Apply hoist optimizations to Geometry.c's transforms

[akx]

Like the other recent optimization PRs: hoist im*->xsize and im*->ysize into locals so autovectorization can do its thing.

[codspeed-hq]

Merging this PR will improve performance by 19.6%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 21 improved benchmarks
✅ 582 untouched benchmarks
⏩ 335 skipped benchmarks¹

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	test_rotate_right[1237x811-L-ROTATE_180]	8.5 ms	6.2 ms	+36.32%
⚡	test_rotate_right[1237x811-L-FLIP_LEFT_RIGHT]	8.5 ms	6.2 ms	+36.3%
⚡	test_scale[1237x811-LA-2.14-NEAREST]	101.9 ms	84.6 ms	+20.45%
⚡	test_scale[1237x811-I-2.14-NEAREST]	101.8 ms	84.6 ms	+20.37%
⚡	test_scale[1237x811-F-2.14-NEAREST]	101.9 ms	84.7 ms	+20.33%
⚡	test_scale[1237x811-RGBA-2.14-NEAREST]	101.9 ms	84.7 ms	+20.29%
⚡	test_scale[1237x811-RGB-2.14-NEAREST]	101.9 ms	84.7 ms	+20.28%
⚡	test_rotate_right[1237x811-LA-ROTATE_180]	14.1 ms	11.9 ms	+18.87%
⚡	test_rotate_right[1237x811-RGBA-FLIP_LEFT_RIGHT]	14.1 ms	11.9 ms	+18.87%
⚡	test_rotate_right[1237x811-RGBA-ROTATE_180]	14.1 ms	11.9 ms	+18.86%
⚡	test_rotate_right[1237x811-LA-FLIP_LEFT_RIGHT]	14.1 ms	11.9 ms	+18.86%
⚡	test_rotate_right[1237x811-RGB-FLIP_LEFT_RIGHT]	14.1 ms	11.9 ms	+18.81%
⚡	test_rotate_right[1237x811-RGB-ROTATE_180]	14.1 ms	11.9 ms	+18.8%
⚡	test_scale[1237x811-F-0.8-NEAREST]	16.5 ms	14.1 ms	+17.07%
⚡	test_scale[1237x811-RGB-0.8-NEAREST]	16.5 ms	14.1 ms	+17.07%
⚡	test_scale[1237x811-LA-0.8-NEAREST]	16.5 ms	14.1 ms	+17.07%
⚡	test_scale[1237x811-RGBA-0.8-NEAREST]	16.5 ms	14.1 ms	+17.07%
⚡	test_scale[1237x811-I-0.8-NEAREST]	16.5 ms	14.1 ms	+17.06%
⚡	test_transform[1237x811-LA-BICUBIC-PERSPECTIVE]	385.7 ms	337.7 ms	+14.23%
⚡	test_transform[1237x811-LA-BICUBIC-AFFINE]	347.5 ms	304.3 ms	+14.18%
...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

Tip

Curious why performance improved? Comment @codspeedbot explain why performance improved on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing akx:faster-geometry-1 (4ae228a) with main (0a61c53)}

Footnotes

1. 335 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[radarhere]

Could you explain how this is an improvement? I mean, it wasn't using the values within a loop. There was no repetition at all.

[akx]

You mean out_xsize/out_ysize? Yes, they're not used in the loops, but I think it reads more cleanly for the human reader if it's

if (in_xsize != out_xsize || in_ysize != out_ysize) {

and not

if (in_xsize != imOut->xsize || in_ysize != imOut->ysize) {

-- with the latter, at least my code eye starts wondering why there's two different idioms for the sizes.

[radarhere]

Yes, that is what I was referring to. This appears to be another different of opinion - your instinct is to wonder why the code is written differently, my instinct is to wonder why the code is more complex than it could be.

We'll see how #9962 goes. It is merged, then this discussion becomes irrelevant.

[akx]

My opinion about this isn't strong at all, but it's a consistency thing a tiny bit like #9975 (comment) - it's a little easier to read if all of the similar functions have the same prelude.