Facile Conversion and Optimization of Structured Illumination Image Reconstruction Code into the GPU Environment

Figure 10

Byte differences due to the precision issues of floating-point numbers to represent real numbers. (a) Analysis of an image produced by the original code. (b) Single-precision approach image. (c) Analysis of an image produced by the improved and GPU-enhanced code. The red-colored numbers in (b) and (c) show the byte differences from the original output (baseline) image, which occurred every 4 bytes. We propose that this is acceptable based on the original vanilla code that used the single-precision and double-precision intermixed.

(a) Binary snippet of the output image generated by vanilla code

(b) Binary snippet of the output image generated by a code—a single-precision approach applied

(c) Binary snippet of the output image generated by a code—all approaches including GPU applied