Reimplement _mm_movemask_ps and _mm_movemask_pd without LLVM intrinsics

rust-lang · Oct 10, 2023 · 9b7fb42 · 9b7fb42
1 parent 3b578c3
commit 9b7fb42
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 6 deletions.
diff --git a/crates/core_arch/src/x86/sse.rs b/crates/core_arch/src/x86/sse.rs
@@ -1081,7 +1081,7 @@ pub unsafe fn _mm_movelh_ps(a: __m128, b: __m128) -> __m128 {
 #[cfg_attr(test, assert_instr(movmskps))]
 #[stable(feature = "simd_x86", since = "1.27.0")]
 pub unsafe fn _mm_movemask_ps(a: __m128) -> i32 {
-    movmskps(a)
+    simd_bitmask::<u32x4, u8>(transmute(a)).into()
 }
 
 /// Construct a `__m128` with the lowest element read from `p` and the other
@@ -1885,8 +1885,6 @@ extern "C" {
     fn maxss(a: __m128, b: __m128) -> __m128;
     #[link_name = "llvm.x86.sse.max.ps"]
     fn maxps(a: __m128, b: __m128) -> __m128;
-    #[link_name = "llvm.x86.sse.movmsk.ps"]
-    fn movmskps(a: __m128) -> i32;
     #[link_name = "llvm.x86.sse.cmp.ps"]
     fn cmpps(a: __m128, b: __m128, imm8: i8) -> __m128;
     #[link_name = "llvm.x86.sse.comieq.ss"]

diff --git a/crates/core_arch/src/x86/sse2.rs b/crates/core_arch/src/x86/sse2.rs
@@ -2450,7 +2450,7 @@ pub unsafe fn _mm_setzero_pd() -> __m128d {
 #[cfg_attr(test, assert_instr(movmskpd))]
 #[stable(feature = "simd_x86", since = "1.27.0")]
 pub unsafe fn _mm_movemask_pd(a: __m128d) -> i32 {
-    movmskpd(a)
+    simd_bitmask::<u64x2, u8>(transmute(a)).into()
 }
 
 /// Loads 128-bits (composed of 2 packed double-precision (64-bit)
@@ -2914,8 +2914,6 @@ extern "C" {
     fn ucomigesd(a: __m128d, b: __m128d) -> i32;
     #[link_name = "llvm.x86.sse2.ucomineq.sd"]
     fn ucomineqsd(a: __m128d, b: __m128d) -> i32;
-    #[link_name = "llvm.x86.sse2.movmsk.pd"]
-    fn movmskpd(a: __m128d) -> i32;
     #[link_name = "llvm.x86.sse2.cvtpd2dq"]
     fn cvtpd2dq(a: __m128d) -> i32x4;
     #[link_name = "llvm.x86.sse2.cvtsd2si"]