From patchwork Thu Nov 17 13:06:20 2016 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Rainer Orth X-Patchwork-Id: 696107 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Received: from sourceware.org (server1.sourceware.org [209.132.180.131]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by ozlabs.org (Postfix) with ESMTPS id 3tKLxY2rK0z9t0v for ; Fri, 18 Nov 2016 00:06:57 +1100 (AEDT) Authentication-Results: ozlabs.org; dkim=pass (1024-bit key; unprotected) header.d=gcc.gnu.org header.i=@gcc.gnu.org header.b="ygipEeFE"; dkim-atps=neutral DomainKey-Signature: a=rsa-sha1; c=nofws; d=gcc.gnu.org; h=list-id :list-unsubscribe:list-archive:list-post:list-help:sender:from :to:cc:subject:date:message-id:mime-version:content-type; q=dns; s=default; b=Qin5KDxsdTaTxGuEgMAl0eYgyARRTtpvsxwkMYLwJrDvHwvpMs DmP4CxQe35z4xo5e8rU38HPXfZ/7dcsqmP7uu4Nzr0VHbbLgY0WjibSieZWnzNuy 8VQIJU5L+mNx3XceuyxlaMOG1md3wzEe46tqUKci7MgKzqJi5PhMgvREk= DKIM-Signature: v=1; a=rsa-sha1; c=relaxed; d=gcc.gnu.org; h=list-id :list-unsubscribe:list-archive:list-post:list-help:sender:from :to:cc:subject:date:message-id:mime-version:content-type; s= default; bh=4LZFufzt09LJmFy1vvDlWvmHq68=; b=ygipEeFEdlQThvt1ZiNu 7RPwYpXGFi0ZugbVe4vG3Fg3m9hlpgfHEa8toEexkx//BR3Et7BLWtwKGg1o6ZXR HxRrY5AASqykoxDD5FVDRCk1ounW38W1b6sufwWkBe9smySTvjPbd79ioRWEFaa0 /gZ0TEQ8m2tahpIEyyrjDlw= Received: (qmail 42648 invoked by alias); 17 Nov 2016 13:06:47 -0000 Mailing-List: contact gcc-patches-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Unsubscribe: List-Archive: List-Post: List-Help: Sender: gcc-patches-owner@gcc.gnu.org Delivered-To: mailing list gcc-patches@gcc.gnu.org Received: (qmail 42580 invoked by uid 89); 17 Nov 2016 13:06:46 -0000 Authentication-Results: sourceware.org; auth=none X-Virus-Found: No X-Spam-SWARE-Status: No, score=-0.5 required=5.0 tests=AWL, BAYES_50, KAM_ASCII_DIVIDERS, KAM_LAZY_DOMAIN_SECURITY, RP_MATCHES_RCVD autolearn=ham version=3.3.2 spammy=brands, forums, save_CFLAGS, save_cflags X-HELO: smtp.CeBiTec.Uni-Bielefeld.DE Received: from smtp.CeBiTec.Uni-Bielefeld.DE (HELO smtp.CeBiTec.Uni-Bielefeld.DE) (129.70.160.84) by sourceware.org (qpsmtpd/0.93/v0.84-503-g423c35a) with ESMTP; Thu, 17 Nov 2016 13:06:36 +0000 Received: from localhost (localhost.CeBiTec.Uni-Bielefeld.DE [127.0.0.1]) by smtp.CeBiTec.Uni-Bielefeld.DE (Postfix) with ESMTP id C01EEEC8; Thu, 17 Nov 2016 14:06:32 +0100 (CET) Received: from smtp.CeBiTec.Uni-Bielefeld.DE ([127.0.0.1]) by localhost (malfoy.CeBiTec.Uni-Bielefeld.DE [127.0.0.1]) (amavisd-new, port 10024) with LMTP id 1kZSTSXaGIMD; Thu, 17 Nov 2016 14:06:20 +0100 (CET) Received: from lokon.CeBiTec.Uni-Bielefeld.DE (lokon.CeBiTec.Uni-Bielefeld.DE [129.70.161.152]) (using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits)) (No client certificate requested) by smtp.CeBiTec.Uni-Bielefeld.DE (Postfix) with ESMTPS id DCC11EC6; Thu, 17 Nov 2016 14:06:20 +0100 (CET) Received: (from ro@localhost) by lokon.CeBiTec.Uni-Bielefeld.DE (8.15.2+Sun/8.15.2/Submit) id uAHD6KGZ005240; Thu, 17 Nov 2016 14:06:20 +0100 (MET) From: Rainer Orth To: gcc-patches@gcc.gnu.org Cc: Ilya Verbin Subject: Import libcilkrts Build 4467 (PR target/68945) Date: Thu, 17 