Performance benchmarks about csound-cxx-opcodes

[gogins]

The results are in. The clang opcodes are not quite as fast as the Faust-generated plugin opcode, but it is very close.

[gogins]

faustgen	faust2csound	faust2csound g++ native	faust2csound clang++ native	Clang opcodes
4.4607	5.0320	4.6015	3.8385	3.9381
4.4475	5.0350	4.6022	3.8239	3.9440
4.2371	5.0270	4.6325	3.8237	3.9240
4.4839	4.9800		3.8610	3.9309
4.4357	5.0240		3.8010	3.9460
4.3795	5.0490		3.8296	3.9406
4.4235	4.9984		3.8701	3.9541
4.3682	4.9811		3.8213	3.9503
4.4395	5.0729		3.8716	3.9230
4.3618	5.0629		3.8693	3.9252
4.3783	5.0464		3.8726	3.9323
4.4142	5.0570		3.8365	3.9901
4.3713	5.0134		3.8666	4.0642
4.3180	5.0053		3.8410	4.0044
4.4177	5.0275		3.8287	3.9744
4.3958	5.0275	4.6121	3.8437	3.9561
% of faustgen	114.37%	95.31%	76.45%	90.00%
Student's t test	3.81E-15			2.82E-12
				102.92%

[gogins]

About 3% slower than the fastest implementation, which is faust2csound compiled with clang++ for -march=native.

This is extremely good for a runtime compiler, needless to say.

And this means there is no reason not to use the Clang opcodes unless the ORC compile time gets too big. That would definitely happen at some point. I don't think that would be a problem in the studio, but it might be a problem in a live venue.

And, it might well be that I can speed up the ORC compiler.

[gogins]

About 3% slower than the fastest implementation, which is faust2csound compiled with clang++ for -march=native.

This is extremely good for a runtime compiler, needless to say.

And this means there is no reason not to use the Clang opcodes unless the ORC compile time gets too big. That would definitely happen at some point. I don't think that would be a problem in the studio, but it might be a problem in a live venue.

And, it might well be that I can speed up the ORC compiler.