anisotropy-without-anisotropic-filter

Status: shipped (Phase 3) — see CHANGELOG.

(via ADR 0011)

What it detects

A SamplerState (or SamplerComparisonState) whose descriptor sets MaxAnisotropy > 1 while the Filter field is not one of the anisotropic filter modes (D3D12_FILTER_ANISOTROPIC, D3D12_FILTER_COMPARISON_ANISOTROPIC, D3D12_FILTER_MINIMUM_ANISOTROPIC, D3D12_FILTER_MAXIMUM_ANISOTROPIC). The detector enumerates sampler descriptors via Slang reflection and matches the MaxAnisotropy field against the Filter field on each. It does not fire when MaxAnisotropy is 1 (the trivial default) or when the filter is anisotropic; it fires on the silent-misconfiguration case where MaxAnisotropy was set with the intent of enabling AF but the filter selector was left on linear/point.

Why it matters on a GPU

The MaxAnisotropy field of a sampler descriptor is consumed by the hardware anisotropic filtering path only when the Filter selector requests anisotropic filtering. AMD RDNA 2/3 routes anisotropic samples through the TMU's anisotropic footprint estimator, which computes the elongation of the sample footprint in texture space and issues up to MaxAnisotropy taps along the major axis. NVIDIA Turing/Ada and Intel Xe-HPG document equivalent anisotropic taps. The field is ignored under linear or point filtering: the hardware does not consult MaxAnisotropy because the chosen filter does not invoke the anisotropic footprint estimator.

The cost of the mismatch is purely the lie in the descriptor. A MaxAnisotropy = 16 set against a LINEAR filter has no runtime cost — the hardware honestly ignores it — but it strongly implies to a reader (and to any tooling that scans descriptors for "16x AF" status) that the texture is being sampled with anisotropic filtering. The visual quality is whatever linear or point filtering produces, and that mismatch surfaces as a "why does this surface look poor at oblique angles?" bug that is hard to track down because the anisotropy field looks right.

The fix is one of: change Filter to an anisotropic mode (and accept the AF runtime cost), or set MaxAnisotropy = 1 to make the descriptor honest. The rule does not pick; it surfaces the inconsistency.

Examples

Bad

hlsl

// Descriptor:
//   Filter        = D3D12_FILTER_MIN_MAG_MIP_LINEAR
//   MaxAnisotropy = 16
// MaxAnisotropy is silently ignored; surface looks like trilinear, not 16xAF.
SamplerState LinearLooksAniso : register(s0);

Good

hlsl

// Either commit to anisotropic filtering...
// Descriptor:
//   Filter        = D3D12_FILTER_ANISOTROPIC
//   MaxAnisotropy = 16
SamplerState Aniso16 : register(s0);

// ...or make the descriptor consistent with the filter:
// Descriptor:
//   Filter        = D3D12_FILTER_MIN_MAG_MIP_LINEAR
//   MaxAnisotropy = 1
SamplerState TrilinearHonest : register(s1);

Options

none

Fix availability

suggestion — Both directions of the fix have visual or performance consequences (real AF is more expensive than trilinear). The diagnostic surfaces the descriptor inconsistency; the author edits the root signature.

anisotropy-without-anisotropic-filter ​

What it detects ​

Why it matters on a GPU ​

Examples ​

Bad ​

Good ​

Options ​

Fix availability ​

See also ​