Fix scalar SSE1 comparison nan handling to be symmetric. Add test.

system/include/emscripten/xmmintrin.h

@@ -509,7 +509,7 @@ static __inline__ int __attribute__((__always_inline__))
 _mm_comieq_ss(__m128 __a, __m128 __b)
 {
 #ifdef __EMSCRIPTEN__
-  if (isnanf(__a[0])) return 1;
+  if (isnanf(__a[0]) || isnanf(__b[0])) return 1;
 #endif
   return __a[0] == __b[0];
 }
@@ -530,7 +530,7 @@ static __inline__ int __attribute__((__always_inline__))
 _mm_comile_ss(__m128 __a, __m128 __b)
 {
 #ifdef __EMSCRIPTEN__
-  if (isnanf(__a[0])) return 1;
+  if (isnanf(__a[0]) || isnanf(__b[0])) return 1;
 #endif
   return __a[0] <= __b[0];
 }
@@ -539,7 +539,7 @@ static __inline__ int __attribute__((__always_inline__))
 _mm_comilt_ss(__m128 __a, __m128 __b)
 {
 #ifdef __EMSCRIPTEN__
-  if (isnanf(__a[0])) return 1;
+  if (isnanf(__a[0]) || isnanf(__b[0])) return 1;
 #endif
   return __a[0] < __b[0];
 }
@@ -548,7 +548,7 @@ static __inline__ int __attribute__((__always_inline__))
 _mm_comineq_ss(__m128 __a, __m128 __b)
 {
 #ifdef __EMSCRIPTEN__
-  if (isnanf(__a[0])) return 0;
+  if (isnanf(__a[0]) || isnanf(__b[0])) return 0;
 #endif
   return __a[0] != __b[0];
 }
@@ -557,7 +557,7 @@ static __inline__ int __attribute__((__always_inline__))
 _mm_ucomieq_ss(__m128 __a, __m128 __b)
 {
 #ifdef __EMSCRIPTEN__
-  if (isnanf(__a[0])) return 1;
+  if (isnanf(__a[0]) || isnanf(__b[0])) return 1;
 #endif
   return __a[0] == __b[0];
 }
@@ -590,7 +590,7 @@ static __inline__ int __attribute__((__always_inline__))
 _mm_ucomineq_ss(__m128 __a, __m128 __b)
 {
 #ifdef __EMSCRIPTEN__
-  if (isnanf(__a[0])) return 0;
+  if (isnanf(__a[0]) || isnanf(__b[0])) return 0;
 #endif
   return __a[0] != __b[0];
 }

