svn commit: r353954 - in vendor/llvm-openmp/dist/runtime/src: . thirdparty/ittnotify
Dimitry Andric
dim at FreeBSD.org
Wed Oct 23 17:53:19 UTC 2019
Author: dim
Date: Wed Oct 23 17:53:14 2019
New Revision: 353954
URL: https://svnweb.freebsd.org/changeset/base/353954
Log:
Vendor import of stripped LLVM openmp trunk r375505, the last commit
before the upstream Subversion repository was made read-only, and the
LLVM project migrated to GitHub:
https://llvm.org/svn/llvm-project/openmp/trunk@375505
Added:
vendor/llvm-openmp/dist/runtime/src/thirdparty/ittnotify/ittnotify_static.cpp
Deleted:
vendor/llvm-openmp/dist/runtime/src/kmp_taskq.cpp
vendor/llvm-openmp/dist/runtime/src/thirdparty/ittnotify/ittnotify_static.c
Modified:
vendor/llvm-openmp/dist/runtime/src/kmp.h
vendor/llvm-openmp/dist/runtime/src/kmp_affinity.cpp
vendor/llvm-openmp/dist/runtime/src/kmp_affinity.h
vendor/llvm-openmp/dist/runtime/src/kmp_csupport.cpp
vendor/llvm-openmp/dist/runtime/src/kmp_ftn_os.h
vendor/llvm-openmp/dist/runtime/src/kmp_global.cpp
vendor/llvm-openmp/dist/runtime/src/kmp_gsupport.cpp
vendor/llvm-openmp/dist/runtime/src/kmp_lock.cpp
vendor/llvm-openmp/dist/runtime/src/kmp_lock.h
vendor/llvm-openmp/dist/runtime/src/kmp_os.h
vendor/llvm-openmp/dist/runtime/src/kmp_platform.h
vendor/llvm-openmp/dist/runtime/src/kmp_runtime.cpp
vendor/llvm-openmp/dist/runtime/src/kmp_stub.cpp
vendor/llvm-openmp/dist/runtime/src/kmp_taskdeps.cpp
vendor/llvm-openmp/dist/runtime/src/kmp_wait_release.h
vendor/llvm-openmp/dist/runtime/src/ompt-general.cpp
vendor/llvm-openmp/dist/runtime/src/ompt-internal.h
vendor/llvm-openmp/dist/runtime/src/ompt-specific.cpp
vendor/llvm-openmp/dist/runtime/src/ompt-specific.h
vendor/llvm-openmp/dist/runtime/src/thirdparty/ittnotify/ittnotify_config.h
vendor/llvm-openmp/dist/runtime/src/z_Linux_asm.S
vendor/llvm-openmp/dist/runtime/src/z_Linux_util.cpp
Modified: vendor/llvm-openmp/dist/runtime/src/kmp.h
==============================================================================
--- vendor/llvm-openmp/dist/runtime/src/kmp.h Wed Oct 23 17:53:12 2019 (r353953)
+++ vendor/llvm-openmp/dist/runtime/src/kmp.h Wed Oct 23 17:53:14 2019 (r353954)
@@ -2181,10 +2181,9 @@ struct kmp_dephash_entry {
typedef struct kmp_dephash {
kmp_dephash_entry_t **buckets;
size_t size;
-#ifdef KMP_DEBUG
+ size_t generation;
kmp_uint32 nelements;
kmp_uint32 nconflicts;
-#endif
} kmp_dephash_t;
typedef struct kmp_task_affinity_info {
@@ -3342,7 +3341,7 @@ extern int __kmp_aux_set_affinity_mask_proc(int proc,
extern int __kmp_aux_unset_affinity_mask_proc(int proc, void **mask);
extern int __kmp_aux_get_affinity_mask_proc(int proc, void **mask);
extern void __kmp_balanced_affinity(kmp_info_t *th, int team_size);
-#if KMP_OS_LINUX
+#if KMP_OS_LINUX || KMP_OS_FREEBSD
extern int kmp_set_thread_affinity_mask_initial(void);
#endif
#endif /* KMP_AFFINITY_SUPPORTED */
Modified: vendor/llvm-openmp/dist/runtime/src/kmp_affinity.cpp
==============================================================================
--- vendor/llvm-openmp/dist/runtime/src/kmp_affinity.cpp Wed Oct 23 17:53:12 2019 (r353953)
+++ vendor/llvm-openmp/dist/runtime/src/kmp_affinity.cpp Wed Oct 23 17:53:14 2019 (r353954)
@@ -1968,7 +1968,7 @@ static void __kmp_dispatch_set_hierarchy_values() {
__kmp_hier_max_units[kmp_hier_layer_e::LAYER_THREAD + 1] =
nPackages * nCoresPerPkg * __kmp_nThreadsPerCore;
__kmp_hier_max_units[kmp_hier_layer_e::LAYER_L1 + 1] = __kmp_ncores;
-#if KMP_ARCH_X86_64 && (KMP_OS_LINUX || KMP_OS_WINDOWS)
+#if KMP_ARCH_X86_64 && (KMP_OS_LINUX || KMP_OS_FREEBSD || KMP_OS_WINDOWS)
if (__kmp_mic_type >= mic3)
__kmp_hier_max_units[kmp_hier_layer_e::LAYER_L2 + 1] = __kmp_ncores / 2;
else
@@ -1982,7 +1982,7 @@ static void __kmp_dispatch_set_hierarchy_values() {
__kmp_hier_threads_per[kmp_hier_layer_e::LAYER_THREAD + 1] = 1;
__kmp_hier_threads_per[kmp_hier_layer_e::LAYER_L1 + 1] =
__kmp_nThreadsPerCore;
-#if KMP_ARCH_X86_64 && (KMP_OS_LINUX || KMP_OS_WINDOWS)
+#if KMP_ARCH_X86_64 && (KMP_OS_LINUX || KMP_OS_FREEBSD || KMP_OS_WINDOWS)
if (__kmp_mic_type >= mic3)
__kmp_hier_threads_per[kmp_hier_layer_e::LAYER_L2 + 1] =
2 * __kmp_nThreadsPerCore;
Modified: vendor/llvm-openmp/dist/runtime/src/kmp_affinity.h
==============================================================================
--- vendor/llvm-openmp/dist/runtime/src/kmp_affinity.h Wed Oct 23 17:53:12 2019 (r353953)
+++ vendor/llvm-openmp/dist/runtime/src/kmp_affinity.h Wed Oct 23 17:53:14 2019 (r353954)
@@ -160,6 +160,7 @@ class KMPHwlocAffinity : public KMPAffinity { (public)
};
#endif /* KMP_USE_HWLOC */
+#if KMP_OS_LINUX || KMP_OS_FREEBSD
#if KMP_OS_LINUX
/* On some of the older OS's that we build on, these constants aren't present
in <asm/unistd.h> #included from <sys.syscall.h>. They must be the same on
@@ -234,6 +235,10 @@ class KMPHwlocAffinity : public KMPAffinity { (public)
#endif /* __NR_sched_getaffinity */
#error Unknown or unsupported architecture
#endif /* KMP_ARCH_* */
+#elif KMP_OS_FREEBSD
+#include <pthread.h>
+#include <pthread_np.h>
+#endif
class KMPNativeAffinity : public KMPAffinity {
class Mask : public KMPAffinity::Mask {
typedef unsigned char mask_t;
@@ -294,8 +299,13 @@ class KMPNativeAffinity : public KMPAffinity {
int get_system_affinity(bool abort_on_error) override {
