aboutsummaryrefslogtreecommitdiff
path: root/src/runtime/SchedulerUtils.cpp
diff options
context:
space:
mode:
authormorgolock <pablo.tello@arm.com>2020-08-20 14:51:39 +0100
committerPablo Marquez <pablo.tello@arm.com>2020-08-25 13:17:09 +0000
commit5111264954e2d1a4d3e91d23a0869a0d7105be4c (patch)
tree57c8db6a8911378129364bc6d826a3c0ef7f9283 /src/runtime/SchedulerUtils.cpp
parent96a14008af85725d067cdd8247023474581102ea (diff)
downloadComputeLibrary-5111264954e2d1a4d3e91d23a0869a0d7105be4c.tar.gz
COMPMID-3661: Added multidimension support to OMP scheduler.
Change-Id: Iedacf7094896f08d7c2847c8fb99bd7153deba2c Signed-off-by: morgolock <pablo.tello@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3809 Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Diffstat (limited to 'src/runtime/SchedulerUtils.cpp')
-rw-r--r--src/runtime/SchedulerUtils.cpp79
1 files changed, 79 insertions, 0 deletions
diff --git a/src/runtime/SchedulerUtils.cpp b/src/runtime/SchedulerUtils.cpp
new file mode 100644
index 0000000000..1c12e3ce58
--- /dev/null
+++ b/src/runtime/SchedulerUtils.cpp
@@ -0,0 +1,79 @@
+/*
+ * Copyright (c) 2020 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "arm_compute/core/Error.h"
+
+#include <cmath>
+
+namespace arm_compute
+{
+#ifndef BARE_METAL
+std::pair<unsigned, unsigned> split_2d(unsigned max_threads, std::size_t m, std::size_t n)
+{
+ /*
+ * We want the same ratio of threads in M & N to the ratio of m and n problem size
+ *
+ * Therefore: mt/nt == m/n where mt*nt == max_threads
+ *
+ * max_threads/nt = mt & (max_threads/nt) * (m/n) = nt
+ * nt^2 = max_threads * (m/n)
+ * nt = sqrt( max_threads * (m/n) )
+ */
+ //ratio of m to n in problem dimensions
+ double ratio = m / static_cast<double>(n);
+
+ // nt = sqrt(max_threads * (m / n) )
+ const unsigned adjusted = std::round(
+ std::sqrt(max_threads * ratio));
+
+ //find the nearest factor of max_threads
+ for(unsigned i = 0; i != adjusted; ++i)
+ {
+ //try down
+ const unsigned adj_down = adjusted - i;
+ if(max_threads % adj_down == 0)
+ {
+ return { adj_down, max_threads / adj_down };
+ }
+
+ //try up
+ const unsigned adj_up = adjusted + i;
+ if(max_threads % adj_up == 0)
+ {
+ return { adj_up, max_threads / adj_up };
+ }
+ }
+
+ //we didn't find anything so lets bail out with maxes biased to the largest dimension
+ if(m > n)
+ {
+ return { std::min<unsigned>(m, max_threads), 1 };
+ }
+ else
+ {
+ return { 1, std::min<unsigned>(n, max_threads) };
+ }
+}
+#endif /* #ifndef BARE_METAL */
+} // namespace arm_compute