[libcamera-devel] [PATCH 3/3] ipa: ipu3: af: Remove redundant memcpy

[Kieran Bingham]

Quoting Laurent Pinchart (2022-03-23 01:24:45)
> Hi Kieran,
> 
> Thank you for the patch.
> 
> On Tue, Mar 15, 2022 at 05:53:45PM +0000, Kieran Bingham via libcamera-devel wrote:
> > The af statistics can be accessed directly from the mapped buffer.
> > Remove the redundant memcpy, and simplify the call to
> > afEstimateVariance().
> > 
> > Signed-off-by: Kieran Bingham <kieran.bingham at ideasonboard.com>
> > ---
> >  src/ipa/ipu3/algorithms/af.cpp | 12 ++++--------
> >  src/ipa/ipu3/algorithms/af.h   |  2 +-
> >  2 files changed, 5 insertions(+), 9 deletions(-)
> > 
> > diff --git a/src/ipa/ipu3/algorithms/af.cpp b/src/ipa/ipu3/algorithms/af.cpp
> > index 469762a29937..e10cff968895 100644
> > --- a/src/ipa/ipu3/algorithms/af.cpp
> > +++ b/src/ipa/ipu3/algorithms/af.cpp
> > @@ -336,7 +336,7 @@ void Af::afIgnoreFrameReset()
> >   *
> >   * \return The variance of the values in the data set \a y_item selected by \a isY1
> >   */
> > -double Af::afEstimateVariance(y_table_item_t *y_item, uint32_t len,
> > +double Af::afEstimateVariance(const y_table_item_t *y_item, uint32_t len,
> >                             bool isY1)
> 
> Not a candidate for this patch, but pointer + len immediately makes me
> think the Span class should be used instead.

I agree, I think the Span would help improve the iterators in here too.
Will do on top.

> 
> Reviewed-by: Laurent Pinchart <laurent.pinchart at ideasonboard.com>
> 
> >  {
> >       uint32_t z = 0;
> > @@ -406,25 +406,21 @@ bool Af::afIsOutOfFocus(IPAContext context)
> >   */
> >  void Af::process(IPAContext &context, const ipu3_uapi_stats_3a *stats)
> >  {
> > -     y_table_item_t y_item[IPU3_UAPI_AF_Y_TABLE_MAX_SIZE / sizeof(y_table_item_t)];
> > +     const y_table_item_t *y_item = reinterpret_cast<const y_table_item_t *>(&stats->af_raw_buffer.y_table);
> >       uint32_t afRawBufferLen;
> >  
> >       /* Evaluate the AF buffer length */
> >       afRawBufferLen = context.configuration.af.afGrid.width *
> >                        context.configuration.af.afGrid.height;
> >  
> > -     memcpy(y_item, stats->af_raw_buffer.y_table,
> > -            afRawBufferLen * sizeof(y_table_item_t));
> > +     ASSERT(afRawBufferLen / sizeof(y_table_item_t) < IPU3_UAPI_AF_Y_TABLE_MAX_SIZE);

Looking at 
include/linux/intel-ipu3.h:     __u8 y_table[IPU3_UAPI_AF_Y_TABLE_MAX_SIZE] __attribute__((aligned(32)));

IPU3_UAPI_AF_Y_TABLE_MAX_SIZE is the maximum number of entries of y_table_item_t

So I think this ASSERT is wrong, and should be:

	ASSERT(afRawBufferLen < IPU3_UAPI_AF_Y_TABLE_MAX_SIZE);

> >  
> >       /*
> >        * Calculate the mean and the variance of AF statistics for a given grid.
> >        * For coarse: y1 are used.
> >        * For fine: y2 results are used.
> >        */
> > -     if (coarseCompleted_)
> > -             currentVariance_ = afEstimateVariance(y_item, afRawBufferLen, false);
> > -     else
> > -             currentVariance_ = afEstimateVariance(y_item, afRawBufferLen, true);
> > +     currentVariance_ = afEstimateVariance(y_item, afRawBufferLen, !coarseCompleted_);
> >  
> >       if (!context.frameContext.af.stable) {
> >               afCoarseScan(context);
> > diff --git a/src/ipa/ipu3/algorithms/af.h b/src/ipa/ipu3/algorithms/af.h
> > index 906f2843dd49..3b5758e868ea 100644
> > --- a/src/ipa/ipu3/algorithms/af.h
> > +++ b/src/ipa/ipu3/algorithms/af.h
> > @@ -41,7 +41,7 @@ private:
> >       void afReset(IPAContext &context);
> >       bool afNeedIgnoreFrame();
> >       void afIgnoreFrameReset();
> > -     double afEstimateVariance(y_table_item_t *y_item, uint32_t len,
> > +     double afEstimateVariance(const y_table_item_t *y_item, uint32_t len,
> >                                 bool isY1);
> >       bool afIsOutOfFocus(IPAContext context);
> >  
> 
> -- 
> Regards,
> 
> Laurent Pinchart