KMP_ASSERT2(KMP_AFFINITY_CAPABLE(),
"Illegal get affinity operation when not capable");
+#if KMP_OS_LINUX
int retval =
syscall(__NR_sched_getaffinity, 0, __kmp_affin_mask_size, mask);
+#elif KMP_OS_FREEBSD
+ int retval =
+ pthread_getaffinity_np(pthread_self(), __kmp_affin_mask_size, reinterpret_cast<cpuset_t *>(mask));
+#endif
if (retval >= 0) {
return 0;
}
@@ -308,8 +318,13 @@ class KMPNativeAffinity : public KMPAffinity {
int set_system_affinity(bool abort_on_error) const override {
KMP_ASSERT2(KMP_AFFINITY_CAPABLE(),
"Illegal get affinity operation when not capable");
+#if KMP_OS_LINUX
int retval =
syscall(__NR_sched_setaffinity, 0, __kmp_affin_mask_size, mask);
+#elif KMP_OS_FREEBSD
+ int retval =
+ pthread_setaffinity_np(pthread_self(), __kmp_affin_mask_size, reinterpret_cast<cpuset_t *>(mask));
+#endif
if (retval >= 0) {
return 0;
}
@@ -347,7 +362,7 @@ class KMPNativeAffinity : public KMPAffinity {
}
api_type get_api_type() const override { return NATIVE_OS; }
};
-#endif /* KMP_OS_LINUX */
+#endif /* KMP_OS_LINUX || KMP_OS_FREEBSD */
#if KMP_OS_WINDOWS
class KMPNativeAffinity : public KMPAffinity {
Modified: vendor/llvm-openmp/dist/runtime/src/kmp_csupport.cpp
==============================================================================
--- vendor/llvm-openmp/dist/runtime/src/kmp_csupport.cpp Wed Oct 23 17:53:12 2019 (r353953)
+++ vendor/llvm-openmp/dist/runtime/src/kmp_csupport.cpp Wed Oct 23 17:53:14 2019 (r353954)
@@ -545,7 +545,8 @@ void __kmpc_end_serialized_parallel(ident_t *loc, kmp_
if (ompt_enabled.ompt_callback_parallel_end) {
ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
&(serial_team->t.ompt_team_info.parallel_data), parent_task_data,
- ompt_parallel_invoker_program, OMPT_LOAD_RETURN_ADDRESS(global_tid));
+ ompt_parallel_invoker_program | ompt_parallel_team,
+ OMPT_LOAD_RETURN_ADDRESS(global_tid));
}
__ompt_lw_taskteam_unlink(this_thr);
this_thr->th.ompt_thread_info.state = ompt_state_overhead;
@@ -676,7 +677,8 @@ void __kmpc_flush(ident_t *loc) {
#endif // KMP_COMPILER_ICC
}
#endif // KMP_MIC
-#elif (KMP_ARCH_ARM || KMP_ARCH_AARCH64 || KMP_ARCH_MIPS || KMP_ARCH_MIPS64)
+#elif (KMP_ARCH_ARM || KMP_ARCH_AARCH64 || KMP_ARCH_MIPS || KMP_ARCH_MIPS64 || \
+ KMP_ARCH_RISCV64)
// Nothing to see here move along
#elif KMP_ARCH_PPC64
// Nothing needed here (we have a real MB above).
Modified: vendor/llvm-openmp/dist/runtime/src/kmp_ftn_os.h
==============================================================================
--- vendor/llvm-openmp/dist/runtime/src/kmp_ftn_os.h Wed Oct 23 17:53:12 2019 (r353953)
+++ vendor/llvm-openmp/dist/runtime/src/kmp_ftn_os.h Wed Oct 23 17:53:14 2019 (r353954)
@@ -633,5 +633,25 @@
GOMP_loop_ull_doacross_guided_start
#define KMP_API_NAME_GOMP_LOOP_ULL_DOACROSS_RUNTIME_START \
GOMP_loop_ull_doacross_runtime_start
+#define KMP_API_NAME_GOMP_LOOP_NONMONOTONIC_DYNAMIC_NEXT \
+ GOMP_loop_nonmonotonic_dynamic_next
+#define KMP_API_NAME_GOMP_LOOP_NONMONOTONIC_DYNAMIC_START \
+ GOMP_loop_nonmonotonic_dynamic_start
+#define KMP_API_NAME_GOMP_LOOP_NONMONOTONIC_GUIDED_NEXT \
+ GOMP_loop_nonmonotonic_guided_next
+#define KMP_API_NAME_GOMP_LOOP_NONMONOTONIC_GUIDED_START \
+ GOMP_loop_nonmonotonic_guided_start
+#define KMP_API_NAME_GOMP_LOOP_ULL_NONMONOTONIC_DYNAMIC_NEXT \
+ GOMP_loop_ull_nonmonotonic_dynamic_next
+#define KMP_API_NAME_GOMP_LOOP_ULL_NONMONOTONIC_DYNAMIC_START \
+ GOMP_loop_ull_nonmonotonic_dynamic_start
+#define KMP_API_NAME_GOMP_LOOP_ULL_NONMONOTONIC_GUIDED_NEXT \
+ GOMP_loop_ull_nonmonotonic_guided_next
+#define KMP_API_NAME_GOMP_LOOP_ULL_NONMONOTONIC_GUIDED_START \
+ GOMP_loop_ull_nonmonotonic_guided_start
+#define KMP_API_NAME_GOMP_PARALLEL_LOOP_NONMONOTONIC_DYNAMIC \
+ GOMP_parallel_loop_nonmonotonic_dynamic
+#define KMP_API_NAME_GOMP_PARALLEL_LOOP_NONMONOTONIC_GUIDED \
+ GOMP_parallel_loop_nonmonotonic_guided
#endif /* KMP_FTN_OS_H */
Modified: vendor/llvm-openmp/dist/runtime/src/kmp_global.cpp
==============================================================================
--- vendor/llvm-openmp/dist/runtime/src/kmp_global.cpp Wed Oct 23 17:53:12 2019 (r353953)
+++ vendor/llvm-openmp/dist/runtime/src/kmp_global.cpp Wed Oct 23 17:53:14 2019 (r353954)
@@ -431,7 +431,7 @@ std::atomic<int> __kmp_thread_pool_active_nth = ATOMIC
/* -------------------------------------------------
* GLOBAL/ROOT STATE */
KMP_ALIGN_CACHE
-kmp_global_t __kmp_global = {{0}};
+kmp_global_t __kmp_global;
/* ----------------------------------------------- */
/* GLOBAL SYNCHRONIZATION LOCKS */
Modified: vendor/llvm-openmp/dist/runtime/src/kmp_gsupport.cpp
==============================================================================
--- vendor/llvm-openmp/dist/runtime/src/kmp_gsupport.cpp Wed Oct 23 17:53:12 2019 (r353953)
+++ vendor/llvm-openmp/dist/runtime/src/kmp_gsupport.cpp Wed Oct 23 17:53:14 2019 (r353954)
@@ -22,7 +22,7 @@ extern "C" {
#endif // __cplusplus
#define MKLOC(loc, routine) \
- static ident_t(loc) = {0, KMP_IDENT_KMPC, 0, 0, ";unknown;unknown;0;0;;"};
+ static ident_t loc = {0, KMP_IDENT_KMPC, 0, 0, ";unknown;unknown;0;0;;"};
#include "kmp_ftn_os.h"
@@ -622,10 +622,16 @@ LOOP_START(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_STAT
LOOP_NEXT(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_STATIC_NEXT), {})
