[llvm-branch-commits] [openmp] 6f0f022 - [OpenMP] Update allocator trait key/value definitions

2021-01-12 Thread Hansang Bae via llvm-branch-commits 

Author: Hansang Bae
Date: 2021-01-12T20:09:45-06:00
New Revision: 6f0f0220380f83e8f3bf9832ffa795e9965fda2d

URL: 
https://github.com/llvm/llvm-project/commit/6f0f0220380f83e8f3bf9832ffa795e9965fda2d
DIFF: 
https://github.com/llvm/llvm-project/commit/6f0f0220380f83e8f3bf9832ffa795e9965fda2d.diff

LOG: [OpenMP] Update allocator trait key/value definitions

Use new definitions introduced in 5.1 specification.

Differential Revision: https://reviews.llvm.org/D94277

Added: 


Modified: 
openmp/runtime/src/include/omp.h.var
openmp/runtime/src/include/omp_lib.f90.var
openmp/runtime/src/include/omp_lib.h.var
openmp/runtime/src/kmp.h
openmp/runtime/src/kmp_alloc.cpp

Removed: 




diff  --git a/openmp/runtime/src/include/omp.h.var 
b/openmp/runtime/src/include/omp.h.var
index 8821377e29d8..4d055f905bcb 100644
--- a/openmp/runtime/src/include/omp.h.var
+++ b/openmp/runtime/src/include/omp.h.var
@@ -307,7 +307,7 @@
 typedef uintptr_t omp_uintptr_t;
 
 typedef enum {
-omp_atk_threadmodel = 1,
+omp_atk_sync_hint = 1,
 omp_atk_alignment = 2,
 omp_atk_access = 3,
 omp_atk_pool_size = 4,
@@ -320,10 +320,10 @@
 typedef enum {
 omp_atv_false = 0,
 omp_atv_true = 1,
-omp_atv_default = 2,
 omp_atv_contended = 3,
 omp_atv_uncontended = 4,
-omp_atv_sequential = 5,
+omp_atv_serialized = 5,
+omp_atv_sequential = omp_atv_serialized, // (deprecated)
 omp_atv_private = 6,
 omp_atv_all = 7,
 omp_atv_thread = 8,
@@ -338,6 +338,7 @@
 omp_atv_blocked = 17,
 omp_atv_interleaved = 18
 } omp_alloctrait_value_t;
+#define omp_atv_default ((omp_uintptr_t)-1)
 
 typedef struct {
 omp_alloctrait_key_t key;

diff  --git a/openmp/runtime/src/include/omp_lib.f90.var 
b/openmp/runtime/src/include/omp_lib.f90.var
index 24f8a2af4c5e..2fc8d7c3daa4 100644
--- a/openmp/runtime/src/include/omp_lib.f90.var
+++ b/openmp/runtime/src/include/omp_lib.f90.var
@@ -98,7 +98,7 @@
 integer (kind=omp_control_tool_result_kind), parameter :: 
omp_control_tool_success = 0
 integer (kind=omp_control_tool_result_kind), parameter :: 
omp_control_tool_ignored = 1
 
-integer (kind=omp_alloctrait_key_kind), parameter :: 
omp_atk_threadmodel = 1
+integer (kind=omp_alloctrait_key_kind), parameter :: omp_atk_sync_hint 
= 1
 integer (kind=omp_alloctrait_key_kind), parameter :: omp_atk_alignment 
= 2
 integer (kind=omp_alloctrait_key_kind), parameter :: omp_atk_access = 3
 integer (kind=omp_alloctrait_key_kind), parameter :: omp_atk_pool_size 
= 4
@@ -107,12 +107,13 @@
 integer (kind=omp_alloctrait_key_kind), parameter :: omp_atk_pinned = 7
 integer (kind=omp_alloctrait_key_kind), parameter :: omp_atk_partition 
= 8
 
+integer (kind=omp_alloctrait_val_kind), parameter :: omp_atv_default = 
-1
 integer (kind=omp_alloctrait_val_kind), parameter :: omp_atv_false = 0
 integer (kind=omp_alloctrait_val_kind), parameter :: omp_atv_true = 1
-integer (kind=omp_alloctrait_val_kind), parameter :: omp_atv_default = 
2
 integer (kind=omp_alloctrait_val_kind), parameter :: omp_atv_contended 
= 3
 integer (kind=omp_alloctrait_val_kind), parameter :: 
omp_atv_uncontended = 4
-integer (kind=omp_alloctrait_val_kind), parameter :: 
omp_atv_sequential = 5
+integer (kind=omp_alloctrait_val_kind), parameter :: 
omp_atv_serialized = 5
+integer (kind=omp_alloctrait_val_kind), parameter :: 
omp_atv_sequential = omp_atv_serialized
 integer (kind=omp_alloctrait_val_kind), parameter :: omp_atv_private = 
6
 integer (kind=omp_alloctrait_val_kind), parameter :: omp_atv_all = 7
 integer (kind=omp_alloctrait_val_kind), parameter :: omp_atv_thread = 8

diff  --git a/openmp/runtime/src/include/omp_lib.h.var 
b/openmp/runtime/src/include/omp_lib.h.var
index 05140b04f273..3fb2d25b15f1 100644
--- a/openmp/runtime/src/include/omp_lib.h.var
+++ b/openmp/runtime/src/include/omp_lib.h.var
@@ -131,8 +131,8 @@
   integer(omp_control_tool_result_kind)omp_control_tool_ignored
   parameter(omp_control_tool_ignored=1)
 
