|
b97aab3f
|
2022-03-09T17:36:24
|
|
Vulkan: resync mCurrentElementArrayBuffer when out of lineloop
When glDrawElements is called with GL_UNSIGNED_BYTE type or LineLoop
mode, we will internally allocate an element buffer and copy data to it.
But when we switch out of that mode, we must re-sync
mCurrentElementArrayBuffer to what it should be based on VertexArray
buffer binding. This CL fix the bug that we were previously not updating
it and end up using the wrong element buffer.
Also added three tests:
DrawWithSameBufferButDifferentTypes: that uses GL_UNSIGNED_BYTE data and
GL_UNSIGNED_SHORT data in the same buffer and switch between these two
data types without incurring buffer change.
DrawWithSameBufferButDifferentModes: draw line mode followed by triangle
without the same element buffer.
DrawArraysLineLoopFollowedByDrawElementsTriangle: draw line mode with
glDrawArrays and then followed by DrawElements.
Bug: chromium:1299261
Change-Id: I5c471117d300e9fac9127a9d8fa66d48ac312f03
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3513553
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Amirali Abdolrashidi <abdolrashidi@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
38723c28
|
2022-02-15T16:29:36
|
|
Vulkan: Allocate space for default attrib only if it is enabled
When context's default attributes is dirty, we allocate space for the
default attribute, regardless it is enabled or not. Then we call into
VertexArrayVk::updateDefaultAttrib() which only update its state if the
attribute is enabled. This causes a use-after-free scenario that if it
is disabled, the vertex array may have a pointer to the buffer that is
now becomes inflight which may gets deleted when DynamicBuffer code
think the size no longer matches etc.
Bug: chromium:1296467
Change-Id: Ib9ec8e60ebdb326f9bbfb215b3711c37631fce4b
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3466776
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
1608a956
|
2022-02-02T13:54:53
|
|
Vulkan: Revert client vertex data streaming to use DynamicBuffer
In early CL crrev.com/c/3352489, I switched client vertex data streaming
from using DynamicBuffer to sub-allocating from the buffer pool. That
caused CPU overhead regression due to extra cost of handling the
suballocation object creating and garbage collection etc. Even after all
other optimizations I did since then that significantly improved garbage
collection performance, there is still 6% CPU time regression as
measured with gardenscape. This CL moves StreamVertexData() back to use
DynamicBuffer. In order to do that, I have cleaned up DynamicBuffer
interface to be consistent with suballocation interface by storing the
current allocated offset/size in the suballocation object. With that,
the BufferHelper object that returned from DynamicBuffer will be able to
pass around and referenced exactly like it comes from suballocation code
path, and you can retrieve offset/size from that BufferHelper object
instead of having to pass offset around between various function calls.
Given that streaming vertex data from client memory is only possible for
default vertex array and there is only one default vertex array for each
context, this stream vertex data dynamic buffer is essentially a per
context object. So the other change I made here is that I have merged
mDynamicVertexData with default attribute (which uses per context
dynamic buffers) code to use the same sets of dynamic buffers, since you
will only use one or the other but not both.
Bug: b/205337962
Change-Id: I0ceca5b854069f00afdb9544ee86953b9b773821
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3434645
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
293c0b51
|
2022-01-21T15:53:38
|
|
Vulkan: Cache commonly used 6 ushorts stream index array data
Looking at all app traces that we currently have, 16 out of 100 apps are
making glDrawElements calls without element buffer. And among these
usages, most of them are calling glDrawElements with 6 unsigned shorts,
which makes sense for drawing a quad. This CL caches first four
BufferHelper objects with 6 uint16_t indices in a buffer and reuse them
if the data matches. With this we avoid create/destroy suballocations,
we even save the time of data copy and set DIRTY_BIT_INDEX_BUFFER when
called with same set of indices, which is the case for almost all apps
that uses glDrawElements based on app traces research.
In order to test the effect, I modified the `--minimize-gpu-work` to
keep glDrawElements calls with (count=6, tye=ushort) to pass down
count/type into angle, and only change the mode to point. That way this
new optimization will gets activated with `--minimze-gpu-work` command
line option (see crrev.com/c/3421377). With that, this CL sees cpu
overhead reduced from 2.54ms to 2.37ms on Pixel6 with
vulkan_offscreen_gardenscape.
Bug: b/215768827
Change-Id: I9b682868978e3bef7b5b9d1a596500ead2738d3e
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3404677
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
ad27d5d6
|
2021-12-21T11:22:30
|
|
Reland "Vulkan: Consolidate all vertex conversion buffers to shared pool"
This is a reland of cca412cd8b349b7281727c50f2a59d115fd90a05
Further inspection shows it was red-herring. The original CL does not
have the un-intended diff that I saw in the commit email. This is
try to reland the original CL without any modification.
Original change's description:
> Vulkan: Consolidate all vertex conversion buffers to shared pool
>
> There are various conversion buffers that holds converted vertex or
> element or index data. They are DynamicBuffer for now. This CL switches
> them to use the shared group buffer pool. With this change, all
> allocation is represented by a BufferHelper object instead of an offset.
> I am able to remove the offset arguments from a lot of APIs.
>
> Bug: b/208323792
> Change-Id: Ib611beb0c16cddbdd9ddf7b8961c439da9fa5180
> Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3352489
> Reviewed-by: Tim Van Patten <timvp@google.com>
> Reviewed-by: Jamie Madill <jmadill@chromium.org>
> Commit-Queue: Charlie Lao <cclao@google.com>
Bug: b/208323792
Change-Id: I90852ad38c2b9ac423800bb6854757bcc17cd166
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3370602
Reviewed-by: Tim Van Patten <timvp@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
4e85bdd9
|
2022-01-06T17:06:25
|
|
Revert "Vulkan: Consolidate all vertex conversion buffers to shared pool"
This reverts commit cca412cd8b349b7281727c50f2a59d115fd90a05.
Reason for revert: There is accidental code merge bug left in.
Original change's description:
> Vulkan: Consolidate all vertex conversion buffers to shared pool
>
> There are various conversion buffers that holds converted vertex or
> element or index data. They are DynamicBuffer for now. This CL switches
> them to use the shared group buffer pool. With this change, all
> allocation is represented by a BufferHelper object instead of an offset.
> I am able to remove the offset arguments from a lot of APIs.
>
> Bug: b/208323792
> Change-Id: Ib611beb0c16cddbdd9ddf7b8961c439da9fa5180
> Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3352489
> Reviewed-by: Tim Van Patten <timvp@google.com>
> Reviewed-by: Jamie Madill <jmadill@chromium.org>
> Commit-Queue: Charlie Lao <cclao@google.com>
Bug: b/208323792
Change-Id: I18bba207d1d8bb76dff32d9855a744dba93bc6d6
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3370601
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
cca412cd
|
2021-12-21T11:22:30
|
|
Vulkan: Consolidate all vertex conversion buffers to shared pool
There are various conversion buffers that holds converted vertex or
element or index data. They are DynamicBuffer for now. This CL switches
them to use the shared group buffer pool. With this change, all
allocation is represented by a BufferHelper object instead of an offset.
