Bill Hart wrote:
Leif, is this only exposed when the user manually sets an optimisation level of -O3 or is it the default when MPIR builds on this system?
Nope, as mentioned in my previous post: ;-)
> [...] unfortunately (also) the library gets miscompiled. > > When I build the library with XOP instructions enabled and the test suite without, I'm getting the same five test failures with GCC 4.8.3, one less (i.e., only two) with GCC 4.9.0. > > Note however that this happens only with '-O3', otherwise GCC hardly ever emits XOP instructions at all (in my experience at least).
I would say that if you compile with -O3 you are asking for trouble.
Well, no risk, no fun. Actually the only flaw (in [recent] vanilla FSF GCCs) with '-O3' I'm currently aware of (or running into). (Ubuntu's and Debian's GCCs are similarly broken, but at least Ubuntu's in addition to other bugs...)
But if we do that by default, that's a different story.
I'd rather say it's something to mention in the release notes; if the user specifies C/CXXFLAGS, MPIR's currently get dropped anyway, so adding '-mno-xop' on bdver* wouldn't really help, although you could of course issue a warning.
Or /prepend/ MPIR's to user-specified ones... (That's what we currently do in Sage, but to achieve this, we have to dry-run MPIR's 'configure' once with CFLAGS and CXXFLAGS temporarily unset.)
-leif
On 16 June 2014 22:22, leif <not.rea...@online.de <mailto:not.rea...@online.de>> wrote: Bill Hart wrote: I have uploaded alpha7 to our website (http://mpir.org). Just for the record: MPIR's test suite now exposes GCC's XOP-brokenness (I *think* that's new in alpha7, but probably already in alpha5 or 6). AFAICT, there's not a single GCC version (supporting bdver*) that /isn't/ in some way broken w.r.t. XOP instructions, but so far MPIR wasn't affected (and GMP still isn't). So be prepared for respective bug reports... ;-) (Haven't checked yet whether just the test suite miscompiles, but I at least hope the library itself isn't affected.) -leif GCC 4.9.0, -march=bdver2 -mtune=bdver2 -O3 (all pass with -mno-xop): PASS: t-inv_divappr_q /bin/bash: line 1: 12856 Aborted "$tst" > t-inv_divappr_q_n.log-t 2>&1 FAIL: t-inv_divappr_q_n /bin/bash: line 1: 12887 Aborted "$tst" > t-invert.log-t 2>&1 FAIL: t-invert PASS: t-iord_u PASS: t-lorrshift1 PASS: t-matrix22 /bin/bash: line 1: 13011 Aborted "$tst" > t-mp_bases.log-t 2>&1 FAIL: t-mp_bases PASS: t-mullow_basecase PASS: t-mullowhigh PASS: t-mulmid PASS: t-mulmod_2expm1 PASS: t-mulmod_2expp1 PASS: t-neg PASS: t-perfsqr PASS: t-redc_1 PASS: t-sb_bdiv_q PASS: t-sb_bdiv_qr PASS: t-sb_div_q PASS: t-sb_div_qr PASS: t-sb_divappr_q PASS: t-scan PASS: t-subadd_n PASS: t-tdiv_q PASS: t-tdiv_qr PASS: st_fat PASS: st_instrument ==============================__==============================__= 3 of 49 tests failed See tests/mpn/test-suite.log Please report to http://groups.google.co.uk/__group/mpir-devel/ <http://groups.google.co.uk/group/mpir-devel/> ==============================__==============================__= ==============================__============ MPIR 2.7.0: tests/mpn/test-suite.log ==============================__============ 3 of 49 tests failed. .. contents:: :depth: 2 FAIL: t-inv_divappr_q_n (exit: 134) ==============================__===== failed: nn = 3084, dn = 1542, qn = 1542, rn = 2312 ------------------------------__------------------------------__------------------- GCC 4.8.3, -march=bdver2 -mtune=bdver2 -O3 (all pass with -mno-xop): PASS: t-inv_div_q /bin/bash: line 1: 19233 Aborted "$tst" > t-inv_div_qr.log-t 2>&1 FAIL: t-inv_div_qr /bin/bash: line 1: 19264 Aborted "$tst" > t-inv_div_qr_n.log-t 2>&1 FAIL: t-inv_div_qr_n /bin/bash: line 1: 19295 Aborted "$tst" > t-inv_divappr_q.log-t 2>&1 FAIL: t-inv_divappr_q /bin/bash: line 1: 19326 Aborted "$tst" > t-inv_divappr_q_n.log-t 2>&1 FAIL: t-inv_divappr_q_n /bin/bash: line 1: 19357 Aborted "$tst" > t-invert.log-t 2>&1 FAIL: t-invert PASS: t-iord_u PASS: t-lorrshift1 PASS: t-matrix22 PASS: t-mp_bases PASS: t-mullow_basecase PASS: t-mullowhigh PASS: t-mulmid PASS: t-mulmod_2expm1 PASS: t-mulmod_2expp1 PASS: t-neg PASS: t-perfsqr PASS: t-redc_1 PASS: t-sb_bdiv_q PASS: t-sb_bdiv_qr PASS: t-sb_div_q PASS: t-sb_div_qr PASS: t-sb_divappr_q PASS: t-scan PASS: t-subadd_n PASS: t-tdiv_q PASS: t-tdiv_qr PASS: st_fat PASS: st_instrument ==============================__==============================__= 5 of 49 tests failed See tests/mpn/test-suite.log Please report to http://groups.google.co.uk/__group/mpir-devel/ <http://groups.google.co.uk/group/mpir-devel/> ==============================__==============================__= ==============================__============ MPIR 2.7.0: tests/mpn/test-suite.log ==============================__============ 5 of 49 tests failed. .. contents:: :depth: 2 FAIL: t-inv_div_qr (exit: 134) ============================== failed: remainder does not match -- () The ASCII Ribbon Campaign /\ Help Cure HTML E-Mail
-- () The ASCII Ribbon Campaign /\ Help Cure HTML E-Mail -- You received this message because you are subscribed to the Google Groups "mpir-devel" group. To unsubscribe from this group and stop receiving emails from it, send an email to mpir-devel+unsubscr...@googlegroups.com. To post to this group, send email to mpir-devel@googlegroups.com. Visit this group at http://groups.google.com/group/mpir-devel. For more options, visit https://groups.google.com/d/optout.
