rocFFT
6.1.2
Real data
Supported array type combinations
Setting strides
Examples
rocFFT
Design Documents
View page source
Design Documents
ΒΆ
Contents:
Code Generator Design Document for rocFFT
Copyright and Disclaimer
Proposal
Rationale
Required kernels (scope)
Tiling
Strides and batches
Large twiddle tables
Launching
Implementation
Stockham tiling implementation
Large twiddle tables
Runtime Compilation Design Document for rocFFT
Copyright and Disclaimer
Summary
Problem
Solution
Implementation
Embedding and running the generator
Compilation
Caching kernels at runtime
Interaction with callbacks
Parallel compilation
Code organization
Future work
Moving away from chosen problem sizes
Buffer assignment design document for rocFFT
Copyright and Disclaimer
Summary
Observations
Solution
Implementation
A Structure Storing A Try
Exhaustive Search
Decision Function and Output Lengths
Fusions
Padding
Choosing a winner
Future Work
Strides
Enforcing Read-only Input
Flexibility Between Minimizing Memory or Maximizing Fusions
Make C Buffer as Temp2 Buffer
Bluestein Design Document
Copyright and Disclaimer
Summary
Background and Notation
Bluestein Algorithm
Chirp
Convolution
Zero padding
DFT via Bluestein
Implementation
Device Kernel Configuration
Optimizing Bluestein for large length DFTs