2.4. DSP Builder for Intel® FPGAs
Unless stated otherwise, the following DSP Builder for Intel® FPGAs issues apply only to the Intel® Quartus® Prime Pro Edition software.
- Single-precision floating-point adders in a DSP Builder design can now use a mixture of hard and soft implementations.
- VFFT_btb and VFFT_Light_btb blocks have been added. These support back-to-back operation of the variable-size FFT, so there is no need to flush the FFT pipeline between different-size FFT iterations.
- A defect in the VFFT_cp_btb and VFFT_Light_cp_btb blocks has been corrected. When the 18.1 versions of these blocks were used with multiple subchannels, the output start-of-packet, size and (for VFFT_cp_btb) end-of-packet signals were not correctly synchronized with the output valid signal.
- The optimization of floating-point scalar products is now slightly less aggressive. With default settings, DSP Builder no longer attempts to infer a chain-in input for scalar products. As a result of this change, complex scalar products now consume slightly more DSP blocks but require fewer logic resources for latency-balancing.
- Fixed initialization of SharedMems of type single precision float.
- Added support for loop blocks to the Advanced C-model feature.