aboutsummaryrefslogtreecommitdiffstats
path: root/gcc-4.7
Commit message (Collapse)AuthorAgeFilesLines
* [gcc] Remove "-mstackrealign" option turned on by default on x86.Alexander Ivchenko2014-10-142-2/+2
| | | | | | | | For ndk docs change, please refer to: https://android-review.googlesource.com/#/c/110100/ Change-Id: Icbe13a158511d519312b2a1d3e606c9dd2bff4af Signed-off-by: Alexander Ivchenko <alexander.ivchenko@intel.com>
* Setup x86_64 ABI and add -mssse3 to x86 ABIPavel Chupin2014-06-201-1/+1
| | | | | | | | | | | 32-bit: replace -msse3 by -mssse3 64-bit: setup default options as -msse4.2 -mpopcnt Note: when multilib compiler is used -m32 will match 32-bit options and -m64 or default (neither -m32 nor -m64) will match 64-bit options. Change-Id: Ia20a03f54e3ff5857108e9ab0ae1c4c7c1e6cc7f Signed-off-by: Pavel Chupin <pavel.v.chupin@intel.com>
* Fix GNU4.7/4.8/4.9 libstdc++ for clangAndrew Hsieh2014-04-282-4/+3
| | | | | | | | 1. Undo fix in ed7057a1ece24733fe30423cd94131deb8ccb2ca which precludes __atomic_fetch_add() now supported by clang >= 3.3 2. __float128 is supported by clang >= 3.4 Change-Id: Ic2856368d83d261e3a27e0e24ccc09743afdc68b
* GCC 4.6-4.8: Only quote arguments in pex-win32.c that need it.Ray Donnelly2014-03-171-9/+38
| | | | | | .. it always quoted all arguments, irrespective of whether there was a need to. This means the 32k limit is getting hit more often than it needs to.
* [4.7, 4.8] Backport of two patches for additional SLM-tuning.Alexander Ivchenko2014-01-301-1/+1
| | | | | | | | | | | | | | | | | | | We don't have config/i386/x86-tune.def in 4.8, instead initial_ix86_tune_features is modified. This one is for 4.8 only. 2013-09-12 Yuri Rumyantsev <ysrumyan@gmail.com> * config/i386/x86-tune.def: Turn on X86_TUNE_AVOID_MEM_OPND_FOR_CMOVE for SLM. Update HImode imul cost for Silvermont 2013-12-13 Yuri Rumyantsev <ysrumyan@gmail.com> * config/i386/i386.c (slm_cost): Fix imul cost for HI. Change-Id: I3e6e7b157897e93bc3874738635db4ecf4e7f587 Signed-off-by: Alexander Ivchenko <alexander.ivchenko@intel.com>
* Backport fix to set mips_dbx_regno entries to IGNORED_DWARF_REGNUM by defaultSteve Ellcey2013-10-255-8/+48
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | 2012-12-12 Steve Ellcey <sellcey@mips.com> * config/mips/mips.c (mips_option_override): Set mips_dbx_regno entries to IGNORED_DWARF_REGNUM by default. 2012-12-10 Steve Ellcey <sellcey@mips.com> PR target/54061 rtl.h (IGNORED_DWARF_REGNUM): New. * dwarf2out.c (reg_loc_descriptor): Check for IGNORED_DWARF_REGNUM. (mem_loc_descriptor): Ditto. * config/mips/mips.h (ALL_COP_REG_FIRST): New. (ALL_COP_REG_LAST): New. (ALL_COP_REG_NUM): Redefine using above macros. * config/mips/mips.c (mips_option_override): Set mips_dbx_regno coprocessor entries to IGNORED_DWARF_REGNUM. Error reads: target C++: libicui18n <= external/icu4c/i18n/olsontz.cpp external/icu4c/i18n/decimfmt.cpp: In member function 'virtual icu_50::UnicodeString& icu_50::DecimalFormat::format(const icu_50::StringPiece&, icu_50::UnicodeString&, icu_50::FieldPositionIterator*, UErrorCode&) const': external/icu4c/i18n/decimfmt.cpp:1322:1: internal compiler error: in dbx_reg_number, at dwarf2out.c:10185 See http://gcc.gnu.org/ml/gcc-patches/2012-12/msg00830.html Change-Id: I84110b5b54d8d9262043811eddb0edc1c42303f8