I am able to remove the offset arguments from a lot of APIs.
Bug: b/208323792
Change-Id: Ib611beb0c16cddbdd9ddf7b8961c439da9fa5180
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3352489
Reviewed-by: Tim Van Patten <timvp@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
e354ff1a
|
2021-03-05T04:07:21
|
|
Vulkan: Allow DynamicBuffer suballocation in BufferVk
When allocations are made from DynamicBuffer, they suballocate from a
possibly larger BufferHelper. In BufferVk, the offset of the
suballocation was discarded, which limited the use of DynamicBuffer to a
pool of small buffers.
This change applies any such offset that may arise from suballocations
everywhere, and makes BufferVk use a larger buffer size when the
GL_DYNAMIC_* buffer usage hints are provided.
Bug: angleproject:5719
Change-Id: I3df3317f7acff1b1b06a5e3e2bb707616a7d0512
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2738650
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
f0b02054
|
2020-08-06T20:55:05
|
|
Add a Vulkan feature to compress float32 vertex formats.
Use the vertex conversion pipeline in VertexArrayVk to detect
static vertex data and convert float32 vertices to float16. This
feature is useful for determining if an allication is vertex
bandwidth bound and seeing what gains could be had by using smaller
attributes.
This feature could be implemented in ANGLE's frontend but new
infrastructure for converting and storing the converted attributes
would need to be added to gl::VertexArray. Our backends already
have the functionality needed to handle unsupported attribute formats
and this can be repurposed for compressing vertex formats.
Bug: b/167404532
Bug: b/161716126
Change-Id: I9a09656a72e8499faa4124adf876d7261c8341c9
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2342285
Commit-Queue: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
959037e0
|
2020-05-25T15:40:38
|
|
Vulkan: Preserve RPs on XFB changes when possible.
Instead of unconditonally ending the RenderPass we keep a set of
active XFB buffers in the ContextVk. This lets us re-use RPs when
we don't write to the same buffer repeatedly.
Reduces the RenderPass count in our Manhattan capture from 29->23.
Bug: angleproject:4622
Change-Id: I28c2d4d3db1490e5d07be3c48d21fd2cc6ff85d6
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2196957
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Reviewed-by: Charlie Lao <cclao@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
93577b20
|
2020-05-28T15:16:46
|
|
Vulkan: Move "null" buffer to RendererVk.
This will allow the TransformFeedback and other classes to share
the same buffer. Also should save a bit of memory.
Bug: chromium:1086532
Change-Id: I198170b4e09165a4770b68af6df9aa7b690e8d66
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2219138
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
|
|
67d8b63a
|
2019-12-23T11:35:52
|
|
Vulkan:Update Vulkan Vertex Attribute Offsets
This fixes most of the test for
dEQP 3.1 KHR-GLES31.core.vertex_attrib_binding.basic-input*
The bug fixes is that we were not correctly setting the vertex attribute
offset for vertex attribs that were converted into their own buffer.
We kept the attrib offset from the frontend as if the converted attribs
were in the original buffer, but converted vtx attrib buffers in the VK
backend are tightly packed from the start of the buffer so when converting
the actual offset should be 0.
Also, in order to avoid a VK validation error, this change includes a tmp
workaround to make sure transform feedback output buffer range is never 0,
but sets VK_WHOLE_SIZE in that case. A follow-on fix that focuses on
transform feedback fixes will address this validation error more fully.
Bug: angleproject:4145
Bug: angleproject:4236
Change-Id: I8c218954725945414a8f18beb4f964b90da7062a
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1980906
Commit-Queue: Tobin Ehlis <tobine@google.com>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
c373dfd8
|
2019-11-22T08:48:22
|
|
Vulkan : Handle dirty state correctly when there are muiltiple VAOs
If vertex array object binding is changed, we need to update
the pipeline cache with the attribute information of the newly
bound VAO. We cache the strides of attributes because emulated
attributes will have strides that don't match the stride info
cached in its binding struct. Also added a test case that
switches between multiple VAOs.
Bug: angleproject:4127
Test: angle_end2end_tests.exe --gtest_filter=SimpleStateChangeTestES31.MultipleVertexArrayObjectRendering
Change-Id: I4f23aec33d5aa5988baa41f3c63db5534daf75ca
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1917453
Reviewed-by: Tobin Ehlis <tobine@google.com>
Commit-Queue: Tobin Ehlis <tobine@google.com>
|
|
7f418fc2
|
2019-10-01T07:56:53
|
|
Vulkan: lineloop support for DrawArrayIndirect
Add support for lineloops.
Includes a compute shader for generating an index
buffer to draw lineloop.
Instancing turns out to be a special case for indirect draws if we
have vertex attributes that need to be emulated (e.g. divisor too
large or native vertex format not available).
Test: dEQP.GLES31/functional_draw_indirect_*
angle_end2end_tests --gtest_filter=LineLoopIndirectTest.*/*
dEQP.GLES3/functional_draw_*
Bug: angleproject:3564
Change-Id: I1fdabe2c8a690c8b6df9e252e1e839e08796bcca
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1834682
Commit-Queue: Courtney Goeltzenleuchter <courtneygo@google.com>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
|
|
|
f03259ad
|
2019-09-19T11:31:40
|
|
Vulkan: lineloop support for DrawElementsIndirect
Add support for lineloops.
Includes a compute shader for converting lineloop index
buffer with optional restart into linestrip.
Test:
dEQP.GLES31/functional_draw_indirect_*
angle_end2end_tests --gtest_filter=LineLoopIndirectTest.*/*
Bug: angleproject:3564
Change-Id: I12d08db1c8d99867f0611e53be50193647695260
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1797106
Commit-Queue: Courtney Goeltzenleuchter <courtneygo@google.com>
Reviewed-by: Tobin Ehlis <tobine@google.com>
|
|
|
f10bf6bf
|
2019-09-26T10:27:18
|
|
Vulkan: Implement multi-threaded GL.
The main component of this change is to make vk::BufferHelper,
vk::ImageHelper and vk::SyncHelper use a common path. We introduce a
new "vk::SharedGarbage" helper class that stores small lists of garbage
from individual objects like an ImageHelper or BufferHelper. The
SharedGarbage is stored in the RendererVk with the ResourceUse of the
helper object. The ResourceUse tells RendererVk when it is safe to
destroy the GarbageObjects.
