228012 2021-07-15 18:06:15 -0700 REGRESSION(ANGLE+METAL): WebGL2 content low frame rate 2021-07-22 17:56:59 -0700 1 1 1 Unclassified WebKit WebGL Safari Technology Preview iPhone / iPad iOS 14 RESOLVED FIXED https://noclip.website/#smg/DarkRoomGalaxy;ShareData=AZju&UDsRrT3IyH97Uy9=3k]%5eQY7VbUi%29BDT:JHOVvKe=UwLZ+873GrUDmZ%29=2L InRadar P2 Normal --- 227059 228192 1 jstpierre kpiddington dino gsnedders kbr kkinnunen kpiddington webkit-bug-importer oldest_to_newest 1777401 0 jstpierre 2021-07-15 18:06:15 -0700 URL: https://noclip.website/#smg/DarkRoomGalaxy;ShareData=AZju&UDsRrT3IyH97Uy9=3k]%5eQY7VbUi%29BDT:JHOVvKe=UwLZ+873GrUDmZ%29=2L According to a user of mine, this runs fine on iOS 14 (60 fps), but now runs poorly on iOS 15 Beta (10 fps). The user was on an iPhone 12 Pro. 1777405 1 jstpierre 2021-07-15 18:32:15 -0700 According to the same user: 1. The low performance happens in many different scenes, not just the link provided. 2. The performance seems to act strangely -- e.g. the more things that appear on the screen, the *faster* the application seems to be. 3. This *only* happens in portrait mode. Landscape mode performs just fine. 1777483 2 gsnedders 2021-07-16 03:30:51 -0700 Performance is fine with the Metal backend disabled via experimental features, so regression is from enabling it. To note: at least in the above case, this isn't per-se a regression in iOS 15, given WebGL 2 wasn't enabled in iOS 14. Might affect other WebGL content, however. 1777484 3 webkit-bug-importer 2021-07-16 03:31:03 -0700 <rdar://problem/80677758> 1777551 4 kbr 2021-07-16 11:46:17 -0700 This might have been addressed by Bug 227059. Kyle or other Apple engineers, possible for you to test on iOS with top-of-tree? 1777557 5 kpiddington 2021-07-16 12:15:18 -0700 Repo'd, I'm seeing most of the slowdown happening with a 'bufferSubData' call. As Jasper reports, landscape mode performs just fine, oddly enough. 1777586 6 kpiddington 2021-07-16 14:00:53 -0700 Unfortunately, this has not been addressed by Bug 227059. I'm looking into it, however. Updates soon :). I suspect this is some sort of run-ahead issue, but now we seem to be falling behind rather than running ahead. 1777632 7 jstpierre 2021-07-16 15:46:01 -0700 The bufferSubData is likely uploading my giant UBO for the frame. Either that or vertex data for some dynamic draws. All uploads happening every frame should be marked GL_DYNAMIC_DRAW (which I believe should be correct? Can never remember the difference between that and GL_STREAM_DRAW), and I try to keep UBO data limited to 65536-sized buffer pages. If there's anything I'm doing wrong on my side, please let