cajeta.gfx.Texture2D — a read-only 2-D image of texels, sampled in a kernel through a Sampler with hardware filtering. Like KernelBuffer, it is a small host-side handle over a device image resource, with the same RAII memory model: the constructor acquires the device image, ~Texture2D() releases it at scope exit, and a launch borrows each Texture2D argument until the next KernelStream.sync(). The type parameter T defaults to float32, so every bare Texture2D spelling is exactly Texture2D<float32>; sample/fetch return Vector<T, 4>. The storage format is chosen at construction from TextureFormat (default R32F). sample is float-only; integer textures (T = int32/uint32) are fetch-only — the hardware cannot filter integer texels.

uint32 w = 16;
uint32 hgt = 16;
float32[] pixels = heap float32[w * hgt];
Texture2D tex = heap Texture2D(w, hgt);   // R32F, one float channel
tex.upload(pixels);
// in an @Kernel: tex.sample(samp, u, v) / tex.fetch(x, y)

Methods

Signature
`Texture2D(uint32 width, uint32 height)` ⚑	RAII constructor: allocate a `width` x `height` float32 (`R32F`) device image
`Texture2D(uint32 width, uint32 height, int32 format)` ⚑	RAII constructor with an explicit storage `TextureFormat` ordinal, e.g. `TextureFormat.RGBA8_UNORM`
`Texture2D(uint32 width, uint32 height, int32 format, uint32 levels)` ⚑	RAII constructor with explicit mip levels — a mipmapped texture; upload each level with `uploadLevel`
`uint32 textureWidth()`	Image width in texels
`uint32 textureHeight()`	Image height in texels
`void upload(T[] host)`	Host → device transfer of `width`*`height` row-major, channel-interleaved texels
`void uploadLevel(uint32 level, T[] host)`	Host → device transfer of one mip level (dims `max(1, width>>level)` x `max(1, height>>level)`)
`Vector<T, 4> sample(Sampler s, float32 u, float32 v)`	Device-only intrinsic: filtered sample at normalized (u, v) in [0, 1] through `s`, returning the RGBA texel
`Vector<T, 4> fetch(uint32 x, uint32 y)`	Device-only intrinsic: unfiltered, sampler-free read of the texel at exact integer (x, y), mip level 0
`Vector<T, 4> fetchLod(uint32 x, uint32 y, uint32 lod)`	Device-only intrinsic: explicit-LOD `texelFetch` — the mip-aware twin of `fetch`
`Vector<T, 4> sampleLod(Sampler s, float32 u, float32 v, float32 lod)`	Device-only intrinsic: explicit-LOD sample; fractional LODs blend two levels (trilinear); float-only

⚑ = @EntryPoint

sample, fetch, fetchLod, and sampleLod are lowered at the call site inside an @Kernel / @Device function to the backend’s image ops; calling them on the host is unsupported.

Texture2D<T = float32>

Methods

See also