From b81b3e57ee6e6d38febb68d330e2d09466f914fa Mon Sep 17 00:00:00 2001
From: Jeff Muizelaar <jrmuizel@gmail.com>
Date: Tue, 8 Sep 2020 13:08:07 -0400
Subject: [PATCH] Add vld1q_s32 and vld1q_u32

---
 crates/core_arch/src/aarch64/neon/mod.rs | 46 ++++++++++++++++++++++++
 crates/core_arch/src/arm/neon/mod.rs     | 44 +++++++++++++++++++++++
 2 files changed, 90 insertions(+)

diff --git a/crates/core_arch/src/aarch64/neon/mod.rs b/crates/core_arch/src/aarch64/neon/mod.rs
index 3712c99ba3..cbf3d3638a 100644
--- a/crates/core_arch/src/aarch64/neon/mod.rs
+++ b/crates/core_arch/src/aarch64/neon/mod.rs
@@ -1812,6 +1812,32 @@ pub unsafe fn vld1q_f32(addr: *const f32) -> float32x4_t {
     ))
 }
 
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(ldr))]
+pub unsafe fn vld1q_s32(addr: *const i32) -> int32x4_t {
+    use crate::core_arch::simd::i32x4;
+    transmute(i32x4::new(
+        *addr,
+        *addr.offset(1),
+        *addr.offset(2),
+        *addr.offset(3),
+    ))
+}
+
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(ldr))]
+pub unsafe fn vld1q_u32(addr: *const u32) -> uint32x4_t {
+    use crate::core_arch::simd::u32x4;
+    transmute(u32x4::new(
+        *addr,
+        *addr.offset(1),
+        *addr.offset(2),
+        *addr.offset(3),
+    ))
+}
+
 #[cfg(test)]
 mod tests {
     use crate::core_arch::aarch64::test_support::*;
@@ -1830,6 +1856,26 @@ mod tests {
         assert_eq!(r, e);
     }
 
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vld1q_s32() {
+        let e = i32x4::new(1, 2, 3, 4);
+        let f = [0, 1, 2, 3, 4];
+        // do a load that has 4 byte alignment to make sure we're not
+        // over aligning it
+        let r: i32x4 = transmute(vld1q_s32(f[1..].as_ptr()));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vld1q_u32() {
+        let e = u32x4::new(1, 2, 3, 4);
+        let f = [0, 1, 2, 3, 4];
+        // do a load that has 4 byte alignment to make sure we're not
+        // over aligning it
+        let r: u32x4 = transmute(vld1q_u32(f[1..].as_ptr()));
+        assert_eq!(r, e);
+    }
+
     #[simd_test(enable = "neon")]
     unsafe fn test_vpaddq_s16() {
         let a = i16x8::new(1, 2, 3, 4, 5, 6, 7, 8);
diff --git a/crates/core_arch/src/arm/neon/mod.rs b/crates/core_arch/src/arm/neon/mod.rs
index 03dce6dd1b..c733f01f76 100644
--- a/crates/core_arch/src/arm/neon/mod.rs
+++ b/crates/core_arch/src/arm/neon/mod.rs
@@ -219,6 +219,8 @@ extern "C" {
     ) -> int8x8_t;
     #[cfg_attr(target_arch = "arm", link_name = "llvm.arm.neon.vld1.v4f32.p0i8")]
     fn vld1q_v4f32(addr: *const u8, align: u32) -> float32x4_t;
+    #[cfg_attr(target_arch = "arm", link_name = "llvm.arm.neon.vld1.v4i32.p0i8")]
+    fn vld1q_v4i32(addr: *const u8, align: u32) -> int32x4_t;
 }
 
 /// Absolute value (wrapping).
@@ -1769,6 +1771,26 @@ pub unsafe fn vld1q_u8(addr: *const u8) -> uint8x16_t {
     ptr::read(addr as *const uint8x16_t)
 }
 
+/// Load multiple single-element structures to one, two, three, or four registers
+#[inline]
+#[cfg(target_arch = "arm")]
+#[target_feature(enable = "neon")]
+#[target_feature(enable = "v7")]
+#[cfg_attr(test, assert_instr("vld1.32"))]
+pub unsafe fn vld1q_s32(addr: *const i32) -> int32x4_t {
+    vld1q_v4i32(addr as *const u8, 4)
+}
+
+/// Load multiple single-element structures to one, two, three, or four registers
+#[inline]
+#[cfg(target_arch = "arm")]
+#[target_feature(enable = "neon")]
+#[target_feature(enable = "v7")]
+#[cfg_attr(test, assert_instr("vld1.32"))]
+pub unsafe fn vld1q_u32(addr: *const u32) -> uint32x4_t {
+    transmute(vld1q_v4i32(addr as *const u8, 4))
+}
+
 /// Load multiple single-element structures to one, two, three, or four registers
 #[inline]
 #[cfg(target_arch = "arm")]
@@ -1826,6 +1848,28 @@ mod tests {
         assert_eq!(r, e);
     }
 
+    #[cfg(target_arch = "arm")]
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vld1q_s32() {
+        let e = i32x4::new(1, 2, 3, 4);
+        let f = [0, 1, 2, 3, 4];
+        // do a load that has 4 byte alignment to make sure we're not
+        // over aligning it
+        let r: i32x4 = transmute(vld1q_s32(f[1..].as_ptr()));
+        assert_eq!(r, e);
+    }
+
+    #[cfg(target_arch = "arm")]
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vld1q_u32() {
+        let e = u32x4::new(1, 2, 3, 4);
+        let f = [0, 1, 2, 3, 4];
+        // do a load that has 4 byte alignment to make sure we're not
+        // over aligning it
+        let r: u32x4 = transmute(vld1q_u32(f[1..].as_ptr()));
+        assert_eq!(r, e);
+    }
+
     #[simd_test(enable = "neon")]
     unsafe fn test_vld1q_dup_f32() {
         let e = f32x4::new(1., 1., 1., 1.);