New "onGraphAccess" commands are added in a few places to enable the
common garbage collection path. A couple Context-only resources like
default attributes now are referenced where they were not before.
Also reorganizes some functions so we can add a few helpful ASSERTs
to our graph dependencies. Added "updateCurrentAccessNodes" for this.
Also adds a "RendererScoped" helper to replace many uses of
"ContextScoped".
The multithreading EGL tests mostly pass but have some remaining
flakiness so cannot yet be enabled.
Bug: angleproject:2464
Change-Id: Ia3e3ae8848d731abf3f21ebe04c33e381e130be0
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1808444
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
|
|
|
71c1138d
|
2019-08-16T12:23:04
|
|
Vulkan: Emulate instanced attrib divisor
This sets instancedArrays[ANGLE|EXT] extenstions as always
supported regardless of underlying Vulkan HW's max vertex attrib
divisor.
Then detect instances where app sets a divisor that isn't supported
by hardware and emulate those cases. Emulations is accomplished by
copying the instanced attribs to a new buffer where each attrib is
present once per instance, using the attrib divisor value as a
factor to replicate the attribs, and then setting the actual divisor
value for the draw to "1".
Also, we only store 8 bits for the divisor used in the PSO, so this
code also handles emulation of the case where divisor is > 255.
This is passing all of the drawInstanced/Elements dEQP tests
where divisor has to be emulated.
Also enabled end2end InstancingTestES3 for Vulkan backend.
Bug: angleproject:2672
Change-Id: I9932f9eab49b16a19e8bbd35dacaf3b5a27a213f
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1758689
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
Commit-Queue: Tobin Ehlis <tobine@google.com>
|
|
|
3c2a5230
|
2019-09-23T11:07:18
|
|
Vulkan: 8bit index support for DrawElementsIndirect
Add partial support for DrawElementsIndirect.
This supports all primitives types except lineloop.
Includes a compute shader for converting 8bit index
buffers to 16bit index buffers where the index buffer range
is defined in a GPU buffer.
Test:
dEQP.GLES31/functional_draw_indirect_*
Bug: angleproject:3564
Change-Id: Ibe9c55323e46a398f0b703cd8597a72ba6790570
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1792948
Commit-Queue: Courtney Goeltzenleuchter <courtneygo@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
0136ac37
|
2019-07-17T17:29:52
|
|
Separate dirty bit for attrib's binding VBO change
Make a separate dirty bit DIRTY_ATTRIB_POINTER_BUFFER for vertex
attrib's binding buffer change. So in handling glVertexAttribPointer,
ANGLE will only modify a vulkan graphics pipeline when attrib.format,
attrib.stride or attrib.divisor change. If only the VBO pointer changes,
then Vulkan can update the state via "vkCmdBindVertexBuffers()" without
triggering a pipeline update.
Bug: angleproject:3256
Change-Id: I01e02adde3708963b496a20050a5723e8eb57ab2
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1707614
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jiacheng Lu <lujc@google.com>
|
|
7e48c9eb
|
2019-08-06T17:17:19
|
|
Add explicit integer casts
WebKit uses the -Wshorten-64-to-32 flag which warns on these cases.
Bug: 3439
Change-Id: I8c1de60da0f173ca2036e2120e79b857f5f2775f
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1740866
Commit-Queue: James Darpinian <jdarpinian@chromium.org>
Reviewed-by: Kenneth Russell <kbr@chromium.org>
|
|
bf3d9333
|
2019-07-29T10:00:44
|
|
Vulkan: support relative offset in attrib bindings
Handles the ES 3.1 relative offset parameter in vertex attributes by
adding it to the binding offset.
Test: ./angle_deqp_gles31_no_gtest --deqp-egl-display-type=angle-vulkan -n dEQP-GLES31.functional.vertex_attribute_binding.usage.single_binding.unaligned_offset_elements_1_aligned_elements
Bug: angleproject:3598
Change-Id: Idbbd5ba4868a4dfc8f99188a84a5cd1374e09065
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1724453
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: James Dong <dongja@google.com>
|
|
|
375ddfc5
|
2019-07-12T11:12:14
|
|
Signal different dirty bit for vertex buffer change.
We use new logic to compare if the attribute format changes before
setting dirty bits. This improves performance of VBO-only state changes
significantly. On the VBO change Vulkan microbenchmark gives about a
30% improvement.
Bug: angleproject:3256
Change-Id: Ifaf1c92ed7a09422156ef79b5983e7349de63346
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1684294
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
875509e9
|
2019-06-17T13:44:21
|
|
Vulkan: Minor cleanup to ContextVk::setupIndexedDraw.
This removes VertexArrayVk::updateIndexTranslation. Turns out this
helper function wasn't that helpful.
Bug: angleproject:3539
Change-Id: Ia4573219073261767e9d215ed4227233c23cbfaa
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1660639
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
08b56293
|
2019-06-10T12:55:36
|
|
Vulkan: add LINE_LOOP with primitive restart
Adds support for GL_LINE_LOOP with primitive restart.
Bug: angleproject:3215
Change-Id: Ief1bdf15ef9b108dba025eaf4ce580bba54af623
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1649351
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
44063c80
|
2019-06-04T15:20:30
|
|
Vulkan: Store array buffer conversions in BufferVk.
The intent of this CL is to call convertVertexBuffer*PU only when we
have new data to convert. If the app unbinds and rebinds a vertex
buffer without changing the data we can now retrieve the cached
vertex buffer info from the BufferVk class. Previously we would always
reconvert the data on a rebind. This was slowing down applications and
benchmarks.
To achieve this we add a conversion cache to BufferVk. Each cache entry
stores a key based on the vertex info. Also we store a ring buffer for
each cache entry and a flag to indicate if the entry is dirty. The
cache is dirtied on a bufffer data update or a map call.
Improves performance in the T-Rex benchmark.
Bug: angleproject:3495
Change-Id: Ia999c9187510748ba95bc98362eb332e1990d270
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1638903
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
|
|
|
98f21671
|
2019-05-31T15:34:39
|
|
Vulkan: Refactor index buffer convert functions.
This is a prepratory refactor for converting index buffers on the GPU
using a more generic compute shader. No functional change.
Bug: angleproject:3490
Change-Id: Iadf4b1429314db6850320aee33c4113f38577378
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1639057
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
ee244c77
|
2019-05-06T10:30:18
|
|
Vulkan: Move command graph and garbage to ContextVk.
To support multithreading, contexts should manage their own command graphs
and garbage. This allows safe access to vulkan resources such as command pools
without thread synchronization.
BUG=angleproject:2464
Change-Id: I930149bc9f0793028761ee05ab50b8c0a4dec98a
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1516515
Commit-Queue: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
52047de4
|
2018-11-13T17:22:36
|
|
Vulkan: support instanced draws. (reland)
Enable instanced draws with the Vulkan backend.