Nov 2016 14:06:20 +0100 Message-ID: User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/25.1 (usg-unix-v) MIME-Version: 1.0 X-IsSubscribed: yes I happened to notice that my libcilkrts SPARC port has been applied upstream. So to reach closure on this issue for the GCC 7 release, I'd like to import upstream into mainline which seems to be covered by the free-for-all clause in https://gcc.gnu.org/svnwrite.html#policies, even though https://gcc.gnu.org/codingconventions.html#upstream lists nothing specific and we have no listed maintainer. A few issues are worth mention: * Upstream still has a typo in the git URL in many files. I've corrected that during the import to avoid a massive diff: -# https://bitbucket.org/intelcilkruntime/itnel-cilk-runtime.git are +# https://bitbucket.org/intelcilkruntime/intel-cilk-runtime.git are * libcilkrts.spec.in is missing upstream. I've no idea if this is intentional. * A few of my changes have been lost and I can't tell if this is by accident: ** Lost whitespace: I've done nothing about those, just wanted to point them out. The following patch has passed x86_64-pc-linux-gnu bootstrap without regressions; i386-pc-solaris2.12 and sparc-sun-solaris2.12 bootstraps are currently running. Ok for mainline if they pass? Thanks. Rainer # HG changeset patch # Parent f20811cd2fde8357b5c51307decee8c877c1e16a Import libcilkrts Build 4467 (PR target/68945) diff --git a/libcilkrts/Makefile.am b/libcilkrts/Makefile.am --- a/libcilkrts/Makefile.am +++ b/libcilkrts/Makefile.am @@ -54,7 +54,7 @@ GENERAL_FLAGS = -I$(top_srcdir)/include # Enable Intel Cilk Plus extension GENERAL_FLAGS += -fcilkplus -# Always generate unwind tables +#Always generate unwind tables GENERAL_FLAGS += -funwind-tables AM_CFLAGS = $(XCFLAGS) $(GENERAL_FLAGS) -std=c99 diff --git a/libcilkrts/README b/libcilkrts/README --- a/libcilkrts/README +++ b/libcilkrts/README @@ -1,14 +1,16 @@ -Intel(R) Cilk(TM) Plus runtime library +Intel(R) Cilk(TM) Plus Runtime Library Index: -1. BUILDING -2. USING -3. DOXYGEN DOCUMENTATION -4. QUESTIONS OR BUGS -5. CONTRIBUTIONS +1. BUILDING WITH AUTOMAKE +2. BUILDING WITH CMAKE +3. INSTALLING TO VXWORKS +4. USING +5. DOXYGEN DOCUMENTATION +6. QUESTIONS OR BUGS +7. CONTRIBUTIONS # -# 1. BUILDING: +# 1. BUILDING WITH AUTOMAKE: # To distribute applications that use the Intel Cilk Plus language @@ -40,22 +42,87 @@ configure script: % ./configure --prefix=/your/path/to/lib -It is also possible to use CMake if the above method does not apply -well in your environment. Instruction is available in CMakeLists.txt. +# +# 2. BUILDING WITH CMAKE: +# + +To distribute applications that use the Intel Cilk Plus language +extensions to non-development systems, you need to build the Intel +Cilk Plus runtime library and distribute it with your application. +This instruction describes the build process using CMake*, which +supports Linux*, Windows*, and OS X*. It is fine to use this process +to build a Linux library, but it is highly recommended to use the +more mature build process described above when building on Linux. + +You need the CMake tool and a C/C++ compiler that supports the Intel +Cilk Plus language extensions, and the requirements