me know. 1777672 8 433720 kpiddington 2021-07-16 17:29:32 -0700 Created attachment 433720 Patch 1777687 9 kpiddington 2021-07-16 18:17:47 -0700 (In reply to Jasper St. Pierre from comment #7) > The bufferSubData is likely uploading my giant UBO for the frame. Either > that or vertex data for some dynamic draws. All uploads happening every > frame should be marked GL_DYNAMIC_DRAW (which I believe should be correct? > Can never remember the difference between that and GL_STREAM_DRAW), and I > try to keep UBO data limited to 65536-sized buffer pages. > > If there's anything I'm doing wrong on my side, please let me know. While I'm working on a patch, the best thing you can do for performance is limit the number of draw calls you submit. Try to keep submissions around 60 FPS, and you should see a performance improvement. 1777691 10 jstpierre 2021-07-16 18:24:59 -0700 Hmm, I do try to rely on requestAnimationFrame to manage my scheduling. I should be submitting around 200 draw calls in that link provided, which should hopefully be on the low side. It's possible there's a bug in my animation scheduling. If that doesn't match what you're seeing, I'd be happy to hear what might be going wrong, or how else I can rate-limit my draw calls submission. 1777766 11 433720 dino 2021-07-18 12:09:26 -0700 Comment on attachment 433720 Patch Do we need to do this for macOS too? 1778063 12 kpiddington 2021-07-19 17:29:47 -0700 (In reply to Jasper St. Pierre from comment #10) > Hmm, I do try to rely on requestAnimationFrame to manage my scheduling. I > should be submitting around 200 draw calls in that link provided, which > should hopefully be on the low side. It's possible there's a bug in my > animation scheduling. > > If that doesn't match what you're seeing, I'd be happy to hear what might be > going wrong, or how else I can rate-limit my draw calls submission. We definitely have a problem with our fencing code, though I'm able to address that now. That patch will One thing I'm interested in is your render target size when the iPhone is in portrait vs Landscape mode. I suspect we're getting a significantly larger render target. This patch will address our bugs in the fencing code, so we don't run-ahead too much. 1778064 13 433837 kpiddington 2021-07-19 17:32:08 -0700 Created attachment 433837 Patch 1778067 14 433839 kpiddington 2021-07-19 17:41:07 -0700 Created attachment 433839 Patch 1778214 15 kpiddington 2021-07-20 11:27:28 -0700 (In reply to Kyle Piddington from comment #12) > (In reply to Jasper St. Pierre from comment #10) > > Hmm, I do try to rely on requestAnimationFrame to manage my scheduling. I > > should