cinera_handmade.network/cmuratori/hero/code/code127.hmml

[video member=cmuratori stream_platform=twitch stream_username=handmade_hero project=code title="Aligning Rendering Memory" vod_platform=youtube id=blcNbU70I9o annotator=dspecht annotator=Miblo]
[1:23][Set the stage for this week]
[2:51][Demo the problem with the tile size]
[4:04][Confirm that _mm_sfence is unnecessary]
[5:46][Blackboard: "Write-Combining Memory"]
[13:43][Blackboard: Tiles and Alignment]
[21:02][Suggest rounding TileWidth and TileHeight to the nearest 4 and aligning the tiles to the 4-byte boundary]
[21:35][Think some more, and then unthink it]
[22:11][handmade_render_group.cpp: Ensure that we always get aligned tiles]
[23:27][Blackboard: Calculating and filling the correct number of 4-pixel units]
[26:10][Blackboard: The destination buffer must always allow us to overwrite it by a certain amount]
[29:49][Assert that OutputTarget->Memory is aligned to 16-bytes]
[31:45][Round TileWidth up to the nearest 4]
[33:13][See what numbers we're getting]
[34:14][Ensure that FinalTileWidth accounts for the fact that it'll be smaller]
[34:54][Remove FinalTileWidth and limit ClipRect.MaxX to the OutputTarget->Width]
[35:23][Remove the ClipRect adjustments]
[36:17][Run single-threaded for the moment]
[37:06][Moment of realisation: We're only clipping to the end, but don't handle clipping at the beginning]
[37:34][Blackboard: The clipping problem]
[41:58][Align the MinX and MaxX]
[47:08][Blackboard: Setting the EndClipMasks]
[48:24][Figure out when to use the EndClipMask]
[50:01][Compile and see what's happening]
[50:30][Reverse the sense of the test and run it again]
[50:55][See if we're always aligned]
[51:16][Compile in -O2 and turn on multi-threading to ensure everything's still kosher]
[52:02][win32_handmade.cpp: Double-check that the platform code is allocating the memory aligned]
[53:20][Align Buffer->Pitch to 16-bytes]
[54:10][handmade_platform.h: Introduce Align16]
[55:13][Everything looks fine at 1920x1080]
[56:13][handmade.cpp: Setup the PlatformAddEntry and PlatformCompleteAllWork at the beginning]
[58:00][Q&A][:speech]
[58:43][@robotchocolatedino][The bitmap memory size calculation squares BytesPerPixel which allocates more memory than necessary]
[58:50][win32_handmade.cpp: Remove the multiplication by BytesPerPixel]
[59:19][@ttbjm][Can you test if weird resolutions actually are working?]
[1:00:33][handmade_render_group.cpp: Ensure that it goes all the way to the end if this is the last tile]
[1:07:06][handmade.cpp: Hard-set the game's DrawBuffer dimensions to test our support of arbitrary resolutions]
[1:08:28][@braincruser][Will you check for Cache Aliasing? With this much alignment hitting cache aliasing is much easier]
[1:08:51][@braincruser][Cache Associativity aliasing...]
[1:09:27][@flyingwafflenyc][What about the extra pixel around the edges to deal with bilinear filtering? Will you get rid of that?]
[1:10:33][Compile with that #if 0'd out and glimpse into the future of 4K art]
[1:11:50][@pragmascrypt][Will you keep the rendering of the two lines on separate hyperthreads?]
[1:13:33][@terrorscout][So this game is going to be a 4K Zelda 1-style game?]
[1:14:43][Assess our progress and glimpse into the future of tightening up the renderer]
[1:16:38][Shut things down here][:speech]
[/video]