GitLab

This avoids both function call overhead and extra checks for values that
are outside MININT/MAXINT range, for the uses inside GeglBuffer we only
care about coordinates that are representable as integers - thus these
inline-able shortcut versions are good enough for us.

The speedup seems to consistently make all samplers about 10% faster
according to the tests in perf/

This reverts earlier indentation breaking changes when GeglRectangle was
renamed GeglBufferRectangle this change also retains the existing released API
and ABI more exactly.

The 2x2 matrix is only used by GeglBuffer and its API, gimp-2.10 does not
use the symbols - we are remaining abi compatible but are rearranging
locations of symbols.

The typedef for GeglRectangle is now:
  typedef struct _GeglBufferRectangle GeglRectangle;

This reduces the throughpput of gegl buffer accessors and more slightly in
the single-threaded case, which is now becoming rare - at least when GEGL
is operated through GIMP.

Relying less on this dynamically changing parameter from gegl_config()
will also make it easier to split GeglBuffer out in its own project.

Use the new gegl_tile_read_lock() and gegl_tile_read_unlock()
functions, added in the previous commit, before/after reading from
a tile in a potentially-multithreaded context.

See the previous commit for an explanation of why this is
necessary.

Note, in particular, that this may make 1x1 gegl_buffer_get()s
(and NEAREST gegl_buffer_sample()) more expensive, however, NEAREST
sampler objects amortize the cost of these functions by only
locking/unlocking the sampler's hot-tile when switching tiles.  As
a result, a concurrent write to the same buffer may block while the
sampler object is alive.  We assume that buffers may not be
written-to while there are active sampler objects constructed for
them (which can already lead to stale data in the samplers).

This reverts commit c0d3ce82.

Through profiling - it turns out that always taking the babl fish path, rather
than checking if memcpy is better - is better.

tidies up lose #if 0'd out code, it now is a straightforward per sampler
hot-tile.

Using memcpy directly should be lower overhead than invoking a fish for a
single pixel when pixelformats match.

In GeglSamplerNearest, use a per-instance hot-tile pointer, rather
than using the tile-storage's common hot-tile pointer, so that
sampling from the sampler's hot-tile doesn't require locking the
tile storage.  This notably improves multithreaded performance of
ops that use a nearest sampler with a more-or-less regular sampling
pattern.

Adds a level argument to gegl_tile_indice, this makes more tests pass.

... when not using multithreading

Marking some functions that should be skipped, and updating rename-to syntax.

Some use of mutexes is a performance impact we want to minimize in the
non-threaded case.

This fixes a regression in GIMPs live transform preview.

When using gegl_buffer_get / gegl_buffer_set and gegl_buffer_sample for
small/1x1px sized rectangles the overhead of each involved function call
amounts to more than 2% of time spend processing.

This refactoring involves moving gegl_buffer_sample from gegl-buffer.c to
gegl-sampler.c to make them live in the same compilation unit. Replacing
internal nested chains of function calls with added complexity with an
inlinable implementation using switches.

Implement an adaptive caching policy; that meets the needs of affine
transformations. Running averages are kept of x/y deltas as well as the
magnitudes of the deltas of coordinates leading to cache misses.

The initial rectangle used is a small one; subsequent rectangles have extents
up to 64 pixels wide/tall; when tracked deltas are more diagonal than axis
aligned smaller more square regions are tracked.

The GeglSampler base-class is micro optimized for affine transformations at
the moment, for now, this short-circuit of the nearest sampler implementation
avoids a lot of redundant elbow-room caching.

sampler-nearest: RaGaBaA like other samplers, use local array to pass back values, use gegl_sampler_get_ptr to get values from buffer

sampler-nearest.c: for some strange reason, enlarging width=height to 3 (from 1) make it interact better with GIMP: no half-pixel wide margin when enlarging