Selection is by name, so a non-site 128 array is legitimate output. The shape check is a function and the array-128 variant passes it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
SPIRV-Cross names the macro from the SpecId and offers no option to rename it, so the SpecId is CultMath's own
('CMS1'); partition.py reads it from spvedit.py.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
wrappers.py compiles the library with CULTMATH_MAX_SITES defined as a sentinel no shader file spells, so the three
site arrays get an array length id of their own and spvedit.py converts exactly that constant to the overridable
specialization constant (default from the HLSL guard). A library array of literal length 128 keeps its own id and
stays a plain constant. build-glsl.sh gains the pipeline function and scratch variants: a macro outside a declarator,
a folded expression reaching the capacity id, and a non-site float4 x[128] parameter that a consumer's override must
not resize. The [128] shape arm goes: a non-site 128 array is legitimate output.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CULTMATH_MAX_SITES defaults to 128 under #ifndef. The consumer-written ring-loop and border-loop comment recipes are deleted. The C# mirror takes the array as T[]; the span twins live in SiteArrays; the GLSL lowering keeps the capacity macro; gen.py learns the array leaf; the smooth-step family gains its cap-point draw; CULTMATH_WRITE_GLSL refuses non-Linux.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Per verdict s1: the judge keeps the fixture tolerance as correctness and adds a recorded sha256 of each family's
result per route (a change in the computation fails until re-recorded); a test refuses mediump/lowp/precision
other than highp and min16float/min10float/min16int/half in the committed shaders; phacelle's tolerance text states
a device clause (2^-18) that the judge parses and enforces. The committed bounds file lacks the new column until
the next WRITE_BOUNDS=1 job, so this commit's device run fails and it does not pass DeviceResultsHoldFixture.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tools/device/gen.py spells one HLSL and one GLSL ES 3.10 compute shader per work
item, vkrun.c dispatches it on the first Vulkan device, run-lavapipe.sh runs the
whole conforming job. None of them compares. Also fixes the metric test's family
choice. Not yet run: these may not build or run.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The wrapper built one empty argument for an empty parameter list, so cultmath_site_cells_empty() failed glslang.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
dxc runs with -WX; a ShaderPath that is not a file throws naming the parameter and entry index,
never the path; the GLSL wrapper builds a struct parameter from its declaration's field zeros.
design.md and README state the struct-argument twin rule, the span twins and the missing shader identity.
No C# change; the compiles run on Starfire after the last commit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The spec's negative check greps the script for arm64; the comment matched. Comment
only; script rerun on Starfire (compiles both copies, exit 0; arm64 dxc control throws). Builds.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The b == 0 branch now guards t1 != t0 like the sloped branch (C# and both HLSL
copies), so an infinite bound times a zero length no longer forms NaN.
Edge-grid mirror test and a counting-clamp shadow test replace the text-order
check; the four pin tests assert equality with their measured expression;
tools/compile-hlsl.ps1 is the one dxc compile verdict for the HLSL copies.
Builds. CultMath.glsl and glsl-parity.json are not yet regenerated, so
CommittedGlslEqualsLowering and GoldenFixtureMatchesCSharp are red at this
commit; the next commit regenerates them.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
cycles_power read b - 2 floor(0.5 b) != 0 as odd, which is NaN for b = +-inf, so a negative base got a negative result where Cycles' powf(a, +-inf) is never negative. Odd now means the remainder is exactly 1 (C#, HLSL, Unity copy). oracle.py gains the infinite family (operands made in-kernel as +-exp(100), results read through comparisons with 0) for power, modulo and Map Range Smooth Step; the infinite.json fixture and the regenerated GLSL follow in the next commit. This commit builds; InfiniteOperandsFollowBlender and CommittedGlslEqualsLowering fail until it lands.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fixtures and the regenerated GLSL follow in the next commit, so the tests that read the new rows
(InverseTrigonometry..., OperandStreams...) are red at this commit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The z channel's boundary mutant survived because 0/0 reads as 0 after the clamp. mix.json follows. Builds.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Stryker's survivors in math.Cycles.cs that were not equivalent: Burn's divisor <= 0 boundary, Logarithm's
a <= 0 guard, and rgb_to_hsv's h < 0 wrap. mix.json is regenerated in a later commit, so
FamilyMatchesBlender is unchanged here and the new Burn rows are absent. Builds.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The film zeroes an infinite result, which had hidden both the power quirk and Compare's epsilon floor
from the fixtures (two hand mutants of the HLSL survived). The fixtures are regenerated in a later commit.
