#1038 Compile case ranges as ranges instead of one case per value

[Matt Mower]

Compile time for a switch with a case range rises steeply with the number of values in the range. The attached main.c has one range of N values (default 1024) and builds with both GCC and SDCC:

gcc -c -DN=1024 main.c
sdcc -mmcs51 --std-c2y -DN=1024 -c main.c

Compile time in seconds:

N	gcc 14.2.0	SDCC 4.6.0 mcs51	SDCC r17016 mcs51
256	0.01	< 0.1	< 0.1
512	0.01	2.3	2.3
1024	0.01	16.0	16.2
2048	0.01	> 60 (stopped)	> 60 (stopped)

With r17016 at N=1024: z80 28.7 s, stm8 0.4 s.

Context: QMK keyboard firmware uses ranges such as case QK_MODS ... QK_MODS_MAX: (7936 values). With SDCC, these files do not finish compiling.

SDCC version (sdcc -v):

SDCC : mcs51/z80/z180/r2k/r2ka/r3ka/r4k/r5k/r6k/sm83/tlcs90/ez80/z80n/r800/ds390/pic16/pic14/TININative/ds400/hc08/s08/stm8/pdk13/pdk14/pdk15/mos6502/mos65c02/huc6280/f8/f8l TD- 4.6.4 #17016 (Linux)
published under GNU General Public License (GPL)

[Philipp Klaus Krause]

So for some targets (like stm8), SDCC uses a jump table, for others it uses an if-else chain. Neither is optimal here, but let's ignore that for now. A jump table would already be an improvement for z80.

• The if-else chain then takes more time in (G)CSE. (G)CSE being slow for large control-flow graphs is a known issue.
• On the other hand, a jump table should be much better than an if-else chain here, so I wonder what goes wrong in the heuristic making the jump table vs. if/else chain choice.