Rewrite msvc asm to not use non volatile registers

[ZXShady]

I tested this change by running the as_test suite, and all tests passed.

While learning x64 assembly for a compiler project (with the goal of contributing more to AngelScript, as I really enjoy the language) and studying calling conventions, I spent time reading through AngelScript's implementation. It was a great learning resource, and while doing so I noticed this tiny small optimization.

I tried rewriting the code to not use any non volatile registers saving lots of pops and pushes. I did some small tests that called empty functions with 1 int and 4 int and 8 int and 1 float, 4 float and 8 float args and it seems to be faster by about 10%-15% but this is not representative of the real world as this is just calling an empty function it got slower at calling 0 arg functions since i removed the branch though but I don't think they are that common to special case.

Resources used:
https://learn.microsoft.com/en-us/cpp/build/x64-calling-convention?view=msvc-170#callercallee-saved-registers

[anjo76]

Thanks for the contribution.

Did you actually notice any performance improvement with this optimization? 😄 While your change is accurate I doubt it really affected anything, as this is the type of optimization that the compiler would easily do by itself.

[ZXShady]

Thanks for the contribution.

Did you actually notice any performance improvement with this optimization? 😄 While your change is accurate I doubt it really affected anything, as this is the type of optimization that the compiler would easily do by itself.

It isn't strictly equal so the assembler optimizer cannot change it (also I don't think the assembler has any optimizations) , as it doesn't check The O flag (overflow) while jle does but as no overflow happens (it is a positive signed int) it is 1 less instruction for jle vs 2 seperate jumps.

Performance wise I did a small test but it was noise and I didn't notice anything but I had background apps running so. But for readability I think it is worth it.

[ZXShady]

I also tried rewriting the file to use no volatile registers, but I don't know whether it is readable using less instructions and stack in the process.

14f7521#diff-6bb1a1e324035a147c6f845c43b35e5a302837b28321eea42cf6124ed16fdb03

[anjo76]

As long as the code comments explain at a high level what the assembly instructions are doing it should be OK.

[ZXShady]

Have you took a look at the diff?

Should I make a new pr for it or should I update this pr?