Changes in / [97fed44:f57f6ea0]
- Files:
-
- 5 added
- 39 edited
-
benchmark/io/http/Makefile.am (modified) (1 diff)
-
benchmark/io/http/main.cfa (modified) (4 diffs)
-
benchmark/readyQ/churn.cfa (modified) (1 diff)
-
benchmark/readyQ/locality.cfa (modified) (2 diffs)
-
benchmark/readyQ/transfer.cfa (modified) (6 diffs)
-
libcfa/prelude/Makefile.am (modified) (4 diffs)
-
libcfa/src/bits/random.hfa (modified) (4 diffs)
-
libcfa/src/common.hfa (modified) (2 diffs)
-
libcfa/src/concurrency/clib/cfathread.cfa (modified) (2 diffs)
-
libcfa/src/concurrency/io.cfa (modified) (1 diff)
-
libcfa/src/concurrency/kernel.cfa (modified) (7 diffs)
-
libcfa/src/concurrency/kernel.hfa (modified) (7 diffs)
-
libcfa/src/concurrency/kernel/fwd.hfa (modified) (2 diffs)
-
libcfa/src/concurrency/kernel/startup.cfa (modified) (6 diffs)
-
libcfa/src/concurrency/locks.hfa (modified) (2 diffs)
-
libcfa/src/concurrency/ready_queue.cfa (modified) (26 diffs)
-
libcfa/src/concurrency/stats.cfa (modified) (3 diffs)
-
libcfa/src/concurrency/stats.hfa (modified) (1 diff)
-
libcfa/src/concurrency/thread.cfa (modified) (3 diffs)
-
libcfa/src/iostream.cfa (modified) (24 diffs)
-
libcfa/src/parseconfig.cfa (modified) (3 diffs)
-
libcfa/src/startup.cfa (modified) (3 diffs)
-
libcfa/src/stdlib.cfa (modified) (2 diffs)
-
libcfa/src/stdlib.hfa (modified) (3 diffs)
-
src/AST/Decl.cpp (modified) (2 diffs)
-
src/AST/Decl.hpp (modified) (2 diffs)
-
src/AST/Eval.hpp (modified) (1 diff)
-
src/AST/Print.cpp (modified) (1 diff)
-
src/Validate/InitializerLength.cpp (modified) (1 diff)
-
src/Validate/InitializerLength.hpp (modified) (1 diff)
-
tests/concurrent/.expect/ctor-check.txt (modified) (1 diff)
-
tests/concurrent/.expect/semaphore.txt (added)
-
tests/concurrent/.expect/spinaphore.txt (added)
-
tests/concurrent/preempt.cfa (modified) (3 diffs)
-
tests/concurrent/semaphore.cfa (added)
-
tests/concurrent/spinaphore.cfa (added)
-
tests/device/cpu.cfa (modified) (3 diffs)
-
tests/io/io-acquire-in.cfa (modified) (2 diffs)
-
tests/io/io-acquire-no-io.cfa (modified) (2 diffs)
-
tests/io/io-acquire-out.cfa (modified) (2 diffs)
-
tests/io/io-acquire.cfa (modified) (2 diffs)
-
tests/unified_locking/.expect/locks.txt (modified) (1 diff)
-
tests/unified_locking/locks.cfa (modified) (3 diffs)
-
tests/zombies/fastlock.cfa (added)
Legend:
- Unmodified
- Added
- Removed
-
benchmark/io/http/Makefile.am
r97fed44 rf57f6ea0 21 21 include $(top_srcdir)/tools/build/cfa.make 22 22 23 AM_CFLAGS = -O 3-Wall -Wextra -I$(srcdir) -lrt -pthread # -Werror23 AM_CFLAGS = -O2 -Wall -Wextra -I$(srcdir) -lrt -pthread # -Werror 24 24 AM_CFAFLAGS = -quiet -nodebug 25 25 AM_LDFLAGS = -quiet -nodebug -
benchmark/io/http/main.cfa
r97fed44 rf57f6ea0 1 #define _ GNU_SOURCE1 #define __USE_GNU 2 2 3 3 #include <errno.h> … … 6 6 #include <unistd.h> 7 7 extern "C" { 8 #include <sched.h>9 8 #include <signal.h> 10 9 #include <sys/socket.h> … … 68 67 (this.self){ "Server Cluster", options.clopts.params }; 69 68 70 cpu_set_t fullset;71 CPU_ZERO(&fullset);72 int ret = sched_getaffinity(getpid(), sizeof(fullset), &fullset);73 if( ret != 0 ) abort | "sched_getaffinity failed with" | errno | strerror( errno );74 int cnt = CPU_COUNT(&fullset);75 76 69 this.procs = alloc(options.clopts.nprocs); 77 70 for(i; options.clopts.nprocs) { 78 71 (this.procs[i]){ "Benchmark Processor", this.self }; 79 80 int c = 0;81 int n = 1 + (i % cnt);82 for(int j = 0; j < CPU_SETSIZE; j++) {83 if(CPU_ISSET(j, &fullset)) n--;84 if(n == 0) {85 c = j;86 break;87 }88 }89 cpu_set_t localset;90 CPU_ZERO(&localset);91 CPU_SET(c, &localset);92 ret = pthread_setaffinity_np(this.procs[i].kernel_thread, sizeof(localset), &localset);93 if( ret != 0 ) abort | "sched_getaffinity failed with" | ret | strerror( ret );94 72 95 73 #if !defined(__CFA_NO_STATISTICS__) … … 168 146 int waited = 0; 169 147 for() { 170 int sockfd = server_fd; 171 __CONST_SOCKADDR_ARG addr; 172 addr.__sockaddr__ = (struct sockaddr *)&address; 173 socklen_t addrlen = sizeof(address); 174 ret = bind( sockfd, addr, addrlen ); 148 ret = bind( server_fd, (struct sockaddr *)&address, sizeof(address) ); 175 149 if(ret < 0) { 176 150 if(errno == EADDRINUSE) { -
benchmark/readyQ/churn.cfa
r97fed44 rf57f6ea0 21 21 wait( sem ); 22 22 for() { 23 uint 32_t r = prng();23 uint64_t r = thread_rand(); 24 24 bench_sem * next = __atomic_exchange_n(&spots[r % spot_cnt], &sem, __ATOMIC_SEQ_CST); 25 25 if(next) post( *next ); -
benchmark/readyQ/locality.cfa
r97fed44 rf57f6ea0 128 128 __attribute__((noinline)) void work(MyData & data, size_t cnt_, uint64_t & state) { 129 129 for (cnt_) { 130 access(data, xorshift_13_7_17(state));130 access(data, __xorshift64(state)); 131 131 } 132 132 } 133 133 134 134 void main(MyThread & this) { 135 uint64_t state = prng();135 uint64_t state = thread_rand(); 136 136 137 137 // Wait for start … … 144 144 145 145 // Wait on a random spot 146 uint64_t idx = xorshift_13_7_17(state) % this.spots.len;146 uint64_t idx = __xorshift64(state) % this.spots.len; 147 147 bool closed = put(*this.spots.ptr[idx], this, this.data, this.share); 148 148 -
benchmark/readyQ/transfer.cfa
r97fed44 rf57f6ea0 1 1 #include "rq_bench.hfa" 2 2 #include <fstream.hfa> 3 #include <locale.h>4 3 5 4 Duration default_preemption() { … … 9 8 #define PRINT(...) 10 9 11 __ uint128_t lead_seed;10 __lehmer64_state_t lead_seed; 12 11 volatile unsigned leader; 13 12 volatile size_t lead_idx; … … 69 68 waitgroup(); 70 69 71 unsigned nleader = lehmer64( lead_seed ) % nthreads;70 unsigned nleader = __lehmer64( lead_seed ) % nthreads; 72 71 __atomic_store_n( &leader, nleader, __ATOMIC_SEQ_CST ); 73 72 … … 106 105 // ================================================== 107 106 int main(int argc, char * argv[]) { 108 uint64_t lead_seed = getpid();109 for(10) lehmer64( lead_seed );107 __lehmer64_state_t lead_seed = getpid(); 108 for(10) __lehmer64( lead_seed ); 110 109 unsigned nprocs = 2; 111 110 … … 127 126 128 127 lead_idx = 0; 129 leader = lehmer64( lead_seed ) % nthreads;128 leader = __lehmer64( lead_seed ) % nthreads; 130 129 131 130 size_t rechecks = 0; … … 168 167 } 169 168 170 setlocale( LC_NUMERIC, getenv( "LANG" ) );171 169 sout | "Duration (ms) : " | ws(3, 3, unit(eng((end - start)`dms))); 172 170 sout | "Number of processors : " | nprocs; -
libcfa/prelude/Makefile.am
r97fed44 rf57f6ea0 11 11 ## Created On : Sun May 31 08:54:01 2015 12 12 ## Last Modified By : Peter A. Buhr 13 ## Last Modified On : Thu Jan 13 17:06:27 202214 ## Update Count : 2 1513 ## Last Modified On : Mon Feb 3 21:27:18 2020 14 ## Update Count : 208 15 15 ############################################################################### 16 16 … … 37 37 # create extra forward types/declarations to reduce inclusion of library files 38 38 extras.cf : ${srcdir}/extras.regx ${srcdir}/extras.c 39 @echo '# 2 "${@}" // needed for error messages from this file' > ${@} 40 ${AM_V_GEN}gcc ${AM_CFLAGS} -E ${srcdir}/extras.c | grep -f ${srcdir}/extras.regx >> ${@} 41 ${AM_V_GEN}gcc ${AM_CFLAGS} -E ${srcdir}/extras.c | grep -zo -f ${srcdir}/extras.regx2 | tr '\0' '\n' >> ${@} 39 ${AM_V_GEN}gcc ${AM_CFLAGS} -E ${srcdir}/extras.c | grep -f ${srcdir}/extras.regx > extras.cf 40 ${AM_V_GEN}gcc ${AM_CFLAGS} -E ${srcdir}/extras.c | grep -zo -f ${srcdir}/extras.regx2 | tr '\0' '\n' >> extras.cf 42 41 43 42 # create forward declarations for gcc builtins 44 43 gcc-builtins.cf : gcc-builtins.c ${srcdir}/prototypes.sed 45 @echo '# 2 "${@}" // needed for error messages from this file' > ${@} 46 ${AM_V_GEN}gcc -I${srcdir} -E -P $< | sed -r -f ${srcdir}/prototypes.sed >> ${@} 44 ${AM_V_GEN}gcc -I${srcdir} -E -P $< | sed -r -f ${srcdir}/prototypes.sed > $@ 47 45 48 46 gcc-builtins.c : ${srcdir}/builtins.def ${srcdir}/prototypes.awk ${srcdir}/sync-builtins.cf ${srcdir}/prototypes.c 49 ${AM_V_GEN}gcc -I${srcdir} -E ${srcdir}/prototypes.c | awk -f ${srcdir}/prototypes.awk > $ {@}47 ${AM_V_GEN}gcc -I${srcdir} -E ${srcdir}/prototypes.c | awk -f ${srcdir}/prototypes.awk > $@ 50 48 51 49 prelude.cfa : prelude-gen.cc 52 50 ${AM_V_GEN}${CXX} ${AM_CXXFLAGS} ${CXXFLAGS} ${AM_CFLAGS} ${<} -o prelude-gen -Wall -Wextra -O2 -g -std=c++14 53 @./prelude-gen > $ {@}51 @./prelude-gen > $@ 54 52 @rm ./prelude-gen 55 53 … … 60 58 # create forward declarations for cfa builtins 61 59 builtins.cf : builtins.c @LOCAL_CFACC@ 62 ${AM_V_GEN}gcc ${AM_CFLAGS} -E ${<} -o ${@} -MD -MP -MF $(DEPDIR)/builtins.Po -D__cforall60 ${AM_V_GEN}gcc ${AM_CFLAGS} -E -P ${<} -o ${@} -MD -MP -MF $(DEPDIR)/builtins.Po -D__cforall 63 61 ${AM_V_at}sed -i 's/builtins.o/builtins.cf/g' $(DEPDIR)/builtins.Po 64 62 … … 66 64 67 65 bootloader.c : ${srcdir}/bootloader.cf prelude.cfa extras.cf gcc-builtins.cf builtins.cf @CFACPP@ 68 ${AM_V_GEN}@CFACPP@ --prelude-dir=${builddir} -tpm ${srcdir}/bootloader.cf $ {@}# use src/cfa-cpp as not in lib until after install66 ${AM_V_GEN}@CFACPP@ --prelude-dir=${builddir} -tpm ${srcdir}/bootloader.cf $@ # use src/cfa-cpp as not in lib until after install 69 67 70 68 maintainer-clean-local : -
