diff options
author | Dominic Symes <dominic.symes@arm.com> | 2023-07-27 11:50:57 +0100 |
---|---|---|
committer | Dominic Symes <dominic.symes@arm.com> | 2023-08-16 16:33:20 +0000 |
commit | b5b067819e5de11153b41cf3d26da4f3f9dd23e8 (patch) | |
tree | a347a7b4d7f89d5706ced611c3d60e49c1ddc5da /tosa.xml | |
parent | 830b43b1d1bd82edd57dee1f5cac12e2b5cf0e04 (diff) | |
download | specification-b5b067819e5de11153b41cf3d26da4f3f9dd23e8.tar.gz |
Main conformance: Add local bound flag for convolutions
Adds a local_bound flag to convolution operations.
If the local_bound flag is true then the output accuracy
for floating-point is measured relative to the
input tensor data local to the particular convolution output.
If the local_bound flag is false then the output accuracy
for floating-point is measured relative to the whole input tensor.
This accuracy measure is the default and more approapriate for
implementations with non local optimizations (such as
transform based fast convolutions).
Signed-off-by: Dominic Symes <dominic.symes@arm.com>
Change-Id: I64e3e4981a63e26e6391149e28d5d71e7ef5560a
Diffstat (limited to 'tosa.xml')
-rw-r--r-- | tosa.xml | 40 |
1 files changed, 40 insertions, 0 deletions
@@ -159,6 +159,14 @@ <description>Weight zero point. Must be zero for non-int8 types.</description> <rank min="0" max="0"/> </argument> + <argument category="attribute" name="local_bound" type="tensor_t" shape="-" tensor-element-type="bool_t" optional="true"> + <description> + This optional attribute affects the floating-point compliance error bound. + The default of false allows for direct and transform based, fast convolution algorithms. + Only set to true if direct dot-product calculation precision is required. + </description> + <rank min="0" max="0"/> + </argument> <argument category="output" name="output" type="tensor_t" shape="[N,OH,OW,OC]" tensor-element-type="out_t"> <description>Output tensor</description> <rank min="4" max="4"/> @@ -236,6 +244,14 @@ <description>Weight zero point. Must be zero for non-int8 types.</description> <rank min="0" max="0"/> </argument> + <argument category="attribute" name="local_bound" type="tensor_t" shape="-" tensor-element-type="bool_t" optional="true"> + <description> + This optional attribute affects the floating-point compliance error bound. + The default of false allows for direct and transform based, fast convolution algorithms. + Only set to true if direct dot-product calculation precision is required. + </description> + <rank min="0" max="0"/> + </argument> <argument category="output" name="output" type="tensor_t" shape="[N,OD,OH,OW,OC]" tensor-element-type="out_t"> <description>Output tensor</description> <rank min="5" max="5"/> @@ -309,6 +325,14 @@ <description>Weight zero point. Must be zero for non-int8 types.</description> <rank min="0" max="0"/> </argument> + <argument category="attribute" name="local_bound" type="tensor_t" shape="-" tensor-element-type="bool_t" optional="true"> + <description> + This optional attribute affects the floating-point compliance error bound. + The default of false allows for direct and transform based, fast convolution algorithms. + Only set to true if direct dot-product calculation precision is required. + </description> + <rank min="0" max="0"/> + </argument> <argument category="output" name="output" type="tensor_t" shape="[N,OH,OW,C*M]" tensor-element-type="out_t"> <description>Output tensor</description> <rank min="4" max="4"/> @@ -360,6 +384,14 @@ <description>Real part of the complex output.</description> <rank min="3" max="3"/> </argument> + <argument category="attribute" name="local_bound" type="tensor_t" shape="-" tensor-element-type="bool_t" optional="true"> + <description> + This optional attribute affects the floating-point compliance error bound. + The default of false allows for direct and transform based, fast convolution algorithms. + Only set to true if direct dot-product calculation precision is required. + </description> + <rank min="0" max="0"/> + </argument> <argument category="output" name="output_imag" type="tensor_t" shape="[N,H,W]" tensor-element-type="in_out_t"> <description>Imaginary part of the complex output.</description> <rank min="3" max="3"/> @@ -592,6 +624,14 @@ <description>Weight zero point. Must be zero for non-int8 types.</description> <rank min="0" max="0"/> </argument> + <argument category="attribute" name="local_bound" type="tensor_t" shape="-" tensor-element-type="bool_t" optional="true"> + <description> + This optional attribute affects the floating-point compliance error bound. + The default of false allows for direct and transform based, fast convolution algorithms. + Only set to true if direct dot-product calculation precision is required. + </description> + <rank min="0" max="0"/> + </argument> <argument category="output" name="output" type="tensor_t" shape="[N,OH,OW,OC]" tensor-element-type="out_t"> <description>Output tensor</description> <rank min="4" max="4"/> |