How do you compile Halide for iOS? - ios

The README claims it can compile to armv7, but I cannot find the magic incantation to make it work.
I started down the rabbit hole of changing the Makefile to set the arch=armv7, fixing the resulting compilation errors, etc, but that doesn't seem like the right way to go about it.
There recommended cmake flags are:
cmake -DLLVM_TARGETS_TO_BUILD="X86;ARM;NVPTX" -DLLVM_ENABLE_ASSERTIONS=ON -DCMAKE_BUILD_TYPE=Release ..
But alas, the bin directory contains only a .a and a .so, both of which are compiled for x86_64. There are no dylibs.
I can successfully run the test iOS app in the simulator, linking with the x86 libraries, but I cannot build on a device since there are no arm binaries.
Here is a link to the Halide test app I'm trying to build:
https://github.com/halide/Halide/tree/master/apps/HelloiOS

You should use AOT compilation for iOS. The JIT in principle works on ARM (the architecture), but not on iOS (the OS).

Clarification: are you trying to build Halide to run on ARM, or merely to generate code for ARM? (If the latter, any target will do, as all builds of Halide can generate code for all known targets.)

Related

How can I make Kotlin-native executable with runtime debug information?

In Personally, I am modifying and testing some features of the kotlin-native memory manager.
But, I am struggling to find a way to create an OSX XCode executable, including debugging information of the C++ sources in kotlin-native runtime.
'kotlinc -g ..' generates only debug informations for Kotlin sources.
Please let me know how to build debugging version of runtime and link the debug information into sample kotlin apps.
I am using Kotlin-native 1.3.72, and Xcode 11.5
I think the only available option here is to re-build Kotlin/Native runtime with adding -g compiler option to the Gradle script. See this file and line, probably it should be the correct place.

Is it possible to compile a X86 program on a jailbroken iOS device?

I have an iPad running iOS 13.5 and it is jailbroken.
I installed clang compiler and vim editor on the Cydia store, so I can write code with vim and compile the source code with clang. Both compilation and execution are successful.
Now I have an idea, I know there is a method called cross-compile, so is it possible to compile a C program on iPad that can run in macOS or Windows?
You can certainly compile for macOS. All you need is the SDK with header files and library stubs, which you can either copy from Xcode, or grab from here.
You can either pass its folder to the compiler with -isysroot, or place/symlink it to /usr/share/SDKs/MacOSX.sdk where the compiler will find it automatically.
Once you've done that, you can run:
clang --target=x86_64-apple-darwin -o t t.c
Compiling for Linux or Windows would work in similar fashion with --target=x86_64-linux-gnu and --target=x86_64-windows-msvc respectively, but in addition to a suitable SDK will also require a custom linker. You could presumably build LLVM's lld for iOS and then pass -fuse-ld=ld.lld for Linux or -fuse-ld=lld-link for Windows, but I haven't been able to find this as precompiled binary for iOS.

How does one build a OpenSSL library for Project Catalyst?

I have to support OpenSSL in my project in building my iPad app for UIKitForMac. Currently, I get these errors.
Building for UIKit for Mac, but the linked library 'libssl.a' was built for freestanding. You may need to restrict the platforms for which this library should be linked in the target editor.
Building for UIKit for Mac, but the linked library 'libcrypto.a' was built for freestanding. You may need to restrict the platforms for which this library should be linked in the target editor.
I was reading about XCFrameworks, but Apple really hasn't put out much information here. Has anyone figured out build scenarios?
The solution in the comments doesn't work for me. However, I just build to different libs: iOS as I used to and another one for Catalyst by adding the build parameters: -target x86_64-apple-ios13.0-macabi and defining Mac SDK in -isysroot. After that, I just conditionally add each of the libraries for each build version and it works.
Amid mounting frustration following many failed attempts and Google searches, I successfully built openSSL 1.1.1g for Catalyst, compiled my project, linked openSSL and launched the app on my Mac by doing the following:
I used the same directory in to which I had previously extracted and built openSSL for IOS.
Following instructions here, I edited <openSSL directory>/Configurations/10-main.conf. Scrolling down to the "darwin64-x86_64-cc" section, I added a second CFLAGS line:
CFLAGS => add("-target x86_64-apple-ios13.0-macabi"),
In the openSSL directory, execute ./Configure darwin64-x86_64-cc -shared Note that I've seen several other versions of this Configure statement, some with many more options. This command worked for me, but I'm not sure what all the other variations do. You may want to research this further.
Execute make clean to clear all the objects from the prior IOS build
Execute make This successfully built openSSL.
In Xcode, under -> General -> "Frameworks, Libraries and Embedded Content" I removed both libcrypto.a and libssl.a this was a critical step
Switching tabs to Build Phases -> Link Binary With Libraries, verify both archives are removed from this section as well. (It appeared that removing them in step 5 also cleared them in step 6, but I'm not certain).
Back on General -> Frameworks, click the + to add new entries, select "add other" in the lower left corner of the popup window, then provide the path to the newly built libcrypto.a. Repeat for libssl.a
Delete the derived data in a terminal window (I'm not certain this was necessary, but did it out of an abundance of caution):
cd ~/Library/Developer/Xcode
mv DerivedData DerivedData.old
Build the project in Xcode. This successfully completed.
Steps 5-7 turned out to be critical. Even though I moved and/or completely replaced the prior libraries, when I tried to build in Xcode I would get linker errors that I was building for MacOS Catalina but trying to link something built for MacOS x86.