libcfa/src/bits/random.hfa
r97fed44 rf57f6ea0 1 //2 // Cforall Version 1.0.0 Copyright (C) 2022 University of Waterloo3 //4 // The contents of this file are covered under the licence agreement in the5 // file "LICENCE" distributed with Cforall.6 //7 // random.hfa --8 //9 // Author : Peter A. Buhr10 // Created On : Fri Jan 14 07:18:11 202211 // Last Modified By : Peter A. Buhr12 // Last Modified On : Fri Jan 14 07:18:58 202213 // Update Count : 114 //15 16 1 #pragma once 17 2 18 3 #include <stdint.h> 19 4 20 // Pipelined to allow out-of-order overlap with reduced dependencies. Critically, the current random state is returned 21 // (copied), and then compute and store the next random value. 22 5 //-------------------------------------------------- 23 6 #if defined(__SIZEOF_INT128__) 24 //-------------------------------------------------- 25 static inline uint64_t lehmer64( __uint128_t & state ) { 26 __uint128_t ret = state; 7 typedef __uint128_t __lehmer64_state_t; 8 static inline uint64_t __lehmer64( __lehmer64_state_t & state ) { 27 9 state *= 0xda942042e4dd58b5; 28 return ret>> 64;10 return state >> 64; 29 11 } 30 12 31 13 //-------------------------------------------------- 32 static inline uint64_t wyhash64( uint64_t & state ) { 14 typedef uint64_t __wyhash64_state_t; 15 static inline uint64_t __wyhash64( __wyhash64_state_t & state ) { 33 16 state += 0x60bee2bee120fc15; 34 17 __uint128_t tmp; … … 42 25 43 26 //-------------------------------------------------- 44 static inline uint64_t xorshift_13_7_17( uint64_t & state ) { 45 uint64_t ret = state; 46 state ^= state << 13; 47 state ^= state >> 7; 48 state ^= state << 17; 49 return ret; 27 typedef uint64_t __xorshift64_state_t; 28 static inline uint64_t __xorshift64( __xorshift64_state_t & state ) { 29 uint64_t x = state; 30 x ^= x << 13; 31 x ^= x >> 7; 32 x ^= x << 17; 33 return state = x; 50 34 } 51 52 //--------------------------------------------------53 static inline uint32_t xorshift_6_21_7( uint32_t & state ) {54 uint32_t ret = state;55 state ^= state << 6;56 state ^= state >> 21;57 state ^= state << 7;58 return ret;59 } // xorshift_6_21_760 35 61 36 //-------------------------------------------------- … … 63 38 uint32_t a, b, c, d; 64 39 uint32_t counter; 65 } xorwow__state_t;40 } __xorwow__state_t; 66 41 67 // The state array must be initialized to not be all zero in the first four words. 68 static inline uint32_t xorwow( xorwow__state_t & state ) { 69 // Algorithm "xorwow" from p. 5 of Marsaglia, "Xorshift RNGs". 70 uint32_t ret = state.a + state.counter; 42 /* The state array must be initialized to not be all zero in the first four words */ 43 static inline uint32_t __xorwow( __xorwow__state_t & state ) { 44 /* Algorithm "xorwow" from p. 5 of Marsaglia, "Xorshift RNGs" */ 71 45 uint32_t t = state.d; 72 46 … … 82 56 83 57 state.counter += 362437; 84 return ret;58 return t + state.counter; 85 59 } 86 87 //--------------------------------------------------88 static inline uint32_t LCG( uint32_t & state ) { // linear congruential generator89 uint32_t ret = state;90 state = 36969 * (state & 65535) + (state >> 16); // 36969 is NOT prime! No not change it!91 return ret;92 } // LCG93 94 //--------------------------------------------------95 #define M (1_l64u << 48_l64u)96 #define A (25214903917_l64u)97 #define AI (18446708753438544741_l64u)98 #define C (11_l64u)99 #define D (16_l64u)100 101 // Bi-directional LCG random-number generator102 static inline uint32_t LCGBI_fwd( uint64_t & state ) {103 state = (A * state + C) & (M - 1);104 return state >> D;105 }106 107 static inline uint32_t LCGBI_bck( uint64_t & state ) {108 unsigned int r = state >> D;109 state = AI * (state - C) & (M - 1);110 return r;111 }112 113 #undef M114 #undef A115 #undef AI116 #undef C117 #undef D -
libcfa/src/common.hfa
r97fed44 rf57f6ea0 1 // 1 // 2 2 // Cforall Version 1.0.0 Copyright (C) 2016 University of Waterloo 3 3 // 4 4 // The contents of this file are covered under the licence agreement in the 5 5 // file "LICENCE" distributed with Cforall. 6 // 7 // common .hfa --8 // 6 // 7 // common -- 8 // 9 9 // Author : Peter A. Buhr 10 10 // Created On : Wed Jul 11 17:54:36 2018 … … 12 12 // Last Modified On : Wed May 5 14:02:04 2021 13 13 // Update Count : 18 14 // 14 // 15 15 16 16 #pragma once -
libcfa/src/concurrency/clib/cfathread.cfa
r97fed44 rf57f6ea0 22 22 #include "thread.hfa" 23 23 #include "time.hfa" 24 #include "stdlib.hfa"25 24 26 25 #include "cfathread.h" … … 196 195 eevent.data.u64 = (uint64_t)active_thread(); 197 196 198 int id = prng() % poller_cnt;197 int id = thread_rand() % poller_cnt; 199 198 if(0 != epoll_ctl(poller_fds[id], EPOLL_CTL_ADD, fd, &eevent)) 200 199 { -
libcfa/src/concurrency/io.cfa
r97fed44 rf57f6ea0 144 144 __ioarbiter_flush( ctx ); 145 145 146 if(ctx.sq.to_submit != 0 || min_comp > 0) { 147 148 __STATS__( true, io.calls.flush++; ) 149 int ret = syscall( __NR_io_uring_enter, ctx.fd, ctx.sq.to_submit, min_comp, min_comp > 0 ? IORING_ENTER_GETEVENTS : 0, (sigset_t *)0p, _NSIG / 8); 150 if( ret < 0 ) { 151 switch((int)errno) { 152 case EAGAIN: 153 case EINTR: 154 case EBUSY: 155 // Update statistics 156 __STATS__( false, io.calls.errors.busy ++; ) 157 return false; 158 default: 159 abort( "KERNEL ERROR: IO_URING SYSCALL - (%d) %s\n", (int)errno, strerror(errno) ); 160 } 146 __STATS__( true, io.calls.flush++; ) 147 int ret = syscall( __NR_io_uring_enter, ctx.fd, ctx.sq.to_submit, min_comp, min_comp > 0 ? IORING_ENTER_GETEVENTS : 0, (sigset_t *)0p, _NSIG / 8); 148 if( ret < 0 ) { 149 switch((int)errno) { 150 case EAGAIN: 151 case EINTR: 152 case EBUSY: 153 // Update statistics 154 __STATS__( false, io.calls.errors.busy ++; ) 155 return false; 156 default: 157 abort( "KERNEL ERROR: IO_URING SYSCALL - (%d) %s\n", (int)errno, strerror(errno) ); 161 158 } 162 163 __cfadbg_print_safe(io, "Kernel I/O : %u submitted to io_uring %d\n", ret, ctx.fd); 164 __STATS__( true, io.calls.submitted += ret; ) 165 /* paranoid */ verify( ctx.sq.to_submit <= *ctx.sq.num ); 166 /* paranoid */ verify( ctx.sq.to_submit >= ret ); 167 168 ctx.sq.to_submit -= ret; 169 170 /* paranoid */ verify( ctx.sq.to_submit <= *ctx.sq.num ); 171 172 // Release the consumed SQEs 173 __release_sqes( ctx ); 174 175 /* paranoid */ verify( ! __preemption_enabled() ); 176 177 ctx.proc->io.pending = false; 178 } 179 159 } 160 161 __cfadbg_print_safe(io, "Kernel I/O : %u submitted to io_uring %d\n", ret, ctx.fd); 162 __STATS__( true, io.calls.submitted += ret; ) 163 /* paranoid */ verify( ctx.sq.to_submit <= *ctx.sq.num ); 164 /* paranoid */ verify( ctx.sq.to_submit >= ret ); 165 166 ctx.sq.to_submit -= ret; 167 168 /* paranoid */ verify( ctx.sq.to_submit <= *ctx.sq.num ); 169 170 // Release the consumed SQEs 171 __release_sqes( ctx ); 172 173 /* paranoid */ verify( ! __preemption_enabled() ); 174 175 ctx.proc->io.pending = false; 180 176 ready_schedule_lock(); 181 177 bool ret = __cfa_io_drain( proc ); -
libcfa/src/concurrency/kernel.cfa
r97fed44 rf57f6ea0 205 205 // Don't block if we are done 206 206 if( __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST) ) break MAIN_LOOP; 207 208 #if !defined(__CFA_NO_STATISTICS__) 209 __tls_stats()->ready.sleep.halts++; 210 #endif 207 211 208 212 // Push self to idle stack … … 728 732 // Wake a thread from the front if there are any 729 733 static void __wake_one(cluster * this) { 734 /* paranoid */ verify( ! __preemption_enabled() ); 735 /* paranoid */ verify( ready_schedule_islocked() ); 736 737 // Check if there is a sleeping processor 738 // int fd = __atomic_load_n(&this->procs.fd, __ATOMIC_SEQ_CST); 739 int fd = 0; 740 if( __atomic_load_n(&this->procs.fd, __ATOMIC_SEQ_CST) != 0 ) { 741 fd = __atomic_exchange_n(&this->procs.fd, 0, __ATOMIC_RELAXED); 742 } 743 744 // If no one is sleeping, we are done 745 if( fd == 0 ) return; 746 747 // We found a processor, wake it up 730 748 eventfd_t val; 731 732 /* paranoid */ verify( ! __preemption_enabled() ); 733 /* paranoid */ verify( ready_schedule_islocked() ); 734 735 // Check if there is a sleeping processor 736 struct __fd_waitctx * fdp = __atomic_load_n(&this->procs.fdw, __ATOMIC_SEQ_CST); 737 738 // If no one is sleeping: we are done 739 if( fdp == 0p ) return; 740 741 int fd = 1; 742 if( __atomic_load_n(&fdp->fd, __ATOMIC_SEQ_CST) != 1 ) { 743 fd = __atomic_exchange_n(&fdp->fd, 1, __ATOMIC_RELAXED); 744 } 745 746 switch(fd) { 747 case 0: 748 // If the processor isn't ready to sleep then the exchange will already wake it up 749 #if !defined(__CFA_NO_STATISTICS__) 750 if( kernelTLS().this_stats ) { __tls_stats()->ready.sleep.early++; 751 } else { __atomic_fetch_add(&this->stats->ready.sleep.early, 1, __ATOMIC_RELAXED); } 752 #endif 753 break; 754 case 1: 755 // If someone else already