DS2x86 Progress

Pate has posted more WIP news of his Dos Emulator for the DS

I got my free DSTwo flash cart last Monday, so I have been working on DS2x86 ever since. It took me all of Monday evening to get the DS2x86 framework (tester routine calling the MIPS ASM CPU emulation code, and it returning properly back to the tester routine) to run without crashing, so on Tuesday evening I was able to start working on the actual CPU opcode emulation with MIPS ASM language. I am actually using a strict TDD (Test Driven Development) coding technique when working on DS2x86. With DSx86 I usually coded something, then tested it with some test games, and only if that failed I coded more thorough tests. With DS2x86 I implement the test routines (or improve the old tests I used in DSx86) first, and only then start coding the actual opcode handlers. I do this because the MIPS ASM language is very unfamiliar to me, and also because I am now using the Lazy Flags approach, so I can no longer use the ARM CPU to calculate the correct x86 CPU flags for me.

By Saturday morning I had implemented the first four opcodes 0x00..0x03 (the various ADD opcode versions). Each of these actually have 256 different modrm bytes for all the different memory address modes, and as there are 6 different segment override possibilities, plus the case with no segment override, each of these opcodes actually has 7*256 different cases. My test routine runs each of these cases with random input values, and tests for correct results and correct emulated CPU flags state after the run. So in fact I had over 7000 different cases, including their unit tests, coded and tested by Saturday morning. Pretty good progress I think, considering I was only able to work on it for a few hours every evening.


The images above are screen captures from the default SDK console, using a screen capture code by BassAceGold, another SDK tester (thanks for the code!). The first one is from Saturday morning, with the first four opcodes working and opcode 0x04 stopping with an invalid result (0x59+0x1C=0x1C). The second one is the current (Sunday midday) situation. I have skipped opcode 0x0F, as it contains a lot of different 386 opcodes, and also the 386-versions of the already coded opcodes are mostly missing. The Lazy Flags handling does not yet calculate correct Overflow flag for SUB/SBB opcodes, so that too still needs work. I have the 386-specific FS and GS segment registers already supported, and also the immediate 32-bit versions like ADD EAX,0x1234567 are already coded and tested. But since my first priority is to get Norton Sysinfo running, I can leave supporting most of the actual 386-opcodes for later.

I think the font used by the SDK is not the best possible, especially the small letter g is pretty unclear, but this is certainly good enough for debug printing. The MIPS cpu is set to run at 120MHz by default (I believe that is the lowest speed it can run), which is fine while coding the tester program. There is an API call in the SDK to change the CPU speed, up to 396MHz I believe, so I am thinking of adding a configuration option to DS2x86 where the user can choose the CPU clock speed. Running at higher speeds will probably drain the battery pretty fast, so always running at the highest speed is not a good idea.

Anyways, I am already at opcode 0x27 DAA. This is one of the weird BCD (Binary Coded Decimal) opcodes that use the Adjust Flag (which is not properly supported in DSx86). Now with DS2x86 I can support the Adjust Flag properly, so I can finally code the DAA opcode so that it will always give correct results. Even though this opcode is very rarely used, it will be good to have it working correctly. :-)

http://dsx86.patrickaalto.com/DSblog.html