> I have proted Dhrystone to the FX2LP, and ran it at 48 Mhz. I get the > results below, which translates to 0.0096 DMIPS / Mhz (vs. 0.355 DMIPS / > Mhz for stm8 in current SDCC). So the STM8 is about 37 times faster per > Mhz than the FX2LP. Does that seem about right? > > For the FX2LP I Used --model-large and otherwise default options, while > for the stm8 I used heavy optimization for code speed - AFAIK the > optimization options don't make a big difference for mcs51 yet. > > Does anyone have numbers for other compilers or mcs51 variants? > > Philipp
Hi Philipp, I've never tried to run Dhrystone. But the FX2 is not the fastest mcs51 variant. It is a so-called 4-clocker. The original 8051 used 12 clock cycles for 1 machine cycle. The FX2 uses 4 cycles for 1 machine cycle and is machine cycle compatible. Many modern variants use 1 clock cycle for 1 machine cycle, but they are no longer machine cycle compatible resulting in instructions taking more machine cycles. Further, using the large model makes the code substantially slower. Compare small against large ticks in the regression tests. Maarten ------------------------------------------------------------------------------ _______________________________________________ Sdcc-user mailing list Sdcc-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/sdcc-user
