[x264-devel] [PATCH 02/24] checkasm: Check the right output range for integral_initXh

[Martin Storsjö]

On Mon, 24 Aug 2015, Janne Grunau wrote:

> On 2015-08-24 13:29:35 +0300, Martin Storsjö wrote:
>> On Sat, 22 Aug 2015, Janne Grunau wrote:
>>
>>> On 2015-08-13 23:59:23 +0300, Martin Storsjö wrote:
>>>> These functions write their output into sum+stride, while we previously
>>>> only checked [0..stride-8] within the sum array.
>>>>
>>>> This catches the previously broken aarch64 version of these functions.
>>>> ---
>>>> tools/checkasm.c |   14 +++++++-------
>>>> 1 file changed, 7 insertions(+), 7 deletions(-)
>>>>
>>>> diff --git a/tools/checkasm.c b/tools/checkasm.c
>>>> index a1e8eda..efe874b 100644
>>>> --- a/tools/checkasm.c
>>>> +++ b/tools/checkasm.c
>>>> @@ -1616,7 +1616,7 @@ static int check_mc( int cpu_ref, int cpu_new )
>>>>         report( "lowres init :" );
>>>>     }
>>>>
>>>> -#define INTEGRAL_INIT( name, size, ... )\
>>>> +#define INTEGRAL_INIT( name, size, offset, ... )\
>>>>     if( mc_a.name != mc_ref.name )\
>>>>     {\
>>>>         intptr_t stride = 96;\
>>>> @@ -1628,17 +1628,17 @@ static int check_mc( int cpu_ref, int cpu_new )
>>>>         call_c1( mc_c.name, __VA_ARGS__ );\
>>>>         sum = (uint16_t*)buf4;\
>>>>         call_a1( mc_a.name, __VA_ARGS__ );\
>>>> -        if( memcmp( buf3, buf4, (stride-8)*2 ) \
>>>> -            || (size>9 && memcmp( buf3+18*stride, buf4+18*stride, (stride-8)*2 )))\
>>>> +        if( memcmp( buf3, buf4, (offset+stride-8)*2 ) \
>>>
>>> I'd prefer if we separate the two parameters and move the out of
>>> __VA_ARGS__. Another option would be to use cmp_len which would allow us
>>> to test the last 4 values of integral_init4h. We can not just compare
>>> stride values since all asm variants overwrite and C does not
>>
>> Hmmm, so we'd do call_a1( mc_a.name, sum+offset, __VA_ARGS )
>> instead, and pass cmp_len as a separate parameter? That'd probably
>> work, and would certainly look nicer.
>
> I was more thinking of call_a1( mc_a.name, sum, offset, __VA_ARGS ) and
> I think you have to do that since you can't use sum in the memcmp().  Or
> even omit sum completely since passing the offset is enough.

Umm, either I'm not understanding, or you mean INTEGRAL_INIT( name, size, 
sum, offset, <other params> )? That's what I meant also; that this 
parameter (currently sum+offset) gets moved from __VA_ARGS__ to the fixed 
parameters to the INTEGRAL_INIT macro, and then has to be declared 
manually when calling call_a1 (where it currently is part of __VA_ARGS__).

// Martin