[FFmpeg-devel] [PATCH v3 7/7] avutil/la: Add function performance testing

Shiyou Yin yinshiyou-hf at loongson.cn
Thu May 25 09:58:59 EEST 2023



> 2023年5月25日 10:36,Hao Chen <chenhao at loongson.cn> 写道:
> 
> 
> 在 2023/5/24 下午7:03, Rémi Denis-Courmont 写道:
>> 
>> Le 24 mai 2023 10:39:59 GMT+03:00, Hao Chen <chenhao at loongson.cn> a écrit :
>>> 在 2023/5/20 下午5:38, Rémi Denis-Courmont 写道:
>>>> Le lauantaina 20. toukokuuta 2023, 10.27.19 EEST Hao Chen a écrit :
>>>>> From: yuanhecai <yuanhecai at loongson.cn>
>>>>> 
>>>>> This patch supports the use of the "checkasm --bench" testing feature
>>>>> on loongarch platform.
>>>>> 
>>>>> Change-Id: I42790388d057c9ade0dfa38a19d9c1fd44ca0bc3
>>>>> ---
>>>>>   libavutil/loongarch/timer.h | 48 +++++++++++++++++++++++++++++++++++++
>>>>>   libavutil/timer.h           |  2 ++
>>>>>   2 files changed, 50 insertions(+)
>>>>>   create mode 100644 libavutil/loongarch/timer.h
>>>>> 
>>>>> diff --git a/libavutil/loongarch/timer.h b/libavutil/loongarch/timer.h
>>>>> new file mode 100644
>>>>> index 0000000000..44ed786409
>>>>> --- /dev/null
>>>>> +++ b/libavutil/loongarch/timer.h
>>>>> @@ -0,0 +1,48 @@
>>>>> +/*
>>>>> + * Copyright (c) 2023 Loongson Technology Corporation Limited
>>>>> + * Contributed by Hecai Yuan <yuanhecai at loongson.cn>
>>>>> + *
>>>>> + * This file is part of FFmpeg.
>>>>> + *
>>>>> + * FFmpeg is free software; you can redistribute it and/or
>>>>> + * modify it under the terms of the GNU Lesser General Public
>>>>> + * License as published by the Free Software Foundation; either
>>>>> + * version 2.1 of the License, or (at your option) any later version.
>>>>> + *
>>>>> + * FFmpeg is distributed in the hope that it will be useful,
>>>>> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
>>>>> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
>>>>> + * Lesser General Public License for more details.
>>>>> + *
>>>>> + * You should have received a copy of the GNU Lesser General Public
>>>>> + * License along with FFmpeg; if not, write to the Free Software
>>>>> + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301
>>>>> USA + */
>>>>> +
>>>>> +#ifndef AVUTIL_LOONGARCH_TIMER_H
>>>>> +#define AVUTIL_LOONGARCH_TIMER_H
>>>>> +
>>>>> +#include <stdint.h>
>>>>> +#include "config.h"
>>>>> +
>>>>> +#if HAVE_INLINE_ASM
>>>>> +
>>>>> +#define AV_READ_TIME read_time
>>>>> +
>>>>> +static inline uint64_t read_time(void)
>>>>> +{
>>>>> +
>>>>> +#if ARCH_LOONGARCH64
>>>>> +    uint64_t a, id = 0;
>>>> Initial value is never used.
>>>> 
>>>>> +    __asm__ volatile ( "rdtime.d  %0, %1" : "=r"(a), "=r"(id) :: "memory"
>>>>> ); +    return a;
>>>>> +#else
>>>>> +    uint32_t a, id = 0;
>>>>> +    __asm__ volatile ( "rdtimel.w  %0, %1" : "=r"(a), "=r"(id) :: "memory"
>>>>> ); +    return (uint64_t)a;
>>>>> +#endif
>>>> Why do you clobber memory here?
>>>> 
>>>>> +}
>>>>> +
>>>>> +#endif /* HAVE_INLINE_ASM */
>>>>> +
>>>>> +#endif /* AVUTIL_LOONGARCH_TIMER_H */
>>>>> diff --git a/libavutil/timer.h b/libavutil/timer.h
>>>>> index d3db5a27ef..861ba7e9d7 100644
>>>>> --- a/libavutil/timer.h
>>>>> +++ b/libavutil/timer.h
>>>>> @@ -61,6 +61,8 @@
>>>>>   #   include "riscv/timer.h"
>>>>>   #elif ARCH_X86
>>>>>   #   include "x86/timer.h"
>>>>> +#elif ARCH_LOONGARCH
>>>>> +#   include "loongarch/timer.h"
>>>>>   #endif
>>>>> 
>>>>>   #if !defined(AV_READ_TIME)
>>>>> 
>>>>     Thanks for your advice.  As described in loongarch's instruction manual, the rdtime.d instruction is used as follows:
>>>> rdtime.d rd, rj. The rj register stores the counter ID. In this application, the value of counter ID is equal to 0.
>> You're setting a value, zero, to a variable `id`, that is then used as output operand. As far as the compiler is concerned, the value zero is never used and the initialisation can be elided. The value of register %1 is unspecified.
>> 
>> If you meant for `id` to be an input operand, the constraints are incorrect.
>> 
> 
> 
> You are right! Thank you very much for your reminder. I will correct it.
> 

 id is output operand, the constraints is correct, and initilazation of id is not necessary.



More information about the ffmpeg-devel mailing list