How to compile OpenCV iOS with ENABLE_BITCODE

When I tried to compile my XCode project with OpenCV 2.4 iOS using XCode 7 + iOS SDK 9, XCode complained that
ld:
'opencv2.framework/opencv2(alloc.o)'
does not contain bitcode. You must rebuild it with bitcode enabled
(Xcode setting ENABLE_BITCODE), obtain an updated library from the
vendor, or disable bitcode for this target. for architecture arm64
clang: error: linker command failed with exit code 1 (use -v to see
invocation)
and refused to link. After some googling, it turns out to be because Apple added a new feature named Bitcode for app optimization within App Store. While OpenCV iOS binary hasn't been updated to include Bitcode, it cannot pass the link stage.
Some reference pointed out a temporary solution to disable ENABLE_BITCODE so the linking could be done without Bitcode. This will prevent the app being compiled for Apple Watches because Bitcode is mandatory for Watch Apps. Therefore my question is, are there some (best easy) ways to compile iOS OpenCV with Bitcode enabled? (better with a download link for compiled framework)
After some search and trial, I figured out a way to compile OpenCV iOS from the source with Bitcode. A compiled binary is also provided here: [v3.0] [v2.4]. [Disclaimer: I am not responsible for the integrity of the compiled binary. Use at your own risk.]
The steps of compilation is basically the same as the official document, with only one extra step.
Download the code with git:
cd ~/<my_working_directory>
git clone https://github.com/Itseez/opencv.git
Make symbolic link for Xcode to let OpenCV build scripts find the compiler, header files etc.
cd /
sudo ln -s /Applications/Xcode.app/Contents/Developer Developer
[Key Step] Change the compilation script to add the extra option for Bitcode: edit ~/<my_working_directory>/opencv/platform/ios/build_framework.py, and locate the line containing -DCMAKE_C_FLAGS. Add a flag of -fembed-bitcode. For example, in the source I got, it's line 55, and will look like
"-DCMAKE_C_FLAGS=\"-Wno-implicit-function-declaration -fembed-bitcode\" " +
after the change. [ref]
Build OpenCV framework:
cd ~/<my_working_directory>
python opencv/platforms/ios/build_framework.py ios
If everything’s fine, a few minutes later you will get ~/<my_working_directory>/ios/opencv2.framework. You can add this framework to your Xcode projects.
P.S. Ask a question, even when you already know the answer is encouraged according to this post on Meta Stackchange.
OpenCV is precisely the kind of software (along with audio and video Codecs) that is likely to have hand-rolled ARM NEON optimisations. The documentation suggests that ~40 functions have had this treatment in OpenCV3.0.
If compiling for LLVM bit-code you'll get the generic (less optimised, implemented in C or C++) versions instead.
Use of Bitcode is optional - except when compiling for Apple watch, where it's hard to imagine you'd run computationally complex image processing anyway. If you're bundling a watch app, override the build setting for bitcode on it only.

How to build ASSIMP Library for iOS (Device and Simulator) with boost-library?

