unexpected allocations / slow execution when using @SVector and rand

[BenediktAuer]

Hello,
on v1.9 I experience some wierd allocations and slow execution when using @SVector and rand with a specific type T:

 @benchmark @SVector rand(Float64,5)
BenchmarkTools.Trial: 10000 samples with 174 evaluations.
 Range (min … max):  606.052 ns …  21.025 μs  ┊ GC (min … max): 0.00% … 96.05%
 Time  (median):     626.333 ns               ┊ GC (median):    0.00%
 Time  (mean ± σ):   702.088 ns ± 752.018 ns  ┊ GC (mean ± σ):  4.21% ±  3.84%

  ▂█▇▄▄▄▃▂▂▁  ▁▁    ▂▂   ▁▁                                     ▁
  █████████████████████▇█████▇▆▇▆▇▇▇▇████▇▇▇██▅▆▄▅▄▅▄▅▄▄▄▄▄▄▅▄▄ █
  606 ns        Histogram: log(frequency) by time       1.16 μs <

 Memory estimate: 496 bytes, allocs estimate: 10.

Without a type specifier I get the expected result:

@benchmark @SVector rand(5)
BenchmarkTools.Trial: 10000 samples with 1000 evaluations.
 Range (min … max):  5.640 ns … 90.822 ns  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     5.650 ns              ┊ GC (median):    0.00%
 Time  (mean ± σ):   6.082 ns ±  1.905 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

  █             ▃                                            ▁
  █▅▄▃▄▃▄▄▅▃▄▃▃▃█▅▃▄▅▆▅▅▅▃▅▆▆▅▅▅▅▅▅▅▅▅▅▆▆▆▆▆▆▆▆▅▆▆▅▅▆▄▆▆▆▆█▇ █
  5.64 ns      Histogram: log(frequency) by time     11.3 ns <

 Memory estimate: 0 bytes, allocs estimate: 0.

Note that on v1.8.2 it gives the expected result:

@benchmark (@SVector rand(Float64,5))
BenchmarkTools.Trial: 10000 samples with 1000 evaluations.
 Range (min … max):  5.640 ns … 57.621 ns  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     5.650 ns              ┊ GC (median):    0.00%
 Time  (mean ± σ):   5.992 ns ±  1.498 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

  █             ▃                                            ▁
  █▃▃▂▄▃▄▃▂▃▃▄▄▄█▃▄▄▄▄▄▅▅▄▅▅▅▅▄▅▅▅▅▄▅▅▆▅▅▅▆▅▆▆▆▆▆▅▅▅▃▄▄▅▅▆▇▆ █
  5.64 ns      Histogram: log(frequency) by time     11.3 ns <

Memory estimate: 0 bytes, allocs estimate: 0.