float32-blendable

Full-precision floating-point color targets can store fragment output without supporting the fixed-function blending step. Requesting float32-blendable at device creation allows a render pipeline to configure blend for r32float, rg32float, and rgba32float targets, combining incoming fragments with the values already in the attachment.

Scientific visualization and accumulation passes may need more range or precision than a half-float target can provide. This feature lets them use blending directly, for example to add contributions into a full-precision image. Higher precision still has a bandwidth cost, and floating-point accumulation remains sensitive to rounding and operation order.

This is independent of float32 filtering: blending controls render writes, while filtering controls sampled reads. If blending is unavailable, use a blendable format such as rgba16float when its precision is sufficient, or restructure the accumulation into compute passes with explicit synchronization and separate intermediate resources.