From: Palmer Dabbelt <palmer@rivosinc.com>
To: nelson@rivosinc.com
Cc: lifang_xia@linux.alibaba.com, binutils@sourceware.org,
Greg McGary <gkm@rivosinc.com>
Subject: Re: [PATCH v2] RISC-V: Optimize relax of GP/call with max_alignment.
Date: Tue, 22 Nov 2022 08:19:06 -0800 (PST) [thread overview]
Message-ID: <mhng-ef858f6b-9eba-4720-8174-befb36dbd66e@palmer-ri-x1c9> (raw)
In-Reply-To: <CAPpQWtDQe-Frq+wZFuOUFzSyh2FCjbMagYdtbm0PqOJJHRoX1Q@mail.gmail.com>
On Tue, 22 Nov 2022 02:19:00 PST (-0800), nelson@rivosinc.com wrote:
> On Thu, Oct 27, 2022 at 11:19 AM <lifang_xia@linux.alibaba.com> 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).
>>
>> Likewise, the max_alignment defined out of [call-1M, call+1M),
>> the max_alignment shouldn't affect the relax of call.
>> If the symbol is in [call-1M, call+1M), the max_alignment would be
>> replaced with the max_alignment of the section in [call-1M, call+1M).
>
> I'm thinking that - If it is safe enough to decide which output
> section alignments should be considered in the range of gp/pc, but the
> target symbols or the gp are ABS symbols. If one of them is ABS, then
> that means the output sections in the range of pc/gp may be different
> in the different relax rounds. So maybe we should,
> 1. make sure the target symbol and gp/pc are not ABS symbols, they
> should belong to a section.
> 2. consider the maximum output section alignment between the output
> sections of the target symbol and gp/pc.
> 3. Otherwise, consider all output section alignments as usual.
It's actually a bit screwier than that, we also have to contend with the
psABI's mixing of "PC relative" and "position independent". I've got a
bug on this: https://sourceware.org/bugzilla/show_bug.cgi?id=28789
>
> Thanks
> Nelson
>
>> bfd/
>> * elfnn-riscv.c (_bfd_riscv_get_max_alignment_in_jtype): New.
>> (_bfd_riscv_relax_call): The max_alignment of sections is from
>> [base-1M, base+1M).
>> (_bfd_riscv_get_max_alignment_in_itype): 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-call.*: New tests.
>> * ld/testsuite/ld-riscv-elf/relax-max-align-gp.*: New tests.
>> ---
>> bfd/elfnn-riscv.c | 67 +++++++++++++++++++
>> ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp | 2 +
>> .../ld-riscv-elf/relax-max-align-call.d | 18 +++++
>> .../ld-riscv-elf/relax-max-align-call.ld | 22 ++++++
>> .../ld-riscv-elf/relax-max-align-call.s | 23 +++++++
>> .../ld-riscv-elf/relax-max-align-gp.d | 46 +++++++++++++
>> .../ld-riscv-elf/relax-max-align-gp.s | 29 ++++++++
>> 7 files changed, 207 insertions(+)
>> create mode 100644 ld/testsuite/ld-riscv-elf/relax-max-align-call.d
>> create mode 100644 ld/testsuite/ld-riscv-elf/relax-max-align-call.ld
>> create mode 100644 ld/testsuite/ld-riscv-elf/relax-max-align-call.s
>> create mode 100644 ld/testsuite/ld-riscv-elf/relax-max-align-gp.d
>> create mode 100644 ld/testsuite/ld-riscv-elf/relax-max-align-gp.s
>>
>> diff --git a/bfd/elfnn-riscv.c b/bfd/elfnn-riscv.c
>> index cf852636c9c..ed72704bd80 100644
>> --- a/bfd/elfnn-riscv.c
>> +++ b/bfd/elfnn-riscv.c
>> @@ -4266,6 +4266,28 @@ typedef bool (*relax_func_t) (bfd *, asection *, asection *,
>> riscv_pcgp_relocs *,
>> bool undefined_weak);
>>
>> +/* Traverse all output sections in jtype, and return the max alignment. */
>> +
>> +static bfd_vma
>> +_bfd_riscv_get_max_alignment_in_jtype (asection *sec, bfd_vma base)
>> +{
>> + unsigned int max_alignment_power = 0;
>> + asection *o;
>> +
>> + if (sec == NULL)
>> + return 0;
>> +
>> + for (o = sec->owner->sections; o != NULL; o = o->next)
>> + {
>> + if (VALID_JTYPE_IMM (sec_addr(o) - base)
>> + || VALID_ITYPE_IMM (sec_addr(o) + o->size - base))
>
> Probably should be JTYPE?
