Add comments in NT asm blocks for future reference

Author: sayantn

crates/core_arch/src/x86/avx.rs

@@ -1833,6 +1833,7 @@ pub unsafe fn _mm256_lddqu_si256(mem_addr: *const __m256i) -> __m256i {
 #[cfg_attr(test, assert_instr(vmovntdq))]
 #[stable(feature = "simd_x86", since = "1.27.0")]
 pub unsafe fn _mm256_stream_si256(mem_addr: *mut __m256i, a: __m256i) {
+    // see #1541, we should use inline asm to be sure, because LangRef isn't clear enough
     crate::arch::asm!(
         vps!("vmovntdq", ",{a}"),
         p = in(reg) mem_addr,
@@ -1861,6 +1862,7 @@ pub unsafe fn _mm256_stream_si256(mem_addr: *mut __m256i, a: __m256i) {
 #[stable(feature = "simd_x86", since = "1.27.0")]
 #[allow(clippy::cast_ptr_alignment)]
 pub unsafe fn _mm256_stream_pd(mem_addr: *mut f64, a: __m256d) {
+    // see #1541, we should use inline asm to be sure, because LangRef isn't clear enough
     crate::arch::asm!(
         vps!("vmovntpd", ",{a}"),
         p = in(reg) mem_addr,
@@ -1890,6 +1892,7 @@ pub unsafe fn _mm256_stream_pd(mem_addr: *mut f64, a: __m256d) {
 #[stable(feature = "simd_x86", since = "1.27.0")]
 #[allow(clippy::cast_ptr_alignment)]
 pub unsafe fn _mm256_stream_ps(mem_addr: *mut f32, a: __m256) {
+    // see #1541, we should use inline asm to be sure, because LangRef isn't clear enough
     crate::arch::asm!(
         vps!("vmovntps", ",{a}"),
         p = in(reg) mem_addr,

crates/core_arch/src/x86/avx512f.rs

@@ -29593,6 +29593,7 @@ pub fn _mm_mask_testn_epi64_mask(k: __mmask8, a: __m128i, b: __m128i) -> __mmask
 #[cfg_attr(test, assert_instr(vmovntps))]
 #[allow(clippy::cast_ptr_alignment)]
 pub unsafe fn _mm512_stream_ps(mem_addr: *mut f32, a: __m512) {
+    // see #1541, we should use inline asm to be sure, because LangRef isn't clear enough
     crate::arch::asm!(
         vps!("vmovntps", ",{a}"),
         p = in(reg) mem_addr,
@@ -29619,6 +29620,7 @@ pub unsafe fn _mm512_stream_ps(mem_addr: *mut f32, a: __m512) {
 #[cfg_attr(test, assert_instr(vmovntpd))]
 #[allow(clippy::cast_ptr_alignment)]
 pub unsafe fn _mm512_stream_pd(mem_addr: *mut f64, a: __m512d) {
+    // see #1541, we should use inline asm to be sure, because LangRef isn't clear enough
     crate::arch::asm!(
         vps!("vmovntpd", ",{a}"),
         p = in(reg) mem_addr,
@@ -29645,6 +29647,7 @@ pub unsafe fn _mm512_stream_pd(mem_addr: *mut f64, a: __m512d) {
 #[cfg_attr(test, assert_instr(vmovntdq))]
 #[allow(clippy::cast_ptr_alignment)]
 pub unsafe fn _mm512_stream_si512(mem_addr: *mut __m512i, a: __m512i) {
+    // see #1541, we should use inline asm to be sure, because LangRef isn't clear enough
     crate::arch::asm!(
         vps!("vmovntdq", ",{a}"),
         p = in(reg) mem_addr,