LOOP_START(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_DYNAMIC_START),
kmp_sch_dynamic_chunked)
+LOOP_START(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_NONMONOTONIC_DYNAMIC_START),
+ kmp_sch_dynamic_chunked)
LOOP_NEXT(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_DYNAMIC_NEXT), {})
+LOOP_NEXT(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_NONMONOTONIC_DYNAMIC_NEXT), {})
LOOP_START(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_GUIDED_START),
kmp_sch_guided_chunked)
+LOOP_START(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_NONMONOTONIC_GUIDED_START),
+ kmp_sch_guided_chunked)
LOOP_NEXT(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_GUIDED_NEXT), {})
+LOOP_NEXT(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_NONMONOTONIC_GUIDED_NEXT), {})
LOOP_RUNTIME_START(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_RUNTIME_START),
kmp_sch_runtime)
LOOP_NEXT(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_RUNTIME_NEXT), {})
@@ -892,6 +898,16 @@ LOOP_NEXT_ULL(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_U
LOOP_START_ULL(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_ULL_GUIDED_START),
kmp_sch_guided_chunked)
LOOP_NEXT_ULL(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_ULL_GUIDED_NEXT), {})
+LOOP_START_ULL(
+ KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_ULL_NONMONOTONIC_DYNAMIC_START),
+ kmp_sch_dynamic_chunked)
+LOOP_NEXT_ULL(
+ KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_ULL_NONMONOTONIC_DYNAMIC_NEXT), {})
+LOOP_START_ULL(
+ KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_ULL_NONMONOTONIC_GUIDED_START),
+ kmp_sch_guided_chunked)
+LOOP_NEXT_ULL(
+ KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_ULL_NONMONOTONIC_GUIDED_NEXT), {})
LOOP_RUNTIME_START_ULL(
KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_ULL_RUNTIME_START), kmp_sch_runtime)
LOOP_NEXT_ULL(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_LOOP_ULL_RUNTIME_NEXT), {})
@@ -1487,6 +1503,12 @@ PARALLEL_LOOP(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_PARALL
kmp_sch_static, OMPT_LOOP_PRE, OMPT_LOOP_POST)
PARALLEL_LOOP(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_PARALLEL_LOOP_DYNAMIC),
kmp_sch_dynamic_chunked, OMPT_LOOP_PRE, OMPT_LOOP_POST)
+PARALLEL_LOOP(
+ KMP_EXPAND_NAME(KMP_API_NAME_GOMP_PARALLEL_LOOP_NONMONOTONIC_GUIDED),
+ kmp_sch_guided_chunked, OMPT_LOOP_PRE, OMPT_LOOP_POST)
+PARALLEL_LOOP(
+ KMP_EXPAND_NAME(KMP_API_NAME_GOMP_PARALLEL_LOOP_NONMONOTONIC_DYNAMIC),
+ kmp_sch_dynamic_chunked, OMPT_LOOP_PRE, OMPT_LOOP_POST)
PARALLEL_LOOP(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_PARALLEL_LOOP_GUIDED),
kmp_sch_guided_chunked, OMPT_LOOP_PRE, OMPT_LOOP_POST)
PARALLEL_LOOP(KMP_EXPAND_NAME(KMP_API_NAME_GOMP_PARALLEL_LOOP_RUNTIME),
@@ -1941,6 +1963,26 @@ KMP_VERSION_SYMBOL(KMP_API_NAME_GOMP_LOOP_ULL_DOACROSS
KMP_VERSION_SYMBOL(KMP_API_NAME_GOMP_LOOP_ULL_DOACROSS_GUIDED_START, 45,
"GOMP_4.5");
KMP_VERSION_SYMBOL(KMP_API_NAME_GOMP_LOOP_ULL_DOACROSS_RUNTIME_START, 45,
+ "GOMP_4.5");
+KMP_VERSION_SYMBOL(KMP_API_NAME_GOMP_LOOP_NONMONOTONIC_DYNAMIC_START, 45,
+ "GOMP_4.5");
+KMP_VERSION_SYMBOL(KMP_API_NAME_GOMP_LOOP_NONMONOTONIC_DYNAMIC_NEXT, 45,
+ "GOMP_4.5");
+KMP_VERSION_SYMBOL(KMP_API_NAME_GOMP_LOOP_NONMONOTONIC_GUIDED_START, 45,
+ "GOMP_4.5");
+KMP_VERSION_SYMBOL(KMP_API_NAME_GOMP_LOOP_NONMONOTONIC_GUIDED_NEXT, 45,
+ "GOMP_4.5");
+KMP_VERSION_SYMBOL(KMP_API_NAME_GOMP_LOOP_ULL_NONMONOTONIC_DYNAMIC_START, 45,
+ "GOMP_4.5");
+KMP_VERSION_SYMBOL(KMP_API_NAME_GOMP_LOOP_ULL_NONMONOTONIC_DYNAMIC_NEXT, 45,
+ "GOMP_4.5");
+KMP_VERSION_SYMBOL(KMP_API_NAME_GOMP_LOOP_ULL_NONMONOTONIC_GUIDED_START, 45,
+ "GOMP_4.5");
+KMP_VERSION_SYMBOL(KMP_API_NAME_GOMP_LOOP_ULL_NONMONOTONIC_GUIDED_NEXT, 45,
+ "GOMP_4.5");
+KMP_VERSION_SYMBOL(KMP_API_NAME_GOMP_PARALLEL_LOOP_NONMONOTONIC_DYNAMIC, 45,
+ "GOMP_4.5");
+KMP_VERSION_SYMBOL(KMP_API_NAME_GOMP_PARALLEL_LOOP_NONMONOTONIC_GUIDED, 45,
"GOMP_4.5");
#endif // KMP_USE_VERSION_SYMBOLS
Modified: vendor/llvm-openmp/dist/runtime/src/kmp_lock.cpp
==============================================================================
--- vendor/llvm-openmp/dist/runtime/src/kmp_lock.cpp Wed Oct 23 17:53:12 2019 (r353953)
+++ vendor/llvm-openmp/dist/runtime/src/kmp_lock.cpp Wed Oct 23 17:53:14 2019 (r353954)
@@ -2943,10 +2943,10 @@ static int (*direct_test_check[])(kmp_dyna_lock_t *, k
#undef expand
// Exposes only one set of jump tables (*lock or *lock_with_checks).
-void (*(*__kmp_direct_destroy))(kmp_dyna_lock_t *) = 0;
-int (*(*__kmp_direct_set))(kmp_dyna_lock_t *, kmp_int32) = 0;
-int (*(*__kmp_direct_unset))(kmp_dyna_lock_t *, kmp_int32) = 0;
-int (*(*__kmp_direct_test))(kmp_dyna_lock_t *, kmp_int32) = 0;
+void (**__kmp_direct_destroy)(kmp_dyna_lock_t *) = 0;
+int (**__kmp_direct_set)(kmp_dyna_lock_t *, kmp_int32) = 0;
+int (**__kmp_direct_unset)(kmp_dyna_lock_t *, kmp_int32) = 0;
+int (**__kmp_direct_test)(kmp_dyna_lock_t *, kmp_int32) = 0;
// Jump tables for the indirect lock functions
#define expand(l, op) (void (*)(kmp_user_lock_p)) __kmp_##op##_##l##_##lock,
@@ -2993,10 +2993,10 @@ static int (*indirect_test_check[])(kmp_user_lock_p, k
#undef expand
// Exposes only one jump tables (*lock or *lock_with_checks).
