[jonareyess]

Adds the Mac profile for Trails in the Sky 1st Chapter (01002C9022770000, update 1.0.7) on current main. Uses main's pipeline-selected integer outputs and image-view fitting, with the remaining fixes for invocation-private compute scratch, reflected compute group sizes (runtime and offline archive), and GLSLC compute-code offsets.

Fixes excessive GPU allocation from unbounded buffer lookups. Indirect indexed draws created overlapping 80–100 MB views over a 1,507 MB pool, reaching 4,243 MB of shared buffers and GPU out-of-memory errors on M1. Unbounded lookups now share one whole-pool view; short cached views cannot satisfy them. Known-size accesses retain chunking, and existing views remain valid for encoded GPU addresses.

Validation:

• All 6,602 Mac shaders build, and the corrected app packages successfully.
• Pool tests: 171 checks pass, including a GPU access 96 MB beyond an unbounded address. The pre-fix renderer fails 20 checks.
• The full graphics target list, targeted tool/docs/chain tests, and Python/C lint pass.
• Trails' shader/interface regression against b06f4e0e preserves 6,602/6,602 shaders with no new failures. Two expected MSL changes come from private-scratch lowering.
• A 70-second scripted save load reaches frame 3,900 without GPU-memory errors; loaded-scene Metal allocation stays at 2,685 MB. A subsequent owner walking/combat session lasts about seven minutes, exits normally without GPU-memory errors, and stabilizes near 3,153 MB.
• Visual validation on M1: the owner confirms that menu eyes and bestiary lighting are correct with this integration.
• M1 performance observation: the owner reports substantially fewer severe frame-time drops with DM_RT_FRAME=0. The log confirms the switch and a normal exit without GPU errors. These were manually repeated sessions of different durations; this is a diagnostic result, not an established scheduling fix. No default or profile setting is changed.

Longer-session stability, renewed reflection checks, and the cause of the scheduling-related performance difference remain open. iOS is untested. The game note and shared references record the causes and supporting evidence.

[ricardosousa191]

Tested this PR on my M3 air and the game is working as intended and haven't encountered any type of graphical bugs in my 1 hour of gameplay.

Screenshot:

[seedds]

A note for this PR: main now handles integer colour outputs per pipeline, which overlaps with this PR's "mixed integer/float fragment alternatives selected from the actual attachments".

Trails in the Sky 2nd Chapter (same engine, now on main in 6cb67b8) hit the same red line: a G-buffer with a float target at colour 0 and integer targets at colours 1 to 3, where the integer entry point chosen from the shader's code left out an output that goes to an integer target. Every rule that told integer outputs apart from the shader's code just moved the red line to the next shader.

c398ff3 (chain/dmsh_entry) replaces those rules. main0_u now returns each colour either as a float or as the same bits as an unsigned integer, chosen by a function constant (DMSH_UINT_FC + n, 208 to 215). The host specializes it with the pipeline's own integer-target mask (pso_uint_mask, uint_fn, cached per shader and mask). The alpha-test (main0_u_a) and logic-op (main0_u_l) entry points follow the same switch. The fixed mixed entry point main0_m is gone.

Reference: Integer outputs told by the pipeline's targets.

Checked on the 2nd Chapter (no [RED] in a fresh-save route), Luigi's Mansion 3 (every G-buffer target integer) and Kirby Star Allies (float and integer targets mixed). The logic-op replay (941 checks) and the alpha-test replay (788 checks) pass.

If this PR is rebased onto main, its integer-output alternatives can most likely be dropped in favour of main's. The menu eyes and bestiary lighting would then be worth rechecking on the 1st Chapter. I haven't run the 1st Chapter on main myself.

[jonareyess]

Updated on current main, keeping its pipeline-selected integer outputs and removing our older alternatives. I rechecked the menu eyes and bestiary lighting on M1, and both render correctly. The PR now contains two focused commits, with validation details in the description.