Fixed-rate compressed floating-point arrays

摘要: Current compression schemes for floating-point data commonly take fixed-precision values and compress them to a variable-length bit stream, complicating memory management random access. We present fixed-rate, near-lossless scheme that maps small blocks of 4(d) in d dimensions fixed, user-specified number bits per block, thereby allowing read write access compressed at block granularity. Our approach is inspired by fixed-rate texture methods widely adopted graphics hardware, but has been tailored the high dynamic range precision demands scientific applications. compressor based on new, lifted, orthogonal transform embedded coding, each per-block stream be truncated any point if desired, thus facilitating rate selection using single scheme. To avoid or decompression upon every access, we employ software write-back cache uncompressed blocks. designed with computational simplicity speed mind allow possibility hardware implementation, uses only fixed-point arithmetic operations value. demonstrate viability benefits lossy several applications, including visualization, quantitative analysis, numerical simulation.