# SDK scratch: the 1987 TMS34010 DOS toolchain

This directory is a scratch area for the investigation in
`reference/tms34010-sdk.md`.  The SDK under `tools/tms34010-sdk/` is only read;
everything here is extracted work product.

## Round trip

```sh
./roundtrip.sh 0x1000 32768 3F208000
```

`OFFSET SIZE BASE_BIT`: disassemble that ROM region with
`tools/smart34010/dis34 -o`, convert it to GSPA syntax with `dis2as34.py`,
assemble/link/convert with the 1987 DOS tools under DOSBox, interleave the
high/low Intel-hex files, and diff against the ROM.  It prints `BYTE-EXACT`
and the md5 of the reassembled region on success.

The tools are extracted on first use from:

```
tools/tms34010-sdk/1987-05 TMS34010 SDK/
  1987-11-06 TMS34010 Simulator-C Tools-ASM Tools.img
```

into `work/` (A2 = assembler/linker, C2 = C compiler/runtime).  The ROM is
never written.

## Benchmark

```sh
python3 -u bench2.py    # DOSBox vs dosemu, 3 runs, two workloads
```

`bench2.py` uses `dosbox-bench.conf` (dynamic core, `cycles=max`) and a private
dosemu `HOME` in `dosemu-home/`.  The dosemu batch must end in `exitemu`;
`-E` alone leaves DOS running.  Results are in the report.

## Notes

* `dis2as34.py` needs the `-o` opcode comments from `dis34`; they are what
  disambiguate `MOVI` short/long, long relative `JR` vs absolute `JA`, and the
  B-file `SP`.
* `gsprom -i` writes two files (high and low byte); interleave low then high.
  Its record addresses are 16-bit word indices, not bit addresses.
* DOSBox 8.3 names are uppercased; check for `RT.OBJ`, not `rt.obj`.  dosemu's
  `lredir` keeps lowercase names.