Document the default alignment and padding value.
This CL documents the default value of the preferred alignment and
padding for request memories.
Bug: 179799921
Test: m
Change-Id: I89c16c3e20939a0ad68d9e3b0061c4a16dc00d8b
Merged-In: I89c16c3e20939a0ad68d9e3b0061c4a16dc00d8b
(cherry picked from commit e6dcc9a38d3fbc2cbd957577657493e36783d3ed)
diff --git a/neuralnetworks/aidl/android/hardware/neuralnetworks/DataLocation.aidl b/neuralnetworks/aidl/android/hardware/neuralnetworks/DataLocation.aidl
index 1b2378f..f656360 100644
--- a/neuralnetworks/aidl/android/hardware/neuralnetworks/DataLocation.aidl
+++ b/neuralnetworks/aidl/android/hardware/neuralnetworks/DataLocation.aidl
@@ -35,6 +35,11 @@
* total size of the writable region of the output data, and padding specifies the extra bytes at
* the end of the memory region that may be used by the device to access memory in chunks, for
* efficiency, but must not be used to hold any output data.
+ *
+ * When used in RequestArgument, clients should prefer to align and pad the sub-region to
+ * 64 bytes when possible; this may allow the device to access the sub-region more efficiently.
+ * The sub-region is aligned to 64 bytes if the value of offset is a multiple of 64.
+ * The sub-region is padded to 64 bytes if the sum of length and padding is a multiple of 64.
*/
@VintfStability
parcelable DataLocation {