[hypermoist]

What

Adds 40 libc math exports that AnyPS5 did not have, in a new src/MathExact.cpp. Every one is matched exactly against the console:

• long double: copysignl, fdiml, fmodl, remainderl, remquol, logbl, ilogbl, rintl, roundl, lroundl, lrintl, llrintl, sqrtl, modfl, nanl, scalblnl, fmal
• double/float: rint(f), lrint(f), llrint(f), drem(f), scalbln(f), scalb(f), finite(f)
• classification: __fpclassifyl, __isinfl, __isnanl, __isnormall, _LDtest, _Dclass, _FDclass, _LDclass, _LDsign

Console behaviour that is reproduced (found by running the console binary side by side, see Tested):

• fdiml, roundl, modfl, and fmal with a NaN z, return a signalling NaN operand unquieted; rint/rintf quiet it.
• nanl ignores its argument and always returns the default quiet NaN.
• fmodl sets EDOM whenever x is infinite or NaN, y is zero or y is NaN (NaN inputs included); remquol, remainderl and drem(f) set it only for real domain errors. scalb(f) never touches errno.
• remquol reports the quotient modulo 8 with its sign (glibc can report ±8 where the console reports 0).
• ilogbl(NaN) is INT_MAX; scalblnl truncates its exponent to int (scalblnl(1, 2^32 + 1) == 2).
• fmal is FreeBSD's: x*y + z when x or y is zero or non-finite, x*y when z is zero, z when only z is non-finite, and x*y when z is more than 128 binades below the product (so a midpoint product rounds to even instead of using z's sign).
• __fpclassifyl returns the Dinkumware codes (-2 denormal, -1 finite, 0 zero, 1 infinite, 2 NaN), like _LDtest; they differ only on pseudo-denormals, which _LDtest treats as finite. _LDsign returns the raw sign bit (-32768).

Two known gaps are recorded in docs/dev/debt/libc.md: remquol leaves *quo unchanged when the result is NaN (the console stores a value that depends on earlier x87 state), and non-canonical 80-bit encodings passed to the arithmetic functions follow the host x87.

Tested

Linux (CachyOS, GCC Release build, plus a clang -Wall -Wextra syntax check): new guest_math_exact test and the full ctest suite pass. tools/check_conventions.py passes; tools/pr_overlap.py --branch reports no line overlap.

Differential test against the console: the PPSA08427 libc.prx was mapped into a Linux process (its __error import bound to the test's errno) and all 40 functions were run side by side with this implementation on 200,000 random inputs each (random exponents and mantissas, signalling and quiet NaNs with payloads, infinities, zeros, subnormals, huge long exponents). Results agree bit for bit, and errno agrees, everywhere except the remquol NaN quotient noted above. All 40 NIDs are also exported by a newer libc.prx build (J03848509).

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
• Unblocks: none
• GPUs tested: n/a
• AI-assisted: yes

🤖 Generated with Claude Code

https://claude.ai/code/session_01Q8tinbP1MUpUaDsbNWnGnu

[github-actions]

Progress report (9f5d4bd → 1181068)

📈 System libraries: 91.8% (+0.08%, +40 implemented functions)

