1// Copyright (c) 2021 Qualcomm Technologies, Inc. 2// 3// SPDX-License-Identifier: CC-BY-4.0 4 5include::{generated}/meta/{refprefix}VK_QCOM_image_processing.adoc[] 6 7=== Other Extension Metadata 8 9*Last Modified Date*:: 10 2022-07-08 11*Interactions and External Dependencies*:: 12 - This extension provides API support for 13 {GLSLregistry}/qcom/GLSL_QCOM_image_processing.txt[`GL_QCOM_image_processing`] 14 15*Contributors*:: 16 - Jeff Leger, Qualcomm Technologies, Inc. 17 - Ruihao Zhang, Qualcomm Technologies, Inc. 18 19=== Description 20 21GPUs are commonly used to process images for various applications from 3D 22graphics to UI and from composition to compute applications. 23Simple scaling and filtering can be done with bilinear filtering, which 24comes for free during texture sampling. 25However, as screen sizes get larger and more use cases rely on GPU such as 26camera and video post-processing needs, there is increasing demand for GPU 27to support higher order filtering and other advanced image processing. 28 29This extension introduces a new set of SPIR-V built-in functions for image 30processing. 31It exposes the following new imaging operations 32 33 * The `OpImageSampleWeightedQCOM` instruction takes 3 operands: _sampled 34 image_, _weight image_, and texture coordinates. 35 The instruction computes a weighted average of an MxN region of texels 36 in the _sampled image_, using a set of MxN weights in the _weight image_. 37 * The `OpImageBoxFilterQCOM` instruction takes 3 operands: _sampled 38 image_, _box size_, and texture coordinates. 39 Note that _box size_ specifies a floating point width and height in 40 texels. 41 The instruction computes a weighted average of all texels in the 42 _sampled image_ that are covered (either partially or fully) by a box 43 with the specified size and centered at the specified texture 44 coordinates. 45 * The `OpImageBlockMatchSADQCOM` and `OpImageBlockMatchSSDQCOM` 46 instructions each takes 5 operands: _target image_, _target 47 coordinates_, _reference image_, _reference coordinates_, and _block 48 size_. 49 Each instruction computes an error metric, that describes whether a 50 block of texels in the _target image_ matches a corresponding block of 51 texels in the _reference image_. 52 The error metric is computed per-component. 53 `OpImageBlockMatchSADQCOM` computes "Sum Of Absolute Difference" and 54 `OpImageBlockMatchSSDQCOM` computes "Sum of Squared Difference". 55 56Each of the image processing instructions operate only on 2D images. 57The instructions do not-support sampling of mipmap, multi-plane, 58multi-layer, multi-sampled, or depth/stencil images. 59The instructions can be used in any shader stage. 60 61Implementations of this this extension should support these operations 62natively at the HW instruction level, offering potential performance gains 63as well as ease of development. 64 65include::{generated}/interfaces/VK_QCOM_image_processing.adoc[] 66 67=== Version History 68 69 * Revision 1, 2022-07-08 (Jeff Leger) 70