Code-mapping considerations

There are several challenges associated with mapping code designed for one instruction set and compiled for another instruction set. These including performance, unmappable constructs, limited size of LS, and equivalent precision, as described in this section.

• Code-mapping performance considerations
  Simple remapping of low-level intrinsics can result in less-than-optimal performance, depending upon the intrinsics used.
• Unmappable constructs considerations
  Differences in the processing of intrinsics make simple translation of certain intrinsics unmappable.
• Limited size of LS considerations
  Vector/SIMD Multimedia Extension programs mapped to SPU programs might not fit within the LS of the SPE, either because the program is initially too big or because mapping expands the code.
• Equivalent precision considerations
  The SPU instruction set does not fully implement the IEEE 754 single-precision floating-point standard (default rounding mode is round to zero , denormals are treated as zero, and there are no infinities or NaNs).