Pr feedback for instruction & hookup CI for aarch64_be

Author: Jamesbarford

.github/workflows/main.yml

@@ -56,6 +56,9 @@ jobs:
           os: ubuntu-latest
         - tuple: aarch64-unknown-linux-gnu
           os: ubuntu-latest
+        - target:
+          tuple: aarch64_be-unknown-linux-gnu
+          os: ubuntu-latest
         - tuple: riscv64gc-unknown-linux-gnu
           os: ubuntu-latest
         - tuple: powerpc-unknown-linux-gnu
@@ -125,6 +128,11 @@ jobs:
             tuple: aarch64-unknown-linux-gnu
             os: ubuntu-latest
           test_everything: true
+        - target:
+          tuple: aarch64_be-unknown-linux-gnu
+          os: ubuntu-latest
+          test_everything: true
+          build_std: true
         - target:
             tuple: armv7-unknown-linux-gnueabihf
             os: ubuntu-latest
@@ -199,6 +207,11 @@ jobs:
         rustup update nightly --no-self-update
         rustup default nightly
     - run: rustup target add ${{ matrix.target.tuple }}
+      if: matrix.build_std == ''
+    - run: |
+        rustup component add rust-src
+        echo "CARGO_UNSTABLE_BUILD_STD=std" >> $GITHUB_ENV
+      if: matrix.build_std != ''
     - run: cargo generate-lockfile
 
     # Configure some env vars based on matrix configuration

crates/core_arch/src/aarch64/neon/generated.rs

@@ -24470,6 +24470,73 @@ pub fn vrsrad_n_u64<const N: i32>(a: u64, b: u64) -> u64 {
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vrsubhn_high_s16)"]
 #[inline]
 #[target_feature(enable = "neon")]
+#[cfg(target_endian = "little")]
+#[cfg_attr(test, assert_instr(rsubhn2))]
+#[stable(feature = "neon_intrinsics", since = "1.59.0")]
+pub fn vrsubhn_high_s16(a: int8x8_t, b: int16x8_t, c: int16x8_t) -> int8x16_t {
+    let x: int8x8_t = vrsubhn_s16(b, c);
+    unsafe { simd_shuffle!(a, x, [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]) }
+}
+#[doc = "Rounding subtract returning high narrow"]
+#[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vrsubhn_high_s32)"]
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg(target_endian = "little")]
+#[cfg_attr(test, assert_instr(rsubhn2))]
+#[stable(feature = "neon_intrinsics", since = "1.59.0")]
+pub fn vrsubhn_high_s32(a: int16x4_t, b: int32x4_t, c: int32x4_t) -> int16x8_t {
+    let x: int16x4_t = vrsubhn_s32(b, c);
+    unsafe { simd_shuffle!(a, x, [0, 1, 2, 3, 4, 5, 6, 7]) }
+}
+#[doc = "Rounding subtract returning high narrow"]
+#[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vrsubhn_high_s64)"]
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg(target_endian = "little")]
+#[cfg_attr(test, assert_instr(rsubhn2))]
+#[stable(feature = "neon_intrinsics", since = "1.59.0")]
+pub fn vrsubhn_high_s64(a: int32x2_t, b: int64x2_t, c: int64x2_t) -> int32x4_t {
+    let x: int32x2_t = vrsubhn_s64(b, c);
+    unsafe { simd_shuffle!(a, x, [0, 1, 2, 3]) }
+}
+#[doc = "Rounding subtract returning high narrow"]
+#[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vrsubhn_high_u16)"]
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg(target_endian = "little")]
+#[cfg_attr(test, assert_instr(rsubhn2))]
+#[stable(feature = "neon_intrinsics", since = "1.59.0")]
+pub fn vrsubhn_high_u16(a: uint8x8_t, b: uint16x8_t, c: uint16x8_t) -> uint8x16_t {
+    let x: uint8x8_t = vrsubhn_u16(b, c);
+    unsafe { simd_shuffle!(a, x, [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]) }
+}
+#[doc = "Rounding subtract returning high narrow"]
+#[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vrsubhn_high_u32)"]
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg(target_endian = "little")]
+#[cfg_attr(test, assert_instr(rsubhn2))]
+#[stable(feature = "neon_intrinsics", since = "1.59.0")]
+pub fn vrsubhn_high_u32(a: uint16x4_t, b: uint32x4_t, c: uint32x4_t) -> uint16x8_t {
+    let x: uint16x4_t = vrsubhn_u32(b, c);
+    unsafe { simd_shuffle!(a, x, [0, 1, 2, 3, 4, 5, 6, 7]) }
+}
+#[doc = "Rounding subtract returning high narrow"]
+#[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vrsubhn_high_u64)"]
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg(target_endian = "little")]
+#[cfg_attr(test, assert_instr(rsubhn2))]
+#[stable(feature = "neon_intrinsics", since = "1.59.0")]
+pub fn vrsubhn_high_u64(a: uint32x2_t, b: uint64x2_t, c: uint64x2_t) -> uint32x4_t {
+    let x: uint32x2_t = vrsubhn_u64(b, c);
+    unsafe { simd_shuffle!(a, x, [0, 1, 2, 3]) }
+}
+#[doc = "Rounding subtract returning high narrow"]
+#[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vrsubhn_high_s16)"]
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg(target_endian = "big")]
 #[cfg_attr(test, assert_instr(rsubhn))]
 #[stable(feature = "neon_intrinsics", since = "1.59.0")]
 pub fn vrsubhn_high_s16(a: int8x8_t, b: int16x8_t, c: int16x8_t) -> int8x16_t {
@@ -24480,6 +24547,7 @@ pub fn vrsubhn_high_s16(a: int8x8_t, b: int16x8_t, c: int16x8_t) -> int8x16_t {
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vrsubhn_high_s32)"]
 #[inline]
 #[target_feature(enable = "neon")]
+#[cfg(target_endian = "big")]
 #[cfg_attr(test, assert_instr(rsubhn))]
 #[stable(feature = "neon_intrinsics", since = "1.59.0")]
 pub fn vrsubhn_high_s32(a: int16x4_t, b: int32x4_t, c: int32x4_t) -> int16x8_t {
@@ -24490,6 +24558,7 @@ pub fn vrsubhn_high_s32(a: int16x4_t, b: int32x4_t, c: int32x4_t) -> int16x8_t {
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vrsubhn_high_s64)"]
 #[inline]
 #[target_feature(enable = "neon")]
+#[cfg(target_endian = "big")]
 #[cfg_attr(test, assert_instr(rsubhn))]
 #[stable(feature = "neon_intrinsics", since = "1.59.0")]
 pub fn vrsubhn_high_s64(a: int32x2_t, b: int64x2_t, c: int64x2_t) -> int32x4_t {
@@ -24500,6 +24569,7 @@ pub fn vrsubhn_high_s64(a: int32x2_t, b: int64x2_t, c: int64x2_t) -> int32x4_t {
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vrsubhn_high_u16)"]
 #[inline]
 #[target_feature(enable = "neon")]
+#[cfg(target_endian = "big")]
 #[cfg_attr(test, assert_instr(rsubhn))]
 #[stable(feature = "neon_intrinsics", since = "1.59.0")]
 pub fn vrsubhn_high_u16(a: uint8x8_t, b: uint16x8_t, c: uint16x8_t) -> uint8x16_t {
@@ -24510,6 +24580,7 @@ pub fn vrsubhn_high_u16(a: uint8x8_t, b: uint16x8_t, c: uint16x8_t) -> uint8x16_
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vrsubhn_high_u32)"]
 #[inline]
 #[target_feature(enable = "neon")]
+#[cfg(target_endian = "big")]
 #[cfg_attr(test, assert_instr(rsubhn))]
 #[stable(feature = "neon_intrinsics", since = "1.59.0")]
 pub fn vrsubhn_high_u32(a: uint16x4_t, b: uint32x4_t, c: uint32x4_t) -> uint16x8_t {
@@ -24520,6 +24591,7 @@ pub fn vrsubhn_high_u32(a: uint16x4_t, b: uint32x4_t, c: uint32x4_t) -> uint16x8
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vrsubhn_high_u64)"]
 #[inline]
 #[target_feature(enable = "neon")]
+#[cfg(target_endian = "big")]
 #[cfg_attr(test, assert_instr(rsubhn))]
 #[stable(feature = "neon_intrinsics", since = "1.59.0")]
 pub fn vrsubhn_high_u64(a: uint32x2_t, b: uint64x2_t, c: uint64x2_t) -> uint32x4_t {