be submitting around 200 draw calls in that link provided, which > > should hopefully be on the low side. It's possible there's a bug in my > > animation scheduling. > > > > If that doesn't match what you're seeing, I'd be happy to hear what might be > > going wrong, or how else I can rate-limit my draw calls submission. > > We definitely have a problem with our fencing code, though I'm able to > address that now. That patch will > > One thing I'm interested in is your render target size when the iPhone is in > portrait vs Landscape mode. I suspect we're getting a significantly larger > render target. > > This patch will address our bugs in the fencing code, so we don't run-ahead > too much. One last thing: Locally, I'm seeing more draw detail with the Metal backend. For example, in the Dark Room Galaxy sample, enabling and disabling the metal backend toggles whether or not the floor and floating pillars are textured. Could that account for the performance change? 1778224 16 433839 dino 2021-07-20 11:48:34 -0700 Comment on attachment 433839 Patch This looks good but I think it might be in danger of breaking on future OS releases. 1778285 17 kbr 2021-07-20 14:52:09 -0700 Thanks Kyle for tracking down this tricky sandboxing issue that was affecting the fence implementation! 1778314 18 ews-feeder 2021-07-20 16:38:01 -0700 Committed r280114 (239831@main): <https://commits.webkit.org/239831@main> All reviewed patches have been landed. Closing bug and clearing flags on attachment 433839. 1778913 19 jstpierre 2021-07-22 17:11:48 -0700 Should I open a separate bug for the landscape vs. portrait issue, or should this be reopened to investigate that? 1778916 20 kpiddington 2021-07-22 17:39:34 -0700 (In reply to Jasper St. Pierre from comment #19) > Should I open a separate bug for the landscape vs. portrait issue, or should > this be reopened to investigate that? Let's open a separate bug. Can you confirm with your user that the scenes are identical on iOS14 and iOS 15? In addition, could you check the render target sizes on Landscape vs Portrait? I suspect we're rendering more in Portrait mode. 1778917 21 jstpierre 2021-07-22 17:41:25 -0700 I don't know any way to easily check the render target sizes. Would that be the drawingBufferWidth / drawingBufferHeight on the WebGL render context? I attempt to resize the canvas to fit the window content. 1778922 22 jstpierre 2021-07-22 17:43:49 -0700 Also, I filed https://bugs.webkit.org/show_bug.cgi?id=228208 for the portrait vs. landscape issue. 1778926 23 kbr 2021-07-22 17:56:59 -0700 Yes, drawingBufferWidth/drawingBufferHeight are the real pixel sizes of WebGL's back buffer. 