wrenczes@gmail.com ba07ca2334
Improvement on my last change for reducing the formatted text caching: chopped it out altogether. I ran some profiling to see how much time on the psp it took to process the formatting on the description text (ie word wrapping it into several lines to fit on a card), and it ends up taking an average of 0.14 ms per card. For context, 60 fps is 16 ms. So clearly the formatting time is not the bottleneck here, and we gain no real performance caching the text, but lose memory due to pooling. So I cut it out entirely of the base class and we format on the fly during renders. ...
I also profiled this after the fact, as we have an open bug on poor drawing performance on psp when in text mode - it turns out that in the AlternateRender() function, my numbers look like this: 10.2 seconds were spent in JBLFont::DrawString(); 0.5 seconds were spent in my new helper FormatText() call, so we know where the perf hit is now.
I compared before & after on the psp with this mod, and the difference really isn't perceptible. (It's still juddery, but no worse than before.) I'll look at the DrawString() call next to see if we can make it any faster, although at first glance it looks like a pain.
2011-04-23 09:54:19 +00:00
2011-01-21 18:01:14 +00:00
2011-04-10 00:11:03 +00:00
2011-04-12 20:54:34 +00:00
2011-03-13 21:19:02 +00:00
2011-04-20 21:07:24 +00:00
2011-01-21 18:01:14 +00:00
2011-04-21 15:16:22 +00:00
2011-04-22 13:12:36 +00:00
2011-02-02 18:22:08 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
2011-04-21 03:21:15 +00:00
2011-04-23 09:54:19 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
2011-04-22 13:12:36 +00:00
2011-02-08 15:12:14 +00:00
2011-04-20 07:50:00 +00:00
Resuming on my threading support work with the card caching mechanism. This change unfortunately touches quite a few files, but I needed to get it out of the way before things got out of hand: one significant hurdle is the assumed lifetime of a JQuad pointer. In a single threaded model, the life time of the pointer is clear: you fetch it into the cache, the cache makes room, you use the pointer immediately. In a multithreaded context however, it's unsafe, as the drawing thread can request a few JQuads, and the cache operating on a separate thread can potentially bounce a JQuad out of the cache before the draw routine is done using it, which ends up in an access violation when you attempt to draw using an invalidated quad pointer. To prevent this, the bulk of this change swaps out the use of naked JQuad* pointers in the code with a JQuadPtr, which is basically a typedef to a boost shared_ptr<JQuad>.
2011-02-01 10:37:21 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
2011-01-28 06:00:51 +00:00
2011-02-10 17:19:11 +00:00
Resuming on my threading support work with the card caching mechanism. This change unfortunately touches quite a few files, but I needed to get it out of the way before things got out of hand: one significant hurdle is the assumed lifetime of a JQuad pointer. In a single threaded model, the life time of the pointer is clear: you fetch it into the cache, the cache makes room, you use the pointer immediately. In a multithreaded context however, it's unsafe, as the drawing thread can request a few JQuads, and the cache operating on a separate thread can potentially bounce a JQuad out of the cache before the draw routine is done using it, which ends up in an access violation when you attempt to draw using an invalidated quad pointer. To prevent this, the bulk of this change swaps out the use of naked JQuad* pointers in the code with a JQuadPtr, which is basically a typedef to a boost shared_ptr<JQuad>.
2011-02-01 10:37:21 +00:00
2011-01-21 18:01:14 +00:00
2011-04-20 17:51:40 +00:00
2011-02-10 17:19:11 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
2011-04-20 09:28:28 +00:00
2011-04-20 07:50:00 +00:00
2011-02-12 15:06:28 +00:00
2011-04-22 11:17:20 +00:00
2011-01-30 13:06:21 +00:00
2011-01-21 18:01:14 +00:00
2011-04-20 07:50:00 +00:00
2011-04-20 17:51:40 +00:00
2011-02-06 11:35:40 +00:00
2011-01-21 18:01:14 +00:00
Resuming on my threading support work with the card caching mechanism. This change unfortunately touches quite a few files, but I needed to get it out of the way before things got out of hand: one significant hurdle is the assumed lifetime of a JQuad pointer. In a single threaded model, the life time of the pointer is clear: you fetch it into the cache, the cache makes room, you use the pointer immediately. In a multithreaded context however, it's unsafe, as the drawing thread can request a few JQuads, and the cache operating on a separate thread can potentially bounce a JQuad out of the cache before the draw routine is done using it, which ends up in an access violation when you attempt to draw using an invalidated quad pointer. To prevent this, the bulk of this change swaps out the use of naked JQuad* pointers in the code with a JQuadPtr, which is basically a typedef to a boost shared_ptr<JQuad>.
