Improve launch performance of kernels some more

Created by: mwarusz

Using this script https://github.com/JuliaGPU/KernelAbstractions.jl/pull/80#issuecomment-605210053

Before:

[ Info: Ka Launch
BenchmarkTools.Trial:
  memory estimate:  816 bytes
  allocs estimate:  9
  --------------
  minimum time:     4.301 μs (0.00% GC)
  median time:      4.585 μs (0.00% GC)
  mean time:        4.617 μs (0.46% GC)
  maximum time:     220.415 μs (97.08% GC)
  --------------
  samples:          10000
  evals/sample:     7

After:

BenchmarkTools.Trial:
  memory estimate:  816 bytes
  allocs estimate:  9
  --------------
  minimum time:     98.354 ns (0.00% GC)
  median time:      121.497 ns (0.00% GC)
  mean time:        148.127 ns (17.31% GC)
  maximum time:     1.882 μs (87.36% GC)
  --------------
  samples:          9587
  evals/sample:     951