✅ 40 implemented

Library	Name
libc	_Dclass_nid_postfix
libc	_FDclass_nid_postfix
libc	_LDclass_nid_postfix
libc	_LDsign_nid_postfix
libc	_LDtest_nid_postfix
libc	__fpclassifyl_nid_postfix
libc	__isinfl_nid_postfix
libc	__isnanl_nid_postfix
libc	__isnormall_nid_postfix
libc	copysignl_nid_postfix
libc	drem_nid_postfix
libc	dremf_nid_postfix
libc	fdiml_nid_postfix
libc	finite_nid_postfix
libc	finitef_nid_postfix
libc	fmal_nid_postfix
libc	fmodl_nid_postfix
libc	ilogbl_nid_postfix
libc	llrint_nid_postfix
libc	llrintf_nid_postfix
libc	llrintl_nid_postfix
libc	logbl_nid_postfix
libc	lrint_nid_postfix
libc	lrintf_nid_postfix
libc	lrintl_nid_postfix
libc	lroundl_nid_postfix
libc	modfl_nid_postfix
libc	nanl_nid_postfix
libc	remainderl_nid_postfix
libc	remquol_nid_postfix
libc	rint_nid_postfix
libc	rintf_nid_postfix
libc	rintl_nid_postfix
libc	roundl_nid_postfix
libc	scalb_nid_postfix
libc	scalbf_nid_postfix
libc	scalbln_nid_postfix
libc	scalblnf_nid_postfix
libc	scalblnl_nid_postfix
libc	sqrtl_nid_postfix

[github-actions]

Other open pull requests touch the same code. Merging one will break the other, or both implement the same thing. If this one builds on them, list them in Depends on.

Pull request	Conflicting files (lines on main)	Same exports	Same new files
PR 7545	docs/dev/debt/libc.md L64

Some comparisons could not be completed because a pull request conflicts with the base branch. Review these shared files manually before concluding that there is no overlap.

Pull request	Shared files not compared
PR 3987	docs/dev/debt/libc.md
PR 5021	core/libs/prx/libc/CMakeLists.txt docs/dev/debt/libc.md
PR 5303	docs/dev/debt/libc.md
PR 6044	core/libs/prx/libc/CMakeLists.txt
PR 6226	docs/dev/debt/libc.md
PR 6469	docs/dev/debt/libc.md
PR 7039	docs/dev/debt/libc.md
PR 7081	core/libs/prx/libc/CMakeLists.txt docs/dev/debt/libc.md
PR 7084	core/libs/prx/libc/CMakeLists.txt
PR 7201	core/libs/prx/libc/CMakeLists.txt
PR 7332	core/libs/prx/libc/CMakeLists.txt docs/dev/debt/libc.md
PR 7715	core/libs/prx/libc/CMakeLists.txt docs/dev/debt/libc.md

[DavidSamir]

Title evidence for this PR: Spider-Man: Edge of Time (PPSA99300), added to the homebrew catalog today, stops on llrint on main.

Windows 11 10.0.26300, WinLibs GCC 15.2.0 r7, Release. main 0afa6a6 alone, then with this PR's commit 4f608a8 cherry-picked onto it (4 files, no conflicts); the catalog release converted with relinker --windows and run for 60 s.

main: the title exits after 1 s with 0xC0000139 (entry point not found):

FAIL: unresolved ELF import -431A-YBAks from libSceLibcInternal.prx

-431A-YBAks is llrint (tools/nid_names.py --nid=-431A-YBAks).

With this PR it starts, runs its ReXGlue startup and exits by itself with code 0 after reporting the files it needs from the user:

[Spider-Man Edge Of Time] startup: app: paths
[Spider-Man Edge Of Time] startup: fatal: reeot - game files missing: No Default.xex in /app0/gamedata. Install the game with reeot on a PC, then copy its game folder into the title's gamedata ...
[Spider-Man Edge Of Time] startup: main: returning success
[Spider-Man Edge Of Time] rex_ps5_app_main returned 0

That message is the title's own check for the game data it does not ship, so this PR is all it needs to reach it.

AI-assisted: yes

[jotitadev]

On macOS, drem, dremf and remainderl don't set EDOM. The other paths that the description says set errno (fmodl, remquol) set it explicitly. These three return std::remainder directly, so they rely on the host libm setting errno. glibc does that. Apple's libm never sets errno: math_errhandling is MATH_ERREXCEPT only.