2011-02-01 10:37:21 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
Resuming on my threading support work with the card caching mechanism. This change unfortunately touches quite a few files, but I needed to get it out of the way before things got out of hand: one significant hurdle is the assumed lifetime of a JQuad pointer. In a single threaded model, the life time of the pointer is clear: you fetch it into the cache, the cache makes room, you use the pointer immediately. In a multithreaded context however, it's unsafe, as the drawing thread can request a few JQuads, and the cache operating on a separate thread can potentially bounce a JQuad out of the cache before the draw routine is done using it, which ends up in an access violation when you attempt to draw using an invalidated quad pointer. To prevent this, the bulk of this change swaps out the use of naked JQuad* pointers in the code with a JQuadPtr, which is basically a typedef to a boost shared_ptr<JQuad>.
2011-02-01 10:37:21 +00:00
Resuming on my threading support work with the card caching mechanism. This change unfortunately touches quite a few files, but I needed to get it out of the way before things got out of hand: one significant hurdle is the assumed lifetime of a JQuad pointer. In a single threaded model, the life time of the pointer is clear: you fetch it into the cache, the cache makes room, you use the pointer immediately. In a multithreaded context however, it's unsafe, as the drawing thread can request a few JQuads, and the cache operating on a separate thread can potentially bounce a JQuad out of the cache before the draw routine is done using it, which ends up in an access violation when you attempt to draw using an invalidated quad pointer. To prevent this, the bulk of this change swaps out the use of naked JQuad* pointers in the code with a JQuadPtr, which is basically a typedef to a boost shared_ptr<JQuad>.
2011-02-01 10:37:21 +00:00
2011-01-21 18:01:14 +00:00
Resuming on my threading support work with the card caching mechanism. This change unfortunately touches quite a few files, but I needed to get it out of the way before things got out of hand: one significant hurdle is the assumed lifetime of a JQuad pointer. In a single threaded model, the life time of the pointer is clear: you fetch it into the cache, the cache makes room, you use the pointer immediately. In a multithreaded context however, it's unsafe, as the drawing thread can request a few JQuads, and the cache operating on a separate thread can potentially bounce a JQuad out of the cache before the draw routine is done using it, which ends up in an access violation when you attempt to draw using an invalidated quad pointer. To prevent this, the bulk of this change swaps out the use of naked JQuad* pointers in the code with a JQuadPtr, which is basically a typedef to a boost shared_ptr<JQuad>.
2011-02-01 10:37:21 +00:00
2011-01-21 18:01:14 +00:00
2011-01-24 17:58:14 +00:00
2011-01-21 18:01:14 +00:00
2011-04-16 17:00:51 +00:00
2011-02-09 20:42:19 +00:00
2011-01-21 18:01:14 +00:00
2011-04-22 11:17:20 +00:00
2011-04-23 09:54:19 +00:00
2011-04-23 05:16:53 +00:00
2011-02-14 09:35:43 +00:00
2011-04-22 11:17:20 +00:00
2011-01-21 18:01:14 +00:00
2011-03-02 13:41:24 +00:00
2011-01-21 18:01:14 +00:00
2011-04-22 11:17:20 +00:00
2011-01-21 18:01:14 +00:00
2011-03-13 21:19:02 +00:00
2011-04-23 08:43:34 +00:00
2011-04-20 07:50:00 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
2011-04-12 20:54:34 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
2011-02-15 14:17:34 +00:00
2011-01-21 18:01:14 +00:00
2011-04-20 07:50:00 +00:00
2011-01-21 18:01:14 +00:00
2011-01-28 06:00:51 +00:00
2011-03-02 21:33:42 +00:00
Resuming on my threading support work with the card caching mechanism. This change unfortunately touches quite a few files, but I needed to get it out of the way before things got out of hand: one significant hurdle is the assumed lifetime of a JQuad pointer. In a single threaded model, the life time of the pointer is clear: you fetch it into the cache, the cache makes room, you use the pointer immediately. In a multithreaded context however, it's unsafe, as the drawing thread can request a few JQuads, and the cache operating on a separate thread can potentially bounce a JQuad out of the cache before the draw routine is done using it, which ends up in an access violation when you attempt to draw using an invalidated quad pointer. To prevent this, the bulk of this change swaps out the use of naked JQuad* pointers in the code with a JQuadPtr, which is basically a typedef to a boost shared_ptr<JQuad>.
