Hi Gabe,

Thanks for taking the time to making everything work with 4.6.1. I have a couple of questions about what you're changing in the ARM ISA. It wasn't immediately clear to me what you're doing or why. Have you tried 4.6 support for link time optimization or whole program optimization? It would be nice to get a speedup with it.

src/arch/alpha/isa/decoder.isa (Diff revision 1)

Does the parser still pick this up as a source if it's not in the expression?

1. Gabe Black Sept. 7, 2011, 7 p.m. (Sept. 7, 2011, 7 p.m.)

   Yes. It's not that smart. Also, this way Rb is always used and the preprocessor won't chop out the use.

src/arch/arm/isa/formats/fp.isa (Diff revision 1)

Could you describe what you're doing here? I can't quite grok it from inspection. 

1. Gabe Black Sept. 7, 2011, 7:05 p.m. (Sept. 7, 2011, 7:05 p.m.)

   I don't remember for sure (I did this patch over a long time), but I don't think VpminQ exists as an instruction. I think also the US version of this function changes the versions generated, but I don't remember how. The function names aren't great.

2. Ali Saidi Sept. 10, 2011, 1:09 a.m. (Sept. 10, 2011, 1:09 a.m.)

   Ok... i'm going to have to do some digging/testing because I can't convince myself via inspection

src/arch/arm/isa/formats/fp.isa (Diff revision 1)

Same here. 

1. Gabe Black Sept. 7, 2011, 7:05 p.m. (Sept. 7, 2011, 7:05 p.m.)

   Ditto.

src/arch/arm/isa/insts/m5ops.isa (Diff revision 1)

excellent...

src/arch/arm/isa/insts/m5ops.isa (Diff revision 1)

You don't need to do this, but it would be nice if we added a PseudoInst::initParam() and called it just as every other one.

1. Gabe Black Sept. 7, 2011, 7:07 p.m. (Sept. 7, 2011, 7:07 p.m.)

   That's a good idea, but it should be in a separate change.

src/arch/arm/isa/insts/macromem.isa (Diff revision 1)

short description?

1. Gabe Black Sept. 7, 2011, 7:11 p.m. (Sept. 7, 2011, 7:11 p.m.)

   I don't remember for sure, but I think just so the size is available while iterating in the loops so only valid instructions are generated. Otherwise you get versions that 4.6.1 gets upset about because they overflow arrays. There isn't really much to say. It's a bit ugly, but there isn't really anything that interesting going on.

src/arch/arm/isa/insts/neon.isa (Diff revision 1)

same please

1. Gabe Black Sept. 7, 2011, 7:13 p.m. (Sept. 7, 2011, 7:13 p.m.)

   It's doing whatever it used to be doing. This is just simplifying the code with some basic refactoring.

src/arch/mips/isa/decoder.isa (Diff revision 1)

What is  the point of this?

1. Gabe Black Sept. 7, 2011, 7:18 p.m. (Sept. 7, 2011, 7:18 p.m.)

   Oh, I thought I took this back out. This is to hide the operand name from the parser since it didn't use to ignore names in strings. It's not needed any more, but apparently I forgot to refresh my patch. I guess I won't be able to push this until I get home and get at the current version of the patch.

2. Gabe Black Sept. 7, 2011, 7:20 p.m. (Sept. 7, 2011, 7:20 p.m.)

   Hmm, nope, it looks like I just missed that one.