[PATCH v2 0/2] MR6006: dwmapi: Simulate DwmFlush blocking behavior.
The documentation states it's a blocking call. There are reports of it being used for WaitForVBlank-like purposes [1]. Without blocking Softube VST plugin logic breaks and they never update their UIs. [1] https://news.ycombinator.com/item?id=34501612 Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=56935 -- v2: dwmapi: Simulate DwmFlush blocking behavior. https://gitlab.winehq.org/wine/wine/-/merge_requests/6006
From: Grazvydas Ignotas <notasas(a)gmail.com> --- dlls/dwmapi/dwmapi_main.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) diff --git a/dlls/dwmapi/dwmapi_main.c b/dlls/dwmapi/dwmapi_main.c index adc02552ba7..8d4ccbb1fc4 100644 --- a/dlls/dwmapi/dwmapi_main.c +++ b/dlls/dwmapi/dwmapi_main.c @@ -184,9 +184,9 @@ HRESULT WINAPI DwmEnableBlurBehindWindow(HWND hWnd, const DWM_BLURBEHIND *pBlurB */ BOOL WINAPI DwmDefWindowProc(HWND hWnd, UINT Msg, WPARAM wParam, LPARAM lParam, LRESULT *plResult) { - static int i; + static BOOL once; - if (!i++) FIXME("stub\n"); + if (!once++) FIXME("stub\n"); return FALSE; } @@ -273,7 +273,8 @@ static int get_display_frequency(void) HRESULT WINAPI DwmGetCompositionTimingInfo(HWND hwnd, DWM_TIMING_INFO *info) { LARGE_INTEGER performance_frequency, qpc; - static int i, display_frequency; + int display_frequency; + static BOOL once; if (!info) return E_INVALIDARG; @@ -281,7 +282,7 @@ HRESULT WINAPI DwmGetCompositionTimingInfo(HWND hwnd, DWM_TIMING_INFO *info) if (info->cbSize != sizeof(DWM_TIMING_INFO)) return MILERR_MISMATCHED_SIZE; - if(!i++) FIXME("(%p %p)\n", hwnd, info); + if (!once++) FIXME("(%p %p)\n", hwnd, info); memset(info, 0, info->cbSize); info->cbSize = sizeof(DWM_TIMING_INFO); -- GitLab https://gitlab.winehq.org/wine/wine/-/merge_requests/6006
From: Grazvydas Ignotas <notasas(a)gmail.com> The documentation states it's a blocking call. There are reports of it being used for WaitForVBlank-like purposes [1]. Without blocking Softube VST plugin logic breaks and they never update their UIs. [1] https://news.ycombinator.com/item?id=34501612 Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=56935 --- dlls/dwmapi/dwmapi_main.c | 23 +++++++++++++++++++-- dlls/dwmapi/tests/dwmapi.c | 41 ++++++++++++++++++++++++++++++++++++++ 2 files changed, 62 insertions(+), 2 deletions(-) diff --git a/dlls/dwmapi/dwmapi_main.c b/dlls/dwmapi/dwmapi_main.c index 8d4ccbb1fc4..419216534aa 100644 --- a/dlls/dwmapi/dwmapi_main.c +++ b/dlls/dwmapi/dwmapi_main.c @@ -32,6 +32,7 @@ WINE_DEFAULT_DEBUG_CHANNEL(dwmapi); +static int get_display_frequency(void); /********************************************************************** * DwmIsCompositionEnabled (DWMAPI.@) @@ -88,9 +89,26 @@ HRESULT WINAPI DwmGetColorizationColor(DWORD *colorization, BOOL *opaque_blend) */ HRESULT WINAPI DwmFlush(void) { + static volatile LONG last_time; static BOOL once; + DWORD now, interval, last = last_time, target; + int freq; - if (!once++) FIXME("() stub\n"); + if (!once++) FIXME("() semi-stub\n"); + + // simulate the WaitForVBlank-like blocking behavior + freq = get_display_frequency(); + interval = 1000 / freq; + now = GetTickCount(); + if (now - last < interval) + target = last + interval; + else + { + // act as if we were called midway between 2 vsyncs + target = now + interval / 2; + } + Sleep(target - now); + InterlockedCompareExchange(&last_time, target, last); return S_OK; } @@ -262,7 +280,8 @@ static int get_display_frequency(void) } else { - WARN("Failed to query display frequency, returning a fallback value.\n"); + static BOOL once; + if (!once++) WARN("Failed to query display frequency, returning a fallback value.\n"); return 60; } } diff --git a/dlls/dwmapi/tests/dwmapi.c b/dlls/dwmapi/tests/dwmapi.c index a89a1fd705b..20628578f41 100644 --- a/dlls/dwmapi/tests/dwmapi.c +++ b/dlls/dwmapi/tests/dwmapi.c @@ -140,9 +140,50 @@ cleanup: DestroyWindow(hwnd); } +static void test_DwmFlush(void) +{ + LARGE_INTEGER frequency, ts[2]; + int i, result, ms; + DEVMODEA mode; + BOOL enabled; + HRESULT hr; + + enabled = FALSE; + hr = DwmIsCompositionEnabled(&enabled); + ok(hr == S_OK, "Got hr %#lx.\n", hr); + if (!enabled) + { + skip("DWM is disabled.