crates/core_arch/src/x86/sse.rs

@@ -2022,6 +2022,7 @@ unsafe extern "C" {
 #[stable(feature = "simd_x86", since = "1.27.0")]
 #[allow(clippy::cast_ptr_alignment)]
 pub unsafe fn _mm_stream_ps(mem_addr: *mut f32, a: __m128) {
+    // see #1541, we should use inline asm to be sure, because LangRef isn't clear enough
     crate::arch::asm!(
         vps!("movntps", ",{a}"),
         p = in(reg) mem_addr,

crates/core_arch/src/x86/sse2.rs

@@ -1363,6 +1363,7 @@ pub unsafe fn _mm_storel_epi64(mem_addr: *mut __m128i, a: __m128i) {
 #[cfg_attr(test, assert_instr(movntdq))]
 #[stable(feature = "simd_x86", since = "1.27.0")]
 pub unsafe fn _mm_stream_si128(mem_addr: *mut __m128i, a: __m128i) {
+    // see #1541, we should use inline asm to be sure, because LangRef isn't clear enough
     crate::arch::asm!(
         vps!("movntdq",  ",{a}"),
         p = in(reg) mem_addr,
@@ -1390,6 +1391,7 @@ pub unsafe fn _mm_stream_si128(mem_addr: *mut __m128i, a: __m128i) {
 #[cfg_attr(test, assert_instr(movnti))]
 #[stable(feature = "simd_x86", since = "1.27.0")]
 pub unsafe fn _mm_stream_si32(mem_addr: *mut i32, a: i32) {
+    // see #1541, we should use inline asm to be sure, because LangRef isn't clear enough
     crate::arch::asm!(
         vps!("movnti", ",{a:e}"), // `:e` for 32bit value
         p = in(reg) mem_addr,
@@ -2627,6 +2629,7 @@ pub unsafe fn _mm_loadl_pd(a: __m128d, mem_addr: *const f64) -> __m128d {
 #[stable(feature = "simd_x86", since = "1.27.0")]
 #[allow(clippy::cast_ptr_alignment)]
 pub unsafe fn _mm_stream_pd(mem_addr: *mut f64, a: __m128d) {
+    // see #1541, we should use inline asm to be sure, because LangRef isn't clear enough
     crate::arch::asm!(
         vps!("movntpd", ",{a}"),
         p = in(reg) mem_addr,

crates/core_arch/src/x86/sse4a.rs

@@ -110,6 +110,7 @@ pub fn _mm_inserti_si64<const LEN: i32, const IDX: i32>(x: __m128i, y: __m128i)
 #[cfg_attr(test, assert_instr(movntsd))]
 #[stable(feature = "simd_x86", since = "1.27.0")]
 pub unsafe fn _mm_stream_sd(p: *mut f64, a: __m128d) {
+    // see #1541, we should use inline asm to be sure, because LangRef isn't clear enough
     crate::arch::asm!(
         vps!("movntsd",  ",{a}"),
         p = in(reg) p,
@@ -135,6 +136,7 @@ pub unsafe fn _mm_stream_sd(p: *mut f64, a: __m128d) {
 #[cfg_attr(test, assert_instr(movntss))]
 #[stable(feature = "simd_x86", since = "1.27.0")]
 pub unsafe fn _mm_stream_ss(p: *mut f32, a: __m128) {
+    // see #1541, we should use inline asm to be sure, because LangRef isn't clear enough
     crate::arch::asm!(
         vps!("movntss",  ",{a}"),
         p = in(reg) p,

crates/core_arch/src/x86_64/sse2.rs

@@ -78,6 +78,7 @@ pub fn _mm_cvttsd_si64x(a: __m128d) -> i64 {
 #[cfg_attr(test, assert_instr(movnti))]
 #[stable(feature = "simd_x86", since = "1.27.0")]
 pub unsafe fn _mm_stream_si64(mem_addr: *mut i64, a: i64) {
+    // see #1541, we should use inline asm to be sure, because LangRef isn't clear enough
     crate::arch::asm!(
         vps!("movnti", ",{a}"),
         p = in(reg) mem_addr,