-  integer(kind=omp_alloctrait_key_kind)omp_atk_threadmodel
-  parameter(omp_atk_threadmodel=1)
+  integer(kind=omp_alloctrait_key_kind)omp_atk_sync_hint
+  parameter(omp_atk_sync_hint=1)
   integer(kind=omp_alloctrait_key_kind)omp_atk_alignment
   parameter(omp_atk_alignment=2)
   integer(kind=omp_alloctrait_key_kind)omp_atk_access
@@ -148,18 +148,20 @@
   integer(kind=omp_alloctrait_key_kind)omp_atk_partition
   parameter(omp_atk_partition=8)
 
+  integer(kind=omp_alloctrait_val_kind)omp_atv_default
+  parameter(omp_atv_default=-1)
   ! Reserved for future use
   integer(kind=omp_alloctrait_val_kind)omp

[llvm-branch-commits] [openmp] bba3a82 - [OpenMP] Use persistent memory for omp_large_cap_mem

2021-01-12 Thread Hansang Bae via llvm-branch-commits 

Author: Hansang Bae
Date: 2021-01-12T20:35:27-06:00
New Revision: bba3a82b56c0874757f2c1423bbdff08e2a88967

URL: 
https://github.com/llvm/llvm-project/commit/bba3a82b56c0874757f2c1423bbdff08e2a88967
DIFF: 
https://github.com/llvm/llvm-project/commit/bba3a82b56c0874757f2c1423bbdff08e2a88967.diff

LOG: [OpenMP] Use persistent memory for omp_large_cap_mem

This change enables volatile use of persistent memory for omp_large_cap_mem*
on supported systems. It depends on libmemkind's support for persistent memory,
and requirements/details can be found at the following url.

https://pmem.io/2020/01/20/memkind-dax-kmem.html

Differential Revision: https://reviews.llvm.org/D94353

Added: 


Modified: 
openmp/runtime/src/kmp_alloc.cpp

Removed: 




diff  --git a/openmp/runtime/src/kmp_alloc.cpp 
b/openmp/runtime/src/kmp_alloc.cpp
index b56e71881208..31981d5c1d55 100644
--- a/openmp/runtime/src/kmp_alloc.cpp
+++ b/openmp/runtime/src/kmp_alloc.cpp
@@ -1239,6 +1239,9 @@ static void **mk_hbw_preferred;
 static void **mk_hugetlb;
 static void **mk_hbw_hugetlb;
 static void **mk_hbw_preferred_hugetlb;
+static void **mk_dax_kmem;
+static void **mk_dax_kmem_all;
+static void **mk_dax_kmem_preferred;
 
 #if KMP_OS_UNIX && KMP_DYNAMIC_LIB
 static inline void chk_kind(void ***pkind) {
@@ -1279,25 +1282,21 @@ void __kmp_init_memkind() {
   mk_hbw_preferred_hugetlb =
   (void **)dlsym(h_memkind, "MEMKIND_HBW_PREFERRED_HUGETLB");
   chk_kind(&mk_hbw_preferred_hugetlb);
+  mk_dax_kmem = (void **)dlsym(h_memkind, "MEMKIND_DAX_KMEM");
+  chk_kind(&mk_dax_kmem);
+  mk_dax_kmem_all = (void **)dlsym(h_memkind, "MEMKIND_DAX_KMEM_ALL");
+  chk_kind(&mk_dax_kmem_all);
+  mk_dax_kmem_preferred =
+  (void **)dlsym(h_memkind, "MEMKIND_DAX_KMEM_PREFERRED");
+  chk_kind(&mk_dax_kmem_preferred);
   KE_TRACE(25, ("__kmp_init_memkind: memkind library initialized\n"));
   return; // success
 }
 dlclose(h_memkind); // failure
-h_memkind = NULL;
   }
-  kmp_mk_check = NULL;
-  kmp_mk_alloc = NULL;
-  kmp_mk_free = NULL;
-  mk_default = NULL;
-  mk_interleave = NULL;
-  mk_hbw = NULL;
-  mk_hbw_interleave = NULL;
-  mk_hbw_preferred = NULL;
-  mk_hugetlb = NULL;
-  mk_hbw_hugetlb = NULL;
-  mk_hbw_preferred_hugetlb = NULL;
-#else
+#else // !(KMP_OS_UNIX && KMP_DYNAMIC_LIB)
   kmp_mk_lib_name = "";
+#endif // !(KMP_OS_UNIX && KMP_DYNAMIC_LIB)
   h_memkind = NULL;
   kmp_mk_check = NULL;
   kmp_mk_alloc = NULL;
@@ -1310,7 +1309,9 @@ void __kmp_init_memkind() {
   mk_hugetlb = NULL;
   mk_hbw_hugetlb = NULL;
   mk_hbw_preferred_hugetlb = NULL;
-#endif
+  mk_dax_kmem = NULL;
+  mk_dax_kmem_all = NULL;
+  mk_dax_kmem_preferred = NULL;
 }
 
 void __kmp_fini_memkind() {
@@ -1332,6 +1333,9 @@ void __kmp_fini_memkind() {
   mk_hugetlb = NULL;
   mk_hbw_hugetlb = NULL;
   mk_hbw_preferred_hugetlb = NULL;
+  mk_dax_kmem = NULL;
+  mk_dax_kmem_all = NULL;
+  mk_dax_kmem_preferred = NULL;
 #endif
 }
 
