I'm getting error on compiling x264 for iOS.
I have Xcode Version 5.0 (5A1413) with Apple LLVM version 5.0 (clang-500.2.75) (based on LLVM 3.3svn). I'm compiling x264-snapshot-20130925-2245.
Config:
CC=/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang ./configure \
--host=arm-apple-darwin \
--sysroot=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS7.0.sdk \
--prefix=armv7 \
--extra-cflags='-arch armv7' \
--extra-ldflags="-L/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS7.0.sdk/usr/lib/system -arch armv7" \
--enable-pic \
--enable-static
Getting error:
common/arm/cpu-a.S:29:7: error: unknown token in expression
.align
^
common/arm/cpu-a.S:139:5: error: instruction 'suble' can not set flags, but 's' suffix specified
subles ip, ip, #1
^
The relevant change in the Xcode 5 toolchain is that the LLVM compiler now defaults to using the built-in assembler, and the built-in assembler requires more strict adherence to the ARM Unified Assembly Language.
There are two ways to get it to compile with the Xcode 5 toolchain:
Give clang the flag -no-integrated-as. Adding it to --extra-cflags ought to work. (The flag worked for me compiling individual files but I never worked it into configure.) Consider this a workaround.
Fix the assembly source code in x264's common/arm subdirectory. This is pretty easy actually, and it's what I did. This is the right fix. BTW, I'm about to submit a patch to x264 with these changes.
The assembler emits many errors and they fall into four categories:
In cpu-a.S, the ".align" directive should be ".align 2". (Apparently it used to default to 2, now the 2 must be explicit.)
Several subles and sublts instructions in multiple files. These are variants of "sub" (subtract), followed by a condition (2 characters) and the "s" suffix. Now the "s" has to precede the condition. Thus "subles" => "subsle" and "sublts" => "subslt".
A fair number of ldrd instructions in various files. This instruction means "load register, double (from memory)". It loads 2 32-bit words from memory into registers. It used to be OK to name only the first register; now both need to be named. They're always adjacent. So "ldrd r2, whatever" needs to become "ldrd r2, r3, whatever". "ldrd r6, something" becomes "ldrd r6, r7, something". Etc.
In pixel-a.S, there's an instruction "vmov.32 r0, r1, d0". This is incorrect. vmov.32 means move a 32-bit quantity, yet the arguments say to move d0 (64 bits) into r0 and r1. Apparently the old compiler took the ".32" part as a hint. I believe it should be "vmov r0, r1, d0" and that change works for me - but I don't have absolute proof that is the correct instruction.
Many thanks to gparker on the Apple Developer Forum! I could not have figured this out without his/her help. Link to forum discussion, Apple ID required.
So far the only solution seems to be --disable-asm.
I think disabling assembler optimizations is a bad solution.
After a long research I've found the root of the problem: clang during assembler compilation uses ASFLAGS not CFLAGS, so adding --extra-asflags="-arch armv7" solves the problem
./configure \
--host=arm-apple-darwin \
--sysroot=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS.sdk \
--prefix=armv7 \
--extra-cflags="-arch armv7" \
--extra-asflags="-arch armv7" \
--extra-ldflags="-arch armv7" \
--enable-pic \
--enable-static
NOTE: For bitcode support just add -fembed-bitcode to all extra flags parameters
Related
I am trying to build LLVM compilers so that I can enable OpenMP on the Apple M1.
I am using the LLVM development tree, (since I saw some OpenMP runtime go into that for this recently).
I have ended up with this script to invoke cmake:
# Xcode, Ninja
BUILD_SYSTEM=Ninja
BUILD_TAG=Ninja
cmake ../llvm \
-G$BUILD_SYSTEM -B ${BUILD_TAG}_build \
-DCMAKE_OSX_ARCHITECTURES='arm64' \
-DCMAKE_C_COMPILER=`which clang` \
-DCMAKE_CXX_COMPILER=`which clang++` \
-DCMAKE_BUILD_TYPE=Release \
-DCMAKE_BUILD_WITH_INSTALL_RPATH=1 \
-DCMAKE_INSTALL_PREFIX=$HOME/software/clang-12.0.0/arm64 \
-DLLVM_ENABLE_WERROR=FALSE \
-DLLVM_TARGETS_TO_BUILD='AArch64' \
-DLLVM_ENABLE_PROJECTS='clang;openmp,polly' \
-DLLVM_DEFAULT_TARGET_TRIPLE='aarch64-apple-darwin20.1.0'
The compilers used here are
$ /usr/bin/clang --version
Apple clang version 12.0.0 (clang-1200.0.32.27)
Target: arm64-apple-darwin20.1.0
Thread model: posix
InstalledDir: /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin
ninja can then successfully build clang, clang++ and the OpenMp runtime and install them. (As simple, Arm64 images targeting Arms64)
$ file ~/software/clang-12.0.0/arm64/bin/clang
/Users/jcownie/software/clang-12.0.0/arm64/bin/clang: Mach-O 64-bit executable arm64
$ ~/software/clang-12.0.0/arm64/bin/clang --version
clang version 12.0.0 (https://github.com/llvm/llvm-project.git 879c15e890b4d25d28ea904e92497f091f796019)
Target: aarch64-apple-darwin20.1.0
Thread model: posix
InstalledDir: /Users/jcownie/software/clang-12.0.0/arm64/bin
Which all looks sane, except that when I try to compile anything with them they are missing the include path to get system headers.
