Slow performance

[isovector]

Rendering nontrivial meshes takes forever at small resolutions.

[isovector]

Currently, almost all of our time is being spent in getImplicitShared. We can shave off a lot of that by specializing it:

{-# SPECIALIZE getImplicitShared :: SharedObj SymbolicObj2 ℝ2 -> ℝ2 -> ℝ #-}
{-# SPECIALIZE getImplicitShared :: SharedObj SymbolicObj3 ℝ3 -> ℝ3 -> ℝ #-}

[isovector]

Enabling StrictData and -funbox-strict-fields reduces the allocation by ~7x.

[isovector]

After both of the optimizations above, we're still spending a huge chunk of time inside getImplicitShared:

	Sat Dec 19 18:22 2020 Time and Allocation Profiling Report  (Final)

	   cad-exe +RTS -N -p -RTS

	total time  =       72.96 secs   (270323 ticks @ 1000 us, 8 processors)
	total alloc = 119,409,782,968 bytes  (excludes profiling overheads)

COST CENTRE               MODULE                                         SRC                                                               %time %alloc

getImplicitShared         Graphics.Implicit.ObjectUtil.GetImplicitShared Graphics/Implicit/ObjectUtil/GetImplicitShared.hs:(52,1)-(93,77)   20.5   13.5
shared_translate          Graphics.Implicit.ObjectUtil.GetImplicitShared Graphics/Implicit/ObjectUtil/GetImplicitShared.hs:82:43-91         10.4    6.9
fmap                      Linear.V3                                      src/Linear/V3.hs:128:3-42                                           8.8   15.1
getImplicit3              Graphics.Implicit.ObjectUtil.GetImplicit3      Graphics/Implicit/ObjectUtil/GetImplicit3.hs:(30,1)-(156,69)        7.3    6.5
shared_union              Graphics.Implicit.ObjectUtil.GetImplicitShared Graphics/Implicit/ObjectUtil/GetImplicitShared.hs:60:39-66          6.6   10.0
-                         Linear.V3                                      src/Linear/V3.hs:200:3-18                                           6.0    5.5
dot                       Linear.V3                                      src/Linear/V3.hs:270:3-51                                           5.0    6.0
reflect                   Graphics.Implicit.MathUtil                     Graphics/Implicit/MathUtil.hs:159:1-56                              4.6    4.9
shared_translate_subtract Graphics.Implicit.ObjectUtil.GetImplicitShared Graphics/Implicit/ObjectUtil/GetImplicitShared.hs:82:86-90          3.8    0.0
<*>                       Linear.V3                                      src/Linear/V3.hs:170:3-46                                           3.6    4.2
getImplicit2              Graphics.Implicit.ObjectUtil.GetImplicit2      Graphics/Implicit/ObjectUtil/GetImplicit2.hs:(41,1)-(69,50)         2.1    3.2
shared_mirror             Graphics.Implicit.ObjectUtil.GetImplicitShared Graphics/Implicit/ObjectUtil/GetImplicitShared.hs:86:31-61          2.0    1.8
*                         Linear.Quaternion                              src/Linear/Quaternion.hs:(275,3)-(276,73)                           2.0    3.0
shared_intersection       Graphics.Implicit.ObjectUtil.GetImplicitShared Graphics/Implicit/ObjectUtil/GetImplicitShared.hs:64:36-79          1.4    1.0
rmaximum                  Graphics.Implicit.MathUtil                     Graphics/Implicit/MathUtil.hs:(79,1)-(90,38)                        1.4    0.6
bf                        Graphics.Implicit.Export.TextBuilderUtils      Graphics/Implicit/Export/TextBuilderUtils.hs:38:1-64                1.1    3.5
fmap                      Linear.V2                                      src/Linear/V2.hs:156:3-34                                           1.1    2.3
rminimum                  Graphics.Implicit.MathUtil                     Graphics/Implicit/MathUtil.hs:(97,1)-(108,25)                       1.1    0.0
+                         Linear.V3                                      src/Linear/V3.hs:198:3-18                                           1.0    2.8
-                         Linear.V2                                      src/Linear/V2.hs:226:3-18                                           0.9    1.4
getMesh                   Graphics.Implicit.Export.Render                Graphics/Implicit/Export/Render.hs:(77,1)-(188,50)                  0.8    1.8

I've added cost centers to each case in getImplicitShared, which is what the shared_union et al. CCs are.

profile:

cad-exe.prof.txt

flame graph:

flame.svg.txt

[isovector]

I think the reason for the massive costs of translate and union in this case is twofold:

1. we're calling getImplicit a lot
2. my model has lots of constructions of the form translate _ $ union [translate _ $ union [...], translate $ union [ ...] ]

I think in general we want to add a simplification pass that pushes translates through union/intersection/difference/shell/outset/complement. Sounds crazy, I admit, but this gives us drastically more opportunities to simplify translate . translate.