So far it only works when Vulkan has VK_EXT_vertex_attribute_divisor.
BUG=angleproject:2672
Change-Id: Ib6655625776344305911a1a742c85f17638cee8f
Reviewed-on: https://chromium-review.googlesource.com/c/1469263
Reviewed-by: Frank Henigman <fjhenigman@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Frank Henigman <fjhenigman@chromium.org>
|
|
|
341304d8
|
2019-02-12T20:58:54
|
|
Revert "Vulkan: support instanced draws."
This reverts commit 199a9f385f5489b957fe1e42bf08f3f40edd38ca.
Reason for revert: Causes a validation error when the extension is not present. See failures here: https://ci.chromium.org/p/chromium/builders/luci.chromium.try/win-angle-rel/693
Original change's description:
> Vulkan: support instanced draws.
>
> Enable instanced draws with the Vulkan backend.
> So far it only works when Vulkan has VK_EXT_vertex_attribute_divisor.
>
> BUG=angleproject:2672
>
> Change-Id: I9445ba64282fa00a6eaee207b15efa2c7a9abbd3
> Reviewed-on: https://chromium-review.googlesource.com/c/1334973
> Commit-Queue: Frank Henigman <fjhenigman@chromium.org>
> Reviewed-by: Jamie Madill <jmadill@chromium.org>
TBR=fjhenigman@chromium.org,jmadill@chromium.org,syoussefi@chromium.org
Change-Id: Iffccc2cca259bcd19c068a87a415d4e196901f45
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: angleproject:2672
Reviewed-on: https://chromium-review.googlesource.com/c/1468201
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
|
199a9f38
|
2018-11-13T17:22:36
|
|
Vulkan: support instanced draws.
Enable instanced draws with the Vulkan backend.
So far it only works when Vulkan has VK_EXT_vertex_attribute_divisor.
BUG=angleproject:2672
Change-Id: I9445ba64282fa00a6eaee207b15efa2c7a9abbd3
Reviewed-on: https://chromium-review.googlesource.com/c/1334973
Commit-Queue: Frank Henigman <fjhenigman@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
c81e7bfe
|
2019-01-18T15:35:55
|
|
Vulkan: refactor CommandGraphResource
Merged back RecordableGraphResource into CommandGraphResource. Queries
didn't really need to be a resource, as they always inserted separate
single-command nodes in the graph. The CommandGraph class is augmented
with a few functions that generate such nodes.
This is in preparation for debug markers, as they too require such
nodes.
Bug: angleproject:2853
Change-Id: I5251a0e0fdd42ed1126921b4acc13687a14af9cd
Reviewed-on: https://chromium-review.googlesource.com/c/1422549
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
a63cc59f
|
2019-01-16T13:27:16
|
|
Vulkan: Don't store vertex attrib format and stride.
We only use these values in vertex conversion. We can instead pass more
info to the convertGPU/CPU functions. In client attribute streaming
we can query the format table a second time. This is a bit slower. But
the runtime of attribute streaming would usually be dominated by the
conversion function time. So it shouldn't regress any real work perf.
This saves on a bit of storage and memory overhead in the vertex sync
functions in VertexArrayVk.
Bug: angleproject:3014
Change-Id: I401eeff024664aa0efeea710503c0f619e6d4f22
Reviewed-on: https://chromium-review.googlesource.com/c/1406889
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
|
|
|
c759b8b4
|
2019-01-03T15:16:50
|
|
Vulkan: More Vertex Array optimizations.
Inlines a number of Vulkan vertex array methods.
Also changes the way vertex buffers are bound. Note that Vulkan doesn't
support NULL buffer bindings. Thus we create an emulated NULL buffer
to work around the problem of having gaps in the bound vertex buffers.
This allows us to use a single bind call for ranges of vertex buffers
even when there are gaps.
Also changes how vertex array dirty bits are reset. Instead of calling
memset to clear the affected buffers we pass a mutable pointer to the
Vertex Array sync state. This allows us to only reset the dirty bits
that we sync. This saves on the memory clearing time.
Improves perf by about 10% in the Vulkan VBO state change test.
Bug: angleproject:3014
Change-Id: Ib7b742dff7897fc891606a652ea0b64255a24c86
Reviewed-on: https://chromium-review.googlesource.com/c/1390360
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
|
|
|
80766cfa
|
2019-01-10T10:20:34
|
|
Vulkan: Update individual VAO attribs when dirty.
This simplifies the dirty state updates in VertexArrayVk. It also lets
us use a smaller dirty bit mask when a single attribute is marked dirty
in a vertex array.
Improves performance by about 1-2% in the VBO state change test. Will
allow for better performance using a pipeline transition table.
Bug: angleproject:3013
Change-Id: I25c5172b3f41b7abac6b8273c8f9cd42eb46cc9f
Reviewed-on: https://chromium-review.googlesource.com/c/1403958
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
dbc605ce
|
2019-01-04T16:39:14
|
|
Vulkan: Optimize VBO state changes.
Also has some minor optimizations for the front-end.
12% improvement on the Vulkan VBO change test.
Bug: angleproject:3014
Change-Id: I38e1a8194edfc14bfe57424be348cb9688e928f4
Reviewed-on: https://chromium-review.googlesource.com/c/1369286
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
611bbaab
|
2018-12-06T01:59:53
|
|
Vulkan: Convert vertex attributes in compute
In this commit, VertexArrayVk::convertVertexBuffer() is renamed to
VertexArrayVk::convertVertexBufferCpu() to explicitly show it does a CPU
readback. A new VertexArrayVk::convertVertexBuffer() function is added
that has the same functionality in gpu (with some assumptions, where the
CPU fallback is used should those assumptions fail). Currently, the
only requirement is that buffer offset/stride are divided by the
component size.
ConvertVertex.comp is the shader responsible for this conversion, and it
implements the functionality in renderer/copyvertex.inc, minus a few
functions that are not used in the Vulkan backend.
Bug: angleproject:2958, angleproject:3009
Change-Id: I8ec9a5f4672509bcf7b9e352cd27663970ad4653
Reviewed-on: https://chromium-review.googlesource.com/c/1364451
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
|
|
|
4abdf74f
|
2018-11-28T14:41:10
|
|
Vulkan: Add dynamic index buffers to graph
With DynamicBuffer outputting BufferHelper objects, these objects can
participate in the command graph, i.e. record commands. This means they
need appropriate dependencies in the graph as well as pipeline barriers.
There are a few users of DynamicBuffer for which this change should be
applied to. This change covers index buffers.
This commit includes a fix to BufferHelper::copyFromBuffer for WaW
hazards.
It also includes a fix for a missing pipeline barrier after
BufferVk::copyToBuffer.