@@ -1401,6 +1405,17 @@ omp_allocator_handle_t __kmpc_init_allocator(int gtid, 
omp_memspace_handle_t ms,
 __kmp_free(al);
 return omp_null_allocator;
   }
+} else if (ms == omp_large_cap_mem_space) {
+  if (mk_dax_kmem_all) {
+// All pmem nodes are visited
+al->memkind = mk_dax_kmem_all;
+  } else if (mk_dax_kmem) {
+// Only closest pmem node is visited
+al->memkind = mk_dax_kmem;
+  } else {
+__kmp_free(al);
+return omp_null_allocator;
+  }
 } else {
   if (al->memkind == (void *)omp_atv_interleaved && mk_interleave) {
 al->memkind = mk_interleave;
@@ -1473,6 +1488,8 @@ void *__kmpc_alloc(int gtid, size_t size, 
omp_allocator_handle_t allocator) {
   // pre-defined allocator
   if (allocator == omp_high_bw_mem_alloc && mk_hbw_preferred) {
 ptr = kmp_mk_alloc(*mk_hbw_preferred, desc.size_a);
+  } else if (allocator == omp_large_cap_mem_alloc && mk_dax_kmem_all) {
+ptr = kmp_mk_alloc(*mk_dax_kmem_all, desc.size_a);
   } else {
 ptr = kmp_mk_alloc(*mk_default, desc.size_a);
   }
@@ -1529,6 +1546,8 @@ void *__kmpc_alloc(int gtid, size_t size, 
omp_allocator_handle_t allocator) {
 // pre-defined allocator
 if (allocator == omp_high_bw_mem_alloc) {
   // ptr = NULL;
+} else if (allocator == omp_large_cap_mem_alloc) {
+  // warnings?
 } else {
   ptr = __kmp_thread_malloc(__kmp_thread_from_gtid(gtid), desc.size_a);
 }
@@ -1684,6 +1703,8 @@ void __kmpc_free(int gtid, void *ptr, const 
omp_allocator_handle_t allocator) {
   // pre-defined allocator
   if (oal == omp_high_bw_mem_alloc && mk_hbw_preferred) {
 kmp_mk_free(*mk_hbw_preferred, desc.ptr_alloc);
+  } else if (oal == omp_large_cap_mem_alloc && mk_dax_kmem_all) {
+kmp_mk_free(*mk_dax_kmem_all, desc.ptr_alloc);
   } else {
 kmp_mk_free(*mk_defa

[llvm-branch-commits] [openmp] 2d911f7 - [OpenMP] Fix atomic entries for captured logical operation

2021-01-19 Thread Hansang Bae via llvm-branch-commits 

Author: Hansang Bae
Date: 2021-01-19T09:59:28-06:00
New Revision: 2d911f7c72f9174a34f74abe2909f992b03caaf1

URL: 
https://github.com/llvm/llvm-project/commit/2d911f7c72f9174a34f74abe2909f992b03caaf1
DIFF: 
https://github.com/llvm/llvm-project/commit/2d911f7c72f9174a34f74abe2909f992b03caaf1.diff

LOG: [OpenMP] Fix atomic entries for captured logical operation

Added missing code for the captured atomic operation.

Differential Revision: https://reviews.llvm.org/D94848

Added: 


Modified: 
openmp/runtime/src/kmp_atomic.cpp

Removed: 




diff  --git a/openmp/runtime/src/kmp_atomic.cpp 
b/openmp/runtime/src/kmp_atomic.cpp
index 4d60b550ae62..a9d5257ab2aa 100644
--- a/openmp/runtime/src/kmp_atomic.cpp
+++ b/openmp/runtime/src/kmp_atomic.cpp
@@ -2536,8 +2536,11 @@ ATOMIC_CRITICAL_CPT_MIX(float10, long double, div_cpt, 
/, fp, _Quad, 10r,

\
   if (flag) {  
\
 new_value OP rhs;  
\
-  } else   
\
+(*lhs) = new_value;
\
+  } else { 
\
 new_value = (*lhs);
\
+(*lhs) OP rhs; 
\
+  }
\

\
   __kmp_release_atomic_lock(&ATOMIC_LOCK##LCK_ID, gtid);
 



___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits

[llvm-branch-commits] [openmp] 480cbed - [OpenMP] Remove unnecessary pointer checks in a few locations

2021-01-22 Thread Hansang Bae via llvm-branch-commits 

Author: Hansang Bae
Date: 2021-01-22T19:18:50-06:00
New Revision: 480cbed31e74b0db3d31d78789b639af250ce9fe

URL: 
https://github.com/llvm/llvm-project/commit/480cbed31e74b0db3d31d78789b639af250ce9fe
DIFF: 
https://github.com/llvm/llvm-project/commit/480cbed31e74b0db3d31d78789b639af250ce9fe.diff

LOG: [OpenMP] Remove unnecessary pointer checks in a few locations

Also, return NULL from unsuccessful OMPT function lookup.

Differential Revision: https://reviews.llvm.org/D95277

Added: 


Modified: 
openmp/runtime/src/kmp_taskdeps.cpp
openmp/runtime/src/ompt-general.cpp

Removed: 




