Building PyTorch without AVX2 on MacOS
In order to quickly explore PyTorch internals, I decided to compile and install a Debug build on my local machine. The first problem was that modern Clang surpris...
AVX2 (1 posts)
Building (3 posts)
Embedding Jupyter-quaility rich visual Python into a static website
We all love our CV/blog websites hosted on GitHub Pages. We also love Jupyter notebooks for revolting the look and feel of daily data processing. Now imagine that...
Setting up Jitsi Meet
Web-conferencing platforms are on the raise during these unprecedented times. On the other side, the vulnerablilities of Zoom and lack of privacy motivates us to ...
Building PyTorch without AVX2 on MacOS
In order to quickly explore PyTorch internals, I decided to compile and install a Debug build on my local machine. The first problem was that modern Clang surpris...
CUDA (4 posts)
Enabling GPU device debugging in HIP
In order to debug a GPU kernel with cuda-gdb, we add -G -O0 to nvcc command line, which in case of CMake would be:
Installing CUDA from DEB packages without graphics
You may want to have your NVIDIA GPU not to be involved in any desktop rendering for many reasons. While this is the default on the headless servers, personal sys...
How to get infinite loops to work in CUDA
The CUDA compiler does not handle infinite loops properly. For instance, the loop below will be completely eliminated from the resulting assembly, along with its ...
How to fix CUDA and avx512vlintrin.h incompatibilty issue
Recent 5.x and 6.x GCC compilers are causing NVCC to produce the following kind of weird compile errors:
GCC (1 posts)
How to fix CUDA and avx512vlintrin.h incompatibilty issue
Recent 5.x and 6.x GCC compilers are causing NVCC to produce the following kind of weird compile errors:
GPU (1 posts)
Enabling GPU device debugging in HIP
In order to debug a GPU kernel with cuda-gdb, we add -G -O0 to nvcc command line, which in case of CMake would be:
Git (2 posts)
Migrating from CVS to Git
CVS is still around for many important projects, making it difficult to scale their development. Tutorials available for this topic are not robust enough for ease...
Working with Overleaf via Git
Overleaf deploys Git to track collaborative modifications to projects. Moreover, a user has an option to work with Overleaf’s Git backend directly. It supports Gi...
HIP (1 posts)
Enabling GPU device debugging in HIP
In order to debug a GPU kernel with cuda-gdb, we add -G -O0 to nvcc command line, which in case of CMake would be:
Jitsi (1 posts)
Setting up Jitsi Meet
Web-conferencing platforms are on the raise during these unprecedented times. On the other side, the vulnerablilities of Zoom and lack of privacy motivates us to ...
LLVM (1 posts)
Using C-Reduce to understand Clang compiler bugs
Suppose we have a crash while compiling huge application from source, e.g. a Python package with native C++ code. A source file fails to compile with the followin...
Linux (2 posts)
Migrating from CVS to Git
CVS is still around for many important projects, making it difficult to scale their development. Tutorials available for this topic are not robust enough for ease...
Use multiple mirrors in apt
There is an interesting and not so well-known feature of APT package manager: the ability to automatically choose a download mirror for every individual operation.
Loongson (1 posts)
Networking (1 posts)
VPN tunnelling for ipv6 only providers
The number of providers having problems with ipv4 support is growing. Recently we came across an ISP, which offers only ipv6, and is able to connect only to ipv6 ...
Overleaf (1 posts)
Working with Overleaf via Git
Overleaf deploys Git to track collaborative modifications to projects. Moreover, a user has an option to work with Overleaf’s Git backend directly. It supports Gi...
PostgreSQL (2 posts)
Error-tolerant PostgreSQL for data rescue
Database corruption always happens before we prepare for it. “Back up or give up” is the most frequently recommended solution. The main reason is database engines...
Updating PostgreSQL version 10 to version 14 in dockerized Zulip instance
Upgrade from Zulip 5 to Zulip 6 requires updating PostgreSQL version from 10 to 14.
PyTorch (1 posts)
Building PyTorch without AVX2 on MacOS
In order to quickly explore PyTorch internals, I decided to compile and install a Debug build on my local machine. The first problem was that modern Clang surpris...
Qt (1 posts)
QWebEngineView remains blank whatever I do
In the most recent version of PyQt5, QWebEngineView refuses to draw any page content. Aparently, the solution is to disable sandboxing, as mentioned in this comme...
Software Engineering (22 posts)
Using C-Reduce to understand Clang compiler bugs
Suppose we have a crash while compiling huge application from source, e.g. a Python package with native C++ code. A source file fails to compile with the followin...
