This is a pair of benchmarks that implement a very trivial
raytracer in C and Pascal - the code is 99.9% the
same in both, the main difference is that the C code uses arrays
to represent vectors (e.g. float v[3]) whereas the
Pascal code uses a Vec record (e.g. var v: Vec) and
const/var to pass across functions and procedures (which passes
arguments via reference).
I wrote this benchmark mainly because i was interested in doing some retrocoding for old Pentium PCs and wanted to figure out which compiler to choose (most likely VC6 as it is the one that generates the fastest code while creating executables that work on the original Pentium - some old MinGW version might do better, but i do not have any installed). This is the reason for the "pentium", "-5" and "586" references in the table below. But i also decided to try the benchmark in a few other C and Pascal compilers i have installed in my PC.
Here are the results for generating a 640x480 image with 256
rays per pixel (SUBGRID=16) on Windows 10 with an
AMD Ryzen 2400G CPU:
| Language | Compiler | Options | Time | % Time |
| C | Clang 8.0.0 64bit | -Ofast -fno-signed-zeros -fno-trapping-math -funroll-loops -march=native -mtune=native | 2.075 | 100% |
| C | Clang 8.0.0 64bit | -Ofast -fno-signed-zeros -fno-trapping-math -funroll-loops -march=native -mtune=native | 2.354 | 113% |
| C | GCC 8.3.0 64bit | -Ofast -fno-signed-zeros -fno-trapping-math -frename-registers -funroll-loops -march=native -mtune=native -fprofile-use | 2.377 | 114% |
| C | GCC 8.3.0 64bit | -Ofast -fno-signed-zeros -fno-trapping-math -frename-registers -funroll-loops -march=native -mtune=native | 2.721 | 131% |
| C | GCC 7.4.0 32bit | -Ofast -fno-signed-zeros -fno-trapping-math -frename-registers -funroll-loops -march=native -mtune=native | 3.444 | 165% |
| Pascal | Free Pascal 3.0.4 64bit | -CfAVX2 -CpCOREAVX2 -O4 | 3.687 | 177% |
| Pascal | Free Pascal 3.0.4 64bit | -O4 | 3.875 | 186% |
| Pascal | Free Pascal 3.0.4 32bit | -O4 -CfAVX2 -CpCOREAVX2 | 4.828 | 232% |
| C | GCC 7.4.0 32bit | -Ofast -fno-signed-zeros -fno-trapping-math -frename-registers -funroll-loops -march=pentium -mtune=pentium | 5.362 | 258% |
| C | Clang 8.0.0 32bit | -Ofast -fno-signed-zeros -fno-trapping-math -funroll-loops -march=pentium -mtune=pentium | 6.170 | 297% |
| C | Visual C++ 6 | Release build, __fastcall calling convention, Pentium CPU | 6.587 | 317% |
| C | Visual C++ 6 | Release build, __cdecl calling convention, Pentium CPU | 6.678 | 321% |
| C | OpenWatcom 2.0 (nightly) | -5r -otexan -ol+ | 7.672 | 369% |
| C | Digital Mars C++ 8.42n | -mn -o -5 | 9.171 | 441% |
| Pascal | Delphi 2 | Optimizations enabled, all runtime checks disabled | 9.828 | 473% |
| Pascal | Free Pascal 3.0.4 | -O4 | 9.953 | 479% |
| C | Borland C++ 5.2 | -4 -O2 | 10.641 | 512% |
| C | Borland C++ 5.0 | -4 -O2 | 10.688 | 515% |
Since this was meant for retrocoding on Pentium machines running Windows 95, here are the results for the compilers that generate code working there (percentage time adjusted for the fastest executable in this table):
| Language | Compiler | Options | Time | % Time |
| C | Visual C++ 6 | Release build, __fastcall calling convention, Pentium CPU | 6.587 | 100% |
| C | Visual C++ 6 | Release build, __cdecl calling convention, Pentium CPU | 6.678 | 101% |
| C | OpenWatcom 2.0 (nightly) | -5r -otexan -ol+ | 7.672 | 116% |
| C | Digital Mars C++ 8.42n | -mn -o -5 | 9.171 | 139% |
| Pascal | Delphi 2 | Optimizations enabled, all runtime checks disabled | 9.828 | 149% |
| C | Borland C++ 5.2 | -4 -O2 | 10.641 | 161% |
| C | Borland C++ 5.0 | -4 -O2 | 10.688 | 162% |
FWIW i'd go with Visual C++ 6 for this, but that is because i'll be running it on my main PC. If i was going to code on a real (or emulated) retro computer, i'd use Delphi 2 and just write better code :-P as all C compilers (and especially VC6) are too slow for comfortable use.
C source code and Pascal source code (to compile with FPC use -Mdelphi).
There is no license in the code, but if you care consider it under zlib license.