Entry 5

The port begins: rugrats on retro_rt shows the legal screen, pixel for pixel

2026-10-1412mbuildvideoformattestdesign

rugrats is the port's program: retro_rt's loop at the NTSC rate, the player's disc read through the game's own file table, and the boot as far as it is ported. It shows the legal screen for 250 frames, as show_still (0x80030010) holds it, and then stops where the port ends: the opening cinematic. Nothing from the PlayStation runs in it.

The still is decoded natively. rugrats_data::bs turns LEGAL512.BS into the run-length stream libpress hands the MDEC, and the test legal512_decodes_to_what_libpress_hands_the_mdec holds it to the stream dumped from the original running in the oracle (RUGRATS_MDEC_DUMP): all 42816 halfwords equal. The first attempt ran out of bits at bit 263765 of 263776; libpress leaves a tail shorter than a block undecoded and pads with 0xfe00, and so does the port now. The MDEC's quant and scale tables are not written into the source: libpress keeps them in the executable as ready commands (0x40000001 at 0x800c1f20, 0x60000000 at 0x800c1fa4), and GameDisc::open reads them from the player's SLUS_006.50, refusing an executable whose command words differ.

the_legal_screen_matches_the_original runs the original in the oracle to show_still's hold loop (VSync returning to 0x80030120), waits 30 frames for libgpu's queue to draw, and compares the displayed 512 x 240 with the port's picture: 0 of 122880 pixels differ. That fixes the draw's one subtlety, the quads' colour 0x7f7f7f (0x800c3f9c), which scales each 5-bit channel by 127/128.

Two oracle faults the test found

The test first met a screen of textures, because CD initialisation took 220 budget frames: libcd's CD_cw waits for an acknowledge, and the drive model raised INT3 with its interrupt-enable register at 0. The BIOS leaves it at 0x1f for the game; the model now starts there, and raises IRQ 2 on the rising edge of flag-and-enable, as the line behaves, so a response flagged before interrupts are enabled is not lost. Initialisation now takes about 140000 steps, and that also answers 4's time-out on the first CdlNop. Then the first hold frame was still black: the original's draw is queued in libgpu and lands a frame or more later.

Not an HLE

The oracle exists to check ports like this one; the program is the port. rugrats is a library (disc, still) and a thin program, so tests can call the ported code directly.