said they will wake them: we are done 756 #if !defined(__CFA_NO_STATISTICS__) 757 if( kernelTLS().this_stats ) { __tls_stats()->ready.sleep.seen++; 758 } else { __atomic_fetch_add(&this->stats->ready.sleep.seen, 1, __ATOMIC_RELAXED); } 759 #endif 760 break; 761 default: 762 // If the processor was ready to sleep, we need to wake it up with an actual write 763 val = 1; 764 eventfd_write( fd, val ); 765 766 #if !defined(__CFA_NO_STATISTICS__) 767 if( kernelTLS().this_stats ) { __tls_stats()->ready.sleep.wakes++; 768 } else { __atomic_fetch_add(&this->stats->ready.sleep.wakes, 1, __ATOMIC_RELAXED); } 769 #endif 770 break; 771 } 749 val = 1; 750 eventfd_write( fd, val ); 751 752 #if !defined(__CFA_NO_STATISTICS__) 753 if( kernelTLS().this_stats ) { 754 __tls_stats()->ready.sleep.wakes++; 755 } 756 else { 757 __atomic_fetch_add(&this->stats->ready.sleep.wakes, 1, __ATOMIC_RELAXED); 758 } 759 #endif 772 760 773 761 /* paranoid */ verify( ready_schedule_islocked() ); … … 782 770 783 771 __cfadbg_print_safe(runtime_core, "Kernel : waking Processor %p\n", this); 784 785 this->idle_wctx.fd = 1;786 772 787 773 eventfd_t val; … … 793 779 794 780 static void idle_sleep(processor * this, io_future_t & future, iovec & iov) { 795 // Tell everyone we are ready to go do sleep796 for() {797 int expected = this->idle_wctx.fd;798 799 // Someone already told us to wake-up! No time for a nap.800 if(expected == 1) { return; }801 802 // Try to mark that we are going to sleep803 if(__atomic_compare_exchange_n(&this->idle_wctx.fd, &expected, this->idle_fd, false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST) ) {804 // Every one agreed, taking a nap805 break;806 }807 }808 809 810 781 #if !defined(CFA_WITH_IO_URING_IDLE) 811 782 #if !defined(__CFA_NO_STATISTICS__) … … 854 825 855 826 static bool mark_idle(__cluster_proc_list & this, processor & proc) { 856 #if !defined(__CFA_NO_STATISTICS__)857 __tls_stats()->ready.sleep.halts++;858 #endif859 860 proc.idle_wctx.fd = 0;861 862 827 /* paranoid */ verify( ! __preemption_enabled() ); 863 828 if(!try_lock( this )) return false; … … 867 832 insert_first(this.idles, proc); 868 833 869 __atomic_store_n(&this.fd w, &proc.idle_wctx, __ATOMIC_SEQ_CST);834 __atomic_store_n(&this.fd, proc.idle_fd, __ATOMIC_SEQ_CST); 870 835 unlock( this ); 871 836 /* paranoid */ verify( ! __preemption_enabled() ); … … 883 848 884 849 { 885 struct __fd_waitctx * wctx= 0;886 if(!this.idles`isEmpty) wctx = &this.idles`first.idle_wctx;887 __atomic_store_n(&this.fd w, wctx, __ATOMIC_SEQ_CST);850 int fd = 0; 851 if(!this.idles`isEmpty) fd = this.idles`first.idle_fd; 852 __atomic_store_n(&this.fd, fd, __ATOMIC_SEQ_CST); 888 853 } 889 854 -
libcfa/src/concurrency/kernel.hfa
r97fed44 rf57f6ea0 55 55 }; 56 56 57 58 struct __fd_waitctx {59 volatile int fd;60 };61 62 57 // Wrapper around kernel threads 63 58 struct __attribute__((aligned(128))) processor { … … 72 67 unsigned target; 73 68 unsigned last; 74 signed cpu; 69 unsigned cnt; 70 unsigned long long int cutoff; 75 71 } rdq; 76 72 … … 106 102 int idle_fd; 107 103 108 // Idle waitctx109 struct __fd_waitctx idle_wctx;110 111 104 // Termination synchronisation (user semaphore) 112 105 oneshot terminated; … … 159 152 volatile unsigned long long tv; 160 153 volatile unsigned long long ma; 161 };162 163 struct __attribute__((aligned(16))) __cache_id_t {164 volatile unsigned id;165 154 }; 166 155 … … 175 164 static inline void ^?{}(__timestamp_t & this) {} 176 165 177 struct __attribute__((aligned(128))) __ready_queue_caches_t;178 void ?{}(__ready_queue_caches_t & this);179 void ^?{}(__ready_queue_caches_t & this);180 181 166 //TODO adjust cache size to ARCHITECTURE 182 // Structure holding the re ady queue167 // Structure holding the relaxed ready queue 183 168 struct __ready_queue_t { 184 169 // Data tracking the actual lanes … … 193 178 __timestamp_t * volatile tscs; 194 179 195 __cache_id_t * volatile caches;196 197 180 // Array of stats 198 181 __help_cnts_t * volatile help; … … 215 198 216 199 // FD to use to wake a processor 217 struct __fd_waitctx * volatile fdw;200 volatile int fd; 218 201 219 202 // Total number of processors -
libcfa/src/concurrency/kernel/fwd.hfa
r97fed44 rf57f6ea0 79 79 return 80 80 #if defined(__SIZEOF_INT128__) 81 lehmer64( kernelTLS().rand_seed );81 __lehmer64( kernelTLS().rand_seed ); 82 82 #else 83 xorshift_13_7_17( kernelTLS().rand_seed );83 __xorshift64( kernelTLS().rand_seed ); 84 84 #endif 85 85 } 86 86 87 #define M (1_l64u << 48_l64u) 88 #define A (25214903917_l64u) 89 #define AI (18446708753438544741_l64u) 90 #define C (11_l64u) 91 #define D (16_l64u) 92 87 93 static inline unsigned __tls_rand_fwd() { 88 return LCGBI_fwd( kernelTLS().ready_rng.fwd_seed ); 94 kernelTLS().ready_rng.fwd_seed = (A * kernelTLS().ready_rng.fwd_seed + C) & (M - 1); 95 return kernelTLS().ready_rng.fwd_seed >> D; 89 96 } 90 97 91 98 static inline unsigned __tls_rand_bck() { 92 return LCGBI_bck( kernelTLS().ready_rng.bck_seed ); 93 } 99 unsigned int r = kernelTLS().ready_rng.bck_seed >> D; 100 kernelTLS().ready_rng.bck_seed = AI * (kernelTLS().ready_rng.bck_seed - C) & (M - 1); 101 return r; 102 } 103 104 #undef M 105 #undef A 106 #undef AI 107 #undef C 108 #undef D 94 109 95 110 static inline void __tls_rand_advance_bck(void) { … … 125 140 } 126 141 } 142 143 extern uint64_t thread_rand(); 127 144 128 145 // Semaphore which only supports a single thread -
libcfa/src/concurrency/kernel/startup.cfa
r97fed44 rf57f6ea0 34 34 #include "kernel_private.hfa" 35 35 #include "startup.hfa" // STARTUP_PRIORITY_XXX 36 #include "limits.hfa"37 36 #include "math.hfa" 38 37 … … 178 177 179 178 179 180 180 //============================================================================================= 181 181 // Kernel Setup logic … … 515 515 this.rdq.its = 0; 516 516 this.rdq.itr = 0; 517 this.rdq.id = MAX; 518 this.rdq.target = MAX; 519 this.rdq.last = MAX; 520 this.rdq.cpu = 0; 521 // this.rdq.cutoff = 0ull; 517 this.rdq.id = -1u; 518 this.rdq.target = -1u; 519 this.rdq.last = -1u; 520 this.rdq.cutoff = 0ull; 522 521 do_terminate = false; 523 522 preemption_alarm = 0p; … … 537 536 } 538 537 539 this.idle_wctx.fd = 0;540 541 // I'm assuming these two are reserved for standard input and output542 // so I'm using them as sentinels with idle_wctx.543 /* paranoid */ verify( this.idle_fd != 0 );544 /* paranoid */ verify( this.idle_fd != 1 );545 546 538 #if !defined(__CFA_NO_STATISTICS__) 547 539 print_stats = 0; … … 597 589 // Cluster 598 590 static void ?{}(__cluster_proc_list & this) { 599 this.fd w = 0p;591 this.fd = 0; 600 592 this.idle = 0; 601 593 this.total = 0; … … 694 686 uint_fast32_t last_size; 695 687 [this->unique_id, last_size] = ready_mutate_register(); 696 697 this->rdq.cpu = __kernel_getcpu();698 688 699 689 this->cltr->procs.total += 1u; -
libcfa/src/concurrency/locks.hfa
r97fed44 rf57f6ea0 31 31 32 32 //----------------------------------------------------------------------------- 33 // Semaphores 34 35 // '0-nary' semaphore 36 // Similar to a counting semaphore except the value of one is never reached 37 // as a consequence, a V() that would bring the value to 1 *spins* until 38 // a P consumes it 39 struct Semaphore0nary { 40 __spinlock_t lock; // needed to protect 41 mpsc_queue(thread$) queue; 42 }; 43 44 static inline bool P(Semaphore0nary & this, thread$ * thrd) { 45 /* paranoid */ verify(!thrd`next); 46 /* paranoid */ verify(!(&(*thrd)`next)); 47 48 push(this.queue, thrd); 49 return true; 50 } 51 52 static inline bool P(Semaphore0nary & this) { 53 thread$ * thrd = active_thread(); 54 P(this, thrd); 55 park(); 56 return true; 57 } 58 59 static inline thread$ * V(Semaphore0nary & this, bool doUnpark = true) { 60 thread$ * next; 61 lock(this.lock __cfaabi_dbg_ctx2); 62 for (;;) { 63 next = pop(this.queue); 64 if (next) break; 65 Pause(); 66 } 67 unlock(this.lock); 68 69 if (doUnpark) unpark(next); 70 return next; 71 } 72 73 // Wrapper used on top of any sempahore to avoid potential locking 74 struct BinaryBenaphore { 75 volatile ssize_t counter; 76 }; 77 78 static inline { 79 void ?{}(BinaryBenaphore & this) { this.counter = 0; } 80 void ?{}(BinaryBenaphore & this, zero_t) { this.counter = 0; } 81 void ?{}(BinaryBenaphore & this, one_t ) { this.counter = 1; } 82 83 // returns true if no blocking needed 84 bool P(BinaryBenaphore & this) { 85 return __atomic_fetch_sub(&this.counter, 1, __ATOMIC_SEQ_CST) > 0; 86 } 87 88 bool tryP(BinaryBenaphore & this) { 89 ssize_t c = this.counter; 90 /* paranoid */ verify( c > MIN ); 91 return (c >= 1) && __atomic_compare_exchange_n(&this.counter, &c, c-1, false, __ATOMIC_SEQ_CST, __ATOMIC_RELAXED); 92 } 93 94 // returns true if notify needed 95 bool V(BinaryBenaphore & this) { 96 ssize_t c = 0; 97 for () { 98 /* paranoid */ verify( this.counter < MAX ); 99 if (__atomic_compare_exchange_n(&this.counter, &c, c+1, false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST)) { 100 if (c == 0) return true; 101 /* paranoid */ verify(c < 0); 102 return false; 103 } else { 104 if (c == 1) return true; 105 /* paranoid */ verify(c < 1); 106 Pause(); 107 } 108 } 109 } 110 } 111 112 // Binary Semaphore based on the BinaryBenaphore on top of the 0-nary Semaphore 113 struct ThreadBenaphore { 114 BinaryBenaphore ben; 115 Semaphore0nary sem; 116 }; 117 118 static inline void ?