termux-ndk VS jotai-benchmarks

Next, we need to set up the Android NDK (Native Development Kit) to compile the llama.cpp project. Visit the Termux-NDK repository and download the latest NDK release. Extract the ZIP file, then set the NDK path in Termux:

10 - Install the Android SDK by copying and pasting the following command in the console: wget https://github.com/lzhiyong/termux-ndk/releases/download/ndk-r23/android-ndk-r23c-aarch64.zip unzip android-ndk-r23c-aarch64.zip export NDK=~/android-ndk-r23c-aarch64

You can use termux. r/termux. For ndk you can use https://github.com/Lzhiyong/termux-ndk for cmake below 3.14 (I think) you have to use a prooted distro. For build-tools you cant use sdkmanager unless you use android studio in GUI mode . you have to download manually . there is a website for that . and that's how you build android apps in Android.

And for ndk I guess you can use this if there is no arm official build. https://github.com/Lzhiyong/termux-ndk

And the docs : https://github.com/Lzhiyong/termux-ndk/tree/master/build-app

Found this https://github.com/Lzhiyong/termux-ndk/blob/master/patches/align_fix.py to patch the binary. But why do I need to patch it?

Did you install ndk from here? https://github.com/Lzhiyong/termux-ndk

You can check out this repo

But let me suggest Jotai to you. That's a collection with about 30K C programs. They all run without undefined behavior (as per kcc or frama-C). Each program is self-contained. But each program also runs for a very short time. So, I recommend you use CFGGrind to count the number of instructions executed. CFGGrind can separate instructions per function, and each Jotai benchmark consists of a single function. For instance, to count the total number of instructions executed by function foo, you can do (assuming that profile data was saved into a file called `test.cfg`):

We, at UFMG, have been working on a methodology to generate benchmarks in C. We have a working collection of benchmarks here with a bit more than 30K executable programs. Benchmarks are single functions mined from open-source repositories. We have designed a domain-specific language to generate inputs for them. We use psyche-c to infer missing types and declarations. We use kcc and AddressSanitizier to filter out as much undefined behavior as possible. We use CFGGrind to check input coverage and to count the number of instructions executed. These benchmarks can be used in many ways: to stress test compilers; to autotune predictive compilation tasks; to analyze the dynamic behavior of programs; to improve compiler optimizations; etc. We have a technical report here.