I want to use the ASSIMP library http://assimp.sourceforge.net in an iOS project. Unfortunately, I'm not very experienced with makefiles and that stuff, so I need some help.
I've downloaded the sources and first I tried to build with make (in the code-subfolder)
In the makefile I've added INCLUDEFLAGS = -I/Lib because my boost header-files are in /Lib/boost
Executing make static succeeds with some warnings. A static library (.a) is generated.
Then I tried to add the .a-file to my xcode-project and specified the assimp-header folder as additional include directory (Other Search Paths). Linking failed with the message that the library has not the right architecture (i386 required for the simulator)
file libassimp.a outputs: "libassimp.a: current ar archive random library"
How can I build the library for the i386 architcture and for arm6 or arm7, whatever I need on an iOS device?
Is it ok to use the boost-headers only or is it better/necessary to build boost as a library? Currently I'm using boost headers only, which should be fine since boost is a header only library?!
There is also a cmake - makefile (CMakeLists.txt). cmake is the recommended way of building the library but I don't have any experience with cmake.
Or another thought: Is it possible to build a library via xcode?
The final result should be a library for i386, arm6 and arm7 architecture.
What shall I do? And how?
Edit:
I've just discovered that there are the following preprocessor checks in the file aiDefines.h:
#if defined(_MSC_VER)
// See http://msdn.microsoft.com/en-us/library/b0084kay.
# if defined(_M_IX86)
# define ASSIMP_BUILD_X86_32BIT_ARCHITECTURE
# elif defined(_M_X64)
# define ASSIMP_BUILD_X86_64BIT_ARCHITECTURE
# elif defined(_M_IA64)
# define ASSIMP_BUILD_IA_64BIT_ARCHITECTURE
# else
# error unknown architecture
# endif
#elif defined(__GNUC__)
// See http://gcc.gnu.org/onlinedocs/cpp/Predefined-Macros.html.
# if defined(__x86_32__) || defined(__i386__)
# define ASSIMP_BUILD_X86_32BIT_ARCHITECTURE
# elif defined(__x86_64__)
# define ASSIMP_BUILD_X86_64BIT_ARCHITECTURE
# elif defined(__ppc__)
# define ASSIMP_BUILD_PPC_32BIT_ARCHITECTURE
# else
# error unknown architecture
# endif
#else
# error unknown compiler
#endif
Does this mean, it is not possible to compile the ASSIMP library for ARM architecture?
I personally don't do any iOS development, but I know that others have successfully compiled Assimp for their iDevices using Xcode. An Xcode 3 project should be included with the distribution, although I don't know if you can use it without further modification.
The architecture preprocessor defines are currently only used for logging output (in code/Importer.cpp), and support for ARM has been added to trunk in the meantime (r919, to be exact).
I also had a few issues getting assimp to work on iOS devices.
Here's what I did in case anyone is also having similar issues - similar to Artur Sampaio's example above, but with a few differences:
get the latest assimp from https://github.com/assimp/assimp, i.e., git clone git://github.com/assimp/assimp.git
cd into the assimp directory and open up CMakeLists.txt. For some reason, the make file couldn't find my glut and gl libs, so I commented out the lines in CMakeLists.txt that referred to making the samples, i.e., the "IF ( BUILD_ASSIMP_SAMPLES)" block of code. There's probably a simple way to point to these, but since I didn't need the samples I just did commented those lines out.
cd to port/iOS/ and then sudo ./build_ios.sh (it takes a few minutes to compile all 3 versions of the lib).
now if you cd to assimp/lib/ios and lipo -info libassimp.a you will see that the library is a fat file with i386, arm6 and arm7 architectures (and will work on both the simulator and the arm6 or arm7 device).
to get it to work in XCode 4.3.2, drag&drop the libassimp.a file into my project (from the finder). You don't have to actually copy it over to your project directory, it doesn't hurt if you do though.
for some reason, a build will still fail unless you explicitly link to the libz dynamic library. Click the main project from the XCode file list, select TARGETS, then click on the Build Settings tab, scroll down to the Linking section, and then add "/usr/lib/libz.dylib" under "Other Linker Flags".
the current version of assimp on github seems to have restructured the code somewhat, and while all of the examples I tested from the assimp website work, they all require different, or at least renamed, header files:
Here I've commented out the previous names of the headers and below are the new names:
//#include <assimp.hpp> // C++ importer interface
//#include <aiScene.h> // Output data structure
//#include <aiPostProcess.h> // Post processing flags
#include "Importer.hpp"
#include "scene.h"
#include "postprocess.h"
After this point I was able to use assimp on my iPad.
Also, you probably want to make sure that you have uninstalled other versions of assimp before doing this (eg from macports or brew).
I found branch where one guy already changed project setting for your needs. https://github.com/blandinw/assimp/tree/ios-xcode46/doc
What I did was:
download version 1090 of the Assimp repo: http://assimp.svn.sourceforge.net/viewvc/assimp/trunk/?pathrev=1090
unzip
cd ~/Downloads/trunk/port/iOS
sudo ./build_ios.sh
(replace ~/Downloads/trunk with the path to the unzipped folder)
wait...
check libs at ~/Downloads/trunk/libs/ios/
hope that works

Resources