Bug: angleproject:2958
Change-Id: I3e61af56936580b2da20c28c45defece552d9a39
Reviewed-on: https://chromium-review.googlesource.com/c/1352732
Reviewed-by: Tobin Ehlis <tobine@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
|
8dc27f99
|
2018-11-29T11:45:44
|
|
Use packed enum for DrawElementsType.
The packing and unpacking take a few extra instructions. But it
completely obviates the need for any switches in the validation code.
Speed is slightly faster or the similar depending on the back-end.
Also add gl_angle_ext.xml to GL entry point generator inputs. This was
missing and would cause the code generation to miss certain changes.
Bug: angleproject:2985
Change-Id: I1ea41a71db71135000166ead8305ec42d22ff7b3
Reviewed-on: https://chromium-review.googlesource.com/c/1351729
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
|
c30f45d3
|
2018-11-12T16:37:59
|
|
Vulkan: Rename PipelineDesc/Cache to Graphics&
PipelineDesc describes a Vertex-Fragment pipeline and PipelineCache (not
to be confused with vk::PipelineCache) implements a cache of such
pipeline objects.
In preparation for Compute support, these data structures are prefixed
with Graphics.
Bug: angleproject:2959
Change-Id: I9181586fb946b787216ca0b2ad6340f90c3ab55f
Reviewed-on: https://chromium-review.googlesource.com/c/1333971
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
|
03d1a5ed
|
2018-11-12T11:34:24
|
|
Vulkan: Use global buffer barriers.
This switches from using resource barriers for buffers to using global
barriers. This matches the general advised best practice. It also
allows us to combine multiple barriers into one. On a draw we might
combine all the vertex and index barriers into a single barrier call.
We implement this using a bit of extra state tracking in BufferHelper.
Bug: angleproject:2828
Change-Id: I196b368804ff50e60d085687a643e5566ba1c5b6
Reviewed-on: https://chromium-review.googlesource.com/c/1309977
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
|
193a284d
|
2018-10-30T17:28:41
|
|
Vulkan: Split vk::CommandGraphResource.
This adds two subclasses: RecordableGraphResource and
QueryGraphResource. Each specializes for Buffer/Image/Frambuffer use
cases and Query use cases respectively. No virtual functions are added
to keep best performance.
We also change the CommandGraph API slightly to optimize away the check
for a barrier resource. This requires exposing the set current barrier
API on the CommandGraph.
Bug: angleproject:2828
Change-Id: I1c23f52bfe04cc682a00b245d63c3ac9a651615d
Reviewed-on: https://chromium-review.googlesource.com/c/1305994
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
|
|
|
bfe31c42
|
2018-10-25T17:03:47
|
|
Remove uses of DrawCallParams.
Packing and referencing this structure was causing unnecessary draw
call overhead. This improves performance on all the back-ends. Impacts
the GL back-end the most.
In total this patch series reduces overhead by up to 5%.
Bug: angleproject:2933
Change-Id: Ief416ab874e481baf960d02965978a311214a146
Reviewed-on: https://chromium-review.googlesource.com/c/1299477
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
|
|
|
c1fd7376
|
2018-10-26T22:48:39
|
|
Move index range calculations into VertexArray.
This is in preparation for removing the entire DrawCallParams struct.
This struct was big enough to cause a performance hit on draw call perf
tests just by virtue of initializing the fields. Also dereferencing the
struct members is slower than reading function parameters since it adds
an indirection.
Also includes some error refactoring to enable moving code to a shared
location.
In total this patch series reduces overhead by up to 5%.
Bug: angleproject:2933
Change-Id: Ib663f2538c14ac30d4c31fd10d6350be469626e2
Reviewed-on: https://chromium-review.googlesource.com/c/1298380
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
6f755b21
|
2018-10-09T12:48:54
|
|
Use angle::Result in front-end. (Part 1)
This covers most of the hot paths used in draw calls. Gives in the
order of a 5% reduction in draw call overhead.
Bug: angleproject:2491
Change-Id: I2d53afb1163eaceed61fb9cd9ce6c1267c85c0fa
Reviewed-on: https://chromium-review.googlesource.com/c/1258149
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
88fc6da3
|
2018-08-30T16:18:36
|
|
Vulkan: Mega-refactor to VertexArrayVk.
This moves a lot of the code in VertexArrayVk into ContextVk. Having
the code in a centralized place makes the code a bit more organized
since the Context is reponsible for binding state to the command
buffers. It also makes it easier to use dirty bits to track the command
buffer state.
Bug: angleproject:2786
Change-Id: I5cefbb14028e8f3fe651f26e997ca88f8f1c7628
Reviewed-on: https://chromium-review.googlesource.com/1188953
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
|
|
|
253038d8
|
2018-08-30T16:18:35
|
|
Vulkan: Refactor VertexArrayVk::streamIndexData.
This enables us to use the same code for streaming client side index
arrays and for translating buffers.
Also includes a few more code cleanups.
Bug: angleproject:2786
Change-Id: Ic615d87cb50fd0acd8ab6b63ed334da4b1c40eff
Reviewed-on: https://chromium-review.googlesource.com/1188952
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Frank Henigman <fjhenigman@chromium.org>
|
|
|
a064c27c
|
2018-08-30T16:18:34
|
|
Vulkan: Refactor VertexArrayVk::streamVertexData.
Merges duplicated code in convertVertexBuffer with streamVertexData.
Refactoring change only.
Bug: angleproject:2786
Change-Id: I32baaad42d74918cfb17f04970d3c7e9b88362e7
Reviewed-on: https://chromium-review.googlesource.com/1188950
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
e452382a
|
2018-07-19T16:10:53
|
|
Vulkan: convert/align vertex data from buffers
When we get a buffer with vertex data in an unsupported format or
alignment, correct that as we copy it to a DynamicBuffer, then use
the copy. Enable tests.
BUG=angleproject:2405
Change-Id: I2132abea4d936f6b53d9209be7f99a0e2d8de219
Reviewed-on: https://chromium-review.googlesource.com/1141277
Commit-Queue: Frank Henigman <fjhenigman@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
|
|
|
5a4c932a
|
2018-07-16T11:01:58
|
|
Vulkan: Implement "default" vertex attributes.
Rendering from disabled attributes is implemented using small dynamic
buffers. The buffers use a stride of zero so the same vertex data is
pulled for every index.
This fixes all the disable attribute tests in dEQP.
Bug: angleproject:2444
Change-Id: I04fe139076da3e3ff723bed8eb17e333b4cb0ddf
Reviewed-on: https://chromium-review.googlesource.com/1136664
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
|
|
|
419acc8f
|
2018-06-24T19:57:31
|
|
Vulkan: Convert streamed vertex data as needed.
Add two members to vk::Format:
- vertex data copy function
- flag indicating if the function converts or not
Use the function when streaming vertex data so it gets converted if needed.
