Use SIMD intrinsics for cvtsi{,64}_{ss,sd} intrinsics

Author: sayantn

crates/core_arch/src/x86/sse.rs

@@ -882,7 +882,7 @@ pub fn _mm_cvtss_f32(a: __m128) -> f32 {
 #[cfg_attr(test, assert_instr(cvtsi2ss))]
 #[stable(feature = "simd_x86", since = "1.27.0")]
 pub fn _mm_cvtsi32_ss(a: __m128, b: i32) -> __m128 {
-    unsafe { cvtsi2ss(a, b) }
+    unsafe { simd_insert!(a, 0, b as f32) }
 }
 
 /// Alias for [`_mm_cvtsi32_ss`](fn._mm_cvtsi32_ss.html).
@@ -1989,8 +1989,6 @@ unsafe extern "C" {
     fn cvtss2si(a: __m128) -> i32;
     #[link_name = "llvm.x86.sse.cvttss2si"]
     fn cvttss2si(a: __m128) -> i32;
-    #[link_name = "llvm.x86.sse.cvtsi2ss"]
-    fn cvtsi2ss(a: __m128, b: i32) -> __m128;
     #[link_name = "llvm.x86.sse.sfence"]
     fn sfence();
     #[link_name = "llvm.x86.sse.stmxcsr"]

crates/core_arch/src/x86/sse2.rs

@@ -2422,7 +2422,10 @@ pub fn _mm_cvtsd_f64(a: __m128d) -> f64 {
 #[cfg_attr(test, assert_instr(cvtss2sd))]
 #[stable(feature = "simd_x86", since = "1.27.0")]
 pub fn _mm_cvtss_sd(a: __m128d, b: __m128) -> __m128d {
-    unsafe { cvtss2sd(a, b) }
+    unsafe {
+        let elt: f32 = simd_extract!(b, 0);
+        simd_insert!(a, 0, elt as f64)
+    }
 }
 
 /// Converts packed double-precision (64-bit) floating-point elements in `a` to
@@ -3118,8 +3121,6 @@ unsafe extern "C" {
     fn cvtsd2si(a: __m128d) -> i32;
     #[link_name = "llvm.x86.sse2.cvtsd2ss"]
     fn cvtsd2ss(a: __m128, b: __m128d) -> __m128;
-    #[link_name = "llvm.x86.sse2.cvtss2sd"]
-    fn cvtss2sd(a: __m128d, b: __m128) -> __m128d;
     #[link_name = "llvm.x86.sse2.cvttpd2dq"]
     fn cvttpd2dq(a: __m128d) -> i32x4;
     #[link_name = "llvm.x86.sse2.cvttsd2si"]

crates/core_arch/src/x86_64/sse.rs

@@ -11,8 +11,6 @@ unsafe extern "C" {
     fn cvtss2si64(a: __m128) -> i64;
     #[link_name = "llvm.x86.sse.cvttss2si64"]
     fn cvttss2si64(a: __m128) -> i64;
-    #[link_name = "llvm.x86.sse.cvtsi642ss"]
-    fn cvtsi642ss(a: __m128, b: i64) -> __m128;
 }
 
 /// Converts the lowest 32 bit float in the input vector to a 64 bit integer.
@@ -65,7 +63,7 @@ pub fn _mm_cvttss_si64(a: __m128) -> i64 {
 #[cfg_attr(test, assert_instr(cvtsi2ss))]
 #[stable(feature = "simd_x86", since = "1.27.0")]
 pub fn _mm_cvtsi64_ss(a: __m128, b: i64) -> __m128 {
-    unsafe { cvtsi642ss(a, b) }
+    unsafe { simd_insert!(a, 0, b as f32) }
 }
 
 #[cfg(test)]