Fix manual LOOP_UNROLLING
The issue is caused by the number of iterations passed to
LOOP_UNROLLING. When we use the manual LOOP_UNROLLING, the number of
iterations must be less than or equal to 128.
To overcome this problem, we create a utility function to check if
any of the critical iterations (kernel dimensions) are beyond that
limit. If so, the utility function, disable the manual loop unrolling.
Resolves COMPMID-4609
Change-Id: I7221c967609e462a5abd1cbb74e2a120f344fcb3
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5913
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/arm_compute/core/CL/CLHelpers.h b/arm_compute/core/CL/CLHelpers.h
index 180211c..ce7e675 100644
--- a/arm_compute/core/CL/CLHelpers.h
+++ b/arm_compute/core/CL/CLHelpers.h
@@ -46,6 +46,9 @@
/** Max vector width of an OpenCL vector */
static constexpr unsigned int max_cl_vector_width = 16;
+/** Max number of manual loop unrolling */
+static constexpr int max_manual_loop_unrolling = 128;
+
/** Translates a tensor data type to the appropriate OpenCL type.
*
* @param[in] dt @ref DataType to be translated to OpenCL type.
@@ -244,5 +247,15 @@
*/
bool export_weights_to_cl_image(const ITensorInfo *tensor);
+/* Helper function to force unroll with pragma when any of the input values (iterations) are greater than @ref max_manual_loop_unrolling
+ *
+ * This function passes UNROLL_WITH_PRAGMA at compile time when any of the input values are greater than @ref max_manual_loop_unrolling
+ *
+ * @param[in] built_opts OpenCL kernel build options
+ * @param[in] values Input values (iterations)
+ *
+ */
+void set_unroll_with_pragma(CLBuildOptions &built_opts, std::initializer_list<int> values);
+
} // namespace arm_compute
#endif /* ARM_COMPUTE_CLHELPERS_H */