[clang] [llvm] [clang][amdgpu] Add builtins for raw/struct buffer lds load (PR #137678)

[Jon Chesterfield via cfe-commits]

================
@@ -0,0 +1,12 @@
+// RUN: %clang_cc1 -triple amdgcn-unknown-unknown -target-cpu tahiti -S -verify -o - %s
+// RUN: %clang_cc1 -triple amdgcn-unknown-unknown -target-cpu bonaire -S -verify -o - %s
+// RUN: %clang_cc1 -triple amdgcn-unknown-unknown -target-cpu carrizo -S -verify -o - %s
+// RUN: %clang_cc1 -triple amdgcn-unknown-unknown -target-cpu gfx1100 -S -verify -o - %s
+// RUN: %clang_cc1 -triple amdgcn-unknown-unknown -target-cpu gfx1200 -S -verify -o - %s
+// REQUIRES: amdgpu-registered-target
+
+typedef unsigned int v4u32 __attribute__((ext_vector_type(4)));
+
+void test_amdgcn_struct_buffer_load_lds(v4u32 rsrc, __local void* lds, int index, int offset, int soffset, int x) {
----------------
JonChesterfield wrote:

I'd love to say no, since they're almost identical and share the first N lines, but if I put them in a single file clang bails on the first error and thus the lit test fails (as the later three diagnostics are not emitted). I don't understand that behaviour but after an hour or so iterating random guesswork and looking for hints in other tests I thought I'd post the review like this. Either it ships or someone knows how to make clang+lit do the sane thing here.

https://github.com/llvm/llvm-project/pull/137678