Add fallbacks for integer formats. These formats will now work everywhere,
as long as they are in client memory, not a buffer object.
Adjust test expectations accordingly.
BUG=angleproject:2405
Change-Id: I677221219d933c35740633a0ab7694293e218177
Reviewed-on: https://chromium-review.googlesource.com/1084328
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
21061026
|
2018-07-12T23:56:30
|
|
Vulkan: Use angle::Result error handling.
Introduces a vk::Context class to contain an error handler and Renderer
pointer. This abtracts the common code path for ContextVk + DisplayVk.
Removes vk::Error in favor of the POD angle::Result class. There are a
few remaining usages of gl::Error that will have to be cleaned up when
we can change the front-end APIs.
Bug: angleproject:2713
Change-Id: I5e68f223d595c6c561b59d6a85759e5738ed43c6
Reviewed-on: https://chromium-review.googlesource.com/1128924
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Frank Henigman <fjhenigman@chromium.org>
|
|
|
2ad498eb
|
2018-06-20T13:19:01
|
|
Vulkan: get vertex formats from format table.
Use the Vulkan format table to look up the Vulkan format for vertex data.
This will let us support more vertex formats by adding them to the table.
It also eliminates one usage of gl::VertexFormatType.
No functional change.
BUG=angleproject:2405
BUG=angleproject:2531
Change-Id: I73eb69ccac50d427de3e7d5479f92bb17c49aed3
Reviewed-on: https://chromium-review.googlesource.com/1051028
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Luc Ferron <lucferron@chromium.org>
Commit-Queue: Frank Henigman <fjhenigman@chromium.org>
|
|
6ed167a9
|
2018-06-13T13:45:55
|
|
Vulkan: Implement conversion to uint16 for drawElements with ubytes
Bug: angleproject:2646
Bug: angleproject:2659
Change-Id: If3c7a2b77d6acd18c8ca2522a427a43e10ed6db2
Reviewed-on: https://chromium-review.googlesource.com/1099420
Commit-Queue: Luc Ferron <lucferron@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
|
|
|
49aacad5
|
2018-06-12T08:33:59
|
|
Vulkan: Support the indices offset in drawElements calls
Also fixes an issue in buffer copy synchronization.
Bug: angleproject:2645
Change-Id: Ibca7052daaf1e6fe37913c8a8216ec33c66426b6
Reviewed-on: https://chromium-review.googlesource.com/1096911
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Commit-Queue: Luc Ferron <lucferron@chromium.org>
|
|
|
316c6065
|
2018-05-29T10:49:45
|
|
Vulkan: Call GraphResource instead of GraphNode.
We don't need to use the CommandGraphNode class directly. This CL
consolidates our code so we never call the GraphNodes class directly.
Instead we call operations on GraphResource. This should simplify the
interaction with APIs from the various graph and dependency management
classes in the Vulkan back-end.
A new concept of 'starting' vs 'appending' commands is introduced.
Appending tries to avoid starting new command buffers when possible.
Should not change how the graphs are constructed, and mostly be a
refactoring change. There may be minor behaviour changes to some
commands.
Bug: angleproject:2539
Change-Id: Ia971e5cacb1164b9b3b22fa4a0a55b954d81f10e
Reviewed-on: https://chromium-review.googlesource.com/1052068
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Frank Henigman <fjhenigman@chromium.org>
|
|
|
a9ab0f37
|
2018-05-17T17:03:55
|
|
Vulkan: Fix DynamicBuffer usages, need to use specific alignment
The alignment used to allocate VkBuffers in the VkBuffer needs to be at
least the size of the nonCoherentAtomSize defined in the limits of the
VkDevicePhysicalProperties. The latest roll of the
vulkan-validation-layers added that check and caused a bunch of errors.
This is fixing them.
Bug: angleproject:2565
Change-Id: Ia2ad506dce7966adb6220c52ea891903922c47d0
Reviewed-on: https://chromium-review.googlesource.com/1064950
Commit-Queue: Luc Ferron <lucferron@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
18e323ab
|
2018-05-11T16:54:17
|
|
D3D11: Fix out-of-range access with robust access.
When using a vertex buffer with DYNAMIC usage, with robust buffer
access enabled, we would sometimes read out-of-bounds when using very
large values for the index range. An unchecked signed addition would
overflow and lead to reading a negative offset.
Fix this problem by keeping the value size_t whenever possible. Also do
clamped casts when converting to a smaller values.
Also adds a regression test.
Bug: chromium:842028
Change-Id: Ie630ac857c6acfc0bace849a03eebfbaa2fbe89a
Reviewed-on: https://chromium-review.googlesource.com/1055928
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
|
|
|
bcef3224
|
2018-04-13T15:19:11
|
|
Move client attribs mask to front-end.
The Vulkan and GL back-ends both had a client attributes mask.
This consolidates them into the front-end, where it can also
be used in the validation layer.
Also includes a fix which was incorrectly setting the enabled
mask in setVertexAttribFormatImpl.
Bug: angleproject:1391
Change-Id: I5e45c1e2a56b30a36dec1482d170592c30a16d40
Reviewed-on: https://chromium-review.googlesource.com/1008272
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Luc Ferron <lucferron@chromium.org>
|
|
|
a56467e0
|
2018-04-11T16:19:41
|
|
VertexArray: Use switch macro for faster iteration.
Has a small but noticeable impact on performance on a microbenchmark.
Seems to improve a synthetic score by about 1%. Should have a very
small improvement in real-world performance.
Note that the odd formatting is an idiosyncrasy of clang-format.
Bug: angleproject:2389
Change-Id: I888bf101c6d8b80a0fbafdb9c5a84205c9c8fee6
Reviewed-on: https://chromium-review.googlesource.com/962963
Reviewed-by: Luc Ferron <lucferron@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
22f12fed
|
2018-04-08T14:23:40
|
|
Vulkan: Rename vk::LineLoopHelper.
This more closely follows the general pattern laid out by the naming
in vk_helpers.h. It also changes the dynamic buffer that the helper
wraps to be stored by-value since the header include order problem is
fixed.
Bug: angleproject:2318
Change-Id: I1de9e1edee2125d3afd490b4f9c99cf70c61215c
Reviewed-on: https://chromium-review.googlesource.com/1001654
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Reviewed-by: Luc Ferron <lucferron@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
9cceac42
|
2018-03-31T14:19:16
|
|
Vulkan: Update resource dependency semantics.
This removes passing the Serial around to several methods, so that
dependency management is a bit more automatic.
This makes life a bit easier when dealing with state updates when
resources are in use by Vulkan.
The FramebuffeVk no longer stores an extra serial of the last draw,
instead it will trigger creation of a new writing node on a state
change update.