{}(ThreadBenaphore & this) {} 119 static inline void ?{}(ThreadBenaphore & this, zero_t) { (this.ben){ 0 }; } 120 static inline void ?{}(ThreadBenaphore & this, one_t ) { (this.ben){ 1 }; } 121 122 static inline bool P(ThreadBenaphore & this) { return P(this.ben) ? false : P(this.sem); } 123 static inline bool tryP(ThreadBenaphore & this) { return tryP(this.ben); } 124 static inline bool P(ThreadBenaphore & this, bool wait) { return wait ? P(this) : tryP(this); } 125 126 static inline thread$ * V(ThreadBenaphore & this, bool doUnpark = true) { 127 if (V(this.ben)) return 0p; 128 return V(this.sem, doUnpark); 129 } 130 131 //----------------------------------------------------------------------------- 33 132 // Semaphore 34 133 struct semaphore { … … 72 171 static inline void on_wakeup( owner_lock & this, size_t v ) { on_wakeup ( (blocking_lock &)this, v ); } 73 172 static inline void on_notify( owner_lock & this, struct thread$ * t ) { on_notify( (blocking_lock &)this, t ); } 173 174 struct fast_lock { 175 thread$ * volatile owner; 176 ThreadBenaphore sem; 177 }; 178 179 static inline void ?{}(fast_lock & this) __attribute__((deprecated("use linear_backoff_then_block_lock instead"))); 180 static inline void ?{}(fast_lock & this) { this.owner = 0p; } 181 182 static inline bool $try_lock(fast_lock & this, thread$ * thrd) { 183 thread$ * exp = 0p; 184 return __atomic_compare_exchange_n(&this.owner, &exp, thrd, false, __ATOMIC_SEQ_CST, __ATOMIC_RELAXED); 185 } 186 187 static inline void lock( fast_lock & this ) __attribute__((deprecated("use linear_backoff_then_block_lock instead"), artificial)); 188 static inline void lock( fast_lock & this ) { 189 thread$ * thrd = active_thread(); 190 /* paranoid */verify(thrd != this.owner); 191 192 for (;;) { 193 if ($try_lock(this, thrd)) return; 194 P(this.sem); 195 } 196 } 197 198 static inline bool try_lock( fast_lock & this ) __attribute__((deprecated("use linear_backoff_then_block_lock instead"), artificial)); 199 static inline bool try_lock ( fast_lock & this ) { 200 thread$ * thrd = active_thread(); 201 /* paranoid */ verify(thrd != this.owner); 202 return $try_lock(this, thrd); 203 } 204 205 static inline thread$ * unlock( fast_lock & this ) __attribute__((deprecated("use linear_backoff_then_block_lock instead"), artificial)); 206 static inline thread$ * unlock( fast_lock & this ) { 207 /* paranoid */ verify(active_thread() == this.owner); 208 209 // open 'owner' before unlocking anyone 210 // so new and unlocked threads don't park incorrectly. 211 // This may require additional fencing on ARM. 212 this.owner = 0p; 213 214 return V(this.sem); 215 } 216 217 static inline size_t on_wait( fast_lock & this ) { unlock(this); return 0; } 218 static inline void on_wakeup( fast_lock & this, size_t ) { lock(this); } 219 static inline void on_notify( fast_lock &, struct thread$ * t ) { unpark(t); } 74 220 75 221 struct mcs_node { -
libcfa/src/concurrency/ready_queue.cfa
r97fed44 rf57f6ea0 20 20 21 21 22 //#define USE_RELAXED_FIFO22 #define USE_RELAXED_FIFO 23 23 // #define USE_WORK_STEALING 24 24 // #define USE_CPU_WORK_STEALING 25 #define USE_AWARE_STEALING26 25 27 26 #include "bits/defs.hfa" … … 30 29 31 30 #include "stdlib.hfa" 32 #include "limits.hfa"33 31 #include "math.hfa" 34 32 … … 56 54 #endif 57 55 58 #if defined(USE_AWARE_STEALING) 59 #define READYQ_SHARD_FACTOR 2 60 #define SEQUENTIAL_SHARD 2 61 #elif defined(USE_CPU_WORK_STEALING) 56 #if defined(USE_CPU_WORK_STEALING) 62 57 #define READYQ_SHARD_FACTOR 2 63 58 #elif defined(USE_RELAXED_FIFO) … … 143 138 __kernel_rseq_register(); 144 139 140 __cfadbg_print_safe(ready_queue, "Kernel : Registering proc %p for RW-Lock\n", proc); 145 141 bool * handle = (bool *)&kernelTLS().sched_lock; 146 142 … … 178 174 } 179 175 176 __cfadbg_print_safe(ready_queue, "Kernel : Registering proc %p done, id %lu\n", proc, n); 177 180 178 // Return new spot. 181 179 /* paranoid */ verify(n < ready); … … 192 190 193 191 __atomic_store_n(cell, 0p, __ATOMIC_RELEASE); 192 193 __cfadbg_print_safe(ready_queue, "Kernel : Unregister proc %p\n", proc); 194 194 195 195 __kernel_rseq_unregister(); … … 201 201 uint_fast32_t ready_mutate_lock( void ) with(*__scheduler_lock) { 202 202 /* paranoid */ verify( ! __preemption_enabled() ); 203 /* paranoid */ verify( ! kernelTLS().sched_lock ); 203 204 204 205 // Step 1 : lock global lock … … 206 207 // to simply lock their own lock and enter. 207 208 __atomic_acquire( &write_lock ); 208 209 // Make sure we won't deadlock ourself210 // Checking before acquiring the writer lock isn't safe211 // because someone else could have locked us.212 /* paranoid */ verify( ! kernelTLS().sched_lock );213 209 214 210 // Step 2 : lock per-proc lock … … 248 244 249 245 //======================================================================= 250 // caches handling251 252 struct __attribute__((aligned(128))) __ready_queue_caches_t {253 // Count States:254 // - 0 : No one is looking after this cache255 // - 1 : No one is looking after this cache, BUT it's not empty256 // - 2+ : At least one processor is looking after this cache257 volatile unsigned count;258 };259 260 void ?{}(__ready_queue_caches_t & this) { this.count = 0; }261 void ^?{}(__ready_queue_caches_t & this) {}262 263 static inline void depart(__ready_queue_caches_t & cache) {264 /* paranoid */ verify( cache.count > 1);265 __atomic_fetch_add(&cache.count, -1, __ATOMIC_SEQ_CST);266 /* paranoid */ verify( cache.count != 0);267 /* paranoid */ verify( cache.count < 65536 ); // This verify assumes no cluster will have more than 65000 kernel threads mapped to a single cache, which could be correct but is super weird.268 }269 270 static inline void arrive(__ready_queue_caches_t & cache) {271 // for() {272 // unsigned expected = cache.count;273 // unsigned desired = 0 == expected ? 2 : expected + 1;274 // }275 }276 277 //=======================================================================278 246 // Cforall Ready Queue used for scheduling 279 247 //======================================================================= 280 unsigned long long moving_average(unsigned long long currtsc, unsigned long long instsc, unsigned long long old_avg) { 281 /* paranoid */ verifyf( currtsc < 45000000000000000, "Suspiciously large current time: %'llu (%llx)\n", currtsc, currtsc ); 282 /* paranoid */ verifyf( instsc < 45000000000000000, "Suspiciously large insert time: %'llu (%llx)\n", instsc, instsc ); 283 /* paranoid */ verifyf( old_avg < 15000000000000, "Suspiciously large previous average: %'llu (%llx)\n", old_avg, old_avg ); 284 285 const unsigned long long new_val = currtsc > instsc ? currtsc - instsc : 0; 286 const unsigned long long total_weight = 16; 287 const unsigned long long new_weight = 4; 288 const unsigned long long old_weight = total_weight - new_weight; 289 const unsigned long long ret = ((new_weight * new_val) + (old_weight * old_avg)) / total_weight; 290 return ret; 248 unsigned long long moving_average(unsigned long long nval, unsigned long long oval) { 249 const unsigned long long tw = 16; 250 const unsigned long long nw = 4; 251 const unsigned long long ow = tw - nw; 252 return ((nw * nval) + (ow * oval)) / tw; 291 253 } 292 254 … … 309 271 } 310 272 #else 311 lanes.data = 0p; 312 lanes.tscs = 0p; 313 lanes.caches = 0p; 314 lanes.help = 0p; 315 lanes.count = 0; 273 lanes.data = 0p; 274 lanes.tscs = 0p; 275 lanes.help = 0p; 276 lanes.count = 0; 316 277 #endif 317 278 } … … 324 285 free(lanes.data); 325 286 free(lanes.tscs); 326 free(lanes.caches);327 287 free(lanes.help); 328 288 } 329 289 330 290 //----------------------------------------------------------------------- 331 #if defined(USE_AWARE_STEALING)332 __attribute__((hot)) void push(struct cluster * cltr, struct thread$ * thrd, unpark_hint hint) with (cltr->ready_queue) {333 processor * const proc = kernelTLS().this_processor;334 const bool external = (!proc) || (cltr != proc->cltr);335 const bool remote = hint == UNPARK_REMOTE;336 337 unsigned i;338 if( external || remote ) {339 // Figure out where thread was last time and make sure it's valid340 /* paranoid */ verify(thrd->preferred >= 0);341 if(thrd->preferred * READYQ_SHARD_FACTOR < lanes.count) {342 /* paranoid */ verify(thrd->preferred * READYQ_SHARD_FACTOR < lanes.count);343 unsigned start = thrd->preferred * READYQ_SHARD_FACTOR;344 do {345 unsigned r = __tls_rand();346 i = start + (r % READYQ_SHARD_FACTOR);347 /* paranoid */ verify( i < lanes.count );348 // If we can't lock it retry349 } while( !