diff  --git a/openmp/runtime/src/kmp_taskdeps.cpp 
b/openmp/runtime/src/kmp_taskdeps.cpp
index f580431d0182..5713cb9cb4d8 100644
--- a/openmp/runtime/src/kmp_taskdeps.cpp
+++ b/openmp/runtime/src/kmp_taskdeps.cpp
@@ -529,10 +529,9 @@ kmp_int32 __kmpc_omp_task_with_deps(ident_t *loc_ref, 
kmp_int32 gtid,
   current_task->ompt_task_info.frame.enter_frame.ptr =
   OMPT_GET_FRAME_ADDRESS(0);
 if (ompt_enabled.ompt_callback_task_create) {
-  ompt_data_t task_data = ompt_data_none;
   ompt_callbacks.ompt_callback(ompt_callback_task_create)(
-  current_task ? &(current_task->ompt_task_info.task_data) : 
&task_data,
-  current_task ? &(current_task->ompt_task_info.frame) : NULL,
+  &(current_task->ompt_task_info.task_data),
+  &(current_task->ompt_task_info.frame),
   &(new_taskdata->ompt_task_info.task_data),
   ompt_task_explicit | TASK_TYPE_DETAILS_FORMAT(new_taskdata), 1,
   OMPT_LOAD_OR_GET_RETURN_ADDRESS(gtid));
@@ -646,13 +645,12 @@ kmp_int32 __kmpc_omp_task_with_deps(ident_t *loc_ref, 
kmp_int32 gtid,
 void __ompt_taskwait_dep_finish(kmp_taskdata_t *current_task,
 ompt_data_t *taskwait_task_data) {
   if (ompt_enabled.ompt_callback_task_schedule) {
-ompt_data_t task_data = ompt_data_none;
 ompt_callbacks.ompt_callback(ompt_callback_task_schedule)(
-current_task ? &(current_task->ompt_task_info.task_data) : &task_data,
-ompt_task_switch, taskwait_task_data);
+&(current_task->ompt_task_info.task_data), ompt_task_switch,
+taskwait_task_data);
 ompt_callbacks.ompt_callback(ompt_callback_task_schedule)(
 taskwait_task_data, ompt_task_complete,
-current_task ? &(current_task->ompt_task_info.task_data) : &task_data);
+&(current_task->ompt_task_info.task_data));
   }
   current_task->ompt_task_info.frame.enter_frame.ptr = NULL;
   *taskwait_task_data = ompt_data_none;
@@ -698,11 +696,9 @@ void __kmpc_omp_wait_deps(ident_t *loc_ref, kmp_int32 
gtid, kmp_int32 ndeps,
   current_task->ompt_task_info.frame.enter_frame.ptr =
   OMPT_GET_FRAME_ADDRESS(0);
 if (ompt_enabled.ompt_callback_task_create) {
-  ompt_data_t task_data = ompt_data_none;
   ompt_callbacks.ompt_callback(ompt_callback_task_create)(
-  current_task ? &(current_task->ompt_task_info.task_data) : 
&task_data,
-  current_task ? &(current_task->ompt_task_info.frame) : NULL,
-  taskwait_task_data,
+  &(current_task->ompt_task_info.task_data),
+  &(current_task->ompt_task_info.frame), taskwait_task_data,
   ompt_task_explicit | ompt_task_undeferred | ompt_task_mergeable, 1,
   OMPT_LOAD_OR_GET_RETURN_ADDRESS(gtid));
 }

diff  --git a/openmp/runtime/src/ompt-general.cpp 
b/openmp/runtime/src/ompt-general.cpp
index 36bd6b55f33a..c52a3f28c2e7 100644
--- a/openmp/runtime/src/ompt-general.cpp
+++ b/openmp/runtime/src/ompt-general.cpp
@@ -840,5 +840,5 @@ static ompt_interface_fn_t ompt_fn_lookup(const char *s) {
 
   FOREACH_OMPT_INQUIRY_FN(ompt_interface_fn)
 
-  return (ompt_interface_fn_t)0;
+  return NULL;
 }



___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits

[llvm-branch-commits] [openmp] 82a29a6 - [OpenMP] Add definition/interface for target memory routines

2021-01-04 Thread Hansang Bae via llvm-branch-commits 

Author: Hansang Bae
Date: 2021-01-04T08:12:57-06:00
New Revision: 82a29a62aba52d68d37309cd3025370ba98e37e4

URL: 
https://github.com/llvm/llvm-project/commit/82a29a62aba52d68d37309cd3025370ba98e37e4
DIFF: 
https://github.com/llvm/llvm-project/commit/82a29a62aba52d68d37309cd3025370ba98e37e4.diff

LOG: [OpenMP] Add definition/interface for target memory routines

The change includes new routines introduced in 5.1 and Fortran
interface.

Differential Revision: https://reviews.llvm.org/D93505

Added: 


Modified: 
openmp/runtime/src/include/omp.h.var
openmp/runtime/src/include/omp_lib.f90.var
openmp/runtime/src/include/omp_lib.h.var

Removed: 




diff  --git a/openmp/runtime/src/include/omp.h.var 
b/openmp/runtime/src/include/omp.h.var
index b687ff16eaeb..8821377e29d8 100644
--- a/openmp/runtime/src/include/omp.h.var
+++ b/openmp/runtime/src/include/omp.h.var
@@ -213,6 +213,24 @@
  */
 extern const char * __KAI_KMPC_CONVENTION  omp_get_interop_rc_desc(const 
omp_interop_rc_t, omp_interop_rc_t);
 
+/* OpenMP 5.1 device memory routines */
+
+/*!
+ * The `omp_target_memcpy_async` routine asynchronously performs a copy 
between any combination of host and device pointers.
+ */
+extern int__KAI_KMPC_CONVENTION  omp_target_memcpy_async(void *, const 
void *, size_t, size_t, size_t, int,
+ int, int, omp_depend_t *);
+/*!
+ * The `omp_target_memcpy_rect_async` routine asynchronously performs a 
copy between any combination of host and device pointers.
+ */
+extern int__KAI_KMPC_CONVENTION  omp_target_memcpy_rect_async(void *, 
const void *, size_t, int, const size_t *,
+ const size_t *, const size_t *, 
const size_t *, const size_t *, int, int,
+ int, omp_depend_t *);
+/*!
+ * The `omp_get_mapped_ptr` routine returns the device pointer that is 
associated with a host pointer for a given device.
+ */
+extern void * __KAI_KMPC_CONVENTION  omp_get_mapped_ptr(const void *, int);
+
 /* kmp API functions */
 extern int__KAI_KMPC_CONVENTION  kmp_get_stacksize  (void);
 extern void   __KAI_KMPC_CONVENTION  kmp_set_stacksize  (int);

diff  --git a/openmp/runtime/src/include/omp_lib.f90.var 
b/openmp/runtime/src/include/omp_lib.f90.var
index fbbb7b9df94d..1bde868a505b 100644
--- a/openmp/runtime/src/include/omp_lib.f90.var
+++ b/openmp/runtime/src/include/omp_lib.f90.var
@@ -509,6 +509,103 @@
   end subroutine omp_display_env
 
