Supported Architectures & Tools Limitations¶
Here there is a list of architecture currently supported for profiling:
GPU: Nvidia Tesla V100 (available on AWS as a p3.2xlarge instance)
FPGA: Xilinx VU9P (available on AWS as a f1.2xlarge instance)
CPU: AMD Epyc 7r32 (available on AWS as a f5a.24xlarge instance)
Do you want to request a new architecture for profiling? Contact us here
We are continuously updating Huxelerate Performance Estimator, however, there may be bugs and limitations that we have never experienced. Here’s a list of the ones we know:
The tool is not thread safe, make sure to use a single thread while profiling your application
The tool is able to profile the source code that is compiled to LLVM IR and processed by the
hux-roofline-instrumentcommand. For this reason, external static and dynamic libraries will not be instrumented and profiled except for basic functions from the C/C++ standard math library. When compiling your application warning will be issued if one or more non-profiled functions are detected.
The tool currently does not support profiling of C++ code that makes use of exceptions. To profile C++ code, make sure to compile it using the
-fno-exceptionscompiler flag. Profiling of C++ code with exceptions might result in empty or incorrect performance reports.
Did you experience an issue and you want to report it? Thanks! Please send us an email to email@example.com with all the details and the error provided by the tool.