* Extend MIPS GCC4.6/4.7/4.8 -mldc1-sdc1 to control ldxc1/sdxc1 tooIceberg Fu2013-08-302-3/+3
| | | | | | | | | | | | | -mldc1-sdc1 now also controls codegen of ldxc1/sdxc1, the indexed versions (where address of load/store is computed from sum of two registers) of ldc1/sdc1 which are already controlled by this option. Although double (or struct containing double) is always aligned and the stock does guarantee 8-byte alignment, this option is to workaround issue when double is allocated from a custom allocator w/o honoring 8-byte minimal alignment. Change-Id: I79433976509b885b5699d62693fd3154bb752abf
* Add MXU support in gcc4.6/4.7/4.8 with -mmxuAndrew Hsieh2013-08-152-0/+5
| | | | | | For Ingenic MXU. Change-Id: Ie7b465c971e3642b3244ac1a77b6f86be4ab4fea
* [4.6,4.7] Always enable --eh-frame-hdr for static executableAndrew Hsieh2013-08-0710-4/+122
| | | | | | | See 23e3137ee2897464b051599b85a09f130d3ad05d for the reason why. backport 4.8 patch from http://gcc.gnu.org/ml/gcc-patches/2012-09/msg00969.html Change-Id: I9496e11f314e5e2ec9359b993016723ab6366dad
* Merge "Add an internal switch -minline-thumb1-jumptable to ARM gcc4.6/4.7/4.8"Andrew Hsieh2013-08-074-3/+9
|\
| * Add an internal switch -minline-thumb1-jumptable to ARM gcc4.6/4.7/4.8Lai Wei-Chih2013-08-024-3/+9
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Default is off (ie. no changes of behavior), meaning that gcc may generate thumb1 jumptable helper calls (eg. __gnu_thumb1_case_sqi) for switch/case. Use -minline-thumb1-jumptable to compile code which can be linked with compiler-rt where helper __gnu_thumb1_case_* doesn't exist. This is mainly for use to compile NDK gabi++/stlport/libc++ to be independent of libgcc.a. The only affected binaries are all in armeabi. ABIs armeabi-v7a and x86/mips aren't affected. Size differences are libgabi++_static.a 176286 -> 176538 +0.1% libstlport_static.a 1406346 -> 1413318 +0.5% libc++_static.a 2426814 -> 2439074 +0.5% Change-Id: I582898955b28e53e988a91bc1a64b76fba677e58
* | [4.7] Add stack unwinding directives for __aeabi_idiv0 in libgcc.aBen Cheng2013-08-012-0/+8
|/ | | | Change-Id: I32b4178ec6f09c08c5ecf70aabf7532e03774766
* Fix segfault on const init to object addressAndrew Hsieh2013-06-203-2/+56
| | | | | | | | | | | | | | | See b.android.com/56508 and http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54986 Backport the fix in r193189 2012-11-05 Eric Botcazou <ebotcazou@adacore.com> PR tree-optimization/54986 * gimple-fold.c (canonicalize_constructor_val): Strip again all no-op conversions on entry but add them back on exit if needed. Change-Id: Ibedd829e7780ff78e7b90219ebb7c425f7d2fc01
* [4.7, 4.8] Extended Silvermont tuning.Pavel Chupin2013-06-043-5/+151
