Auto merge of #146148 - Flakebi:global-addrspace-test, r=Mark-Simulacrum

Add amdgpu test for addrspacecasting global vars and the gpu-kernel calling convention

Add two tests that can now be added, as the amdgpu is merged.

- Global variables are casted to the default address space since #135026
- gpu-kernel calling convention, translatos to amdgpu_kernel #135047

Tracking issue: #135024

Author: bors

tests/auxiliary/minicore.rs

@@ -179,7 +179,14 @@ impl Add<isize> for isize {
 
 #[lang = "sync"]
 trait Sync {}
-impl Sync for u8 {}
+impl_marker_trait!(
+    Sync => [
+        char, bool,
+        isize, i8, i16, i32, i64, i128,
+        usize, u8, u16, u32, u64, u128,
+        f16, f32, f64, f128,
+    ]
+);
 
 #[lang = "drop_in_place"]
 fn drop_in_place<T>(_: *mut T) {}

tests/codegen-llvm/amdgpu-addrspacecast.rs

@@ -16,3 +16,12 @@ pub fn ref_of_local(f: fn(&i32)) {
     let i = 0;
     f(&i);
 }
+
+// CHECK-LABEL: @ref_of_global
+// CHECK: addrspacecast (ptr addrspace(1) @I to ptr)
+#[no_mangle]
+pub fn ref_of_global(f: fn(&i32)) {
+    #[no_mangle]
+    static I: i32 = 0;
+    f(&I);
+}

tests/codegen-llvm/gpu-kernel-abi.rs

@@ -1,7 +1,9 @@
 // Checks that the gpu-kernel calling convention correctly translates to LLVM calling conventions.
 
 //@ add-core-stubs
-//@ revisions: nvptx
+//@ revisions: amdgpu nvptx
+//@ [amdgpu] compile-flags: --crate-type=rlib --target=amdgcn-amd-amdhsa -Ctarget-cpu=gfx900
+//@ [amdgpu] needs-llvm-components: amdgpu
 //@ [nvptx] compile-flags: --crate-type=rlib --target=nvptx64-nvidia-cuda
 //@ [nvptx] needs-llvm-components: nvptx
 #![feature(no_core, lang_items, abi_gpu_kernel)]
@@ -10,6 +12,7 @@
 extern crate minicore;
 use minicore::*;
 
+// amdgpu: define amdgpu_kernel void @fun(i32
 // nvptx: define ptx_kernel void @fun(i32
 #[no_mangle]
 pub extern "gpu-kernel" fn fun(_: i32) {}