Re: Problems with arrays of floats (lttng 2.13)

[Herman Bruyninckx via lttng-dev <lttng-dev@lists.lttng.org>]

[-- Attachment #1: Type: text/plain, Size: 7362 bytes --]

On Tue, 13 May 2025, Kienan Stewart wrote:

> Hi Herman,
>
> On 5/10/25 4:14 AM, Herman Bruyninckx wrote:
>> On Fri, 9 May 2025, Kienan Stewart wrote:
>> 
>>> Hi Herman,
>>> 
>>> On 5/9/25 6:03 AM, Herman Bruyninckx via lttng-dev wrote:
>>>> Dear all,
>>>> 
>>>> I want to trace an array of floats, but I always get bogus output at the
>>>> (Python) Babeltrace2 site...
>>> 
>>> could you give an example of the tracepoint you have setup in your code 
>>> that doesn't work, with the corresponding output?
>> 
>> Thank you for your swift reaction. Much appreaciated.
>> 
>> First of all, my apologies: I should have provided the required information 
>> the first time around already.
>> 
>> ---
>> This is the information about the installed versions of the relevant 
>> software:
>> 
>> $ uname -a
>> Linux SET-L-ME-P22061 6.11.0-24-generic #24-Ubuntu SMP PREEMPT_DYNAMIC Fri 
>> Mar 14 18:13:56 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
>> 
>> $ lttng -V
>> lttng (LTTng Trace Control) 2.13.13 - Nordicité
>> 
>> $ babeltrace2 -V
>> Babeltrace 2.0.6 "Amqui"
>> ---
>> 
>> 
>> ---
>> My application code generates a trace as follows:
>> 
>> Tracepoint "-tp.h":
>> LTTNG_UST_TRACEPOINT_EVENT(
>>      comm_activity,
>>      range_scan,
>>
>>      /* Input arguments */
>>      LTTNG_UST_TP_ARGS(
>>          int *, distances,
>>      ),
>>
>>      /* Output event fields */
>>      LTTNG_UST_TP_FIELDS(
>>          lttng_ust_field_array(int, distances,distances,5)
>> )
>> 
>> 
>> int testmeas[5] = {1, 2, 3, 4, 5};
>> // double testmeas[5] = {1., 2., 3., 4., 5.};
>> // (one array for test with integers, one with doubles):
>> 
>> lttng_ust_tracepoint(comm_activity,range_scan,5,testmeas, "range_scan");
>> 
>> ---
>> 
>>> Could you explain what you mean by (Python) babeltrace2 site?
>> This is the Python code I use to display the trace generated above:
>> (inspired by <https://babeltrace.org/docs/v2.0/python/bt2/examples.html>)
>> 
>> ---
>> file tt.py:
>> 
>> import bt2
>> import sys
>> 
>> # Get the trace path from the first command-line argument.
>> path = sys.argv[1]
>> 
>> # Create a trace collection message iterator with this path.
>> msg_it = bt2.TraceCollectionMessageIterator(path)
>> 
>> # Iterate the trace messages.
>> for msg in msg_it:
>>      # `bt2._EventMessageConst` is the Python type of an event message.
>>      if type(msg) is bt2._EventMessageConst:
>>          # An event message holds a trace event.
>>          event = msg.event
>>          # Only check `range_scan` events.
>>          if event.name != 'comm_activity:range_scan':
>>              continue
>>          print(event.payload_field)
>> ---
>> 
>> and these are the resulting traces (for int and double, respectively, and 
>> with the C and Python versions of Babeltrace):
>> 
>> $ babeltrace2  ~/git-repositories/communication_5c/build/test_session/
>> [12:19:11.381152129] (+?.?????????) SET-L-ME-P22061 
>> comm_activity:range_scan: { cpu_id = 6 }, { distances = [ [0] = 1, [1] = 2, 
>> [2] = 3, [3] = 4, [4] = 5 ] }
>> 
>> $ python3 tt.py  ~/git-repositories/communication_5c/build/test_session/
>> {'distances': [1, 2, 3, 4, 5]}
>> 
>> ----
>> 
>> bt2$ babeltrace2  ~/git-repositories/communication_5c/build/test_session/
>> [10:04:51.068836215] (+?.?????????) SET-L-ME-P22061 
>> comm_activity:range_scan: { cpu_id = 6 }, { distances = [ [0] = 0, [1] = 
>> 1072693248, [2] = 0, [3] = 1073741824, [4] = 0 ] }
>> 
>> $ python3 tt.py  ~/git-repositories/communication_5c/build/test_session/
>> {'distances': [0, 1072693248, 0, 1073741824, 0]}
