Hi,
in LLVM CVS the afore-mentioned function has 'const TargetRegisterClass*'
parameter, that is not documented.
Can somebody explain what does it mean?
Thanks,
Volodya
Basically, it gives the target more information about the spill. In particular, it specifies the register class to use for the copy. The target can choose to ignore this if it wants.
-Chris
Chris Lattner wrote:
in LLVM CVS the afore-mentioned function has 'const TargetRegisterClass*'
parameter, that is not documented.Can somebody explain what does it mean?
Basically, it gives the target more information about the spill. In
particular, it specifies the register class to use for the copy.
I'm still missing something. The 'storeRegToStackSlot' saves
'SrcReg' (already specified) to stack (which is not a register). So, what
does this register class applies to?
Examining RegAllocLocal.cpp suggests that the argument actually specifies
the register class of the spilled virtual register. Can you give some
examples how that information can be helpful?
I'd like to send doc patch, but can't do that without understanding the
semantics 
- Volodya
Chris Lattner wrote:
in LLVM CVS the afore-mentioned function has 'const TargetRegisterClass*'
parameter, that is not documented.
Can somebody explain what does it mean?Basically, it gives the target more information about the spill. In
particular, it specifies the register class to use for the copy.I'm still missing something. The 'storeRegToStackSlot' saves
'SrcReg' (already specified) to stack (which is not a register). So, what
does this register class applies to?
It applies to the register class. Memory (the stack slot) doesn't have a register class.
Examining RegAllocLocal.cpp suggests that the argument actually specifies
the register class of the spilled virtual register.
Right.
Can you give some examples how that information can be helpful?
Take a look at how some other targets use implement this. PPC, for example, uses this:
if (RC == PPC::GPRCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::STW, 3).addReg(SrcReg),FrameIdx);
} else if (RC == PPC::G8RCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::STD, 3).addReg(SrcReg),FrameIdx);
} else if (RC == PPC::F8RCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::STFD, 3).addReg(SrcReg),FrameIdx);
} else if (RC == PPC::F4RCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::STFS, 3).addReg(SrcReg),FrameIdx);
} else if (RC == PPC::VRRCRegisterClass) {
...
If you didn't have the register class (in the "bad old days") the target had to do two things:
1. For virtual registers, it had to look up the regclass in the SSARegMap.
This is an extra step, but was efficient and not too problematic.
2. For physregs, you either had to switch on every physreg (ugly!) or scan
through all the register classes to find out which regclass a physreg
was in (slow!). Additionally, physregs can be in multiple register
classes, which could potentially be a problem.
Again, as mentioned before, this is just extra information provided to the target. The target can figure out, however it wants, how to handle the register. The regclass is just a convenient way to do this. 
I'd like to send doc patch, but can't do that without understanding the
semantics
Thanks!!
-Chris