tree 8964d6923de11f3eba9b2a99382c3fdfe66cbdbd
parent 844fc9b5eeef46edab49c04516010db283e420da
author Nigel Tao <nigeltao@golang.org> 1590711479 +1000
committer Nigel Tao <nigeltao@golang.org> 1590715292 +1000

Performance mystery: delete unused string constant

This commit deletes the definition of the `const char*
wuffs_base__note__i_o_redirect` global variable. This variable is not
used anywhere (after the previous commit removed references to it).

Deleting this one line of code (two if you count the declaration, not
just the definition) can have a dramatic effect on seemingly unrelated
performance micro-benchmarks. Some numbers get better (e.g. +5%), some
numbers get worse (e.g. -10%). The same micro-benchmark can get faster
on one C compiler but slower on another.

It is a mystery why this happens. This small, self-contained commit (as
well as the previous one, which set it up) will be immediately followed
by a roll back, but it is committed anyway so that we can refer to these
numbers in the git log. To reproduce:

----
$ git clone https://github.com/google/wuffs.git
$ cd wuffs
$ gcc --version | head -n 1
gcc (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0

$ # Check out the parent commit.
$ git checkout --quiet THE_HASH_OF_THIS_COMMIT^
$ gcc -O3 -Wall -std=c99 test/c/std/json.c -DWUFFS_MIMIC
$ ./a.out -bench -focus=wuffs_json_decode_1k | grep MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1125 ns/op	     756.762 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1131 ns/op	     752.652 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1131 ns/op	     752.785 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1131 ns/op	     752.697 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1132 ns/op	     752.365 MB/s

$ # Check out this commit.
$ git checkout --quiet THE_HASH_OF_THIS_COMMIT
$ gcc -O3 -Wall -std=c99 test/c/std/json.c -DWUFFS_MIMIC
$ ./a.out -bench -focus=wuffs_json_decode_1k | grep MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1200 ns/op	     709.617 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1203 ns/op	     707.651 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1203 ns/op	     707.976 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1233 ns/op	     690.663 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1203 ns/op	     708.153 MB/s

$ # Dropping the "-std=c99" also changes the numbers.

$ # Check out the parent commit.
$ git checkout --quiet THE_HASH_OF_THIS_COMMIT^
$ gcc -O3 -Wall          test/c/std/json.c -DWUFFS_MIMIC
$ ./a.out -bench -focus=wuffs_json_decode_1k | grep MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1229 ns/op	     693.226 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1230 ns/op	     692.254 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1230 ns/op	     692.486 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1234 ns/op	     689.986 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1232 ns/op	     691.480 MB/s

$ # Check out this commit.
$ git checkout --quiet THE_HASH_OF_THIS_COMMIT
$ gcc -O3 -Wall          test/c/std/json.c -DWUFFS_MIMIC
$ ./a.out -bench -focus=wuffs_json_decode_1k | grep MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1129 ns/op	     754.103 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1131 ns/op	     752.849 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1135 ns/op	     750.622 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1150 ns/op	     740.731 MB/s
Benchmarkwuffs_json_decode_1k/gcc7	 1000000	    1135 ns/op	     750.609 MB/s

$ ldd ./a.out
	linux-vdso.so.1 (0x00007ffd9b1f2000)
	libgtk3-nocsd.so.0 => /usr/lib/x86_64-linux-gnu/libgtk3-nocsd.so.0 (0x00007f9b46f60000)
	libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f9b46b6f000)
	libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007f9b4696b000)
	libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f9b4674c000)
	/lib64/ld-linux-x86-64.so.2 (0x00007f9b5b59a000)
$ rm a.out
----

Some more benchmark numbers before/after this commit:

----
name                                                old speed      new speed      delta

wuffs_deflate_decode_1k_full_init/clang5             103MB/s ± 1%   105MB/s ± 1%   +2.25%  (p=0.000 n=9+10)
wuffs_deflate_decode_1k_part_init/clang5             121MB/s ± 0%   122MB/s ± 0%     ~     (p=1.000 n=9+9)
wuffs_deflate_decode_10k_full_init/clang5            148MB/s ± 1%   149MB/s ± 0%     ~     (p=0.059 n=9+8)
wuffs_deflate_decode_10k_part_init/clang5            151MB/s ± 0%   151MB/s ± 0%     ~     (p=0.968 n=9+10)
wuffs_deflate_decode_100k_just_one_read/clang5       172MB/s ± 0%   171MB/s ± 1%   -0.44%  (p=0.001 n=9+9)
wuffs_deflate_decode_100k_many_big_reads/clang5      149MB/s ± 0%   149MB/s ± 1%   -0.56%  (p=0.000 n=10+9)

wuffs_deflate_decode_1k_full_init/gcc7               104MB/s ± 1%   102MB/s ± 1%   -2.45%  (p=0.000 n=9+10)
wuffs_deflate_decode_1k_part_init/gcc7               121MB/s ± 0%   121MB/s ± 0%   +0.46%  (p=0.000 n=9+10)
wuffs_deflate_decode_10k_full_init/gcc7              147MB/s ± 0%   153MB/s ± 0%   +3.75%  (p=0.000 n=8+10)
wuffs_deflate_decode_10k_part_init/gcc7              150MB/s ± 1%   155MB/s ± 0%   +3.98%  (p=0.000 n=9+10)
wuffs_deflate_decode_100k_just_one_read/gcc7         184MB/s ± 0%   194MB/s ± 1%   +5.43%  (p=0.000 n=9+10)
wuffs_deflate_decode_100k_many_big_reads/gcc7        154MB/s ± 0%   158MB/s ± 0%   +2.45%  (p=0.000 n=9+10)

