This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [patch] Disable if_conversion2 for Og
- From: Richard Earnshaw <rearnsha at arm dot com>
- To: Joey Ye <Joey dot Ye at arm dot com>
- Cc: "gcc-patches at gcc dot gnu dot org" <gcc-patches at gcc dot gnu dot org>
- Date: Wed, 16 Apr 2014 11:32:56 +0100
- Subject: Re: [patch] Disable if_conversion2 for Og
- Authentication-results: sourceware.org; auth=none
- References: <000001cf584e$66102060$32306120$ at arm dot com> <534E50F2 dot 2000606 at arm dot com> <000401cf595a$ee875910$cb960b30$ at arm dot com> <534E558E dot 6010206 at arm dot com> <45520D6299C11E4588128526465332BB436C04503E at SAROVARA dot Asiapac dot Arm dot com> <534E5996 dot 4060509 at arm dot com> <000601cf595e$db964510$92c2cf30$ at arm dot com>
On 16/04/14 11:30, Joey Ye wrote:
>
>
>> -----Original Message-----
>> From: Richard Earnshaw
>> Sent: Wednesday, April 16, 2014 6:21 PM
>> To: Joey Ye
>> Cc: gcc-patches@gcc.gnu.org
>> Subject: Re: [patch] Disable if_conversion2 for Og
>>
>> On 16/04/14 11:17, Joey Ye wrote:
>>>> -----Original Message-----
>>>> From: Richard Earnshaw
>>>> Sent: Wednesday, April 16, 2014 6:04 PM
>>>> To: Joey Ye
>>>> Cc: gcc-patches@gcc.gnu.org
>>>> Subject: Re: [patch] Disable if_conversion2 for Og
>>>>
>>>> On 16/04/14 11:02, Joey Ye wrote:
>>>>>
>>>>>
>>>>>> -----Original Message-----
>>>>>> From: Richard Earnshaw
>>>>>> Sent: Wednesday, April 16, 2014 5:44 PM
>>>>>> To: Joey Ye
>>>>>> Cc: gcc-patches@gcc.gnu.org
>>>>>> Subject: Re: [patch] Disable if_conversion2 for Og
>>>>>>
>>>>>> Arguably, this is a bug in gdb. The debugger should understand
>>>>>> when a breakpointed conditional instruction is not going to execute
>>>>>> and silently continue. That preserves the illusion of not
>>>>>> executing the code without requiring the compiler to de-optimize
>> things.
>>>>>>
>>>>>> R.
>>>>> Or compiler just optimizes it, and emits generic DWARFx information
>>>>> to help GDB handle it in more target independently?
>>>>>
>>>>> - Joey
>>>>>
>>>>
>>>> I'm not sure extra dwarf info would help much. The debugger still
>>>> has to understand that the breakpoint has not really been hit.
>>>>
>>>> R.
>>> Yes, it is inevitable. But without extra dwarf info it will be even more
> painful:
>> each time setting break-point or break-point hits it has to decode the
> break-
>> pointed instructions and its context to search for conditional execution
> and IT
>> blocks.
>>>
>>
>> For thumb code it can get the conditional information it needs from the IT
>> state in the PSR; for ARM code it has to look no further than the
> instruction
>> itself.
>>
>> R.
> The thing is, debugger has to do this for every breakpoint, even though more
> of them are not conditional execution, which isn't efficient.
>
>
Then cache whether the instruction might be conditional when it's
created. That's more work when creating the bp (and you do have to scan
back to check if a thumb instruction might be made conditional with IT),
but would save time when hitting it.
Anyway, this is getting off-topic for the GCC-patches list.
R.