lld | Project moved to https | Compiler library

I have implemented a Convolutional Neural Network in C and have been studying what parts of it have the longest latency.

Based on my research, the massive amounts of matricial multiplication required by CNNs makes running them on CPUs and even GPUs very inefficient. However, when I actually profiled my code (on an unoptimized build) I found out that something other than the multiplication itself was the bottleneck of the implementation.

After turning on optimization (-O3 -march=native -ffast-math, gcc cross compiler), the Gprof result was the following:

Clearly, the convolution2D function takes the largest amount of time to run, followed by the batch normalization and depthwise convolution functions.

The convolution function in question looks like this:

I am working on a p2p application and to make testing simple, I am currently using udp broadcast for the peer discovery in my local network. Each peer binds one udp socket to port 29292 of the ip address of each local network interface (discovered via GetAdaptersInfo) and each socket periodically sends a packet to the broadcast address of its network interface/local address. The sockets are set to allow port reuse (via setsockopt SO_REUSEADDR), which enables me to run multiple peers on the same local machine without any conflicts. In this case there is only a single peer on the entire network though.

This all works perfectly fine (tested with 2 peers on 1 machine and 2 peers on 2 machines) UNTIL a network interface is disconnected. When deactivacting the network adapter of either my wifi or an USB-to-LAN adapter in the windows dialog, or just plugging the usb cable of the adapter, the next call to sendto will fail with return code 10049. It doesn't matter if the other adapter is still connected, or was at the beginning, it will fail. The only thing that doesn't make it fail is deactivating wifi through the fancy win10 dialog through the taskbar, but that isn't really a surprise because that doesn't deactivate or remove the adapter itself.

I initially thought that this makes sense because when the nic is gone, how should the system route the packet. But: The fact that the packet can't reach its target has absolutely nothing to do with the address itsself being invalid (which is what the error means), so I suspect I am missing something here. I was looking for any information I could use to detect this case and distinguish it from simply trying to sendto INADDR_ANY, but I couldn't find anything. I started to log every bit of information which I suspected could have changed, but its all the same on a successfull sendto and the one that crashes (retrieved via getsockopt):

I came across this post discussing how to print file permissions from the stat library, but I'm confused on how the definition of the struct works.

When I've defined structs, I've always defined them like struct struct_type *name. But in the example the struct is defined like struct struct_type name (without being a pointer). From what I can tell, there should be two different ways to access the information depending on how you define the struct.

My command uses return codes and a string reference argument instead of exceptions to detect an error in the execution of a function.

We have some invariant checking that looks like this:

I am trying to link Cortex-M4 firmware with clang + lld. The gcc build works fine. I am using the stock CMSIS linker script with only RAM & ROM size adjusted (bases are the same). Beginning of the script (without comments):

I have an AVPlayer that is playing a moderately large video (~150mb). When loading the video initially, I find that the player remains idle for upwards of 10-15 seconds in the AVPlayerWaitingWhileEvaluatingBufferingRateReason state. My question is simple: how can I prevent AVPlayer from "evaluating the buffering rate reason" for this long and instead move to immediately playing the video?

I am using a custom resource loader (although this same behaviour is exhibited without using a custom resource loader). Here is the relevant code for creating the AVPlayer (all standard boilerplate):

Although High Sierra is no longer supported by Homebrew, but I need to install llvm@13 formula as a dependency for other formulas. So I tried to install it this way:

I wrote a program that converts decimal numbers (date of birth) to binary. Converting the day and month works smoothly, but when it comes to converting the year a problem occurs, for example 2001 is converted to 2521075409 instead of 11111010001. Could you tell me where the problem is?

I have the following (simplified) function using inline assembly, targeting mips: