Implement arm64 vaddlvq_u8 and vld1q_u8_x4 vendor intrinsics

This is required for using the bytecount crate on arm64.

Author: bjorn3

src/intrinsics/llvm_aarch64.rs

@@ -17,6 +17,14 @@ pub(crate) fn codegen_aarch64_llvm_intrinsic_call<'tcx>(
             fx.bcx.ins().fence();
         }
 
+        "llvm.aarch64.neon.ld1x4.v16i8.p0i8" => {
+            intrinsic_args!(fx, args => (ptr); intrinsic);
+
+            let ptr = ptr.load_scalar(fx);
+            let val = CPlace::for_ptr(Pointer::new(ptr), ret.layout()).to_cvalue(fx);
+            ret.write_cvalue(fx, val);
+        }
+
         _ if intrinsic.starts_with("llvm.aarch64.neon.abs.v") => {
             intrinsic_args!(fx, args => (a); intrinsic);
 
@@ -115,6 +123,22 @@ pub(crate) fn codegen_aarch64_llvm_intrinsic_call<'tcx>(
             );
         }
 
+        "llvm.aarch64.neon.uaddlv.i32.v16i8" => {
+            intrinsic_args!(fx, args => (v); intrinsic);
+
+            let mut res_val = fx.bcx.ins().iconst(types::I16, 0);
+            for lane_idx in 0..16 {
+                let lane = v.value_lane(fx, lane_idx).load_scalar(fx);
+                let lane = fx.bcx.ins().uextend(types::I16, lane);
+                res_val = fx.bcx.ins().iadd(res_val, lane);
+            }
+            let res = CValue::by_val(
+                fx.bcx.ins().uextend(types::I32, res_val),
+                fx.layout_of(fx.tcx.types.u32),
+            );
+            ret.write_cvalue(fx, res);
+        }
+
         _ if intrinsic.starts_with("llvm.aarch64.neon.faddv.f32.v") => {
             intrinsic_args!(fx, args => (v); intrinsic);