Commits on Source (68)
-
Jose Fonseca authored
-
Piotr Podsiadły authored
-
Jose Fonseca authored
As the latter is not in widely available CMake versions.
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
In antecipation for migrating to Python 3. But this time around, don't remove Python 2, so we can continue to build old branches.
-
Jose Fonseca authored
This reverts commit 0bee2047.
-
Jose Fonseca authored
This reverts commit 56439173.
-
Jose Fonseca authored
-
Jose Fonseca authored
Fixes https://github.com/apitrace/apitrace/issues/574
-
Jose Fonseca authored
-
i.Dark_Templar authored
-
Jose Fonseca authored
And ensure it's only included internally. This addresses the issue described in https://github.com/apitrace/apitrace/pull/611#issuecomment-488087078
-
Danylo Piliaiev authored
Pointers have unsigned type thus casting them to intptr_t could lead to a negative value which in one part of the expression is being implicitly converted to size_t resulting in: nBlocks = big_positive_number - negative_number This rarely happens on x64 but would often happen on x32 resulting in memory access violation. Fixes: bac0ea03 Signed-off-by:
Danylo Piliaiev <danylo.piliaiev@globallogic.com>
-
Jose Fonseca authored
-
Andrii Simiklit authored
According to directx 11 spec all device methods are thread safe: "All ID3D11Device interface methods are free-threaded, which means it is safe to have multiple threads call the functions at the same time." But all device-context methods according to directx 11 spec: "Any context – immediate or deferred – can be used on any thread as long as the context is only used in one thread at a time." So without this fix WrapID3D11Device*::CreateBuffer methods are not thread safe. It could lead to the crash in some games which are create buffers in different threads. Suggested-by:
Jose Fonseca <jfonseca@vmware.com> Signed-off-by:
Andrii Simiklit <andrii.simiklit@globallogic.com>
-
Andrii Simiklit authored
Signed-off-by:
Andrii Simiklit <andrii.simiklit@globallogic.com>
-
Danylo Piliaiev authored
CreateDeviceAndSwapChain family of functions can accept null pSwapChainDesc. Signed-off-by:
Danylo Piliaiev <danylo.piliaiev@globallogic.com>
-
Andrii Kryvytskyi authored
During each ::Map different pointers could be returned by DirectX. Thus for the same resource with the same subresource having the same size pointer may change between calls of MemoryShadow::cover. With the pointer change number of blocks also may change, which may later cause memory access violation in MemoryShadow::update. Signed-off-by:
Danylo Piliaiev <danylo.piliaiev@globallogic.com> Signed-off-by:
Andrii Kryvytskyi <andrii.o.kryvytskyi@globallogic.com>
-
Rene Sugar authored
-
Jose Fonseca authored
Fixes https://github.com/apitrace/apitrace/issues/599
-
Jose Fonseca authored
-
Jose Fonseca authored
They easily lead to hangs, as we don't capture multiple processes.
-
Zack Rusin authored
It's usefull to know what's the currently set topology so lets dump that information as well.
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
We should still use D3D11_RESOURCE_MISC_SHARED regardless.
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
IDXGIKeyedMutex::ReleaseSync method has side effects, which we need to emulate, when emulating D3D11_RESOURCE_MISC_SHARED_KEYEDMUTEX on top of D3D11_RESOURCE_MISC_SHARED.
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
As this seems to be the maximum supported in practice.
-
Danylo Piliaiev authored
For all writable coherent buffers we create respective shadow memory of the same size aligned to page. When memory is being mapped we save GL memory pointer and return readonly shadow memory with the offset. When write into shadow memory is caught we remove protection from the respective page, mark page as dirty and add this shadow memory to the dirty list in the current context. When removing protection if previous page is also dirty we double the number of page to remove protection from. Thus reducing amount of page exceptions to log(n) where n is a number of consecutive pages being written to. On all draw and unpack functions all dirty memory shadows are being commited to corresponding gl memory and made protected again. When fence become signaled we sync all readable buffers. Current shortcomings: - Buffer's contents may be changed from shaders however this wouldn't be reflected in shadow memory and if the same page is changed on CPU side and on GPU side even if offsets are different the data on the GPU will be overwtitten. - Small changes in buffer result in full page write. - All readable-writable buffers are synced back to cpu on fence signal instead of only potentially changed ones. Signed-off-by:
Danylo Piliaiev <danylo.piliaiev@globallogic.com>
-
Danylo Piliaiev authored
Signed-off-by:
Danylo Piliaiev <danylo.piliaiev@globallogic.com>
-
Illia Iorin authored
Signed-off-by:
Illia Iorin <illia.iorin@globallogic.com>
-
Jose Fonseca authored
Unfortunately this often happens due to application sloppyness when terminating, whereby GL resources are destroyed without first binding the associated GL context.
-
Jose Fonseca authored
Fix the issue that megai2 <megai2@ya.ru> reported on https://github.com/apitrace/apitrace/pull/625
-
megai2 authored
-
Jose Fonseca authored
Based on megai2@ya.ru's comment.
-
Jose Fonseca authored
-
Michal Krol authored
Also, fix tabs in dxgitrace.def.
-
Michal Krol authored
Uppercase "DXGI" is not accepted by apitrace trace.
-
Jose Fonseca authored
-
Jose Fonseca authored
D3D11CreateDevice can return E_FALSE when pointers are null. Apps seem to use this to check supported feature level. This change makes d3dretrace more robust against that.
-
Charlie Birks authored
-
Jeff Gilbert authored
v2: Keep old behavior for older CMake; plus tie a few loose-ends -- Jose
-
Jose Fonseca authored
Somehow auto-detection is not working on Travis builder images.
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Emil Velikov authored
Bump the waffle version requirement to 1.6 and handle the new platform token. Signed-off-by:
Emil Velikov <emil.velikov@collabora.com> Signed-off-by:
Alexandros Frantzis <alexandros.frantzis@collabora.com>
-
Jose Fonseca authored
-
Gleb Mazovetskiy authored
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Jose Fonseca authored
-
Joshua Ashton authored
Implement env var APITRACE_FORCE_MODULE_PATH which allows the user to force override the directory to the dirs we want to trace against.This is useful for tracing against reimplementations (ie. d9vk/dxvk) on their native platforms.