 
+  function omp_target_alloc(size, device_num) bind(c)
+use omp_lib_kinds
+type(c_ptr) omp_target_alloc
+integer (kind=kmp_size_t_kind), value :: size
+integer (kind=omp_integer_kind), value :: device_num
+  end function omp_target_alloc
+
+  subroutine omp_target_free(device_ptr, device_num) bind(c)
+use omp_lib_kinds
+type(c_ptr), value :: device_ptr
+integer (kind=omp_integer_kind), value :: device_num
+  end subroutine omp_target_free
+
+  function omp_target_is_present(ptr, device_num) bind(c)
+use omp_lib_kinds
+integer (kind=omp_integer_kind) omp_target_is_present
+type(c_ptr), value :: ptr
+integer (kind=omp_integer_kind), value :: device_num
+  end function omp_target_is_present
+
+  function omp_target_memcpy(dst, src, length, dst_offset, src_offset, 
&
+  dst_device_num, src_device_num) bind(c)
+use omp_lib_kinds
+integer (kind=omp_integer_kind) omp_target_memcpy
+type(c_ptr), value :: dst, src
+integer (kind=kmp_size_t_kind), value :: length, dst_offset,   
&
+src_offset
+integer (kind=omp_integer_kind), value :: dst_device_num,  
&
+src_device_num
+  end function omp_target_memcpy
+
+  function omp_target_memcpy_rect(dst, src, element_size, num_dims,
&
+  volume, dst_offsets, src_offsets, dst_dimensions,
&
+  src_dimensions, dst_device_num, src_device_num) bind(c)
+use omp_lib_kinds
+integer (kind=omp_integer_kind) omp_target_memcpy_rect
+type(c_ptr), value :: dst, src
+integer (kind=kmp_size_t_kind), value :: element_size
+integer (kind=omp_integer_kind), value :: num_dims,
&
+dst_device_num, src_device_num
+integer (kind=kmp_size_t_kind), intent(in) :: volume(*),   
&
+dst_offsets(*), src_offsets(*), dst_dimensions(*), 
&
+src_dimensions(*)
+  end function omp_target_memcpy_rect
+
+  function omp_target_memcpy_async(dst, src,

[llvm-branch-commits] [openmp] fb1c528 - [OpenMP] Use c_int/c_size_t in Fortran target memory routine interface

2021-01-06 Thread Hansang Bae via llvm-branch-commits 

Author: Hansang Bae
Date: 2021-01-06T16:28:30-06:00
New Revision: fb1c52852690cb6ad6f7f2a7b5c35bf4d0c56e44

URL: 
https://github.com/llvm/llvm-project/commit/fb1c52852690cb6ad6f7f2a7b5c35bf4d0c56e44
DIFF: 
https://github.com/llvm/llvm-project/commit/fb1c52852690cb6ad6f7f2a7b5c35bf4d0c56e44.diff

LOG: [OpenMP] Use c_int/c_size_t in Fortran target memory routine interface

The Fortran interface is now in line with 5.1 specification.

Differential Revision: https://reviews.llvm.org/D94042

Added: 


Modified: 
openmp/runtime/src/include/omp_lib.f90.var
openmp/runtime/src/include/omp_lib.h.var

Removed: 




diff  --git a/openmp/runtime/src/include/omp_lib.f90.var 
b/openmp/runtime/src/include/omp_lib.f90.var
index 1bde868a505b..24f8a2af4c5e 100644
--- a/openmp/runtime/src/include/omp_lib.f90.var
+++ b/openmp/runtime/src/include/omp_lib.f90.var
@@ -512,59 +512,54 @@
   function omp_target_alloc(size, device_num) bind(c)
 use omp_lib_kinds
 type(c_ptr) omp_target_alloc
-integer (kind=kmp_size_t_kind), value :: size
-integer (kind=omp_integer_kind), value :: device_num
+integer(c_size_t), value :: size
+integer(c_int), value :: device_num
   end function omp_target_alloc
 
   subroutine omp_target_free(device_ptr, device_num) bind(c)
 use omp_lib_kinds
 type(c_ptr), value :: device_ptr
-integer (kind=omp_integer_kind), value :: device_num
+integer(c_int), value :: device_num
   end subroutine omp_target_free
 
   function omp_target_is_present(ptr, device_num) bind(c)
 use omp_lib_kinds
-integer (kind=omp_integer_kind) omp_target_is_present
+integer(c_int) omp_target_is_present
 type(c_ptr), value :: ptr
-integer (kind=omp_integer_kind), value :: device_num
+integer(c_int), value :: device_num
   end function omp_target_is_present
 
   function omp_target_memcpy(dst, src, length, dst_offset, src_offset, 
&
   dst_device_num, src_device_num) bind(c)
 use omp_lib_kinds
-integer (kind=omp_integer_kind) omp_target_memcpy
+integer(c_int) omp_target_memcpy
 type(c_ptr), value :: dst, src
-integer (kind=kmp_size_t_kind), value :: length, dst_offset,   
&
-src_offset
-integer (kind=omp_integer_kind), value :: dst_device_num,  
&
-src_device_num
+integer(c_size_t), value :: length, dst_offset, src_offset
+integer(c_int), value :: dst_device_num, src_device_num
   end function omp_target_memcpy
 