Bug: angleproject:2318
Change-Id: Ie58ec66e6e8644ba4d402c509255c3795d363dd3
Reviewed-on: https://chromium-review.googlesource.com/985201
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Frank Henigman <fjhenigman@chromium.org>
|
|
|
6c7ab7fe
|
2018-03-31T14:19:15
|
|
Vulkan: Reorganize helper classes.
This renames ResourceVk to vk::CommandGraphResource, which should help
clarify its usage. This also moves LineLoopHandler, ImageHelper, and
the DynamicBuffer and DynamicCommandPool classes into a new vk_helpers
module. This file contains helper classes that manage other resources.
Also this makes DynamicBuffer and DynamicDescriptorPool no longer
inherit from CommandGraphResource. In the future, only Impl objects
will be allowed to be graph resources.
Bug: angleproject:2318
Change-Id: I0fa23da2ac853d90f3c822547a4a314f247cc757
Reviewed-on: https://chromium-review.googlesource.com/985200
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Frank Henigman <fjhenigman@chromium.org>
|
|
|
b8e39660
|
2018-04-04T11:41:42
|
|
Vulkan: Remove Observer from LineLoopHandler.
This can now use the VertexArrayVk dirty bits. It also
seems as though there are a couple caching bugs with the
LineLoopHandler.
Bug: angleproject:2389
Change-Id: I8af73f4acf56768ed9c68395349ba96acfbe9666
Reviewed-on: https://chromium-review.googlesource.com/989259
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Frank Henigman <fjhenigman@chromium.org>
Reviewed-by: Luc Ferron <lucferron@google.com>
|
|
|
c3755fc5
|
2018-04-05T08:39:13
|
|
Vulkan: Move Streaming data to VertexArrayVk.
Instead of the ContextVk owning the translations for the various
attributes, make the VertexArrayVk own them. This way they can
handle the dirty bit state notifications directly instead of
needing their own Observers.
Bug: angleproject:2389
Change-Id: I5e571ba6c563e820a4c0d5f92db35031e6f2428a
Reviewed-on: https://chromium-review.googlesource.com/989258
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
|
|
|
f3614370
|
2018-03-31T14:19:14
|
|
Vulkan: Rename StreamingBuffer to DynamicBuffer.
This makes it consistent with DynamicDescriptorPool, and gives a bit
more precise definition.
Bug: angleproject:2318
Change-Id: I8953113165ebe2d0dcfc0fc923d94280180442ce
Reviewed-on: https://chromium-review.googlesource.com/985199
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Luc Ferron <lucferron@chromium.org>
|
|
|
32fd63bc
|
2018-03-31T11:20:35
|
|
Vulkan: Use DrawCallParams in draw methods.
This cleans up some of the vertex streaming logic.
Bug: angleproject:2389
Change-Id: I8ed2f8acd06bbdd97db40acac35e5692112a3efe
Reviewed-on: https://chromium-review.googlesource.com/989257
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
|
|
|
0af5b86a
|
2018-03-27T20:19:33
|
|
Return gl::Error from VertexArray::syncState().
No functional change.
When we add vertex data format conversion to Vulkan we will need to be
able to return an error from VertexArray::syncState().
BUG=angleproject:2405
Change-Id: I4b537946ecbb6593280b6510c5cd8d8e3c65e8dd
Reviewed-on: https://chromium-review.googlesource.com/982897
Commit-Queue: Frank Henigman <fjhenigman@chromium.org>
Reviewed-by: Luc Ferron <lucferron@chromium.org>
|
|
|
e858cb1d
|
2018-03-27T09:44:32
|
|
Split VAO dirty bits to speed iteration.
Using > 64 bits (we had over 90) would use a much slower dirty bit
iteration. Speed this up by splitting the dirty bits into two levels.
The first top level only has a single dirty bit per attrib, per
binding, and one bit for the element array buffer. The next level has
separate dirty bits for attribs and bindings.
The D3D11 back-end doesn't actually care about individual dirty bits
of attribs or bindings, since it resets entire attributes at a time,
but the GL back-end only refreshes the necessary info.
Improves the score of a simple state change microbenchmark by 15% on
the D3D11 and GL back-ends with a no-op driver. Real-world impact will
be smaller.
Also includes a test suppression for an NVIDIA bug that surfaced when
we changed the order of that GL commands were sent to the driver.
BUG=angleproject:2389
Change-Id: If8d5e5eb0b27e2a77e20535e33626183d372d311
Reviewed-on: https://chromium-review.googlesource.com/556799
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
a53d0e18
|
2018-02-13T00:06:06
|
|
Vulkan: support indices in client memory.
glDrawElements will now work with indices and/or vertex data in client memory,
as well as in a buffer object.
Enable corresponding tests.
BUG=angleproject:1683
Change-Id: Iefb9796a48b21ed6f9a837b08b0ad3218ff6dd6b
Reviewed-on: https://chromium-review.googlesource.com/915721
Commit-Queue: Frank Henigman <fjhenigman@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
6dd4a92a
|
2018-03-02T16:35:13
|
|
Vulkan: track attributes in client memory.
To support indexed draws it's useful to have a cheap query for the
presence of vertex attribute data in client memory.
This patch adds a bit mask to keep track of such attributes.
It also lets us simplify VertexArrayVk::streamVertexData() slightly.
BUG=angleproject:1683
Change-Id: I871bfb885112650b025e110c383db3c391eafb90
Reviewed-on: https://chromium-review.googlesource.com/947927
Commit-Queue: Frank Henigman <fjhenigman@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
78e39b3f
|
2018-02-26T07:42:44
|
|
Vulkan: Line loops for indexed draw calls
Bug: angleproject:2335
Change-Id: Iabd6ae8181c6d3fb487f953a6fbf699db568a1c9
Reviewed-on: https://chromium-review.googlesource.com/941261
Commit-Queue: Luc Ferron <lucferron@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
1f46bc12
|
2018-02-20T16:09:43
|
|
Vulkan: Add CommandGraph class.
This also renames CommandBufferNode to CommandGraphNode. It also
renames some of the intenal members to more closely represent the
tree relationships (parents/children). This should clean up the
command graph classes and make them a bit easier to understand.
Bug: angleproject:2361
Change-Id: I024bffcc7f4157c78072ef902a3c40a07a08b18a
Reviewed-on: https://chromium-review.googlesource.com/922121
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
17448956
|
2017-01-05T15:48:26
|
|
Vulkan: vertex attributes in client memory.
Support vertex data stored in client memory passed to glVertexAttribPointer.
Only GL_FLOAT data is supported at this time. Includes a simple test.
