IntenseWebs/openvino
3
0 Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

update number of threads per stream on Ecore to 2 for aggressive model on hybrid platform (#16857)

Browse Source 
* update number of threads per stream on Ecore to 2 when  aggressive model run on hybrid platform

* update for corner case and add test case
This commit is contained in:
Shen, Wanglei authored and GitHub committed 2023-04-17 18:42:55 +08:00
1 parent 816c0f76e2
commit 98393c0da1
2 files changed
+66 -4
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/plugins/intel_cpu/src/cpu_streams_calculation.cpp
+11
View File
@@ -137,6 +137,14 @@ std::vector<std::vector<int>> get_streams_info_table(const int input_streams,
}
}
}
} else if ((1 == model_prefer_threads) && (proc_type_table[0][EFFICIENT_CORE_PROC] > 0) &&
(proc_type_table[0][MAIN_CORE_PROC] > 0) && (n_threads > proc_type_table[0][MAIN_CORE_PROC])) {
n_streams = (n_threads >= proc_type_table[0][MAIN_CORE_PROC] + proc_type_table[0][EFFICIENT_CORE_PROC])
? static_cast<int>(n_threads - proc_type_table[0][EFFICIENT_CORE_PROC] / 2)
: static_cast<int>(proc_type_table[0][MAIN_CORE_PROC] +
(n_threads - proc_type_table[0][MAIN_CORE_PROC]) / 2);
n_streams = (input_infer_requests > 0) ? std::min(n_streams, input_infer_requests) : n_streams;
n_threads_per_stream = -1;
} else {
n_streams = ((n_threads + model_prefer_threads - 1) / model_prefer_threads);
n_streams = (input_infer_requests > 0) ? std::min(n_streams, input_infer_requests) : n_streams;
@@ -150,6 +158,9 @@ std::vector<std::vector<int>> get_streams_info_table(const int input_streams,
while (1) {
for (int n = MAIN_CORE_PROC; n < PROC_TYPE_TABLE_SIZE; n++) {
if (0 != proc_type_table[0][n]) {
if (n_threads_per_stream == -1) {
stream_info[THREADS_PER_STREAM] = (n == EFFICIENT_CORE_PROC) ? 2 : 1;
}
stream_info[PROC_TYPE] = n;
stream_info[NUMBER_OF_STREAMS] =
static_cast<int>(proc_type_table[0][n] / stream_info[THREADS_PER_STREAM]);
src/plugins/intel_cpu/tests/unit/streams_info_table_test.cpp
+55 -4
View File
@@ -463,7 +463,7 @@ StreamsCalculationTestCase _1sockets_14cores_tput_5 = {
0,
1,
{{20, 6, 8, 6}},
{{6, MAIN_CORE_PROC, 1}, {8, EFFICIENT_CORE_PROC, 1}, {6, HYPER_THREADING_PROC, 1}},
{{6, MAIN_CORE_PROC, 1}, {4, EFFICIENT_CORE_PROC, 2}, {6, HYPER_THREADING_PROC, 1}},
};
StreamsCalculationTestCase _1sockets_14cores_tput_6 = {
@@ -529,6 +529,43 @@ StreamsCalculationTestCase _1sockets_14cores_tput_12 = {
{{1, MAIN_CORE_PROC, 6}, {1, EFFICIENT_CORE_PROC, 6}},
};
StreamsCalculationTestCase _1sockets_14cores_tput_13 = {
0,
1,
0,
1,
{{20, 6, 8, 6}},
{{1, MAIN_CORE_PROC, 1}},
};
StreamsCalculationTestCase _1sockets_14cores_tput_14 = {
0,
9,
0,
1,
{{20, 6, 8, 6}},
{{6, MAIN_CORE_PROC, 1}, {1, EFFICIENT_CORE_PROC, 2}},
};
StreamsCalculationTestCase _1sockets_14cores_tput_15 = {
0,
12,
0,
1,
{{20, 6, 8, 6}},
{{6, MAIN_CORE_PROC, 1}, {3, EFFICIENT_CORE_PROC, 2}},
};
StreamsCalculationTestCase _1sockets_14cores_tput_16 = {
0,
15,
0,
1,
{{20, 6, 8, 6}},
{{6, MAIN_CORE_PROC, 1}, {4, EFFICIENT_CORE_PROC, 2}, {1, HYPER_THREADING_PROC, 1}},
};
StreamsCalculationTestCase _1sockets_10cores_latency_1 = {
1,
0,
@@ -607,7 +644,7 @@ StreamsCalculationTestCase _1sockets_10cores_tput_5 = {
0,
1,
{{12, 2, 8, 2}},
{{2, MAIN_CORE_PROC, 1}, {8, EFFICIENT_CORE_PROC, 1}, {2, HYPER_THREADING_PROC, 1}},
{{2, MAIN_CORE_PROC, 1}, {4, EFFICIENT_CORE_PROC, 2}, {2, HYPER_THREADING_PROC, 1}},
};
StreamsCalculationTestCase _1sockets_10cores_tput_6 = {
@@ -715,7 +752,7 @@ StreamsCalculationTestCase _1sockets_8cores_tput_7 = {
0,
1,
{{12, 4, 4, 4}},
{{4, MAIN_CORE_PROC, 1}, {4, EFFICIENT_CORE_PROC, 1}, {4, HYPER_THREADING_PROC, 1}},
{{4, MAIN_CORE_PROC, 1}, {2, EFFICIENT_CORE_PROC, 2}, {4, HYPER_THREADING_PROC, 1}},
};
StreamsCalculationTestCase _1sockets_6cores_latency_1 = {
@@ -853,6 +890,15 @@ StreamsCalculationTestCase _1sockets_ecores_tput_5 = {
{{2, EFFICIENT_CORE_PROC, 8}},
};
StreamsCalculationTestCase _1sockets_mock_tput_1 = {
0,
15,
0,
1,
{{20, 6, 7, 6}},
{{6, MAIN_CORE_PROC, 1}, {3, EFFICIENT_CORE_PROC, 2}, {3, HYPER_THREADING_PROC, 1}},
};
TEST_P(StreamsCalculationTests, StreamsCalculation) {}
INSTANTIATE_TEST_SUITE_P(StreamsInfoTable,
@@ -894,6 +940,10 @@ INSTANTIATE_TEST_SUITE_P(StreamsInfoTable,
_1sockets_14cores_tput_10,
_1sockets_14cores_tput_11,
_1sockets_14cores_tput_12,
_1sockets_14cores_tput_13,
_1sockets_14cores_tput_14,
_1sockets_14cores_tput_15,
_1sockets_14cores_tput_16,
_1sockets_10cores_latency_1,
_1sockets_10cores_latency_2,
_1sockets_10cores_latency_3,
@@ -929,6 +979,7 @@ INSTANTIATE_TEST_SUITE_P(StreamsInfoTable,
_1sockets_ecores_tput_2,
_1sockets_ecores_tput_3,
_1sockets_ecores_tput_4,
_1sockets_ecores_tput_5));
_1sockets_ecores_tput_5,
_1sockets_mock_tput_1));
} // namespace