>
>
>> + if (o->alignment_power > max_alignment_power)
>> + max_alignment_power = o->alignment_power;
>> + }
>> +
>> + return (bfd_vma) 1 << max_alignment_power;
>> +}
>> +
>> /* Relax AUIPC + JALR into JAL. */
>>
>> static bool
>> @@ -4294,6 +4316,15 @@ _bfd_riscv_relax_call (bfd *abfd, asection *sec, asection *sym_sec,
>> if (sym_sec->output_section == sec->output_section
>> && sym_sec->output_section != bfd_abs_section_ptr)
>> max_alignment = (bfd_vma) 1 << sym_sec->output_section->alignment_power;
>> + else
>> + {
>> + bfd_vma new_max_alignment =
>> + _bfd_riscv_get_max_alignment_in_jtype (sec,
>> + (sec_addr (sec)
>> + + rel->r_offset));
>> + if (new_max_alignment)
>> + max_alignment = new_max_alignment;
>> + }
>> foff += ((bfd_signed_vma) foff < 0 ? -max_alignment : max_alignment);
>> }
>>
>> @@ -4360,6 +4391,28 @@ _bfd_riscv_get_max_alignment (asection *sec)
>> return (bfd_vma) 1 << max_alignment_power;
>> }
>>
>> +/* Traverse all output sections in [gp-2K, gp+2K) and return the max alignment. */
>> +
>> +static bfd_vma
>> +_bfd_riscv_get_max_alignment_in_itype (asection *sec, bfd_vma gp)
>> +{
>> + unsigned int max_alignment_power = 0;
>> + asection *o;
>> +
>> + if (sec == NULL)
>> + return 0;
>> +
>> + for (o = sec->owner->sections; o != NULL; o = o->next)
>> + {
>> + if (VALID_ITYPE_IMM (sec_addr(o) - gp)
>> + || VALID_ITYPE_IMM (sec_addr(o) + o->size - gp))
>> + if (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
>> @@ -4391,6 +4444,13 @@ _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
>> + {
>> + /* Otherwise, consider the alignment of sections in [gp-2K,gp+2K). */
>> + bfd_vma new_max_alignment = _bfd_riscv_get_max_alignment_in_itype (sec, gp);
>> + if (new_max_alignment)
>> + max_alignment = new_max_alignment;
>> + }
>> }
>>
>> /* Is the reference in range of x0 or gp?
