[Openmp-commits] [openmp] Update OpenMP runtime to adopt taskgraph clause from 6.0 Specs (PR #130751)
Josep Pinot via Openmp-commits
openmp-commits at lists.llvm.org
Thu Mar 13 05:21:50 PDT 2025
https://github.com/jpinot updated https://github.com/llvm/llvm-project/pull/130751
>From 6dd9437005956f70c3e566ad007962629e1eeb04 Mon Sep 17 00:00:00 2001
From: jpinot <josep.pinot at bsc.es>
Date: Tue, 11 Mar 2025 14:30:49 +0100
Subject: [PATCH 1/3] [OpenMP] Fix task record/replay comments
---
openmp/runtime/src/kmp_tasking.cpp | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/openmp/runtime/src/kmp_tasking.cpp b/openmp/runtime/src/kmp_tasking.cpp
index 563aa29f6265e..ea123dcb05a63 100644
--- a/openmp/runtime/src/kmp_tasking.cpp
+++ b/openmp/runtime/src/kmp_tasking.cpp
@@ -5452,7 +5452,6 @@ bool __kmpc_omp_has_task_team(kmp_int32 gtid) {
#if OMPX_TASKGRAPH
// __kmp_find_tdg: identify a TDG through its ID
-// gtid: Global Thread ID
// tdg_id: ID of the TDG
// returns: If a TDG corresponding to this ID is found and not
// its initial state, return the pointer to it, otherwise nullptr
@@ -5505,7 +5504,7 @@ void __kmp_print_tdg_dot(kmp_tdg_info_t *tdg, kmp_int32 gtid) {
KA_TRACE(10, ("__kmp_print_tdg_dot(exit): T#%d tdg_id=%d \n", gtid, tdg_id));
}
-// __kmp_start_record: launch the execution of a previous
+// __kmp_exec_tdg: launch the execution of a previous
// recorded TDG
// gtid: Global Thread ID
// tdg: ID of the TDG
>From 29181c9076fe67c98fd84b176c3fe201afcd178b Mon Sep 17 00:00:00 2001
From: jpinot <josep.pinot at bsc.es>
Date: Mon, 10 Mar 2025 12:18:25 +0100
Subject: [PATCH 2/3] [OpenMP] Fix tdg_id in task record and replay mechanism
Previously, tdg_id was used as an array index and an id for the task replay
and record mechanism, but to adder to the OpenMP 6.0 specifications it
should be trated uniquly as a id. This makes the mechanism fully
compliant with OpenMP 6.0 graph_id clause without breaking
compatibility.
With this change, tdg_id more accurately represents an opaque
identifier, accommodating the scenario where the compiler generates a
hashed ID that accounts for the region/graph ID.
---
openmp/runtime/src/kmp.h | 2 +-
openmp/runtime/src/kmp_global.cpp | 3 +-
openmp/runtime/src/kmp_tasking.cpp | 80 ++++++++++++-------
.../tasking/omp_record_replay_random_id.cpp | 47 +++++++++++
4 files changed, 100 insertions(+), 32 deletions(-)
create mode 100644 openmp/runtime/test/tasking/omp_record_replay_random_id.cpp
diff --git a/openmp/runtime/src/kmp.h b/openmp/runtime/src/kmp.h
index 9b8c6102dbee2..811049547bca9 100644
--- a/openmp/runtime/src/kmp.h
+++ b/openmp/runtime/src/kmp.h
