[FFmpeg-devel] [PATCH 1/3] avutil/mem: make max_alloc_size an atomic type

James Almer jamrial at gmail.com
Mon May 31 17:02:30 EEST 2021


On 5/31/2021 5:09 AM, Anton Khirnov wrote:
> Quoting James Almer (2021-05-23 00:09:02)
>> This is in preparation for the following commit.
>>
>> Signed-off-by: James Almer <jamrial at gmail.com>
>> ---
>>   libavutil/mem.c | 17 +++++++++++------
>>   1 file changed, 11 insertions(+), 6 deletions(-)
>>
>> diff --git a/libavutil/mem.c b/libavutil/mem.c
>> index fa227f5e12..c12c24aa90 100644
>> --- a/libavutil/mem.c
>> +++ b/libavutil/mem.c
>> @@ -31,6 +31,7 @@
>>   #include <limits.h>
>>   #include <stdint.h>
>>   #include <stdlib.h>
>> +#include <stdatomic.h>
>>   #include <string.h>
>>   #if HAVE_MALLOC_H
>>   #include <malloc.h>
>> @@ -68,17 +69,17 @@ void  free(void *ptr);
>>    * dynamic libraries and remove -Wl,-Bsymbolic from the linker flags.
>>    * Note that this will cost performance. */
>>   
>> -static size_t max_alloc_size= INT_MAX;
>> +static atomic_size_t max_alloc_size = ATOMIC_VAR_INIT(INT_MAX);
>>   
>>   void av_max_alloc(size_t max){
>> -    max_alloc_size = max;
>> +    atomic_store_explicit(&max_alloc_size, max, memory_order_relaxed);
> 
> Any specific reason for using a non-default memory order?
> AFAIK it is recommended by the spec authors to use default (sequentially
> consistent) operations unless there is a very good reason to do
> something else.

I copied the existing behavior of av_force_cpu_flags() and 
av_get_cpu_flags(), which seems to be enough for a similar atomic 
variable as max_alloc_size.


More information about the ffmpeg-devel mailing list