Merge pull request godotengine#59595 from Calinou/scons-use-sse4.2

Use SSE 4.2 as a baseline when compiling Godot

Author: akien-mga

SConstruct

@@ -731,11 +731,25 @@ elif env.msvc:
         )
         Exit(255)
 
-# Default architecture flags.
-if env["arch"] == "x86_32":
-    if env.msvc:
+# Set x86 CPU instruction sets to use by the compiler's autovectorization.
+if env["arch"] == "x86_64":
+    # On 64-bit x86, enable SSE 4.2 and prior instruction sets (SSE3/SSSE3/SSE4/SSE4.1) to improve performance.
+    # This is supported on most CPUs released after 2009-2011 (Intel Nehalem, AMD Bulldozer).
+    # AVX and AVX2 aren't enabled because they aren't available on more recent low-end Intel CPUs.
+    if env.msvc and not methods.using_clang(env):
+        # https://stackoverflow.com/questions/64053597/how-do-i-enable-sse4-1-and-sse3-but-not-avx-in-msvc/69328426
+        env.Append(CCFLAGS=["/d2archSSE42"])
+    else:
+        # `-msse2` is implied when compiling for x86_64.
+        env.Append(CCFLAGS=["-msse4.2"])
+elif env["arch"] == "x86_32":
+    # Be more conservative with instruction sets on 32-bit x86 to improve compatibility.
+    # SSE and SSE2 are present on all CPUs that support 64-bit, even if running a 32-bit OS.
+    if env.msvc and not methods.using_clang(env):
         env.Append(CCFLAGS=["/arch:SSE2"])
     else:
+        # Use `-mfpmath=sse` to use SSE for floating-point math, which is more stable than x87.
+        # `-mstackrealign` is needed for it to work.
         env.Append(CCFLAGS=["-msse2", "-mfpmath=sse", "-mstackrealign"])
 
 # Explicitly specify colored output.

modules/raycast/SCsub

@@ -80,9 +80,16 @@ if env["builtin_embree"]:
     env_thirdparty.disable_warnings()
     env_thirdparty.add_source_files(thirdparty_obj, thirdparty_sources)
 
+    # Set x86 CPU instruction sets to use when building Embree's own intrinsics.
+    # Keep this in sync with Godot's main SConstruct file.
+    # This is only needed on MSVC, as GCC/Clang will set those defines automatically
+    # according to compiler instruction set flags.
     if env["arch"] != "x86_64" or env.msvc:
-        # Embree needs those, it will automatically use SSE2NEON in ARM
-        env_thirdparty.Append(CPPDEFINES=["__SSE2__", "__SSE__"])
+        # Embree needs those; it will automatically use SSE2NEON in ARM.
+        env_thirdparty.Append(CPPDEFINES=["__SSE__", "__SSE2__"])
+
+    if env["arch"] == "x86_64" and env.msvc:
+        env_thirdparty.Append(CPPDEFINES=["__SSE3__", "__SSSE3__", "__SSE4_1__", "__SSE4_2__"])
 
     if env["platform"] == "web":
         env_thirdparty.Append(CXXFLAGS=["-msimd128"])