Snippet, macOS 27.2, Apple clang (Xcode 27), M2 Max:

volatile double one = 1.0, zero = 0.0; volatile long double onel = 1.0L, zerol = 0.0L;
errno = 0; std::remainder(one, zero);   // errno
errno = 0; std::remainder(onel, zerol); // errno

== x86_64 (Rosetta)
math_errhandling=2 (MATH_ERRNO=1)
remainder(1,0)=nan errno=0 (EDOM=33)
remainderf(1,0)=nan errno=0
remainderl(1,0)=nan errno=0
== arm64
(same: errno=0 for all three)

So on a macOS host a guest gets errno == 0 where the console sets EDOM. CheckErrno in GuestMathExact.cpp would also abort at Require(std::isnan(drem_nid_postfix(1.0, 0.0)) && errno == EDOM). The fix would be the same explicit check that remquol already uses (!isnan(x) && !isnan(y) && (isinf(x) || y == 0) → errno = EDOM) in remainderl, drem and dremf. That makes them independent of the host libm.

Not tested: I did not build this PR. libc doesn't build on macOS from main, so this is the host-libm behaviour checked with a standalone snippet. Head 4f608a8.

Written with AI assistance (Claude Code). I ran the check above on my machine.

[nicolasfalesy]

A title that reaches this: Spider-Man: Edge of Time (PPSA99300) from the homebrew catalog, on Linux (Intel Arc, Mesa ANV).

• main 92e5041: stops at load with ./app.elf: symbol lookup error: ./app.elf: undefined symbol: -431A-YBAks (llrint, libc).
• main + this PR: it loads and runs to its own check, startup: fatal: reeot - game files missing: No Default.xex in /app0/gamedata, then exits with status 0 (no game data was installed for the test).

[adriadam10]

Thanks for the thorough differential testing. One thing to fix before merge, already reported by @jotitadev:

• drem, dremf and remainderl (MathExact.cpp:83-85, :179-185) return std::remainder directly, so errno == EDOM for x infinite or y == 0 depends on the host libm. glibc sets it; Apple's libm (and likely MinGW's) does not, so a guest gets errno == 0 where the console sets EDOM, and CheckErrno (GuestMathExact.cpp:71) aborts on those hosts. Use the same explicit check as remquol (!isnan(x) && !isnan(y) && (isinf(x) || y == 0) → errno = EDOM) in all three, and add remainderl and dremf cases to the test.

[cadenasynyster]

Windows results for this head (4f608a8), merged onto current main without conflicts. Windows 11, WinLibs GCC 15.2.0 r7 (x86_64-ucrt-posix-seh), Release.

1. guest_math_exact aborts on Windows, 3 of 3 runs (exit code 0xc0000409, SIGABRT inside the test). It is not CheckErrno. I ran a local copy of the test where Require prints its line number and carries on, so every failing check shows: exactly one fails, GuestMathExact.cpp:77 in CheckConsoleQuirks, the pair scalblnl(1.0L, 0x100000001L) == 2.0L && scalblnl(1.0L, LONG_MAX) == 0.5L. I did not isolate which half. The LONG_MAX half cannot hold on Windows, since long is 32 bits there: a standalone program prints sizeof(long) = 4, LONG_MAX = 2147483647, and std::scalbln(1.0L, LONG_MAX) = inf. On the console (and Linux), truncating LONG_MAX to int gives -1, hence 0.5. Everything else in the test passes on Windows, including the drem errno row (CheckErrno, line 71). The check probably needs an exponent that does not depend on the width of the host long.

2. errno of the host remainder family, for the review point on drem, dremf and remainderl. A standalone program calling std::remainder, std::fmod and std::remquo (not this PR's code), same toolchain:

• float and double: errno = EDOM (33) for (1,0) and for (inf,1); errno stays 0 for NaN arguments. So drem and dremf get EDOM from the host here.
• long double: errno stays 0 for every case, including (1,0) and (inf,1), for remainder, fmod and remquo.
  So on Windows/MinGW only remainderl has the gap; "likely MinGW's" holds for long double only. Going by this, a remainderl row in CheckErrno would abort on Windows without the explicit check (not run).

Not run: Linux and macOS, and the fix itself.

AI-assisted: yes