433720 2021-07-16 17:29:32 -0700 2021-07-19 17:32:06 -0700 Patch bug-228012-20210716172932.patch text/plain 1845 kpiddington U3VidmVyc2lvbiBSZXZpc2lvbjogMjgwMDA4CmRpZmYgLS1naXQgYS9Tb3VyY2UvV2ViS2l0L0No YW5nZUxvZyBiL1NvdXJjZS9XZWJLaXQvQ2hhbmdlTG9nCmluZGV4IGM5ODliNjBkZDNlNmFlNWE5 MTEzZDEyMDdmN2Q1MmI3NzZjODYwMDUuLmVkY2JhODM2MTJmOTI3ZjgyMzVmZmZkZjU5YjUwOTk4 NmYxMWE4ZjMgMTAwNjQ0Ci0tLSBhL1NvdXJjZS9XZWJLaXQvQ2hhbmdlTG9nCisrKyBiL1NvdXJj ZS9XZWJLaXQvQ2hhbmdlTG9nCkBAIC0xLDMgKzEsMjEgQEAKKzIwMjEtMDctMTYgIEt5bGUgUGlk ZGluZ3RvbiAgPGtwaWRkaW5ndG9uQGFwcGxlLmNvbT4KKworICAgICAgICBSRUdSRVNTSU9OKEFO R0xFK01FVEFMKTogV2ViR0wyIGNvbnRlbnQgbG93IGZyYW1lIHJhdGUKKyAgICAgICAgaHR0cHM6 Ly9idWdzLndlYmtpdC5vcmcvc2hvd19idWcuY2dpP2lkPTIyODAxMgorCisgICAgICAgIFdlYkdM J3MgRnJhbWUtcGFjaW5nIGNvZGUgcmVsaWVzIG9uIE1UTFNoYXJlZEV2ZW50cy4gSG93ZXZlciwg dGhlIGV2ZW50IGxpc3RlbmVyIGlzIG5vdCBjcmVhdGVkCisgICAgICAgIGR1ZSB0byBhIHNhbmRi b3ggdmlvbGF0aW9uLiBUaGlzIG1lYW5zIHRoYXQgYW55IHRpbWUgd2UgaW50ZW5kIHRvIHdhaXQg b24gYSBNVExGZW5jZSwgd2UgaW5zdGVhZCAKKyAgICAgICAgd2FpdCBhbGwgZml2ZSBzZWNvbmRz IGZvciBhIGZpdmUgc2Vjb25kIHRpbWVvdXQuIAorCisgICAgICAgIE5vdCBhbGwgYXBwcyBhcmUg aWxsLWJlaGF2ZWQuIEFwcHMgdGhhdCBydW4gYWhlYWQgYXQgYW4gdW5jYXBwZWQgZnJhbWVyYXRl LCBob3dldmVyLCB3aWxsIHJ1biBpbnRvIG91ciBmZW5jaW5nCisgICAgICAgIHF1aWNrbHksIGFu ZCBoYXZlIHRoZWlyIHBlcmZvcm1hbmNlIGRlZ3JhZGUuIAorCisgICAgICAgIFJldmlld2VkIGJ5 IE5PQk9EWSAoT09QUyEpLgorCisgICAgICAgIFRlc3Q6IHJ1biBhdHRhY2hlZCBVUkwsIGVuc3Vy ZSBwZXJmb3JtYW5jZSBkb2VzIG5vdCBkZWdyYWRlLiAKKworICAgICAgICAqIFJlc291cmNlcy9T YW5kYm94UHJvZmlsZXMvaW9zL2NvbS5hcHBsZS5XZWJLaXQuV2ViQ29udGVudC5zYi5pbjoKKwog MjAyMS0wNy0xNiAgV2Vuc29uIEhzaWVoICA8d2Vuc29uX2hzaWVoQGFwcGxlLmNvbT4KIAogICAg ICAgICBSRUdSRVNTSU9OIChyMjc3ODIwKTogQ2FuJ3Qgc2Nyb2xsIHVwIGFuZCBkb3duIHVzaW5n IHRyYWNrcGFkIGluIE1haWwgbWVzc2FnZSB2aWV3ZXIgd2hlbiBpbiBTbGlkZSBPdmVyCmRpZmYg LS1naXQgYS9Tb3VyY2UvV2ViS2l0L1Jlc291cmNlcy9TYW5kYm94UHJvZmlsZXMvaW9zL2NvbS5h cHBsZS5XZWJLaXQuV2ViQ29udGVudC5zYi5pbiBiL1NvdXJjZS9XZWJLaXQvUmVzb3VyY2VzL1Nh bmRib3hQcm9maWxlcy9pb3MvY29tLmFwcGxlLldlYktpdC5XZWJDb250ZW50LnNiLmluCmluZGV4 IGYxNjE0NTU4MWQ4YzJjYjMyNDg3MzNjMWJlYTg5NDkwNWZiNmYzNDcuLjU2MWZjYjE2NWI4OGNh MDlhNDc3OGMyZTBhNWFiNzc0Y2I4MTA5ZWYgMTAwNjQ0Ci0tLSBhL1NvdXJjZS9XZWJLaXQvUmVz b3VyY2VzL1NhbmRib3hQcm9maWxlcy9pb3MvY29tLmFwcGxlLldlYktpdC5XZWJDb250ZW50LnNi LmluCisrKyBiL1NvdXJjZS9XZWJLaXQvUmVzb3VyY2VzL1NhbmRib3hQcm9maWxlcy9pb3MvY29t LmFwcGxlLldlYktpdC5XZWJDb250ZW50LnNiLmluCkBAIC00NTEsNiArNDUxLDcgQEAKICAgICAg ICAgICAgICAgICAgICAgMzUKICAgICAgICAgICAgICAgICAgICAgMzYKICAgICAgICAgICAgICAg ICAgICAgMzgKKyAgICAgICAgICAgICAgICAgICAgNDEKICAgICAgICAgICAgICAgICApCiAgICAg ICAgICAgICApCiAgICAgICAgICkK 433837 