[all] Build all release builds with -Os

[jfirebaugh]

Refs #8035.

[jfirebaugh]

Bloaty output on linux:

john@john-VirtualBox:~/mapbox-gl-native$ ../bloaty/bloaty -d compileunits build-os/linux-x86_64/Release/mbgl-glfw -- build-g/linux-x86_64/Release/mbgl-glfw 
     VM SIZE                                                                   FILE SIZE
 ++++++++++++++ GROWING                                                     ++++++++++++++
  +4.7%     +20 ../../../src/mbgl/style/layers/custom_layer.cpp                 +20  +4.7%
  +1.9%      +9 ../../../platform/default/webp_reader.cpp                        +9  +1.9%

 -------------- SHRINKING                                                   --------------
 -54.2%  -688Ki [Other]                                                      -673Ki -53.8%
 -40.4%  -122Ki ../../../src/mbgl/style/parser.cpp                           -122Ki -40.4%
 -48.2% -41.2Ki ../../../src/mbgl/style/layers/symbol_layer.cpp             -41.2Ki -48.2%
 -42.0% -41.1Ki ../../../src/clipper/clipper.cpp                            -41.1Ki -42.0%
 -68.8% -40.0Ki ../../../platform/glfw/glfw_view.cpp                        -40.0Ki -68.8%
 -77.5% -39.9Ki ../../../src/mbgl/tile/vector_tile.cpp                      -39.9Ki -77.5%
 -89.9% -38.9Ki ../../../src/mbgl/style/layers/raster_layer.cpp             -38.9Ki -89.9%
 -90.6% -38.9Ki ../../../src/mbgl/style/group_by_layout.cpp                 -38.9Ki -90.6%
 -74.8% -38.0Ki ../../../src/mbgl/sprite/sprite_parser.cpp                  -38.0Ki -74.8%
 -61.5% -33.6Ki ../../../src/mbgl/renderer/painter_background.cpp           -31.2Ki -59.7%
 -41.7% -31.8Ki ../../../src/mbgl/annotation/annotation_manager.cpp         -31.8Ki -41.7%
 -41.8% -30.7Ki ../../../src/mbgl/layout/symbol_layout.cpp                  -30.7Ki -41.8%
 -94.3% -30.6Ki ../../../src/mbgl/style/layers/raster_layer_impl.cpp        -30.6Ki -94.3%
 -70.0% -29.9Ki ../../../src/mbgl/tile/geometry_tile_worker.cpp             -29.9Ki -70.0%
 -92.0% -27.5Ki ../../../src/mbgl/annotation/fill_annotation_impl.cpp       -27.5Ki -92.0%
 -43.5% -26.3Ki ../../../src/mbgl/annotation/shape_annotation_impl.cpp      -26.3Ki -43.5%
 -73.6% -26.0Ki ../../../platform/default/mbgl/storage/offline_download.cpp -26.0Ki -73.6%
 -80.3% -26.0Ki ../../../src/mbgl/style/layers/background_layer.cpp         -26.0Ki -80.3%
 -48.9% -25.7Ki ../../../src/mbgl/style/layers/circle_layer.cpp             -25.7Ki -49.3%
 -55.4% -25.2Ki ../../../src/mbgl/renderer/circle_bucket.cpp                -25.2Ki -55.4%
 -50.2% -24.4Ki ../../../src/mbgl/style/layers/line_layer.cpp               -24.4Ki -50.2%

 -+-+-+-+-+-+-+ MIXED                                                       +-+-+-+-+-+-+-
  +5.5%  +202Ki [None]                                                      -26.0Mi -20.7%

 -19.4% -1.19Mi TOTAL                                                       -27.3Mi -21.4%

[kkaefer]

We've been using -Os before on Android too, and our binary size chart also noted a massive increase for it. Why do you think this would improve the situation on these platforms?

[jfirebaugh]

Right, Android was already using -Os, so it won't help there, but it will help linux, macOS, and iOS. This PR is just the first low-hanging fruit I noticed for #8035.

[friedbunny]

Assuming the command line builds for iOS and macOS respect the GCC_OPTIMIZATION_LEVEL setting in their respective Xcode projects, they’re both also already set to -Os.

[kkaefer]

Assuming the command line builds for iOS and macOS respect the GCC_OPTIMIZATION_LEVEL setting in their respective Xcode projects, they’re both also already set to -Os.

They do; we're using xcodebuild, which uses the Xcode project and its settings.

[jfirebaugh]

Assuming the command line builds for iOS and macOS respect the GCC_OPTIMIZATION_LEVEL setting in their respective Xcode projects, they’re both also already set to -Os.

They might use -Os for the code that's included in the checked-in Xcode projects, but they currently use -O3 for mbgl-core, since that's built from a cmake-generated project. (I verified by removing | xcpretty.)

Here's a before/after bloaty comparison for macOS:

     VM SIZE                               FILE SIZE
 ++++++++++++++ GROWING                 ++++++++++++++
   +16% +9.04Ki __TEXT,__unwind_info    +9.04Ki   +16%
  +0.1%    +128 __TEXT,__const             +128  +0.1%
  +0.1%    +125 __TEXT,__cstring           +125  +0.1%
  +0.6%     +16 __DATA,__bss                  0  [ = ]

 -------------- SHRINKING               --------------
 -19.2%  -614Ki __TEXT,__text            -614Ki -19.2%
  -4.9% -14.9Ki __TEXT,__gcc_except_tab -14.9Ki  -4.9%
  -4.4% -7.56Ki [None]                  -5.32Ki  -3.0%
  -1.5%     -32 __DATA,__got                -32  -1.5%

 -14.7%  -628Ki TOTAL                    -625Ki -14.7%

And iOS:

     VM SIZE                               FILE SIZE
 ++++++++++++++ GROWING                 ++++++++++++++
 +10.0% +6.00Ki __TEXT,__unwind_info    +6.00Ki +10.0%
  +0.1%    +176 __TEXT,__const             +176  +0.1%
  +0.6%     +16 __DATA,__bss                  0  [ = ]

 -------------- SHRINKING               --------------
  -0.1%    -994 [None]                  -3.41Mi  -5.9%
 -19.7%  -655Ki __TEXT,__text            -655Ki -19.7%
  -5.9% -19.5Ki __TEXT,__gcc_except_tab -19.5Ki  -5.9%
  -0.7%     -32 __DATA,__got                -32  -0.7%
  -0.0%     -10 __TEXT,__cstring            -10  -0.0%

 -12.8%  -669Ki TOTAL                   -4.06Mi  -6.5%

So 10-20% size savings across the board (except for Android).

[springmeyer]

So 10-20% size savings across the board (except for Android).

Great improvement. Is there any performance cost to this?