| Message ID | 561d01a4-84b7-b70d-b9b9-517c1528c0c2@redhat.com |
|---|---|
| State | New |
| Headers | show |
On Fri, Apr 07, 2017 at 12:04:16PM -0400, Vladimir Makarov wrote: > The following patch fixes > > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70478 > > The patch was successfully bootstrapped and tested on x86-64/ppc64/arm64. > > Committed as rev. 246764. > > Index: ChangeLog > =================================================================== > --- ChangeLog (revision 246763) > +++ ChangeLog (working copy) > @@ -1,3 +1,9 @@ > +2017-04-07 Vladimir Makarov <vmakarov@redhat.com> > + > + PR rtl-optimization/70478 > + * lra-constraints.c (process_alt_operands): Disfavor alternative > + insn memory operands. This has regressed: +UNRESOLVED: gfortran.dg/pr68627.f -O scan-assembler-not vbroadcastsd[ \\\\t]+%xmm[0-9]+, %ymm[0-9]+ +FAIL: gfortran.dg/pr68627.f -O (internal compiler error) +FAIL: gfortran.dg/pr68627.f -O (test for excess errors) on x86_64-linux, starting with r246764 there is ICE: Error: unable to find a register to spill pr68627.f:16:0: Error: this is the insn: (insn 202 1300 1177 12 (set (reg:V2DF 1306 [785]) (vec_concat:V2DF (reg:DF 1307 [orig:259 _282 ] [259]) (reg:DF 1421 [orig:263 _313 ] [263]))) "pr68627.f":11 2727 {vec_concatv2df} (expr_list:REG_DEAD (reg:DF 1421 [orig:263 _313 ] [263]) (expr_list:REG_DEAD (reg:DF 1307 [orig:259 _282 ] [259]) (nil)))) pr68627.f:16:0: internal compiler error: in assign_by_spills, at lra-assigns.c:1476 Jakub
On 04/07/2017 05:45 PM, Jakub Jelinek wrote: > On Fri, Apr 07, 2017 at 12:04:16PM -0400, Vladimir Makarov wrote: > >> Index: ChangeLog >> =================================================================== >> --- ChangeLog (revision 246763) >> +++ ChangeLog (working copy) >> @@ -1,3 +1,9 @@ >> +2017-04-07 Vladimir Makarov <vmakarov@redhat.com> >> + >> + PR rtl-optimization/70478 >> + * lra-constraints.c (process_alt_operands): Disfavor alternative >> + insn memory operands. > This has regressed: > +UNRESOLVED: gfortran.dg/pr68627.f -O scan-assembler-not vbroadcastsd[ \\\\t]+%xmm[0-9]+, %ymm[0-9]+ > +FAIL: gfortran.dg/pr68627.f -O (internal compiler error) > +FAIL: gfortran.dg/pr68627.f -O (test for excess errors) > on x86_64-linux, starting with r246764 there is ICE: > Error: unable to find a register to spill > pr68627.f:16:0: Error: this is the insn: > (insn 202 1300 1177 12 (set (reg:V2DF 1306 [785]) > (vec_concat:V2DF (reg:DF 1307 [orig:259 _282 ] [259]) > (reg:DF 1421 [orig:263 _313 ] [263]))) "pr68627.f":11 2727 {vec_concatv2df} > (expr_list:REG_DEAD (reg:DF 1421 [orig:263 _313 ] [263]) > (expr_list:REG_DEAD (reg:DF 1307 [orig:259 _282 ] [259]) > (nil)))) > pr68627.f:16:0: internal compiler error: in assign_by_spills, at lra-assigns.c:1476 > > Thanks, Jakub. I used only check-gcc on 3 platforms to check the patch. I should have checked other languages too. Meanwhile, I've reversed the patch. I'll continue work on the PR.
Index: ChangeLog =================================================================== --- ChangeLog (revision 246763) +++ ChangeLog (working copy) @@ -1,3 +1,9 @@ +2017-04-07 Vladimir Makarov <vmakarov@redhat.com> + + PR rtl-optimization/70478 + * lra-constraints.c (process_alt_operands): Disfavor alternative + insn memory operands. + 2017-04-07 Jeff Law <law@redhat.com> * config/iq2000/iq2000.c (final_prescan_insn): Do not separate a Index: testsuite/ChangeLog =================================================================== --- testsuite/ChangeLog (revision 246763) +++ testsuite/ChangeLog (working copy) @@ -1,3 +1,8 @@ +2017-04-07 Vladimir Makarov <vmakarov@redhat.com> + + PR rtl-optimization/70478 + * gcc.target/s390/pr70478.c: New. + 2017-04-07 Martin Liska <mliska@suse.cz> PR target/79889 Index: lra-constraints.c =================================================================== --- lra-constraints.c (revision 246763) +++ lra-constraints.c (working copy) @@ -2685,6 +2685,21 @@ process_alt_operands (int only_alternati } } + /* When we use memory operand, the insn should read the + value from memory and even if we just wrote a value + into the memory it is costly in comparison with an + insn alternative which does not use memory + (e.g. register or immediate operand). */ + if (no_regs_p && offmemok) + { + if (lra_dump_file != NULL) + fprintf + (lra_dump_file, + " Using memory insn operand %d: reject+=3\n", + nop); + reject += 3; + } + #ifdef SECONDARY_MEMORY_NEEDED /* If reload requires moving value through secondary memory, it will need one more insn at least. */ Index: testsuite/gcc.target/s390/pr70478.c =================================================================== --- testsuite/gcc.target/s390/pr70478.c (nonexistent) +++ testsuite/gcc.target/s390/pr70478.c (working copy) @@ -0,0 +1,8 @@ +/* { dg-do compile } */ +/* { dg-options "-w -O3 -march=z9-109" } */ +/* { dg-final { scan-assembler-not "stg" } } */ + +void foo(unsigned int *a, unsigned char *b) +{ + *a &= *b; +}