Patent №
US 8,558,842
Granted
2013-10-15
Filed 2010
Owner
NVIDIA CORPORATION
Lab
AI components
1
hardware
Assignment
Recorded
Dataset
AIPD
2023_r1 edition
Application
12723266
One embodiment of the present invention sets forth a technique for detecting duplicate vertex indices in parallel and batching indices defining multiple primitives for parallel primitive processing. A lookback cache breaks the dependent loop for the miss processing. Because each index is compared to all previous indices (duplicate or not), each index is not dependent on whether the previous indices have hit or missed. This allows the comparison operation that detects the duplicate vertex indices to be fully pipelined. The duplicate vertex indices are removed to reduce the number of indices that define the primitives in the batch. Multiple, independent rasterizer units operate concurrently on the different batches of graphics primitives to render multiple primitives per system clock.
AI classification
Ownership
NVIDIA CORPORATION
assignment · 240740747
Assignors
JOHNSON, PHILIP B., BOWMAN, JAMES C.
On an employer assignment, the assignors are typically the inventors.