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Backport r199546 from trunk: 2013-05-31 Yuri Rumyantsev <yuri.s.rumyantsev@intel.com> Igor Zamyatin <igor.zamyatin@intel.com> Silvermont (SLM) architecture performance tuning. * config/i386/i386.h (enum ix86_tune_indices): Add X86_TUNE_SPLIT_MEM_OPND_FOR_FP_CONVERTS. (TARGET_SPLIT_MEM_OPND_FOR_FP_CONVERTS): New define. * config/i386/i386.c (initial_ix86_tune_features) <X86_TUNE_SPLIT_MEM_OPND_FOR_FP_CONVERTS>: Initialize. (ix86_lea_outperforms): Handle Silvermont tuning. (ix86_avoid_lea_for_add): Add new argument to ix86_lea_outperforms call. (ix86_use_lea_for_mov): Likewise. (ix86_avoid_lea_for_addr): Likewise. (ix86_lea_for_add_ok): Likewise. (exact_dependency_1): New function. (exact_store_load_dependency): Likewise. (ix86_adjust_cost): Handle Silvermont tuning. (do_reoder_for_imul): Likewise. (swap_top_of_ready_list): New function. (ix86_sched_reorder): Changed to handle Silvermont tuning. * config/i386/i386.md (peepholes that split memory operand in fp converts): New. Also backport r199611 with fixes for the patch above and previous SLM patch: 2013-06-03 Yuri Rumyantsev <yuri.s.rumyantsev@intel.com> * config/i386/i386.c (ix86_lea_outperforms): Fix formatting. (ix86_avoid_lea_for_addr): Likewise. (exact_dependency_1): Likewise. (ix86_adjust_cost): Likewise. (swap_top_of_ready_list): Fix formatting and !reload_completed check removed. (do_reorder_for_imul): Fix typo, formatting and !reload_completed check removed. (ix86_sched_reorder): Fix typo and formatting. (fold_builtin_cpu): Move M_INTEL_SLM at the end of processor types list. * config/i386/cpuinfo.c (INTEL_SLM): New enum value. Note that [4.7] part of the patch doesn't contain some of optimizations (IMUL) due to missed dependencies. [4.8] part of this backport is complete. Change-Id: I4b5f92b025aab217046f5b393527636f3cf25669 Signed-off-by: Pavel Chupin <pavel.v.chupin@intel.com>
* [4.7, 4.8] Release basic tuning for new Silvermont architecturePavel Chupin2013-05-307-32/+896
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Support new switches: -march=slm/-mtune=slm This is backport of trunk r199444: 2013-05-30 Yuri Rumyantsev <yuri.s.rumyantsev@intel.com> Igor Zamyatin <igor.zamyatin@intel.com> Silvermont (SLM) architecture pipeline model, tuning and insn selection. * config.gcc: Add slm config options and target. * config/i386/slm.md: New. * config/i386/driver-i386.c (host_detect_local_cpu): Check * movbe. * gcc/config/i386/i386-c.c (ix86_target_macros_internal): New * case PROCESSOR_SLM. (ix86_target_macros_internal): Likewise. * gcc/config/i386/i386.c (slm_cost): New cost. (m_SLM): New macro flag. (initial_ix86_tune_features): Set m_SLM. (x86_accumulate_outgoing_args): Likewise. (x86_arch_always_fancy_math_387): Likewise. (processor_target_table): Add slm cost. (cpu_names): Add slm cpu name. (x86_option_override_internal): Set SLM ISA. (ix86_issue_rate): New case PROCESSOR_SLM. (ia32_multipass_dfa_lookahead): Likewise. (fold_builtin_cpu): Add slm. * config/i386/i386.h (TARGET_SLM): New target macro. (target_cpu_default): Add TARGET_CPU_DEFAULT_slm. (processor_type): Add PROCESSOR_SLM. * config/i386/i386.md (cpu): Add new value "slm". (slm.md): Include slm.md. * libgcc/config/i386/cpuinfo.c (INTEL_SLM): New enum value. Change-Id: I3ad6f5584e3fd5de52ac608dc699daaad24f2fe4 Signed-off-by: Pavel Chupin <pavel.v.chupin@intel.com>
* [4.6], [4.7], [4.8] Backport of the patch from trunk/4.8 for fixingAlexander Ivchenko2013-05-242-0/+11