@@ -2650,7 +2650,7 @@ typedef struct kmp_tdg_info {
extern int __kmp_tdg_dot;
extern kmp_int32 __kmp_max_tdgs;
extern kmp_tdg_info_t **__kmp_global_tdgs;
-extern kmp_int32 __kmp_curr_tdg_idx;
+extern kmp_tdg_info_t *__kmp_curr_tdg;
extern kmp_int32 __kmp_successors_size;
extern std::atomic<kmp_int32> __kmp_tdg_task_id;
extern kmp_int32 __kmp_num_tdg;
diff --git a/openmp/runtime/src/kmp_global.cpp b/openmp/runtime/src/kmp_global.cpp
index 52e0fdbdfb1da..7b6bfff7c54ea 100644
--- a/openmp/runtime/src/kmp_global.cpp
+++ b/openmp/runtime/src/kmp_global.cpp
@@ -554,8 +554,7 @@ int *__kmp_nesting_nth_level;
int __kmp_tdg_dot = 0;
kmp_int32 __kmp_max_tdgs = 100;
kmp_tdg_info_t **__kmp_global_tdgs = NULL;
-kmp_int32 __kmp_curr_tdg_idx =
- 0; // Id of the current TDG being recorded or executed
+kmp_tdg_info_t *__kmp_curr_tdg = NULL; // Current TDG being recorded or executed
kmp_int32 __kmp_num_tdg = 0;
kmp_int32 __kmp_successors_size = 10; // Initial succesor size list for
// recording
diff --git a/openmp/runtime/src/kmp_tasking.cpp b/openmp/runtime/src/kmp_tasking.cpp
index ea123dcb05a63..761d133997275 100644
--- a/openmp/runtime/src/kmp_tasking.cpp
+++ b/openmp/runtime/src/kmp_tasking.cpp
@@ -1651,11 +1651,11 @@ kmp_task_t *__kmp_task_alloc(ident_t *loc_ref, kmp_int32 gtid,
}
#if OMPX_TASKGRAPH
- kmp_tdg_info_t *tdg = __kmp_find_tdg(__kmp_curr_tdg_idx);
+ kmp_tdg_info_t *tdg = __kmp_curr_tdg;
if (tdg && __kmp_tdg_is_recording(tdg->tdg_status) &&
(task_entry != (kmp_routine_entry_t)__kmp_taskloop_task)) {
taskdata->is_taskgraph = 1;
- taskdata->tdg = __kmp_global_tdgs[__kmp_curr_tdg_idx];
+ taskdata->tdg = tdg;
taskdata->td_task_id = KMP_GEN_TASK_ID();
taskdata->td_tdg_task_id = KMP_ATOMIC_INC(&__kmp_tdg_task_id);
}
@@ -2577,14 +2577,11 @@ without help of the runtime library.
*/
void *__kmpc_task_reduction_init(int gtid, int num, void *data) {
#if OMPX_TASKGRAPH
- kmp_tdg_info_t *tdg = __kmp_find_tdg(__kmp_curr_tdg_idx);
+ kmp_tdg_info_t *tdg = __kmp_curr_tdg;
if (tdg && __kmp_tdg_is_recording(tdg->tdg_status)) {
- kmp_tdg_info_t *this_tdg = __kmp_global_tdgs[__kmp_curr_tdg_idx];
- this_tdg->rec_taskred_data =
- __kmp_allocate(sizeof(kmp_task_red_input_t) * num);
- this_tdg->rec_num_taskred = num;
- KMP_MEMCPY(this_tdg->rec_taskred_data, data,
- sizeof(kmp_task_red_input_t) * num);
+ tdg->rec_taskred_data = __kmp_allocate(sizeof(kmp_task_red_input_t) * num);
+ tdg->rec_num_taskred = num;
+ KMP_MEMCPY(tdg->rec_taskred_data, data, sizeof(kmp_task_red_input_t) * num);
}
#endif
return __kmp_task_reduction_init(gtid, num, (kmp_task_red_input_t *)data);
@@ -2604,14 +2601,11 @@ has two parameters, pointer to object to be initialized and pointer to omp_orig
