CHANGELOG

# Changelog

This file contains all changes made to the source code for each release.

## 2.6

#### Changed:
- Fix a bug that leads to wrong performance numbers when using MPI

#### Added:
- Support for half and double precision floating point numbers

## 2.5

#### Added:
- Configuration files for 520N, U280 and PAC D5005

#### Changed:
- Use registers between load and store operations from global memory to local memory to overcome destroyed memory bursts on Xilinx FPGAs when unrolling the inner loops.

## 2.4

#### Added:
- MPI support for EP execution

## 2.3

#### Added:
- Base implementation tests are now build and linked with the unit test binary
- Support for custom kernel designs

#### Changed:
- Replaced flag `single-kernel` with `multi-kernel`. Now the single kernel implementation is considered the default. The legacy multi kernel implementation can be executed using the new flag.

## 2.2

#### Added:
- Support for Intel HBM using the USE_HBM flag. This will use the CL_MEM_HETEROGENEOUS_INTELFPGA flag to allocate OpenCL buffers.

## 2.1.5

#### Changed:
- Converted host code to new OO code

## 2.1.4.1

#### Added:
- Support for SVM in host code

## 2.1.4

#### Modified:
- Reduce BRAM usage of kernels by combining two local memory buffers to one. This allows larger sizes of the local memory buffer with the same resurce usage.


## 2.1.3

#### Modified:
- Optimize kernels for Xilinx toolchain
- Remove volatile keyword of kernel parameters to enable bursts on Xilinx boards
- Change buffer placement on memory banks for Intel FPGAs

## 2.1.2

#### Added:
- Optimize build system for Xilinx boards

## 2.1.1

#### Added:
- Add support for kernel replication for Intel devices and host code
- Add single kernel with operation switch

## 2.1

#### Added:
- Add basic support for Xilinx boards

#### Modified:
- Change from CMake 2.8 to CMake 3.1 as build system

## 2.0

#### Added:
- Replace Makefile with CMake as a build system