agx: Use a transfer graph for parallel copies

Lifted from ir3. Algorithm is the same; the data structures and interface are
lightly modified to decouple from ir3's IR.

Sequentializing parallel copies after RA is tricky. ir3's implementation works
well enough, so I use that one.

Original implementation by Connor Abbott.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16268>

src/asahi/compiler/agx_compiler.h

@@ -674,6 +674,23 @@ void agx_pack_binary(agx_context *ctx, struct util_dynarray *emission);
 
 unsigned agx_write_registers(agx_instr *I, unsigned d);
 
+struct agx_copy {
+   /* Base register destination of the copy */
+   unsigned dest;
+
+   /* Base register source of the copy */
+   unsigned src;
+
+   /* Size of the copy */
+   enum agx_size size;
+
+   /* Whether the copy has been handled. Callers must leave to false. */
+   bool done;
+};
+
+void
+agx_emit_parallel_copies(agx_builder *b, struct agx_copy *copies, unsigned n);
+
 void agx_compute_liveness(agx_context *ctx);
 void agx_liveness_ins_update(BITSET_WORD *live, agx_instr *I);