>> 
>>> Are you instrumenting python code, or are you instrumenting a C/C++ 
>>> application with lttng-ust?
>> 
>> The latter.
>
> thanks for the extra info, it's much clearer to me now!
>
> When you're passing the pointer to the array of floats, you get the weird 
> values because there's not an explicit type conversion so the raw data of the 
> floats is being interpreted as integer values.
>
> As there is no lttng-ust function to handle an array of floats at this point 
> there's a few other options you could try.
>
> In your example you're using a fixed length array, so it's possible to define 
> a tracepoint with 5 floats like this
>
> ```
> /* note, this version isn't doing any bounds checking */
> LTTNG_UST_TRACEPOINT_EVENT(
>  my_provider
>  my_tracepoint
>  LTTNG_UST_TP_ARGS(
>    float*, my_float_array_ptr
>  )
>  LTTNG_UST_TP_FIELDS(
>    lttng_ust_field_float(float, float0, my_float_array_ptr[0])
>    lttng_ust_field_float(float, float1, my_float_array_ptr[1])
>    ...
>    lttng_ust_field_float(float, float4, my_float_array_ptr[4])
>  )
> )
> ```
>
> Here's a couple of ideas for a more general approach
>
> 1. Define a tracepoint with an integer length, maybe the base pointer 
> identifier (for correlation) and the offset. You'd have to have a smaller 
> helper that calls the tracepoint one or more times depending on the length of 
> your array.
>
> Something a bit like the following. Note that I didn't do a full test.
>
> ```
> LTTNG_UST_TRACEPOINT_EVENT(
>  my_provider,
>  my_tracepoint,
>  LTTNG_UST_TP_ARGS(
>    float*, my_float_array_ptr,
>    size_t, my_float_array_len,
>    size_t, my_float_array_offset,
>  )
>  LTTNG_UST_TP_FIELDS(
>   lttng_ust_field_integer_hex(void*, array_ptr, my_float_array_ptr)
>   lttng_ust_field_integer(size_t, array_len, my_float_array_len,
>   lttng_ust_field_integer(size_t, array_offset, my_float_array_offset),
>   lttng_ust_field_float(float, float1, (my_float_array_offset + 0 >= 
> my_float_array_len) ? 0.0 : my_float_array_ptr[my_float_array_offset + 0])
>   lttng_ust_field_float(float, float1, (my_float_array_offset + 1 >= 
> my_float_array_len) ? 0.0 : my_float_array_ptr[my_float_array_offset + 1])
>   ...
>   lttng_ust_field_float(float, float1, (my_float_array_offset + n >= 
> my_float_array_len) ? 0.0 : my_float_array_ptr[my_float_array_offset + n])
>  )
> )
> ```
>
> 2. or, use a structure with N floats and a process similar to the above.
>
> It's a bit inelegant until there is support for array of floats in lttng-ust.
>
> 3. Use the int or char array and rebuild the float values in post
>
>
>
> Hope this gives you an option to work with for the moment.

Definitely! Much appreciated :-)

> thanks,
> kienan

Herman

>>>> I have no problems with arrays of integers, though.
>>>> 
>>>> Is the float array feature not yet implemented in lttng 2.13?
>>>> Will it be in 2.14? (Also the "BLOB" feature is useful for me.)
>>>> 
>>> 
>>> I don't see any changes in lttng-ust for 2.14 that allow for tracepoints 
>>> that have float arrays or blobs. lttng-modules 2.14 does add support for 
>>> CTF 2. The support for CTF 2 in lttng-tools 2.14 will remain at an 
>>> experimental level.
>>> 
>>> From the release notes:
>>> 
>>>      CTF 2 output isn't enabled by default and is intended solely for
>>>      preview and feedback purposes: having an experimental support will
>>>      allow users to test and evaluate CTF 2 until we make it the
>>>      default output format.
>>> 
>>> 
>>>> Thanks for your help.
>>>> 
>>>> Herman
>>> 
>>> thanks,
>>> kienan
>> 
>> Herman
>> 
>>> [1]: https://git.lttng.org/?p=lttng- 
>>> tools.git;a=blob_plain;f=ChangeLog;hb=11f963148df9597de6d891d96d748bae5ced7149
>>> 
>>> 
>
>