Luminescent said:
So let me get this straight: Xenos cannot blend or AA rendertargets/buffers at FP16 per component precision (per developer confirmation), but it can output to a rendertarget/buffer at FP16. Can it output at FP32 (w/o AA or blending of course)?
The following render target formats are supported:
- 8:8:8:8 fixed point with blending
- 2:10:10:10 fixed point with blending
- 2:7e3:7e3:7e3 packed floating point with blending (add-only at full rate, other modes at half-rate)
- 16:16 fixed point with blending
- 16:16:16:16 fixed point with blending (at half-rate)
- 16:16 floating point without blending
- 16:16:16:16 floating point without blending (at half-rate)
- 32 floating point without blending
- 32:32 floating point without blending (at half-rate)
The 32-bit-per-pixel formats all operate at full rate; the 64-bit-per-pixel formats operate at half rate. The 32-bit packed floating-point format supports additive blending (SRC + DEST) at full rate, while other blending modes are half rate.
Both 2× and 4× multisampling are supported. On resolve, the GPU can downsample multisampled render targets and do format conversion. However, multisample downsampling is limited to blendable target formats.
The 8:8:8:8 fixed-point format can be gamma corrected using an approximation to the sRGB gamma 2.2 curve to convert from linear light space to gamma space on writes and to convert from gamma space to linear light space on reads.