| | | | | | | | | | | | | | this issue: https://code.google.com/p/android/issues/detail?id=54623 2013-05-24 Alexander Ivchenko <alexander.ivchenko@intel.com> PR tree-ssa/57385 * tree-ssa-sccvn.c (fully_constant_vn_reference_p): Check that index is not negative. Change-Id: I8c83a3f9992b60d3f59c2b8e123ecda7c398d07e Signed-off-by: Alexander Ivchenko <alexander.ivchenko@intel.com>
* Remove hardcoded -m32 in compiler optionsPavel Chupin2013-05-221-1/+1
| | | | | | | | | | | | | It's required for x86_64 targets where we obviously don't want to enforce -m32. It shouldn't affect standard i686 targets because i686 is enforced on configure stage and it effectively removes 64-bit support, keeping -m32 default. Patch changes 4.7 and 4.8 only. x86_64 targets are not expected to be supported in older gcc versions due to many issues and dependencies. Change-Id: Id15a677a9d3b1712eb95aba92c2c5bd98f9ed1cc Signed-off-by: Pavel Chupin <pavel.v.chupin@intel.com>
* Fix ARM/GCC-4.7,4.8 generates insufficient alignment for NEON vst/vldAndrew Hsieh2013-05-151-2/+6
| | | | | | | | | | | | | | GCC allocates memory buffer and passes it as the first hidden argument for function return large composite type (ie. > 4 bytes for all NDK toolchain). Problem is that GCC doesn't observe the aligement required by the type, and ARM EABI only requires stack to be aligned to 8-byte. Please see http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57271 for external bug tracking this issue and testcase. This CL offers temp relief before formal one Change-Id: I8004bc4208487e539ba8b0c9686c44ac86c37d83
* Add new GCC/MIPS option -mldc1-sdc1Andrew Hsieh2013-04-262-1/+5
| | | | | | | | ldc1/sdc1 needs memory to be 8-byte aligned. This option is on by default for arch supports it, but can be turned off by "-mno-ldc1-sdc1" to workaround issue, for example. Change-Id: I14291dc4280e75f6c765c933b60eb1b7cd44f974
* Merge "[4.7] x32: Backport x32 support into 4.7"Andrew Hsieh2013-04-2287-395/+1483
|\
| * [4.7] x32: Backport x32 support into 4.7Pavel Chupin2013-04-1887-395/+1483
| | | | | | | | | | | | | | | | This patch contains all gcc changes required to build x32 compiler. They are backported from 4.8/trunk. Change-Id: I923f639c1f0cee5812b0f555a39bab0bd0723865 Signed-off-by: Pavel Chupin <pavel.v.chupin@intel.com>
* | Fix GNU4.7/4.8 libstdc++ for clangAndrew Hsieh2013-04-192-2/+3
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Two fixes 1) Undefine GCC4.7/4.8 _GLIBCXX_ATOMIC_BUILTINS Clang arm/mips can't compile the following atomic intrinsics yet __atomic_fetch_add(__mem, __val, __ATOMIC_ACQ_REL); __atomic_fetch_add(__mem, __val, __ATOMIC_ACQ_REL); Use library calls instead 2) Undefine GCC4.7/4.8 _GLIBCXX_USE_FLOAT128. Clang doesn't support __float128 yet Change-Id: I0ff0eb208665beeb3e5e09917fd4c7dd378fc54e
* | Fix GCC4.4.3/4.6/4.7 to supress -export-dynamic for statically linked programsAndrew Hsieh2013-04-121-2/+3
|/ | | | | | | | | r191401 | jingyu | 2012-09-18 01:33:05 +0800 (Tue, 18 Sep 2012) | 6 lines 2012-09-17 Ben Cheng <bccheng@google.com * config/arm/linux-elf.h (LINUX_TARGET_LINK_SPEC): Suppress the dynamic linker commands for statically linked programs. Change-Id: Ib8a27fb054ddb12242d500ef8b3ceb2b0215cb3e
* Misc fixesAndrew Hsieh2013-04-041-1/+1
