mirror of
https://github.com/ddidderr/wuffs.git
synced 2026-11-17 16:32:11 +01:00
The algorithm is discussed at https://lemire.me/blog/2020/03/10/fast-float-parsing-in-practice/ ---- Binary size, before: 27848 gen/lib/c/clang-5.0-dynamic/wuffs-base-f64conv.lo 28040 gen/lib/c/clang-5.0-static/wuffs-base-f64conv.o 30392 gen/lib/c/gcc-dynamic/wuffs-base-f64conv.lo 30392 gen/lib/c/gcc-static/wuffs-base-f64conv.o After: 29416 gen/lib/c/clang-5.0-dynamic/wuffs-base-f64conv.lo 29688 gen/lib/c/clang-5.0-static/wuffs-base-f64conv.o 31792 gen/lib/c/gcc-dynamic/wuffs-base-f64conv.lo 31792 gen/lib/c/gcc-static/wuffs-base-f64conv.o The binary size difference does not account for the powers-of-10 table, which was expanded in a previous commit in anticipation of this one. A future commit will also remove some of the other f64 parser code used before implementing the Eisel algorithm. ---- Processing citylots.json from github.com/zemirco/sf-city-lots-json $ g++ -O3 script/process-json-numbers.c $ time ./a.out -parse-number-f64 < citylots.json Before/After: real 0m1.231s real 0m1.031s The difference with and without -parse-number-f64 measures how much time is spent in the parse_number_f64 function. Before: 424ms, after: 224ms. $ time ./a.out < citylots.json real 0m0.807s Similarly, for canada.json from github.com/miloyip/nativejson-benchmark $ time ./a.out -parse-number-f64 < canada.json Before/After: real 0m0.023s real 0m0.017s $ time ./a.out < canada.json real 0m0.013s Time spent in parse_number_f64, before: 10ms, after: 4ms. ---- name old time/op new time/op delta wuffs_strconv_parse_number_f64_1_lsh53_add0/clang5 77.0ns ± 0% 37.0ns ± 0% -51.95% (p=0.029 n=4+4) wuffs_strconv_parse_number_f64_1_lsh53_add1/clang5 690ns ± 0% 725ns ± 2% +4.95% (p=0.008 n=5+5) wuffs_strconv_parse_number_f64_pi_long/clang5 92.0ns ± 0% 136.0ns ± 2% +47.83% (p=0.016 n=4+5) wuffs_strconv_parse_number_f64_pi_short/clang5 35.0ns ± 3% 18.0ns ± 0% -48.57% (p=0.008 n=5+5) wuffs_strconv_parse_number_f64_1_lsh53_add0/gcc7 67.0ns ± 0% 34.0ns ± 0% -49.25% (p=0.000 n=5+4) wuffs_strconv_parse_number_f64_1_lsh53_add1/gcc7 712ns ± 0% 760ns ± 2% +6.72% (p=0.008 n=5+5) wuffs_strconv_parse_number_f64_pi_long/gcc7 88.4ns ± 1% 116.2ns ± 3% +31.45% (p=0.008 n=5+5) wuffs_strconv_parse_number_f64_pi_short/gcc7 32.6ns ± 2% 21.4ns ± 7% -34.36% (p=0.008 n=5+5)