Explain the discrepancy in the mask type for _mm_shuffle_ps (#879)

Author: georgio

crates/core_arch/src/x86/sse.rs

@@ -992,6 +992,12 @@ pub const fn _MM_SHUFFLE(z: u32, y: u32, x: u32, w: u32) -> i32 {
 /// `b`. Mask is split to 2 control bits each to index the element from inputs.
 ///
 /// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_shuffle_ps)
+///
+/// Note that there appears to be a mistake within Intel's Intrinsics Guide.
+/// `_mm_shuffle_ps` is supposed to take an `i32` instead of an `u32`
+/// as is the case for [other shuffle intrinsics](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm_shuffle_).
+/// Performing an implicit type conversion between an unsigned integer and a signed integer
+/// does not cause a problem in C, however Rust's commitment to strong typing does not allow this.
 #[inline]
 #[target_feature(enable = "sse")]
 #[cfg_attr(test, assert_instr(shufps, mask = 3))]