Re: [PATCH] RISC-V: Optimize relax of GP with max_alignment.

[Lifang Xia <lifang_xia@linux.alibaba.com>]

On 2022/9/23 18:44, Nelson Chu wrote:
> Unfortunately, I got lots of gcc failures for rv64gc-elf regression,
>
>                 ========= Summary of gcc testsuite =========
>
>                              | # of unexpected case / # of unique unexpected case
>
>                              |          gcc |          g++ |     gfortran |
>
>       rv64gc/  lp64d/ medlow |38068 /  4059 |20265 /  2738 |      - |
>
> And I guess that is because we should pass "sec" rather than "sym_sec"
> to the _bfd_riscv_get_max_alignment_in_gp.
Agree with that.

>
> Besides, not really sure if that may happen: do some sections that
> don't be considered in the range of gp in the first round, but will be
> considered in the later relax passes?  If so, then does that make the
> max_alignment returned by the _bfd_riscv_get_max_alignment_in_gp have
> different values?  Not really sure if that will break something.

It would be occurence if the gp is assigned with a absolute address, 
such as "__global_point = 0x4000".

>   An
> idea is that - we probably can reserve the original max_alignment when
> determine if the section is in the range of gp, like this,
> https://github.com/bminor/binutils-gdb/blob/master/bfd/elfnn-riscv.c#L4566.
> Though it seems a bit weird to estimate the alignment twice, but the
> original max alignment is used to estimate and try to cover most of
> the sections in the range of gp, and then we will get the new max
> alignment, which should be always smaller than the original one, and
> use it to estimate if the symbol is in the range if gp.  However, I
> don't have any test case for now to approve what I am thinking, but
> that at least works for your testcase, too.
That's is conservative idea, it would work fine for both of the test cases.
> Moreover, not sure if we can estimate the new max alignment for gp
> once also in here,
> https://github.com/bminor/binutils-gdb/blob/master/bfd/elfnn-riscv.c#L4708.
> If we just use the original max alignment to estimate it, then doing
> once is probably enough.

It is not a good choice. Because the max alignment is only used for gp, 
but not all of the relax need it.

