public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [PATCH] gcc/config/c6x/c6x.md: Remove "clobber (match_scratch ...)" in "movmisalign<mode>_store".
@ 2015-03-27  0:05 Chen Gang
  2015-03-27  0:18 ` Chen Gang
  2015-03-27 13:04 ` Bernd Schmidt
  0 siblings, 2 replies; 5+ messages in thread
From: Chen Gang @ 2015-03-27  0:05 UTC (permalink / raw)
  To: Jeff Law, bernds; +Cc: gcc-patches List

For misalignment memory access, c6x gcc will cause issue, so need remove
"clobber (match_scratch ...)" which will be symmetric with "movmisalign
<mode>_load", then pass compiling and generate correct assembly code.

The related issue:

  [root@localhost c6x]# cat test.i
  int oxu_driver_init(void)
  {
   *(volatile unsigned int *)(-1) = *(unsigned int *)(-1);
  }

  [root@localhost c6x]# /upstream/release-c6x/libexec/gcc/tic6x-gchen-elf/5.0.0/cc1 -Os test.i
   oxu_driver_init
  test.i: In function 'oxu_driver_init':
  test.i:4:1: error: unrecognizable insn:
   }
   ^
  (insn 9 8 12 2 (set (mem/v:SI (reg/f:SI 77) [1 MEM[(volatile unsigned int *)4294967295B]+0 S4 A8])
          (unspec:SI [
                  (reg:SI 73 [ D.1542 ])
              ] UNSPEC_MISALIGNED_ACCESS)) test.i:3 -1
       (nil))
  test.i:4:1: internal compiler error: in extract_insn, at recog.c:2343
  0x9f8645 _fatal_insn(char const*, rtx_def const*, char const*, int, char const*)
          ../../gcc-c6x/gcc/rtl-error.c:110
  0x9f8679 _fatal_insn_not_found(rtx_def const*, char const*, int, char const*)
          ../../gcc-c6x/gcc/rtl-error.c:118
  0x9ce127 extract_insn(rtx_insn*)
          ../../gcc-c6x/gcc/recog.c:2343
  0x7c759b instantiate_virtual_regs_in_insn
          ../../gcc-c6x/gcc/function.c:1598
  0x7c759b instantiate_virtual_regs
          ../../gcc-c6x/gcc/function.c:1966
  0x7c759b execute
          ../../gcc-c6x/gcc/function.c:2015
  Please submit a full bug report,

Related output after fix (I guess it is OK, please check):

        .file   "test.i"
        .c6xabi_attribute Tag_ABI_array_object_alignment, 0
        .c6xabi_attribute Tag_ABI_array_object_align_expected, 0
        .c6xabi_attribute Tag_ABI_stack_align_needed, 0
        .c6xabi_attribute Tag_ABI_stack_align_preserved, 0
        .c6xabi_attribute Tag_ABI_conformance, "1.0"
  .text;
        .align  2
        .global oxu_driver_init
        .type   oxu_driver_init, @function
  oxu_driver_init:
                mvk     .d1     11, A4
                ldnw    .d1t1   *A4, A3
                nop     4
                stnw    .d1t1   A3, *A4
                ret     .s2     B3
                nop     5
        .size   oxu_driver_init, .-oxu_driver_init
        .ident  "GCC: (GNU) 5.0.0 20150321 (experimental)"



2015-03-27  Chen Gang  <gang.chen.5i5j@gmail.com>

	* config/c6x/c6x.md (movmisalign<mode>_store): Remove "clobber
	(match_scratch ...)".
---
 gcc/config/c6x/c6x.md | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/gcc/config/c6x/c6x.md b/gcc/config/c6x/c6x.md
index 892aac4..6a3157f 100644
--- a/gcc/config/c6x/c6x.md
+++ b/gcc/config/c6x/c6x.md
@@ -785,8 +785,7 @@
 (define_insn_and_split "movmisalign<mode>_store"
   [(set (match_operand:SIDIVM 0 "memory_operand" "=W,Q,T,Q,T")
 	(unspec:SIDIVM [(match_operand:SIDIVM 1 "register_operand" "r,a,b,b,a")]
-		       UNSPEC_MISALIGNED_ACCESS))
-   (clobber (match_scratch:SI 2 "=r,X,X,X,X"))]
+		       UNSPEC_MISALIGNED_ACCESS))]
   "TARGET_INSNS_64"
   "@
    #
-- 
1.9.3

