[DonAurelio]

Fixed

• Rename libPyTorchInterval to libPYTORCHInterval to match the lib${backend^^}Interval naming convention expected by the test wrapper script.
• Validate non-LIFO exit order in btx_pytorchinterval_callbacks.cpp. The current implementation only validated exit-with-no-entry.

Added (tests)

• Added pretty-print support for the pytorch backend's iprof -t trace mode.
• Added eight tests: basic op name, overload name, nested LIFO, non-LIFO, deep nesting, stack underflow, reentracy, multithread.
• Read /backends/pytorch/tests/READMDE.MD for tests description.

[TApplencourt]

"aten::empty\x00" pretty print hsould not print \x00 AFAIK.
Does other backend print the last \x00?

[TApplencourt]

FOr historical reason, pretty print use this pattern for testing:
https://github.com/argonne-lcf/THAPI/tree/devel/backends/itt/tests

[DonAurelio]

This is a pre-existing condition, not introduced in this PR.

"aten::empty\x00" pretty print hsould not print \x00 AFAIK. Does other backend print the last \x00?

iprof -v
v0.0.16

iprof -t clinfo

22:06:05.774998101 - x4710c0s2b0n0 - vpid: 1894843, vtid: 1894843 - lttng_ust_opencl_devices:device_name: device: 0xc1974e0, name: "Intel(R) Data Center GPU Max 1550\x00"
...
22:06:05.775076638 - x4710c0s2b0n0 - vpid: 1894843, vtid: 1894843 - lttng_ust_opencl:clGetDeviceInfo_exit: errcode_ret_val: CL_SUCCESS, param_value_size_ret_val: 4, param_value_vals: "\x00\x00\xC0\x00"
...
22:06:05.775083611 - x4710c0s2b0n0 - vpid: 1894843, vtid: 1894843 - lttng_ust_opencl:clGetDeviceInfo_exit: errcode_ret_val: CL_SUCCESS, param_value_size_ret_val: 22, param_value_vals: "2026.21.6.0.17_160000"
...
22:06:05.775028251 - x4710c0s2b0n0 - vpid: 1894843, vtid: 1894843 - lttng_ust_opencl:clGetDeviceInfo_exit: errcode_ret_val: CL_SUCCESS, param_value_size_ret_val: 1237, param_value_vals: ""