nir/opt_intrinsic: fix sample mask opt with demote

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

Fixes: d3ce8a7f6b ("nir: optimize gl_SampleMaskIn to gl_HelperInvocation for radeonsi when possible")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32327>
(cherry picked from commit 3f26e9ca19)

.pick_status.json

@@ -2914,7 +2914,7 @@
         "description": "nir/opt_intrinsic: fix sample mask opt with demote",
         "nominated": true,
         "nomination_type": 2,
-        "resolution": 0,
+        "resolution": 1,
         "main_sha": null,
         "because_sha": "d3ce8a7f6b93e893929b81f4d6605a2a968730a1",
         "notes": null

src/compiler/nir/nir_opt_intrinsics.c

@@ -329,6 +329,20 @@ opt_intrinsics_intrin(nir_builder *b, nir_intrinsic_instr *intrin,
             if (!nir_scalar_is_const(other) || nir_scalar_as_uint(other))
                continue;
 
+            nir_cf_node *cf_node = &intrin->instr.block->cf_node;
+            while (cf_node->parent)
+               cf_node = cf_node->parent;
+
+            nir_function_impl *func_impl = nir_cf_node_as_function(cf_node);
+
+            /* We need to insert load_helper before any demote,
+             * which is only possible in the entry point function
+             */
+            if (func_impl != nir_shader_get_entrypoint(b->shader))
+               break;
+
+            b->cursor = nir_before_impl(func_impl);
+
             nir_def *new_expr = nir_load_helper_invocation(b, 1);
 
             if (alu->op == nir_op_ine)