[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH 01/11] tci: Fix build regression

From: Stefan Weil
Subject: Re: [Qemu-devel] [PATCH 01/11] tci: Fix build regression
Date: Thu, 7 Apr 2016 22:37:26 +0200
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Icedove/38.7.0

Am 07.04.2016 um 21:16 schrieb Stefan Weil:
> Am 07.04.2016 um 20:15 schrieb Richard Henderson:
>> On 04/07/2016 08:53 AM, Sergey Fedorov wrote:
>>> +/* Enable TCI assertions only when debugging TCG (and without NDEBUG
>>> defined).
>>> + * Without assertions, the interpreter runs much faster. */
>>> +#if defined(CONFIG_DEBUG_TCG)
>>> +# define tci_assert(cond) assert(cond)
>>> +#else
>>> +# define tci_assert(cond) ((void)0)
>>>   #endif
>> Please just use tcg_debug_assert.
>> r~
> Hi Richard,
> that's a good suggestion, but maybe a little late for 2.6-rc2.
> I already sent a pull request an hour ago after Michael had added
> his tested-by.
> My first priority is fixing the build regression in 2.6. I can
> try to prepare a new patch, wait for reviews and send a pull
> request, but I am afraid this might not be finished in time
> for 2.6.
> Regards
> Stefan

I just tested a variant with tcg_debug_assert. It creates less efficient
code when debugging is disabled. Here is the code size for x86_64:

with normal tcg_debug_assert:

   text       data        bss        dec        hex    filename
   8293          0        128       8421       20e5   

In gdb I can also see assembler code for op_size at the beginning
of the for loop in tcg_qemu_tb_exec. This slows down the interpreter.

with ((void)0):

   text       data        bss        dec        hex    filename
   8135          0        128       8263       2047   

Therefore I'd prefer using my pull request for 2.6.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]