Adds macros to the MIPS assembler for detecting overflow.

Adds macros to the MIPS assembler for detecting overflow.
Committed: https://code.google.com/p/dart/source/detail?r=21085

[regis, 2013-04-05 23:23:45]

https://codereview.chromium.org/13722022/diff/4001/runtime/vm/assembler_mips.cc
File runtime/vm/assembler_mips.cc (right):

https://codereview.chromium.org/13722022/diff/4001/runtime/vm/assembler_mips....
runtime/vm/assembler_mips.cc:91: ASSERT(rd != T9);
This macro should take a scratch register argument that would be used here
instead of implicitly using T9: ASSERT(scratch != kNoRegister)
The caller is free to pass kNoRegister if no scratch register will be required.

https://codereview.chromium.org/13722022/diff/4001/runtime/vm/assembler_mips....
runtime/vm/assembler_mips.cc:119: const Immediate& imm, Register ro) {
ditto, scratch argument instead of T8.

https://codereview.chromium.org/13722022/diff/4001/runtime/vm/assembler_mips_...
File runtime/vm/assembler_mips_test.cc (right):

https://codereview.chromium.org/13722022/diff/4001/runtime/vm/assembler_mips_...
runtime/vm/assembler_mips_test.cc:1046: Label error, success;
"success" is actually "done".

[zra, 2013-04-06 00:04:19]

https://codereview.chromium.org/13722022/diff/4001/runtime/vm/assembler_mips.cc
File runtime/vm/assembler_mips.cc (right):

https://codereview.chromium.org/13722022/diff/4001/runtime/vm/assembler_mips....
runtime/vm/assembler_mips.cc:91: ASSERT(rd != T9);
On 2013/04/05 23:23:45, regis wrote:
> This macro should take a scratch register argument that would be used here
> instead of implicitly using T9: ASSERT(scratch != kNoRegister)
> The caller is free to pass kNoRegister if no scratch register will be
required.

I thought it might be clearer to require a scratch register and to use that
instead of TMP, and then to use TMP instead of T9 in the case that rd = rs = rt.
I think it might be too hard to remember to provide a scratch register only in
this one weird case.

https://codereview.chromium.org/13722022/diff/4001/runtime/vm/assembler_mips....
runtime/vm/assembler_mips.cc:119: const Immediate& imm, Register ro) {
On 2013/04/05 23:23:45, regis wrote:
> ditto, scratch argument instead of T8.

I decided to remove this function. Otherwise it would need multiple scratch
registers, and it's probably more readable at call sites just to load the
immediate by hand.

https://codereview.chromium.org/13722022/diff/4001/runtime/vm/assembler_mips_...
File runtime/vm/assembler_mips_test.cc (right):

https://codereview.chromium.org/13722022/diff/4001/runtime/vm/assembler_mips_...
runtime/vm/assembler_mips_test.cc:1046: Label error, success;
On 2013/04/05 23:23:45, regis wrote:
> "success" is actually "done".

Done.

[regis, 2013-04-06 00:33:24]

LGTM with one comment and one idea for a later cl.

Regarding the function you removed:
I expect that adding an immediate will be very frequent (e.g. in for-loops).
Since we have so many registers on MIPS, we should consider sacrificing 2 of
them as implicit scratch registers (TMP0 and TMP1, or T0 and T1, ...) instead of
just one.
This would remove the complication of passing a scratch register to possibly
many macros.

https://codereview.chromium.org/13722022/diff/8001/runtime/vm/assembler_mips.h
File runtime/vm/assembler_mips.h (right):

https://codereview.chromium.org/13722022/diff/8001/runtime/vm/assembler_mips....
runtime/vm/assembler_mips.h:511: // Clobbers TMP.
By definition, TMP can be clobbered by any asm macro.
You should add a more detailed comment about register usage, otherwise the
caller will waste a register.
You should explain when a scratch register is required and when not.

[zra, 2013-04-08 16:45:25]

I added to the comment, and also added a test for the case that the scratch
register is needed.

In my next CL, I'll change to using T8 and T9 as macro temps, and I'll also add
some handy macros for comparison and branches.

https://codereview.chromium.org/13722022/diff/8001/runtime/vm/assembler_mips.h
File runtime/vm/assembler_mips.h (right):

https://codereview.chromium.org/13722022/diff/8001/runtime/vm/assembler_mips....
runtime/vm/assembler_mips.h:511: // Clobbers TMP.
On 2013/04/06 00:33:24, regis wrote:
> By definition, TMP can be clobbered by any asm macro.
> You should add a more detailed comment about register usage, otherwise the
> caller will waste a register.
> You should explain when a scratch register is required and when not.

Done.

[zra, 2013-04-08 16:51:17]

Committed patchset #4 manually as r21085 (presubmit successful).