Thanks,
Lifang

>
> Happy to see any thoughts!
>
> Thanks
> Nelson
>
> On Fri, Sep 23, 2022 at 5:47 PM lifang_xia--- via Binutils
> <binutils@sourceware.org> wrote:
>> From: Lifang Xia <lifang_xia@linux.alibaba.com>
>>
>> The max_alignment defined out of [gp-2K, gp+2k), the max_alignment
>> shouldn't affect the relax of gp.
>> If the symbol is in [gp-2K, gp+2k), the max_alignment would be
>> replaced with the max_alignment of the section in [gp-2k, gp+2k).
>>
>> bfd/
>>          * elfnn-riscv.c (_bfd_riscv_get_max_alignment_in_gp): New.
>>          (_bfd_riscv_relax_lui): The max_alignment of sections is from
>>          [gp-2K, gp+2K).
>>          (_bfd_riscv_relax_pc): Likewise.
>> ld/
>>          * ld/testsuite/ld-riscv-elf/relax-max-align.*: New tests.
>> ---
>>   bfd/elfnn-riscv.c                           | 37 +++++++++++++++++
>>   ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp  |  1 +
>>   ld/testsuite/ld-riscv-elf/relax-max-align.d | 46 +++++++++++++++++++++
>>   ld/testsuite/ld-riscv-elf/relax-max-align.s | 29 +++++++++++++
>>   4 files changed, 113 insertions(+)
>>   create mode 100644 ld/testsuite/ld-riscv-elf/relax-max-align.d
>>   create mode 100644 ld/testsuite/ld-riscv-elf/relax-max-align.s
>>
>> diff --git a/bfd/elfnn-riscv.c b/bfd/elfnn-riscv.c
>> index 0e0a0b09e24..83afc95ccab 100644
>> --- a/bfd/elfnn-riscv.c
>> +++ b/bfd/elfnn-riscv.c
>> @@ -4259,6 +4259,29 @@ _bfd_riscv_get_max_alignment (asection *sec)
>>     return (bfd_vma) 1 << max_alignment_power;
>>   }
>>
>> +/* Traverse all output sections and return the max alignment
>> +   in [gp-2K, gp+2K) .  */
>> +
>> +static bfd_vma
>> +_bfd_riscv_get_max_alignment_in_gp (asection *sec, bfd_vma gp)
>> +{
>> +  unsigned int max_alignment_power = 0;
>> +  asection *o;
>> +
>> +  if (sec == NULL)
>> +    return 0;
>> +
>> +  for (o = sec->output_section->owner->sections; o != NULL; o = o->next)
>> +    {
>> +      if ((VALID_ITYPE_IMM (sec_addr(o) - gp)
>> +          || VALID_ITYPE_IMM (sec_addr(o) + o->size - gp))
>> +         && (o->alignment_power > max_alignment_power))
>> +       max_alignment_power = o->alignment_power;
>> +    }
>> +
>> +  return (bfd_vma) 1 << max_alignment_power;
>> +}
>> +
>>   /* Relax non-PIC global variable references to GP-relative references.  */
>>
>>   static bool
>> @@ -4290,6 +4313,13 @@ _bfd_riscv_relax_lui (bfd *abfd,
>>         if (h->u.def.section->output_section == sym_sec->output_section
>>            && sym_sec->output_section != bfd_abs_section_ptr)
>>          max_alignment = (bfd_vma) 1 << sym_sec->output_section->alignment_power;
>> +      else if (!undefined_weak)
>> +       {
>> +         /* Otherwise, consider the alignment of sections in [gp-2K,gp+2K). */
>> +         bfd_vma new_max_alignment = _bfd_riscv_get_max_alignment_in_gp (sym_sec, gp);
>> +         if (new_max_alignment)
>> +           max_alignment = new_max_alignment;
>> +       }
>>       }
>>
>>     /* Is the reference in range of x0 or gp?
>> @@ -4556,6 +4586,13 @@ _bfd_riscv_relax_pc (bfd *abfd ATTRIBUTE_UNUSED,
>>         if (h->u.def.section->output_section == sym_sec->output_section
>>            && sym_sec->output_section != bfd_abs_section_ptr)
>>          max_alignment = (bfd_vma) 1 << sym_sec->output_section->alignment_power;
>> +      else if (!undefined_weak)
>> +       {
>> +         /* Otherwise, consider the alignment of sections in [gp-2K,gp+2K). */
>> +         bfd_vma new_max_alignment = _bfd_riscv_get_max_alignment_in_gp (sym_sec, gp);
>> +         if (new_max_alignment)
>> +           max_alignment = new_max_alignment;
>> +       }
>>       }
>>
>>     /* Is the reference in range of x0 or gp?
>> diff --git a/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp b/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
>> index df89e0ee68b..e53f16facfa 100644
>> --- a/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
>> +++ b/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
>> @@ -169,6 +169,7 @@ if [istarget "riscv*-*-*"] {
>>       run_dump_test "attr-merge-priv-spec-failed-05"
>>       run_dump_test "attr-merge-priv-spec-failed-06"
>>       run_dump_test "attr-phdr"
>> +    run_dump_test "relax-max-align"
>>       run_ld_link_tests [list \
>>          [list "Weak reference 32" "-T weakref.ld -m[riscv_choose_ilp32_emul]" "" \
>>              "-march=rv32i -mabi=ilp32" {weakref32.s} \
>> diff --git a/ld/testsuite/ld-riscv-elf/relax-max-align.d b/ld/testsuite/ld-riscv-elf/relax-max-align.d
>> new file mode 100644
>> index 00000000000..4ba920da970
>> --- /dev/null
>> +++ b/ld/testsuite/ld-riscv-elf/relax-max-align.d
>> @@ -0,0 +1,46 @@
>> +#source: relax-max-align.s
>> +#ld:
>> +#objdump: -d
>> +
>> +.*:[   ]+file format .*
>> +
>> +
>> +Disassembly of section .text:
>> +
>> +0+[0-9a-f]+ <_start>:
>> +.*:[   ]+[0-9a-f]+[    ]+addi[         ]+.*<gdata>
>> +.*:[   ]+[0-9a-f]+[    ]+jal[  ]+.*
>> +.*:[   ]+[0-9a-f]+[    ]+j[    ]+.*
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +.*:[   ]+[0-9a-f]+[    ]+nop
>> +
>> +0+[0-9a-f]+ <func>:
>> +.*:[   ]+[0-9a-f]+[    ]+ret
>> +[      ]+...
>> diff --git a/ld/testsuite/ld-riscv-elf/relax-max-align.s b/ld/testsuite/ld-riscv-elf/relax-max-align.s
>> new file mode 100644
>> index 00000000000..0d162ff4d93
>> --- /dev/null
>> +++ b/ld/testsuite/ld-riscv-elf/relax-max-align.s
>> @@ -0,0 +1,29 @@
>> +
>> +.global _start
>> +_start:
>> +       lui   a0, %hi(gdata)
>> +       addi  a0, a0, %lo(gdata)
>> +       call func
>> +       j .
>> +       .size _start, . - _start
>> +
>> +.global func
>> +.align 7
>> +func:
>> +       ret
>> +       .size func, . - func
>> +
>> +.data
>> +padding:
>> +       .long 0
>> +       .long 0
>> +       .long 0
>> +       .long 0
>> +       .size padding, . - padding
>> +
>> +.global gdata
>> +.type gdata, object
>> +gdata:
>> +       .zero 4
>> +       .size gdata, . - gdata
>> +
>> --
>> 2.32.1 (Apple Git-133)
>>