Upstream clang3.6 still doesn't allow change of calling convension of
__builtin_isnan from the default to "softfp" (via __attribute__((pcs("aapcs"))))
as required by bionic libm.so for 32-bit ARM. Let's hope clang3.7
fix that. See upstream bug http://llvm.org/bugs/show_bug.cgi?id=20958
Change-Id: Ibdcc1783ffa7cf2f36e733983bd6d912ad38df70
Also
1. Header singal.h now includes <sys/ucontext.h>
2. asm/user.h include sys/user.h
3. change use of user_i387_struct to user_fpregs_struct
See bionic 26a8eb50a84e131d34d10d5d167d67e9995399bd
Change-Id: I9592e98b90d773ef5babb428a6a6ef208fe86327
When GCC is built bionic's stdio.h (among other headers) is fixed: __va_list
is replaced with __gnuc_va_list, and put in include-fixed/stdio.h,
unless stdio.h already contains __gnuc_va_list. When GCC is later
used bionic's stdio.h is never included.
The problem in the context of NDK is that all 32-bit Android toolchain
are built with android-9 header and expect to work with all API levels
sometime with sustantially different stdio.h.
This CL stops GCC from generating include-fixed/stdio.h by putting
__gnuc_va_list in it.
See bionic df85f50b and b.android.com/73728
Change-Id: I7a490496efbd491a0376608b9ae6e9e2dfc15fb3
1) Add ldexp. ldexp used to co-exist in both libc.so and libm.so until
the removal from libm.so in ICS. Unfortunately the sole copy of libm_hard.a
in NDK was compiled from JB source and miss ldexp. Code compiled with
hard-float using ldexp may use libc.so which follows soft-float and misbehave
2) Turns out the following (all in fpclassify.o) also exists in recent libc.so
too. Add them in libm_hard.a as well:
__fpclassifyd
__fpclassifyf
__fpclassifyl
__isfinitef
__isfinite
__isfinitel
__isinff
__isinf
__isinfl
__isnanl
__isnormalf
__isnormal
__isnormall
isinf
isnan
isnanf
Change-Id: Ia157fb2911341988de102e55d0c6f6141a9c3dda
1) ssize_t is int, instead of long int
2) WCHAR_MIN and WCHAR_MAX be consistent with bionic's:
respond to __WCHAR_UNSIGNED__, work for both signed (X86/MIPS)
and unsigned (ARM) wchar_t, and work for -fshort-wchar too
Change-Id: Ief1be007571b50c5d63a819df489f9d3f9b47bc8
$NDK/tests/standalone/run.sh fail to run with last update on stdint.h
to android-20 level. Restore stdint.h for fear of the new one breaking
the existing apps
Also fix signal_portable.h to use SIGRTMIN/MAX in old headers
(which got renamed to __SIGRTMIN in new header)
Also change _JBLEN to 1024 which shall be large enough to store CPU
states for forseable future.
Change-Id: Ia9c03558a258c10434eb8e6985adc122532feaa3
It appears that upstream 3.5 still can't accept __builtin_isnan redeclared
with calling convention, eg.
cat > isnan.c <<EOF
int __builtin_isnan(double) __attribute__((pcs("aapcs")));
EOF
clang -target armv5te-none-linux-androideabi -c isnan.c
isnan.c:1:5: error: function declared 'aapcs' here was previously declared without calling convention
int __builtin_isnan(double) __attribute__((pcs("aapcs")));
^
isnan.c:1:5: note: previous declaration is here
Change-Id: Iff466fd913c3deec27aa3ff355a208543028d368
* Add i386/elf_machdep.h, it is used during the build of 32-bit part of
multilib compiler;
* Remove page.h. See https://android-review.googlesource.com/#/c/83299;
* Update other x86 headers from bionic for compatibility which are
required for successful build;
Change-Id: Ia66301849e699b79c657146424a806d0aa22342b
Signed-off-by: Pavel Chupin <pavel.v.chupin@intel.com>
This change moves _types.h from common area to all 32-bits archs.
