On 08/21/2012 01:26 PM, Kostya Shishkov wrote:
> On Tue, Aug 21, 2012 at 12:36:49PM -0400, Justin Ruggles wrote:
>> Fixes crash on x86 due to alignment requirements for w_data in
>> lpc_apply_welch_window_sse2().
>> ---
>>  libavcodec/lpc.c |   15 +++++++--------
>>  libavcodec/lpc.h |    1 +
>>  2 files changed, 8 insertions(+), 8 deletions(-)
>>
>> diff --git a/libavcodec/lpc.c b/libavcodec/lpc.c
>> index 6ce5dde..98011d9 100644
>> --- a/libavcodec/lpc.c
>> +++ b/libavcodec/lpc.c
>> @@ -178,11 +178,9 @@ int ff_lpc_calc_coefs(LPCContext *s,
>>      }
>>  
>>      if (lpc_type == FF_LPC_TYPE_LEVINSON) {
>> -        double *windowed_samples = s->windowed_samples + max_order;
>> +        s->lpc_apply_welch_window(samples, blocksize, s->windowed_samples);
>>  
>> -        s->lpc_apply_welch_window(samples, blocksize, windowed_samples);
>> -
>> -        s->lpc_compute_autocorr(windowed_samples, blocksize, max_order, 
>> autoc);
>> +        s->lpc_compute_autocorr(s->windowed_samples, blocksize, max_order, 
>> autoc);
>>  
>>          compute_lpc_coefs(autoc, max_order, &lpc[0][0], MAX_LPC_ORDER, 0, 
>> 1);
>>  
>> @@ -248,10 +246,11 @@ av_cold int ff_lpc_init(LPCContext *s, int blocksize, 
>> int max_order,
>>      s->lpc_type  = lpc_type;
>>  
>>      if (lpc_type == FF_LPC_TYPE_LEVINSON) {
>> -        s->windowed_samples = av_mallocz((blocksize + max_order + 2) *
>> -                                         sizeof(*s->windowed_samples));
>> -        if (!s->windowed_samples)
>> +        s->windowed_buffer = av_mallocz((blocksize + 2 + FFALIGN(max_order, 
>> 2)) *
>> +                                        sizeof(*s->windowed_samples));
>> +        if (!s->windowed_buffer)
>>              return AVERROR(ENOMEM);
>> +        s->windowed_samples = s->windowed_buffer + FFALIGN(max_order, 2);
> 
> nit: maybe align to 4? Because you'll step on the same rakes with AVX 
> probably.
> 
> in general LGTM

Anyone who is modifying a function to add an AVX version will still have
to look at all uses of that function to check for proper alignment...
but I suppose it can't hurt to go ahead and account for it now.

-Justin
_______________________________________________
libav-devel mailing list
libav-devel@libav.org
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to