Add missing ARM-v7A CRC intrinsics (#1515)

* Move aarch64 crc into arm shared module

* Add missing 32-bit arm crc intrinsics

On 32-bit ARM, this intrinsic emits two instructions and splits its 64-bit input parameter between them.

https://gcc.gnu.org/onlinedocs/gcc-4.9.4/gcc/ARM-ACLE-Intrinsics.html

Author: eupn

crates/core_arch/src/aarch64/crc.rs

@@ -17,10 +17,6 @@ mod tme;
 #[unstable(feature = "stdarch_aarch64_tme", issue = "117216")]
 pub use self::tme::*;
 
-mod crc;
-#[unstable(feature = "stdarch_arm_crc32", issue = "117215")]
-pub use self::crc::*;
-
 mod prefetch;
 #[unstable(feature = "stdarch_aarch64_prefetch", issue = "117217")]
 pub use self::prefetch::*;

crates/core_arch/src/aarch64/mod.rs

@@ -18,6 +18,10 @@ extern "unadjusted" {
     #[cfg_attr(target_arch = "aarch64", link_name = "llvm.aarch64.crc32cw")]
     #[cfg_attr(target_arch = "arm", link_name = "llvm.arm.crc32cw")]
     fn crc32cw_(crc: u32, data: u32) -> u32;
+    #[cfg_attr(target_arch = "aarch64", link_name = "llvm.aarch64.crc32x")]
+    fn crc32x_(crc: u32, data: u64) -> u32;
+    #[cfg_attr(target_arch = "aarch64", link_name = "llvm.aarch64.crc32cx")]
+    fn crc32cx_(crc: u32, data: u64) -> u32;
 }
 
 #[cfg(test)]
@@ -95,12 +99,82 @@ pub unsafe fn __crc32cw(crc: u32, data: u32) -> u32 {
     crc32cw_(crc, data)
 }
 
+/// CRC32 single round checksum for quad words (64 bits).
+///
+/// [Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/__crc32d)
+#[inline]
+#[target_feature(enable = "crc")]
+#[cfg(target_arch = "aarch64")]
+#[cfg_attr(test, assert_instr(crc32x))]
+#[unstable(feature = "stdarch_arm_crc32", issue = "117215")]
+pub unsafe fn __crc32d(crc: u32, data: u64) -> u32 {
+    crc32x_(crc, data)
+}
+
+/// CRC32 single round checksum for quad words (64 bits).
+///
+/// [Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/__crc32d)
+#[inline]
+#[target_feature(enable = "crc")]
+#[cfg(target_arch = "arm")]
+#[cfg_attr(test, assert_instr(crc32w))]
+#[unstable(feature = "stdarch_arm_crc32", issue = "117215")]
+pub unsafe fn __crc32d(crc: u32, data: u64) -> u32 {
+    // On 32-bit ARM this intrinsic emits a chain of two `crc32_w` instructions
+    // and truncates the data to 32 bits in both clang and gcc
+    crc32w_(
+        crc32w_(crc, (data & 0xffffffff) as u32),
+        (data >> 32) as u32,
+    )
+}
+
+/// CRC32 single round checksum for quad words (64 bits).
+///
+/// [Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/__crc32cd)
+#[inline]
+#[target_feature(enable = "crc")]
+#[cfg(target_arch = "aarch64")]
+#[cfg_attr(test, assert_instr(crc32cx))]
+#[unstable(feature = "stdarch_arm_crc32", issue = "117215")]
+pub unsafe fn __crc32cd(crc: u32, data: u64) -> u32 {
+    crc32cx_(crc, data)
+}
+
+/// CRC32 single round checksum for quad words (64 bits).
+///
+/// [Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/__crc32cd)
+#[inline]
+#[target_feature(enable = "crc")]
+#[cfg(target_arch = "arm")]
+#[cfg_attr(test, assert_instr(crc32cw))]
+#[unstable(feature = "stdarch_arm_crc32", issue = "117215")]
+pub unsafe fn __crc32cd(crc: u32, data: u64) -> u32 {
+    // On 32-bit ARM this intrinsic emits a chain of two `crc32_cw` instructions
+    // and truncates the data to 32 bits in both clang and gcc
+    crc32cw_(
+        crc32cw_(crc, (data & 0xffffffff) as u32),
+        (data >> 32) as u32,
+    )
+}
+
 #[cfg(test)]
 mod tests {
     use crate::core_arch::{arm_shared::*, simd::*};
     use std::mem;
     use stdarch_test::simd_test;
 
+    #[simd_test(enable = "crc")]
+    unsafe fn test_crc32d() {
+        assert_eq!(__crc32d(0, 0), 0);
+        assert_eq!(__crc32d(0, 18446744073709551615), 1147535477);
+    }
+
+    #[simd_test(enable = "crc")]
+    unsafe fn test_crc32cd() {
+        assert_eq!(__crc32cd(0, 0), 0);
+        assert_eq!(__crc32cd(0, 18446744073709551615), 3293575501);
+    }
+
     #[simd_test(enable = "crc")]
     unsafe fn test_crc32b() {
         assert_eq!(__crc32b(0, 0), 0);

crates/core_arch/src/arm_shared/crc.rs

@@ -14,8 +14,6 @@ vbfmlaltq_laneq_f32
 vbfmmlaq_f32
 
 # Implemented in Clang and stdarch for A64 only even though CSV claims A32 support
-__crc32d
-__crc32cd
 vaddq_p64
 vbsl_p64
 vbslq_p64