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH] gcc/config/c6x/c6x.md: Remove "clobber (match_scratch ...)" in "movmisalign<mode>_store".
  2015-03-27  0:05 [PATCH] gcc/config/c6x/c6x.md: Remove "clobber (match_scratch ...)" in "movmisalign<mode>_store" Chen Gang
@ 2015-03-27  0:18 ` Chen Gang
  2015-03-27 13:04 ` Bernd Schmidt
  1 sibling, 0 replies; 5+ messages in thread
From: Chen Gang @ 2015-03-27  0:18 UTC (permalink / raw)
  To: Jeff Law, bernds; +Cc: gcc-patches List



On 3/27/15 08:05, Chen Gang wrote:
> For misalignment memory access, c6x gcc will cause issue, so need remove
> "clobber (match_scratch ...)" which will be symmetric with "movmisalign
> <mode>_load", then pass compiling and generate correct assembly code.
> 
> The related issue:
> 
>   [root@localhost c6x]# cat test.i
>   int oxu_driver_init(void)
>   {
>    *(volatile unsigned int *)(-1) = *(unsigned int *)(-1);

Oh, sorry, I use 11 instead of -1 for generating the assembly code (-1
and 11 lead the same result).

[...]

>   oxu_driver_init:
>                 mvk     .d1     11, A4
>                 ldnw    .d1t1   *A4, A3
>                 nop     4
>                 stnw    .d1t1   A3, *A4
>                 ret     .s2     B3
>                 nop     5
>         .size   oxu_driver_init, .-oxu_driver_init
>         .ident  "GCC: (GNU) 5.0.0 20150321 (experimental)"
> 
[...]

Thanks.
-- 
Chen Gang

Open, share, and attitude like air, water, and life which God blessed

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH] gcc/config/c6x/c6x.md: Remove "clobber (match_scratch ...)" in "movmisalign<mode>_store".
  2015-03-27  0:05 [PATCH] gcc/config/c6x/c6x.md: Remove "clobber (match_scratch ...)" in "movmisalign<mode>_store" Chen Gang
  2015-03-27  0:18 ` Chen Gang
@ 2015-03-27 13:04 ` Bernd Schmidt
  2015-03-28  1:32   ` Chen Gang
  1 sibling, 1 reply; 5+ messages in thread
From: Bernd Schmidt @ 2015-03-27 13:04 UTC (permalink / raw)
  To: Chen Gang, Jeff Law; +Cc: gcc-patches List

[-- Attachment #1: Type: text/plain, Size: 612 bytes --]

On 03/27/2015 01:05 AM, Chen Gang wrote:
> For misalignment memory access, c6x gcc will cause issue, so need remove
> "clobber (match_scratch ...)" which will be symmetric with "movmisalign
> <mode>_load", then pass compiling and generate correct assembly code.
>

> 	* config/c6x/c6x.md (movmisalign<mode>_store): Remove "clobber
> 	(match_scratch ...)".

No, that just will make the compiler confuse loads and stores. I've 
committed the following to fix it (I thought I'd done so a year ago, but 
probably it was one of those commit against an out-of-date tree 
situations and it didn't go through).