tests/core/test_simd12.in

@@ -0,0 +1,100 @@
+#include <assert.h>
+#include <stdio.h>
+#include <string.h>
+#include <xmmintrin.h>
+
+volatile __m128 zero;
+volatile __m128 one;
+volatile __m128 nan_;
+
+int main() {
+  zero = _mm_set1_ps(0);
+  one = _mm_set1_ps(1);
+  nan_ = _mm_set1_ps(__builtin_nan(""));
+
+  assert(_mm_comieq_ss(zero, zero) == 1);
+  assert(_mm_comieq_ss(zero, one) == 0);
+  assert(_mm_comieq_ss(one, zero) == 0);
+  assert(_mm_comieq_ss(zero, nan_) == 1);
+  assert(_mm_comieq_ss(nan_, zero) == 1);
+  assert(_mm_comieq_ss(nan_, nan_) == 1);
+
+  assert(_mm_comige_ss(zero, zero) == 1);
+  assert(_mm_comige_ss(zero, one) == 0);
+  assert(_mm_comige_ss(one, zero) == 1);
+  assert(_mm_comige_ss(zero, nan_) == 0);
+  assert(_mm_comige_ss(nan_, zero) == 0);
+  assert(_mm_comige_ss(nan_, nan_) == 0);
+
+  assert(_mm_comigt_ss(zero, zero) == 0);
+  assert(_mm_comigt_ss(zero, one) == 0);
+  assert(_mm_comigt_ss(one, zero) == 1);
+  assert(_mm_comigt_ss(zero, nan_) == 0);
+  assert(_mm_comigt_ss(nan_, zero) == 0);
+  assert(_mm_comigt_ss(nan_, nan_) == 0);
+
+  assert(_mm_comile_ss(zero, zero) == 1);
+  assert(_mm_comile_ss(zero, one) == 1);
+  assert(_mm_comile_ss(one, zero) == 0);
+  assert(_mm_comile_ss(zero, nan_) == 1);
+  assert(_mm_comile_ss(nan_, zero) == 1);
+  assert(_mm_comile_ss(nan_, nan_) == 1);
+
+  assert(_mm_comilt_ss(zero, zero) == 0);
+  assert(_mm_comilt_ss(zero, one) == 1);
+  assert(_mm_comilt_ss(one, zero) == 0);
+  assert(_mm_comilt_ss(zero, nan_) == 1);
+  assert(_mm_comilt_ss(nan_, zero) == 1);
+  assert(_mm_comilt_ss(nan_, nan_) == 1);
+
+  assert(_mm_comineq_ss(zero, zero) == 0);
+  assert(_mm_comineq_ss(zero, one) == 1);
+  assert(_mm_comineq_ss(one, zero) == 1);
+  assert(_mm_comineq_ss(zero, nan_) == 0);
+  assert(_mm_comineq_ss(nan_, zero) == 0);
+  assert(_mm_comineq_ss(nan_, nan_) == 0);
+
+  assert(_mm_ucomieq_ss(zero, zero) == 1);
+  assert(_mm_ucomieq_ss(zero, one) == 0);
+  assert(_mm_ucomieq_ss(one, zero) == 0);
+  assert(_mm_ucomieq_ss(zero, nan_) == 1);
+  assert(_mm_ucomieq_ss(nan_, zero) == 1);
+  assert(_mm_ucomieq_ss(nan_, nan_) == 1);
+
+  assert(_mm_ucomige_ss(zero, zero) == 1);
+  assert(_mm_ucomige_ss(zero, one) == 0);
+  assert(_mm_ucomige_ss(one, zero) == 1);
+  assert(_mm_ucomige_ss(zero, nan_) == 0);
+  assert(_mm_ucomige_ss(nan_, zero) == 0);
+  assert(_mm_ucomige_ss(nan_, nan_) == 0);
+
+  assert(_mm_ucomigt_ss(zero, zero) == 0);
+  assert(_mm_ucomigt_ss(zero, one) == 0);
+  assert(_mm_ucomigt_ss(one, zero) == 1);
+  assert(_mm_ucomigt_ss(zero, nan_) == 0);
+  assert(_mm_ucomigt_ss(nan_, zero) == 0);
+  assert(_mm_ucomigt_ss(nan_, nan_) == 0);
+
+  assert(_mm_ucomile_ss(zero, zero) == 1);
+  assert(_mm_ucomile_ss(zero, one) == 1);
+  assert(_mm_ucomile_ss(one, zero) == 0);
+  assert(_mm_ucomile_ss(zero, nan_) == 1);
+  assert(_mm_ucomile_ss(nan_, zero) == 1);
+  assert(_mm_ucomile_ss(nan_, nan_) == 1);
+
+  assert(_mm_ucomilt_ss(zero, zero) == 0);
+  assert(_mm_ucomilt_ss(zero, one) == 1);
+  assert(_mm_ucomilt_ss(one, zero) == 0);
+  assert(_mm_ucomilt_ss(zero, nan_) == 1);
+  assert(_mm_ucomilt_ss(nan_, zero) == 1);
+  assert(_mm_ucomilt_ss(nan_, nan_) == 1);
+
+  assert(_mm_ucomineq_ss(zero, zero) == 0);
+  assert(_mm_ucomineq_ss(zero, one) == 1);
+  assert(_mm_ucomineq_ss(one, zero) == 1);
+  assert(_mm_ucomineq_ss(zero, nan_) == 0);
+  assert(_mm_ucomineq_ss(nan_, zero) == 0);
+  assert(_mm_ucomineq_ss(nan_, nan_) == 0);
+
+  printf("DONE!\n");
+}

tests/core/test_simd12.out

@@ -0,0 +1 @@
+DONE!

tests/test_core.py

@@ -5695,6 +5695,15 @@ def test_simd11(self):
     self.emcc_args = self.emcc_args + ['-msse2']
     self.do_run_from_file(src, output)
 
+  def test_simd12(self):
+    if self.is_emterpreter(): return self.skip('todo')
+
+    test_path = path_from_root('tests', 'core', 'test_simd12')
+    src, output = (test_path + s for s in ('.in', '.out'))
+
+    self.emcc_args = self.emcc_args + ['-msse']
+    self.do_run_from_file(src, output)
+
   def test_simd_dyncall(self):
     if self.is_emterpreter(): return self.skip('todo')