-void (*(*__kmp_indirect_destroy))(kmp_user_lock_p) = 0;
-int (*(*__kmp_indirect_set))(kmp_user_lock_p, kmp_int32) = 0;
-int (*(*__kmp_indirect_unset))(kmp_user_lock_p, kmp_int32) = 0;
-int (*(*__kmp_indirect_test))(kmp_user_lock_p, kmp_int32) = 0;
+void (**__kmp_indirect_destroy)(kmp_user_lock_p) = 0;
+int (**__kmp_indirect_set)(kmp_user_lock_p, kmp_int32) = 0;
+int (**__kmp_indirect_unset)(kmp_user_lock_p, kmp_int32) = 0;
+int (**__kmp_indirect_test)(kmp_user_lock_p, kmp_int32) = 0;
// Lock index table.
kmp_indirect_lock_table_t __kmp_i_lock_table;
Modified: vendor/llvm-openmp/dist/runtime/src/kmp_lock.h
==============================================================================
--- vendor/llvm-openmp/dist/runtime/src/kmp_lock.h Wed Oct 23 17:53:12 2019 (r353953)
+++ vendor/llvm-openmp/dist/runtime/src/kmp_lock.h Wed Oct 23 17:53:14 2019 (r353954)
@@ -1122,18 +1122,18 @@ typedef struct {
// Function tables for direct locks. Set/unset/test differentiate functions
// with/without consistency checking.
extern void (*__kmp_direct_init[])(kmp_dyna_lock_t *, kmp_dyna_lockseq_t);
-extern void (*(*__kmp_direct_destroy))(kmp_dyna_lock_t *);
-extern int (*(*__kmp_direct_set))(kmp_dyna_lock_t *, kmp_int32);
-extern int (*(*__kmp_direct_unset))(kmp_dyna_lock_t *, kmp_int32);
-extern int (*(*__kmp_direct_test))(kmp_dyna_lock_t *, kmp_int32);
+extern void (**__kmp_direct_destroy)(kmp_dyna_lock_t *);
+extern int (**__kmp_direct_set)(kmp_dyna_lock_t *, kmp_int32);
+extern int (**__kmp_direct_unset)(kmp_dyna_lock_t *, kmp_int32);
+extern int (**__kmp_direct_test)(kmp_dyna_lock_t *, kmp_int32);
// Function tables for indirect locks. Set/unset/test differentiate functions
// with/withuot consistency checking.
extern void (*__kmp_indirect_init[])(kmp_user_lock_p);
-extern void (*(*__kmp_indirect_destroy))(kmp_user_lock_p);
-extern int (*(*__kmp_indirect_set))(kmp_user_lock_p, kmp_int32);
-extern int (*(*__kmp_indirect_unset))(kmp_user_lock_p, kmp_int32);
-extern int (*(*__kmp_indirect_test))(kmp_user_lock_p, kmp_int32);
+extern void (**__kmp_indirect_destroy)(kmp_user_lock_p);
+extern int (**__kmp_indirect_set)(kmp_user_lock_p, kmp_int32);
+extern int (**__kmp_indirect_unset)(kmp_user_lock_p, kmp_int32);
+extern int (**__kmp_indirect_test)(kmp_user_lock_p, kmp_int32);
// Extracts direct lock tag from a user lock pointer
#define KMP_EXTRACT_D_TAG(l) \
Modified: vendor/llvm-openmp/dist/runtime/src/kmp_os.h
==============================================================================
--- vendor/llvm-openmp/dist/runtime/src/kmp_os.h Wed Oct 23 17:53:12 2019 (r353953)
+++ vendor/llvm-openmp/dist/runtime/src/kmp_os.h Wed Oct 23 17:53:14 2019 (r353954)
@@ -69,7 +69,7 @@
#error Unknown compiler
#endif
-#if (KMP_OS_LINUX || KMP_OS_WINDOWS) && !KMP_OS_CNK
+#if (KMP_OS_LINUX || KMP_OS_WINDOWS || KMP_OS_FREEBSD) && !KMP_OS_CNK
#define KMP_AFFINITY_SUPPORTED 1
#if KMP_OS_WINDOWS && KMP_ARCH_X86_64
#define KMP_GROUP_AFFINITY 1
@@ -165,7 +165,8 @@ typedef unsigned long long kmp_uint64;
#if KMP_ARCH_X86 || KMP_ARCH_ARM || KMP_ARCH_MIPS
#define KMP_SIZE_T_SPEC KMP_UINT32_SPEC
-#elif KMP_ARCH_X86_64 || KMP_ARCH_PPC64 || KMP_ARCH_AARCH64 || KMP_ARCH_MIPS64
+#elif KMP_ARCH_X86_64 || KMP_ARCH_PPC64 || KMP_ARCH_AARCH64 || \
+ KMP_ARCH_MIPS64 || KMP_ARCH_RISCV64
#define KMP_SIZE_T_SPEC KMP_UINT64_SPEC
#else
#error "Can't determine size_t printf format specifier."
@@ -840,7 +841,7 @@ extern kmp_real64 __kmp_xchg_real64(volatile kmp_real6
#endif /* KMP_OS_WINDOWS */
#if KMP_ARCH_PPC64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64 || KMP_ARCH_MIPS || \
- KMP_ARCH_MIPS64
+ KMP_ARCH_MIPS64 || KMP_ARCH_RISCV64
#define KMP_MB() __sync_synchronize()
#endif
Modified: vendor/llvm-openmp/dist/runtime/src/kmp_platform.h
==============================================================================
--- vendor/llvm-openmp/dist/runtime/src/kmp_platform.h Wed Oct 23 17:53:12 2019 (r353953)
+++ vendor/llvm-openmp/dist/runtime/src/kmp_platform.h Wed Oct 23 17:53:14 2019 (r353954)
@@ -98,6 +98,7 @@
#define KMP_ARCH_PPC64 (KMP_ARCH_PPC64_LE || KMP_ARCH_PPC64_BE)
#define KMP_ARCH_MIPS 0
#define KMP_ARCH_MIPS64 0
+#define KMP_ARCH_RISCV64 0
#if KMP_OS_WINDOWS
#if defined(_M_AMD64) || defined(__x86_64)
@@ -135,6 +136,9 @@
#undef KMP_ARCH_MIPS
#define KMP_ARCH_MIPS 1
#endif
+#elif defined __riscv && __riscv_xlen == 64
+#undef KMP_ARCH_RISCV64
+#define KMP_ARCH_RISCV64 1
#endif
#endif
@@ -199,7 +203,7 @@
// TODO: Fixme - This is clever, but really fugly
#if (1 != \
KMP_ARCH_X86 + KMP_ARCH_X86_64 + KMP_ARCH_ARM + KMP_ARCH_PPC64 + \
- KMP_ARCH_AARCH64 + KMP_ARCH_MIPS + KMP_ARCH_MIPS64)
+ KMP_ARCH_AARCH64 + KMP_ARCH_MIPS + KMP_ARCH_MIPS64 + KMP_ARCH_RISCV64)
#error Unknown or unsupported architecture
#endif
Modified: vendor/llvm-openmp/dist/runtime/src/kmp_runtime.cpp
==============================================================================
--- vendor/llvm-openmp/dist/runtime/src/kmp_runtime.cpp Wed Oct 23 17:53:12 2019 (r353953)
+++ vendor/llvm-openmp/dist/runtime/src/kmp_runtime.cpp Wed Oct 23 17:53:14 2019 (r353954)
@@ -1190,8 +1190,8 @@ void __kmp_serialized_parallel(ident_t *loc, kmp_int32
ompt_callbacks.ompt_callback(ompt_callback_parallel_begin)(
&(parent_task_info->task_data), &(parent_task_info->frame),
- &ompt_parallel_data, team_size, ompt_parallel_invoker_program,
- codeptr);
+ &ompt_parallel_data, team_size,
+ ompt_parallel_invoker_program | ompt_parallel_team, codeptr);
}
}
#endif // OMPT_SUPPORT
@@ -1481,9 +1481,13 @@ int __kmp_fork_call(ident_t *loc, int gtid,
int team_size = master_set_numthreads
? master_set_numthreads
: get__nproc_2(parent_team, master_tid);
