28166cb35d
Add "allocation slack" to our pushbuffers. Fixes a memory overwrite bug
...
Reported by Joseph on Discord.
Sometimes, things could align perfectly so allocations happend exactly
at the end of a pushbuffer. At the same time, we allow our vertex decoder to write an extra few
bytes if it needs to for speed. Unfortunately I missed this interaction,
resulting in some uncommon crashes that were especially common with
heavy-geometry things like modified GTA LCS with PS2 assets, for
example.
The problem was reported with Vulkan, but our OpenGL backend had the
same issue too.
2026-05-13 15:55:34 +02:00
4c2be7f8af
More work. Some initialization order problem.
2026-04-23 13:31:51 +02:00
4335a327cd
Hackery to prepare for constant buffers in compute
2026-04-23 13:31:51 +02:00
33b87578c5
Revive old alignment utility functions, give them better names.
2025-09-03 20:48:51 +02:00
f36173e17f
Vulkan: Remove mac specific hack for specifying host coherent memory - we need this everywhere.
2025-04-16 08:42:07 +02:00
c91169e702
Restore removed <algorithm> includes.
...
Turns out these were needed after all. For some reason, on Windows and
Mac, <algorithm> gets auto-included by something else so I don't notice
when it's missing, and MSVC's include dependency tracker doesn't see it
either.
2024-12-19 09:53:07 +01:00
4dd3621fa0
Remove some unnecessary <algorithm> includes.
2024-12-18 17:04:27 +01:00
49553bcf8d
Simplify MacOS AMD GPU workaround
...
Based on https://github.com/KhronosGroup/MoltenVK/issues/960 ,
expand MacOS AMD GPU workaround to all dGPU and instead of
changing usage, just append VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
2024-12-09 14:22:56 +01:00
286580a6a0
Switch to VMA_MEMORY_USAGE_CPU_ONLY on MacOS AMD
...
The ideal allocation would be VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
like in 1.12.3, but vma picked an index that's actually not mappable
VMA_MEMORY_USAGE_GPU_TO_CPU selects VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
which seems to hurt performance
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT + VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
selection with VMA_MEMORY_USAGE_CPU_ONLY would suffice for the
workaround, and get better performance
2024-12-09 12:32:31 +01:00
93009a3178
Work around metal buffer bug on MacOS + AMD GPU
...
With VMA_MEMORY_USAGE_CPU_TO_GPU buffers, metal buffer appears 0
filled in metal trace during vkCmdCopyBufferToImage triggered
MTLBlitCommandEncoder instance method.
Allocate VMA_MEMORY_USAGE_GPU_TO_CPU instead on MacOS + AMD GPU
2024-12-08 22:13:33 +01:00
2c283fbb07
Minor cleanups, crashfixes
2024-10-14 23:57:19 +02:00
f1cbf1ba96
Android TextRenderer: Add try {} around loading the Roboto font. Also improve an unrelated assert.
2024-10-10 16:01:00 +02:00
e01ca5b057
Logging API change (refactor) ( #19324 )
...
* Rename LogType to Log
* Explicitly use the Log:: enum when logging. Allows for autocomplete when editing.
* Mac/ARM64 buildfix
* Do the same with the hle result log macros
* Rename the log names to mixed case while at it.
* iOS buildfix
* Qt buildfix attempt, ARM32 buildfix
2024-07-14 14:42:59 +02:00
126d88ecfc
Back out clearly inconsequential/useless .reserve() calls
2023-12-29 08:27:56 +01:00
b8748ae9e5
[Common/File/GPU/Render/Input] Using reserve if possible
2023-12-15 14:00:51 +03:00
2ac14f555d
Remove VulkanPushBuffer (keeping our newer replacement VulkanPushPool)
2023-10-11 09:06:24 +02:00
34fbbf2c2a
Split out the descriptorset pool from VulkanMemory.cpp/h
2023-10-08 11:45:00 +02:00
6a8f65b566
Some assert paranoia, remove unused "failed_" variable
2023-09-23 10:09:32 +02:00
875112e9c2
Improve an assert
2023-09-11 14:18:55 +02:00
dfb446f89d
Allow other backends than Vulkan to have GPU memory stats. Implement for GL.
2023-05-24 14:33:01 +02:00
7e1f907826
VulkanPushPool: Sprinkle a couple of asserts.
2023-05-04 10:25:16 +02:00
a41fc74dba
Add a debug assert in VulkanPushPool
2023-05-01 11:41:25 +02:00
ae8c804328
Merge pull request #17134 from hrydgard/more-replacement-work
...
Refactor the replacement cache
2023-03-16 12:37:48 +01:00
de4eb64298
Build: Fix MSVC 2019 again.
2023-03-15 21:28:46 -07:00
34926472aa
Log fix
2023-03-16 00:30:20 +01:00
87a16b0386
Fix the stats
2023-03-15 10:19:00 +01:00
089b63c22f
Address feedback, except the stats.
2023-03-15 10:09:39 +01:00
2ac74e3d37
Remove the PushBufferType enum
2023-03-15 09:59:00 +01:00
f3bdf324f9
Garbage collect old buffers in push pool.
2023-03-15 01:25:03 +01:00
c8b25e50b0
Better debug vis for push pools
2023-03-15 01:25:03 +01:00
9fcd6d6612
Use thin3d's push pool in the draw engine too.
2023-03-15 01:25:03 +01:00
56d792f422
Sketch the new VulkanPushPool
2023-03-15 01:25:00 +01:00
ef4f17e998
Resurrect the Vulkan memory visualizer, but now it's global stats and pushbuffer stats.
2023-02-27 01:11:11 +01:00
31fd928a5c
Name more image views. Very useful with a little patch to the Vulkan debug layers.
2022-10-27 11:05:59 +02:00
974cd38b1f
Some more debug names to vulkan objects
2022-09-08 09:16:40 +02:00
b85a7e9a46
Name uniform buffers, add more asserts. Used this to track down the bug fixed in the previous commit.
2021-12-10 21:01:01 +01:00
8d06431ea7
Vulkan: Use new desc sets in Draw and VulkanUtil.
...
It seems like a bug that destory device objects wasn't clearing descSets.
2021-12-09 19:03:17 -08:00
1010d7ac0e
Vulkan: Refactor common desc pool handling.
2021-12-09 18:39:55 -08:00
f991f6a789
Remove the old allocator visualizer. Keep the setting but hide it. Other feedback
2021-11-23 08:55:12 +01:00
8d5247ddf3
Convert the PushBuffer to use VMA.
2021-11-22 18:54:09 +01:00
0cbb7ab027
Change the PushBuffer API a bit to not take explicit memory types.
2021-11-22 18:54:09 +01:00
c2f594cc73
Remove the VulkanDeviceAllocator
2021-11-22 18:54:09 +01:00
a31cbe42d2
Rename namespace to PPSSPP_VK, remove indentation
2021-08-21 12:39:15 +02:00
cc00a4132d
Prevent dlopen collisions between ppsspp_libretro.so and libvulkan.so
2021-08-20 11:22:57 +02:00
f32f89dd90
Global: Remove some unused variables.
2021-02-15 11:59:45 -08:00
0e3a84b4a8
Move most GPU things to Common.
...
It works after the move, on Windows and Android at least.
Deletes the D3DX9 shader compiler loader, which was not used.
2020-10-04 23:39:02 +02:00
Henrik Rydgård