Input filtering and sampler acceleration for supersampling
Input filtering and sampler acceleration for supersampling is described. An example of a graphics processor comprises a set of processing resources configured to perform a supersampling operation via a convolutional neural network, the set of processing resources including circuitry configured to receive input data for supersampling processing, the input data including data sampled according to a jitter pattern that varies locations for data samples; apply an image filter to the received input data, wherein the image filter includes weighting for pixels that is based at least in part on the jitter pattern; process the input data to generate upsampled data; and apply supersampling processing to the upsampled data.
1 . An apparatus comprising:
graphics processing circuitry coupled to a memory, the graphics processing circuitry having processing resources configured to perform a supersampling operation via a convolutional neural network, wherein the graphics processing circuitry is configured to:
receive input data for supersampling processing, the input data including data sampled according to a jitter pattern that varies locations for data samples;
apply an image filter to the received input data, wherein the image filter includes weighting for pixels that is based at least in part on the jitter pattern;
process the input data to generate upsampled data; and
apply, via the convolutional neural network, supersampling processing to the upsampled data.
2 . The apparatus of claim 1 , wherein the circuitry is further configured to receive the input data at a first resolution and to upsample the received input data to a second resolution, the second resolution being higher than the first resolution.
3 . The apparatus of claim 1 , wherein the image filter comprises a filter generated utilizing an approximation of filtering required for image processing, wherein the processing circuitry further comprises texture sampler hardware, wherein the image filter includes application of the texture sampler hardware.
4 . The graphics processor of claim 1 , wherein application of the input filtering occurs in upsampling of the input data, wherein the weighting for a pixel is further based on sampling of neighboring pixels.
5 . The graphics processor of claim 1 , wherein the image filter is a Gaussian filter.
6 . A method comprising:
receiving, by graphics processing circuitry of a computing device, input data for supersampling processing, the input data including data sampled according to a jitter pattern that varies locations for data samples;
applying an image filter to the received input data, wherein the image filter includes weighting for pixels that is based at least in part on the jitter pattern;
processing the input data to generate upsampled data; and
applying, via the convolutional neural network, supersampling processing to the upsampled data.
7 . The method of claim 6 , further comprising:
receiving the input data includes receiving the input data at a first resolution; and
generating the upsampled data includes generating the upsampled data at a second resolution, the second resolution being greater than the first resolution.
8 . The method of claim 6 , wherein the image filter comprises a filter generated utilizing an approximation of filtering required for image processing, wherein the image filter includes application of texture sampler hardware of the graphics processor.
9 . The method of claim 6 , wherein applying the input filtering comprises applying the filtering in upsampling of the input data, wherein the weighting for a pixel is further based on sampling of neighboring pixels.
10 . The method of claim 6 , wherein the image filter is a Gaussian filter.
11 . At least one computer-readable medium having stored thereon instructions which, when executed, cause a computing device to perform operations comprising:
receiving, by graphics processing circuitry of the computing device, input data for supersampling processing, the input data including data sampled according to a jitter pattern that varies locations for data samples;
applying an image filter to the received input data, wherein the image filter includes weighting for pixels that is based at least in part on the jitter pattern;
processing the input data to generate upsampled data; and
applying, via the convolutional neural network, supersampling processing to the upsampled data.
12 . The non-transitory computer-readable medium of claim 11 , wherein the operations further comprise:
receiving the input data includes receiving the input data at a first resolution; and
generating the upsampled data includes generating the upsampled data at a second resolution, the second resolution being greater than the first resolution.
13 . The non-transitory computer-readable medium of claim 11 , wherein the image filter is a filter generated utilizing an approximation of filtering required for image processing, wherein the image filter includes application of texture sampler hardware of the graphics processor.
14 . The non-transitory computer-readable medium of claim 11 , wherein applying the input filtering including applying the filtering in upsampling of the input data.
15 . The non-transitory computer-readable medium of claim 11 , wherein the weighting for a pixel is further based on sampling of neighboring pixels wherein the image filter is a Gaussian filter.