| | | | | | | | | | 1. Fixed darwin build of GCC 4.8/X86 by removing trailing ',' from the last item of enum. Not discovered before because GCC 4.8 now builds with -Wall -Werror, and gcc-4.2.1 in darwin is picky about that 2. Fixed -fuse-ld=mcld Change-Id: I7b65edfb76841f14c3290acb5a98f556d00e1139
* Use READELF_FOR_TARGET in GCC 4.6/4.7Andrew Hsieh2013-03-262-3/+7
| | | | | | | | | | | | | | | | | | | | | | In darwin where readelf isn't available, gcc/configure fails to set HAVE_LD_PERSONALITY_RELAXATION for MIPS, causes TARGET_WRITABLE_EH_FRAME=1 and flag_dwarf2_cfi_asm=0 (ie. disable CFI directive), and fails ld.mcld It's curious that the successful detection of the /tmp/ndk-$USER/gcc/binutils-2.21/readelf requires binutils to exist in the GCC's source tree, according to gcc/configure. Other tools (objdump, nm, ld, as, etc) can be successfully located because of OBJDUMP_FOR_TARGET, NM_FOR_TARGET, etc. Fix to look for READELF_FOR_TARGET in gcc/configure See https://android-review.googlesource.com/#/c/54760 https://android-review.googlesource.com/#/c/54770 https://android-review.googlesource.com/#/c/54710 Change-Id: I8b057a70d43389d3d9d02b5c54b8365ebe6487d1
* Merge "More fix to handle NULL return value of setlocale() in libstdc++/GCC4.7"Andrew Hsieh2013-03-092-12/+24
|\
| * More fix to handle NULL return value of setlocale() in libstdc++/GCC4.7Andrew Hsieh2013-03-082-12/+24
| | | | | | | | | | | | | | | | | | | | Bionic setlocale() always returns NULL. Check return value before passing it to other libraries which don't expect NULL in argument. See 20ffc596a2771a96ac9b4a4838edc7779a48f0af https://code.google.com/p/android/issues/detail?id=46718 Change-Id: Ie2bf30d5f54c77401519ecb27cb97fb905639bba
* | Fixed GCC 4.7 ICE in redirect_eh_edge_1, at tree-eh.c:2214Andrew Hsieh2013-03-083-0/+34
|/ | | | | | | | | | | | | | http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=191571 2012-09-17 Jakub Jelinek <jakub@redhat.com> PR tree-optimization/54563 * tree-ssa-math-opts.c (execute_cse_sincos): Call gimple_purge_dead_eh_edges if last call has been changed. * g++.dg/torture/pr54563.C: New test. See https://code.google.com/p/android/issues/detail?id=52909 Change-Id: Ib0f8d0dd2faa5a8f0bdeb160da2e2cd0c24f5757
* Merge "Fix GCC 4.4.3/4.6/4.7 openmp support to use libgomp/config/linux"Andrew Hsieh2013-03-084-3/+28
|\
| * Fix GCC 4.4.3/4.6/4.7 openmp support to use libgomp/config/linuxAndrew Hsieh2013-03-074-3/+28
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Android does support SYS_gettid and SYS_futex which allows libgomp to use libgomp/config/linux/* instead of libgomp/config/posix and enables more features such as thread affinity. For GCC4.4.3/ARM, the missing linux/futex.h is copied from GCC 4.6 Also guard the new sc_nprocessors_actu with __ANDROID__, and fix return value in proc.c Change-Id: Ib22496075ac5e41c5f44f5160d1b34a5609c168b
* | GNU libstdc++ --enable-linux-futexAndrew Hsieh2013-03-072-0/+17
|/ | | | Change-Id: Ib97053768252b05decd513825ae45757b220f7d4
* Merge "OpenMP: Better CPU count detection for Linux"Andrew Hsieh2013-03-061-3/+186
|\
| * OpenMP: Better CPU count detection for LinuxGeunsik Lim2013-03-061-3/+186
