Remove Into/From between x86 and portable types (rust-lang#292)

This is primarily doing to avoid falling into a portability trap by accident,
and in general makes the vendor types (on x86) going towards as minimal as they
can be. Along the way some tests were cleaned up which were still using the
portable types.

coresimd/src/v128.rs

@@ -55,8 +55,7 @@ define_from!(
     u16x8,
     i16x8,
     u8x16,
-    i8x16,
-    __m128i
+    i8x16
 );
 define_from!(
     i64x2,
@@ -66,8 +65,7 @@ define_from!(
     u16x8,
     i16x8,
     u8x16,
-    i8x16,
-    __m128i
+    i8x16
 );
 define_from!(
     u32x4,
@@ -77,8 +75,7 @@ define_from!(
     u16x8,
     i16x8,
     u8x16,
-    i8x16,
-    __m128i
+    i8x16
 );
 define_from!(
     i32x4,
@@ -88,8 +85,7 @@ define_from!(
     u16x8,
     i16x8,
     u8x16,
-    i8x16,
-    __m128i
+    i8x16
 );
 define_from!(
     u16x8,
@@ -99,8 +95,7 @@ define_from!(
     i32x4,
     i16x8,
     u8x16,
-    i8x16,
-    __m128i
+    i8x16
 );
 define_from!(
     i16x8,
@@ -110,8 +105,7 @@ define_from!(
     i32x4,
     u16x8,
     u8x16,
-    i8x16,
-    __m128i
+    i8x16
 );
 define_from!(
     u8x16,
@@ -121,22 +115,9 @@ define_from!(
     i32x4,
     u16x8,
     i16x8,
-    i8x16,
-    __m128i
-);
-define_from!(
-    i8x16,
-    u64x2,
-    i64x2,
-    u32x4,
-    i32x4,
-    u16x8,
-    i16x8,
-    u8x16,
-    __m128i
+    i8x16
 );
 define_from!(
-    __m128i,
     i8x16,
     u64x2,
     i64x2,

coresimd/src/v256.rs

@@ -80,8 +80,7 @@ define_from!(
     u16x16,
     i16x16,
     u8x32,
-    i8x32,
-    __m256i
+    i8x32
 );
 define_from!(
     i64x4,
@@ -91,8 +90,7 @@ define_from!(
     u16x16,
     i16x16,
     u8x32,
-    i8x32,
-    __m256i
+    i8x32
 );
 define_from!(
     u32x8,
@@ -102,8 +100,7 @@ define_from!(
     u16x16,
     i16x16,
     u8x32,
-    i8x32,
-    __m256i
+    i8x32
 );
 define_from!(
     i32x8,
@@ -113,8 +110,7 @@ define_from!(
     u16x16,
     i16x16,
     u8x32,
-    i8x32,
-    __m256i
+    i8x32
 );
 define_from!(
     u16x16,
@@ -124,8 +120,7 @@ define_from!(
     i32x8,
     i16x16,
     u8x32,
-    i8x32,
-    __m256i
+    i8x32
 );
 define_from!(
     i16x16,
@@ -135,8 +130,7 @@ define_from!(
     i32x8,
     u16x16,
     u8x32,
-    i8x32,
-    __m256i
+    i8x32
 );
 define_from!(
     u8x32,
@@ -146,22 +140,9 @@ define_from!(
     i32x8,
     u16x16,
     i16x16,
-    i8x32,
-    __m256i
-);
-define_from!(
-    i8x32,
-    u64x4,
-    i64x4,
-    u32x8,
-    i32x8,
-    u16x16,
-    i16x16,
-    u8x32,
-    __m256i
+    i8x32
 );
 define_from!(
-    __m256i,
     i8x32,
     u64x4,
     i64x4,

coresimd/src/v64.rs

@@ -35,13 +35,12 @@ define_ty_doc! {
     /// 64-bit wide integer vector type.
 }
 
-define_from!(u32x2, i32x2, u16x4, i16x4, u8x8, i8x8, __m64);
-define_from!(i32x2, u32x2, u16x4, i16x4, u8x8, i8x8, __m64);
-define_from!(u16x4, u32x2, i32x2, i16x4, u8x8, i8x8, __m64);
-define_from!(i16x4, u32x2, i32x2, u16x4, u8x8, i8x8, __m64);
-define_from!(u8x8, u32x2, i32x2, u16x4, i16x4, i8x8, __m64);
-define_from!(i8x8, u32x2, i32x2, u16x4, i16x4, u8x8, __m64);
-define_from!(__m64, i8x8, u32x2, i32x2, u16x4, i16x4, u8x8);
+define_from!(u32x2, i32x2, u16x4, i16x4, u8x8, i8x8);
+define_from!(i32x2, u32x2, u16x4, i16x4, u8x8, i8x8);
+define_from!(u16x4, u32x2, i32x2, i16x4, u8x8, i8x8);
+define_from!(i16x4, u32x2, i32x2, u16x4, u8x8, i8x8);
+define_from!(u8x8, u32x2, i32x2, u16x4, i16x4, i8x8);
+define_from!(i8x8, u32x2, i32x2, u16x4, i16x4, u8x8);
 
 define_common_ops!(f32x2, u32x2, i32x2, u16x4, i16x4, u8x8, i8x8);
 define_float_ops!(f32x2);

coresimd/src/x86/i586/sse3.rs

@@ -68,7 +68,7 @@ pub unsafe fn _mm_hsub_ps(a: __m128, b: __m128) -> __m128 {
 #[target_feature(enable = "sse3")]
 #[cfg_attr(test, assert_instr(lddqu))]
 pub unsafe fn _mm_lddqu_si128(mem_addr: *const __m128i) -> __m128i {
-    __m128i::from(lddqu(mem_addr as *const _))
+    mem::transmute(lddqu(mem_addr as *const _))
 }
 
 /// Duplicate the low double-precision (64-bit) floating-point element