[Ducengarth]

What

Implement 26 Dinkumware extended-precision helpers in libSceLibcInternal, replacing their parameterless throwing placeholders in Export.cpp:

• _Dtento, _FDtento, _LDtento: *value * 10^power, where value points to a three-word extended-precision accumulator that is updated in place. On overflow or underflow they set the guest errno to ERANGE (34) and OR 1 into *error when it is not null.
• _FXp_addh, _FXp_addx, _FXp_invx, _FXp_ldexpx, _FXp_movx, _FXp_mulh, _FXp_mulx, _FXp_setn, _FXp_setw, _FXp_sqrtx, _FXp_subx (float).
• _LXp_addh, _LXp_addx, _LXp_getw, _LXp_invx, _LXp_ldexpx, _LXp_movx, _LXp_mulh, _LXp_mulx, _LXp_setn, _LXp_setw, _LXp_sqrtx, _LXp_subx (long double).

An extended-precision value is an array of words in decreasing magnitude, each holding at most half the significand (12 bits for float, 26 for double, 32 for long double), ended by a zero word. One template in the new ExtendedPrecision.cpp serves all three types.

Source: Dinkumware V6.00 as shipped with Visual C++ 2013 (xxxprec.h, xxxdtent.h, prototypes in xmath.h), the reference the debt file already uses for _Thrd_sleep. The code is written from that algorithm, not copied. long arguments are std::int64_t, as in the PS5 ABI. Differences from V6.00, all recorded in docs/dev/debt/libSceLibcInternal.md:

• _LXp_getw and the reads inside _Dtento follow the console's _FXp_getw from feat(libSceLibcInternal): answer _Daysto, _FXp_getw and _Isdst as firmware 13.42 does #14945, read from firmware 13.42: they add the fourth word when there are at least four, instead of V6.00's sticky bit. This also means the console runs a later revision than V6.00, so other helpers may differ in details.
• Inputs on which V6.00 reads past an array or never returns are defined here. A one-word _Xp_mulx multiplier reads only that word (same result). A word count below 1 does nothing in _Xp_movx, _Xp_addh, _Xp_invx and _Xp_sqrtx, where V6.00 calls memcpy with a negative size for a finite invx or sqrtx argument. A _Dtento power of LONG_MIN underflows instead of looping on n = -n. The _Xp_addh carry bit count is not truncated to short.

Scope: _FXp_getw is already in main (#14945). The double _Xp_* exports are still unnamed placeholders, and #15777 names them; they can follow once it merges. _FAtan, _FSincos, _FTan, _FTgamma, _FLogpoly, _FQuad, _FQuadph and their _L and double versions still throw: no source I found gives their signatures or code.

No title's trace or disassembly was used, and no export was measured on a PS5.

pr_overlap reports no other pull request with the same exports or new files, only line conflicts: #15767 adds other math declarations between these in MathInternals.hpp and next to the _Dtento stubs in Export.cpp, #13881 and #14085 edit neighbouring debt lines, and #15665 and my #16146 add a test block at the same place in CMakeLists.txt. Whichever merges second resolves them; I will rebase this one as needed. Nine pull requests that conflict with main could not be compared by the tool (#5021, #13486, #13778, #13782, #14212, #14664, #14936, #15581, #15899); none of their diffs adds any of these 26 names.

Tested

Windows 11, MinGW-w64 GCC (WinLibs), Release build. No title or homebrew, and not built on Linux.

• guest_libc_internal_extended_precision (new) passes, and passed 3 more runs in a row. It calls every one of the 26 exports, and _FXp_getw from main, and checks:
  • setw splits 20000 random floats into two exact half-width words, plus zero, one word, infinity, no room and subnormals.
  • setn holds 32-bit integers exactly and rounds the quotient by 10000 to float first for wider ones.
  • addh keeps exact sums of half-precision addends in decreasing words, and handles cancellation, infinity, Inf - Inf (FE_INVALID), NaN and a full array.
  • mulh and mulx keep exact or six-word products.
  • movx, ldexpx, addx and subx behave as above, and so do invx and sqrtx, within 2^-40 and 2^-36 of the exact result, read back within one ulp. sqrtx also covers negative values, -0 and ±infinity.
  • _Dtento gives correctly rounded powers of ten from 10^-22 to 10^22, the exact one-tenth words for each type, overflow, underflow, a null error pointer and a power of INT64_MIN.
  • _LXp_getw adds a fourth word where V6.00's sticky bit would round differently.
• Making getw ignore the fourth word, making mulh drop the low half of a partial product, or changing the last word of the long double one-tenth table each makes the test fail with the matching message.
• guest_libc_internal_firmware_answers, guest_libc_internal_aligned_new, guest_libc_internal_exception_destructors, guest_libc_internal_mtx, guest_libc_internal_pad and guest_statvfs pass. The full suite was not run.
• objdump -p on libSceLibcInternal.prx lists all 26 names.
• python3 tools/check_conventions.py --base upstream/main passes.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed under Silent stubs in their library's technical debt file
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• Fixes: none
• Unblocks: none
• GPUs tested: n/a
• AI-assisted: yes

🤖 Generated with Claude Code

[JJenkx]

Windows check of head c3798b1149, merged onto main bf80054e7c (76c6252c44, clean merge).

Setup: Windows 11 Enterprise 22621, WinLibs GCC 15.2 UCRT, Release. The hosted Windows check does not run these tests on pull requests.

• guest_libc_internal_extended_precision passes in 3 of 3 runs, none skipped.

AI-assisted: yes.

[github-actions]

Progress report (6bf3a19 → bfbd4f9)

📈 System libraries: 29.91% (+0.06%, +26 implemented functions)

✅ 26 implemented

Library	Name
libSceLibcInternal	_Dtento_nid_postfix
libSceLibcInternal	_FDtento_nid_postfix
libSceLibcInternal	_FXp_addh_nid_postfix
libSceLibcInternal	_FXp_addx_nid_postfix
libSceLibcInternal	_FXp_invx_nid_postfix
libSceLibcInternal	_FXp_ldexpx_nid_postfix
libSceLibcInternal	_FXp_movx_nid_postfix
libSceLibcInternal	_FXp_mulh_nid_postfix
libSceLibcInternal	_FXp_mulx_nid_postfix
libSceLibcInternal	_FXp_setn_nid_postfix
libSceLibcInternal	_FXp_setw_nid_postfix
libSceLibcInternal	_FXp_sqrtx_nid_postfix
libSceLibcInternal	_FXp_subx_nid_postfix
libSceLibcInternal	_LDtento_nid_postfix
libSceLibcInternal	_LXp_addh_nid_postfix
libSceLibcInternal	_LXp_addx_nid_postfix
libSceLibcInternal	_LXp_getw_nid_postfix
libSceLibcInternal	_LXp_invx_nid_postfix
libSceLibcInternal	_LXp_ldexpx_nid_postfix
libSceLibcInternal	_LXp_movx_nid_postfix
libSceLibcInternal	_LXp_mulh_nid_postfix
libSceLibcInternal	_LXp_mulx_nid_postfix
libSceLibcInternal	_LXp_setn_nid_postfix
libSceLibcInternal	_LXp_setw_nid_postfix
libSceLibcInternal	_LXp_sqrtx_nid_postfix
libSceLibcInternal	_LXp_subx_nid_postfix

Changed libraries (implemented exports)

Library	Before	After	Change
libSceLibcInternal	106/1141 (9.29%)	132/1141 (11.57%)	+2.28%