\n"); + return; + } + + memset(&mode, 0, sizeof(mode)); + mode.dmSize = sizeof(mode); + result = EnumDisplaySettingsA(NULL, ENUM_CURRENT_SETTINGS, &mode); + ok(result, "Failed to get display mode %#lx.\n", GetLastError()); + ok(mode.dmDisplayFrequency != 0, "dmDisplayFrequency is 0.\n"); + + result = QueryPerformanceFrequency(&frequency); + ok(result, "Failed to get performance counter frequency.\n"); + + result = QueryPerformanceCounter(&ts[0]); + ok(result, "Failed to read performance counter.\n"); + for (i = 0; i < 2; i++) + { + hr = DwmFlush(); + ok(hr == S_OK, "Got hr %#lx.\n", hr); + } + result = QueryPerformanceCounter(&ts[1]); + ok(result, "Failed to read performance counter.\n"); + ms = (ts[1].QuadPart - ts[0].QuadPart) * 1000 / frequency.QuadPart; + ok(ms >= 1000 / mode.dmDisplayFrequency, + "DwmFlush() took %dms with dmDisplayFrequency %ld.\n", ms, mode.dmDisplayFrequency); +} + START_TEST(dwmapi) { test_DwmIsCompositionEnabled(); test_DwmGetCompositionTimingInfo(); test_DWMWA_EXTENDED_FRAME_BOUNDS(); + test_DwmFlush(); } -- GitLab https://gitlab.winehq.org/wine/wine/-/merge_requests/6006
Hi, It looks like your patch introduced the new failures shown below. Please investigate and fix them before resubmitting your patch. If they are not new, fixing them anyway would help a lot. Otherwise please ask for the known failures list to be updated. The tests also ran into some preexisting test failures. If you know how to fix them that would be helpful. See the TestBot job for the details: The full results can be found at: https://testbot.winehq.org/JobDetails.pl?Key=146865 Your paranoid android. === debian11b (64 bit WoW report) === user32: win.c:4037: Test failed: Expected active window 00000000010B0162, got 0000000004A2017C. win.c:4038: Test failed: Expected focus window 00000000010B0162, got 0000000004A2017C.
v2: adjusted for simultaneous multithreaded DwmFlush() call behavior to be closer to some testing on windows. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/6006#note_75383
There is my MR for that from a year ago: https://gitlab.winehq.org/wine/wine/-/merge_requests/2463. It didn't get much attention for some reason although the specific issue it was back then motivated by was fixed in Wine Gecko. There are functional differences between my MR and this one which I suppose are important: - sleep time should not depend on the last DwmFlush() call time. Failing to know (or have) the actual next composition update time I think it is better to stick to synthetic equal intervals (my patch just rounds the QPC time to estimate the next period end); - Sleep() is probably too coarse with 1ms granularily. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/6006#note_75447
1. Your approach works for me and the mentioned plugin. And yes it looks better, but question is why was it rejected? 2. I'm seeing quite a lot of jitter on windows so I'd say it doesn't matter. Anyway would be great if any solution would be merged... -- https://gitlab.winehq.org/wine/wine/-/merge_requests/6006#note_75489
There's a few places where waitforvblank is not implemented either, should we have a real implementation for all of these or is sleep like that really enough? -- https://gitlab.winehq.org/wine/wine/-/merge_requests/6006#note_75513
I am not sure all these are really the same thing, as well as that we can sensibly implement it. It is not like Linux gives us an easy and unviersal way to wait for "vblank". Also I am not sure this specific call is directly linked to vblank really, this is about waiting for presented d3d changes to appear on screen. which, I think, might be related to DWM screen update but probably should not necessarily happen on the nearest screen update. In other words, one can't properly implement this DwmFlush with wait before implementing DWM compositor behaving similar to Windows. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/6006#note_75514
!2463 got merged, closing. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/6006#note_94684
This merge request was closed by Grazvydas Ignotas. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/6006
participants (5)
-
Fabian Maurer (＠DarkShadow44) -
Grazvydas Ignotas
-
Grazvydas Ignotas (＠notaz)
-
Marvin -
Paul Gofman (＠gofman)