for each operating +systems are: + +Common: + CMake 3.0.0 or later + Make tools such as make (Linux, OS X) or nmake (Windows) +Linux: + GCC* 4.9.2 or later, or Intel(R) C++ Compiler v12.1 or later +Windows: + Intel C++ Compiler v12.1 or later + Visual Studio* 2010 or later +OS X: + Cilk-enabled branch of Clang*/LLVM* (http://cilkplus.github.io), + or Intel C++ Compiler v12.1 or later + +The common steps to build the libraries are 1) invoke cmake with +appropriate options, 2) invoke a make tool available on the system. +The following examples show build processes on OS X and Windows. + +OS X: + % mkdir ./build && cd ./build + % cmake -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ \ + -DCMAKE_INSTALL_PREFIX=./install .. + % make && make install + +Windows: + % mkdir .\build && cd .\build + % cmake -G "NMake Makefiles" -DCMAKE_C_COMPILER=icl \ + -DCMAKE_CXX_COMPILER=icl -DCMAKE_INSTALL_PREFIX=.\install .. + % nmake && nmake install # -# 2. USING: +# 3. INSTALLING TO VXWORKS OS +# + +For Windows host, run VxWorks_Install.bat. +For Linux host, run VxWorks_Install.sh. + +You may need to give environment variable WIND_BASE to indicate VxWorks +installation path. + +Create a VSB project, and you will see a layer named "CILKPLUS_KERNEL". +Enable it and build the project. + +Create a VIP project with ICC and add component INCLUDE_CILKPLUS, then you +will get support of Intel Cilk Plus features in VxWorks. + +# +# 4. USING: # The Intel(R) C++ Compiler will automatically try to bring in the Intel Cilk Plus runtime in any program that uses the relevant -features. GCC requires explicit linking of both the library and -its dependencies (libpthread, libdl). For example: +features. GCC and Clang requires an explicit compiler option, +-fcilkplus, to enable Intel Cilk Plus language extensions. +For example, + +% gcc -fcilkplus -o foo.exe foo.c +% clang -fcilkplus -o foo.exe foo.c + +Older GCC versions (e.g., 4.8 cilkplus branch) requires explicit linking +of both the library and its dependencies (libpthread, libdl). +For example: % gcc foo.c -lcilkrts -lpthread -ldl # -# 3. DOXYGEN DOCUMENTATION: +# 5. DOXYGEN DOCUMENTATION: # The library source has Doxygen markup. Generate HTML documentation @@ -64,7 +131,7 @@ based on the markup by changing director % doxygen doxygen.cfg # -# 4. QUESTIONS OR BUGS: +# 6. QUESTIONS OR BUGS: # Issues with the Intel Cilk Plus runtime can be addressed in the Intel @@ -72,7 +139,7 @@ Cilk Plus forums: http://software.intel.com/en-us/forums/intel-cilk-plus/ # -# 5. CONTRIBUTIONS: +# 7. CONTRIBUTIONS: # The Intel Cilk Plus runtime library is dual licensed. The upstream copy @@ -85,6 +152,13 @@ contributed to the upstream version via Thanks to Tobias Burnus for showing us the magic to make gcc and g++ automatically include the Cilk Plus runtime. +Thanks to Eric Olson for sharing his patch for Raspberry Pi* with us. + +Thanks to Rainer Orth for submitting patches for exception handling and +enabling Cilk Plus on the SPARC* architecture. + ------------------------ Intel and Cilk are trademarks of Intel Corporation in the U.S. and/or other countries. + +*Other names and brands may be claimed as the property of others. diff --git a/libcilkrts/configure.ac b/libcilkrts/configure.ac --- a/libcilkrts/configure.ac +++ b/libcilkrts/configure.ac @@ -69,6 +69,10 @@ AC_PROG_CXX AC_CONFIG_FILES([Makefile libcilkrts.spec]) AC_FUNC_ALLOCA +# Check for dl functions +AC_CHECK_LIB(dl, dladdr, + [AC_DEFINE(HAVE_DLADDR, 1, [Define if you have dladdr()])]) + # Check whether the target supports protected visibility. save_CFLAGS="$CFLAGS" CFLAGS="$CFLAGS -Werror" @@ -141,11 +145,7 @@ esac # contains information on what's needed case "${target}" in - x86_64-*-*) - config_dir="x86" - ;; - - i?86-*-*) + i?86-*-* | x86_64-*-*) config_dir="x86" ;; @@ -153,6 +153,10 @@ case "${target}" in config_dir="arm" ;; + sparc*-*-*) + config_dir="sparc" + ;; + *) config_dir="generic" ;; diff --git a/libcilkrts/configure.tgt b/libcilkrts/configure.tgt --- a/libcilkrts/configure.tgt +++ b/libcilkrts/configure.tgt @@ -44,12 +44,12 @@ # Disable Cilk Runtime library for unsupported architectures. case "${target}" in - x86_64-*-*) - ;; - i?86-*-*) + i?86-*-* | x86_64-*-*) ;; arm-*-*) ;; + sparc*-*-*) + ;; *-*-*) UNSUPPORTED=1 ;; diff --git a/libcilkrts/include/cilk/cilk_undocumented.h b/libcilkrts/include/cilk/cilk_undocumented.h --- a/libcilkrts/include/cilk/cilk_undocumented.h +++ b/libcilkrts/include/cilk/cilk_undocumented.h @@ -106,9 +106,6 @@ size_t __cilkrts_get_stack_size(void); CILK_EXPORT __CILKRTS_NOTHROW void __cilkrts_dump_stats(void); -CILK_EXPORT __CILKRTS_NOTHROW -int __cilkrts_irml_version(void); - struct __cilk_tbb_unwatch_thunk; struct __cilk_tbb_stack_op_thunk; diff --git a/libcilkrts/include/internal/cilk_version.h b/libcilkrts/include/internal/cilk_version.h --- a/libcilkrts/include/internal/cilk_version.h +++ b/libcilkrts/include/internal/cilk_version.h @@ -49,10 +49,10 @@ #define VERSION_MAJOR 2 #define VERSION_MINOR 0 -#define VERSION_BUILD 4420 +#define VERSION_BUILD 4467 #define VERSION_REV 0 -#define VERSION_STRING "2,0,4420,0" -#define VERSION_HASH "3b2d6aa9059c" +#define VERSION_STRING "2,0,4467,0" +#define VERSION_HASH "b7e54d87bd17" #define VERSION_BRANCH "eng" -#define TBB_REV_NUMBER "14788" -#define VERSION_YEAR "2015" +#define TBB_REV_NUMBER "" +#define VERSION_YEAR "2016" diff --git a/libcilkrts/include/internal/rev.mk b/libcilkrts/include/internal/rev.mk --- a/libcilkrts/include/internal/rev.mk +++ b/libcilkrts/include/internal/rev.mk @@ -49,4 +49,4 @@ # # It was automatically generated by cilkrts/include/internal/Makefile -CILK_REVISION = 4420 +CILK_REVISION = 4467 diff --git a/libcilkrts/runtime/cilk-abi.c b/libcilkrts/runtime/cilk-abi.c --- a/libcilkrts/runtime/cilk-abi.c +++ b/libcilkrts/runtime/cilk-abi.c @@ -709,7 +709,6 @@ CILK_API_TBB_RETCODE return 0; /* Success! */ } - // This function must be called only within a continuation, within the stack // frame of the continuation itself. CILK_API_INT __cilkrts_synched(void) diff --git a/libcilkrts/runtime/cilk_fiber-unix.cpp b/libcilkrts/runtime/cilk_fiber-unix.cpp --- a/libcilkrts/runtime/cilk_fiber-unix.cpp +++ b/libcilkrts/runtime/cilk_fiber-unix.cpp @@ -220,7 +220,7 @@ NORETURN cilk_fiber_sysdep::run() // enough extra space from the top of the stack we are // switching to for any temporaries required for this