Increase efficiency of accessing members of struct pointers

[galenlynch]

The current code to access a member of a struct pointer uses a package function, fieldposition, that has identical functionality to a function provided in base, fieldindex. While this latter function is not exported, it is documented and will hopefully remain a part of Julia's API. I imagine that the base function is more performant than the implementation in fieldposition, which compares each field name to the target field name in order to find the position of a field. As such, I have replaced calls to fieldposition with calls to Base.fieldindex, and deleted the function fieldposition.

Additionally, when setting a member of a struct pointer, the current implementation calculates the pointer position, wraps it in a julia array with unsafe_wrap, and then sets the first element of that array to the desired value. The same result is achieved, perhaps more simply, by using unsafe_store! and not using an intermediate array. I have therefore replaced the call to unsafe_wrap and subsequent code with unsafe_store!.

While I do not know if the existing code causes a measurable decrease in performance, using the base functions reduces the amount of code, and most likely makes it faster.

[codecov[bot]]

Codecov Report

Merging #250 into master will decrease coverage by 0.02%. The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #250      +/-   ##
==========================================
- Coverage   76.83%   76.80%   -0.03%     
==========================================
  Files          14       14              
  Lines         600      595       -5     
==========================================
- Hits          461      457       -4     
+ Misses        139      138       -1     

Impacted Files	Coverage Δ
src/util.jl	70.37% <100.00%> (-1.51%)	:arrow_down:

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 8849d0f...82ea173. Read the comment docs.

[IanButterworth]

Good catch. It would be good to benchmark this before merging, I want to be sure that we don't impede performance (I agree with your logic, but still worth checking). Would you mind constructing BenchmarTools.@btime tests for frame read/write and compare to master? thanks!

[galenlynch]

I only measured the write functionality, which is the majority of this PR (though reads are also affected), but the PR is about 10 times faster, and uses less memory:

on master:

julia> using VideoIO, BenchmarkTools

julia> using VideoIO: avcodec_find_encoder_by_name, avcodec_alloc_context3, AVCodecContext, av_setfield

julia> codec_name = "libx264";

julia> codec = avcodec_find_encoder_by_name(codec_name);

julia> apCodecContext = Ptr{AVCodecContext}[avcodec_alloc_context3(codec)];

julia> width = 1024;

julia> @benchmark av_setfield(apCodecContext[1],:width,width)
BenchmarkTools.Trial: 
  memory estimate:  1.77 KiB
  allocs estimate:  8
  --------------
  minimum time:     3.070 μs (0.00% GC)
  median time:      3.141 μs (0.00% GC)
  mean time:        3.364 μs (1.17% GC)
  maximum time:     146.271 μs (95.39% GC)
  --------------
  samples:          10000
  evals/sample:     8

for this PR:

...
julia> @benchmark av_setfield(apCodecContext[1],:width,width)
BenchmarkTools.Trial: 
  memory estimate:  96 bytes
  allocs estimate:  6
  --------------
  minimum time:     384.736 ns (0.00% GC)
  median time:      401.701 ns (0.00% GC)
  mean time:        414.897 ns (0.39% GC)
  maximum time:     4.677 μs (89.38% GC)
  --------------
  samples:          10000
  evals/sample:     201

[IanButterworth]

Ok great. Thanks