__atomic_try_acquire( &lanes.data[i].lock ) );350 } else {351 do {352 i = __tls_rand() % lanes.count;353 } while( !__atomic_try_acquire( &lanes.data[i].lock ) );354 }355 } else {356 do {357 unsigned r = proc->rdq.its++;358 i = proc->rdq.id + (r % READYQ_SHARD_FACTOR);359 /* paranoid */ verify( i < lanes.count );360 // If we can't lock it retry361 } while( !__atomic_try_acquire( &lanes.data[i].lock ) );362 }363 364 // Actually push it365 push(lanes.data[i], thrd);366 367 // Unlock and return368 __atomic_unlock( &lanes.data[i].lock );369 370 #if !defined(__CFA_NO_STATISTICS__)371 if(unlikely(external || remote)) __atomic_fetch_add(&cltr->stats->ready.push.extrn.success, 1, __ATOMIC_RELAXED);372 else __tls_stats()->ready.push.local.success++;373 #endif374 }375 376 static inline unsigned long long calc_cutoff(const unsigned long long ctsc, const processor * proc, __ready_queue_t & rdq) {377 unsigned start = proc->rdq.id;378 unsigned long long max = 0;379 for(i; READYQ_SHARD_FACTOR) {380 unsigned long long ptsc = ts(rdq.lanes.data[start + i]);381 if(ptsc != -1ull) {382 /* paranoid */ verify( start + i < rdq.lanes.count );383 unsigned long long tsc = moving_average(ctsc, ptsc, rdq.lanes.tscs[start + i].ma);384 if(tsc > max) max = tsc;385 }386 }387 return (max + 2 * max) / 2;388 }389 390 __attribute__((hot)) struct thread$ * pop_fast(struct cluster * cltr) with (cltr->ready_queue) {391 /* paranoid */ verify( lanes.count > 0 );392 /* paranoid */ verify( kernelTLS().this_processor );393 /* paranoid */ verify( kernelTLS().this_processor->rdq.id < lanes.count );394 395 processor * const proc = kernelTLS().this_processor;396 unsigned this = proc->rdq.id;397 /* paranoid */ verify( this < lanes.count );398 __cfadbg_print_safe(ready_queue, "Kernel : pop from %u\n", this);399 400 // Figure out the current cpu and make sure it is valid401 const int cpu = __kernel_getcpu();402 /* paranoid */ verify(cpu >= 0);403 /* paranoid */ verify(cpu < cpu_info.hthrd_count);404 unsigned this_cache = cpu_info.llc_map[cpu].cache;405 406 // Super important: don't write the same value over and over again407 // We want to maximise our chances that his particular values stays in cache408 if(lanes.caches[this / READYQ_SHARD_FACTOR].id != this_cache)409 __atomic_store_n(&lanes.caches[this / READYQ_SHARD_FACTOR].id, this_cache, __ATOMIC_RELAXED);410 411 const unsigned long long ctsc = rdtscl();412 413 if(proc->rdq.target == MAX) {414 uint64_t chaos = __tls_rand();415 unsigned ext = chaos & 0xff;416 unsigned other = (chaos >> 8) % (lanes.count);417 418 if(ext < 3 || __atomic_load_n(&lanes.caches[other / READYQ_SHARD_FACTOR].id, __ATOMIC_RELAXED) == this_cache) {419 proc->rdq.target = other;420 }421 }422 else {423 const unsigned target = proc->rdq.target;424 __cfadbg_print_safe(ready_queue, "Kernel : %u considering helping %u, tcsc %llu\n", this, target, lanes.tscs[target].tv);425 /* paranoid */ verify( lanes.tscs[target].tv != MAX );426 if(target < lanes.count) {427 const unsigned long long cutoff = calc_cutoff(ctsc, proc, cltr->ready_queue);428 const unsigned long long age = moving_average(ctsc, lanes.tscs[target].tv, lanes.tscs[target].ma);429 __cfadbg_print_safe(ready_queue, "Kernel : Help attempt on %u from %u, age %'llu vs cutoff %'llu, %s\n", target, this, age, cutoff, age > cutoff ? "yes" : "no");430 if(age > cutoff) {431 thread$ * t = try_pop(cltr, target __STATS(, __tls_stats()->ready.pop.help));432 if(t) return t;433 }434 }435 proc->rdq.target = MAX;436 }437 438 for(READYQ_SHARD_FACTOR) {439 unsigned i = this + (proc->rdq.itr++ % READYQ_SHARD_FACTOR);440 if(thread$ * t = try_pop(cltr, i __STATS(, __tls_stats()->ready.pop.local))) return t;441 }442 443 // All lanes where empty return 0p444 return 0p;445 446 }447 __attribute__((hot)) struct thread$ * pop_slow(struct cluster * cltr) with (cltr->ready_queue) {448 unsigned i = __tls_rand() % lanes.count;449 return try_pop(cltr, i __STATS(, __tls_stats()->ready.pop.steal));450 }451 __attribute__((hot)) struct thread$ * pop_search(struct cluster * cltr) {452 return search(cltr);453 }454 #endif455 291 #if defined(USE_CPU_WORK_STEALING) 456 292 __attribute__((hot)) void push(struct cluster * cltr, struct thread$ * thrd, unpark_hint hint) with (cltr->ready_queue) { … … 514 350 /* paranoid */ verify( kernelTLS().this_processor ); 515 351 516 processor * const proc = kernelTLS().this_processor;517 352 const int cpu = __kernel_getcpu(); 518 353 /* paranoid */ verify(cpu >= 0); … … 525 360 /* paranoid */ verifyf((map.start + map.count) * READYQ_SHARD_FACTOR <= lanes.count, "have %zu lanes but map can go up to %u", lanes.count, (map.start + map.count) * READYQ_SHARD_FACTOR); 526 361 362 processor * const proc = kernelTLS().this_processor; 527 363 const int start = map.self * READYQ_SHARD_FACTOR; 528 364 const unsigned long long ctsc = rdtscl(); 529 365 530 366 // Did we already have a help target 531 if(proc->rdq.target == MAX) {367 if(proc->rdq.target == -1u) { 532 368 unsigned long long max = 0; 533 369 for(i; READYQ_SHARD_FACTOR) { 534 unsigned long long tsc = moving_average(ctsc ,ts(lanes.data[start + i]), lanes.tscs[start + i].ma);370 unsigned long long tsc = moving_average(ctsc - ts(lanes.data[start + i]), lanes.tscs[start + i].ma); 535 371 if(tsc > max) max = tsc; 536 372 } 537 //proc->rdq.cutoff = (max + 2 * max) / 2;373 proc->rdq.cutoff = (max + 2 * max) / 2; 538 374 /* paranoid */ verify(lanes.count < 65536); // The following code assumes max 65536 cores. 539 375 /* paranoid */ verify(map.count < 65536); // The following code assumes max 65536 cores. … … 548 384 } 549 385 550 /* paranoid */ verify(proc->rdq.target != MAX);386 /* paranoid */ verify(proc->rdq.target != -1u); 551 387 } 552 388 else { 553 389 unsigned long long max = 0; 554 390 for(i; READYQ_SHARD_FACTOR) { 555 unsigned long long tsc = moving_average(ctsc ,ts(lanes.data[start + i]), lanes.tscs[start + i].ma);391 unsigned long long tsc = moving_average(ctsc - ts(lanes.data[start + i]), lanes.tscs[start + i].ma); 556 392 if(tsc > max) max = tsc; 557 393 } … … 559 395 { 560 396 unsigned target = proc->rdq.target; 561 proc->rdq.target = MAX;397 proc->rdq.target = -1u; 562 398 lanes.help[target / READYQ_SHARD_FACTOR].tri++; 563 if(moving_average(ctsc ,lanes.tscs[target].tv, lanes.tscs[target].ma) > cutoff) {399 if(moving_average(ctsc - lanes.tscs[target].tv, lanes.tscs[target].ma) > cutoff) { 564 400 thread$ * t = try_pop(cltr, target __STATS(, __tls_stats()->ready.pop.help)); 565 401 proc->rdq.last = target; 566 402 if(t) return t; 403 else proc->rdq.target = -1u; 567 404 } 568 proc->rdq.target = MAX;405 else proc->rdq.target = -1u; 569 406 } 570 407 571 408 unsigned last = proc->rdq.last; 572 if(last != MAX && moving_average(ctsc, lanes.tscs[last].tv, lanes.tscs[last].ma) >cutoff) {409 if(last != -1u && lanes.tscs[last].tv < cutoff && ts(lanes.data[last]) < cutoff) { 573 410 thread$ * t = try_pop(cltr, last __STATS(, __tls_stats()->ready.pop.help)); 574 411 if(t) return t; 575 412 } 576 413 else { 577 proc->rdq.last = MAX;414 proc->rdq.last = -1u; 578 415 } 579 416 } … … 591 428 processor * const proc = kernelTLS().this_processor; 592 429 unsigned last = proc->rdq.last; 593 if(last != MAX) {430 if(last != -1u) { 594 431 struct thread$ * t = try_pop(cltr, last __STATS(, __tls_stats()->ready.pop.steal)); 595 432 if(t) return t; 596 proc->rdq.last = MAX;433 proc->rdq.last = -1u; 597 434 } 598 435 … … 723 560 #else 724 561 unsigned preferred = thrd->preferred; 725 const bool external = (hint != UNPARK_LOCAL) || (!kernelTLS().this_processor) || preferred == MAX|| thrd->curr_cluster != cltr;562 const bool external = (hint != UNPARK_LOCAL) || (!kernelTLS().this_processor) || preferred == -1u || thrd->curr_cluster != cltr; 726 563 /* paranoid */ verifyf(external || preferred < lanes.count, "Invalid preferred queue %u for %u lanes", preferred, lanes.count ); 727 564 … … 775 612 processor * proc = kernelTLS().this_processor; 776 613 777 if(proc->rdq.target == MAX) {614 if(proc->rdq.target == -1u) { 778 615 unsigned long long min = ts(lanes.data[proc->rdq.id]); 779 616 for(int i = 0; i < READYQ_SHARD_FACTOR; i++) { … … 786 623 else { 787 624 unsigned target = proc->rdq.target; 788 proc->rdq.target = MAX;625 proc->rdq.target = -1u; 789 626 const unsigned long long bias = 0; //2_500_000_000; 790 627 const unsigned long long cutoff = proc->rdq.cutoff > bias ? proc->rdq.cutoff - bias : proc->rdq.cutoff; … … 821 658 // try to pop from a lane given by index w 822 659 static inline struct thread$ * try_pop(struct cluster * cltr, unsigned w __STATS(, __stats_readyQ_pop_t & stats)) with (cltr->ready_queue) { 823 /* paranoid */ verify( w < lanes.count );824 660 __STATS( stats.attempt++; ) 825 661 … … 845 681 // Actually pop the list 846 682 struct thread$ * thrd; 