commit a1422fbf985c89ffebc8f5af8093e9cd987cfe29
author Michele Di Giorgio <michele.digiorgio@arm.com> Wed Oct 24 12:20:19 2018 +0100
committer Anthony Barbier <anthony.barbier@arm.com> Fri Nov 02 16:55:45 2018 +0000
tree 4536c22bb96859e8d80cc0c11e4d106b780b7539
parent 33893c3e5a8d298f1a9fcc36ab89b73382fc1245

COMPMID-1673: Collapse window in CLArithmeticAddition when one operand is a vector

When one of the operands is a vector, the kernel does a broadcast addition and
the window is not collapsed. This represent an issue because it leads to a lot
of enqueues that increases the time taken by the OpenCL driver. This patch
allows to collapse the window when one of the two operands is a vector.

Furthermore, it adds LWS tuner to the kernel.

It also changes the number of elements processed per iteration to 8 to make
better usage of the cache.

Change-Id: I5f09ab0ddcffb3b7f9326a987c79a997b2d7fa8c
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/155003
Reviewed-by: Giuseppe Rossini <giuseppe.rossini@arm.com>
Tested-by: bsgcomp <bsgcomp@arm.com>

tests/validation/CL/ArithmeticAddition.cpp

diff --git a/tests/validation/CL/ArithmeticAddition.cpp b/tests/validation/CL/ArithmeticAddition.cpp
index e327769..09f1b7c 100644
--- a/tests/validation/CL/ArithmeticAddition.cpp
+++ b/tests/validation/CL/ArithmeticAddition.cpp
@@ -43,6 +43,7 @@
 {
 namespace
 {
+constexpr unsigned int num_elems_processed_per_iteration = 8;
 /** Input data sets **/
 const auto ArithmeticAdditionU8Dataset = combine(combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::U8)), framework::dataset::make("DataType",
                                                  DataType::U8));
@@ -110,7 +111,7 @@
     validate(dst.info()->valid_region(), valid_region);
 
     // Validate padding
-    const PaddingSize padding = PaddingCalculator(shape.x(), 16).required_padding();
+    const PaddingSize padding = PaddingCalculator(shape.x(), num_elems_processed_per_iteration).required_padding();
     validate(ref_src1.info()->padding(), padding);
     validate(ref_src2.info()->padding(), padding);
     validate(dst.info()->padding(), padding);
@@ -146,7 +147,7 @@
     validate(dst.info()->valid_region(), valid_region);
 
     // Validate padding
-    const PaddingSize padding = PaddingCalculator(shape.x(), 16).required_padding();
+    const PaddingSize padding = PaddingCalculator(shape.x(), num_elems_processed_per_iteration).required_padding();
     validate(ref_src1.info()->padding(), padding);
     validate(ref_src2.info()->padding(), padding);
     validate(dst.info()->padding(), padding);
@@ -186,7 +187,7 @@
     validate(dst.info()->valid_region(), valid_region);
 
     // Validate padding
-    const PaddingSize padding = PaddingCalculator(shape.x(), 16).required_padding();
+    const PaddingSize padding = PaddingCalculator(shape.x(), num_elems_processed_per_iteration).required_padding();
     validate(ref_src1.info()->padding(), padding);
     validate(ref_src2.info()->padding(), padding);
     validate(dst.info()->padding(), padding);
@@ -235,7 +236,7 @@
     validate(dst.info()->valid_region(), valid_region);
 
     // Validate padding
-    const PaddingSize padding = PaddingCalculator(shape.x(), 16).required_padding();
+    const PaddingSize padding = PaddingCalculator(shape.x(), num_elems_processed_per_iteration).required_padding();
     validate(ref_src1.info()->padding(), padding);
     validate(ref_src2.info()->padding(), padding);
     validate(dst.info()->padding(), padding);