Qualcomm Incorporated

Jian Liang

Tao Wang

Andrew Evan Gruber

Chun Yu

Kalyan Kumar Bhiravabhatla

Minjie Huang

Chunlin Wang

Xuefeng Tang

Yun Du

Guofang Jiao

Christopher E Koob

Alexei Vladimirovich Bourd

Jian Shen

Gang Zhong

Jonnala Gadda Nagendra Kumar

Srihari Babu Alla

Yang Yu

Eric Demers

Zilin Ying

Nigel Terence Poole

Vishwanath Shashikant Nikam

Shambhoo Khandelwal

Yang Xia

Yibin Zhang

Fei Xu

Nagarjuna Duvvuru

William L Licea-Kane

Fredrick Alan Hickman

Christopher Edward Koob

Heng Qi

Jiexia Yu

Rahul Sunil Kukreja

The present disclosure relates to methods and apparatus for graphics processing, e.g., a GPU. The apparatus may receive an image including a plurality of pixels associated with one or more workgroups and one or more pixel tiles, each of the workgroups and the pixel tiles including one or more pixels of the plurality of pixels. The apparatus may determine whether the one or more workgroups are misaligned with the one or more pixel tiles. The apparatus may determine a conversion order of the one or more workgroups when the one or more workgroups are misaligned with the one or more pixel tiles, the conversion order corresponding to a common multiple of one of the one or more workgroups and one of the one or more pixel tiles. The apparatus may convert each of the one or more workgroups based on the conversion order of the one or more workgroups.

Rasterization of compute workloads

This disclosure provides systems, devices, apparatus, and methods, including computer programs encoded on storage media, for improving visibility generation in tile-based GPU architectures. A graphics processor may perform a first binning pass associated with visibility information for each of a plurality of primitives in at least one frame. The visibility information for each of the plurality of primitives may correspond to a visible indication or an invisible indication. The graphics processor may update a depth buffer based on the visibility information for all of the plurality of primitives in the at least one frame. The graphics processor may perform a second binning pass for each of the visible set of primitives based on the updated depth buffer. The graphics processor may store at least one of the updated visibility information or updated position data for all primitives in the visible set of primitives from the second binning pass.

Visibility generation improvements in tile based GPU architectures

Methods, systems, and devices for graphics processing are described. A device may receive an image including a set of pixels. The device may render a first subset of pixels in each bin of a set of bins during a first rendering pass, and defer rendering a second subset of pixels and a third subset of pixels in each bin of the set of bins during the first rendering pass. The second subset of pixels may include edge pixels and the third subset of pixels may be between the first subset of pixels and the second subset of pixels. The device may render the second subset of pixels and the third subset of pixels in each bin of the set of bins during a second rendering pass based on rendering the first subset of pixels. The device may then output the image based on the first and second rendering pass.

Bin filtering

The present disclosure relates to methods and apparatus of operation of a processing unit. The apparatus can update a first context register of one or more context registers based on a first programming state. In some aspects, the one or more context registers can be associated with at least one processing unit cluster in a graphics processing pipeline of the processing unit. The apparatus can execute a first draw call function corresponding to the first programming state. The apparatus can determine whether at least one additional first draw call function corresponds to the first programming state. In some aspects, the at least one additional first draw call function can follow the first draw call function in the graphics processing pipeline. Also, the apparatus can execute the at least one additional first draw call function when the at least one additional first draw call function corresponds to the first programming state.

Methods and apparatus for GPU context register management

This disclosure describes an adaptive memory address scanning technique that defines an address scanning pattern, to be used for a particular surface, based on one or more properties of the surface. In addition, a number, shape, and arrangement of sub-primitives of a surface to process in parallel may be determined. In one example of the disclosure, a memory accessing method for graphics processing comprises, determining, by a graphics processing unit (GPU), properties of a surface, determining, by the GPU, a memory address scanning technique based on the determined properties of the surface, and performing, by the GPU, at least one of a read or a write of data associated with the surface in a memory based on the determined memory address scanning technique.