$ ~/software/clang-12.0.0/arm64/bin/clang hello.c
hello.c:1:10: fatal error: 'stdio.h' file not found
#include <stdio.h>
^~~~~~~~~
1 error generated.
So, after all that,
Does anyone know how to fix that include path problem?
Does anyone know how to configure and build a fat binary for the compilers (and libraries) so that the x86_64 embedded compiler targets x86_64 and the aarch64 binary aarch64? (This is what the Xcode clang and clang++ do...)
My attempt at this ended up with a compiler fat binary where both architectures targeted x86_64 :-(
Thanks
You can set -DDEFAULT_SYSROOT=/path/to/MacOSX11.1.sdk at build time or do export SDKROOT=/path/to/MacOSX11.1.sdk at runtime.
You need to compile with clang -arch arm64 -arch x86_64 to get a fat binary out of clang. You need to do this for Apple clang as well.
UPDATED 8 Feb 2021
Homebrew now supports the M1 based Arm machines, so using that is a better answer than the one below.
The info below is potentially still useful if you want to do this on your own, but using brew is likely to be much simpler.
Pre-brew answer
I haven't found a clean solution, but in case it helps anyone else, I do have a horrible hack.
The full recipe, then is configure with this script, then build and install.
# Xcode, Ninja
BUILD_SYSTEM=Ninja
BUILD_TAG=ninja
INSTALLDIR=$HOME/software/clang-12.0.0/arm64
cmake ../llvm \
-G$BUILD_SYSTEM -B ${BUILD_TAG}_build \
-DCMAKE_OSX_ARCHITECTURES='arm64' \
-DCMAKE_C_COMPILER=`which clang` \
-DCMAKE_CXX_COMPILER=`which clang++` \
-DCMAKE_BUILD_TYPE=Release \
-DCMAKE_INSTALL_PREFIX=$INSTALLDIR \
-DLLVM_LOCAL_RPATH=$INSTALLDIR/lib \
-DLLVM_ENABLE_WERROR=FALSE \
-DLLVM_TARGETS_TO_BUILD='AArch64' \
-DLLVM_DEFAULT_TARGET_TRIPLE='aarch64-apple-darwin20.1.0' \
-DDEFAULT_SYSROOT="$(xcrun --show-sdk-path)" \
-DLLVM_ENABLE_PROJECTS='clang;openmp;polly;clang-tools-extra;libcxx;libcxxabi' \
# -DLLVM_ENABLE_PROJECTS='clang;openmp;polly'
That gives a compiler that finds the right headers, but won't link successfully if OpenMP is used because it doesn't pass on any useful -L path or add a necessary rpath.
To overcome that I created a small shell script that sits in my ~/bin, at the front of my $PATH, which adds those extra linker flags.
#
# A truly awful hack, but it seems necessary.
# Install this with execute permissions as clang and clang++ in
# a directory early in your path, so that it is executed when clang or
# clang++ is needed.
#
# For brew...
INSTALLDIR=/usr/local/opt/llvm
# For a local build.
INSTALLDIR=${HOME}/software/clang-12.0.0/arm64/
# Find out the name of this file, and then invoke the same file in the
# compiler installation, adding the necessary linker directives
CMD=`echo $0 | sed "s/\/.*\///"`
${INSTALLDIR}/bin/${CMD} -L${INSTALLDIR}/lib -Wl,-rpath,${INSTALLDIR}/lib $*
I am not recommending this particularly; there should clearly be a better way to make it work, but it'll do for now, and lets me get back to using the compiler rather than building it!