+ int flags = OMPT_INVOKER(call_context) |
+ ((microtask == (microtask_t)__kmp_teams_master)
+ ? ompt_parallel_league
+ : ompt_parallel_team);
ompt_callbacks.ompt_callback(ompt_callback_parallel_begin)(
- parent_task_data, ompt_frame, &ompt_parallel_data, team_size,
- OMPT_INVOKER(call_context), return_address);
+ parent_task_data, ompt_frame, &ompt_parallel_data, team_size, flags,
+ return_address);
}
master_th->th.ompt_thread_info.state = ompt_state_overhead;
}
@@ -1512,19 +1516,17 @@ int __kmp_fork_call(ident_t *loc, int gtid,
// AC: we are in serialized parallel
__kmpc_serialized_parallel(loc, gtid);
KMP_DEBUG_ASSERT(parent_team->t.t_serialized > 1);
- // AC: need this in order enquiry functions work
- // correctly, will restore at join time
- parent_team->t.t_serialized--;
+
#if OMPT_SUPPORT
void *dummy;
- void **exit_runtime_p;
+ void **exit_frame_p;
ompt_lw_taskteam_t lw_taskteam;
if (ompt_enabled.enabled) {
__ompt_lw_taskteam_init(&lw_taskteam, master_th, gtid,
&ompt_parallel_data, return_address);
- exit_runtime_p = &(lw_taskteam.ompt_task_info.frame.exit_frame.ptr);
+ exit_frame_p = &(lw_taskteam.ompt_task_info.frame.exit_frame.ptr);
__ompt_lw_taskteam_link(&lw_taskteam, master_th, 0);
// don't use lw_taskteam after linking. content was swaped
@@ -1532,19 +1534,23 @@ int __kmp_fork_call(ident_t *loc, int gtid,
/* OMPT implicit task begin */
implicit_task_data = OMPT_CUR_TASK_DATA(master_th);
if (ompt_enabled.ompt_callback_implicit_task) {
- ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
- ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th),
- implicit_task_data, 1, __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial?
OMPT_CUR_TASK_INFO(master_th)
->thread_num = __kmp_tid_from_gtid(gtid);
+ ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
+ ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th),
+ implicit_task_data, 1,
+ OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit);
}
/* OMPT state */
master_th->th.ompt_thread_info.state = ompt_state_work_parallel;
} else {
- exit_runtime_p = &dummy;
+ exit_frame_p = &dummy;
}
#endif
+ // AC: need to decrement t_serialized for enquiry functions to work
+ // correctly, will restore at join time
+ parent_team->t.t_serialized--;
{
KMP_TIME_PARTITIONED_BLOCK(OMP_parallel);
@@ -1552,26 +1558,27 @@ int __kmp_fork_call(ident_t *loc, int gtid,
__kmp_invoke_microtask(microtask, gtid, 0, argc, parent_team->t.t_argv
#if OMPT_SUPPORT
,
- exit_runtime_p
+ exit_frame_p
#endif
);
}
#if OMPT_SUPPORT
- *exit_runtime_p = NULL;
if (ompt_enabled.enabled) {
+ *exit_frame_p = NULL;
OMPT_CUR_TASK_INFO(master_th)->frame.exit_frame = ompt_data_none;
if (ompt_enabled.ompt_callback_implicit_task) {
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
ompt_scope_end, NULL, implicit_task_data, 1,
- OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial?
+ OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit);
}
+ ompt_parallel_data = *OMPT_CUR_TEAM_DATA(master_th);
__ompt_lw_taskteam_unlink(master_th);
-
if (ompt_enabled.ompt_callback_parallel_end) {
ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
- OMPT_CUR_TEAM_DATA(master_th), OMPT_CUR_TASK_DATA(master_th),
- OMPT_INVOKER(call_context), return_address);
+ &ompt_parallel_data, OMPT_CUR_TASK_DATA(master_th),
+ OMPT_INVOKER(call_context) | ompt_parallel_team,
+ return_address);
}
master_th->th.ompt_thread_info.state = ompt_state_overhead;
}
@@ -1586,6 +1593,15 @@ int __kmp_fork_call(ident_t *loc, int gtid,
parent_team->t.t_level++;
parent_team->t.t_def_allocator = master_th->th.th_def_allocator; // save
+#if OMPT_SUPPORT
+ if (ompt_enabled.enabled) {
+ ompt_lw_taskteam_t lw_taskteam;
+ __ompt_lw_taskteam_init(&lw_taskteam, master_th, gtid,
+ &ompt_parallel_data, return_address);
+ __ompt_lw_taskteam_link(&lw_taskteam, master_th, 1, true);
+ }
+#endif
+
/* Change number of threads in the team if requested */
if (master_set_numthreads) { // The parallel has num_threads clause
if (master_set_numthreads < master_th->th.th_teams_size.nth) {
@@ -1714,7 +1730,7 @@ int __kmp_fork_call(ident_t *loc, int gtid,
#if OMPT_SUPPORT
void *dummy;
- void **exit_runtime_p;
+ void **exit_frame_p;
ompt_task_info_t *task_info;
ompt_lw_taskteam_t lw_taskteam;
@@ -1727,19 +1743,21 @@ int __kmp_fork_call(ident_t *loc, int gtid,
// don't use lw_taskteam after linking. content was swaped
task_info = OMPT_CUR_TASK_INFO(master_th);
- exit_runtime_p = &(task_info->frame.exit_frame.ptr);
+ exit_frame_p = &(task_info->frame.exit_frame.ptr);
if (ompt_enabled.ompt_callback_implicit_task) {
- ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
- ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th),
- &(task_info->task_data), 1, __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial?
OMPT_CUR_TASK_INFO(master_th)
->thread_num = __kmp_tid_from_gtid(gtid);
+ ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
+ ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th),
+ &(task_info->task_data), 1,
+ OMPT_CUR_TASK_INFO(master_th)->thread_num,
+ ompt_task_implicit);
}
/* OMPT state */
master_th->th.ompt_thread_info.state = ompt_state_work_parallel;
} else {
- exit_runtime_p = &dummy;
+ exit_frame_p = &dummy;
}
#endif
@@ -1750,25 +1768,27 @@ int __kmp_fork_call(ident_t *loc, int gtid,
parent_team->t.t_argv
#if OMPT_SUPPORT
,
- exit_runtime_p
+ exit_frame_p
#endif
);
}
#if OMPT_SUPPORT
if (ompt_enabled.enabled) {
- exit_runtime_p = NULL;
+ *exit_frame_p = NULL;
if (ompt_enabled.ompt_callback_implicit_task) {
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
ompt_scope_end, NULL, &(task_info->task_data), 1,
- OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial?
