SCons: Properly set SSE2 as baseline on x86_32

Setting it only for release templates on Windows and macOS was inconsistent,
and Jolt requires it as a minimum.

Drop the `-mxsave` flag from the raycast module, this doesn't seem to be
used explicitly by Embree, and unnecessarily makes our config and baseline
muddy.

SConstruct

@@ -713,6 +713,12 @@ elif env.msvc:
         )
         Exit(255)
 
+# Default architecture flags.
+if env["arch"] == "x86_32":
+    if env.msvc:
+        env.Append(CCFLAGS=["/arch:SSE2"])
+    else:
+        env.Append(CCFLAGS=["-msse2"])
 
 # Set optimize and debug_symbols flags.
 # "custom" means do nothing and let users set their own optimization flags.

modules/raycast/SCsub

@@ -66,18 +66,12 @@ if env["builtin_embree"]:
     env_raycast.Append(CPPDEFINES=["EMBREE_TARGET_SSE2", "EMBREE_LOWEST_ISA", "TASKING_INTERNAL"])
     env_raycast.AppendUnique(CPPDEFINES=["NDEBUG"])  # No assert() even in debug builds.
 
-    if not env.msvc:
-        if env["arch"] in ["x86_64", "x86_32"]:
-            env_raycast.Append(CCFLAGS=["-msse2", "-mxsave"])
-
-        if env["platform"] == "windows":
-            env_raycast.Append(CCFLAGS=["-mstackrealign"])
-
     if env["platform"] == "windows":
         if env.msvc:
             env.Append(LINKFLAGS=["psapi.lib"])
         else:
             env.Append(LIBS=["psapi"])
+            env_raycast.Append(CCFLAGS=["-mstackrealign"])
 
     if env.msvc:  # Disable bogus warning about intentional struct padding.
         env_raycast.Append(CCFLAGS=["/wd4324"])

platform/macos/detect.py

@@ -70,14 +70,6 @@ def configure(env: "SConsEnvironment"):
     supported_arches = ["x86_64", "arm64"]
     validate_arch(env["arch"], get_name(), supported_arches)
 
-    ## Build type
-
-    if env["target"] == "template_release":
-        if env["arch"] != "arm64":
-            env.Prepend(CCFLAGS=["-msse2"])
-    elif env.dev_build:
-        env.Prepend(LINKFLAGS=["-Xlinker", "-no_deduplicate"])
-
     ## Compiler configuration
 
     # Save this in environment for use by other modules
@@ -97,6 +89,7 @@ def configure(env: "SConsEnvironment"):
         env.Append(LINKFLAGS=["-arch", "x86_64", "-mmacosx-version-min=10.13"])
 
     env.Append(CCFLAGS=["-ffp-contract=off"])
+    env.Append(CCFLAGS=["-fobjc-arc"])
 
     cc_version = get_compiler_version(env)
     cc_version_major = cc_version["apple_major"]
@@ -106,7 +99,8 @@ def configure(env: "SConsEnvironment"):
     if is_apple_clang(env) and cc_version_major == 1500 and cc_version_minor == 0:
         env.Prepend(LINKFLAGS=["-ld_classic"])
 
-    env.Append(CCFLAGS=["-fobjc-arc"])
+    if env.dev_build:
+        env.Prepend(LINKFLAGS=["-Xlinker", "-no_deduplicate"])
 
     if "osxcross" not in env:  # regular native build
         if env["macports_clang"] != "no":

platform/windows/detect.py

@@ -701,11 +701,7 @@ def configure_mingw(env: "SConsEnvironment"):
         print("Detected GCC to be a wrapper for Clang.")
         env["use_llvm"] = True
 
-    # TODO: Re-evaluate the need for this / streamline with common config.
-    if env["target"] == "template_release":
-        if env["arch"] != "arm64":
-            env.Append(CCFLAGS=["-msse2"])
-    elif env.dev_build:
+    if env.dev_build:
         # Allow big objects. It's supposed not to have drawbacks but seems to break
         # GCC LTO, so enabling for debug builds only (which are not built with LTO
         # and are the only ones with too big objects).