>> @@ -4656,6 +4716,13 @@ _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_itype (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..f38807b991e 100644
>> --- a/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
>> +++ b/ld/testsuite/ld-riscv-elf/ld-riscv-elf.exp
>> @@ -169,6 +169,8 @@ 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-gp"
>> + run_dump_test "relax-max-align-call"
>> 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-call.d b/ld/testsuite/ld-riscv-elf/relax-max-align-call.d
>> new file mode 100644
>> index 00000000000..490ad8c865e
>> --- /dev/null
>> +++ b/ld/testsuite/ld-riscv-elf/relax-max-align-call.d
>> @@ -0,0 +1,18 @@
>> +#source: relax-max-align-call.s
>> +#as: -march=rv64gc
>> +#ld: -Trelax-max-align-call.ld
>> +#objdump: -d
>> +
>> +.*:[ ]+file format .*
>> +
>> +
>> +Disassembly of section .text1:
>> +
>> +0000000000100000 <_start>:
>> +[ ]+100000:[ ]+79dff0ef[ ]+jal[ ]+1fff9c <foo>
>> +[ ]+100004:[ ]+a001[ ]+j[ ]+100004 <_start\+0x4>
>> +
>> +Disassembly of section .text2:
>> +
>> +00000000001fff9c <foo>:
>> +[ ]+1fff9c:[ ]+8082[ ]+ret
>> diff --git a/ld/testsuite/ld-riscv-elf/relax-max-align-call.ld b/ld/testsuite/ld-riscv-elf/relax-max-align-call.ld
>> new file mode 100644
>> index 00000000000..91991ed0f41
>> --- /dev/null
>> +++ b/ld/testsuite/ld-riscv-elf/relax-max-align-call.ld
>> @@ -0,0 +1,22 @@
>> +
>> +ENTRY(_start)
>> +
>> +SECTIONS
>> +{
>> + . = 0x100000;
>> +
>> + .text1 : {
>> + *(.text1)
>> + }
>> +
>> + . = 0x200000 - 100;
>> +
>> + .text2 : {
>> + *(.text2)
>> + }
>> +
>> + . = 0x300000;
>> + .data : {
>> + *(.data)
>> + }
>> +}
>> diff --git a/ld/testsuite/ld-riscv-elf/relax-max-align-call.s b/ld/testsuite/ld-riscv-elf/relax-max-align-call.s
>> new file mode 100644
>> index 00000000000..3335c0970e7
>> --- /dev/null
>> +++ b/ld/testsuite/ld-riscv-elf/relax-max-align-call.s
>> @@ -0,0 +1,23 @@
>> +
>> +.section .text1, "ax"
>> +.global _start
>> +.align 2
>> +_start:
>> + call foo
>> + j .
>> + .size _start, . - _start
>> +
>> +.section .text2, "ax"
>> +.align 2
>> +.global foo
>> +.type foo, @function
>> +foo:
>> + ret
>> + .size foo, . - foo
>> +
>> +.data
>> +.global gdata
>> +.align 10
>> +gdata:
>> + .long 0
>> + .size gdata, . - gdata
>> diff --git a/ld/testsuite/ld-riscv-elf/relax-max-align-gp.d b/ld/testsuite/ld-riscv-elf/relax-max-align-gp.d
>> new file mode 100644
>> index 00000000000..637de426ee4
>> --- /dev/null
>> +++ b/ld/testsuite/ld-riscv-elf/relax-max-align-gp.d
>> @@ -0,0 +1,46 @@
>> +#source: relax-max-align-gp.s
>> +#ld:
>> +#objdump: -d
>> +
>> +.*:[ ]+file format .*
>> +
>> +
>> +Disassembly of section .text:
>> +
>> +0+[0-9a-f]+ <_start>:
>> +.*:[ ]+[0-9a-f]+[ ]+add[ ]+.*<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-gp.s b/ld/testsuite/ld-riscv-elf/relax-max-align-gp.s
>> new file mode 100644
>> index 00000000000..0d162ff4d93
>> --- /dev/null
>> +++ b/ld/testsuite/ld-riscv-elf/relax-max-align-gp.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
>> +
>> --
>> 2.37.0 (Apple Git-136)
>>
prev parent reply other threads:[~2022-11-22 16:19 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-09-23 6:47 [PATCH] RISC-V: Optimize relax of GP " lifang_xia
2022-09-23 8:05 ` Nelson Chu
2022-09-23 9:09 ` Lifang Xia
2022-09-23 9:46 ` lifang_xia
2022-09-23 10:44 ` Nelson Chu
2022-09-24 1:54 ` Lifang Xia
2022-09-27 8:51 ` Nelson Chu
2022-10-27 3:19 ` [PATCH v2] RISC-V: Optimize relax of GP/call " lifang_xia
2022-11-09 2:52 ` Lifang Xia
2022-11-22 10:19 ` Nelson Chu
2022-11-22 16:19 ` Palmer Dabbelt [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=mhng-ef858f6b-9eba-4720-8174-befb36dbd66e@palmer-ri-x1c9 \
--to=palmer@rivosinc.com \
--cc=binutils@sourceware.org \
--cc=gkm@rivosinc.com \
--cc=lifang_xia@linux.alibaba.com \
--cc=nelson@rivosinc.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).