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | .PURPOSE This patch is to improve the detection of total CPU count for OpenMP/Linux. This patch helps that OpenMP recognizes the actual CPUs correctly for the task parallelism. Recent mobile devices dynamically disable or enable CPU cores on demand to optimize battery life. This impacts the output of '/proc/cpuinfo' (which only reports active ones). This patch provides a way to compute the total number of cores by parsing the content of '/sys/devices/system/cpu/' instead. Actual CPUs: total installed CPUs Online's CPUs: CPUs of the turn-on status Offline's CPUs: CPUs of the turn-off status Therefore, We need to enhance the existing approach using procfs like '/proc/cpuinfo/' or '/proc/stat'. This patch is suitable for Linux kernel. This patch is based on GCC 4.7.2. .RELATED LINKS: https://android-review.googlesource.com/#/c/52821/ https://android-review.googlesource.com/#/c/51950/ https://android-review.googlesource.com/#/c/51370/ https://android.googlesource.com/toolchain/gcc/ .UNIT TEST int main (int argc, char *argv[]) { int nthreads, tid; printf("SC_NPROCESSORS_ONLN: %d\n", sysconf (_SC_NPROCESSORS_ONLN)); /* Fork a team of threads giving them their own copies of variables */ { /* Obtain thread number */ tid = omp_get_thread_num(); printf("Hello World from thread = %d\n", tid); /* Only master thread does this */ if (tid == 0) { nthreads = omp_get_num_threads(); printf("Number of threads = %d\n", nthreads); } } /* All threads join master thread and disband */ } .LIMITATION OF SCOPE These functions have to be included in Bionic C library as the resource of sysconf(). Examples: _SC_NPROCESSORS_ONLN /proc/stat (= /sys/devices/system/cpu/online) _SC_NPROCESSORS_CONF /proc/cpuinfo (= /sys/devices/system/cpu/online) _SC_NPROCESSORS_ACTU /sys/devices/system/cpu/present (* Future work) .COST EVALUATION OF SYSCALLS (On 32bit Quad core) ------------------------------------------ time seconds usecs/call calls syscall ------------------------------------------ -nan 0.000000 0 3 read -nan 0.000000 0 9 write -nan 0.000000 0 4 open -nan 0.000000 0 4 close -nan 0.000000 0 1 execve -nan 0.000000 0 1 access -nan 0.000000 0 1 brk -nan 0.000000 0 1 munmap -nan 0.000000 0 2 mprotect -nan 0.000000 0 7 mmap2 -nan 0.000000 0 3 fstat64 -nan 0.000000 0 1 set_thread_area ------------------------------------------ 100 0.000000 0 37 total Change-Id: I2c6e41616f309eee89ee1145e92201ffe73463f4 Signed-off-by: Geunsik Lim <leemgs@gmail.com> Signed-off-by: Geunsik Lim <geunsik.lim@samsung.com> Acked-by: Changwoo Min <multics69@gmail.com> CC: David Turner <digit@android.com> CC: Keith Obenschain <obenschaink@gmail.com> CC: Pavel Chupin <pavel.v.chupin@intel.com> CC: Andrew Hsieh <andrewhsieh@google.com>
* | Merge "Use default visibility for the __cxa functions in eh_arm.cc."Andrew Hsieh2013-03-061-0/+12
|\ \
| * | Use default visibility for the __cxa functions in eh_arm.cc.Ben Cheng2013-03-051-0/+12
| | | | | | | | | | | | Change-Id: I45ecaa7c20aece451664c631c01a3fefe054293a
* | | Fixed GCC 4.6/4.7 ICE gen_thumb_movhi_clobber at config/arm/arm.md:5832Andrew Hsieh2013-03-052-1/+2
| |/ |/| | | | | | | | | | | See issue https://code.google.com/p/android/issues/detail?id=52732, and fix http://gcc.gnu.org/ml/gcc-patches/2010-12/msg00732.html Change-Id: If4759a637e5fb6f9c351c11b79d7f9f682ca2ab2
