<html> <head> <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> </head> <body text="#000000" bgcolor="#FFFFFF"> <div class="moz-cite-prefix">On 08/08/18 21:55, Rafael Antognolli wrote: </div> <blockquote type="cite" cite="mid:20180808205549.xga3v57vcya7lnb4@rantogno-mobl4.amr.corp.intel.com"> <blockquote type="cite" style="color: #000000;"> <pre wrap="">t @@ -790,6 +892,7 @@ aub_viewer_render_batch(struct aub_viewer_decode_ctx *ctx, for (unsigned i = 0; i < ARRAY_SIZE(info_decoders); i++) { if (strcmp(inst_name, info_decoders[i].cmd_name) == 0) { + ctx->stage = info_decoders[i].stage; info_decoders[i].decode(ctx, inst, p); break; </pre> </blockquote> <pre wrap="">Looks like you run the info_decoders before the display decoders (even though some of them decode the same type of instructions) because you want to first do a pass storing all the information required for the display_decoders, right? And in this case, you also want to store the stage we are at, so when we run the info_decoders that information is available. Also, is there a chance we use these same info decoders for storing other information than urb related stuff? If so, maybe we should just call them handle_3ds_gs, handle_3ds_constant_vs, etc... Or something along those lines. But of course we could just rename them in the future. </pre> </blockquote> The UI code (decode_* functions) run conditionally to the instruction having their fields visible (expanded by clicking on them). info_decoders runs unconditionally so that we can have a consistent accumulated state that doesn't depend on the UI state. For example, you might not want to display urb instructions & their content. Yet you still need to parse their content so that URB state can be displayed on a latter instruction (like 3DPRIMITIVE). This is poorly named indeed and could be done more efficiently. This can be done later (once you get the hang of ImGui. It's really fun!). - Lionel </body> </html>