run() // function. - JMPBUF_SP(m_resume_jmpbuf) = m_stack_base - frame_size; + JMPBUF_SP(m_resume_jmpbuf) = CILK_ADJUST_SP(m_stack_base - frame_size); // GCC doesn't allow us to call __builtin_longjmp in the same function // that calls __builtin_setjmp, so it's been moved into it's own diff --git a/libcilkrts/runtime/cilk_fiber.h b/libcilkrts/runtime/cilk_fiber.h --- a/libcilkrts/runtime/cilk_fiber.h +++ b/libcilkrts/runtime/cilk_fiber.h @@ -73,9 +73,12 @@ * * A value of 0 means no debugging. * Higher values generate more debugging output. + * */ + +#ifndef FIBER_DEBUG #define FIBER_DEBUG 0 - +#endif /** * @brief Flag for validating reference counts. * diff --git a/libcilkrts/runtime/config/arm/os-fence.h b/libcilkrts/runtime/config/arm/os-fence.h --- a/libcilkrts/runtime/config/arm/os-fence.h +++ b/libcilkrts/runtime/config/arm/os-fence.h @@ -47,6 +47,14 @@ * for your assistance in helping us improve Cilk Plus. **************************************************************************/ +// __atomic_* intrinsics are available since GCC 4.7. +#define HAVE_ATOMIC_INTRINSICS defined(__GNUC__) && \ + (__GNUC__ * 10 + __GNUC_MINOR__ >= 47) + +// GCC before 4.4 does not implement __sync_synchronize properly +#define HAVE_SYNC_INTRINSICS defined(__GNUC__) && \ + (__GNUC__ * 10 + __GNUC_MINOR__ >= 44) + /* * void __cilkrts_fence(void) * @@ -60,5 +68,12 @@ * the CPUID instruction). */ -// COMMON_SYSDEP void __cilkrts_fence(void); ///< MFENCE instruction -# define __cilkrts_fence() __asm__ __volatile__ ("mcr p15,0,%[t],c7,c10,4\n" :: [t] "r" (0) : "memory"); +#if HAVE_ATOMIC_INTRINSICS +# define __cilkrts_fence() __atomic_thread_fence(__ATOMIC_SEQ_CST) +#elif HAVE_SYNC_INTRINSICS +# define __cilkrts_fence() __sync_synchronize() +#else +# define __cilkrts_fence() +// Leaving this code just in case. +//# define __cilkrts_fence() __asm__ __volatile__ ("mcr p15,0,%[t],c7,c10,4\n" :: [t] "r" (0) : "memory"); +#endif diff --git a/libcilkrts/runtime/except-gcc.cpp b/libcilkrts/runtime/except-gcc.cpp --- a/libcilkrts/runtime/except-gcc.cpp +++ b/libcilkrts/runtime/except-gcc.cpp @@ -60,7 +60,9 @@ #include #include +#ifndef DEBUG_EXCEPTIONS #define DEBUG_EXCEPTIONS 0 +#endif struct pending_exception_info { diff --git a/libcilkrts/runtime/global_state.cpp b/libcilkrts/runtime/global_state.cpp --- a/libcilkrts/runtime/global_state.cpp +++ b/libcilkrts/runtime/global_state.cpp @@ -579,7 +579,6 @@ global_state_t* cilkg_init_global_state( g->stack_size = cilkos_validate_stack_size(g->stack_size); g->failure_to_allocate_stack = 0; - return g; } diff --git a/libcilkrts/runtime/global_state.h b/libcilkrts/runtime/global_state.h --- a/libcilkrts/runtime/global_state.h +++ b/libcilkrts/runtime/global_state.h @@ -215,7 +215,6 @@ struct global_state_t { /* COMMON_PORTAB /// Global fiber pool cilk_fiber_pool fiber_pool; - /** * @brief Track whether the runtime has failed to allocate a * stack. diff --git a/libcilkrts/runtime/jmpbuf.h b/libcilkrts/runtime/jmpbuf.h --- a/libcilkrts/runtime/jmpbuf.h +++ b/libcilkrts/runtime/jmpbuf.h @@ -107,6 +107,18 @@ */ #define SP(SF) JMPBUF_SP((SF)->ctx) +/** + * @brief Some architecture-dependent stack adjustment. + */ +#if