at https://github.com/dlang/dmd https://github.com/dlang/dmd/commit/a5bffa8b35bb519ca7a5cd41fee86f541a1c2e76 fix Issue 16679 - prefetch on old pentium d results in an illegal instruction https://github.com/dlang/dmd/commit/f5a73f36360fc5b2b2e93f2dd17082c4fed2b0bc Merge pull request #6266 from WalterBright/fix16679 fix Issue 16679 - prefetch on old pentium d results in an illegal ins…

Reopened and lowered to major from regression until the cpuid work is done and the test is re-enabled. I would have suggested waiting for the right fix rather than the quick fix, but in neither case should we loose track of the disabled test.

(In reply to Brad Roberts from comment #4) > Reopened and lowered to major from regression until the cpuid work is done > and the test is re-enabled. I would have suggested waiting for the right > fix rather than the quick fix, but in neither case should we loose track of > the disabled test. This is already covered by https://issues.dlang.org/show_bug.cgi?id=16704

Commits pushed to scope at https://github.com/dlang/dmd https://github.com/dlang/dmd/commit/a5bffa8b35bb519ca7a5cd41fee86f541a1c2e76 fix Issue 16679 - prefetch on old pentium d results in an illegal instruction https://github.com/dlang/dmd/commit/f5a73f36360fc5b2b2e93f2dd17082c4fed2b0bc Merge pull request #6266 from WalterBright/fix16679

Commits pushed to newCTFE at https://github.com/dlang/dmd https://github.com/dlang/dmd/commit/a5bffa8b35bb519ca7a5cd41fee86f541a1c2e76 fix Issue 16679 - prefetch on old pentium d results in an illegal instruction https://github.com/dlang/dmd/commit/f5a73f36360fc5b2b2e93f2dd17082c4fed2b0bc Merge pull request #6266 from WalterBright/fix16679