I was able to build with -DDEFAULT_SYSROOT="$(xcrun --show-sdk-path)" -DCMAKE_INSTALL_PREFIX=/Users/foo/lokal/ and install into the lokal/bin lokal/lib path. Once that is done you can use LD_LIBRARY_PATH=/Users/foo/lokal/lib and all the libraries should be found without mucking with anything else rpath related.
I have been using clang-3.5 to happily build bitcode versions of musl libc and
use the result to produce nice stand alone executables.
Recent attempts with clang-3.8 have not been so happy. It seems that
the bitcode clang-3.8 generates uses functions defined in
compiler-rt/lib/builtins
Typical examples of functions I find polluting the bitcode are mulxc3, mulsc3, and muldc3. I can solve this by linking against libgcc, or even the llvm alternative if I had any clear idea of what that was. Though I would rather prevent the problem from happening in the first place.
I have seen mention of flags like rtlib=compiler-rt etc, but have found precious little documentation on the subject.
So here are some simple questions.
Is it possible to prevent clang from using the compiler-rt/lib/builtins
in the emitted bitcode? Or if not
Does llvm produce a version of libgcc that I could use. Actually I would
probably build a bitcode version of it, but that is besides the point.
Love to hear some guidance on this.
Added 12/8/2016: So I will illustrate my issues with a particular workflow that
people can reproduce if they wish, or, more likely, just point out where I am being stupid.
So start by checking out:
musllv
and follow the instructions in the README.to compile (here I am using clang-3.8 on ubuntu 14.04)
WLLVM_CONFIGURE_ONLY=1 CC=wllvm ./configure --target=LLVM --build=LLVM
make
cd lib
extract-bc -b libc.a
you will also need the bitcode of a simple executable. I will use nweb.c here.
wllvm nweb.c -o nweb
extract-bc nweb
Now we can do things like:
clang -static -nostdlib nweb.bc libc.a.bc crt1.o libc.a -o nweb
This workflow goes smoothly for clang-3.5 but for clang-3.8 we get:
clang -static -nostdlib nweb.bc libc.a.bc crt1.o libc.a -o nweb
/tmp/libc-f734a3.o: In function `cpowl':
libc.a.bc:(.text+0xbb9a): undefined reference to `__mulxc3'
/tmp/libc-f734a3.o: In function `cpowf':
libc.a.bc:(.text+0x38f7d): undefined reference to `__mulsc3'
/tmp/libc-f734a3.o: In function `csqrt':
libc.a.bc:(.text+0x78fc3): undefined reference to `__muldc3'
/tmp/libc-f734a3.o: In function `cpow':
libc.a.bc:(.text+0xafafc): undefined reference to `__muldc3'
clang-3.8: error: linker command failed with exit code 1 (use -v to seeinvocation)
So as #paul-brannan points out we could try
clang -static -nostdlib --rtlib=compiler-rt nweb.bc libc.a.bc crt1.o libc.a -o nweb
But this is where I am probably being stupid, because I get:
clang-3.8: warning: argument unused during compilation: '--rtlib=compiler-rt'
irregardless of whether I use it as a linking or compiling flag.
OK so I finally managed to make headway on this. I built llvm-3.8.1 together with the compiler-rt project using wllvm and wllvm++.
One of the build products was libclang_rt.builtins-x86_64.a,
and from this archive I was able to extract the bitcode module
libclang_rt.builtins-x86_64.bc
using the command:
extract-bc -b libclang_rt.builtins-x86_64.a
This bitcode module has definitions for those pesky instrinsics like
__mulxc3, __mulsc3, and __muldc3.
Hallelujah!
I have been trying to enable the gold linker on FreeBSD to use the link time optimizations. I made gold from the binutils under /usr/ports. After building binutils using make -k install clean i got ld under /usr/bin and in the directory /usr/local/bin i got ld, ld.gold and ld.bfd.
Now while trying to use link time optimization for the simple example programs here http://llvm.org/docs/GoldPlugin.html (a.c and b.c under the heading 'Examples of Link Time Optimization') i entered the four commands as follows:
clang -flto a.c -c -o a.o
ar q a.a a.o
clang b.c -c -o b.o
clang -flto a.a b.o -o main
I got the following error:
usr/bin/ld: unrecogonized option '-plugin'
usr/bin/ld: use the --help option for usage information
clang: error: linker command failed with exit code 1 (use -v to see invocation)
Is there the problem with the linker that ld.gold is not being called. Should I replace the ld with ld.gold? Does the linker looks in the right directiry for the .so plugins?
