mips eabi64 support

[ethteck]

Very much a WIP, but this PR brings us from

void func_001099D8(void) {
    sceGsSetDefLoadImage(&D_002B8700, 0xE00, 8, 1);
    FlushCache(0);
    sceGsSyncPath(0, 0);
    sceGsExecLoadImage(&D_002B8700, &func_F20000);
    sceGsSetDefLoadImage(&D_002B8760, 0x1400, 8, 1);
    FlushCache(0);
    sceGsSyncPath(0, 0);
    sceGsExecLoadImage(&D_002B8760, &func_F20000 + 0x48000);
}

to the more accurate

void func_001099D8(s64 arg0, s64 arg2, s64 arg4) {
    sceGsSetDefLoadImage(&D_002B8700, 0xE00, 8, 1, 0, 0, 0x200, 0xC0);
    FlushCache(0);
    sceGsSyncPath(0, 0);
    sceGsExecLoadImage(&D_002B8700, &func_F20000);
    sceGsSetDefLoadImage(&D_002B8760, 0x1400, 8, 1, 0, 0, 0x200, 0xC0);
    FlushCache(0);
    sceGsSyncPath(0, 0);
    sceGsExecLoadImage(&D_002B8760, &func_F20000 + 0x48000);
}

That being said, the signature of this function is broken now, and some actual code changes will probably be needed here

[simonlindholm]

Sorry for extreme slowness in getting to this.

[simonlindholm]

this is not really o32 abi, is it? is there a reference for this list?

[ethteck]

you're right, it's n32, whoops. I renamed it accordingly

https://gist.github.com/EllipticEllipsis/27eef11205c7a59d8ea85632bc49224d#general-purpose-registers

[simonlindholm]

Is the convention that you set the assembler to assemble using n32 register names when targeting eabi? Or is this maybe not something that you specify at all, and all the register name mappings are unique and as is able to just take the union of them? The gist seems to suggest though that $12 is $t0 rather than $t4 for n32, which doesn't quite match that.

[ethteck]

We just pass the following flags to as for Kingdom Hearts: -no-pad-sections -EL -march=5900 -mabi=eabi. I am not totally sure how to answer your question unfortunately..

[simonlindholm]

It's a vague question, I was basically just hoping to get some more information about how these aliases work. I gather from experimenting with the flags you mention that as doesn't actually accept these register names as inputs, nor does objdump output them (-Mreg-names=eabi is not a thing). And the same seems to be true for integer registers.

So then, I take it all the names are just made up by us? And we made a decision to stay backwards compatible with o32 names, given the choice of naming $12 $t4 instead of $t0? Are eabi projects using some common prelude.inc that define these names for use with as, or is everyone sticking to numbered registers? Or are there projects assembling code using o32 register names? If so, should we support that, and make $t0 an alias for $a4?

Using the n32 naming of fp regs seems like a poor choice: it names $f21 $ft8 despite it being a saved register. I guess we have a few choices here:

• don't support abi names of fp regs
• use suboptimal names from o32 or n32 (but it's a bit sad to enshrine this in m2c, it makes it harder to change)
• come up with our set of names (and patching https://github.com/decompals/binutils-mips-ps2-decompals?)

[ethteck]

For now, I'm fine with option 1.. does it mean just getting rid of this dict?

[simonlindholm]

Yes, should be. Do you have any thoughts on the questions about integer registers?

[simonlindholm]

I'll merge this and then push some follow-ups on top.

[simonlindholm]

5bd7558