ebiten: improve DrawImage memory usage

[hajimehoshi]

Operating System

• Windows
• macOS
• Linux
• FreeBSD
• OpenBSD
• Android
• iOS
• Nintendo Switch
• PlayStation 5
• Xbox
• Web Browsers

What feature would you like to be added?

Now BenchmarkDrawImage takes too much memory like 4GB:

go test -bench=DrawImage -run=^$ -benchmem -memprofile mem.prof

The profiler says

Showing nodes accounting for 6360.44MB, 100% of 6361.44MB total
Dropped 25 nodes (cum <= 31.81MB)
Showing top 10 nodes out of 16
      flat  flat%   sum%        cum   cum%
 4705.71MB 73.97% 73.97%  6360.44MB   100%  github.com/hajimehoshi/ebiten/v2/internal/graphicscommand.(*commandQueue).EnqueueDrawTrianglesCommand
 1023.99MB 16.10% 90.07%  1023.99MB 16.10%  github.com/hajimehoshi/ebiten/v2/internal/graphicscommand.(*uint32sBuffer).alloc (inline)
  630.75MB  9.92%   100%   630.75MB  9.92%  github.com/hajimehoshi/ebiten/v2/internal/graphicscommand.(*commandQueue).appendIndices (inline)
         0     0%   100%  6360.44MB   100%  github.com/hajimehoshi/ebiten/v2.(*Image).DrawImage
         0     0%   100%  6360.44MB   100%  github.com/hajimehoshi/ebiten/v2/internal/atlas.(*Image).DrawTriangles
         0     0%   100%  6360.44MB   100%  github.com/hajimehoshi/ebiten/v2/internal/atlas.(*Image).drawTriangles
         0     0%   100%  6360.44MB   100%  github.com/hajimehoshi/ebiten/v2/internal/buffered.(*Image).DrawTriangles
         0     0%   100%  6360.44MB   100%  github.com/hajimehoshi/ebiten/v2/internal/graphicscommand.(*Image).DrawTriangles
         0     0%   100%  1023.99MB 16.10%  github.com/hajimehoshi/ebiten/v2/internal/graphicscommand.(*commandQueue).prependPreservedUniforms
         0     0%   100%  6360.44MB   100%  github.com/hajimehoshi/ebiten/v2/internal/graphicscommand.(*commandQueueManager).enqueueDrawTrianglesCommand

This can be an issue when we want to use BenchmarkDrawImage with -count=10.

Why is this needed?

No response

[hajimehoshi]

I thought this would work simply:

	b.Cleanup(func() {
		img0.At(0, 0)
		img0.Deallocate()
		runtime.GC()
	})

However, -count=10 still accumulates memory usages and finally the test used 20GB memory... Hmm