847 #if defined(USE_ AWARE_STEALING) || defined(USE_WORK_STEALING) || defined(USE_CPU_WORK_STEALING)683 #if defined(USE_WORK_STEALING) || defined(USE_CPU_WORK_STEALING) 848 684 unsigned long long tsc_before = ts(lane); 849 685 #endif … … 861 697 __STATS( stats.success++; ) 862 698 863 #if defined(USE_AWARE_STEALING) || defined(USE_WORK_STEALING) || defined(USE_CPU_WORK_STEALING) 864 if (tsv != MAX) { 865 unsigned long long now = rdtscl(); 866 unsigned long long pma = __atomic_load_n(&lanes.tscs[w].ma, __ATOMIC_RELAXED); 867 __atomic_store_n(&lanes.tscs[w].tv, tsv, __ATOMIC_RELAXED); 868 __atomic_store_n(&lanes.tscs[w].ma, moving_average(now, tsc_before, pma), __ATOMIC_RELAXED); 869 } 699 #if defined(USE_WORK_STEALING) || defined(USE_CPU_WORK_STEALING) 700 unsigned long long now = rdtscl(); 701 lanes.tscs[w].tv = tsv; 702 lanes.tscs[w].ma = moving_average(now > tsc_before ? now - tsc_before : 0, lanes.tscs[w].ma); 870 703 #endif 871 704 872 #if defined(USE_ AWARE_STEALING) || defined(USE_CPU_WORK_STEALING)705 #if defined(USE_CPU_WORK_STEALING) 873 706 thrd->preferred = w / READYQ_SHARD_FACTOR; 874 707 #else … … 969 802 /* paranoid */ verifyf( it, "Unexpected null iterator, at index %u of %u\n", i, count); 970 803 it->rdq.id = value; 971 it->rdq.target = MAX;804 it->rdq.target = -1u; 972 805 value += READYQ_SHARD_FACTOR; 973 806 it = &(*it)`next; … … 982 815 983 816 static void fix_times( struct cluster * cltr ) with( cltr->ready_queue ) { 984 #if defined(USE_ AWARE_STEALING) || defined(USE_WORK_STEALING)817 #if defined(USE_WORK_STEALING) 985 818 lanes.tscs = alloc(lanes.count, lanes.tscs`realloc); 986 819 for(i; lanes.count) { 987 lanes.tscs[i].tv = rdtscl(); 988 lanes.tscs[i].ma = 0; 820 unsigned long long tsc1 = ts(lanes.data[i]); 821 unsigned long long tsc2 = rdtscl(); 822 lanes.tscs[i].tv = min(tsc1, tsc2); 989 823 } 990 824 #endif … … 1032 866 // Update original 1033 867 lanes.count = ncount; 1034 1035 lanes.caches = alloc( target, lanes.caches`realloc );1036 868 } 1037 869 … … 1110 942 fix(lanes.data[idx]); 1111 943 } 1112 1113 lanes.caches = alloc( target, lanes.caches`realloc );1114 944 } 1115 945 1116 946 fix_times(cltr); 1117 1118 947 1119 948 reassign_cltr_id(cltr); -
libcfa/src/concurrency/stats.cfa
r97fed44 rf57f6ea0 31 31 stats->ready.sleep.halts = 0; 32 32 stats->ready.sleep.cancels = 0; 33 stats->ready.sleep.early = 0;34 33 stats->ready.sleep.wakes = 0; 35 stats->ready.sleep.seen = 0;36 34 stats->ready.sleep.exits = 0; 37 35 … … 93 91 tally_one( &cltr->ready.sleep.halts , &proc->ready.sleep.halts ); 94 92 tally_one( &cltr->ready.sleep.cancels , &proc->ready.sleep.cancels ); 95 tally_one( &cltr->ready.sleep.early , &proc->ready.sleep.early );96 93 tally_one( &cltr->ready.sleep.wakes , &proc->ready.sleep.wakes ); 97 tally_one( &cltr->ready.sleep.seen , &proc->ready.sleep.wakes );98 94 tally_one( &cltr->ready.sleep.exits , &proc->ready.sleep.exits ); 99 95 … … 157 153 | " (" | eng3(ready.pop.search.attempt) | " try)"; 158 154 159 sstr | "- Idle Slp : " | eng3(ready.sleep.halts) | "halt," | eng3(ready.sleep.cancels) | "cancel," 160 | eng3(ready.sleep.wakes + ready.sleep.early) | '(' | eng3(ready.sleep.early) | ',' | eng3(ready.sleep.seen) | ')' | " wake(early, seen)," 161 | eng3(ready.sleep.exits) | "exit"; 155 sstr | "- Idle Slp : " | eng3(ready.sleep.halts) | "halt," | eng3(ready.sleep.cancels) | "cancel," | eng3(ready.sleep.wakes) | "wake," | eng3(ready.sleep.exits) | "exit"; 162 156 sstr | nl; 163 157 } -
libcfa/src/concurrency/stats.hfa
r97fed44 rf57f6ea0 69 69 volatile uint64_t halts; 70 70 volatile uint64_t cancels; 71 volatile uint64_t early;72 71 volatile uint64_t wakes; 73 volatile uint64_t seen;74 72 volatile uint64_t exits; 75 73 } sleep; -
libcfa/src/concurrency/thread.cfa
r97fed44 rf57f6ea0 10 10 // Created On : Tue Jan 17 12:27:26 2017 11 11 // Last Modified By : Peter A. Buhr 12 // Last Modified On : Thu Jan 13 20:11:55202213 // Update Count : 4212 // Last Modified On : Wed Jan 12 18:46:48 2022 13 // Update Count : 36 14 14 // 15 15 … … 24 24 #define __CFA_INVOKE_PRIVATE__ 25 25 #include "invoke.h" 26 27 uint64_t thread_rand(); 26 28 27 29 extern uint32_t __global_random_seed; … … 172 174 } 173 175 174 //----------------------------------------------------------------------------- 176 uint64_t thread_rand() { 177 disable_interrupts(); 178 uint64_t ret = __tls_rand(); 179 enable_interrupts(); 180 return ret; 181 } 182 175 183 #define GENERATOR LCG 184 185 static inline uint32_t MarsagliaXor( uint32_t & state ) { 186 uint32_t ret = state; 187 state ^= state << 6; 188 state ^= state >> 21; 189 state ^= state << 7; 190 return ret; 191 } // MarsagliaXor 192 193 static inline uint32_t LCG( uint32_t & state ) { // linear congruential generator 194 uint32_t ret = state; 195 state = 36969 * (state & 65535) + (state >> 16); // 36969 is NOT prime! No not change it! 196 return ret; 197 } // LCG 176 198 177 199 void set_seed( uint32_t seed ) { -
libcfa/src/iostream.cfa
r97fed44 rf57f6ea0 10 10 // Created On : Wed May 27 17:56:53 2015 11 11 // Last Modified By : Peter A. Buhr 12 // Last Modified On : Wed Jan 19 08:15:53 202213 // Update Count : 13 5212 // Last Modified On : Sun Oct 10 09:28:17 2021 13 // Update Count : 1345 14 14 // 15 15 … … 57 57 ostype & ?|?( ostype & os, signed char sc ) { 58 58 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 59 fmt( os, "% 'hhd", sc );59 fmt( os, "%hhd", sc ); 60 60 return os; 61 61 } // ?|? … … 66 66 ostype & ?|?( ostype & os, unsigned char usc ) { 67 67 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 68 fmt( os, "% 'hhu", usc );68 fmt( os, "%hhu", usc ); 69 69 return os; 70 70 } // ?|? … … 75 75 ostype & ?|?( ostype & os, short int si ) { 76 76 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 77 fmt( os, "% 'hd", si );77 fmt( os, "%hd", si ); 78 78 return os; 79 79 } // ?|? … … 84 84 ostype & ?|?( ostype & os, unsigned short int usi ) { 85 85 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 86 fmt( os, "% 'hu", usi );86 fmt( os, "%hu", usi ); 87 87 return os; 88 88 } // ?|? … … 93 93 ostype & ?|?( ostype & os, int i ) { 94 94 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 95 fmt( os, "% 'd", i );95 fmt( os, "%d", i ); 96 96 return os; 97 97 } // ?|? … … 102 102 ostype & ?|?( ostype & os, unsigned int ui ) { 103 103 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 104 fmt( os, "% 'u", ui );104 fmt( os, "%u", ui ); 105 105 return os; 106 106 } // ?|? … … 111 111 ostype & ?|?( ostype & os, long int li ) { 112 112 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 113 fmt( os, "% 'ld", li );113 fmt( os, "%ld", li ); 114 114 return os; 115 115 } // ?|? … … 120 120 ostype & ?|?( ostype & os, unsigned long int uli ) { 121 121 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 122 fmt( os, "% 'lu", uli );122 fmt( os, "%lu", uli ); 123 123 return os; 124 124 } // ?|? … … 129 129 ostype & ?|?( ostype & os, long long int lli ) { 130 130 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 131 fmt( os, "% 'lld", lli );131 fmt( os, "%lld", lli ); 132 132 return os; 133 133 } // ?|? … … 138 138 ostype & ?|?( ostype & os, unsigned long long int ulli ) { 139 139 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 140 fmt( os, "% 'llu", ulli );140 fmt( os, "%llu", ulli ); 141 141 return os; 142 142 } // ?|? … … 205 205 ostype & ?|?( ostype & os, float f ) { 206 206 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 207 PrintWithDP( os, "% 'g", f );207 PrintWithDP( os, "%g", f ); 208 208 return os; 209 209 } // ?|? … … 214 214 ostype & ?|?( ostype & os, double d ) { 215 215 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 216 PrintWithDP( os, "% '.*lg", d, DBL_DIG );216 PrintWithDP( os, "%.*lg", d, DBL_DIG ); 217 217 return os; 218 218 } // ?|? … … 223 223 ostype & ?|?( ostype & os, long double ld ) { 224 224 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 225 PrintWithDP( os, "% '.*Lg", ld, LDBL_DIG );225 PrintWithDP( os, "%.*Lg", ld, LDBL_DIG ); 226 226 return os; 227 227 } // ?|? … … 233 233 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 234 234 // os | crealf( fc ) | nonl; 235 PrintWithDP( os, "% 'g", crealf( fc ) );236 PrintWithDP( os, "% '+g", cimagf( fc ) );235 PrintWithDP( os, "%g", crealf( fc ) ); 236 PrintWithDP( os, "%+g", cimagf( fc ) ); 237 237 fmt( os, "i" ); 238 238 return os; … … 245 245 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 246 246 // os | creal( dc ) | nonl; 247 PrintWithDP( os, "% '.*lg", creal( dc ), DBL_DIG );248 PrintWithDP( os, "% '+.*lg", cimag( dc ), DBL_DIG );247 PrintWithDP( os, "%.*lg", creal( dc ), DBL_DIG ); 248 PrintWithDP( os, "%+.*lg", cimag( dc ), DBL_DIG ); 249 249 fmt( os, "i" ); 250 250 return os; … … 257 257 if ( sepPrt$( os ) ) fmt( os, "%s", sepGetCur$( os ) ); 258 258 // os | creall( ldc ) || nonl; 259 PrintWithDP( os, "% '.