mmx requires i686

Author: gnzlbg

coresimd/src/x86/i586/mod.rs

@@ -3,7 +3,6 @@
 pub use self::cpuid::*;
 pub use self::xsave::*;
 
-pub use self::mmx::*;
 pub use self::sse::*;
 pub use self::sse2::*;
 pub use self::sse3::*;
@@ -23,7 +22,6 @@ pub use self::tbm::*;
 mod cpuid;
 mod xsave;
 
-mod mmx;
 mod sse;
 mod sse2;
 mod sse3;

coresimd/src/x86/i586/mmx.rs renamed to coresimd/src/x86/i686/mmx.rs

@@ -18,7 +18,8 @@ use stdsimd_test::assert_instr;
 /// Constructs a 64-bit integer vector initialized to zero.
 #[inline(always)]
 #[target_feature = "+mmx,+sse"]
-#[cfg_attr(all(test, target_arch = "x86"), assert_instr(xorps))]
+// FIXME: this produces a movl instead of xorps on x86
+// FIXME: this produces a xor intrinsic instead of xorps on x86_64
 #[cfg_attr(all(test, target_arch = "x86_64"), assert_instr(xor))]
 pub unsafe fn _mm_setzero_si64() -> __m64 {
     mem::transmute(0_i64)
@@ -59,7 +60,7 @@ extern "C" {
 #[cfg(test)]
 mod tests {
     use v64::{i16x4, i32x2, i8x8};
-    use x86::i586::mmx;
+    use x86::i686::mmx;
     use x86::__m64;
     use stdsimd_test::simd_test;
 

coresimd/src/x86/i686/mod.rs

@@ -1,5 +1,8 @@
 //! `i686` intrinsics
 
+mod mmx;
+pub use self::mmx::*;
+
 mod sse;
 pub use self::sse::*;
 

coresimd/src/x86/i686/sse.rs

@@ -4,6 +4,8 @@ use v128::f32x4;
 use v64::{i16x4, i32x2, i8x8, u8x8};
 use x86::__m64;
 use core::mem;
+use x86::i586;
+use x86::i686::mmx;
 
 #[cfg(test)]
 use stdsimd_test::assert_instr;
@@ -139,9 +141,9 @@ pub unsafe fn _mm_cvt_ps2pi(a: f32x4) -> i32x2 {
 #[cfg_attr(test, assert_instr(cvtps2pi))]
 pub unsafe fn _mm_cvtps_pi16(a: f32x4) -> i16x4 {
     let b = _mm_cvtps_pi32(a);
-    let a = ::x86::_mm_movehl_ps(a, a);
+    let a = i586::_mm_movehl_ps(a, a);
     let c = _mm_cvtps_pi32(a);
-    ::x86::_mm_packs_pi32(b, c)
+    mmx::_mm_packs_pi32(b, c)
 }
 
 /// Convert packed single-precision (32-bit) floating-point elements in `a` to
@@ -152,8 +154,8 @@ pub unsafe fn _mm_cvtps_pi16(a: f32x4) -> i16x4 {
 #[cfg_attr(test, assert_instr(cvtps2pi))]
 pub unsafe fn _mm_cvtps_pi8(a: f32x4) -> i8x8 {
     let b = _mm_cvtps_pi16(a);
-    let c = ::x86::_mm_setzero_si64();
-    ::x86::_mm_packs_pi16(b, mem::transmute(c))
+    let c = mmx::_mm_setzero_si64();
+    mmx::_mm_packs_pi16(b, mem::transmute(c))
 }
 
 #[cfg(test)]