*/
void *__kmpc_taskred_init(int gtid, int num, void *data) {
#if OMPX_TASKGRAPH
- kmp_tdg_info_t *tdg = __kmp_find_tdg(__kmp_curr_tdg_idx);
+ kmp_tdg_info_t *tdg = __kmp_curr_tdg;
if (tdg && __kmp_tdg_is_recording(tdg->tdg_status)) {
- kmp_tdg_info_t *this_tdg = __kmp_global_tdgs[__kmp_curr_tdg_idx];
- this_tdg->rec_taskred_data =
- __kmp_allocate(sizeof(kmp_task_red_input_t) * num);
- this_tdg->rec_num_taskred = num;
- KMP_MEMCPY(this_tdg->rec_taskred_data, data,
- sizeof(kmp_task_red_input_t) * num);
+ tdg->rec_taskred_data = __kmp_allocate(sizeof(kmp_task_red_input_t) * num);
+ tdg->rec_num_taskred = num;
+ KMP_MEMCPY(tdg->rec_taskred_data, data, sizeof(kmp_task_red_input_t) * num);
}
#endif
return __kmp_task_reduction_init(gtid, num, (kmp_taskred_input_t *)data);
@@ -2662,8 +2656,7 @@ void *__kmpc_task_reduction_get_th_data(int gtid, void *tskgrp, void *data) {
#if OMPX_TASKGRAPH
if ((thread->th.th_current_task->is_taskgraph) &&
- (!__kmp_tdg_is_recording(
- __kmp_global_tdgs[__kmp_curr_tdg_idx]->tdg_status))) {
+ (!__kmp_tdg_is_recording(__kmp_curr_tdg->tdg_status))) {
tg = thread->th.th_current_task->td_taskgroup;
KMP_ASSERT(tg != NULL);
KMP_ASSERT(tg->reduce_data != NULL);
@@ -5464,9 +5457,40 @@ static kmp_tdg_info_t *__kmp_find_tdg(kmp_int32 tdg_id) {
__kmp_global_tdgs = (kmp_tdg_info_t **)__kmp_allocate(
sizeof(kmp_tdg_info_t *) * __kmp_max_tdgs);
- if ((__kmp_global_tdgs[tdg_id]) &&
- (__kmp_global_tdgs[tdg_id]->tdg_status != KMP_TDG_NONE))
- res = __kmp_global_tdgs[tdg_id];
+ for (kmp_int32 tdg_idx = 0; tdg_idx < __kmp_max_tdgs; tdg_idx++) {
+ if (__kmp_global_tdgs[tdg_idx] &&
+ __kmp_global_tdgs[tdg_idx]->tdg_id == tdg_id) {
+ if (__kmp_global_tdgs[tdg_idx]->tdg_status != KMP_TDG_NONE)
+ res = __kmp_global_tdgs[tdg_idx];
+ break;
+ }
+ }
+ return res;
+}
+
+// __kmp_alloc_tdg: Allocates a TDG if it doesn't already exist.
+// tdg_id: ID of the TDG.
+// returns: A pointer to the TDG if it already exists. Otherwise,
+// allocates a new TDG if the maximum limit has not been reached.
+// Returns nullptr if no TDG can be allocated.
+static kmp_tdg_info_t *__kmp_alloc_tdg(kmp_int32 tdg_id) {
+ kmp_tdg_info_t *res = nullptr;
+ if ((res = __kmp_find_tdg(tdg_id)))
+ return res;
+
+ if (__kmp_num_tdg > __kmp_max_tdgs)
+ return res;
+
+ for (kmp_int32 tdg_idx = 0; tdg_idx < __kmp_max_tdgs; tdg_idx++) {
+ if (!__kmp_global_tdgs[tdg_idx]) {
+ kmp_tdg_info_t *tdg =
+ (kmp_tdg_info_t *)__kmp_allocate(sizeof(kmp_tdg_info_t));
+ __kmp_global_tdgs[tdg_idx] = tdg;
+ __kmp_curr_tdg = tdg;
+ res = __kmp_global_tdgs[tdg_idx];
+ break;
+ }
+ }
return res;
}