wuffs_gif_decode_1k_bw/clang5                        281MB/s ± 1%   278MB/s ± 1%   -0.92%  (p=0.003 n=10+9)
wuffs_gif_decode_1k_color_full_init/clang5          97.8MB/s ± 1%  93.7MB/s ± 3%   -4.26%  (p=0.000 n=10+10)
wuffs_gif_decode_1k_color_part_init/clang5           117MB/s ± 1%   115MB/s ± 1%   -1.41%  (p=0.000 n=9+10)
wuffs_gif_decode_10k_bgra/clang5                     463MB/s ± 1%   455MB/s ± 1%   -1.68%  (p=0.000 n=10+10)
wuffs_gif_decode_10k_indexed/clang5                  124MB/s ± 1%   121MB/s ± 1%   -2.31%  (p=0.000 n=9+9)
wuffs_gif_decode_20k/clang5                          149MB/s ± 1%   146MB/s ± 1%   -1.95%  (p=0.000 n=10+10)
wuffs_gif_decode_100k_artificial/clang5              313MB/s ± 1%   307MB/s ± 0%   -1.94%  (p=0.000 n=9+7)
wuffs_gif_decode_100k_realistic/clang5               135MB/s ± 1%   133MB/s ± 1%   -1.18%  (p=0.000 n=10+8)
wuffs_gif_decode_1000k_full_init/clang5              136MB/s ± 1%   132MB/s ± 4%   -2.27%  (p=0.023 n=10+10)
wuffs_gif_decode_1000k_part_init/clang5              136MB/s ± 1%   132MB/s ± 3%   -2.96%  (p=0.000 n=9+10)
wuffs_gif_decode_anim_screencap/clang5               644MB/s ± 1%   621MB/s ± 3%   -3.51%  (p=0.000 n=10+10)

wuffs_gif_decode_1k_bw/gcc7                          290MB/s ± 1%   287MB/s ± 1%   -1.18%  (p=0.000 n=10+10)
wuffs_gif_decode_1k_color_full_init/gcc7            91.7MB/s ± 1%  88.2MB/s ± 1%   -3.85%  (p=0.000 n=10+10)
wuffs_gif_decode_1k_color_part_init/gcc7             107MB/s ± 1%   103MB/s ± 1%   -4.12%  (p=0.000 n=9+10)
wuffs_gif_decode_10k_bgra/gcc7                       370MB/s ± 1%   344MB/s ± 1%   -6.98%  (p=0.000 n=10+9)
wuffs_gif_decode_10k_indexed/gcc7                    113MB/s ± 2%   104MB/s ± 1%   -8.31%  (p=0.000 n=10+8)
wuffs_gif_decode_20k/gcc7                            139MB/s ± 1%   127MB/s ± 3%   -8.49%  (p=0.000 n=10+10)
wuffs_gif_decode_100k_artificial/gcc7                299MB/s ± 0%   280MB/s ± 1%   -6.43%  (p=0.000 n=9+8)
wuffs_gif_decode_100k_realistic/gcc7                 123MB/s ± 1%   111MB/s ± 1%  -10.21%  (p=0.000 n=10+8)
wuffs_gif_decode_1000k_full_init/gcc7                124MB/s ± 1%   112MB/s ± 1%   -9.34%  (p=0.000 n=10+8)
wuffs_gif_decode_1000k_part_init/gcc7                125MB/s ± 1%   112MB/s ± 1%  -10.19%  (p=0.000 n=10+9)
wuffs_gif_decode_anim_screencap/gcc7                 622MB/s ± 1%   599MB/s ± 3%   -3.68%  (p=0.000 n=9+10)

wuffs_json_decode_1k/clang5                          628MB/s ± 1%   628MB/s ± 1%     ~     (p=0.842 n=9+10)
wuffs_json_decode_21k_formatted/clang5               331MB/s ± 1%   331MB/s ± 0%     ~     (p=0.720 n=10+9)
wuffs_json_decode_26k_compact/clang5                 521MB/s ± 1%   521MB/s ± 0%     ~     (p=0.549 n=10+9)
wuffs_json_decode_217k_stringy/clang5                432MB/s ± 1%   432MB/s ± 1%     ~     (p=0.780 n=10+9)

wuffs_json_decode_1k/gcc7                            745MB/s ± 0%   703MB/s ± 0%   -5.70%  (p=0.000 n=9+10)
wuffs_json_decode_21k_formatted/gcc7                 418MB/s ± 1%   400MB/s ± 1%   -4.44%  (p=0.000 n=10+8)
wuffs_json_decode_26k_compact/gcc7                   554MB/s ± 1%   511MB/s ± 1%   -7.72%  (p=0.000 n=9+10)
wuffs_json_decode_217k_stringy/gcc7                  452MB/s ± 1%   422MB/s ± 1%   -6.58%  (p=0.000 n=9+10)
----