+ OMPT_CUR_TASK_INFO(master_th)->thread_num,
+ ompt_task_implicit);
}
-
+ ompt_parallel_data = *OMPT_CUR_TEAM_DATA(master_th);
__ompt_lw_taskteam_unlink(master_th);
if (ompt_enabled.ompt_callback_parallel_end) {
ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
- OMPT_CUR_TEAM_DATA(master_th), parent_task_data,
- OMPT_INVOKER(call_context), return_address);
+ &ompt_parallel_data, parent_task_data,
+ OMPT_INVOKER(call_context) | ompt_parallel_team,
+ return_address);
}
master_th->th.ompt_thread_info.state = ompt_state_overhead;
}
@@ -1800,6 +1820,23 @@ int __kmp_fork_call(ident_t *loc, int gtid,
team->t.t_level--;
// AC: call special invoker for outer "parallel" of teams construct
invoker(gtid);
+#if OMPT_SUPPORT
+ if (ompt_enabled.enabled) {
+ ompt_task_info_t *task_info = OMPT_CUR_TASK_INFO(master_th);
+ if (ompt_enabled.ompt_callback_implicit_task) {
+ ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
+ ompt_scope_end, NULL, &(task_info->task_data), 0,
+ OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_initial);
+ }
+ if (ompt_enabled.ompt_callback_parallel_end) {
+ ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
+ &ompt_parallel_data, parent_task_data,
+ OMPT_INVOKER(call_context) | ompt_parallel_league,
+ return_address);
+ }
+ master_th->th.ompt_thread_info.state = ompt_state_overhead;
+ }
+#endif
} else {
argv = args;
for (i = argc - 1; i >= 0; --i)
@@ -1813,7 +1850,7 @@ int __kmp_fork_call(ident_t *loc, int gtid,
#if OMPT_SUPPORT
void *dummy;
- void **exit_runtime_p;
+ void **exit_frame_p;
ompt_task_info_t *task_info;
ompt_lw_taskteam_t lw_taskteam;
@@ -1824,14 +1861,15 @@ int __kmp_fork_call(ident_t *loc, int gtid,
__ompt_lw_taskteam_link(&lw_taskteam, master_th, 0);
// don't use lw_taskteam after linking. content was swaped
task_info = OMPT_CUR_TASK_INFO(master_th);
- exit_runtime_p = &(task_info->frame.exit_frame.ptr);
+ exit_frame_p = &(task_info->frame.exit_frame.ptr);
/* OMPT implicit task begin */
implicit_task_data = OMPT_CUR_TASK_DATA(master_th);
if (ompt_enabled.ompt_callback_implicit_task) {
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
ompt_scope_begin, OMPT_CUR_TEAM_DATA(master_th),
- implicit_task_data, 1, __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial?
+ implicit_task_data, 1, __kmp_tid_from_gtid(gtid),
+ ompt_task_implicit);
OMPT_CUR_TASK_INFO(master_th)
->thread_num = __kmp_tid_from_gtid(gtid);
}
@@ -1839,7 +1877,7 @@ int __kmp_fork_call(ident_t *loc, int gtid,
/* OMPT state */
master_th->th.ompt_thread_info.state = ompt_state_work_parallel;
} else {
- exit_runtime_p = &dummy;
+ exit_frame_p = &dummy;
}
#endif
@@ -1849,18 +1887,19 @@ int __kmp_fork_call(ident_t *loc, int gtid,
__kmp_invoke_microtask(microtask, gtid, 0, argc, args
#if OMPT_SUPPORT
,
- exit_runtime_p
+ exit_frame_p
#endif
);
}
#if OMPT_SUPPORT
if (ompt_enabled.enabled) {
- *exit_runtime_p = NULL;
+ *exit_frame_p = NULL;
if (ompt_enabled.ompt_callback_implicit_task) {
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
ompt_scope_end, NULL, &(task_info->task_data), 1,
- OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial?
+ OMPT_CUR_TASK_INFO(master_th)->thread_num,
+ ompt_task_implicit);
}
ompt_parallel_data = *OMPT_CUR_TEAM_DATA(master_th);
@@ -1868,7 +1907,8 @@ int __kmp_fork_call(ident_t *loc, int gtid,
if (ompt_enabled.ompt_callback_parallel_end) {
ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
&ompt_parallel_data, parent_task_data,
- OMPT_INVOKER(call_context), return_address);
+ OMPT_INVOKER(call_context) | ompt_parallel_team,
+ return_address);
}
master_th->th.ompt_thread_info.state = ompt_state_overhead;
}
@@ -2225,12 +2265,11 @@ static inline void __kmp_join_restore_state(kmp_info_t
static inline void __kmp_join_ompt(int gtid, kmp_info_t *thread,
kmp_team_t *team, ompt_data_t *parallel_data,
- fork_context_e fork_context, void *codeptr) {
+ int flags, void *codeptr) {
ompt_task_info_t *task_info = __ompt_get_task_info_object(0);
if (ompt_enabled.ompt_callback_parallel_end) {
ompt_callbacks.ompt_callback(ompt_callback_parallel_end)(
- parallel_data, &(task_info->task_data), OMPT_INVOKER(fork_context),
- codeptr);
+ parallel_data, &(task_info->task_data), flags, codeptr);
}
task_info->frame.enter_frame = ompt_data_none;
@@ -2263,6 +2302,7 @@ void __kmp_join_call(ident_t *loc, int gtid
master_th->th.th_ident = loc;
#if OMPT_SUPPORT
+ void *team_microtask = (void *)team->t.t_pkfn;
if (ompt_enabled.enabled) {
master_th->th.ompt_thread_info.state = ompt_state_overhead;
}
@@ -2352,10 +2392,25 @@ void __kmp_join_call(ident_t *loc, int gtid
if (master_th->th.th_teams_microtask && !exit_teams &&
team->t.t_pkfn != (microtask_t)__kmp_teams_master &&
team->t.t_level == master_th->th.th_teams_level + 1) {
- // AC: We need to leave the team structure intact at the end of parallel
- // inside the teams construct, so that at the next parallel same (hot) team
- // works, only adjust nesting levels
-
+// AC: We need to leave the team structure intact at the end of parallel
+// inside the teams construct, so that at the next parallel same (hot) team
+// works, only adjust nesting levels
+#if OMPT_SUPPORT
+ ompt_data_t ompt_parallel_data = ompt_data_none;
+ if (ompt_enabled.enabled) {
+ ompt_task_info_t *task_info = __ompt_get_task_info_object(0);
+ if (ompt_enabled.ompt_callback_implicit_task) {
+ int ompt_team_size = team->t.t_nproc;
+ ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
+ ompt_scope_end, NULL, &(task_info->task_data), ompt_team_size,
+ OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit);
+ }
+ task_info->frame.exit_frame = ompt_data_none;
+ task_info->task_data = ompt_data_none;
+ ompt_parallel_data = *OMPT_CUR_TEAM_DATA(master_th);
+ __ompt_lw_taskteam_unlink(master_th);
+ }
+#endif
/* Decrement our nested depth level */
team->t.t_level--;
team->t.t_active_level--;
@@ -2394,8 +2449,8 @@ void __kmp_join_call(ident_t *loc, int gtid
#if OMPT_SUPPORT
if (ompt_enabled.enabled) {
- __kmp_join_ompt(gtid, master_th, parent_team, parallel_data, fork_context,
- codeptr);
+ __kmp_join_ompt(gtid, master_th, parent_team, &ompt_parallel_data,
+ OMPT_INVOKER(fork_context) | ompt_parallel_team, codeptr);
}
#endif
@@ -2424,12 +2479,14 @@ void __kmp_join_call(ident_t *loc, int gtid
if (ompt_enabled.enabled) {
ompt_task_info_t *task_info = __ompt_get_task_info_object(0);
if (ompt_enabled.ompt_callback_implicit_task) {
- int ompt_team_size = team->t.t_nproc;
+ int flags = (team_microtask == (void *)__kmp_teams_master)
+ ? ompt_task_initial
+ : ompt_task_implicit;
+ int ompt_team_size = (flags == ompt_task_initial) ? 0 : team->t.t_nproc;
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
ompt_scope_end, NULL, &(task_info->task_data), ompt_team_size,
- OMPT_CUR_TASK_INFO(master_th)->thread_num, ompt_task_implicit); // TODO: Can this be ompt_task_initial?
