[Mips] Permit tail calls on the R3000A (PS2-184) #10

Merged

techwritescode merged 1 commit from r3000a-tail-calls into r3000a-shared-global-base 2026-10-13 19:16:44 +00:00

Conversation 0 Commits 1 Files changed 5 +61 -4

techwritescode commented 2026-10-13 19:13:58 +00:00

Owner

Copy link

Stacked on #9 (which is stacked on #8); the diff against r5900-ee includes their commits until those merge.

GCC turns a call in tail position into a jump (jr $25 / j callee) when the callee's arguments fit in the caller's incoming argument area. LLVM's MIPS backend disables tail calls by default (mips-tail-calls, because of $gp handling in position independent code), so a wrapper that forwards its arguments keeps them in callee-saved registers across a normal call, needing a frame plus a save and restore for each.

A new tail-calls subtarget feature, implied by r3000a only, turns tail calls on. For a CPU with the feature and no SVR4 abicalls the callee no longer has to be DSO-local, since there is no $gp to restore. Other CPUs are unchanged (the existing tailcall.ll test passes unmodified); the mips-tail-calls option still overrides the feature either way. Test: CodeGen/Mips/r3000a-tail-calls.ll.

Results (-Os/-Oz): sdk-core IOP modules at -Oz 57,808 -> 56,836 bytes (-1.7%, +9.5% -> +7.6% over GCC). RiptOPL IOP modules at -Oz: the 46 IRX files 684,894 -> 681,166 (-0.5%), now +4.7% over GCC's 650,850. Each forwarding wrapper loses its frame and its register saves.

Tested

• Mips CodeGen and MC lit suites: 1,517 passed, 23 expected failures, 0 failed.
• sdk-core regression tests (9, including a new IOP tail-call test) and diff runs (15) in PCSX2 with the new compiler.
• The new test covers forwarding with 0 to 6 arguments, reordered and rotated stack arguments, growing and shrinking argument lists, calls through function pointers, and a tail call after another call.
• RiptOPL builds completely (892 steps) and all 46 IRX pass iopfixup --irx1.
• On hardware (Deckard): the tail-call test, iop-mmio, the sdsq/sdhd/reboot modules at -Os and -Oz, and audsrv.
• Not tested: RiptOPL itself on hardware.

🤖 Generated with Claude Code

Stacked on #9 (which is stacked on #8); the diff against `r5900-ee` includes their commits until those merge. GCC turns a call in tail position into a jump (`jr $25` / `j callee`) when the callee's arguments fit in the caller's incoming argument area. LLVM's MIPS backend disables tail calls by default (`mips-tail-calls`, because of $gp handling in position independent code), so a wrapper that forwards its arguments keeps them in callee-saved registers across a normal call, needing a frame plus a save and restore for each. A new `tail-calls` subtarget feature, implied by `r3000a` only, turns tail calls on. For a CPU with the feature and no SVR4 abicalls the callee no longer has to be DSO-local, since there is no $gp to restore. Other CPUs are unchanged (the existing `tailcall.ll` test passes unmodified); the `mips-tail-calls` option still overrides the feature either way. Test: `CodeGen/Mips/r3000a-tail-calls.ll`. **Results** (-Os/-Oz): sdk-core IOP modules at -Oz 57,808 -> 56,836 bytes (-1.7%, +9.5% -> +7.6% over GCC). RiptOPL IOP modules at -Oz: the 46 IRX files 684,894 -> 681,166 (-0.5%), now +4.7% over GCC's 650,850. Each forwarding wrapper loses its frame and its register saves. **Tested** - Mips CodeGen and MC lit suites: 1,517 passed, 23 expected failures, 0 failed. - sdk-core regression tests (9, including a new IOP tail-call test) and diff runs (15) in PCSX2 with the new compiler. - The new test covers forwarding with 0 to 6 arguments, reordered and rotated stack arguments, growing and shrinking argument lists, calls through function pointers, and a tail call after another call. - RiptOPL builds completely (892 steps) and all 46 IRX pass `iopfixup --irx1`. - On hardware (Deckard): the tail-call test, iop-mmio, the sdsq/sdhd/reboot modules at -Os and -Oz, and audsrv. - Not tested: RiptOPL itself on hardware. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

techwritescode added 1 commit 2026-10-13 19:13:58 +00:00

[Mips] Permit tail calls on the R3000A ... 267389ace7

Tail calls are off by default in the MIPS backend because of $gp handling in
position independent code. Add a tail-calls subtarget feature, implied by
r3000a, that turns them on, and drop the requirement for a local callee when
the code has no SVR4 abicalls: without $gp there is nothing to restore. A caller
that ends in a call then needs neither a frame nor callee-saved registers to
hold its arguments across it. The mips-tail-calls option still overrides the
feature.

techwritescode merged commit 267389ace7 into r3000a-shared-global-base 2026-10-13 19:16:44 +00:00

techwritescode deleted branch r3000a-tail-calls 2026-10-13 19:16:46 +00:00

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

ps2/llvm!10

No description provided.