*Lg", creall( ldc ), LDBL_DIG );260 PrintWithDP( os, "% '+.*Lg", cimagl( ldc ), LDBL_DIG );259 PrintWithDP( os, "%.*Lg", creall( ldc ), LDBL_DIG ); 260 PrintWithDP( os, "%+.*Lg", cimagl( ldc ), LDBL_DIG ); 261 261 fmt( os, "i" ); 262 262 return os; … … 282 282 }; // mask 283 283 284 if ( s == 0p ) { fmt( os, "%s", "0p" ); return os; } // null pointer285 284 if ( s[0] == '\0' ) { sepOff( os ); return os; } // null string => no separator 286 285 … … 497 496 if ( ! f.flags.pc ) memcpy( &fmtstr, IFMTNP, sizeof(IFMTNP) ); \ 498 497 else memcpy( &fmtstr, IFMTP, sizeof(IFMTP) ); \ 499 int star = 5; /* position before first '*' */ \498 int star = 4; /* position before first '*' */ \ 500 499 \ 501 500 /* Insert flags into spaces before '*', from right to left. */ \ … … 504 503 if ( f.flags.sign ) { fmtstr[star] = '+'; star -= 1; } \ 505 504 if ( f.flags.pad0 && ! f.flags.pc ) { fmtstr[star] = '0'; star -= 1; } \ 506 fmtstr[star] = '\''; star -= 1; /* locale */ \507 505 fmtstr[star] = '%'; \ 508 506 \ … … 523 521 } // distribution 524 522 525 IntegralFMTImpl( signed char, " *hh ", "*.*hh " )526 IntegralFMTImpl( unsigned char, " *hh ", "*.*hh " )527 IntegralFMTImpl( signed short int, " *h ", "*.*h " )528 IntegralFMTImpl( unsigned short int, " *h ", "*.*h " )529 IntegralFMTImpl( signed int, " * ", "*.* " )530 IntegralFMTImpl( unsigned int, " * ", "*.* " )531 IntegralFMTImpl( signed long int, " *l ", "*.*l " )532 IntegralFMTImpl( unsigned long int, " *l ", "*.*l " )533 IntegralFMTImpl( signed long long int, " *ll ", "*.*ll " )534 IntegralFMTImpl( unsigned long long int, " *ll ", "*.*ll " )523 IntegralFMTImpl( signed char, " *hh ", " *.*hh " ) 524 IntegralFMTImpl( unsigned char, " *hh ", " *.*hh " ) 525 IntegralFMTImpl( signed short int, " *h ", " *.*h " ) 526 IntegralFMTImpl( unsigned short int, " *h ", " *.*h " ) 527 IntegralFMTImpl( signed int, " * ", " *.* " ) 528 IntegralFMTImpl( unsigned int, " * ", " *.* " ) 529 IntegralFMTImpl( signed long int, " *l ", " *.*l " ) 530 IntegralFMTImpl( unsigned long int, " *l ", " *.*l " ) 531 IntegralFMTImpl( signed long long int, " *ll ", " *.*ll " ) 532 IntegralFMTImpl( unsigned long long int, " *ll ", " *.*ll " ) 535 533 536 534 … … 694 692 if ( ! f.flags.pc ) memcpy( &fmtstr, DFMTNP, sizeof(DFMTNP) ); \ 695 693 else memcpy( &fmtstr, DFMTP, sizeof(DFMTP) ); \ 696 int star = 5; /* position before first '*' */ \694 int star = 4; /* position before first '*' */ \ 697 695 \ 698 696 /* Insert flags into spaces before '*', from right to left. */ \ … … 700 698 if ( f.flags.sign ) { fmtstr[star] = '+'; star -= 1; } \ 701 699 if ( f.flags.pad0 ) { fmtstr[star] = '0'; star -= 1; } \ 702 fmtstr[star] = '\''; star -= 1; /* locale */ \703 700 fmtstr[star] = '%'; \ 704 701 \ … … 718 715 } // distribution 719 716 720 FloatingPointFMTImpl( double, " * ", "*.* " )721 FloatingPointFMTImpl( long double, " *L ", "*.*L " )717 FloatingPointFMTImpl( double, " * ", " *.* " ) 718 FloatingPointFMTImpl( long double, " *L ", " *.*L " ) 722 719 723 720 // *********************************** character *********************************** -
libcfa/src/parseconfig.cfa
r97fed44 rf57f6ea0 1 2 3 #pragma GCC diagnostic push4 //#pragma GCC diagnostic ignored "-Wunused-parameter"5 //#pragma GCC diagnostic ignored "-Wunused-function"6 //#pragma GCC diagnostic ignored "-Wuninitialized"7 //#pragma GCC diagnostic ignored "-Wmaybe-uninitialized"8 9 1 #include <fstream.hfa> 10 2 #include <parseargs.hfa> … … 27 19 // TODO: use string interface when it's ready (and implement exception msg protocol) 28 20 [ void ] msg( * Missing_Config_Entries ex ) { 29 serr | "The config file is missing " | ex->num_missing | "entr" | sepOff | (ex->num_missing == 1 ? "y." : "ies."); 21 serr | nlOff; 22 serr | "The config file is missing " | ex->num_missing; 23 serr | nlOn; 24 if ( ex->num_missing == 1 ) { 25 serr | " entry."; 26 } else { 27 serr | " entries."; 28 } 30 29 } // msg 31 30 … … 224 223 return value < zero_val; 225 224 } 226 #pragma GCC diagnostic pop227 225 228 226 -
libcfa/src/startup.cfa
r97fed44 rf57f6ea0 10 10 // Created On : Tue Jul 24 16:21:57 2018 11 11 // Last Modified By : Peter A. Buhr 12 // Last Modified On : Mon Jan 17 16:41:54 202213 // Update Count : 5 512 // Last Modified On : Wed Jan 12 18:51:24 2022 13 // Update Count : 51 14 14 // 15 15 … … 17 17 #include <locale.h> // setlocale 18 18 #include <stdlib.h> // getenv 19 #include "bits/defs.hfa" // rdtscl20 19 #include "startup.hfa" 20 #include "bits/defs.hfa" 21 21 22 extern uint32_t __global_random_seed; // sequential/concurrent 23 extern uint32_t __global_random_state; // sequential 22 extern uint32_t __global_random_seed, __global_random_state; 24 23 25 24 extern "C" { … … 27 26 void __cfaabi_appready_startup( void ) { 28 27 tzset(); // initialize time global variables 28 setlocale( LC_NUMERIC, getenv("LANG") ); 29 29 #ifdef __CFA_DEBUG__ 30 30 extern void heapAppStart(); -
libcfa/src/stdlib.cfa
r97fed44 rf57f6ea0 10 10 // Created On : Thu Jan 28 17:10:29 2016 11 11 // Last Modified By : Peter A. Buhr 12 // Last Modified On : Thu Jan 13 21:38:30202213 // Update Count : 5 9312 // Last Modified On : Wed Jan 12 18:52:41 2022 13 // Update Count : 582 14 14 // 15 15 16 16 #include "stdlib.hfa" 17 #include "bits/random.hfa"17 //#include "concurrency/kernel/fwd.hfa" 18 18 #include "concurrency/invoke.h" // random_state 19 19 … … 223 223 //--------------------------------------- 224 224 225 // Pipelined to allow OoO overlap with reduced dependencies. Critically, return the current value, and compute and store 226 // the next value. 227 225 228 #define GENERATOR LCG 226 229 230 static inline uint32_t MarsagliaXor( uint32_t & state ) { 231 uint32_t ret = state; 232 state ^= state << 6; 233 state ^= state >> 21; 234 state ^= state << 7; 235 return ret; 236 } // MarsagliaXor 237 238 static inline uint32_t LCG( uint32_t & state ) { // linear congruential generator 239 uint32_t ret = state; 240 state = 36969 * (state & 65535) + (state >> 16); // 36969 is NOT prime! No not change it! 241 return ret; 242 } // LCG 243 227 244 uint32_t __global_random_seed; // sequential/concurrent 228 uint32_t __global_random_state; // sequential only245 uint32_t __global_random_state; // sequential only 229 246 230 247 void set_seed( PRNG & prng, uint32_t seed_ ) with( prng ) { state = seed = seed_; GENERATOR( state ); } // set seed -
libcfa/src/stdlib.hfa
r97fed44 rf57f6ea0 10 10 // Created On : Thu Jan 28 17:12:35 2016 11 11 // Last Modified By : Peter A. Buhr 12 // Last Modified On : Thu Jan 13 21:34:46202213 // Update Count : 6 3612 // Last Modified On : Wed Jan 12 18:56:13 2022 13 // Update Count : 621 14 14 // 15 15 … … 208 208 209 209 forall( TT... | { T * alloc_internal$( void *, T *, size_t, size_t, S_fill(T), TT ); } ) { 210 210 211 T * alloc_internal$( void * , T * Realloc, size_t Align, size_t Dim, S_fill(T) Fill, T_resize Resize, TT rest) { 211 212 return alloc_internal$( Resize, (T*)0p, Align, Dim, Fill, rest); … … 231 232 return alloc_internal$( (void*)0p, (T*)0p, (_Alignof(T) > libAlign() ? _Alignof(T) : libAlign()), dim, (S_fill(T)){'0'}, all); 232 233 } 234 233 235 } // distribution TT 234 236 } // distribution T -
src/AST/Decl.cpp
r97fed44 rf57f6ea0 26 26 #include "Node.hpp" // for readonly 27 27 #include "Type.hpp" // for readonly 28 #include "Expr.hpp"29 28 30 29 namespace ast { … … 66 65 for (auto & tp : this->type_params) { 67 66 ftype->forall.emplace_back(new TypeInstType(tp->name, tp)); 68 for (auto & ap: tp->assertions) {69 ftype->assertions.emplace_back(new VariableExpr(loc, ap));70 }71 67 } 72 68 this->type = ftype; -
src/AST/Decl.hpp
r97fed44 rf57f6ea0 34 34 // Must be included in *all* AST classes; should be #undef'd at the end of the file 35 35 #define MUTATE_FRIEND \ 36 36 template<typename node_t> friend node_t * mutate(const node_t * node); \ 37 37 template<typename node_t> friend node_t * shallowCopy(const node_t * node); 38 38 … … 135 135 std::vector< ptr<Expr> > withExprs; 136 136 137 137 138 FunctionDecl( const CodeLocation & loc, const std::string & name, std::vector<ptr<TypeDecl>>&& forall, 138 139 std::vector<ptr<DeclWithType>>&& params, std::vector<ptr<DeclWithType>>&& returns, -
src/AST/Eval.hpp
r97fed44 rf57f6ea0 24 24 template< typename... Args > 25 25 UntypedExpr * call( const CodeLocation & loc, const std::string & name, Args &&... args ) { 26 return new UntypedExpr { 27 loc, new NameExpr { loc, name }, 26 return new UntypedExpr { 27 loc, new NameExpr { loc, name }, 28 28 std::vector< ptr< Expr > > { std::forward< Args >( args )... } }; 29 29 } -
src/AST/Print.cpp