* | Fix MIPS GCC to emit -m elf32ltsmip for linkerAndrew Hsieh2013-02-281-0/+1
| | | | | | | | | | | | | | | | | | | | ARM/X86 GCC in NDK explicitly emit -m armelf_linux_eabi and -m elf_i386, respectively, for linker. Fix MIPS to do the same, to help mclinker which is built vanilla and relies on either filename (eg. eg. arm-linux-androideabi-ld.mcd implies arm) or -mtriple/-march/-m to figure out the target Change-Id: I236786d0e90a3152225de1eea5e73e3de4ac3f84
* | Add mclinker support: -fuse-ld=mcldAndrew Hsieh2013-02-213-0/+21
| | | | | | | | | | | | http://code.google.com/p/mclinker/ Change-Id: Ic9d88a86298eddaccf3c7da8405efb2b2f0d3e70
* | Use memalign instead of posix_memalign in GCC x86 mm_malloc.hAndrew Hsieh2013-02-201-0/+6
|/ | | | | | | | | | | | | | posix_memalign doesn't exist in NDK. Code inludes ?mmintrin.h which in turn includes mm_malloc.h may fail to link For AOSP platform build which uses the same compiler, add -DHAVE_POSIX_MEMALIGN to restore the original behavior. Other than non-zero return value which _mm_alloc already ignores, both paths are functioanlly identical (under the hood dlmalloc.c in 32-bit ensure alignment is at least 16-byte) Change-Id: I55e9bb8b80e1b55baa9744920df10fcf83218300
* Merge "Remove march/mtune/mfpmath hardcoded values"Andrew Hsieh2013-01-311-2/+1
|\
| * Remove march/mtune/mfpmath hardcoded valuesPavel Chupin2013-01-311-2/+1
| | | | | | | | | | | | | | | | They should be passed in configure to work correctly. See https://android-review.googlesource.com/50815 Change-Id: I2aeb0375132ba985a51c397ee386af1ff4a47c32 Signed-off-by: Pavel Chupin <pavel.v.chupin@intel.com>
* | Fix GCC 4.7 ICE in build_data_member_initialization, at cp/semantics.c:5790Andrew Hsieh2013-01-253-2/+26
|/ | | | | | | | 2013-01-03 Jason Merrill <jason@redhat.com> PR c++/55856 * semantics.c (build_data_member_initialization): Handle DECL_EXPR. Change-Id: Ib34c16c4bc714875a49831ef8fb1691b392640cd
* Fix x86 GCC4.7 segfault in i386.c distance_non_agu_define_in_bb()Andrew Hsieh2013-01-225-33/+43
| | | | | | | | | | | | | | | | | | | | | | | | line #16514 "if (prev == BB_HEAD(bb))" causes segfault due to bb==NULL when compiling external/llvm/lib/Transforms/Scalar/SROA.cpp BUG: 8047767 ------------------------------------------------------------------ r186451 | vries | 2012-04-14 05:17:41 -0700 (Sat, 14 Apr 2012) | 13 lines 2012-04-14 Tom de Vries <tom@codesourcery.com> * cfgcleanup.c (try_optimize_cfg): Replace call to delete_insn_chain by call to delete_insn. Remove code to reorder BASIC_BLOCK note and DELETED_LABEL note, and move it to ... * cfgrtl.c (delete_insn): ... here. Change return type to void. (delete_insn_and_edges): Likewise. (delete_insn_chain): Handle new return type of delete_insn. Delete chain backwards rather than forwards. * rtl.h (delete_insn, delete_insn_and_edges): Change return type to void. * cfglayout.c (fixup_reorder_chain): Delete unused label. Change-Id: Ibc069a4267d063586003d49a805fb794f4041465
* Disable libstdc versioningAndrew Hsieh2013-01-112-2/+2
| | | | | | | Used to be first half of NDK local patch at $NDK/build/tools/ toolchain-patches/gcc/0006-Disable-libstdc-versioning.patch Change-Id: I8ac0f14d20f32a4f946ffd6fd5756a2ab556e99b