Adaptive memory address scanning based on surface format for graphics processing

A graphics processing unit (GPU) may include a triangle setup engine (TSE) configured to determine coordinates of a triangle, rotate coordinates of the triangle based on an angle. To rotate the coordinates, the TSE generates coordinates of the triangle in a rotated domain, and determines coordinates of a bounding box in the rotated domain based on the coordinates of the triangle in the rotated domain. The TSE determines a first plurality of parallel scanlines in the rotated domain, and a second plurality of parallel scanlines in the rotated domain. The first and second pluralities of scanlines are perpendicular. The TSE determines whether the bounding box coordinates are located within two adjacent scanlines. If the bounding box coordinates are located within the two adjacent scanlines, the TSE removes the triangle from the scene.

Zero pixel culling for graphics processing

A graphics processing unit (GPU) may determine a workload of a fragment shader program that executes on the GPU. The GPU may compare the workload of the fragment shader program to a threshold. In response to determining that the workload of the fragment shader program is lower than a specified threshold, the fragment shader program may process one or more fragments without the GPU performing early depth testing of the one or more fragments before the processing by the fragment shader program. The GPU may perform, after processing by the fragment shader program, late depth testing of the one or more fragments to result in one or more non-occluded fragments. The GPU may write pixel values for the one or more non-occluded fragments into a frame buffer.

Dynamically switching between late depth testing and conservative depth testing

A cache memory system includes a cache memory including a plurality of cache memory lines and a dirty buffer including a plurality of dirty masks. A cache controller is configured to allocate one of the dirty masks to each of the cache memory lines when a write to the respective cache memory line is not a full write to that cache memory line. Each of the dirty masks indicates dirty states of data units in one of the cache memory lines. The cache controller may include a dirty buffer index which stores an identification (ID) information that associates the dirty masks with the cache memory lines to which the dirty masks are allocated. A cache line may include a fully dirty flag indicating when each byte in that cache line is dirty, so that a dirty mask does not need to be allocated for that cache line.

Cache memory system and method using dynamically allocated dirty mask space

In general, this disclosure describes techniques for increasing the throughput of multi-bank cache memory systems accessible by multiple clients. Requests for data from a client may be stored in a pending buffer associated with the client for a first cache memory bank. For each of the requests for data, a determination may be made as to if the request is able to be fulfilled by a cache memory within the first cache memory bank regardless of a status of requests by the client for data at a second cache memory bank. Data requested from the cache memory by the client may be stored in a read data buffer associated with the client according to an order of receipt of the requests for data in the pending buffer.

Multi-bank cache memory

A method of transferring data between two caches comprises sending a first message from a first processor to a second processor indicating that data is available for transfer from a first cache associated with the first processor, requesting, from the second processor, a data transfer of the data from the first cache to a second cache associated with the second processor, transferring the data from the first cache to the second cache in response to the request, and sending a second message from the second processor to the first processor indicating that the data transfer is complete.

Cache data migration in a multicore processing system

Disclosed herein is a supplemental cache for use with a graphics processing unit. The supplemental cache can be used to supplement a vertex cache used with a graphics processing unit. The supplemental cache stores vertex values generated in assembling primitives from vertices provided to the graphics processing unit as part of an image geometry. Generated vertex values associated with a vertex determined to be shared by two or more primitives can be retrieved from the supplemental cache, so as to reduce the need to perform duplicative operations to generate vertex values for shared vertices.

Supplemental cache in a graphics processing unit, and apparatus and method thereof

A system and method is disclosed and includes an execution unit that can be used to count the leading zeros in a data word. During operation, the execution unit can receive a data word that has a width of 2 to the Nth power. Further, the execution unit can sign extend the data word to a temporary data word that has a width of 2 to the Mth power, wherein M is greater than N. The temporary data word can be input to a counter that has a width of 2 to the Mth power and the counter can count the leading zeros within the temporary data word to get a result.

Growing community of inventors

San Diego, CA, United States of America

Jian Liang