   function omp_target_memcpy_rect(dst, src, element_size, num_dims,
&
   volume, dst_offsets, src_offsets, dst_dimensions,
&
   src_dimensions, dst_device_num, src_device_num) bind(c)
 use omp_lib_kinds
-integer (kind=omp_integer_kind) omp_target_memcpy_rect
+integer(c_int) omp_target_memcpy_rect
 type(c_ptr), value :: dst, src
-integer (kind=kmp_size_t_kind), value :: element_size
-integer (kind=omp_integer_kind), value :: num_dims,
&
-dst_device_num, src_device_num
-integer (kind=kmp_size_t_kind), intent(in) :: volume(*),   
&
-dst_offsets(*), src_offsets(*), dst_dimensions(*), 
&
-src_dimensions(*)
+integer(c_size_t), value :: element_size
+integer(c_int), value :: num_dims, dst_device_num, src_device_num
+integer(c_size_t), intent(in) :: volume(*), dst_offsets(*),
&
+src_offsets(*), dst_dimensions(*), src_dimensions(*)
   end function omp_target_memcpy_rect
 
   function omp_target_memcpy_async(dst, src, length, dst_offset,   
&
   src_offset, dst_device_num, src_device_num, depobj_count,
&
   depobj_list) bind(c)
 use omp_lib_kinds
-integer (kind=omp_integer_kind) omp_target_memcpy_async
+integer(c_int) omp_target_memcpy_async
 type(c_ptr), value :: dst, src
-integer (kind=kmp_size_t_kind), value :: length, dst_offset,   
&
-src_offset
-integer (kind=omp_integer_kind), value :: dst_device_num,  
&
-src_device_num, depobj_count
-integer (kind=omp_depend_kind), optional :: depobj_list(*)
+integer(c_size_t), value :: length, dst_offset, src_offset
+integer(c_int), value :: dst_device_num, src_device_num,   
&
+depobj_count
+integer(omp_depend_kind), optional :: depobj_list(*)
   end function omp_target_memcpy_asyn

[llvm-branch-commits] [openmp] 171ca93 - [OpenMP] Initialize runtime in the forked child process

2020-12-15 Thread Hansang Bae via llvm-branch-commits 

Author: Hansang Bae
Date: 2020-12-15T07:29:28-06:00
New Revision: 171ca93c543098d5e0bef459847de14cf17b5faf

URL: 
https://github.com/llvm/llvm-project/commit/171ca93c543098d5e0bef459847de14cf17b5faf
DIFF: 
https://github.com/llvm/llvm-project/commit/171ca93c543098d5e0bef459847de14cf17b5faf.diff

LOG: [OpenMP] Initialize runtime in the forked child process

This patch enables serial initialization in the forked child process
to fix unstable runtime behavior when used with Python-based AI tools.

Differential Revision: https://reviews.llvm.org/D93230

Added: 


Modified: 
openmp/runtime/src/z_Linux_util.cpp

Removed: 




diff  --git a/openmp/runtime/src/z_Linux_util.cpp 
b/openmp/runtime/src/z_Linux_util.cpp
index 15237d631958..d039a454f00d 100644
--- a/openmp/runtime/src/z_Linux_util.cpp
+++ b/openmp/runtime/src/z_Linux_util.cpp
@@ -1272,8 +1272,8 @@ static void __kmp_atfork_prepare(void) {
 }
 
 static void __kmp_atfork_parent(void) {
-  __kmp_release_bootstrap_lock(&__kmp_initz_lock);
   __kmp_release_bootstrap_lock(&__kmp_forkjoin_lock);
+  __kmp_release_bootstrap_lock(&__kmp_initz_lock);
 }
 
 /* Reset the library so execution in the child starts "all over again" with
@@ -1281,6 +1281,7 @@ static void __kmp_atfork_parent(void) {
allocated by parent, just abandon it to be safe. */
 static void __kmp_atfork_child(void) {
   __kmp_release_bootstrap_lock(&__kmp_forkjoin_lock);
+  __kmp_release_bootstrap_lock(&__kmp_initz_lock);
   /* TODO make sure this is done right for nested/sibling */
   // ATT:  Memory leaks are here? TODO: Check it and fix.
   /* KMP_ASSERT( 0 ); */
@@ -1302,7 +1303,6 @@ static void __kmp_atfork_child(void) {
   }
 #endif // KMP_AFFINITY_SUPPORTED
 
-  __kmp_init_runtime = FALSE;
 #if KMP_USE_MONITOR
   __kmp_init_monitor = 0;
 #endif
@@ -1355,6 +1355,8 @@ static void __kmp_atfork_child(void) {
   __kmp_itt_reset(); // reset ITT's global state
 #endif /* USE_ITT_BUILD */
 
+  __kmp_serial_initialize();
+
   /* This is necessary to make sure no stale data is left around */
   /* AC: customers complain that we use unsafe routines in the atfork
  handler. Mathworks: dlsym() is unsafe. We call dlsym and dlopen



___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits

[llvm-branch-commits] [openmp] e1fd202 - [OpenMP] Add definitions for 5.1 interop to omp.h

2020-12-17 Thread Hansang Bae via llvm-branch-commits 

Author: Hansang Bae
Date: 2020-12-17T13:03:59-06:00
New Revision: e1fd202489e184e32a23fe01af8a61e48af186a8

URL: 
https://github.com/llvm/llvm-project/commit/e1fd202489e184e32a23fe01af8a61e48af186a8
DIFF: 
https://github.com/llvm/llvm-project/commit/e1fd202489e184e32a23fe01af8a61e48af186a8.diff

LOG: [OpenMP] Add definitions for 5.1 interop to omp.h

Added: 


Modified: 
openmp/runtime/src/include/omp.h.var

Removed: 




diff  --git a/openmp/runtime/src/include/omp.h.var 
b/openmp/runtime/src/include/omp.h.var
index 510bfc225f89..b687ff16eaeb 100644
--- a/openmp/runtime/src/include/omp.h.var
+++ b/openmp/runtime/src/include/omp.h.var
@@ -152,6 +152,67 @@
 extern int   __KAI_KMPC_CONVENTION  omp_get_device_num (void);
 typedef void * omp_depend_t;
 