@@ -5564,9 +5588,7 @@ void __kmp_exec_tdg(kmp_int32 gtid, kmp_tdg_info_t *tdg) {
static inline void __kmp_start_record(kmp_int32 gtid,
kmp_taskgraph_flags_t *flags,
kmp_int32 tdg_id) {
- kmp_tdg_info_t *tdg =
- (kmp_tdg_info_t *)__kmp_allocate(sizeof(kmp_tdg_info_t));
- __kmp_global_tdgs[__kmp_curr_tdg_idx] = tdg;
+ kmp_tdg_info_t *tdg = __kmp_alloc_tdg(tdg_id);
// Initializing the TDG structure
tdg->tdg_id = tdg_id;
tdg->map_size = INIT_MAPSIZE;
@@ -5591,7 +5613,7 @@ static inline void __kmp_start_record(kmp_int32 gtid,
KMP_ATOMIC_ST_RLX(&this_record_map[i].npredecessors_counter, 0);
}
- __kmp_global_tdgs[__kmp_curr_tdg_idx]->record_map = this_record_map;
+ tdg->record_map = this_record_map;
}
// __kmpc_start_record_task: Wrapper around __kmp_start_record to mark
@@ -5599,7 +5621,7 @@ static inline void __kmp_start_record(kmp_int32 gtid,
// loc_ref: Location of TDG, not used yet
// gtid: Global Thread ID of the encountering thread
// input_flags: Flags associated with the TDG
-// tdg_id: ID of the TDG to record, for now, incremental integer
+// tdg_id: ID of the TDG to record
// returns: 1 if we record, otherwise, 0
kmp_int32 __kmpc_start_record_task(ident_t *loc_ref, kmp_int32 gtid,
kmp_int32 input_flags, kmp_int32 tdg_id) {
@@ -5625,8 +5647,7 @@ kmp_int32 __kmpc_start_record_task(ident_t *loc_ref, kmp_int32 gtid,
__kmp_exec_tdg(gtid, tdg);
res = 0;
} else {
- __kmp_curr_tdg_idx = tdg_id;
- KMP_DEBUG_ASSERT(__kmp_curr_tdg_idx < __kmp_max_tdgs);
+ KMP_DEBUG_ASSERT(__kmp_num_tdg < __kmp_max_tdgs);
__kmp_start_record(gtid, flags, tdg_id);
__kmp_num_tdg++;
res = 1;
@@ -5689,10 +5710,11 @@ void __kmpc_end_record_task(ident_t *loc_ref, kmp_int32 gtid,
kmp_int32 input_flags, kmp_int32 tdg_id) {
kmp_tdg_info_t *tdg = __kmp_find_tdg(tdg_id);
+ KMP_DEBUG_ASSERT(tdg != NULL);
KA_TRACE(10, ("__kmpc_end_record_task(enter): T#%d loc=%p finishes recording"
" tdg=%d with flags=%d\n",
gtid, loc_ref, tdg_id, input_flags));
- if (__kmp_max_tdgs) {
+ if (__kmp_max_tdgs && tdg) {
// TODO: use input_flags->nowait
__kmpc_end_taskgroup(loc_ref, gtid);
if (__kmp_tdg_is_recording(tdg->tdg_status))
diff --git a/openmp/runtime/test/tasking/omp_record_replay_random_id.cpp b/openmp/runtime/test/tasking/omp_record_replay_random_id.cpp
new file mode 100644
index 0000000000000..58e90da4d782a
--- /dev/null
+++ b/openmp/runtime/test/tasking/omp_record_replay_random_id.cpp
@@ -0,0 +1,47 @@
+// REQUIRES: ompx_taskgraph
+// RUN: %libomp-cxx-compile-and-run
+#include <iostream>
+#include <cassert>
+#define NT 10
+
+// Compiler-generated code (emulation)
+typedef struct ident {
+ void *dummy;
+} ident_t;
+
+#ifdef __cplusplus
+extern "C" {