2021-07-19 17:32:08 -0700 2021-07-19 17:41:06 -0700 Patch bug-228012-20210719173207.patch text/plain 1880 kpiddington U3VidmVyc2lvbiBSZXZpc2lvbjogMjgwMDQ1CmRpZmYgLS1naXQgYS9Tb3VyY2UvV2ViS2l0L0No YW5nZUxvZyBiL1NvdXJjZS9XZWJLaXQvQ2hhbmdlTG9nCmluZGV4IDFlZDA5YmI2M2JjYjAwYzEw ODRmMTM0OGMxZDcyYTljYTI1NDRjY2YuLjNkN2QyNjQ2ZTBkNDcxNTQ1Njk0MTkxMDUzNjcyNDhk NWNlNmQ1YjUgMTAwNjQ0Ci0tLSBhL1NvdXJjZS9XZWJLaXQvQ2hhbmdlTG9nCisrKyBiL1NvdXJj ZS9XZWJLaXQvQ2hhbmdlTG9nCkBAIC0xLDMgKzEsMTQgQEAKKzIwMjEtMDctMTkgIEt5bGUgUGlk ZGluZ3RvbiAgPGtwaWRkaW5ndG9uQGFwcGxlLmNvbT4KKworICAgICAgICBSRUdSRVNTSU9OKEFO R0xFK01FVEFMKTogV2ViR0wyIGNvbnRlbnQgbG93IGZyYW1lIHJhdGUKKyAgICAgICAgaHR0cHM6 Ly9idWdzLndlYmtpdC5vcmcvc2hvd19idWcuY2dpP2lkPTIyODAxMgorCisgICAgICAgIFJldmll d2VkIGJ5IERlYW4gSmFja3Nvbi4KKyAgICAgICAgCisgICAgICAgIEFkZCBBZGRpdGlvbmFsIHJ1 bGVzLCBhbGxvd2luZyBzaGFyZWQgZXZlbnRzIHRvIHdvcmsgaW4gV2Via2l0CisKKyAgICAgICAg KiBSZXNvdXJjZXMvU2FuZGJveFByb2ZpbGVzL2lvcy9jb20uYXBwbGUuV2ViS2l0LldlYkNvbnRl bnQuc2IuaW46CisKIDIwMjEtMDctMTkgIFdlbnNvbiBIc2llaCAgPHdlbnNvbl9oc2llaEBhcHBs ZS5jb20+CiAKICAgICAgICAgUmV2ZXJ0IGEgZGVidWcgYXNzZXJ0aW9uIHRoYXQgd2FzIGFkZGVk IGluIHIyODAwMTkKZGlmZiAtLWdpdCBhL1NvdXJjZS9XZWJLaXQvUmVzb3VyY2VzL1NhbmRib3hQ cm9maWxlcy9pb3MvY29tLmFwcGxlLldlYktpdC5XZWJDb250ZW50LnNiLmluIGIvU291cmNlL1dl YktpdC9SZXNvdXJjZXMvU2FuZGJveFByb2ZpbGVzL2lvcy9jb20uYXBwbGUuV2ViS2l0LldlYkNv bnRlbnQuc2IuaW4KaW5kZXggZjE2MTQ1NTgxZDhjMmNiMzI0ODczM2MxYmVhODk0OTA1ZmI2ZjM0 Ny4uZThiNTJhNjY3MzM4NDU2MTVhMzRhOTUyYzk5MzA4ZTQwMDJjZjY2ZSAxMDA2NDQKLS0tIGEv U291cmNlL1dlYktpdC9SZXNvdXJjZXMvU2FuZGJveFByb2ZpbGVzL2lvcy9jb20uYXBwbGUuV2Vi S2l0LldlYkNvbnRlbnQuc2IuaW4KKysrIGIvU291cmNlL1dlYktpdC9SZXNvdXJjZXMvU2FuZGJv eFByb2ZpbGVzL2lvcy9jb20uYXBwbGUuV2ViS2l0LldlYkNvbnRlbnQuc2IuaW4KQEAgLTQyNSw2 ICs0MjUsMTIgQEAKICAgICAgICAgICAgIChkZW55ICh3aXRoIHRlbGVtZXRyeSkKICAgICAgICAg ICAgICAgICBpb2tpdC1hc3luYy1leHRlcm5hbC1tZXRob2QKICAgICAgICAgICAgICAgICBpb2tp dC1leHRlcm5hbC10cmFwKQorICAgICAgICAgICAgKGFsbG93IGlva2l0LWFzeW5jLWV4dGVybmFs LW1ldGhvZAorICAgICAgICAgICAgICAgICAgICAoaW9raXQtbWV0aG9kLW51bWJlcgorICAgICAg ICAgICAgICAgICAgICAgICAgNDAKKyAgICAgICAgICAgICAgICAgICAgICAgIDQzCisgICAgICAg ICAgICAgICAgICAgICkKKyAgICAgICAgICAgICkKICAgICAgICAgICAgIChkZW55ICh3aXRoIHRl bGVtZXRyeSkgKHdpdGggbWVzc2FnZSAiSU9TdXJmYWNlUm9vdFVzZXJDbGllbnQiKQogICAgICAg ICAgICAgICAgIGlva2l0LWV4dGVybmFsLW1ldGhvZCkKICAgICAgICAgICAgIChhbGxvdyBpb2tp dC1leHRlcm5hbC1tZXRob2QKQEAgLTQ1MSw2ICs0NTcsOSBAQAogICAgICAgICAgICAgICAgICAg ICAzNQogICAgICAgICAgICAgICAgICAgICAzNgogICAgICAgICAgICAgICAgICAgICAzOAorICAg ICAgICAgICAgICAgICAgICAzOQorICAgICAgICAgICAgICAgICAgICA0MAorICAgICAgICAgICAg ICAgICAgICA0MQogICAgICAgICAgICAgICAgICkKICAgICAgICAgICAgICkKICAgICAgICAgKQo= 