r97fed44 rf57f6ea0 333 333 print( node->funcSpec ); 334 334 335 336 337 if ( node->type && node->isTypeFixed ) { 335 if ( node->type ) { 338 336 node->type->accept( *this ); 339 337 } else { 340 if (!node->type_params.empty()) { 341 os << "forall" << endl; 342 ++indent; 343 printAll(node->type_params); 344 os << indent; 345 --indent; 346 347 if (!node->assertions.empty()) { 348 os << "with assertions" << endl; 349 ++indent; 350 printAll(node->assertions); 351 os << indent; 352 --indent; 353 } 354 } 355 356 os << "function" << endl; 357 if ( ! node->params.empty() ) { 358 os << indent << "... with parameters" << endl; 359 ++indent; 360 printAll( node->params ); 361 if ( node->type->isVarArgs ) { 362 os << indent << "and a variable number of other arguments" << endl; 363 } 364 --indent; 365 } else if ( node->type->isVarArgs ) { 366 os << indent+1 << "accepting unspecified arguments" << endl; 367 } 368 369 os << indent << "... returning"; 370 if ( node->returns.empty() ) { 371 os << " nothing" << endl; 372 } else { 373 os << endl; 374 ++indent; 375 printAll( node->returns ); 376 --indent; 377 } 338 os << "untyped entity"; 378 339 } 379 340 -
src/Validate/InitializerLength.cpp
r97fed44 rf57f6ea0 14 14 // 15 15 16 #include "InitializerLength.hpp"16 //#include "InitializerLength.hpp" 17 17 18 18 #include "AST/Expr.hpp" -
src/Validate/InitializerLength.hpp
r97fed44 rf57f6ea0 14 14 // 15 15 16 #pragma once17 18 namespace ast {19 class TranslationUnit;20 }21 22 16 namespace Validate { 23 17 -
tests/concurrent/.expect/ctor-check.txt
r97fed44 rf57f6ea0 2 2 ?{}: function 3 3 ... with parameters 4 this:lvalue reference to instance of struct Empty with body4 lvalue reference to instance of struct Empty with body 5 5 ... returning nothing 6 6 with body -
tests/concurrent/preempt.cfa
r97fed44 rf57f6ea0 1 #include <fstream.hfa>2 1 #include <kernel.hfa> 3 2 #include <thread.hfa> … … 26 25 thread worker_t { 27 26 int value; 28 unsigned spin;29 27 }; 30 28 31 29 void ?{}( worker_t & this, int value ) { 32 30 this.value = value; 33 this.spin = 0;34 31 } 35 32 36 33 void main(worker_t & this) { 37 34 while(TEST(counter < N)) { 38 if(this.spin > 100_000_000) abort | "Worker" | this.value | "has been spinning too long! (" | this.spin | ")";39 35 __cfaabi_check_preemption(); 40 36 if( (counter % 7) == this.value ) { … … 44 40 if( (next % 100) == 0 ) printf("%d\n", (int)next); 45 41 __cfaabi_check_preemption(); 46 this.spin = 0;47 42 } 48 43 __cfaabi_check_preemption(); 49 44 KICK_WATCHDOG; 50 this.spin++;51 45 } 52 46 } -
tests/device/cpu.cfa
r97fed44 rf57f6ea0 15 15 16 16 17 #include <fstream.hfa> 17 18 #include <device/cpu.hfa> 18 #include <limits.hfa>19 #include <fstream.hfa>20 19 #include <stdlib.hfa> 21 20 … … 119 118 120 119 unsigned found_level = 0; 121 unsigned found = MAX;120 unsigned found = -1u; 122 121 for(i; idxs) { 123 122 unsigned idx = idxs - 1 - i; … … 137 136 } 138 137 139 /* paranoid */ verify(found != MAX);138 /* paranoid */ verify(found != -1u); 140 139 return found; 141 140 } -
tests/io/io-acquire-in.cfa
r97fed44 rf57f6ea0 10 10 // Created On : Mon Mar 1 18:40:09 2021 11 11 // Last Modified By : Peter A. Buhr 12 // Last Modified On : Fri Jan 14 09:13:43 202213 // Update Count : 7 312 // Last Modified On : Wed Oct 6 18:04:58 2021 13 // Update Count : 72 14 14 // 15 15 … … 17 17 #include <thread.hfa> 18 18 #include <mutex_stmt.hfa> 19 20 Duration default_preemption() { return 0; }21 19 22 20 // above output used as input to parallel threads -
tests/io/io-acquire-no-io.cfa
r97fed44 rf57f6ea0 10 10 // Created On : Mon Mar 1 18:40:09 2021 11 11 // Last Modified By : Peter A. Buhr 12 // Last Modified On : Fri Jan 14 09:13:54 202213 // Update Count : 7 312 // Last Modified On : Wed Oct 6 18:04:58 2021 13 // Update Count : 72 14 14 // 15 15 … … 17 17 #include <thread.hfa> 18 18 #include <mutex_stmt.hfa> 19 20 Duration default_preemption() { return 0; }21 19 22 20 multiple_acquisition_lock soutLock, sinLock; -
tests/io/io-acquire-out.cfa
r97fed44 rf57f6ea0 10 10 // Created On : Mon Mar 1 18:40:09 2021 11 11 // Last Modified By : Peter A. Buhr 12 // Last Modified On : Fri Jan 14 09:14:06 202213 // Update Count : 7 312 // Last Modified On : Wed Oct 6 18:04:58 2021 13 // Update Count : 72 14 14 // 15 15 … … 17 17 #include <thread.hfa> 18 18 #include <mutex_stmt.hfa> 19 20 Duration default_preemption() { return 0; }21 19 22 20 thread T {}; -
tests/io/io-acquire.cfa
r97fed44 rf57f6ea0 10 10 // Created On : Mon Mar 1 18:40:09 2021 11 11 // Last Modified By : Peter A. Buhr 12 // Last Modified On : Fri Jan 14 09:13:18202213 // Update Count : 7 412 // Last Modified On : Mon Jan 10 07:57:12 2022 13 // Update Count : 73 14 14 // 15 15 … … 17 17 #include <thread.hfa> 18 18 #include <mutex_stmt.hfa> 19 20 Duration default_preemption() { return 0; }21 19 22 20 thread T {}; -
tests/unified_locking/.expect/locks.txt
r97fed44 rf57f6ea0 11 11 Start Test 6: owner lock and condition variable 3 wait/notify all 12 12 Done Test 6 13 Start Test 7: linear backofflock and condition variable single wait/notify13 Start Test 7: fast lock and condition variable single wait/notify 14 14 Done Test 7 15 Start Test 8: linear backofflock and condition variable 3 wait/notify all15 Start Test 8: fast lock and condition variable 3 wait/notify all 16 16 Done Test 8 17 Start Test 9: multi acquisiton lock and condition variable multiple acquire andwait/notify17 Start Test 9: linear backoff lock and condition variable single wait/notify 18 18 Done Test 9 19 Start Test 10: owner lock and condition variable multiple acquire and wait/notify19 Start Test 10: linear backoff lock and condition variable 3 wait/notify all 20 20 Done Test 10 21 Start Test 11: no lock condition variablewait/notify21 Start Test 11: multi acquisiton lock and condition variable multiple acquire and wait/notify 22 22 Done Test 11 23 Start Test 12: locked condition variable wait/notify with front()23 Start Test 12: owner lock and condition variable multiple acquire and wait/notify 24 24 Done Test 12 25 Start Test 13: no lock condition variable wait/notify 26 Done Test 13 27 Start Test 14: locked condition variable wait/notify with front() 28 Done Test 14 -
tests/unified_locking/locks.cfa
r97fed44 rf57f6ea0 15 15 condition_variable( owner_lock ) c_o; 16 16 17 fast_lock f; 18 condition_variable( fast_lock ) c_f; 19 17 20 linear_backoff_then_block_lock l; 18 21 condition_variable( linear_backoff_then_block_lock ) c_l; … … 71 74 } 72 75 unlock(s); 76 } 77 } 78 79 thread T_C_F_WS1 {}; 80 81 void main( T_C_F_WS1 & this ) { 82 for (unsigned int i = 0; i < num_times; i++) { 83 lock(f); 84 if(empty(c_f) && i != num_times - 1) { 85 wait(c_f,f); 86 }else{ 87 notify_one(c_f); 88 } 89 unlock(f); 90 } 91 } 92 93 thread T_C_F_WB1 {}; 94 95 void main( T_C_F_WB1 & this ) { 96 for (unsigned int i = 0; i < num_times; i++) { 97 lock(f); 98 if(counter(c_f) == 3 || i == num_times - 1) { 99 notify_all(c_f); 100 }else{ 101 wait(c_f,f); 102 } 103 unlock(f); 73 104 } 74 105 } … … 286 317 printf("Done Test 6\n"); 287 318 288 printf("Start Test 7: linear backoff lock and condition variable single wait/notify\n"); 319 printf("Start Test 7: fast lock and condition variable single wait/notify\n"); 320 { 321 T_C_F_WS1 t1[2]; 322 } 323 printf("Done Test 7\n"); 324 325 printf("Start Test 8: fast lock and condition variable 3 wait/notify all\n"); 326 { 327 T_C_F_WB1 t1[4]; 328 } 329 printf("Done Test 8\n"); 330 331 printf("Start Test 9: linear backoff lock and condition variable single wait/notify\n"); 289 332 { 290 333 T_C_L_WS1 t1[2]; 291 334 } 292 printf("Done Test 7\n");293 294 printf("Start Test 8: linear backoff lock and condition variable 3 wait/notify all\n");335 printf("Done Test 9\n"); 336 337 printf("Start Test 10: linear backoff lock and condition variable 3 wait/notify all\n"); 295 338 { 296 339 T_C_L_WB1 t1[4]; 297 340 } 298 printf("Done Test 8\n");299 300 printf("Start Test 9: multi acquisiton lock and condition variable multiple acquire and wait/notify\n");341 printf("Done Test 10\n"); 342 343 printf("Start Test 11: multi acquisiton lock and condition variable multiple acquire and wait/notify\n"); 301 344 { 302 345 T_C_M_WS2 t1[2]; 303 346 } 304 printf("Done Test 9\n");305 306 printf("Start Test 1 0: owner lock and condition variable multiple acquire and wait/notify\n");347 printf("Done Test 11\n"); 348 349 printf("Start Test 12: owner lock and condition variable multiple acquire and wait/notify\n"); 307 350 { 308 351 T_C_O_WS2 t1[2]; 309 352 } 310 printf("Done Test 1 0\n");311 312 printf("Start Test 1 1: no lock condition variable wait/notify\n");353 printf("Done Test 12\n"); 354 355 printf("Start Test 13: no lock condition variable wait/notify\n"); 313 356 { 314 357 T_C_NLW t1; 315 358 T_C_NLS t2; 316 359 } 317 printf("Done Test 1 1\n");318 319 printf("Start Test 1 2: locked condition variable wait/notify with front()\n");360 printf("Done Test 13\n"); 361 362 printf("Start Test 14: locked condition variable wait/notify with front()\n"); 320 363 { 321 364 T_C_S_WNF t1[2]; 322 365 } 323 printf("Done Test 1 2\n");324 } 366 printf("Done Test 14\n"); 367 }
Note:
See TracChangeset
for help on using the changeset viewer.
