Bug 1358149 use fdlibm for inexact math functions in ffvpx/libavcodec r=padenot,firefox-build-system-reviewers,glandium

This includes the rdft FFT implementation.

Differential Revision: https://phabricator.services.mozilla.com/D185781

js/src/build/moz.build

@@ -49,6 +49,7 @@ if CONFIG["JS_HAS_INTL_API"]:
     ]
 
 USE_LIBS += [
+    "fdlibm",
     "nspr",
     "zlib",
 ]

media/ffvpx/libavcodec/moz.build

@@ -15,6 +15,8 @@ if CONFIG['FFVPX_ASFLAGS']:
 if CONFIG['CPU_ARCH'] == 'aarch64':
     DIRS += ['aarch64']
 
+LOCAL_INCLUDES += ['/modules/fdlibm/inexact-math-override']
+
 SharedLibrary('mozavcodec')
 SOURCES += [
     'allcodecs.c',
@@ -144,7 +146,8 @@ SYMBOLS_FILE = 'avcodec.symbols'
 NoVisibilityFlags()
 
 USE_LIBS += [
-     'mozavutil'
+    'fdlibm',
+    'mozavutil',
 ]
 
 if CONFIG['OS_TARGET'] != 'WINNT':

modules/fdlibm/src/moz.build

@@ -8,7 +8,7 @@ EXPORTS += [
     'fdlibm.h',
 ]
 
-FINAL_LIBRARY = 'js'
+Library('fdlibm')
 
 if CONFIG['CC_TYPE'] in ('clang', 'gcc'):
     CXXFLAGS += [