Update XRay to support x86-64 (RDTSC), use dladdr() as an optional way to

Update XRay to support x86-64 (RDTSC), use dladdr() as an optional way to
import symbols for glibc builds.  More options for report generation.
BUG=none
TEST=experimental
R=bradnelson@google.com

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=209051

[bradn, 2013-06-27 23:01:24]

Bunch of nits.
LGTM

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
File native_client_sdk/src/libraries/xray/library.dsc (right):

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
native_client_sdk/src/libraries/xray/library.dsc:17: '-DXRAY -DXRAY_ANNOTATE
-O2'
Why there's a tab here? (Or it that just code review being weird?)

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
File native_client_sdk/src/libraries/xray/symtable.c (right):

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
native_client_sdk/src/libraries/xray/symtable.c:149: r =
XRaySymbolTableAddByName(symtab, info.dli_sname, addr);
Why cache these in your hashtable?
Worried dladdr is slow?
Did you also want to demangle here?
You can do it at runtime with:
abi::__cxa_demangle from cxxabi.h
Of course that'd be your first C++ file here :-)

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
File native_client_sdk/src/libraries/xray/xray.c (right):

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
native_client_sdk/src/libraries/xray/xray.c:642: if (depth > 250) bad_depth++;
Might be nice to give this (250) a name and move up top (optional).

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
File native_client_sdk/src/libraries/xray/xray.h (right):

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
native_client_sdk/src/libraries/xray/xray.h:23: #define XRAY_INLINE
__attribute__((always_inline,no_instrument_function))
Semi-ambiguous given this an attribute, but normally a space after always_inline
would be the done thing.

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
native_client_sdk/src/libraries/xray/xray.h:64: int frame_count, const char*
mapfilename) {}
Maybe a comment mentioning NULL for no mapfile?

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
native_client_sdk/src/libraries/xray/xray.h:72: inline void XRayReport(FILE *f,
float percent_cutoff, int ticks_cutoff);
FILE*

[nfullagar1, 2013-06-27 23:41:55]

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
File native_client_sdk/src/libraries/xray/library.dsc (right):

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
native_client_sdk/src/libraries/xray/library.dsc:17: '-DXRAY -DXRAY_ANNOTATE
-O2'
On 2013/06/27 23:01:24, bradn wrote:
> Why there's a tab here? (Or it that just code review being weird?)
good catch, removed

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
File native_client_sdk/src/libraries/xray/symtable.c (right):

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
native_client_sdk/src/libraries/xray/symtable.c:149: r =
XRaySymbolTableAddByName(symtab, info.dli_sname, addr);
On 2013/06/27 23:01:24, bradn wrote:
> Why cache these in your hashtable?
> Worried dladdr is slow?
> Did you also want to demangle here?
> You can do it at runtime with:
> abi::__cxa_demangle from cxxabi.h
> Of course that'd be your first C++ file here :-)
dladdr (as I understand it) can take any address and return nearest symbol.  For
xray, I only need address of the function -> symbol name, and the xray symbol
table is fast for this, and can also be populated via other means (map files,
etc)

For demangling, agree, talked with Roland yesterday about this, and yes, c++
would get pulled in.  Still considering it, just not for this CL :)

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
File native_client_sdk/src/libraries/xray/xray.c (right):

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
native_client_sdk/src/libraries/xray/xray.c:642: if (depth > 250) bad_depth++;
On 2013/06/27 23:01:24, bradn wrote:
> Might be nice to give this (250) a name and move up top (optional).

This code isn't needed, I'll remove it.

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
File native_client_sdk/src/libraries/xray/xray.h (right):

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
native_client_sdk/src/libraries/xray/xray.h:23: #define XRAY_INLINE
__attribute__((always_inline,no_instrument_function))
On 2013/06/27 23:01:24, bradn wrote:
> Semi-ambiguous given this an attribute, but normally a space after
always_inline
> would be the done thing.

space added

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
native_client_sdk/src/libraries/xray/xray.h:64: int frame_count, const char*
mapfilename) {}
On 2013/06/27 23:01:24, bradn wrote:
> Maybe a comment mentioning NULL for no mapfile?

Done.

https://codereview.chromium.org/18120003/diff/5001/native_client_sdk/src/libr...
native_client_sdk/src/libraries/xray/xray.h:72: inline void XRayReport(FILE *f,
float percent_cutoff, int ticks_cutoff);
On 2013/06/27 23:01:24, bradn wrote:
> FILE*

Done.

[nfullagar1, 2013-06-28 01:12:50]

Committed patchset #3 manually as r209051 (presubmit successful).