Add PC descriptor for ret instruction

Add PC descriptor for ret instruction

Adding a new kind of PcDescriptor to mark the location of
function returns in generated code. This will be needed to
put a single-step breakpoint just before the function return.

Also adding a NOP instruction after the ret, so that the
function return code pattern adds up to 5 bytes, which is needed
to patch in a breakpoint call.
Committed: https://code.google.com/p/dart/source/detail?r=4159

[siva, 2012-02-11 00:49:25]

LGTM

http://codereview.chromium.org/9385022/diff/1/runtime/vm/code_generator_ia32.h
File runtime/vm/code_generator_ia32.h (right):

http://codereview.chromium.org/9385022/diff/1/runtime/vm/code_generator_ia32....
runtime/vm/code_generator_ia32.h:128: void GenerateReturnEpilog(ReturnNode*
node);
Do we have to make a similar change for the x64 version too. I am not sure how
we track changes made in the ia32 version only that need propagation to the x64
version.

[hausner, 2012-02-11 01:01:27]

Thank you.

http://codereview.chromium.org/9385022/diff/1/runtime/vm/code_generator_ia32.h
File runtime/vm/code_generator_ia32.h (right):

http://codereview.chromium.org/9385022/diff/1/runtime/vm/code_generator_ia32....
runtime/vm/code_generator_ia32.h:128: void GenerateReturnEpilog(ReturnNode*
node);
On 2012/02/11 00:49:25, asiva wrote:
> Do we have to make a similar change for the x64 version too. I am not sure how
> we track changes made in the ia32 version only that need propagation to the
x64
> version.

I don't understand patching in 64bit code yet. I made the interface to
GenerateReturnEpilog identical in the 32 and 64 bit version and added a todo to
add code later.

[srdjan, 2012-02-11 11:58:50]

DBC

http://codereview.chromium.org/9385022/diff/9/runtime/vm/code_generator_ia32.cc
File runtime/vm/code_generator_ia32.cc (right):

http://codereview.chromium.org/9385022/diff/9/runtime/vm/code_generator_ia32....
runtime/vm/code_generator_ia32.cc:792: // in breakpoints during debugging.
Since we do not execute this code, would it make sense to add int3 instead?

http://codereview.chromium.org/9385022/diff/9/runtime/vm/code_generator_ia32....
runtime/vm/code_generator_ia32.cc:795: AstNode::kNoId,
Why not use return node id?

[hausner, 2012-02-13 17:27:10]

Thanks Srdjan.

http://codereview.chromium.org/9385022/diff/9/runtime/vm/code_generator_ia32.cc
File runtime/vm/code_generator_ia32.cc (right):

http://codereview.chromium.org/9385022/diff/9/runtime/vm/code_generator_ia32....
runtime/vm/code_generator_ia32.cc:792: // in breakpoints during debugging.
On 2012/02/11 11:58:50, srdjan wrote:
> Since we do not execute this code, would it make sense to add int3 instead?
If it isn't executed, it seems irrelevant whether it's NOP ir INT3. the nop
documents that it has no function other than take up space.

http://codereview.chromium.org/9385022/diff/9/runtime/vm/code_generator_ia32....
runtime/vm/code_generator_ia32.cc:795: AstNode::kNoId,
On 2012/02/11 11:58:50, srdjan wrote:
> Why not use return node id?
No particular reason. Will do in next change.