+/* OpenMP 5.1 interop */
+typedef intptr_t omp_intptr_t;
+
+/* 0..omp_get_num_interop_properties()-1 are reserved for 
implementation-defined properties */
+typedef enum omp_interop_property {
+omp_ipr_fr_id = -1,
+omp_ipr_fr_name = -2,
+omp_ipr_vendor = -3,
+omp_ipr_vendor_name = -4,
+omp_ipr_device_num = -5,
+omp_ipr_platform = -6,
+omp_ipr_device = -7,
+omp_ipr_device_context = -8,
+omp_ipr_targetsync = -9,
+omp_ipr_first = -9
+} omp_interop_property_t;
+
+#define omp_interop_none 0
+
+typedef enum omp_interop_rc {
+omp_irc_no_value = 1,
+omp_irc_success = 0,
+omp_irc_empty = -1,
+omp_irc_out_of_range = -2,
+omp_irc_type_int = -3,
+omp_irc_type_ptr = -4,
+omp_irc_type_str = -5,
+omp_irc_other = -6
+} omp_interop_rc_t;
+
+typedef void * omp_interop_t;
+
+/*!
+ * The `omp_get_num_interop_properties` routine retrieves the number of 
implementation-defined properties available for an `omp_interop_t` object.
+ */
+extern int  __KAI_KMPC_CONVENTION  
omp_get_num_interop_properties(const omp_interop_t);
+/*!
+ * The `omp_get_interop_int` routine retrieves an integer property from an 
`omp_interop_t` object.
+ */
+extern omp_intptr_t __KAI_KMPC_CONVENTION  omp_get_interop_int(const 
omp_interop_t, omp_interop_property_t, int *);
+/*!
+ * The `omp_get_interop_ptr` routine retrieves a pointer property from an 
`omp_interop_t` object.
+ */
+extern void *   __KAI_KMPC_CONVENTION  omp_get_interop_ptr(const 
omp_interop_t, omp_interop_property_t, int *);
+/*!
+ * The `omp_get_interop_str` routine retrieves a string property from an 
`omp_interop_t` object.
+ */
+extern const char * __KAI_KMPC_CONVENTION  omp_get_interop_str(const 
omp_interop_t, omp_interop_property_t, int *);
+/*!
+ * The `omp_get_interop_name` routine retrieves a property name from an 
`omp_interop_t` object.
+ */
+extern const char * __KAI_KMPC_CONVENTION  omp_get_interop_name(const 
omp_interop_t, omp_interop_property_t);
+/*!
+ * The `omp_get_interop_type_desc` routine retrieves a description of the 
type of a property associated with an `omp_interop_t` object.
+ */
+extern const char * __KAI_KMPC_CONVENTION  omp_get_interop_type_desc(const 
omp_interop_t, omp_interop_property_t);
+/*!
+ * The `omp_get_interop_rc_desc` routine retrieves a description of the 
return code associated with an `omp_interop_t` object.
+ */
+extern const char * __KAI_KMPC_CONVENTION  omp_get_interop_rc_desc(const 
omp_interop_rc_t, omp_interop_rc_t);
+
 /* kmp API functions */
 extern int__KAI_KMPC_CONVENTION  kmp_get_stacksize  (void);
 extern void   __KAI_KMPC_CONVENTION  kmp_set_stacksize  (int);



___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits

[llvm-branch-commits] [openmp] c4a2222 - [OpenMP] Add __kmpc_omp_target_task_alloc to dllexport

2020-12-04 Thread Hansang Bae via llvm-branch-commits 

Author: Hansang Bae
Date: 2020-12-04T08:11:14-06:00
New Revision: c4a4d97120df55f5d1d753e7f839fdb1da2f

URL: 
https://github.com/llvm/llvm-project/commit/c4a4d97120df55f5d1d753e7f839fdb1da2f
DIFF: 
https://github.com/llvm/llvm-project/commit/c4a4d97120df55f5d1d753e7f839fdb1da2f.diff

LOG: [OpenMP] Add __kmpc_omp_target_task_alloc to dllexport

This patch enables use of the entry on Windows.

Differential Revision: https://reviews.llvm.org/D92618

Added: 


Modified: 
openmp/runtime/src/dllexports

Removed: 




diff  --git a/openmp/runtime/src/dllexports b/openmp/runtime/src/dllexports
index da5a0c6e04d2..6e41376a16b9 100644
--- a/openmp/runtime/src/dllexports
+++ b/openmp/runtime/src/dllexports
@@ -387,6 +387,7 @@ kmpc_set_disp_num_buffers   267
 __kmpc_task_allow_completion_event  276
 __kmpc_taskred_init 277
 __kmpc_taskred_modifier_init278
+__kmpc_omp_target_task_alloc279
 %endif
 
 # User API entry points that have both lower- and upper- case versions for 
Fortran.



___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits

[llvm-branch-commits] [openmp] c3b5009 - [OpenMP] Use RTM lock for OMP lock with synchronization hint

2020-12-09 Thread Hansang Bae via llvm-branch-commits 

Author: Hansang Bae
Date: 2020-12-09T19:14:53-06:00
New Revision: c3b5009aa7f42ab8376b39d96e762e2d2e98ab5e

URL: 
https://github.com/llvm/llvm-project/commit/c3b5009aa7f42ab8376b39d96e762e2d2e98ab5e
DIFF: 
https://github.com/llvm/llvm-project/commit/c3b5009aa7f42ab8376b39d96e762e2d2e98ab5e.diff

LOG: [OpenMP] Use RTM lock for OMP lock with synchronization hint

This patch introduces a new RTM lock type based on spin lock which is
used for OMP lock with speculative hint on supported architecture.