Required for successful NDK build.
Change-Id: Ife676a285f4b77aef4c33e465a0dbc7cab1481a3
Signed-off-by: Pavel Chupin <pavel.v.chupin@intel.com>
Define __PTHREAD_MUTEX_INIT_VALUE, __PTHREAD_RECURSIVE_MUTEX_INIT_VALUE,
and __PTHREAD_ERRORCHECK_MUTEX_INIT_VALUE in pthread.h for API < 9.
Pthread.h in API >= 9 have those already since bionic
9831ad3ce6bd5b22da16a275ed67e7236eae3d1f
Change-Id: Ia97960ecbfed0811dde70f75cb23951ecdf803af
1. Fixed missing #include <sys/types.h> in android/asset_manager.h for API >= 13
2. Fixed missing #include <stdint.h> in android/rect_manager.h for API >= 14
3. Added JNICALL to JNI_OnLoad and JNI_OnUnload in jni.h
Change-Id: I875ec0b6e6f7c2f433c176f0a738eb78f3c33062
Similar to clang 3.3, it seems that clang 3.4 declares an incompatible
function prototypes for __builtin_isnan(). To workaround this issue,
we should not declare the function prototype for __builtin_isnan().
Note: Though it may be good to fix this in clang type checker, like
__builtin_nexttoward(), however it seems non-trivial since __builtin_isnsn()
might be overloaded. We can revert this after we have fixed the
clang builtin definition.
Change-Id: I329b56225f3d229b0c141bb8414c4369f154efda
wait4() was incorrectly declared in <sys/wait.h>, but only added to the
C library in Android 4.3 (API level 18) [1]
This patch:
- Removes the wait4() declarations for any <sys/wait.h> header before
API level 19.
- Brings the API level 19 <sys/wait.h> which correctly declares wait4().
- Update the libc.so.functions.txt for API level 19 of all platforms
to include the exported symbol.
NOTE: This does not bring the static libc.a to API level 19, which means
it is stuck at API level 9 for ARM.
See [2] for the corresponding NDK tests.
http://b.android.com/19854
[1] 17a8b0db63
[2] https://android-review.googlesource.com/#/c/72000/
Change-Id: I7735473d177c44f43b09bb6738e9914945cd4d0a
In armeabi-v7a, VM still expects JNI functions to follow soft-abi.
Annotate it with __NDK_FPABI__ so compiler follows soft-abi in JNI
functions regardless the presence of -mhard-float in command-line.
It's up to developer to apply JNICALL consistently on JNI functions,
ie. those functions whose first arg has type "JNIEnv*" is likely to be
Change-Id: Ica7cca8ab9498ce00de6ffc7fff35f8cc13aedbf
Annotate __builtin math functions with __NDK_FPABI_MATH__ too.
1. __attribute__((pcs("aapcs"))) has no effect on codegen when __buitlin
function is emitted inlined.
2. When __builtin call into regular function in Android's libm, it has
proper attribute which observes soft float-abi
Change-Id: I11e8b74fc22f3072b9ce849193e4a56e98c92d64
from poll(struct pollfd *, nfds_t, long) to
poll(struct pollfd *, nfds_t, int)
See 11952073af22568bba0b661f7a9d4402c443a888
and ebfbb367fb
Change-Id: Idba600d4b172404f0a1d5d715febcba8a495a770
The mkdtemp() function has always been implemented in the C library
(see [1] for proof) but was missing a declaration in <stdlib.h>, until
Android 4.0.1 [2]
This patch simply adds the missing declaration to former <stdlib.h>
versions. This makes it easier to build certain open-source programs
for Android, because auto-tools probing typically looks at the symbols
in libc.so to define HAVE_MKDTEMP, then the source fails to compile
because of the lack of declaration.