2011-02-01 10:37:21 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
2011-01-28 06:00:51 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
Performance improvements/cleanup: card->hasType("foo") performs a string lookup for an integer id. Now that our primitives have over 1000 types (1018, I believe is the number I saw while debugging), this is not the cheapest lookup. Over the course of time, a lot of new rules have crept in the code - support for 'aura', 'equipment', and recently 'planeswalker'. While profiling, I saw an inordinate amount of time being used doing map lookups for these strings. Since they're used so frequently, I've added them to the SubTypes enumeration, and swapped out doing the string lookup in favour of using the int ID directly. I also found a few places where we were using hasType("Creature") or hasType("land") when we already had helper functions to do the int lookup, so I swapped these in as appropriate, as well as a couple of places where we had hasType("instant") instead of hasType(Subtypes::TYPE_INSTANT), etc.
2011-04-10 00:07:34 +00:00
2011-01-21 18:01:14 +00:00
2011-04-12 20:54:34 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
2011-01-21 18:01:14 +00:00
2011-01-28 06:00:51 +00:00
2011-04-14 17:06:44 +00:00
2011-01-28 06:00:51 +00:00
2011-01-21 18:01:14 +00:00
2010-02-27 18:49:03 +00:00
2011-01-21 18:01:14 +00:00
2011-04-21 13:16:11 +00:00
2011-04-20 06:27:44 +00:00
Resuming on my threading support work with the card caching mechanism. This change unfortunately touches quite a few files, but I needed to get it out of the way before things got out of hand: one significant hurdle is the assumed lifetime of a JQuad pointer. In a single threaded model, the life time of the pointer is clear: you fetch it into the cache, the cache makes room, you use the pointer immediately. In a multithreaded context however, it's unsafe, as the drawing thread can request a few JQuads, and the cache operating on a separate thread can potentially bounce a JQuad out of the cache before the draw routine is done using it, which ends up in an access violation when you attempt to draw using an invalidated quad pointer. To prevent this, the bulk of this change swaps out the use of naked JQuad* pointers in the code with a JQuadPtr, which is basically a typedef to a boost shared_ptr<JQuad>.
2011-02-01 10:37:21 +00:00
2011-04-12 20:54:34 +00:00
2011-01-30 11:14:36 +00:00
2011-01-21 18:01:14 +00:00
2011-01-31 10:24:06 +00:00
Resuming on my threading support work with the card caching mechanism. This change unfortunately touches quite a few files, but I needed to get it out of the way before things got out of hand: one significant hurdle is the assumed lifetime of a JQuad pointer. In a single threaded model, the life time of the pointer is clear: you fetch it into the cache, the cache makes room, you use the pointer immediately. In a multithreaded context however, it's unsafe, as the drawing thread can request a few JQuads, and the cache operating on a separate thread can potentially bounce a JQuad out of the cache before the draw routine is done using it, which ends up in an access violation when you attempt to draw using an invalidated quad pointer. To prevent this, the bulk of this change swaps out the use of naked JQuad* pointers in the code with a JQuadPtr, which is basically a typedef to a boost shared_ptr<JQuad>.
2011-02-01 10:37:21 +00:00
2011-04-21 22:48:22 +00:00
2011-04-22 13:12:36 +00:00
wrenczes@gmail.com
ba07ca2334