+ OMPT_CUR_TASK_INFO(master_th)->thread_num, flags);
}
-
task_info->frame.exit_frame = ompt_data_none;
task_info->task_data = ompt_data_none;
}
@@ -2503,8 +2560,12 @@ void __kmp_join_call(ident_t *loc, int gtid
__kmp_release_bootstrap_lock(&__kmp_forkjoin_lock);
#if OMPT_SUPPORT
+ int flags =
+ OMPT_INVOKER(fork_context) |
+ ((team_microtask == (void *)__kmp_teams_master) ? ompt_parallel_league
+ : ompt_parallel_team);
if (ompt_enabled.enabled) {
- __kmp_join_ompt(gtid, master_th, parent_team, parallel_data, fork_context,
+ __kmp_join_ompt(gtid, master_th, parent_team, parallel_data, flags,
codeptr);
}
#endif
@@ -4432,7 +4493,7 @@ static void __kmp_initialize_team(kmp_team_t *team, in
KF_TRACE(10, ("__kmp_initialize_team: exit: team=%p\n", team));
}
-#if KMP_OS_LINUX && KMP_AFFINITY_SUPPORTED
+#if (KMP_OS_LINUX || KMP_OS_FREEBSD) && KMP_AFFINITY_SUPPORTED
/* Sets full mask for thread and returns old mask, no changes to structures. */
static void
__kmp_set_thread_affinity_mask_full_tmp(kmp_affin_mask_t *old_mask) {
@@ -4980,7 +5041,7 @@ __kmp_allocate_team(kmp_root_t *root, int new_nproc, i
__kmp_partition_places(team);
#endif
} else { // team->t.t_nproc < new_nproc
-#if KMP_OS_LINUX && KMP_AFFINITY_SUPPORTED
+#if (KMP_OS_LINUX || KMP_OS_FREEBSD) && KMP_AFFINITY_SUPPORTED
kmp_affin_mask_t *old_mask;
if (KMP_AFFINITY_CAPABLE()) {
KMP_CPU_ALLOC(old_mask);
@@ -5029,7 +5090,7 @@ __kmp_allocate_team(kmp_root_t *root, int new_nproc, i
__kmp_reinitialize_team(team, new_icvs, NULL);
}
-#if KMP_OS_LINUX && KMP_AFFINITY_SUPPORTED
+#if (KMP_OS_LINUX || KMP_OS_FREEBSD) && KMP_AFFINITY_SUPPORTED
/* Temporarily set full mask for master thread before creation of
workers. The reason is that workers inherit the affinity from master,
so if a lot of workers are created on the single core quickly, they
@@ -5064,7 +5125,7 @@ __kmp_allocate_team(kmp_root_t *root, int new_nproc, i
}
}
-#if KMP_OS_LINUX && KMP_AFFINITY_SUPPORTED
+#if (KMP_OS_LINUX || KMP_OS_FREEBSD) && KMP_AFFINITY_SUPPORTED
if (KMP_AFFINITY_CAPABLE()) {
/* Restore initial master thread's affinity mask */
__kmp_set_system_affinity(old_mask, TRUE);
@@ -5600,7 +5661,7 @@ void __kmp_free_thread(kmp_info_t *this_th) {
void *__kmp_launch_thread(kmp_info_t *this_thr) {
int gtid = this_thr->th.th_info.ds.ds_gtid;
/* void *stack_data;*/
- kmp_team_t *(*volatile pteam);
+ kmp_team_t **volatile pteam;
KMP_MB();
KA_TRACE(10, ("__kmp_launch_thread: T#%d start\n", gtid));
@@ -5618,18 +5679,15 @@ void *__kmp_launch_thread(kmp_info_t *this_thr) {
this_thr->th.ompt_thread_info.state = ompt_state_overhead;
this_thr->th.ompt_thread_info.wait_id = 0;
this_thr->th.ompt_thread_info.idle_frame = OMPT_GET_FRAME_ADDRESS(0);
+ this_thr->th.ompt_thread_info.parallel_flags = 0;
if (ompt_enabled.ompt_callback_thread_begin) {
ompt_callbacks.ompt_callback(ompt_callback_thread_begin)(
ompt_thread_worker, thread_data);
}
- }
-#endif
-
-#if OMPT_SUPPORT
- if (ompt_enabled.enabled) {
this_thr->th.ompt_thread_info.state = ompt_state_idle;
}
#endif
+
/* This is the place where threads wait for work */
while (!TCR_4(__kmp_global.g.g_done)) {
KMP_DEBUG_ASSERT(this_thr == __kmp_threads[gtid]);
@@ -5647,7 +5705,7 @@ void *__kmp_launch_thread(kmp_info_t *this_thr) {
}
#endif
- pteam = (kmp_team_t * (*))(&this_thr->th.th_team);
+ pteam = &this_thr->th.th_team;
/* have we been allocated? */
if (TCR_SYNC_PTR(*pteam) && !TCR_4(__kmp_global.g.g_done)) {
@@ -6956,16 +7014,16 @@ int __kmp_invoke_task_func(int gtid) {
#if OMPT_SUPPORT
void *dummy;
- void **exit_runtime_p;
+ void **exit_frame_p;
ompt_data_t *my_task_data;
ompt_data_t *my_parallel_data;
int ompt_team_size;
if (ompt_enabled.enabled) {
- exit_runtime_p = &(
+ exit_frame_p = &(
team->t.t_implicit_task_taskdata[tid].ompt_task_info.frame.exit_frame.ptr);
} else {
- exit_runtime_p = &dummy;
+ exit_frame_p = &dummy;
}
my_task_data =
@@ -6975,7 +7033,7 @@ int __kmp_invoke_task_func(int gtid) {
ompt_team_size = team->t.t_nproc;
ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
ompt_scope_begin, my_parallel_data, my_task_data, ompt_team_size,
- __kmp_tid_from_gtid(gtid), ompt_task_implicit); // TODO: Can this be ompt_task_initial?
