prepare_output_image()

void MayaFlux::Yantra::TextureExecutionContext::prepare_output_image ( uint32_t  width, uint32_t  height  )

inline

Allocates the output storage image and stages it at the declared output binding.

Only called in CONTAINER or IMAGE mode, from on_before_gpu_dispatch or per-step by callers chaining multi-pass sequences. The output slot keeps the images it has used, keyed by their creation parameters, so a caller that alternates between a few sizes, such as a sequence that downsamples every run, reuses them instead of allocating on every change. Images beyond a byte budget are dropped least recently used first, and asking for a dropped shape allocates again.

TextureLoom retains every image it creates for the life of the process, so an image that is dropped from the cache is not freed.

Definition at line 375 of file TextureExecutionContext.hpp.

    {
        auto& slot = output_slot();
        if (m_avoid_output_alias) {
            auto primary = acquire_output_image(slot, width, height);
            const auto input = input_slot().image;
            slot.image = primary && primary == input
                ? acquire_output_image(slot, width, height, true)
                : std::move(primary);
            slot.width = width;
            slot.height = height;
        } else if (!slot.image || slot.width != width || slot.height != height) {
            slot.image = acquire_output_image(slot, width, height);
            slot.width = width;
            slot.height = height;
        }
        stage_image_at(slot.binding.binding, slot.image, GpuBufferBinding::ElementType::IMAGE_STORAGE);
    }

References height, input, and width.