We need the compiled shared to be fast as possible.

Glandium, any suggestions for how to best accomplish this?

One option is adding something like:

diff --git a/gfx/wr/swgl/build.rs b/gfx/wr/swgl/build.rs
index 5a2ed57f55c91..7c2b454a4147c 100644
--- a/gfx/wr/swgl/build.rs
+++ b/gfx/wr/swgl/build.rs
@@ -134,11 +134,14 @@ fn main() {
     println!("cargo:rerun-if-changed=src/swgl_ext.h");
     println!("cargo:rerun-if-changed=src/texture.h");
     println!("cargo:rerun-if-changed=src/vector_type.h");
     println!("cargo:rerun-if-changed=src/gl.cc");
-    cc::Build::new()
-        .cpp(true)
-        .file("src/gl.cc")
+    let mut build = cc::Build::new();
+    build.cpp(true);
+    if build.get_compiler().args().contains(&std::ffi::OsString::from("-Oz")) {
+        build.flag("-O2");
+    }
+    build.file("src/gl.cc")
         .flag("-std=c++17")
         .flag("-UMOZILLA_CONFIG_H")
         .flag("-fno-exceptions")
         .flag("-fno-rtti")

https://hg.mozilla.org/mozilla-central/rev/f87afcc0d272