Riley Performance: PSF resolution and high SSAA tile sizes

[ScepticalRabbit]

• Resolve of PSF is very slow and can be almost as slow as the raster loop itself
• Resolve of high SSAA is also slow even with tile sizes down to 1x1 pixels

See reporting block below:

================================================================================
Raster Frame Times: Frame 1, Camera 0
================================================================================
Camera Pixels              = 1020x252
Camera Distortion          = none
Camera PSF                 = gaussian
Camera Subsamples          = 8x8
Tile Size Pixels           = 1x1
Tile Size Subpixels        = 8x8
Tile Halo Pixels           = 4
--------------------------------------------------------------------------------
Vis Elems                  = 792
Total Elems                = 792
Vis %                      = 100.00%
Total SubPx                = 16450560
Shaded SubPx               = 1327638986
Shaded %                   = 8070.48%
--------------------------------------------------------------------------------
Setup Frame Buff           = 0.073058 ms
Prep Frame                 = 0.000881 ms
Geometry Preparation       = 0.516605 ms
  Coord Ops                = 0.069151 ms
  Cull Ops                 = 0.278516 ms
  Prep Hulls Shaders       = 0.163890 ms
  Remap Inds               = 0.000481 ms
Elem/Tile Overlap          = 18.447992 ms
Cam Invert Time            = 454.066107 ms
Elem Loop Time             = 45562.797939 ms
Scratch Resolve Time       = 1507.198537 ms
Raster Loop                = 6169.085995 ms
Save Frame                 = 0.552604 ms
--------------------------------------------------------------------------------
ACTIVE FRAME TIME          = 6188.676 ms
FRAME LATENCY              = 6188.678 ms
--------------------------------------------------------------------------------
Geom. Node Throughput      = 0.17 MNodes/s
Geom. Elem. Throughput     = 0.04 MElem/s
Subpx Raster Throughput    = 2.67 MSubPx/s
Raster Throughput          = 0.04 MPx/s
Active Frame Throughput    = 0.04 MPx/s
================================================================================
Frame Output Path =
    out/exp3_riley_render_texfloat_psf_im1020x252/plate260x65_cam256_quad9_chirp/diskaddsat_blackfrac0.6_uniform_j0.25_seed3_linear_os4_ss8_f
================================================================================

Solution:

• Use a full in memory sub-pixel buffer or striped sub-pixel buffer to raster into and then do the resolve as a separate step to avoid re-rastering the halo