MSVC opt /fp:fast /Oi, intrinsic sqrt() w/ XBRZ.

For MSVC, add the optimization flags:

/fp:fast /Oi

, to favor speed for floating point operations and use intrinsic
operations where possible.

For release builds also add:

/O2

. Use sqrt() instead of std::sqrt() in XBRZ for MSVC x64 just in case to
make sure it uses the intrinsic version.

Test build is reported to run much faster.

Signed-off-by: Rafael Kitover <rkitover@gmail.com>

CMakeLists.txt

@@ -744,12 +744,12 @@ elseif(MSVC)
     string(REGEX REPLACE "/[Ww][^ ]+" "" CMAKE_C_FLAGS   ${CMAKE_C_FLAGS})
     string(REGEX REPLACE "/[Ww][^ ]+" "" CMAKE_CXX_FLAGS ${CMAKE_CXX_FLAGS})
 
-    add_compile_options(/std:c++17 -D__STDC_LIMIT_MACROS)
+    add_compile_options(/std:c++17 -D__STDC_LIMIT_MACROS /fp:fast /Oi)
 
     if(CMAKE_BUILD_TYPE STREQUAL Debug)
         add_compile_options(/W4)
     else()
-        add_compile_options(/w)
+        add_compile_options(/w /O2)
 
         if(ENABLE_LTO)
             add_compile_options(/GL)

src/filters/xBRZ/xbrz.cpp

@@ -76,8 +76,10 @@ inline double fastSqrt(double n)
         fld n
         fsqrt
     }
-#else // defined(_MSC_VER) && defined(_M_X64) OR other platforms
-    // VisualStudio x86_64 does not allow inline ASM
+#elif defined(_MSC_VER)
+    // On MSVC x64 use intrinsic with /Oi and /fp:fast
+    return sqrt(n);
+#else // Other platforms.
     return std::sqrt(n);
 #endif
 }