Note that the mkdtemp is also already listed in
platforms/android-*/arch-*/symbols/libc.so.functions.txt, which were
generated by parsing actual system libraries, and thus don't need to
be updated here.
[1] https://android.googlesource.com/platform/bionic/+/android-1.6_r1/libc/stdio/mktemp.chttps://android.googlesource.com/platform/bionic/+/android-1.6_r1/libc/Android.mk
[2] ad1ff2fb26
Change-Id: I838b54bcb5cbfd2fed24cf7649897889ec7a5559
A math library compiled with -mhard-float. Build instructions:
1. cd to a recent Android source tree: bionic.git
fbefb252b09634114977dbd1b48dd42bb2629b83 (internal) in this case
2. Temporarily remove "-mfloat-abi=softfp" from build/core/combo/arch/arm/armv7*
and build/core/combo/TARGET_linux-arm.mk
3. Temporarily add "-mhard-float -mfloat-abi=hard" to LOCAL_CFLAGS
for libm.a in bionic/libm/Android.mk
4. make -B -j32 out/target/product/generic/obj/STATIC_LIBRARIES/libm_intermediates/libm.a
5. Rename it
Change-Id: I107eea48af75e97b448862cfc7aae2e8f8911a63
Allow user code to be compiled with "-mhard-float" for armeabi-v7a
(which implies -mfloat-abi=hard), and either link with
1. Android native APIs which use softfp, or
2. A customized libm_hard.a which is compiled with -mhard-float. Need
-D_NDK_MATH_NO_SOFTFP=1
See tests/device/hard-float/jni/Android.mk for details and restriction
on Clang
Change-Id: I773a842c73368e08b9a6cda0441e95a96fa303b2
1. Sync bionic's, remove mips-specific one
2. Fix statfs.c to transfer the new f_flags too
3. Revise comment about --wrap=symbol
Change-Id: If68818e3dcb2a4056cd0b2e100fcfb8b2156c33a
Bionic does not have actual implementation of those functions.
Further implementations should be introduced in new
API-level in order to prevent break of compatibility.
Change-Id: I0aa65ea28082cf4a88c383ce0d8eceaa5288c027
Signed-off-by: Grigoriy Kraynov <grigoriy.kraynov@intel.com>
This fixes WCHAR_MIN / WCHAR_MAX to be signed / unsigned based on
the target ABI.
This only impacts ARM where wchar_t is really 'unsigned int'.
One can define _WCHAR_IS_ALWAYS_SIGNED to restore the old behaviour.
See http://b.android.com/55749
Change-Id: Idb63b7b11aecd3cfee37f5b6dd2eba78b308876f
This patch ensures that wchar_t is a 32-bit number when targetting
API level < 9 (i.e. Froyo or older).
You can restore the previous (and broken) behaviour by defining
_WCHAR_IS_8BIT at compile time. See http://b.android.com/57267
for more context.
BUG=57267
Change-Id: Ib334e8fe95a3f592d5d4bb157b0f123984133597
1. __set_errno is deprecated and subject to removal in future release
2. elf.h is seen to be included alone, causing compilation error due
to undefined uint32_t and uint64_t
Change-Id: I4ca348a8ba0689eb3880622dcf5c53be470c57e8
To be the same as bionic's. See
ec37237d69289e2bb9836bfa0d071ab958234d00
The new one also no longer undef offsetof and replaces it with
"((size_t) &((TYPE *)0)->MEMBER)" if __compiler_offsetof isn't
defined. The problem with the old behavior are that
1. __compiler_offsetof isn't defined in linux/compiler.h even though
all Android compilers support __builtin_offsetof
2. As as result, offsetof defined in stddef.h to use __builtin_offsetof
is undefined and replaced here. Problem is,
"((size_t) &((TYPE *)0)->MEMBER)" isn't considered a compile-time
constant, and fails some code expect offsetof to be.
Change-Id: I3ebb0de9e1d95305a0890a099360ff6936d24b2f