+int __kmpc_global_thread_num(ident_t *);
+int __kmpc_start_record_task(ident_t *, int, int, int);
+void __kmpc_end_record_task(ident_t *, int, int, int);
+}
+#endif
+
+static void func(int *num_exec) { (*num_exec)++; }
+
+int main() {
+ int num_exec = 0;
+ int num_tasks = 0;
+ int hash_id = 135343854;
+#pragma omp parallel
+#pragma omp single
+ for (int iter = 0; iter < NT; ++iter) {
+ int gtid = __kmpc_global_thread_num(nullptr);
+ int res = __kmpc_start_record_task(nullptr, gtid, /* kmp_tdg_flags */ 0,
+ /* tdg_id */ hash_id);
+ if (res) {
+ num_tasks++;
+#pragma omp task
+ func(&num_exec);
+ }
+ __kmpc_end_record_task(nullptr, gtid, /* kmp_tdg_flags */ 0,
+ /* tdg_id */ hash_id);
+ }
+
+ assert(num_tasks == 1);
+ assert(num_exec == NT);
+
+ std::cout << "Passed" << std::endl;
+ return 0;
+}
+// CHECK: Passed
>From 6d9d8efccbcf37c7739b85abda1c8435502f3343 Mon Sep 17 00:00:00 2001
From: jpinot <josep.pinot at bsc.es>
Date: Tue, 11 Mar 2025 11:16:43 +0100
Subject: [PATCH 3/3] [OpenMP] Add reset parameter to __kmpc_start_record_task
This patch extends the record/replay infrastructure with the graph_reset
clause introduced in OpenMP 6.0 by adding a new reset parameter to
__kmpc_start_record_task. When reset is true, any existing task dependency
graph (TDG) for the specified ID is freed before re-recording.
---
openmp/runtime/src/kmp.h | 4 +-
openmp/runtime/src/kmp_tasking.cpp | 47 +++++++++++++++----
.../test/tasking/omp_record_replay_reset.cpp | 47 +++++++++++++++++++
3 files changed, 88 insertions(+), 10 deletions(-)
create mode 100644 openmp/runtime/test/tasking/omp_record_replay_reset.cpp
diff --git a/openmp/runtime/src/kmp.h b/openmp/runtime/src/kmp.h
index 811049547bca9..856f14e5f057f 100644
--- a/openmp/runtime/src/kmp.h
+++ b/openmp/runtime/src/kmp.h
@@ -2606,7 +2606,9 @@ typedef struct {
typedef struct kmp_taskgraph_flags { /*This needs to be exactly 32 bits */
unsigned nowait : 1;
unsigned re_record : 1;
- unsigned reserved : 30;
+ unsigned graph_reset : 1; /* 1==discard taskgraph record, 0==use taskgraph
+ record */
+ unsigned reserved : 29;
} kmp_taskgraph_flags_t;
/// Represents a TDG node
diff --git a/openmp/runtime/src/kmp_tasking.cpp b/openmp/runtime/src/kmp_tasking.cpp
index 761d133997275..90004bfc8afe0 100644
--- a/openmp/runtime/src/kmp_tasking.cpp
+++ b/openmp/runtime/src/kmp_tasking.cpp
@@ -5494,6 +5494,34 @@ static kmp_tdg_info_t *__kmp_alloc_tdg(kmp_int32 tdg_id) {
return res;
}
+// __kmp_free_tdg: Frees a TDG if it exists.
+// tdg_id: ID of the TDG to be freed.
+// returns: true if a TDG with the given ID was found and successfully freed,
+// false if no such TDG exists.
