[rtextures] Optimize ImageClearBackground and ImageDrawRectangleRec with doubling strategy

[ChocolateChipKookie]

Background

I use the ImageDraw* functions in a slightly unconventional way, and while doing so I noticed that ImageClearBackground is over an order of magnitude slower than simply calling ImageDrawRectangle over the entire image.

Issue

ImageClearBackground calls memcpy once per pixel and doesn’t take advantage of the fact that memcpy is very fast when copying large contiguous blocks of memory. As a result, it incurs the overhead of a function call for every pixel.

Fix

Instead of copying one pixel at a time, we can iteratively copy the data that has already been written to the image buffer, effectively doubling the amount of initialized data with each iteration. This reduces the number of memcpy calls from O(N) to O(log N).

I applied the same optimization to ImageDrawRectangleRec. The improvement there is smaller because only the first row is written pixel-by-pixel.

Result

15-20 times faster execution times on my PC.

Before change:
Image Size: 1024x1024 (1048576 pixels)
--------------------------------------------------------------------------------
  GRAYSCALE                     :  3.615 ms/call,      276.65 MB/s
  GRAY_ALPHA                    :  3.323 ms/call,      601.83 MB/s
  R5G6B5                        :  3.323 ms/call,      601.79 MB/s
  R8G8B8                        :  3.335 ms/call,      899.66 MB/s
  R5G5B5A1                      :  3.324 ms/call,      601.67 MB/s
  R4G4B4A4                      :  3.324 ms/call,      601.67 MB/s
  R8G8B8A8                      :  3.646 ms/call,     1097.08 MB/s
  R32                           :  3.646 ms/call,     1097.15 MB/s
  R32G32B32                     :  3.395 ms/call,     3535.04 MB/s
  R32G32B32A32                  :  3.403 ms/call,     4701.50 MB/s
  R16                           :  3.318 ms/call,      602.84 MB/s
  R16G16B16                     :  3.661 ms/call,     1638.94 MB/s
  R16G16B16A16                  :  3.355 ms/call,     2384.82 MB/s

After change:
Image Size: 1024x1024 (1048576 pixels)
--------------------------------------------------------------------------------
  GRAYSCALE                     :  0.038 ms/call,    26481.94 MB/s
  GRAY_ALPHA                    :  0.074 ms/call,    26984.73 MB/s
  R5G6B5                        :  0.075 ms/call,    26812.89 MB/s
  R8G8B8                        :  0.126 ms/call,    23838.65 MB/s
  R5G5B5A1                      :  0.074 ms/call,    26926.22 MB/s
  R4G4B4A4                      :  0.074 ms/call,    26945.42 MB/s
  R8G8B8A8                      :  0.172 ms/call,    23277.30 MB/s
  R32                           :  0.172 ms/call,    23304.70 MB/s
  R32G32B32                     :  0.492 ms/call,    24401.23 MB/s
  R32G32B32A32                  :  0.715 ms/call,    22390.98 MB/s
  R16                           :  0.074 ms/call,    27102.76 MB/s
  R16G16B16                     :  0.255 ms/call,    23522.38 MB/s
  R16G16B16A16                  :  0.338 ms/call,    23692.66 MB/s

[raysan5]

@ChocolateChipKookie Nice improvement! Thanks!