+void cl_ensure_space(struct vc4_cl *cl, uint32_t size);
+
+#define cl_packet_header(packet) V3D21_ ## packet ## _header
+#define cl_packet_length(packet) V3D21_ ## packet ## _length
+#define cl_packet_pack(packet) V3D21_ ## packet ## _pack
+#define cl_packet_struct(packet) V3D21_ ## packet
+
+static inline void *
+cl_get_emit_space(struct vc4_cl_out **cl, size_t size)
+{
+ void *addr = *cl;
+ cl_advance(cl, size);
+ return addr;
+}
+
+/* Macro for setting up an emit of a CL struct. A temporary unpacked struct
+ * is created, which you get to set fields in of the form:
+ *
+ * cl_emit(bcl, FLAT_SHADE_FLAGS, flags) {
+ * .flags.flat_shade_flags = 1 << 2,
+ * }
+ *
+ * or default values only can be emitted with just:
+ *
+ * cl_emit(bcl, FLAT_SHADE_FLAGS, flags);
+ *
+ * The trick here is that we make a for loop that will execute the body
+ * (either the block or the ';' after the macro invocation) exactly once.
+ * Also, *dst is actually of the wrong type, it's the
+ * uint8_t[cl_packet_length()] in the CL, not a cl_packet_struct(packet).
+ */
+#define cl_emit(cl, packet, name) \
+ for (struct cl_packet_struct(packet) name = { \
+ cl_packet_header(packet) \
+ }, \
+ *_loop_terminate = &name; \
+ __builtin_expect(_loop_terminate != NULL, 1); \
+ ({ \
+ struct vc4_cl_out *cl_out = cl_start(cl); \
+ cl_packet_pack(packet)(cl, (uint8_t *)cl_out, &name); \
+ VG(VALGRIND_CHECK_MEM_IS_DEFINED(cl_out, \
+ cl_packet_length(packet))); \
+ cl_advance(&cl_out, cl_packet_length(packet)); \
+ cl_end(cl, cl_out); \
+ _loop_terminate = NULL; \
+ })) \
+
+#define cl_emit_prepacked(cl, packet) do { \
+ memcpy((cl)->next, packet, sizeof(*packet)); \
+ cl_advance(&(cl)->next, sizeof(*packet)); \
+} while (0)
+
+/**
+ * Helper function called by the XML-generated pack functions for filling in
+ * an address field in shader records.
+ *
+ * Relocations for shader recs and texturing involve the packet (or uniforms
+ * stream) being preceded by the handles to the BOs, and the offset within the
+ * BO being in the stream (the output of this function).
+ */
+static inline void
+cl_pack_emit_reloc(struct vc4_cl *cl, const struct vc4_cl_reloc *reloc)
+{
+ *(uint32_t *)cl->reloc_next = vc4_gem_hindex(cl->job, reloc->bo);
+ cl_advance(&cl->reloc_next, 4);
+
+#ifndef NDEBUG
+ cl->reloc_count--;
+#endif