+static bool __kmp_free_tdg(kmp_int32 tdg_id) {
+ kmp_tdg_info_t *tdg = nullptr;
+ if (__kmp_global_tdgs == NULL)
+ return false;
+
+ for (kmp_int32 tdg_idx = 0; tdg_idx < __kmp_max_tdgs; tdg_idx++) {
+ if (__kmp_global_tdgs[tdg_idx] &&
+ __kmp_global_tdgs[tdg_idx]->tdg_id == tdg_id) {
+ tdg = __kmp_global_tdgs[tdg_idx];
+ for (kmp_int map_idx = 0; map_idx < tdg->map_size; map_idx++) {
+ __kmp_free(tdg->record_map[map_idx].successors);
+ }
+ __kmp_free(tdg->record_map);
+ if (tdg->root_tasks)
+ __kmp_free(tdg->root_tasks);
+
+ __kmp_free(tdg);
+ __kmp_global_tdgs[tdg_idx] = NULL;
+ return true;
+ }
+ }
+ return false;
+}
+
// __kmp_print_tdg_dot: prints the TDG to a dot file
// tdg: ID of the TDG
// gtid: Global Thread ID
@@ -5625,23 +5653,24 @@ static inline void __kmp_start_record(kmp_int32 gtid,
// returns: 1 if we record, otherwise, 0
kmp_int32 __kmpc_start_record_task(ident_t *loc_ref, kmp_int32 gtid,
kmp_int32 input_flags, kmp_int32 tdg_id) {
-
kmp_int32 res;
kmp_taskgraph_flags_t *flags = (kmp_taskgraph_flags_t *)&input_flags;
- KA_TRACE(10,
- ("__kmpc_start_record_task(enter): T#%d loc=%p flags=%d tdg_id=%d\n",
- gtid, loc_ref, input_flags, tdg_id));
+ KA_TRACE(10, ("__kmpc_start_record_task(enter): T#%d loc=%p flags=%d "
+ "tdg_id=%d\n",
+ gtid, loc_ref, input_flags, tdg_id));
if (__kmp_max_tdgs == 0) {
- KA_TRACE(
- 10,
- ("__kmpc_start_record_task(abandon): T#%d loc=%p flags=%d tdg_id = %d, "
- "__kmp_max_tdgs = 0\n",
- gtid, loc_ref, input_flags, tdg_id));
+ KA_TRACE(10, ("__kmpc_start_record_task(abandon): T#%d loc=%p flags=%d "
+ "tdg_id = %d, __kmp_max_tdgs = 0\n",
+ gtid, loc_ref, input_flags, tdg_id));
return 1;
}
__kmpc_taskgroup(loc_ref, gtid);
+ if (flags->graph_reset) {
+ __kmp_free_tdg(tdg_id);
+ __kmp_num_tdg--;
+ }
if (kmp_tdg_info_t *tdg = __kmp_find_tdg(tdg_id)) {
// TODO: use re_record flag
__kmp_exec_tdg(gtid, tdg);
diff --git a/openmp/runtime/test/tasking/omp_record_replay_reset.cpp b/openmp/runtime/test/tasking/omp_record_replay_reset.cpp
new file mode 100644
index 0000000000000..123a9fa5a72f0
--- /dev/null
+++ b/openmp/runtime/test/tasking/omp_record_replay_reset.cpp
@@ -0,0 +1,47 @@
+// REQUIRES: ompx_taskgraph
+// RUN: %libomp-cxx-compile-and-run
+#include <iostream>
+#include <cassert>
+#define NT 10
+
+// Compiler-generated code (emulation)
+typedef struct ident {
+ void *dummy;
+} ident_t;
+
+#ifdef __cplusplus
+extern "C" {
+int __kmpc_global_thread_num(ident_t *);
+int __kmpc_start_record_task(ident_t *, int, int, int);
+void __kmpc_end_record_task(ident_t *, int, int, int);
+}
+#endif
+
+static void func(int *num_exec) { (*num_exec)++; }
+
+int main() {
+ int num_exec = 0;
+ int num_tasks = 0;
+ int flags = 1 << 2;
+#pragma omp parallel
+#pragma omp single
+ for (int iter = 0; iter < NT; ++iter) {
+ int gtid = __kmpc_global_thread_num(nullptr);
+ int res = __kmpc_start_record_task(nullptr, gtid, /* kmp_tdg_flags */ flags,
+ /* tdg_id */ 0);
+ if (res) {
+ num_tasks++;
+#pragma omp task
+ func(&num_exec);
+ }
+ __kmpc_end_record_task(nullptr, gtid, /* kmp_tdg_flags */ 0,
+ /* tdg_id */ 0);
+ }
+
+ assert(num_tasks == NT);
+ assert(num_exec == NT);
+
+ std::cout << "Passed" << std::endl;
+ return 0;
+}
+// CHECK: Passed
More information about the Openmp-commits
mailing list