Builds.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The oracle's first fixtures showed Cycles' AVX2 kernel fuses multiply-adds and its film zeroes
non-finite results; the tool now exits 4 off the SSE4.2 kernel (CYCLES_CPU_NO_AVX2=1) and the
test skips cases Cycles cannot report. The Unity copy of CultMath.hlsl carries the new include.
Builds. Fixtures and CultMath.Cycles.glsl are still to come.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
oracle.py bakes each Cycles node under Blender 5.2.2 LTS (build d13f752e3b9c, refusing any other)
into float images; CyclesOracleTests holds math.cycles_* to the fixtures it writes. No fixtures
exist at this commit, so the tests fail here; builds.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
compile-glsl.ps1 writes the wrapper a WebGL2 consumer writes (#version 300 es, highp precision,
the library by concatenation, one call of every public function) and compiles it with glslang;
-Body replaces the calls, for the negative control.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Aetheria Cut 8a needs int2(float2), clamp(int, int, int), and
frac(double). Fill them in CultMath as dxc compiles them:
- generate-swizzles.ps1 emits same-size conversion constructors and
math functions between float, int, and bool vectors. dxc lowers
float to int as fptosi (toward zero; NaN and out-of-range undefined),
numeric to bool as != 0 (fcmp une, NaN is true), bool to numeric as
0 or 1. floatN(int) splat keeps floatN(0) from going ambiguous
against floatN(intN).
- Int and int-vector clamp are min(max(x, a), b), matching dxc's
IMin(IMax(x, a), b); inverted bounds return b.
- frac(double) keeps double precision. dxc demotes double frac to
float, as it does exp and lerp; CultMath's existing double overloads
already diverge that way, and design.md now says so.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
get-dxc.ps1 fetched releases/latest with no pin or hash, so the compiler that
defines parity could change under the tests. It now downloads v1.9.2607's
dxc_2026_07_29.zip (dxcompiler 1.9.0.5402) and refuses a zip whose SHA256 is
not A1DFB116...CD241. The hash was taken from the zip downloaded from
Microsoft's GitHub release and matches the digest GitHub publishes for the
asset. Script ran end to end and installed 1.9.0.5402.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
tools/generate-swizzles.ps1 writes the checked-in src/CultMath/Swizzles.g.cs
as partial structs plus a partial math: for float2/3/4, int2/3/4 and bool2/3/4
every swizzle of length 2-4 including repeats, in xyzw and rgba naming, with
setters only on non-repeating masks, plus single-component r/g/b/a. double2/3
get lengths 2-3 (there is no double4). Every HLSL mixed constructor
(float4(float2, float2), float4(float, float2, float), copy forms, ...) exists
as a struct constructor and, for float/int/bool, a math function. The
hand-written swizzles and mixed constructors it supersedes are deleted.
Matrices store private rows and return ref float2/float3 from the row
indexer ([UnscopedRef], polyfilled for netstandard2.1), so m[1][2] = s
compiles as in HLSL; _mRC are get/set properties over the rows. Equality and
hashing stay value-based.
Intrinsics follow HLSL: sign returns int/intN and yields 0 for NaN; step is
x >= y ? 1 : 0 so NaN yields 0; any/all accept float and int vectors;
min/max/abs on int vectors return int vectors, and abs(int) returns int.
Strips the UTF-8 BOMs this branch added to math.cs, CultMath.csproj and
HlslSemanticsTests.cs.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
