[llvm] [SPIRV] Add tests documenting incorrect lowering of load/store atomic (PR #185628)

[Juan Manuel Martinez Caamaño via llvm-commits]

================
@@ -0,0 +1,63 @@
+; RUN: llc -O0 -mtriple=spirv64-- %s -o - | FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv64-- %s -o - -filetype=obj | spirv-val %}
+
+; RUN: llc -O0 -mtriple=spirv32-- %s -o - | FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv32-- %s -o - -filetype=obj | spirv-val %}
+
+;; Check that 'load atomic' LLVM IR instructions are lowered.
+;; NOTE: The current lowering is incorrect: 'load atomic' should produce
+;; OpAtomicLoad but currently produces OpLoad, silently dropping the atomic
+;; ordering. This test documents the broken behaviour so it can be fixed.
+
+; CHECK-DAG: %[[#Int32:]] = OpTypeInt 32 0
+; CHECK-DAG: %[[#Int64:]] = OpTypeInt 64 0
+
+; CHECK-LABEL: OpFunction %[[#]]
+; CHECK:       %[[#ptr:]] = OpFunctionParameter %[[#]]
+; CHECK:       %[[#]] = OpLoad %[[#Int32]] %[[#ptr]] Aligned 4
+; CHECK:       OpReturnValue
+
+define i32 @load_i32_unordered(ptr addrspace(1) %ptr) {
+  %val = load atomic i32, ptr addrspace(1) %ptr unordered, align 4
----------------
jmmartinez wrote:

This is also missing tests that set `syncscope`

https://github.com/llvm/llvm-project/pull/185628