Bernd


[-- Attachment #2: misalign.diff --]
[-- Type: text/x-patch, Size: 789 bytes --]

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 8e4b6c1..d5535f9 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,5 +1,8 @@
 2015-03-27  Bernd Schmidt  <bernds@codesourcery.com>
 
+	* config/c6x/c6x.md (movmisalign<mode>): Use MEM_P, not
+	memory_operand.
+
 	PR target/65052
         * config/c6x/constraints.md (S3): New constraint.
         * config/c6x/c6x.md (real_jump): Use it.
diff --git a/gcc/config/c6x/c6x.md b/gcc/config/c6x/c6x.md
index fafefa6..e957eca 100644
--- a/gcc/config/c6x/c6x.md
+++ b/gcc/config/c6x/c6x.md
@@ -775,7 +775,7 @@
 		       UNSPEC_MISALIGNED_ACCESS))]
   "TARGET_INSNS_64"
 {
-  if (memory_operand (operands[0], <MODE>mode))
+  if (MEM_P (operands[0]))
     {
       emit_insn (gen_movmisalign<mode>_store (operands[0], operands[1]));
       DONE;

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH] gcc/config/c6x/c6x.md: Remove "clobber (match_scratch ...)" in "movmisalign<mode>_store".
  2015-03-27 13:04 ` Bernd Schmidt
@ 2015-03-28  1:32   ` Chen Gang
  2015-03-28  6:24     ` Chen Gang
  0 siblings, 1 reply; 5+ messages in thread
From: Chen Gang @ 2015-03-28  1:32 UTC (permalink / raw)
  To: Bernd Schmidt, Jeff Law; +Cc: gcc-patches List

On 3/27/15 21:03, Bernd Schmidt wrote:
> On 03/27/2015 01:05 AM, Chen Gang wrote:
>> For misalignment memory access, c6x gcc will cause issue, so need remove
>> "clobber (match_scratch ...)" which will be symmetric with "movmisalign
>> <mode>_load", then pass compiling and generate correct assembly code.
>>
> 
>>     * config/c6x/c6x.md (movmisalign<mode>_store): Remove "clobber
>>     (match_scratch ...)".
> 
> No, that just will make the compiler confuse loads and stores. I've committed the following to fix it (I thought I'd done so a year ago, but probably it was one of those commit against an out-of-date tree situations and it didn't go through).
> 

OK, thanks. Your patch is OK to me. And I shall try to find another c6x
patches within this month (2015-03-31).


Thanks.
-- 
Chen Gang

Open, share, and attitude like air, water, and life which God blessed

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH] gcc/config/c6x/c6x.md: Remove "clobber (match_scratch ...)" in "movmisalign<mode>_store".
  2015-03-28  1:32   ` Chen Gang
@ 2015-03-28  6:24     ` Chen Gang
  0 siblings, 0 replies; 5+ messages in thread
From: Chen Gang @ 2015-03-28  6:24 UTC (permalink / raw)
  To: Bernd Schmidt, Jeff Law; +Cc: gcc-patches List

On 3/28/15 09:32, Chen Gang wrote:
> On 3/27/15 21:03, Bernd Schmidt wrote:
>> On 03/27/2015 01:05 AM, Chen Gang wrote:
>>> For misalignment memory access, c6x gcc will cause issue, so need remove
>>> "clobber (match_scratch ...)" which will be symmetric with "movmisalign
>>> <mode>_load", then pass compiling and generate correct assembly code.
>>>
>>
>>>     * config/c6x/c6x.md (movmisalign<mode>_store): Remove "clobber
>>>     (match_scratch ...)".
>>
>> No, that just will make the compiler confuse loads and stores. I've committed the following to fix it (I thought I'd done so a year ago, but probably it was one of those commit against an out-of-date tree situations and it didn't go through).
>>
> 
> OK, thanks. Your patch is OK to me. And I shall try to find another c6x
> patches within this month (2015-03-31).
> 

I have reported the related bug 65510, please continue for it, if I need
to continue, please let me know.

Thanks.
-- 
Chen Gang

Open, share, and attitude like air, water, and life which God blessed

^ permalink raw reply	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2015-03-28  6:24 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-03-27  0:05 [PATCH] gcc/config/c6x/c6x.md: Remove "clobber (match_scratch ...)" in "movmisalign<mode>_store" Chen Gang
2015-03-27  0:18 ` Chen Gang
2015-03-27 13:04 ` Bernd Schmidt
2015-03-28  1:32   ` Chen Gang
2015-03-28  6:24     ` Chen Gang

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).