crates/stdarch-gen-arm/spec/neon/aarch64.spec.yml

@@ -59,6 +59,14 @@ neon-unstable-f16: &neon-unstable-f16
 neon-unstable-feat-lut: &neon-unstable-feat-lut
   FnCall: [unstable, ['feature = "stdarch_neon_feat_lut"', 'issue = "138050"']]
 
+# #[cfg(target_endian = "little")]
+little-endian: &little-endian
+  FnCall: [cfg, ['target_endian = "little"']]
+
+# #[cfg(target_endian = "big")]
+big-endian: &big-endian
+  FnCall: [cfg, ['target_endian = "big"']]
+
 intrinsics:
   - name: "vaddd_{type}"
     doc: Add
@@ -8906,6 +8914,30 @@ intrinsics:
     arguments: ["a: {neon_type[0]}", "b: {neon_type[1]}", "c: {neon_type[1]}"]
     return_type: "{neon_type[3]}"
     attr:
+      - *little-endian
+      - FnCall: [cfg_attr, [test, {FnCall: [assert_instr, [rsubhn2]]}]]
+      - FnCall: [stable, ['feature = "neon_intrinsics"', 'since = "1.59.0"']]
+    safety: safe
+    types:
+      - [int8x8_t, int16x8_t, int16x8_t, int8x16_t, '[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]']
+      - [int16x4_t, int32x4_t, int32x4_t, int16x8_t, '[0, 1, 2, 3, 4, 5, 6, 7]']
+      - [int32x2_t, int64x2_t, int64x2_t, int32x4_t, '[0, 1, 2, 3]']
+      - [uint8x8_t, uint16x8_t, uint16x8_t, uint8x16_t, '[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]']
+      - [uint16x4_t, uint32x4_t, uint32x4_t, uint16x8_t, '[0, 1, 2, 3, 4, 5, 6, 7]']
+      - [uint32x2_t, uint64x2_t, uint64x2_t, uint32x4_t, '[0, 1, 2, 3]']
+    compose:
+      - Let:
+          - x
+          - "{neon_type[0]}"
+          - FnCall: ["vrsubhn_{neon_type[1]}", [b, c]]
+      - FnCall: [simd_shuffle!, [a, x, "{type[4]}"]]
+
+  - name: "vrsubhn_high_{neon_type[1]}"
+    doc: "Rounding subtract returning high narrow"
+    arguments: ["a: {neon_type[0]}", "b: {neon_type[1]}", "c: {neon_type[1]}"]
+    return_type: "{neon_type[3]}"
+    attr:
+      - *big-endian
       - FnCall: [cfg_attr, [test, {FnCall: [assert_instr, [rsubhn]]}]]
       - FnCall: [stable, ['feature = "neon_intrinsics"', 'since = "1.59.0"']]
     safety: safe