Differential Revision: https://reviews.llvm.org/D92615

Added: 


Modified: 
openmp/runtime/src/kmp_csupport.cpp
openmp/runtime/src/kmp_lock.cpp
openmp/runtime/src/kmp_lock.h
openmp/runtime/src/kmp_runtime.cpp
openmp/runtime/src/kmp_settings.cpp

Removed: 




diff  --git a/openmp/runtime/src/kmp_csupport.cpp 
b/openmp/runtime/src/kmp_csupport.cpp
index 1a8db51a667b..fbe7c3d646d6 100644
--- a/openmp/runtime/src/kmp_csupport.cpp
+++ b/openmp/runtime/src/kmp_csupport.cpp
@@ -1249,7 +1249,7 @@ static __forceinline kmp_dyna_lockseq_t 
__kmp_map_hint_to_lock(uintptr_t hint) {
   if (hint & kmp_lock_hint_hle)
 return KMP_TSX_LOCK(hle);
   if (hint & kmp_lock_hint_rtm)
-return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(rtm) : __kmp_user_lock_seq;
+return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(rtm_queuing) : __kmp_user_lock_seq;
   if (hint & kmp_lock_hint_adaptive)
 return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(adaptive) : __kmp_user_lock_seq;
 
@@ -1268,9 +1268,9 @@ static __forceinline kmp_dyna_lockseq_t 
__kmp_map_hint_to_lock(uintptr_t hint) {
   if ((hint & omp_lock_hint_uncontended) && !(hint & 
omp_lock_hint_speculative))
 return lockseq_tas;
 
-  // HLE lock for speculation
+  // Use RTM lock for speculation
   if (hint & omp_lock_hint_speculative)
-return KMP_TSX_LOCK(hle);
+return KMP_CPUINFO_RTM ? KMP_TSX_LOCK(rtm_spin) : __kmp_user_lock_seq;
 
   return __kmp_user_lock_seq;
 }
@@ -1291,6 +1291,7 @@ __ompt_get_mutex_impl_type(void *user_lock, 
kmp_indirect_lock_t *ilock = 0) {
   return kmp_mutex_impl_spin;
 #if KMP_USE_TSX
 case locktag_hle:
+case locktag_rtm_spin:
   return kmp_mutex_impl_speculative;
 #endif
 default:
@@ -1302,7 +1303,7 @@ __ompt_get_mutex_impl_type(void *user_lock, 
kmp_indirect_lock_t *ilock = 0) {
   switch (ilock->type) {
 #if KMP_USE_TSX
   case locktag_adaptive:
-  case locktag_rtm:
+  case locktag_rtm_queuing:
 return kmp_mutex_impl_speculative;
 #endif
   case locktag_nested_tas:
@@ -1336,7 +1337,8 @@ static kmp_mutex_impl_t __ompt_get_mutex_impl_type() {
 return kmp_mutex_impl_queuing;
 #if KMP_USE_TSX
   case lk_hle:
-  case lk_rtm:
+  case lk_rtm_queuing:
+  case lk_rtm_spin:
   case lk_adaptive:
 return kmp_mutex_impl_speculative;
 #endif
@@ -2144,7 +2146,8 @@ __kmp_init_nest_lock_with_hint(ident_t *loc, void **lock,
kmp_dyna_lockseq_t seq) {
 #if KMP_USE_TSX
   // Don't have nested lock implementation for speculative locks
-  if (seq == lockseq_hle || seq == lockseq_rtm || seq == lockseq_adaptive)
+  if (seq == lockseq_hle || seq == lockseq_rtm_queuing ||
+  seq == lockseq_rtm_spin || seq == lockseq_adaptive)
 seq = __kmp_user_lock_seq;
 #endif
   switch (seq) {

diff  --git a/openmp/runtime/src/kmp_lock.cpp b/openmp/runtime/src/kmp_lock.cpp
index 6fa6bf060673..38b43a36fcb8 100644
--- a/openmp/runtime/src/kmp_lock.cpp
+++ b/openmp/runtime/src/kmp_lock.cpp
@@ -2764,20 +2764,22 @@ static int 
__kmp_test_hle_lock_with_checks(kmp_dyna_lock_t *lck,
   return __kmp_test_hle_lock(lck, gtid); // TODO: add checks
 }
 
-static void __kmp_init_rtm_lock(kmp_queuing_lock_t *lck) {
+static void __kmp_init_rtm_queuing_lock(kmp_queuing_lock_t *lck) {
   __kmp_init_queuing_lock(lck);
 }
 
-static void __kmp_destroy_rtm_lock(kmp_queuing_lock_t *lck) {
+static void __kmp_destroy_rtm_queuing_lock(kmp_queuing_lock_t *lck) {
   __kmp_destroy_queuing_lock(lck);
 }
 
-static void __kmp_destroy_rtm_lock_with_checks(kmp_queuing_lock_t *lck) {
+static void
+__kmp_destroy_rtm_queuing_lock_with_checks(kmp_queuing_lock_t *lck) {
   __kmp_destroy_queuing_lock_with_checks(lck);
 }
 
 KMP_ATTRIBUTE_TARGET_RTM
-static void __kmp_acquire_rtm_lock(kmp_queuing_lock_t *lck, kmp_int32 gtid) {
+static void __kmp_acquire_rtm_queuing_lock(kmp_queuing_lock_t *lck,
+   kmp_int32 gtid) {
   unsigned retries = 3, status;
   do {
 status = _xbegin();
@@ -2799,13 +2801,14 @@ static void __kmp_acquire_rtm_lock(kmp_queuing_lock_t 
*lck, kmp_int32 gtid) {
   __kmp_acquire_queuing_lock(lck, gtid);
 }
 
-static void __kmp_acquire_rtm_lock_with_checks(kmp_queuing_lock_t *lck,
-   kmp_int32 gtid) {
-  __kmp_acquire_rtm_lock(lck, gtid);
+static void __kmp_acquire_rtm_queuing_lock_with_checks(kmp_queuing_lock_t *lck,
+