+ __kmp_tid_from_gtid(gtid), ompt_task_implicit);
OMPT_CUR_TASK_INFO(this_thr)->thread_num = __kmp_tid_from_gtid(gtid);
}
#endif
@@ -6994,11 +7052,12 @@ int __kmp_invoke_task_func(int gtid) {
tid, (int)team->t.t_argc, (void **)team->t.t_argv
#if OMPT_SUPPORT
,
- exit_runtime_p
+ exit_frame_p
#endif
);
#if OMPT_SUPPORT
- *exit_runtime_p = NULL;
+ *exit_frame_p = NULL;
+ this_thr->th.ompt_thread_info.parallel_flags |= ompt_parallel_team;
#endif
#if KMP_STATS_ENABLED
@@ -7077,7 +7136,22 @@ int __kmp_invoke_teams_master(int gtid) {
(void *)__kmp_teams_master);
#endif
__kmp_run_before_invoked_task(gtid, 0, this_thr, team);
+#if OMPT_SUPPORT
+ int tid = __kmp_tid_from_gtid(gtid);
+ ompt_data_t *task_data =
+ &team->t.t_implicit_task_taskdata[tid].ompt_task_info.task_data;
+ ompt_data_t *parallel_data = &team->t.ompt_team_info.parallel_data;
+ if (ompt_enabled.ompt_callback_implicit_task) {
+ ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
+ ompt_scope_begin, parallel_data, task_data, team->t.t_nproc, tid,
+ ompt_task_initial);
+ OMPT_CUR_TASK_INFO(this_thr)->thread_num = tid;
+ }
+#endif
__kmp_teams_master(gtid);
+#if OMPT_SUPPORT
+ this_thr->th.ompt_thread_info.parallel_flags |= ompt_parallel_league;
+#endif
__kmp_run_after_invoked_task(gtid, 0, this_thr, team);
return 1;
}
@@ -7118,19 +7192,32 @@ void __kmp_push_num_teams(ident_t *id, int gtid, int n
thr->th.th_set_nproc = thr->th.th_teams_size.nteams = num_teams;
// Remember the number of threads for inner parallel regions
+ if (!TCR_4(__kmp_init_middle))
+ __kmp_middle_initialize(); // get internal globals calculated
+ KMP_DEBUG_ASSERT(__kmp_avail_proc);
+ KMP_DEBUG_ASSERT(__kmp_dflt_team_nth);
if (num_threads == 0) {
- if (!TCR_4(__kmp_init_middle))
- __kmp_middle_initialize(); // get __kmp_avail_proc calculated
num_threads = __kmp_avail_proc / num_teams;
+ // adjust num_threads w/o warning as it is not user setting
+ // num_threads = min(num_threads, nthreads-var, thread-limit-var)
+ // no thread_limit clause specified - do not change thread-limit-var ICV
+ if (num_threads > __kmp_dflt_team_nth) {
+ num_threads = __kmp_dflt_team_nth; // honor nthreads-var ICV
+ }
+ if (num_threads > thr->th.th_current_task->td_icvs.thread_limit) {
+ num_threads = thr->th.th_current_task->td_icvs.thread_limit;
+ } // prevent team size to exceed thread-limit-var
if (num_teams * num_threads > __kmp_teams_max_nth) {
- // adjust num_threads w/o warning as it is not user setting
num_threads = __kmp_teams_max_nth / num_teams;
}
} else {
// This thread will be the master of the league masters
// Store new thread limit; old limit is saved in th_cg_roots list
thr->th.th_current_task->td_icvs.thread_limit = num_threads;
-
+ // num_threads = min(num_threads, nthreads-var)
+ if (num_threads > __kmp_dflt_team_nth) {
+ num_threads = __kmp_dflt_team_nth; // honor nthreads-var ICV
+ }
if (num_teams * num_threads > __kmp_teams_max_nth) {
int new_threads = __kmp_teams_max_nth / num_teams;
if (!__kmp_reserve_warn) { // user asked for too many threads
@@ -8023,7 +8110,8 @@ __kmp_determine_reduction_method(
int atomic_available = FAST_REDUCTION_ATOMIC_METHOD_GENERATED;
-#if KMP_ARCH_X86_64 || KMP_ARCH_PPC64 || KMP_ARCH_AARCH64 || KMP_ARCH_MIPS64
+#if KMP_ARCH_X86_64 || KMP_ARCH_PPC64 || KMP_ARCH_AARCH64 || \
+ KMP_ARCH_MIPS64 || KMP_ARCH_RISCV64
#if KMP_OS_LINUX || KMP_OS_DRAGONFLY || KMP_OS_FREEBSD || KMP_OS_NETBSD || \
KMP_OS_OPENBSD || KMP_OS_WINDOWS || KMP_OS_DARWIN || KMP_OS_HURD
Modified: vendor/llvm-openmp/dist/runtime/src/kmp_stub.cpp
==============================================================================
--- vendor/llvm-openmp/dist/runtime/src/kmp_stub.cpp Wed Oct 23 17:53:12 2019 (r353953)
+++ vendor/llvm-openmp/dist/runtime/src/kmp_stub.cpp Wed Oct 23 17:53:14 2019 (r353954)
@@ -164,7 +164,7 @@ void *kmp_aligned_malloc(size_t sz, size_t a) {
#if KMP_OS_WINDOWS
res = _aligned_malloc(sz, a);
#else
- if (err = posix_memalign(&res, a, sz)) {
+ if ((err = posix_memalign(&res, a, sz))) {
errno = err; // can be EINVAL or ENOMEM
res = NULL;
}
@@ -277,7 +277,7 @@ void __kmps_get_schedule(kmp_sched_t *kind, int *modif
kmp_proc_bind_t __kmps_get_proc_bind(void) {
i;
- return 0;
+ return proc_bind_false;
} // __kmps_get_proc_bind
double __kmps_get_wtime(void) {
Modified: vendor/llvm-openmp/dist/runtime/src/kmp_taskdeps.cpp
==============================================================================
--- vendor/llvm-openmp/dist/runtime/src/kmp_taskdeps.cpp Wed Oct 23 17:53:12 2019 (r353953)
+++ vendor/llvm-openmp/dist/runtime/src/kmp_taskdeps.cpp Wed Oct 23 17:53:14 2019 (r353954)
@@ -54,12 +54,64 @@ static inline kmp_depnode_t *__kmp_node_ref(kmp_depnod
enum { KMP_DEPHASH_OTHER_SIZE = 97, KMP_DEPHASH_MASTER_SIZE = 997 };
+size_t sizes[] = { 997, 2003, 4001, 8191, 16001, 32003, 64007, 131071, 270029 };
+const size_t MAX_GEN = 8;
+
static inline kmp_int32 __kmp_dephash_hash(kmp_intptr_t addr, size_t hsize) {
// TODO alternate to try: set = (((Addr64)(addrUsefulBits * 9.618)) %
// m_num_sets );
return ((addr >> 6) ^ (addr >> 2)) % hsize;
}
+static kmp_dephash_t *__kmp_dephash_extend(kmp_info_t *thread,
+ kmp_dephash_t *current_dephash) {
+ kmp_dephash_t *h;
+
+ size_t gen = current_dephash->generation + 1;
+ if (gen >= MAX_GEN)
+ return current_dephash;
+ size_t new_size = sizes[gen];
+
+ kmp_int32 size_to_allocate =
+ new_size * sizeof(kmp_dephash_entry_t *) + sizeof(kmp_dephash_t);
+
+#if USE_FAST_MEMORY
+ h = (kmp_dephash_t *)__kmp_fast_allocate(thread, size_to_allocate);
+#else
+ h = (kmp_dephash_t *)__kmp_thread_malloc(thread, size_to_allocate);
+#endif
+
+ h->size = new_size;
+ h->nelements = current_dephash->nelements;
*** DIFF OUTPUT TRUNCATED AT 1000 LINES ***
More information about the svn-src-all
mailing list