defined(__sparcv9) + // Subtract sparc v9 stack bias so the actual stack starts at the + // allocated area. +# define CILK_ADJUST_SP(SP) ((SP) - 2047) +# define CILK_UNADJUST_SP(SP) ((SP) + 2047) +#else +# define CILK_ADJUST_SP(SP) (SP) +# define CILK_UNADJUST_SP(SP) (SP) +#endif __CILKRTS_BEGIN_EXTERN_C @@ -120,7 +132,7 @@ */ inline char *__cilkrts_get_sp(__cilkrts_stack_frame *sf) { - return (char *)SP(sf); + return (char *)CILK_UNADJUST_SP(SP(sf)); } /** diff --git a/libcilkrts/runtime/linux-symbols.ver b/libcilkrts/runtime/linux-symbols.ver --- a/libcilkrts/runtime/linux-symbols.ver +++ b/libcilkrts/runtime/linux-symbols.ver @@ -71,7 +71,6 @@ CILKABI0 __cilkrts_hyperobject_dealloc; __cilkrts_hyperobject_noop_destroy; __cilkrts_init; - __cilkrts_irml_version; __cilkrts_leave_frame; __cilkrts_metacall; __cilkrts_rethrow; diff --git a/libcilkrts/runtime/local_state.h b/libcilkrts/runtime/local_state.h --- a/libcilkrts/runtime/local_state.h +++ b/libcilkrts/runtime/local_state.h @@ -360,6 +360,13 @@ struct local_state /* COMMON_PORTABLE * unsigned int steal_failure_count; /** + * Record indicating that the worker stole work at least once. + * + * [local read/write] + */ + int has_stolen; + + /** * 1 if work was stolen from another worker. When true, this will flag * setup_for_execution_pedigree to increment the pedigree when we resume * execution to match the increment that would have been done on a return diff --git a/libcilkrts/runtime/mac-symbols.txt b/libcilkrts/runtime/mac-symbols.txt --- a/libcilkrts/runtime/mac-symbols.txt +++ b/libcilkrts/runtime/mac-symbols.txt @@ -33,7 +33,6 @@ ___cilkrts_hyperobject_dealloc ___cilkrts_hyperobject_noop_destroy ___cilkrts_init - ___cilkrts_leave_frame ___cilkrts_metacall ___cilkrts_resume diff --git a/libcilkrts/runtime/os-unix.c b/libcilkrts/runtime/os-unix.c --- a/libcilkrts/runtime/os-unix.c +++ b/libcilkrts/runtime/os-unix.c @@ -432,7 +432,9 @@ COMMON_SYSDEP void __cilkrts_idle(void) #elif defined(__MIC__) _mm_delay_32(1024); #elif defined(__linux__) || \ - defined(__APPLE__) + defined(__APPLE__) || \ + defined(__CYGWIN__) + usleep(10000); #else # error "Unsupported architecture" @@ -452,6 +454,7 @@ COMMON_SYSDEP void __cilkrts_yield(void) { #if defined(__ANDROID__) || \ defined(__APPLE__) || \ + defined(__CYGWIN__) || \ defined(__FreeBSD__) || \ defined(__VXWORKS__) || \ (defined(__sun__) && defined(__svr4__)) diff --git a/libcilkrts/runtime/record-replay.cpp b/libcilkrts/runtime/record-replay.cpp --- a/libcilkrts/runtime/record-replay.cpp +++ b/libcilkrts/runtime/record-replay.cpp @@ -262,7 +262,7 @@ char * walk_pedigree_nodes(char *p, cons if (pnode->parent) { p = walk_pedigree_nodes(p, pnode->parent); - p += cilk_snprintf_s(p, PEDIGREE_BUFF_SIZE, "%s", (char *) "_"); + p += cilk_snprintf_s(p, PEDIGREE_BUFF_SIZE, "%s", (char *)"_"); } return p + cilk_snprintf_l(p, PEDIGREE_BUFF_SIZE, "%" PRIu64, pnode->rank); } diff --git a/libcilkrts/runtime/scheduler.c b/libcilkrts/runtime/scheduler.c --- a/libcilkrts/runtime/scheduler.c +++ b/libcilkrts/runtime/scheduler.c @@ -1789,20 +1789,27 @@ static full_frame* check_for_work(__cilk if (NULL == ff) { // Punish the worker for failing to steal. // No quantum for