radv: apply the indexing workaround for atomic buffer operations on GFX9

Because the new raw/struct intrinsics are buggy with LLVM 8
(they weren't marked as source of divergence), we fallback to the
old instrinsics for atomic buffer operations only. This means we need
to apply the indexing workaround for GFX9. The load/store
operations still use the new LLVM 8 intrinsics.

The fact that we need another workaround is painful but we should
be able to clean up that a bit once LLVM 7 support will be dropped.

This fixes a GPU hang with AC Odyssey and some rendering problems
with Nioh.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=110573
Fixes: 31164cf5f7 ("ac/nir: only use the new raw/struct image atomic intrinsics with LLVM 9+")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

src/amd/vulkan/radv_nir_to_llvm.c

@@ -3687,6 +3687,12 @@ LLVMModuleRef ac_translate_nir_to_llvm(struct ac_llvm_compiler *ac_llvm,
 	ctx.abi.clamp_shadow_reference = false;
 	ctx.abi.gfx9_stride_size_workaround = ctx.ac.chip_class == GFX9 && HAVE_LLVM < 0x800;
 
+	/* Because the new raw/struct atomic intrinsics are buggy with LLVM 8,
+	 * we fallback to the old intrinsics for atomic buffer image operations
+	 * and thus we need to apply the indexing workaround...
+	 */
+	ctx.abi.gfx9_stride_size_workaround_for_atomic = ctx.ac.chip_class == GFX9 && HAVE_LLVM < 0x900;
+
 	if (shader_count >= 2)
 		ac_init_exec_full_mask(&ctx.ac);