26# include <tbb/blocked_range.h>
27# include <tbb/enumerable_thread_specific.h>
28# include <tbb/parallel_for.h>
29# include <tbb/parallel_reduce.h>
40 const TaskParallelSettings *settings;
45 RangeTask(TaskParallelRangeFunc func,
void *userdata,
const TaskParallelSettings *settings)
46 : func(func), userdata(userdata), settings(settings)
48 init_chunk(settings->userdata_chunk);
52 RangeTask(
const RangeTask &other)
53 : func(other.func), userdata(other.userdata), settings(other.settings)
55 init_chunk(settings->userdata_chunk);
59 RangeTask(RangeTask &other, tbb::split )
60 : func(other.func), userdata(other.userdata), settings(other.settings)
62 init_chunk(settings->userdata_chunk);
67 if (settings->func_free !=
nullptr && userdata_chunk !=
nullptr) {
68 settings->func_free(userdata, userdata_chunk);
70 MEM_SAFE_DELETE_VOID(userdata_chunk);
73 void init_chunk(
void *from_chunk)
77 memcpy(userdata_chunk, from_chunk, settings->userdata_chunk_size);
80 userdata_chunk =
nullptr;
84 void operator()(
const tbb::blocked_range<int> &r)
const
87 tls.userdata_chunk = userdata_chunk;
88 for (
int i = r.begin(); i != r.end(); ++i) {
89 func(userdata, i, &tls);
93 void join(
const RangeTask &other)
95 settings->func_reduce(userdata, userdata_chunk, other.userdata_chunk);
110 RangeTask task(func, userdata, settings);
112 const tbb::blocked_range<int> range(start, stop, grainsize);
117 parallel_reduce(range, task);
123 parallel_for(range, task);
133 for (
int i = start; i < stop; i++) {
134 func(userdata, i, &tls);
147 static tbb::enumerable_thread_specific<int> tbb_thread_id(-1);
148 static int tbb_thread_id_counter = 0;
150 int &thread_id = tbb_thread_id.local();
151 if (thread_id == -1) {
154 BLI_assert_msg(0,
"Maximum number of threads exceeded for sculpting");
164namespace threading::detail {
167static void parallel_for_impl_static_size(
const IndexRange range,
169 const FunctionRef<
void(IndexRange)> function)
171 tbb::parallel_for(tbb::blocked_range<int64_t>(range.first(), range.one_after_last(), grain_size),
172 [function](
const tbb::blocked_range<int64_t> &subrange) {
173 function(IndexRange(subrange.begin(), subrange.size()));
179static void parallel_for_impl_individual_size_lookup(
180 const IndexRange range,
182 const FunctionRef<
void(IndexRange)> function,
188 const int64_t outer_grain_size = std::min<int64_t>(grain_size, 512);
191 Array<int64_t, 1024> task_sizes(sub_range.size());
192 size_hints.lookup_individual_sizes(sub_range, task_sizes);
195 Vector<int64_t, 256> offsets_vec;
196 offsets_vec.append(0);
198 for (
const int64_t i : sub_range.index_range()) {
199 counter += task_sizes[i];
200 if (counter >= grain_size) {
201 offsets_vec.append(i + 1);
205 if (offsets_vec.last() < sub_range.size()) {
206 offsets_vec.append(sub_range.size());
208 const OffsetIndices<int64_t> offsets = offsets_vec.as_span();
212 for (const int64_t i : offsets_range) {
213 const IndexRange actual_range = offsets[i].shift(sub_range.start());
214 function(actual_range);
229 if (range.
size() == 1) {
235 if (total_size <= grain_size) {
258 switch (size_hints.
type) {
261 const int64_t final_grain_size = task_size == 1 ?
263 std::max<int64_t>(1, grain_size / task_size);
264 parallel_for_impl_static_size(range, final_grain_size, function);
268 parallel_for_impl_individual_size_lookup(
300 const int num_threads = 8;
306 static tbb::task_arena arena{num_threads};
313 arena.execute(function);
329 tbb::task_arena arena{max_threads};
333 arena.execute(function);
#define BLENDER_MAX_THREADS
Read Blender memory allocation and freeing.
Provides wrapper around system-specific atomic primitives, and some extensions (faked-atomic operatio...
ATOMIC_INLINE int32_t atomic_fetch_and_add_int32(int32_t *p, int32_t x)
#define BLI_assert_msg(a, msg)
constexpr int64_t size() const
constexpr bool is_empty() const
constexpr IndexRange take_front(int64_t n) const
constexpr IndexRange drop_front(int64_t n) const
virtual int64_t lookup_accumulated_size(IndexRange range) const =0
void * MEM_new_uninitialized(size_t len, const char *str)
void parallel_for_impl(IndexRange range, int64_t grain_size, FunctionRef< void(IndexRange)> function, const TaskSizeHints &size_hints)
void max_threads_task_impl(int max_threads, FunctionRef< void()> function)
void memory_bandwidth_bound_task_impl(FunctionRef< void()> function)
static void parallel_for_impl_accumulated_size_lookup(const IndexRange range, const int64_t grain_size, const FunctionRef< void(IndexRange)> function, const TaskSizeHints_AccumulatedLookup &size_hints)
void parallel_invoke(Functions &&...functions)
void parallel_for(const IndexRange range, const int64_t grain_size, const Function &function, const TaskSizeHints &size_hints=detail::TaskSizeHints_Static(1))
BLI_assert(slot->slot_subtype.map==BMO_OP_SLOT_SUBTYPE_MAP_FLT)
void(*) TaskParallelRangeFunc(void *__restrict userdata, int iter, const TaskParallelTLS *__restrict tls)
int BLI_task_scheduler_num_threads()
void BLI_task_parallel_range(int start, int stop, void *userdata, TaskParallelRangeFunc func, const TaskParallelSettings *settings)
int BLI_task_parallel_thread_id(const TaskParallelTLS *tls)
TaskParallelReduceFunc func_reduce
size_t userdata_chunk_size
TaskParallelFreeFunc func_free