433839 2021-07-19 17:41:07 -0700 2021-07-20 16:38:02 -0700 Patch bug-228012-20210719174107.patch text/plain 1852 kpiddington U3VidmVyc2lvbiBSZXZpc2lvbjogMjgwMDQ1CmRpZmYgLS1naXQgYS9Tb3VyY2UvV2ViS2l0L0No YW5nZUxvZyBiL1NvdXJjZS9XZWJLaXQvQ2hhbmdlTG9nCmluZGV4IDFlZDA5YmI2M2JjYjAwYzEw ODRmMTM0OGMxZDcyYTljYTI1NDRjY2YuLjNkN2QyNjQ2ZTBkNDcxNTQ1Njk0MTkxMDUzNjcyNDhk NWNlNmQ1YjUgMTAwNjQ0Ci0tLSBhL1NvdXJjZS9XZWJLaXQvQ2hhbmdlTG9nCisrKyBiL1NvdXJj ZS9XZWJLaXQvQ2hhbmdlTG9nCkBAIC0xLDMgKzEsMTQgQEAKKzIwMjEtMDctMTkgIEt5bGUgUGlk ZGluZ3RvbiAgPGtwaWRkaW5ndG9uQGFwcGxlLmNvbT4KKworICAgICAgICBSRUdSRVNTSU9OKEFO R0xFK01FVEFMKTogV2ViR0wyIGNvbnRlbnQgbG93IGZyYW1lIHJhdGUKKyAgICAgICAgaHR0cHM6 Ly9idWdzLndlYmtpdC5vcmcvc2hvd19idWcuY2dpP2lkPTIyODAxMgorCisgICAgICAgIFJldmll d2VkIGJ5IERlYW4gSmFja3Nvbi4KKyAgICAgICAgCisgICAgICAgIEFkZCBBZGRpdGlvbmFsIHJ1 bGVzLCBhbGxvd2luZyBzaGFyZWQgZXZlbnRzIHRvIHdvcmsgaW4gV2Via2l0CisKKyAgICAgICAg KiBSZXNvdXJjZXMvU2FuZGJveFByb2ZpbGVzL2lvcy9jb20uYXBwbGUuV2ViS2l0LldlYkNvbnRl bnQuc2IuaW46CisKIDIwMjEtMDctMTkgIFdlbnNvbiBIc2llaCAgPHdlbnNvbl9oc2llaEBhcHBs ZS5jb20+CiAKICAgICAgICAgUmV2ZXJ0IGEgZGVidWcgYXNzZXJ0aW9uIHRoYXQgd2FzIGFkZGVk IGluIHIyODAwMTkKZGlmZiAtLWdpdCBhL1NvdXJjZS9XZWJLaXQvUmVzb3VyY2VzL1NhbmRib3hQ cm9maWxlcy9pb3MvY29tLmFwcGxlLldlYktpdC5XZWJDb250ZW50LnNiLmluIGIvU291cmNlL1dl YktpdC9SZXNvdXJjZXMvU2FuZGJveFByb2ZpbGVzL2lvcy9jb20uYXBwbGUuV2ViS2l0LldlYkNv bnRlbnQuc2IuaW4KaW5kZXggZjE2MTQ1NTgxZDhjMmNiMzI0ODczM2MxYmVhODk0OTA1ZmI2ZjM0 Ny4uOTRkMGQyNDExNzc0MTZhMTQyMGMzNmNhNTkxZjdkNjdkNDExYzAwZiAxMDA2NDQKLS0tIGEv U291cmNlL1dlYktpdC9SZXNvdXJjZXMvU2FuZGJveFByb2ZpbGVzL2lvcy9jb20uYXBwbGUuV2Vi S2l0LldlYkNvbnRlbnQuc2IuaW4KKysrIGIvU291cmNlL1dlYktpdC9SZXNvdXJjZXMvU2FuZGJv eFByb2ZpbGVzL2lvcy9jb20uYXBwbGUuV2ViS2l0LldlYkNvbnRlbnQuc2IuaW4KQEAgLTQyNSw2 ICs0MjUsMTEgQEAKICAgICAgICAgICAgIChkZW55ICh3aXRoIHRlbGVtZXRyeSkKICAgICAgICAg ICAgICAgICBpb2tpdC1hc3luYy1leHRlcm5hbC1tZXRob2QKICAgICAgICAgICAgICAgICBpb2tp dC1leHRlcm5hbC10cmFwKQorICAgICAgICAgICAgKGFsbG93IGlva2l0LWFzeW5jLWV4dGVybmFs LW1ldGhvZAorICAgICAgICAgICAgICAgICAgICAoaW9raXQtbWV0aG9kLW51bWJlcgorICAgICAg ICAgICAgICAgICAgICAgICAgNDAKKyAgICAgICAgICAgICAgICAgICAgKQorICAgICAgICAgICAg KQogICAgICAgICAgICAgKGRlbnkgKHdpdGggdGVsZW1ldHJ5KSAod2l0aCBtZXNzYWdlICJJT1N1 cmZhY2VSb290VXNlckNsaWVudCIpCiAgICAgICAgICAgICAgICAgaW9raXQtZXh0ZXJuYWwtbWV0 aG9kKQogICAgICAgICAgICAgKGFsbG93IGlva2l0LWV4dGVybmFsLW1ldGhvZApAQCAtNDUxLDYg KzQ1Niw5IEBACiAgICAgICAgICAgICAgICAgICAgIDM1CiAgICAgICAgICAgICAgICAgICAgIDM2 CiAgICAgICAgICAgICAgICAgICAgIDM4CisgICAgICAgICAgICAgICAgICAgIDM5CisgICAgICAg ICAgICAgICAgICAgIDQwCisgICAgICAgICAgICAgICAgICAgIDQxCiAgICAgICAgICAgICAgICAg KQogICAgICAgICAgICAgKQogICAgICAgICApCg==