Merge pull request rust-lang#1566 from rust-lang/arm64_fixes

Fix the half and bytecount crates on arm64

Author: bjorn3

src/inline_asm.rs

@@ -612,6 +612,15 @@ impl<'tcx> InlineAssemblyGenerator<'_, 'tcx> {
             generated_asm.push_str(".att_syntax\n");
         }
 
+        if self.arch == InlineAsmArch::AArch64 {
+            for feature in &self.tcx.codegen_fn_attrs(self.enclosing_def_id).target_features {
+                if feature.name == sym::neon {
+                    continue;
+                }
+                writeln!(generated_asm, ".arch_extension {}", feature.name).unwrap();
+            }
+        }
+
         // The actual inline asm
         for piece in self.template {
             match piece {
@@ -679,6 +688,15 @@ impl<'tcx> InlineAssemblyGenerator<'_, 'tcx> {
         }
         generated_asm.push('\n');
 
+        if self.arch == InlineAsmArch::AArch64 {
+            for feature in &self.tcx.codegen_fn_attrs(self.enclosing_def_id).target_features {
+                if feature.name == sym::neon {
+                    continue;
+                }
+                writeln!(generated_asm, ".arch_extension no{}", feature.name).unwrap();
+            }
+        }
+
         if is_x86 && self.options.contains(InlineAsmOptions::ATT_SYNTAX) {
             generated_asm.push_str(".intel_syntax noprefix\n");
         }

src/intrinsics/llvm_aarch64.rs

@@ -21,7 +21,7 @@ pub(crate) fn codegen_aarch64_llvm_intrinsic_call<'tcx>(
             fx.bcx.ins().fence();
         }
 
-        "llvm.aarch64.neon.ld1x4.v16i8.p0i8" => {
+        "llvm.aarch64.neon.ld1x4.v16i8.p0" => {
             intrinsic_args!(fx, args => (ptr); intrinsic);
 
             let ptr = ptr.load_scalar(fx);
@@ -253,7 +253,7 @@ pub(crate) fn codegen_aarch64_llvm_intrinsic_call<'tcx>(
             }
             let res = CValue::by_val(
                 fx.bcx.ins().uextend(types::I32, res_val),
-                fx.layout_of(fx.tcx.types.u32),
+                fx.layout_of(fx.tcx.types.i32),
             );
             ret.write_cvalue(fx, res);
         }