-
Notifications
You must be signed in to change notification settings - Fork 8
[FEATURE] Migrate from OpenCL ecosystem #32
Copy link
Copy link
Open
Labels
dependenciesIssues related to library dependenciesIssues related to library dependenciesenhancementFeature requests and improvements to existing functionalityFeature requests and improvements to existing functionalitygpu-accelerationIssues related to GPU support or accelerationIssues related to GPU support or accelerationhigh-priorityIssues requiring urgent attentionIssues requiring urgent attentionnative-integrationIssues related to C/native code integrationIssues related to C/native code integrationperformanceIssues related to speed, memory usage, or efficiencyIssues related to speed, memory usage, or efficiency
Milestone
Description
Feature Description
OpenCL sucks.
We need a more robust, scalable, portable and sustainable approach to the GPU ecosystem, possibly with native support for different backends.
Implementation Ideas
In the spare time I'm working on a unified bridge between CUDA, Metal and OpenCL, allowing for the same code to execute on different backends, with minor changes to the kernels.
Reactions are currently unavailable
Metadata
Metadata
Labels
dependenciesIssues related to library dependenciesIssues related to library dependenciesenhancementFeature requests and improvements to existing functionalityFeature requests and improvements to existing functionalitygpu-accelerationIssues related to GPU support or accelerationIssues related to GPU support or accelerationhigh-priorityIssues requiring urgent attentionIssues requiring urgent attentionnative-integrationIssues related to C/native code integrationIssues related to C/native code integrationperformanceIssues related to speed, memory usage, or efficiencyIssues related to speed, memory usage, or efficiency
Type
Projects
Status
In Progress