Performance compared to wasm32

[sbc100]

trafficstars

I got some initial performance numbers by running the emscripten benchmark suite under v8:

                        v8 v8-wasm64
                  base64 1.000	 1.234   size: 1.065
            conditionals 1.000	 1.003   size: 1.112 
                    copy 1.000	 0.997   size: 1.110
             corrections 1.000	 0.985   size: 1.109
           corrections64 1.000	 1.003   size: 1.110
                fannkuch 1.000	 1.161   size: 1.075
             fasta_float 1.000	 1.149   size: 1.075
                  havlak 1.000	 1.370   size: 1.110
                     ifs 1.000	 1.018   size: 1.110
         matrix_multiply 1.000	 1.162   size: 1.087
                  memops 1.000	 1.163   size: 1.088
                  primes 1.000	 1.016   size: 1.112
          primes_nocheck 1.000	 1.000   size: 2.020
                skinning 1.000	 1.040   size: 1.067
               zzz_box2d 1.000	 1.182   size: 1.043
                    mean         1.098         1.152  (-primes_nocheck: 1.090)

Looks like about 10% performance hit (presumably a combination of extra bounds checks and larger data structures), and about 10% code size increase. The size increase probably warrants more investigation since I wouldn't expect the larger pointer sizes to show up very much in the binary (just the static data structures of which I doubt there are that many).

[sbc100]

benchmarks.txt

[sbc100]

I think the size regressions are because I was measuring both JS and wasm sizes. I will re-measure with just wasm sizes.

[tbuchs]

Do you know if this can still be considered as a good reference value?