The LLVMgold.so and libLTO.so shared objects are in the directory /usr/local/llvm-devel/lib/.
I cannot find the directory where clang is installed. I am not sure where to make the bfd-plugins directory and add the symlinks to LLVMgold.so and libLTO.so.
I am using freebsd 10.1 release. How to enable the gold linker for link time optimizations?
also how can I enable it to be the default linker?
You may want to use ld.gold instead of ld. It is installed at /usr/local/bin/ld.gold. If you are using a Makefile, it would work by setting LD variable to ld.gold, either by modifying your Makefile or specifying it on command line. Example in case you are using lang/clang37:
gmake all CC=clang37 LD=ld.gold
EDIT:
It would be even more neat if you add -fuse-ld=gold to your LDFLAGS:
LDFLAGS=-fuse-ld=gold
I'm not sure ld.bfd allows plugins, but I could be wrong.
Your /usr/bin/ld should be a symlink to whatever linker you want. You can change which linker is used by using binutils-config. Check the man-page here: http://www.linuxhowtos.org/manpages/8/binutils-config.htm. I realise this is a Linux link, but it's directed at binutils itself rather than linux-specifically.
It should be something along the lines binutils-config --gold.
On my Gentoo box it is binutils --linker=gold
EDIT: As pointed out, binutils-config doesn't work on BSD it seems. You can still manually update the symlinks though, the downside is that there might be a few of them.
You can find out which ld is used by your compiler by using gcc -print-prog-name=ld or clang -print-prog-name=ld. The file printed should be a symlink you can re-create to point to ld.gold as oposed to ld.bfd.
I'm trying to build a C library (GMP 6.0.0) for arm64 for use on iOS. I'm running the configure script with the invocation below (compiler is as found using xcrun --find).
./configure \
CC="/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang" \
CPP="/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang -E" \
CPPFLAGS="-target arm64-apple-darwin -isysroot /Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS7.1.sdk/ -miphoneos-version-min=7.0" \
--host=aarch64-apple-darwin
However this fails at the following line ("long long reliability test 1"):
checking compiler /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang -O2 -pedantic -target arm64-apple-darwin -isysroot /Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS7.1.sdk/ -miphoneos-version-min=7.0... no, long long reliability test 1
configure: error: could not find a working compiler, see config.log for details
Full config.log available here. It shows multiple warning and errors for the long long reliability test compile, including the following:
conftest.c:9:1: warning: type specifier missing, defaults to 'int' [-Wimplicit-int]
f(){static const struct{t1 n;t1 src[9];t1 want[9];}d[]={{1,{0},{1}},};t1 got[9];int i;
^
conftest.c:10:44: error: implicit declaration of function 'h' is invalid in C99 [-Werror,-Wimplicit-function-declaration]
for(i=0;i<1;i++){if(e(got,got,9,d[i].n)==0)h();g(i,d[i].src,d[i].n,got,d[i].want,9);if(d[i].n)h();}}
^
conftest.c:10:48: error: implicit declaration of function 'g' is invalid in C99 [-Werror,-Wimplicit-function-declaration]
for(i=0;i<1;i++){if(e(got,got,9,d[i].n)==0)h();g(i,d[i].src,d[i].n,got,d[i].want,9);if(d[i].n)h();}}
^
conftest.c:10:100: warning: control reaches end of non-void function [-Wreturn-type]
for(i=0;i<1;i++){if(e(got,got,9,d[i].n)==0)h();g(i,d[i].src,d[i].n,got,d[i].want,9);if(d[i].n)h();}}
^
Using --host=none it works fine, but I'd really like to figure out how to build it with assembly optimized for arm64.
My system is x86_64-apple-darwin13.1.0 (or coreisbr-apple-darwin13.1.0 according to config.guess), an early '11 Core i7 MBP running OS X v10.9.2. I'm using Xcode 5.1 (5B130a).
Any help appreciated.
EDIT 1
Compiling for ARMv7 passes configure, but fails on make (full configure/make output here), apparently while compiling some assembly:
tmp-dive_1.s:165:18: error: unexpected token in '.section' directive
.section .rodata
^
EDIT 2
#MarcGlisse: By forcing clang to ignore the errors as suggested (-Wno-...) arm64 passes configure, but then fails on make (full output here):
tmp-mul_1.s:59:2: error: unrecognized instruction mnemonic
bcc Lfi1
^
tmp-mul_1.s:60:2: error: unrecognized instruction mnemonic
beq Lfi2
^
As a side note: these commits, I assume intended to remove the need for suppressing the error, don't seem to work i.e. I get the same error when removing the supression.
