Hi, Thanks Tristan, GHDL no longer chokes on my dirty design. That rocks!
Simulation speed: 1740 cycles/s with gcc backend, default settings 2450 cycles/s vith -O2 (+35s analysis/elabo time, -91s simu time) 1560 cycles/s with llvm backend, default settings 1560 cycles/s with -O2 (no change??? including analysis/elabo time) 550 cycles/s with mcode backend (x86-64) And xSim from Vivado 2015.3 simulates at 400 cycles/s, that's again a large win for GHDL. Until next time, for even more creative, larger and dirtier muxes... xD But just to detach ourselves from this particular mux case, I think the increasing use of HLS techniques will abundantly create code of inhuman "style" and unprecedented per-statement size. Unfortunately, even if particular cases can/could/should be addressed with some code generator tricks, in the general case it's extremely difficult to guarantee the generated code respects the simulators' right to breathe... (at least for non-commercial HLS tools xD) And to make matters worse, it's precisely for the ugliest pieces of generated code that validation by simulation is most important, because it reflects the HLS tool's ability to correctly manipulate, transform and merge deep and large operations. Have a good day, Adrien _______________________________________________ Ghdl-discuss mailing list Ghdl-discuss@gna.org https://mail.gna.org/listinfo/ghdl-discuss
