Where should I look then !
To do this:
On Dec 9, 2016, at 9:34 AM, Shubha Ramani via pypy-dev <pypy-dev@python.org> wrote:
Armin you articulated exactly what I want. Please tell me the right place to look then ?
Shubha
Sent from Shubha Ramani's iPhone 7
On Dec 9, 2016, at 8:29 AM, Armin Rigo <armin.rigo@gmail.com> wrote:
Hi,
On 9 December 2016 at 15:55, Shubha Ramani via pypy-dev <pypy-dev@python.org> wrote:
I am no longer confused about that.
I think the rest of your mail shows there is still confusion.
But so far, in order to be able to dump jitcodes, I have to instrument python code with "JitDriver", "merge_point", "meta_interp", etc...under the test directory -
Maybe you should start again by describing in detail what you want to do, from the start.
Let me take a wild guess, completely unrelated to the questions you're asking. Maybe your goal is to run a "pypy" binary, which JITs some parts of the user Python code. What you want then is to map the raw machine code emitted in memory, back to Python- level information. For example, from an IP, you want an answer like "this 'add' instruction comes from JIT-compiling through this Python function, precisely here [Python bytecode of a CALL], with the call inlined, and running this other Python function, precisely here [Python bytecode of a BINARY_ADD].
If that's anywhere close to what you're looking for, then you are looking at the wrong place.
A bientôt,
Armin.
Sent from Shubha Ramani's iPhone 7
Hi Shubha, Sorry, you are asking many questions and I think you are not carefully reading the answers. Similarly, you are insisting that you've read *all* the docs---which would be good, except that at the same time some of yours questions have clear answers in the docs. About your precise question: you should re-read some past answers, e.g. Richard's December 9th, and then look in the source code how the vmprof integration is done through RPython. You might also be interested in the pypy-dev thread from August about VTune, which contains the answers to similar questions asked by other Intel people at the time. In the source code, all relevant places can be found from the high-level pypy.module._vmprof, to the RPython-level rpython.rlib.rvmprof; and x86.assembler._call_header_vmprof() which generates in the assembler frames some structure with VMPROF_JITTED_TAG, which is read by logic in rpython.rlib.rvmprof. In general, grepping for "vmprof" or "rvmprof" throughout the source code of RPython is a good start. I will now assume that you have read the answer above and actually seriously dig into the various parts I mention, before asking more questions. Of course, you can keep asking anyway, but you need to accept that we may not have the time and energy to keep answering you. Asking a deluge of questions and then pressuring to get prompt answers, that is the perfect way to get ignored in the open source world---nobody *has* to answer you. Also, I have nothing against helping someone building VTune support inside PyPy, but if you read the August thread I mentioned above you'd understand that I see VTune in a negative light right now: it pretends to be more open-source-friendly that it is. As far as I am concerned, I am helping you build support specifically for an Intel commercial product, and I am doing that without getting paid. I am fine with doing that to some extend, but it is an unstable situation. A bientôt, Armin.
Armin I don't blame you for being upset with me. I'm sorry...I will lay low and dive deep into the pointers you've just nowmentioned in this email. I'm sorry, I didn't know about the August Vtune thread.Please accept my apologies. Shubha On Monday, December 12, 2016 11:11 AM, Armin Rigo <armin.rigo@gmail.com> wrote: Hi Shubha, Sorry, you are asking many questions and I think you are not carefully reading the answers. Similarly, you are insisting that you've read *all* the docs---which would be good, except that at the same time some of yours questions have clear answers in the docs. About your precise question: you should re-read some past answers, e.g. Richard's December 9th, and then look in the source code how the vmprof integration is done through RPython. You might also be interested in the pypy-dev thread from August about VTune, which contains the answers to similar questions asked by other Intel people at the time. In the source code, all relevant places can be found from the high-level pypy.module._vmprof, to the RPython-level rpython.rlib.rvmprof; and x86.assembler._call_header_vmprof() which generates in the assembler frames some structure with VMPROF_JITTED_TAG, which is read by logic in rpython.rlib.rvmprof. In general, grepping for "vmprof" or "rvmprof" throughout the source code of RPython is a good start. I will now assume that you have read the answer above and actually seriously dig into the various parts I mention, before asking more questions. Of course, you can keep asking anyway, but you need to accept that we may not have the time and energy to keep answering you. Asking a deluge of questions and then pressuring to get prompt answers, that is the perfect way to get ignored in the open source world---nobody *has* to answer you. Also, I have nothing against helping someone building VTune support inside PyPy, but if you read the August thread I mentioned above you'd understand that I see VTune in a negative light right now: it pretends to be more open-source-friendly that it is. As far as I am concerned, I am helping you build support specifically for an Intel commercial product, and I am doing that without getting paid. I am fine with doing that to some extend, but it is an unstable situation. A bientôt, Armin.
participants (2)
-
Armin Rigo
-
Shubha Ramani