Add proper support for -msse and -msse2 build flags. Don't emit SSE2 code unless explicitly enabled, since it is not implemented in Firefox and does not validate.

emcc

@@ -415,6 +415,8 @@ try:
   proxy_to_worker = False
   default_object_extension = '.o'
   valid_abspaths = []
+  sse1 = False
+  sse2 = False
 
   def is_valid_abspath(path_name):
     # Any path that is underneath the emscripten repository root must be ok.
@@ -662,6 +664,16 @@ try:
       if not default_object_extension.startswith('.'):
         default_object_extension = '.' + default_object_extension
       newargs[i+1] = ''
+    elif newargs[i] == '-msse':
+      settings_changes.append('SSE1=1')
+      newargs.append('-D__SSE__=1')
+      newargs[i] = ''
+    elif newargs[i] == '-msse2':
+      settings_changes.append('SSE1=1')
+      settings_changes.append('SSE2=1')
+      newargs.append('-D__SSE__=1')
+      newargs.append('-D__SSE2__=1')
+      newargs[i] = ''
 
   if should_exit:
     sys.exit(0)

emscripten.py

@@ -450,8 +450,14 @@ function _emscripten_asm_const_%d(%s) {
         asm_setup += '\nvar debug_table_' + sig + ' = ' + json.dumps(debug_tables[sig]) + ';'
 
     maths = ['Math.' + func for func in ['floor', 'abs', 'sqrt', 'pow', 'cos', 'sin', 'tan', 'acos', 'asin', 'atan', 'atan2', 'exp', 'log', 'ceil', 'imul', 'min', 'clz32']]
-    simdfloattypes = ['Float32x4', 'Float64x2']
-    simdinttypes = ['Int8x16', 'Int16x8', 'Int32x4']
+    simdfloattypes = []
+    simdinttypes = []
+
+    if settings['SSE1'] or settings['SIMD']:
+      simdfloattypes += ['Float32x4']
+    if settings['SSE2'] or settings['SIMD']:
+      simdfloattypes += ['Float64x2']
+      simdinttypes = ['Int8x16', 'Int16x8', 'Int32x4']
     simdtypes = simdfloattypes + simdinttypes
     simdfuncs = ['check', 'add', 'sub', 'neg', 'mul',
                  'equal', 'lessThan', 'greaterThan',
@@ -479,7 +485,7 @@ function _emscripten_asm_const_%d(%s) {
     if settings['ALLOW_MEMORY_GROWTH']: fundamentals.append('byteLength')
     math_envs = []
 
-    provide_fround = settings['PRECISE_F32'] or settings['SIMD']
+    provide_fround = settings['PRECISE_F32'] or settings['SIMD'] or settings['SSE1']
 
     if provide_fround: maths += ['Math.fround']
 

src/settings.js

@@ -120,6 +120,9 @@ var SIMD = 0; // Whether to allow autovectorized SIMD code ( https://github.com/
               // (In older versions of emscripten, in particular pre-fastcomp, SIMD=1 was needed to get
               // any SIMD output at all.)
 
+var SSE1 = 0; // If true, enables generating code to target the SSE1 instruction set. Build with -msse to enable this.
+var SSE2 = 0; // If true, enables generating code to target the SSE2 instruction set. Build with -msse2 to enable this.
+
 var USE_CLOSURE_COMPILER = 0; // Whether closure compiling is being run on this output
 
 var SKIP_STACK_IN_SMALL = 1; // When enabled, does not push/pop the stack at all in

system/include/emscripten/emmintrin.h

@@ -43,14 +43,12 @@
 #define _mm_cvttsd_si64x _mm_cvttsd_si64
 #define _mm_store_pd1 _mm_store1_pd
 
-#else
+#endif
 
 #ifndef __SSE2__
 #error "SSE2 instruction set not enabled"
 #endif
 
-#endif
-
 #include <xmmintrin.h>
 
 typedef double __m128d __attribute__((__vector_size__(16)));

system/include/emscripten/xmmintrin.h

@@ -6,6 +6,10 @@
 #include <math.h>
 #include <string.h>
 
+#ifndef __SSE__
+#error "SSE instruction set not enabled"
+#endif
+
 // Emscripten SIMD support doesn't support MMX/float32x2/__m64.
 // However, we support loading and storing 2-vectors, so
 // treat "__m64 *" as "void *" for that purpose.