getting some initial set of x86 intrinsics mapped to JS SIMD ops; update existing SIMD code to use updated JS SIMD API (SIMD.type.op instead of SIMD.op & SIMD.opu32; int32x4 instead of uint32x4); bug fix: right type of withX() for int32x4 initialization; bug fix: splat(0) instead of zero() for int32x4 initialization

2 files changed, 218 insertions, 0 deletions

diff --git a/system/include/emscripten/emmintrin.h b/system/include/emscripten/emmintrin.h
new file mode 100644
index 00000000..31265db8
--- /dev/null
+++ b/system/include/emscripten/emmintrin.h
@@ -0,0 +1,87 @@
+#include <xmmintrin.h>
+
+typedef int32x4 __m128i;
+
+static __inline__ __m128i __attribute__((__always_inline__))
+_mm_set_epi32(int z, int y, int x, int w)
+{
+  return (__m128i){ w, x, y, z };
+}
+
+static __inline__ __m128i __attribute__((__always_inline__))
+_mm_set1_epi32(int w)
+{
+  return (__m128i){ w, w, w, w };
+}
+
+static __inline__ __m128i __attribute__((__always_inline__))
+_mm_setzero_si128()
+{
+  return (__m128i){ 0, 0, 0, 0 };
+}
+
+static __inline__ void __attribute__((__always_inline__))
+_mm_store_si128(__m128i *p, __m128i a)
+{
+  *p = a;
+}
+
+static __inline__ __m128i __attribute__((__always_inline__))
+_mm_and_si128(__m128i a, __m128i b)
+{
+  return a & b;
+}
+
+static __inline__ __m128i __attribute__((__always_inline__))
+_mm_andnot_si128(__m128i a, __m128i b)
+{
+  return ~a & b;
+}
+
+static __inline__ __m128i __attribute__((__always_inline__))
+_mm_or_si128(__m128i a, __m128i b)
+{
+  return a | b;
+}
+
+static __inline__ __m128i __attribute__((__always_inline__))
+_mm_xor_si128(__m128i a, __m128i b)
+{
+  return a ^ b;
+}
+
+static __inline__ __m128i __attribute__((__always_inline__))
+_mm_add_epi32(__m128i a, __m128i b)
+{
+  return a + b;
+}
+
+static __inline__ __m128i __attribute__((__always_inline__))
+_mm_sub_epi32(__m128i a, __m128i b)
+{
+  return a - b;
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_castsi128_ps(__m128i a)
+{
+  return emscripten_int32x4_bitsToFloat32x4(a);
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_cvtepi32_ps(__m128i a)
+{
+  return emscripten_int32x4_toFloat32x4(a);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__))
+_mm_castps_si128(__m128 a)
+{
+  return emscripten_float32x4_bitsToInt32x4(a);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__))
+_mm_cvtps_epi32(__m128 a)
+{
+  return emscripten_float32x4_toInt32x4(a);
+}
\ No newline at end of file
diff --git a/system/include/emscripten/xmmintrin.h b/system/include/emscripten/xmmintrin.h
new file mode 100644
index 00000000..1b9108fa
--- /dev/null
+++ b/system/include/emscripten/xmmintrin.h
@@ -0,0 +1,131 @@
+#include <vector.h>
+
+typedef float32x4 __m128;
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_set_ps(float z, float y, float x, float w)
+{
+  return (__m128){ w, x, y, z };
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_set1_ps(float w)
+{
+  return (__m128){ w, w, w, w };
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_setzero_ps(void)
+{
+  return (__m128){ 0.0, 0.0, 0.0, 0.0 };
+}
+
+static __inline__ void __attribute__((__always_inline__))
+_mm_store_ps(float *p, __m128 a)
+{
+  *(__m128 *)p = a;
+}
+
+static __inline__ int __attribute__((__always_inline__))
+_mm_movemask_ps(__m128 a)
+{
+  return emscripten_float32x4_signmask(a);
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_add_ps(__m128 a, __m128 b)
+{
+  return a + b;
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_sub_ps(__m128 a, __m128 b)
+{
+  return a - b;
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_mul_ps(__m128 a, __m128 b)
+{
+  return a * b;
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_div_ps(__m128 a, __m128 b)
+{
+  return a / b;
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_min_ps(__m128 a, __m128 b)
+{
+  return emscripten_float32x4_min(a, b);
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_max_ps(__m128 a, __m128 b)
+{
+  return emscripten_float32x4_max(a, b);
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_sqrt_ps(__m128 a)
+{
+  return emscripten_float32x4_sqrt(a);
+}
+
+/* TODO: shuffles */
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_cmplt_ps(__m128 a, __m128 b)
+{
+  return emscripten_float32x4_lessThan(a, b);
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_cmple_ps(__m128 a, __m128 b)
+{
+  return emscripten_float32x4_lessThanOrEqual(a, b);
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_cmpeq_ps(__m128 a, __m128 b)
+{
+  return emscripten_float32x4_equal(a, b);
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_cmpge_ps(__m128 a, __m128 b)
+{
+  return emscripten_float32x4_greaterThanOrEqual(a, b);
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_cmpgt_ps(__m128 a, __m128 b)
+{
+  return emscripten_float32x4_greaterThan(a, b);
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_and_ps(__m128 a, __m128 b)
+{
+  return emscripten_float32x4_and(a, b);
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_andnot_ps(__m128 a, __m128 b)
+{
+  return emscripten_float32x4_andNot(a, b);
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_or_ps(__m128 a, __m128 b)
+{
+  return emscripten_float32x4_or(a, b);
+}
+
+static __inline__ __m128 __attribute__((__always_inline__))
+_mm_xor_ps(__m128 a, __m128 b)
+{
+  return emscripten_float32x4_xor(a, b);
+}