Fine tune OpenMP some more

libc/BUILD.mk

@@ -142,6 +142,9 @@ libc/isystem/nsync_once.h \
 libc/isystem/nsync_time.h \
 libc/isystem/nsync_waiter.h \
 libc/isystem/numeric \
+libc/isystem/omp-tools.h \
+libc/isystem/omp.h \
+libc/isystem/ompx.h \
 libc/isystem/optional \
 libc/isystem/ostream \
 libc/isystem/paths.h \
@@ -274,6 +277,8 @@ LIBC_INCS = $(filter %.inc,$(LIBC_FILES))
 LIBC_CHECKS = $(LIBC_HDRS_H:%=o/$(MODE)/%.ok)
 LIBC_FILES := $(wildcard libc/*)
 
+o/$(MODE)/libc/isystem/ompx.h.ok: private CPPFLAGS += -Wno-unknown-pragmas
+
 .PHONY:		o/$(MODE)/libc
 o/$(MODE)/libc:	o/$(MODE)/libc/calls		\
 		o/$(MODE)/libc/crt		\

libc/isystem/omp-tools.h

@@ -0,0 +1,3 @@
+#ifndef _OMP_TOOLS_H
+#include "third_party/openmp/omp-tools.h"
+#endif /* _OMP_TOOLS_H */

libc/isystem/omp.h

@@ -0,0 +1,3 @@
+#ifndef _OMP_H
+#include "third_party/openmp/omp.h"
+#endif /* _OMP_H */

libc/isystem/ompx.h

@@ -0,0 +1,3 @@
+#ifndef _OMPX_H
+#include "third_party/openmp/ompx.h"
+#endif /* _OMPX_H */

third_party/openmp/BUILD.mk

@@ -58,10 +58,11 @@ $(THIRD_PARTY_OPENMP_A_OBJS): private				\
 			-ffunction-sections			\
 			-Wno-maybe-uninitialized		\
 			-Wno-stringop-truncation		\
-			-Wno-class-memaccess			\
 			-Wno-unused-but-set-variable		\
+			-Wno-class-memaccess			\
+			-fno-strict-aliasing			\
 			-Wno-frame-address			\
-			-fno-strict-aliasing
+			-Wno-sign-compare
 
 o/$(MODE)/third_party/openmp/util1.o: private COPTS += -fportcosmo
 

third_party/openmp/README.cosmo

@@ -13,5 +13,8 @@ LOCAL CHANGES
   - Use Cosmo's gettid() function
   - Ran third_party/openmp/generate.sh
   - Removed usage of syscall() function
+  - Only enable hidden helper on GNU/Systemd
+  - Made __kmp_get_load_balance() portable at runtime
+  - Made __kmp_affinity_get_offline_cpus() portable at runtime
   - Turned off quad floating point support (why does openmp have it?)
   - Remove bloat for checking if multiple OpenMP libraries are linked

third_party/openmp/kmp_affinity.cpp

@@ -1547,7 +1547,6 @@ kmp_affin_mask_t *__kmp_affinity_get_offline_cpus() {
   kmp_affin_mask_t *offline;
   KMP_CPU_ALLOC(offline);
   KMP_CPU_ZERO(offline);
-#if KMP_OS_LINUX
   int n, begin_cpu, end_cpu;
   kmp_safe_raii_file_t offline_file;
   auto skip_ws = [](FILE *f) {
@@ -1595,7 +1594,6 @@ kmp_affin_mask_t *__kmp_affinity_get_offline_cpus() {
       KMP_CPU_SET(cpu, offline);
     }
   }
-#endif
   return offline;
 }
 

third_party/openmp/kmp_config.h

@@ -22,7 +22,6 @@
 #define KMP_DEBUG 1
 #endif
 
-#define KMP_USE_FUTEX 0
 #define KMP_FTN_ENTRIES KMP_FTN_PLAIN
 #define syscall {{openmp_shall_not_use_syscall}}
 
@@ -63,9 +62,9 @@
 #define OMPX_TASKGRAPH LIBOMP_OMPX_TASKGRAPH
 #define LIBOMP_PROFILING_SUPPORT 0
 #define OMP_PROFILING_SUPPORT LIBOMP_PROFILING_SUPPORT
-#define LIBOMP_OMPT_OPTIONAL 0
+#define LIBOMP_OMPT_OPTIONAL 1
 #define OMPT_OPTIONAL LIBOMP_OMPT_OPTIONAL
-#define LIBOMP_USE_ADAPTIVE_LOCKS 0
+#define LIBOMP_USE_ADAPTIVE_LOCKS 1
 #define KMP_USE_ADAPTIVE_LOCKS LIBOMP_USE_ADAPTIVE_LOCKS
 #define KMP_DEBUG_ADAPTIVE_LOCKS 0
 #define LIBOMP_USE_INTERNODE_ALIGNMENT 0

third_party/openmp/kmp_ftn_stdcall.cpp

@@ -1,32 +0,0 @@
-/*
- * kmp_ftn_stdcall.cpp -- Fortran __stdcall linkage support for OpenMP.
- */
-
-//===----------------------------------------------------------------------===//
-//
-// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-// See https://llvm.org/LICENSE.txt for license information.
-// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-//
-//===----------------------------------------------------------------------===//
-
-#include "kmp.h"
-
-// Note: This string is not printed when KMP_VERSION=1.
-char const __kmp_version_ftnstdcall[] =
-    KMP_VERSION_PREFIX "Fortran __stdcall OMP support: "
-#ifdef USE_FTN_STDCALL
-                       "yes";
-#else
-                       "no";
-#endif
-
-#ifdef USE_FTN_STDCALL
-
-#define FTN_STDCALL KMP_STDCALL
-#define KMP_FTN_ENTRIES USE_FTN_STDCALL
-
-#include "kmp_ftn_entry.h"
-#include "kmp_ftn_os.h"
-
-#endif /* USE_FTN_STDCALL */

third_party/openmp/kmp_lock.cpp

@@ -22,6 +22,9 @@
 #include "kmp_wrapper_getpid.h"
 
