diff --git a/ReleaseNotes.txt b/ReleaseNotes.txt index 7efa23c5..135670ff 100644 --- a/ReleaseNotes.txt +++ b/ReleaseNotes.txt @@ -8,17 +8,18 @@ implementation of discrete Fast Fourier Transforms. It: discrete FFTs * Supports 1D, 2D, and 3D transforms with a batch size that can be greater than 1 +* Supports complex and real transforms * Supports planar (real and complex components in separate arrays) and interleaved (real and complex components as a pair contiguous in memory) formats * Supports dimension lengths that can be any mix of - powers of 2, 3, and 5 + powers of 2, 3, 5 and 7 * Supports single and double precision floating-point formats * Supports in-place or out-of-place transforms -clFFT - Release Notes - version 2.7 +clFFT - Release Notes - version 2.8.0 -------------------------------------- New features of this release: diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt index 36f242aa..d95cc10e 100644 --- a/src/CMakeLists.txt +++ b/src/CMakeLists.txt @@ -40,11 +40,11 @@ if( NOT DEFINED CLFFT_VERSION_MAJOR ) endif( ) if( NOT DEFINED CLFFT_VERSION_MINOR ) - set( CLFFT_VERSION_MINOR 6 ) + set( CLFFT_VERSION_MINOR 8 ) endif( ) if( NOT DEFINED CLFFT_VERSION_PATCH ) - set( CLFFT_VERSION_PATCH 1 ) + set( CLFFT_VERSION_PATCH 0 ) endif( ) set( CLFFT_VERSION "${CLFFT_VERSION_MAJOR}.${CLFFT_VERSION_MINOR}.${CLFFT_VERSION_PATCH}") diff --git a/src/library/mainpage.h b/src/library/mainpage.h index a0dc6b56..23185a9d 100644 --- a/src/library/mainpage.h +++ b/src/library/mainpage.h @@ -127,6 +127,18 @@ explicitly flush the command queues that are passed by reference to it. It pushe command queues and returns the modified queues to the client. The client is free to issue its own blocking logic using OpenCL synchronization mechanisms or push further work onto the queue to continue processing. +@subsection Environment variables +The clFFT library looks for the definition of 2 environment varibles. One is CLFFT_CACHE_PATH. If this +variable is defined, then the library caches OpenCL binaries. This will enable a subsequent application run +of the same type of transforms to avoid going through the expensive compile step. Instead, the stored +binaries are loaded and executed. The CLFFT_CACHE_PATH must point to a folder location where the +library can store binaries. The other environment variable is CLFFT_REQUEST_LIB_NOMEMALLOC. This +variable when defined asks the library to do all computations in-place and avoid allocating extra +device memory whenever possible. This feature is experimental and currently works only for certain types +of transforms, and where the input can be decomposed into square matrices by the library. Currently it +works for 1D complex transforms of size of even powers of 2,3,5 and 7. + + @section clFFTPlans clFFT plans A plan is the collection of (almost) all the parameters needed to specify an FFT computation.