| Message ID | 20180619141104.GA28745@arm.com |
|---|---|
| State | New |
| Headers | show |
| Series | [mid-end] Correct subreg no-op handling for big-endian vec_select. | expand |
Ping. > -----Original Message----- > From: gcc-patches-owner@gcc.gnu.org <gcc-patches-owner@gcc.gnu.org> > On Behalf Of Tamar Christina > Sent: Tuesday, June 19, 2018 15:11 > To: gcc-patches@gcc.gnu.org > Cc: nd <nd@arm.com>; law@redhat.com; rguenther@suse.de; ian@airs.com > Subject: [PATCH][GCC][mid-end] Correct subreg no-op handling for big- > endian vec_select. > > Hi All, > > Previously GCC's no-op detection could would consider something a no-op > even when the mode change is not directly possible. This caused subregs > that shouldn't be removed to be treated as a no-op and deleted. > > Regtested on armeb-none-eabi and no regressions. > Bootstrapped on arm-none-linux-gnueabihf and no issues. > > Ok for trunk? and for backport to GCC 8? > > Thanks, > Tamar > > gcc/ > 2018-06-19 Tamar Christina <tamar.christina@arm.com> > > PR target/84711 > * rtlanal.c (set_noop_p): Constrain on mode change, > include hard-reg-set.h > > --
On 06/19/2018 08:11 AM, Tamar Christina wrote: > Hi All, > > Previously GCC's no-op detection could would consider something a no-op even when the > mode change is not directly possible. This caused subregs that shouldn't be removed > to be treated as a no-op and deleted. > > Regtested on armeb-none-eabi and no regressions. > Bootstrapped on arm-none-linux-gnueabihf and no issues. > > Ok for trunk? and for backport to GCC 8? > > Thanks, > Tamar > > gcc/ > 2018-06-19 Tamar Christina <tamar.christina@arm.com> > > PR target/84711 > * rtlanal.c (set_noop_p): Constrain on mode change, > include hard-reg-set.h > OK. Though please include a testcase. I believe you had big-endian-subreg.c in the original target specific approach to fixing this problem. That'd be fine to re-use here. jeff
On Thu, 28 Jun 2018 at 05:11, Jeff Law <law@redhat.com> wrote: > > On 06/19/2018 08:11 AM, Tamar Christina wrote: > > Hi All, > > > > Previously GCC's no-op detection could would consider something a no-op even when the > > mode change is not directly possible. This caused subregs that shouldn't be removed > > to be treated as a no-op and deleted. > > > > Regtested on armeb-none-eabi and no regressions. > > Bootstrapped on arm-none-linux-gnueabihf and no issues. > > > > Ok for trunk? and for backport to GCC 8? > > > > Thanks, > > Tamar > > > > gcc/ > > 2018-06-19 Tamar Christina <tamar.christina@arm.com> > > > > PR target/84711 > > * rtlanal.c (set_noop_p): Constrain on mode change, > > include hard-reg-set.h > > > OK. Though please include a testcase. I believe you had > big-endian-subreg.c in the original target specific approach to fixing > this problem. That'd be fine to re-use here. > Hi, It looks like this new test fails on x86_32 (i686-pc-linux-gnu): /gcc/testsuite/gcc.dg/vect/pr84711.c: In function 'fn1': /gcc/testsuite/gcc.dg/vect/pr84711.c:8:5: note: The ABI for passing parameters with 16-byte alignment has changed in GCC 4.6 /gcc/testsuite/gcc.dg/vect/pr84711.c:8:5: warning: SSE vector argument without SSE enabled changes the ABI [-Wpsabi] FAIL: gcc.dg/vect/pr84711.c -flto -ffat-lto-objects (test for excess errors) Excess errors: /gcc/testsuite/gcc.dg/vect/pr84711.c:8:5: warning: SSE vector argument without SSE enabled changes the ABI [-Wpsabi] > jeff
diff --git a/gcc/rtlanal.c b/gcc/rtlanal.c index ac3662de3ce0c9fcb58a042138f461c538fc1522..7af516378b38e043849cc2e1326a43837856ae4d 100644 --- a/gcc/rtlanal.c +++ b/gcc/rtlanal.c @@ -35,6 +35,7 @@ along with GCC; see the file COPYING3. If not see #include "recog.h" #include "addresses.h" #include "rtl-iter.h" +#include "hard-reg-set.h" /* Forward declarations */ static void set_of_1 (rtx, const_rtx, void *); @@ -1620,8 +1621,9 @@ set_noop_p (const_rtx set) if (INTVAL (XVECEXP (par, 0, i)) != c0 + i) return 0; return - simplify_subreg_regno (REGNO (src0), GET_MODE (src0), - offset, GET_MODE (dst)) == (int) REGNO (dst); + REG_CAN_CHANGE_MODE_P (REGNO (dst), GET_MODE (src0), GET_MODE (dst)) + && simplify_subreg_regno (REGNO (src0), GET_MODE (src0), + offset, GET_MODE (dst)) == (int) REGNO (dst); } return (REG_P (src) && REG_P (dst)