Introduce DontEmitDebugCodeScope to fix the x64 nosnapshot build.

Introduce DontEmitDebugCodeScope to fix the x64 nosnapshot build.

R=yangguo@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=20784

[haitao.feng, 2014-04-10 09:24:41]

Sorry my commit (https://codereview.chromium.org/42973002) broke the "make
x64.release.check -j8 snapshot=off" because there are two kSmiRegister
uninitialization checks which takes 96 bytes, plus the other generated code, it
exceeds the code size limit (127 bytes from JumpPatchSite).

There are still 3 fails after this fix, I will investigate them and upload
another CL later.

[haitao.feng, 2014-04-10 12:47:04]

Add Yang to review the EmitProfilingCounterReset change. When emit_debug_code is
on, the kJnsOffset is 0x4e instead of 0x1d.

[Yang, 2014-04-11 07:19:22]

On 2014/04/10 12:47:04, haitao.feng wrote:
> Add Yang to review the EmitProfilingCounterReset change. When emit_debug_code
is
> on, the kJnsOffset is 0x4e instead of 0x1d.

How about introducing an assertion in EmitBackEdgeBookkeeping to make sure the
jump distance is as expected?

[Yang, 2014-04-11 09:34:26]

On 2014/04/11 07:19:22, Yang wrote:
> On 2014/04/10 12:47:04, haitao.feng wrote:
> > Add Yang to review the EmitProfilingCounterReset change. When
emit_debug_code
> is
> > on, the kJnsOffset is 0x4e instead of 0x1d.
> 
> How about introducing an assertion in EmitBackEdgeBookkeeping to make sure the
> jump distance is as expected?

Also, just remembered, we have something called PredictableCodeSizeScope, which
is what you are looking for!

[haitao.feng, 2014-04-11 12:07:24]

On 2014/04/11 09:34:26, Yang wrote:
> On 2014/04/11 07:19:22, Yang wrote:
> > On 2014/04/10 12:47:04, haitao.feng wrote:
> > > Add Yang to review the EmitProfilingCounterReset change. When
> emit_debug_code
> > is
> > > on, the kJnsOffset is 0x4e instead of 0x1d.
> > 
> > How about introducing an assertion in EmitBackEdgeBookkeeping to make sure
the
> > jump distance is as expected?
> 
> Also, just remembered, we have something called PredictableCodeSizeScope,
which
> is what you are looking for!

Done. Thank Yang for the comments and recommendation.

[Yang, 2014-04-11 12:46:05]

On 2014/04/11 12:07:24, haitao.feng wrote:
> On 2014/04/11 09:34:26, Yang wrote:
> > On 2014/04/11 07:19:22, Yang wrote:
> > > On 2014/04/10 12:47:04, haitao.feng wrote:
> > > > Add Yang to review the EmitProfilingCounterReset change. When
> > emit_debug_code
> > > is
> > > > on, the kJnsOffset is 0x4e instead of 0x1d.
> > > 
> > > How about introducing an assertion in EmitBackEdgeBookkeeping to make sure
> the
> > > jump distance is as expected?
> > 
> > Also, just remembered, we have something called PredictableCodeSizeScope,
> which
> > is what you are looking for!
> 
> Done. Thank Yang for the comments and recommendation.

PredictableCodeSizeScope also sets a flag on the assembler object. I would
rather you use that scope instead of introducing another one.

[Yang, 2014-04-11 13:18:59]

On 2014/04/11 12:46:05, Yang wrote:
> On 2014/04/11 12:07:24, haitao.feng wrote:
> > On 2014/04/11 09:34:26, Yang wrote:
> > > On 2014/04/11 07:19:22, Yang wrote:
> > > > On 2014/04/10 12:47:04, haitao.feng wrote:
> > > > > Add Yang to review the EmitProfilingCounterReset change. When
> > > emit_debug_code
> > > > is
> > > > > on, the kJnsOffset is 0x4e instead of 0x1d.
> > > > 
> > > > How about introducing an assertion in EmitBackEdgeBookkeeping to make
sure
> > the
> > > > jump distance is as expected?
> > > 
> > > Also, just remembered, we have something called PredictableCodeSizeScope,
> > which
> > > is what you are looking for!
> > 
> > Done. Thank Yang for the comments and recommendation.
> 
> PredictableCodeSizeScope also sets a flag on the assembler object. I would
> rather you use that scope instead of introducing another one.

(you could use it with an unspecified code size as argument, like
kUnknownCodeSize = -1, if you don't have it).

[haitao.feng, 2014-04-11 14:22:13]

Yang, thanks for the comments. I introduced a DontEmitDebugCodeScope to wrapper
the emit_debug_code_ variable. It is more general than
NoSmiConstantRegisterUninitializationCheckScope and semantically clearer than
using PredictableCodeSizeScope.

[Yang, 2014-04-15 14:41:37]

On 2014/04/11 14:22:13, haitao.feng wrote:
> Yang, thanks for the comments. I introduced a DontEmitDebugCodeScope to
wrapper
> the emit_debug_code_ variable. It is more general than
> NoSmiConstantRegisterUninitializationCheckScope and semantically clearer than
> using PredictableCodeSizeScope.

LGTM. But why do we need this in VisitCountOperation?

[mvstanton, 2014-04-15 15:17:18]

Hi Haitao,
I just ran into issues myself with x64 nosnap, and am happy to see your fix. You
might also run into this issue, which I'm checking in the fix for today:
https://codereview.chromium.org/237673007/

All the best, thanks!
--Michael

[haitao.feng, 2014-04-16 02:02:21]

Hi Michael,

Thanks for your fix. This makes the DontEmitDebugCodeScope unnecessary in
VisitCountOperation.

Thanks
-Haitao

[haitao.feng, 2014-04-16 02:06:35]

Committed patchset #7 manually as r20784 (presubmit successful).