Modern SAR sensors exhibit data rates of the order of hundreds Mbits/s; as a consequence, on board data compression is mandatory. The compression algorithms must be characterized by high computational efficiency and satisfactory compression ratios; on the other hand, losses must be kept within limits that take into account the interpretability of the decompressed data. The compressor presented in this paper is flexible, and allows for compression on 2, 3 or 4 bit/sample, useful respectively for topographic monitoring and interferometric applications. The implementation on a rad tolerant DSP permits to achieve high efficiency levels exploiting parallelization at the instrumentation and possibly board level. Classical SAR raw data compression is based on BAQ (Block Adaptive Quantizer) that yields a fixed compression ratio, or its flexible version FBAQ. This latter is a non-uniform quantizer, in which compression is performed comparing the input samples with proper thresholds. The achievable throughput is limited by the need for multiple comparisons for each data sample. For this reason, a variant of FBAQ is presented, based on the key idea of transforming the input samples so that they exhibit uniform distribution; this feature makes possible to perform the compression via simple truncation. The algorithm has been implemented and optimized from the point of view of the achievable throughput. With respect to FBAQ, it exhibits larger memory requirements (16384 extra words of 32-bit), but significantly improved processing speed without appreciable performance degradation in terms of SQNR and phase error. For example, in the case of 4 bit/sample, a throughput 40% larger than FBAQ can be achieved. The algorithm, implemented on high-frequency, radiation tolerant DSPs, will be able to match the requirements of modern SAR for most practical applications.