For armv7, using these commits as suggested fixes the .section error, but make fails later on with the following (full output here):
tmp-mode1o.s:64:2: error: unknown directive
.protected ___gmp_binvert_limb_table
^
EDIT 3
Using the suggested edits, armv7, armv7s, i386 and x86_64 now all compile with assembly!
For arm64 the edits get it past the previous error, but now gives several errors about an invalid input constraint 'rZ', all in the same file (full output here):
divrem_1.c:237:5: error: invalid input constraint 'rZ' in asm
udiv_qrnnd_preinv (*qp, r, r, nshift, d, dinv);
^
../gmp-impl.h:3062:2: note: expanded from macro 'udiv_qrnnd_preinv'
add_ssaaaa (_qh, _ql, _qh, _ql, (nh) + 1, (nl)); \
^
../longlong.h:551:7: note: expanded from macro 'add_ssaaaa'
: "rZ" (ah), "rZ" (bh), "%r" (al), "rI" (bl) __CLOBBER_CC)
EDIT 4
After commenting out add_ssaaaa and sub_ddmmss in longlong.h and editing some more assembly instructions in gcd_1.asm (blo to b.lo etc.), it now fails with several of the following errors (full output here):
tmp-invert_limb.s:75:22: error: immediate value expected for shifter operand
add x1, x1, x2, lsr 1
^
tmp-invert_limb.s:75:22: error: invalid operand for instruction
add x1, x1, x2, lsr 1
^
I'll post a total diff later.
EDIT 5
Ok, that gets us another step further, but it now hits into (full output here):
tmp-invert_limb.s:52:2: error: ADR/ADRP relocations must be GOT relative
adrp x1, approx_tab
^
If this keeps going it might be better to continue this via email.
"Support for ARM64 alias Aarch64 alias ARMv8"
https://gmplib.org/gmp6.0
Copyright issues aside...
GMP may not be compatible with Apple's proprietary CPU, and may not be compatible with Clang/LLVM. Being a GNU project it is probably more thoroughly tested with GCC on non-proprietary chipsets.
A quick search shows historically at least there have been issues compiling GMP with clang.
You really should email the GMP community and ask for help on this one.
I've been trying to compile syn64k--for use in Executor, to run System's Twilight (a game I played as a kid). I'm doing this on my macbook pro (lion 10.7.4 with the latest Xcode and command line tools). I mapped CC to gcc -m32 to fix a couple problems I was having, but I got the following:
Making all in native/i386
make[2]: Nothing to be done for `all'.
outgoing=;\
gcc -m32 -maccumulate-outgoing-args -c -x c /dev/null 2> /dev/null && outgoing=-maccumulate-outgoing-args; \
gcc -m32 -S -O2 -fomit-frame-pointer -Wall -static -fno-defer-pop -Wno-unused\ -I./include -I./../include -I. -I../include $outgoing syn68k.c -o ./syn68k.s
syn68k.c: In function ‘s68k_handle_opcode_0x07A3’:
syn68k.c:52968: internal compiler error: in EmitLV_DECL, at llvm-convert.cpp:7475
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://developer.apple.com/bugreporter> for instructions.
make[2]: *** [syn68k.o] Error 1
make[1]: *** [all-recursive] Error 1
make: *** [all-recursive] Error 1
I went to URL, and it doesn't look like a place to really put a bug report about LLVM. I the file 'syn68k.c' is about 50k lines long after the preprocessor is done with it, so I really don't thing I can make a reproducible test case to show the bug.
Turns out this issue is already on the github project (it is the only issue listed, ha). MaddTheSane says that this happens because clang does not support global register variables (as I verified here).
I don't really have much more than a basic understanding of how compilation works after you type make, so is there a way I can skip clang or something like that? What do you suggest?
Unfortunately, you need to use gcc, not llvm-gcc to compile this program. The LLVM backend of llvm-gcc does not support global register variables. Switching to clang won't help because it too will choke on the global register variables, for the same reason.
you definitely can skip building clang if its not necessary for what using llvm for.
llvm uses cmake for configuration and the generation of makefiles. you can modify the .cmake files to create a configuration that disables any modules you don't want to create.
http://developer.apple.com/bugreporter is definitely the right place to report a bug against any compiler distributed by Apple, even if it may not look like it. That said, there's no point to filing a bug report against anything other than clang. (BTW, the size of the testcase doesn't really matter.)
If you don't know enough C to hack the source code, it might be worth a shot to grab a newer version of gcc from macports or something like that.