 #if KMP_USE_FUTEX
+#ifdef __COSMOPOLITAN__
+#include "third_party/nsync/futex.internal.h"
+#else
 #include <sys/syscall.h>
 #include <unistd.h>
 // We should really include <futex.h>, but that causes compatibility problems on
@@ -36,6 +39,7 @@
 #define FUTEX_WAKE 1
 #endif
 #endif
+#endif
 
 /* Implement spin locks for internal library use.             */
 /* The algorithm implemented is Lamport's bakery lock [1974]. */
@@ -375,8 +379,12 @@ __kmp_acquire_futex_lock_timed_template(kmp_futex_lock_t *lck, kmp_int32 gtid) {
          lck, gtid, poll_val));
 
     long rc;
+#ifdef __COSMOPOLITAN__
+    if ((rc = nsync_futex_wait_(&(lck->lk.poll), poll_val, false, NULL)) != 0) {
+#else
     if ((rc = syscall(__NR_futex, &(lck->lk.poll), FUTEX_WAIT, poll_val, NULL,
                       NULL, 0)) != 0) {
+#endif
       KA_TRACE(1000, ("__kmp_acquire_futex_lock: lck:%p, T#%d futex_wait(0x%x) "
                       "failed (rc=%ld errno=%d)\n",
                       lck, gtid, poll_val, rc, errno));
@@ -453,8 +461,12 @@ int __kmp_release_futex_lock(kmp_futex_lock_t *lck, kmp_int32 gtid) {
     KA_TRACE(1000,
              ("__kmp_release_futex_lock: lck:%p, T#%d futex_wake 1 thread\n",
               lck, gtid));
+#ifdef __COSMOPOLITAN__
+    nsync_futex_wake_(&(lck->lk.poll), 1, false);
+#else
     syscall(__NR_futex, &(lck->lk.poll), FUTEX_WAKE, KMP_LOCK_BUSY(1, futex),
             NULL, NULL, 0);
+#endif
   }
 
   KMP_MB(); /* Flush all pending memory write invalidates.  */

third_party/openmp/kmp_runtime.cpp

@@ -9202,7 +9202,7 @@ kmp_info_t *__kmp_hidden_helper_main_thread;
 std::atomic<kmp_int32> __kmp_unexecuted_hidden_helper_tasks;
 #if KMP_OS_LINUX
 kmp_int32 __kmp_hidden_helper_threads_num = 8;
-kmp_int32 __kmp_enable_hidden_helper = TRUE;
+kmp_int32 __kmp_enable_hidden_helper = IsLinux();
 #else
 kmp_int32 __kmp_hidden_helper_threads_num = 0;
 kmp_int32 __kmp_enable_hidden_helper = FALSE;

third_party/openmp/util1.cpp

@@ -2236,15 +2236,12 @@ int __kmp_is_address_mapped(void *addr) {
 
 #ifdef USE_LOAD_BALANCE
 
-#if KMP_OS_DARWIN || KMP_OS_DRAGONFLY || KMP_OS_FREEBSD || KMP_OS_NETBSD ||    \
-    KMP_OS_OPENBSD || KMP_OS_SOLARIS
-
 // The function returns the rounded value of the system load average
 // during given time interval which depends on the value of
 // __kmp_load_balance_interval variable (default is 60 sec, other values
 // may be 300 sec or 900 sec).
 // It returns -1 in case of error.
-int __kmp_get_load_balance(int max) {
+static int __kmp_get_load_balance_getloadavg(int max) {
   double averages[3];
   int ret_avg = 0;
 
@@ -2268,13 +2265,11 @@ int __kmp_get_load_balance(int max) {
   return ret_avg;
 }
 
-#else // Linux* OS
-
 // The function returns number of running (not sleeping) threads, or -1 in case
 // of error. Error could be reported if Linux* OS kernel too old (without
 // "/proc" support). Counting running threads stops if max running threads
 // encountered.
-int __kmp_get_load_balance(int max) {
+static int __kmp_get_load_balance_linux(int max) {
   static int permanent_error = 0;
   static int glb_running_threads = 0; // Saved count of the running threads for
   // the thread balance algorithm
@@ -2495,9 +2490,16 @@ finish: // Clean up and exit.
 
   return running_threads;
 
-} // __kmp_get_load_balance
+} // __kmp_get_load_balance_linux
 
+int __kmp_get_load_balance(int max) {
+#if KMP_OS_DARWIN || KMP_OS_DRAGONFLY || KMP_OS_FREEBSD || KMP_OS_NETBSD ||    \
+    KMP_OS_OPENBSD || KMP_OS_SOLARIS
+  return __kmp_get_load_balance_getloadavg(max);
+#else // Linux* OS
+  return __kmp_get_load_balance_linux(max);
 #endif // KMP_OS_DARWIN
+} // __kmp_get_load_balance
 
 #endif // USE_LOAD_BALANCE