Back to results
US 12,737,326 B2Granted
Compute graph optimization
AssigneeNVIDIA Corporation
InventorsStephen Jones, Steven Arthur Gurfinkel, David Anthony Fontaine, Sally Tessa Stevenson, Parag Shashikant Kulkarni
Life of this patent
Estimating the term…
Abstract
Systems, processors, and techniques to execute graph code. In at least one embodiment, graph code is instantiated and executed using a first set of parameters, and executed a second time, using a second set of parameters, without reinstantiation.
Claims30
A processor, comprising:
- one or more circuits to:
- identify one or more portions of an optimized graph on which to perform one or more re-optimizations, the identification based at least in part on changes to one or more operands to the optimized graph made subsequent to instantiation, and based at least in part on a determination that identified portions of the optimized graph can be re-optimized, using less than a threshold amount of computing capacity, to use the one or more changes to the one or more operands;
- perform the one or more re-optimizations on the one or more identified portions;
- and
- cause the optimized graph, comprising the one or more identified portions to which re-optimizations have been applied, to be performed.
2 more elements
A non-transitory machine-readable medium having stored thereon a set of instructions, which if performed by one or more processors, cause the one or more processors to at least:
- instantiate a graph based, at least in part, on one or more operands to be operated on by the graph;
- identify one or more first portions of an instantiated graph on which to perform one or more optimizations to replace one or more optimizations made to the graph, the identification based at least in part on changes to one or more operands to the instantiated graph made subsequent to instantiation;
- perform the one or more optimizations on the identified one or more first portions;
- identify one or more second portions of the instantiated graph on which one or more optimizations do not need to be performed, based at least in part on the one or more changes to the one or more operands;
- and
- cause the instantiated graph, comprising the one or more identified first portions to which optimizations have been applied and the one or more second portions which do not need optimizations, to be performed.
3 more elements
A system, comprising:
- one or more processors to:
- identify one or more portions of an optimized graph on which to perform one or more re-optimizations, the identification based at least in part on changes to one or more operands to the optimized graph made subsequent to instantiation, and based at least in part on a determination that identified portions of the optimized graph can be re-optimized, using less than a threshold amount of computing capacity, to use the one or more changes to the one or more operands;
- perform the one or more re-optimizations on the one or more identified portions;
- and
- cause the optimized graph, comprising the one or more identified portions to which re-optimizations have been applied, to be performed.
2 more elements
A system, comprising:
- a memory comprising a first buffer and a second buffer;
- and
- at least one processor to:
- identify one or more portions of an optimized graph on which to perform one or more re-optimizations, the identification based at least in part on changes to one or more operands to the optimized graph made subsequent to instantiation, and based at least in part on a determination that identified portions of the optimized graph can be re-optimized, using less than a threshold amount of computing capacity, to use the one or more changes to the one or more operands;
- perform the one or more re-optimizations on the one or more identified portions;
- and
- cause the optimized graph, comprising the one or more identified portions to which re-optimizations have been applied, to be performed.
4 more elements
Dependencies are inferred from the claim text.