This is the mail archive of the
mailing list for the GCC project.
Re: RFC: patch to build GCC for arm with LRA
- From: Ramana Radhakrishnan <ramrad01 at arm dot com>
- To: Yvan Roux <yvan dot roux at linaro dot org>
- Cc: Vladimir Makarov <vmakarov at redhat dot com>, Richard Earnshaw <Richard dot Earnshaw at arm dot com>, gcc-patches <gcc-patches at gcc dot gnu dot org>, Marcus Shawcroft <Marcus dot Shawcroft at arm dot com>, Matthew Gretton-Dann <matthew dot gretton-dann at linaro dot org>, Richard Henderson <rth at redhat dot com>, "rdsandiford at googlemail dot com" <rdsandiford at googlemail dot com>
- Date: Mon, 23 Sep 2013 15:53:49 +0100
- Subject: Re: RFC: patch to build GCC for arm with LRA
- Authentication-results: sourceware.org; auth=none
- References: <CAD57uCfSsANZvkMHmKphMHRnAxJNXGBxGEmSfhHSYwR7Jg9H-Q at mail dot gmail dot com> <877geri40k dot fsf at talisman dot default> <522D06E4 dot 1050602 at redhat dot com> <CAD57uCfpzd-7fKODP1oyWhY9f1QW2J=3BpMT3LAUMqC82Xvy0g at mail dot gmail dot com> <87zjrmgytu dot fsf at talisman dot default> <87vc2agymo dot fsf at talisman dot default> <CAD57uCfU3LMX+JM3yaJopqsp_rkJ3oyJv5g5VmJ5-9iyQvEMvQ at mail dot gmail dot com> <87wqmpg5n4 dot fsf at talisman dot default> <CAD57uCe-GrDqMHTNWtFcLRNDDVVr0_ZUg5E+zwaJdkCquNHagQ at mail dot gmail dot com> <87txhsmql0 dot fsf at talisman dot default> <CAD57uCchQSHKvaaA_Uq4rroiV9sopPgZgfTXnBJM-YpmWaHwdA at mail dot gmail dot com> <87li33n78h dot fsf at talisman dot default> <CAD57uCfsX4o39hPsgMBJ6K6E7p4YFrdqVFQdKAQHPHBADYAwnA at mail dot gmail dot com> <8738pbmep7 dot fsf at talisman dot default> <CAD57uCfqqQbdbHjkVV09KnLqb6tJLH5kNE4ogPY=hh8Ef8dK0Q at mail dot gmail dot com> <CAD57uCfFxt8-LZVDmY8jY89_o8nAWvpbPGx2FF8hC_G0OUOwrg at mail dot gmail dot com>
No more issues in libstdc++ as well (same as reasons as AArch64), and
only 3 failures in the testsuite:
- The first one is invalid as the test sans the assembler for
"ldaex\tr\[0-9\]+..." and it fails because with LRA the chosen
register is r12 and thus the instruction is "ldaex ip,..."
Fair enough - we should just fix the test and move on.
- The two others are the same bug where LRA keeps some REG_NOTES (DEAD
and UNUSED) on a comparison pattern where reload removes then, and the
results is that the comparison is removed. I'm currently working on
- Thumb still doesn't bootstrap.
I would suggest in addition a transitional command-line option to switch
between LRA and reload as a temporary measure so that folks can do some
more experimenting for AArch32.
I don't see why such a command-line option is wrong for A64 in principle
but that's something for Marcus / Richard to comment.