Embedding Jupyter-quaility rich visual Python into a static website
We all love our CV/blog websites hosted on GitHub Pages. We also love Jupyter notebooks for revolting the look and feel of daily data processing. Now imagine that...
Setting up Jitsi Meet
Web-conferencing platforms are on the raise during these unprecedented times. On the other side, the vulnerablilities of Zoom and lack of privacy motivates us to ...
Building PyTorch without AVX2 on MacOS
In order to quickly explore PyTorch internals, I decided to compile and install a Debug build on my local machine. The first problem was that modern Clang surpris...
How to get infinite loops to work in CUDA
The CUDA compiler does not handle infinite loops properly. For instance, the loop below will be completely eliminated from the resulting assembly, along with its ...
How to fix CUDA and avx512vlintrin.h incompatibilty issue
Recent 5.x and 6.x GCC compilers are causing NVCC to produce the following kind of weird compile errors:
Remote profiling with NVIDIA Visual Profiler on a SLURM-based cluster
GPU-equipped clusters are often managed by SLURM job control system. Essentially, developer logs into the frontend node by SSH, builds the application and then qu...
Using CUDA device functions from OpenACC
OpenACC enables rapid transition of serial C/C++/Fortran into GPU-enabled parallel code. However, due to high-level nature, OpenACC does not offer access to GPU-s...
CUDA-like runtime interface for Xeon Phi
The performance power of GPUs could be exposed to applications using two principal kinds of programming interfaces: with manual parallel programming (CUDA or Open...
OpenMP 4.0 on NVIDIA CUDA GPUs
Multiple presentations about OpenMP 4.0 support on NVIDIA GPUs date back to 2012. There is however still very limited OpenMP 4.0 production-ready tools availabili...
Use CUDA 7.0 NVRTC with Thrust
Runtime Compilation (NVRTC) introduced in CUDA 7.0 allows to dynamically compile CUDA kernels during program execution (see example). This functionality allows to...
Get extra 8% perf in bilinear interpolation on GPU using __restrict__ keyword
Starting from GK110 (Tesla Kepler), “const restrict” annotation on kernel argument has an extra GPU-specific meaning: accesses to that argument should go through ...
Thrust/CUDA tip: reuse temporary buffer across multiple transforms
Thrust is a very handy STL-like template library for rapid data processing on GPUs.
On-the-fly modification of LLVM IR code of CUDA sources
Largely thanks to LLVM, in recent years we’ve seen a significant increase of interest to domain-specific compilation tools research & development. With the re...
How to find CUDA's version of LLVM backend
It is well-known that CUDA toolkit uses LLVM backend, but the used version number is not shown. We can use gdb and LLVM API function to print the version string:
NVIDIA Visual Profiler allows to connect 64-bit Linux server from 32-bit Windows
In CUDA 6.0 release an extremely handy feature has been added to Visual Profiler: support for remote profiling. This means that you can run the profiler GUI from ...
Calling CUDA device function from OpenACC Fortran kernel
OpenACC is known to be a fast method of developing quite efficient GPU-enabled applications. It is also possible to mix CUDA kernels and libraries with OpenACC ke...
Jetson K1: bandwidthTest
Chart on the left shows the bandwidths of memory transfers on Jetson K1 (Click to enlarge). For the baseline we also added GTX680M’s host-device and device-host (...
Jetson K1: from unboxing straight to CUDA in 5 steps
We finally got the most wanted Jetson K1 board in the house! In this post we show how to turn a just unboxed tiny board into fully-functional CUDA development nod...
How to break Ubuntu 13.04/14.04 with vanilla CUDA driver and unbreak it back
After installing CUDA driver from NVIDIA website, Ubuntu 13.04/14.04 window manager decorations (Unity, via Compiz) may stop working properly on Optimus machines ...
Improving CUDA profiler output of the MPI-CUDA program
Consider we need to profile the following MPI-CUDA program on GPU cluster. The most obvious way to profile this code on console-only cluster would be to invoke th...
One non-obvious reason of 'Illegal instruction' in GPU code
If cuda-gdb throws Program received signal CUDA_EXCEPTION_4, Warp Illegal Instruction. for the following code line:
Ubuntu (1 posts)
Use multiple mirrors in apt
There is an interesting and not so well-known feature of APT package manager: the ability to automatically choose a download mirror for every individual operation.
Web Conferencing (1 posts)
Setting up Jitsi Meet
Web-conferencing platforms are on the raise during these unprecedented times. On the other side, the vulnerablilities of Zoom and lack of privacy motivates us to ...
Zulip (1 posts)
Updating PostgreSQL version 10 to version 14 in dockerized Zulip instance
Upgrade from Zulip 5 to Zulip 6 requires updating PostgreSQL version from 10 to 14.