you! - if (w->l->steal_failure_count > 30000) { - // Punish more if the worker has been doing unsuccessful steals - // for a long time. After return from the idle state, it will - // be given a grace period to react quickly. + unsigned int max_fails = w->g->max_steal_failures << 1; + if (w->l->has_stolen == 0 && + w->l->steal_failure_count % max_fails == max_fails - 1) { + // Idle briefly if the worker has never stolen anything for + // the given grace period __cilkrts_idle(); - w->l->steal_failure_count -= 300; } else { __cilkrts_yield(); } w->l->steal_failure_count++; + if (w->l->steal_failure_count > (max_fails << 8)) { + // Reset the flag after certain amount of failures + // - This will reduce cpu time in top-level synched regions + // - max_fails can be controlled by user (CILK_STEAL_FAILURES) + w->l->has_stolen = 0; + } } else { // Reset steal_failure_count since there is obviously still work to // be done. w->l->steal_failure_count = 0; + w->l->has_stolen = 1; } } return ff; @@ -2912,6 +2919,7 @@ void __cilkrts_restore_stealing( w->l->stats = NULL; #endif w->l->steal_failure_count = 0; + w->l->has_stolen = 0; w->l->work_stolen = 0; diff --git a/libcilkrts/runtime/scheduler.h b/libcilkrts/runtime/scheduler.h --- a/libcilkrts/runtime/scheduler.h +++ b/libcilkrts/runtime/scheduler.h @@ -85,7 +85,10 @@ * Print debugging messages and assertions for parallel reducers. 0 is * no debugging. A higher value generates more output. */ + +#ifndef REDPAR_DEBUG #define REDPAR_DEBUG 0 +#endif /** * @brief Lock the worker mutex to allow exclusive access to the diff --git a/libcilkrts/runtime/sysdep-unix.c b/libcilkrts/runtime/sysdep-unix.c --- a/libcilkrts/runtime/sysdep-unix.c +++ b/libcilkrts/runtime/sysdep-unix.c @@ -465,7 +465,7 @@ char* sysdep_reset_jump_buffers_for_resu CILK_ASSERT(fiber); void* sp = (void*)get_sp_for_executing_sf(cilk_fiber_get_stack_base(fiber), ff, sf); - SP(sf) = sp; + SP(sf) = CILK_ADJUST_SP(sp); /* Debugging: make sure stack is accessible. */ ((volatile char *)sp)[-1]; @@ -495,7 +495,7 @@ NORETURN sysdep_longjmp_to_sf(char* new_ #endif // Set the stack pointer. - SP(sf) = new_sp; + SP(sf) = CILK_ADJUST_SP(new_sp); #ifdef RESTORE_X86_FP_STATE if (CILK_FRAME_VERSION_VALUE(sf->flags) >= 1) { @@ -568,7 +568,7 @@ static const char *get_runtime_path () { // dladdr is a glibc extension. If it's available, use it to find the path // for libcilkrts.so -#ifdef _GNU_SOURCE +#if HAVE_DLADDR Dl_info info; if (0 != dladdr(dummy_function, &info)) return info.dli_fname; @@ -689,7 +689,6 @@ static void write_version_file (global_s // ================== // System cores: 8 // Cilk workers requested: 8 - // Thread creator: Private fprintf(fp, "\nThread information\n"); fprintf(fp, "==================\n"); @@ -699,11 +698,6 @@ static void write_version_file (global_s fprintf(fp, "System cores: %d\n", (int)sysconf(_SC_NPROCESSORS_ONLN)); #endif fprintf(fp, "Cilk workers requested: %d\n", n); -#if (PARALLEL_THREAD_CREATE) - fprintf(fp, "Thread creator: Private (parallel)\n"); -#else - fprintf(fp, "Thread creator: Private\n"); -#endif if (fp != stderr && fp != stdout) fclose(fp); else fflush(fp); // flush the handle buffer if it is stdout or stderr.