* Revert "Enable _GLIBCXX_USE_CLOCK_MONOTONIC in libstdc++"Andrew Hsieh2013-01-032-2/+2
| | | | | | | | This reverts commit bd342f0491128ad15e6ea4ff241e243b753137c9 In favor of later/better fix in 75d889ea23f0ff1ecaa5cb79b435aca25460a4c9 Change-Id: Iae68d49b39b528d790c7404e657f3daa32a19864
* Enable _GLIBCXX_USE_CLOCK_MONOTONIC in libstdc++Andrew Hsieh2012-12-292-2/+2
| | | | | | | | | This is needed to provide "steady_clock" (or monotonic_clock in GCC 4.6/4.4.3) in <chrono> See http://code.google.com/p/android/issues/detail?id=39680 Change-Id: I0c31704971dc15751ce936f563de7c42a4e645a6
* Fix OpenMPAndrew Hsieh2012-12-276-11/+15
| | | | | | | | | | | | | | Based on 1271761f530c0050154e8d526b95f952df551751, 92c478dba755a1a2f6f00ff390666acbffd41982, and 51df2e98d22e2c6f5d2a16860bc8fc3644179c1d. With the following modifications: 1. Translate -pthread to -lc instead of -lpthread Android doesn't have (has pthread* in libc.so instead) 2. Because of 1., we can restore to the original order of LINUX_OR_ANDROID_LD Change-Id: I505250c32b9908cb17bb269dc26e73c91669c07f
* Merge "Support OpenMP+Bionic on ICS/GCC-4.7.2 for task parallelism"Andrew Hsieh2012-12-265-4/+63
|\
| * Support OpenMP+Bionic on ICS/GCC-4.7.2 for task parallelismGeunsik Lim2012-12-255-4/+63
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The patch supports OpenMP library to parallelize the existing C/C++ code (e.g: for task parallelism) rapidly/easily without any modification of android platform on multicore embedded devices. The original patch is made by Keith Obenschain (using GCC 4.4.3). . Archive - https://android-review.googlesource.com/#/c/34491/ I just moved GCC version from 4.4.3 to 4.7.2 for ktoolchain version 2.5. The patch is good to me when I evaluate a lot of C/C++ source code with OpenMP's #parama. we move the version of OpenMP from version 2.5 to version 3.1. Please, refer to the "kandroid toolchain" menu in www.kandroid.org to compile source code including bionic and openmp library on real Android devices. http://kandroid.org/board/board.php?board=toolchain&command=body&no=16 I tested Openmp library with the latest GCC 4.7.2 on Android/ARM devices as following. Please refer to the below example. geunsik@rhel6$> ./arm-linux-androideabi-gcc openmptest.c -L /usr/local/ktoolchain-cortexa9-ver2.5-20120515-bionic/arm-linux-androideabi/lib -lgomp -o openmptest [ENTER] geunsik@rhel6$> geunsik@rhel6$> geunsik@rhel6$> file ./openmptest ./openmptest: ELF 32-bit LSB executable, ARM, version 1 (SYSV), dynamically linked (uses shared libs), not stripped Change-Id: I20cb3c1ac5c000445c043310158179723bd69fe2 Signed-off-by: Geunsik Lim <leemgs@gmail.com> Acked-by: Geunsik Lim <geunsik.lim@gmail.com>
* | Enable _GLIBCXX_HAS_GTHREADS in GCC 4.7Andrew Hsieh2012-12-251-1/+1
|/ | | | | | | | | | | Fix a test in libstdc++/confiugre to include $glibcxx_thread_h instead of gthr.h otherwise it will fail to detect gthread lib because the new gthr.h only include gthr-default.h which doesn't exist in source directory See issue http://code.google.com/p/android/issues/detail?id=41770 Change-Id: I736b7ba2b472a26f9e43097bb88fb68325114a7d