Add stdcbench results at https://github.com/z88dk/z88dk/wiki/Benchmarks?

[spth]

At https://github.com/z88dk/z88dk/wiki/Benchmarks, a few benchmarks are used to compare compilers.

Due to the shortcomings of Dhrystone, Whetstone, Coremark and the unsuitability of major benchmarks, such as SPEC for small systems, I wrote stdcbench:

https://dl.acm.org/doi/abs/10.1145/3207719.3207726
https://sourceforge.net/projects/stdcbench/

stdcbench is one of the benchmarks used to track code size and speed over SDCC revisions:

https://sourceforge.net/p/sdcc/code/HEAD/tree/trunk/sdcc-extra/historygraphs/stdcbench-stm8-size.svg
https://sourceforge.net/p/sdcc/code/HEAD/tree/trunk/sdcc-extra/historygraphs/stdcbench-stm8-score.svg

How about adding stdcbench results at https://github.com/z88dk/z88dk/wiki/Benchmarks?

[suborb]

I've started to take a look, it's showing up a few edge cases that I need to handle better in sccz80.

So far the conclusion is (to no great surprise), that loops in sccz80 aren't that great.

[suborb]

Because I'm running under ticks the number produced is a measure of fast (well slow!) my laptop is, so I've made a minor tweak to just run through the benchmarks once - I've only got c90base to work with zsdcc (4.0.3 #11877) at present unfortunately. Sadly this means we can't easily compare vanilla sdcc and z88dk:

zcc +test -compiler=sccz80 -O2 -m -> 173_691_034 ticks, 9417 bytes
zcc +test -compiler=sccz80 -O2 -m --opt-code-speed=all -> 145_485_000 ticks, 9599 bytes
zcc +test -compiler=sccz80 -O2 -m --opt-code-speed=all -lfastmath -> 122_275_012 ticks, 11877 bytes
zcc +test -compiler=sccz80 -O2 -m -clib=8080 -> 168_723_490 ticks, 9612 bytes
zcc +test -compiler=sccz80 -O2 -m -clib=rabbit -> 56_163_266 ticks, 8632 bytes
zcc +test -compiler=sccz80 -O2 -m -clib=gbz80 -> 170_860_224 ticks, 9461 bytes

zcc +test -compiler=sdcc -O2 --max-allocs-per-node100000 --opt-code-speed -m -> 94_211_835 ticks, 8749 bytes
zcc +test -compiler=sdcc -O2 --max-allocs-per-node100000 --opt-code-speed -m -lfastmath -> 86_213_355 ticks, 11041 bytes
zcc +test -compiler=sdcc -O2 --max-allocs-per-node100000 --opt-code-speed -m -clib=rabbit -> 49_282_502 ticks, 8320 bytes

There's obviously something wrong with the first sccz80 test - I'm not sure how the 8080 has managed to outpace it.

Given the lack of a context-aware optimiser (or indeed any code analysis) in sccz80 and the difference in compilation time (on my machine sdcc is taking 47 seconds, sccz80 takes just 3 seconds) it's not as bad as it could have been.

Any long tests will be interesting: the sccz80 generated code for longs is much more compact - see https://github.com/z88dk/z88dk/wiki/WritingOptimalCode#sample-results

EDIT: it's the 16 bit multiplication routine causing problems on the default z80 options: 8080 routine here: https://github.com/z88dk/z88dk/blob/master/libsrc/_DEVELOPMENT/l/sccz80/crt0_8080/l_mult.asm, z80 routine here: https://github.com/z88dk/z88dk/blob/master/libsrc/_DEVELOPMENT/math/integer/small/l_small_divu_16_16x16.asm)

[feilipu]

@suborb Did you publish the modifications required to get STDC Bench to compile with z80? I had a fleeting attempt, but didn't get far.

It would be nice to add at least some compiler runs for this benchmark.

Initially raised in #111