Abstract: The techniques disclosed herein enable systems to perform repeatable and iterative load testing and performance benchmarking for artificial intelligence models deployed in a cloud computing environment. This is achieved by utilizing load profiles and representative workloads generated based on the load profiles to evaluate an artificial intelligence model under various workload contexts. The representative workload is then executed by the artificial intelligence model utilizing available computing infrastructure. Performance metrics are extracted from the execution and analyzed to provide insight into various performance dynamics such as the relationship between latency and data throughput. In addition, load profiles and input datasets are dynamically adjusted to evaluate different scenarios and use cases enabling the system to automatically test the artificial intelligence model across diverse applications.

Abstract: Processing per-use requests for user data is disclosed, including: receiving a use-specific request to read a set of user data; determining whether to grant the use-specific request based at least in part on an access policy associated with an accessor definition associated with the use-specific request; and in response to a determination to grant the use-specific request, transforming the set of user data based at least in part on the accessor definition.

Abstract: Token generation and management are disclosed, including: generating a token corresponding to a set of user data based at least in part on a token generation policy; storing a mapping between the token and the set of user data; and determining whether to grant a token resolution request associated with the token based at least in part on a token access policy associated with the token and a context parameter associated with the token resolution request.

Abstract: Processing per-use requests for user data is disclosed, including: receiving a use-specific request to read a set of user data; determining whether to grant the use-specific request based at least in part on an access policy associated with an accessor definition associated with the use-specific request; and in response to a determination to grant the use-specific request, transforming the set of user data based at least in part on the accessor definition.

Abstract: In a state management system of an effects system implemented in a Graphics Processing Unit (GPU), techniques and technologies are provided for setting a value to particular variables at application run-time without validating the variables. For example, a compiled effects file comprising a number of variables can be loaded at application load time, and a generic, variable interface pointer for a particular variable of the effects file can be retrieved. A specialized variable interface pointer can then be generated which is associated with the particular variable by specifying a desired type of access that will be performed on the particular variable. At application run-time, the specialized variable interface can be used to set a value to each of the particular variables without validating the particular variables at application run-time.

Abstract: Techniques and technologies are provided for binding resources to particular slots associated with shaders in a graphics pipeline. Resource dependencies between resources being utilized by respective shaders can be determined, and, based on these resource dependencies, common resource/slot associations can be computed. Respective common resource/slot associations identify a particular one of the resources to be associated with a particular one of the slots.

Abstract: Techniques and technologies are provided for binding resources to particular slots associated with shaders in a graphics pipeline. Resource dependencies between resources being utilized by each shader can be determined, and, based on these resource dependencies, common resource/slot associations can be computed. Each common resource/slot association identifies a particular one of the resources to be associated with a particular one of the slots.

Abstract: Techniques and technologies are provided for binding resources to particular slots associated with shaders in a graphics pipeline. Resource dependencies between resources being utilized by respective shaders can be determined, and, based on these resource dependencies, common resource/slot associations can be computed. Respective common resource/slot associations identify a particular one of the resources to be associated with a particular one of the slots.

Abstract: Techniques are provided for efficiently arranging shader constants into constant buffers to more efficiently set shader constants to a GPU. Techniques are provided for tracking runtime usage patterns of shader constants, and then generating usage pattern data regarding usage patterns of shader constants. This usage pattern data can then be fed into an optimization algorithm to efficiently lay out the data for maximal performance given certain heuristics. Factors considered include, for example, frequency of client update of the values, usage by certain shaders, and size and number of the resulting buffers.

Abstract: Techniques and technologies are provided for binding resources to particular slots associated with shaders in a graphics pipeline. Resource dependencies between resources being utilized by each shader can be determined, and, based on these resource dependencies, common resource/slot associations can be computed. Each common resource/slot association identifies a particular one of the resources to be associated with a particular one of the slots.

Abstract: In a state management system of an effects system implemented in a Graphics Processing Unit (GPU), techniques and technologies are provided for setting a value to particular variables at application run-time without validating the variables. For example, a compiled effects file comprising a number of variables can be loaded at application load time, and a generic, variable interface pointer for a particular variable of the effects file can be retrieved. A specialized variable interface pointer can then be generated which is associated with the particular variable by specifying a desired type of access that will be performed on the particular variable. At application run-time, the specialized variable interface can be used to set a value to each of the particular variables without validating the particular variables at application run-time.

Abstract: Techniques are provided for efficiently arranging shader constants into constant buffers to more efficiently set shader constants to a GPU. Techniques are provided for tracking runtime usage patterns of shader constants, and then generating usage pattern data regarding usage patterns of shader constants. This usage pattern data can then be fed into an optimization algorithm to efficiently lay out the data for maximal performance given certain heuristics. Factors considered include, for example, frequency of client update of the values, usage by certain shaders, and size and number of the resulting buffers.