BUG=angleproject:1683
Change-Id: I3bc0cdefe02b02c046b0e85822019a0f1762235e
Reviewed-on: https://chromium-review.googlesource.com/425137
Commit-Queue: Frank Henigman <fjhenigman@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
112a3a8e
|
2018-01-23T13:04:06
|
|
Vulkan: De-couple Program from VertexArrayVk dirtyiness.
The VertexArrayVk is responsible for filling out the packed shader
input info in ANGLE's packed PipelineDesc info structure. This
packed info structure is used for Pipeline init and caching lookup.
The prior design had this info depend on the active inputs in the
current Program. This was undesirable because then, on a Program
change, the ContextVk would have to call into the VertexArrayVk
to invalidate this info.
Instead, keep a working copy of the VertexArrayVk bits and only
update the bits corresponding to dirty vertex attributes. This
simplifies the cached state management a little bit for ContextVk.
This also means we don't have to update the cached copy in the
VertexArray on a change in VertexArray binding.
Bug: angleproject:2163
Change-Id: I5ba74535367aed74957d17bdc61f882508562d0e
Reviewed-on: https://chromium-review.googlesource.com/881703
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Frank Henigman <fjhenigman@chromium.org>
|
|
|
3c424b48
|
2018-01-19T12:35:09
|
|
Vulkan: Add vk_cache_utils.h.
This file contains the Pipeline and RenderPass cache utils.
Also renames renderervk_utils.h to vk_utils.h and the format utils
file.
Refactoring change only.
Bug: angleproject:2163
Change-Id: I5113a9a2c6f0b0960d38e6c2d8e391fa2d9f5f6a
Reviewed-on: https://chromium-review.googlesource.com/876505
Reviewed-by: Frank Henigman <fjhenigman@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
f2f6d379
|
2018-01-10T21:37:23
|
|
Vulkan: Add PipelineDesc.
The PipelineDesc class is a 512-byte packed description of the entire
Vulkan pipeline state. It uses the alignas keyword and some static
asserts to verify that the structures are packed. This ensures that
when ANGLE uses MurmurHash to hash the entire struct, and memcmp to
check for identity, that there are no garbage padding bits.
This CL does not implement the Pipeline cache, but it will help, since
now we have a packed type that can be used as the key to a hash map.
Bug: angleproject:2163
Change-Id: I16efa927f08d30d89a9c4c8943edd211c6878ac8
Reviewed-on: https://chromium-review.googlesource.com/829893
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Frank Henigman <fjhenigman@chromium.org>
|
|
|
49ac74bd
|
2017-12-21T14:42:33
|
|
Vulkan: Implement command re-ordering.
This introduces a new CommandBufferNode class. Nodes are linked
together to form a graph based on their dependencies. When the app
triggers a readback or swap, the graph is flushed entirely. This
sends the queued ANGLE Vulkan work to the Vulkan queue which is
then processed on the GPU with the right dependencies.
This design allows us to save on some unnecessary RenderPass creation
and also allows us to know what load/store ops to use. It also allows
us to take advantage of the Vulkan automatic RenderPass transitions
for performance. Load/Store ops and automatic transitions will be
implemented in later patches.
Bug: angleproject:2264
Change-Id: I0e729c719e38254202c6fedcede4e63125eb4810
Reviewed-on: https://chromium-review.googlesource.com/780849
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Frank Henigman <fjhenigman@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
da854a27
|
2017-11-30T17:24:21
|
|
Vulkan: Clean up VAO cached resources.
We can actually store a pointer to the base ResourceVk instead of
BufferVk for updating serials. This will work a little nicer with
streaming vertex data, which won't have a BufferVk but will have an
accessible ResourceVk pointer.
Also add an element array resource pointer for serial update. This was
missing and could lead to incorrect behaviour. Also change the types
of the caches from std::vector to gl::AttribArray, which is a
std::array.
Bug: angleproject:2264
Change-Id: Ibd79b7676b5dbc3875ae9d110be477d228e01c5c
Reviewed-on: https://chromium-review.googlesource.com/798170
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
|
|
|
acf2f3ad
|
2017-11-21T19:22:44
|
|
Apply Chromium style fixes.
This addresses several minor code quality issues that are validated
in Chromium, but not yet applied to ANGLE:
* constructors and destructors must be defined out-of-line
* auto is not allowed for simple pointer types
* use override everywhere instead of virtual
* virtual functions must also be defined out-of-line
Slightly reduces binary size for me (~2k on Win, 150k on Linux).
Bug: angleproject:1569
Change-Id: I073ca3365188caf5f29fb28d9eb207903c1843e6
Reviewed-on: https://chromium-review.googlesource.com/779959
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
ebf7299e
|
2017-10-13T14:09:45
|
|
Vulkan: Minimal dirty bits for ContextVk.
Currently this won't speed up performance much, if at all, since we
don't even really support state changes. It sets the stage for using
a pipeline cache later, with better state change support. It also
makes implementing descriptor sets for Textures a bit simpler, since
we can just update descriptor sets when the dirty bits tell us of a
Texture change.
Add cache structures to VertexArrayVk and ContextVk so we only need
to update the structures before we create a new pipeline. When we
support pipeline caching, we will most likely be updating a compact
representation for fast cache query.
BUG=angleproject:1898
BUG=angleproject:2167
Change-Id: Id545f2c67c06d8b6e8b7eb63ca70464f6b9a51f6
Reviewed-on: https://chromium-review.googlesource.com/713586
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Frank Henigman <fjhenigman@chromium.org>
|
|
|
bd159f06
|
2017-10-09T19:39:06
|
|
Vulkan: Use minimal dirty bits in VertexArrayVk.
This should slightly reduce draw call overhead.
BUG=angleproject:1898
Change-Id: I0e515bf2868f237f1d6948c12942f8cb6637c0c0
Reviewed-on: https://chromium-review.googlesource.com/707690
Reviewed-by: Frank Henigman <fjhenigman@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
492f58ec
|
2017-10-09T19:41:33
|
|
Rename VertexArrayImpl::mData to mState.
Refactoring change only.
BUG=angleproject:1898
Change-Id: I9f55651f923ff930c395a9bb575b4f86ad5d9cbd
Reviewed-on: https://chromium-review.googlesource.com/707689
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
4928b7ca
|
2017-06-20T12:57:39
|
|
Proliferate gl::Context everywhere.
This gives the D3D back-end access to the GL state almost anywhere.
This uses the onDestroy hook for Textures to push errors up from
destructors, although they still don't quite make it to the Context.
There are places, such as in EGL object (Context/Surface) destruction,
where we end up calling through to GL implementation internals without
having access to a gl::Context. We handle this via a proxy Context
to a Display, basically a null context, that has access to impl-side
state like the Renderer pointer if necessary. It does not have access
to the normal GL state.
Also Pass gl::Context to RefCountObject::release(). Since we're using
destroy() methods now, we should not ever call the destructor directly.
BUG=angleproject:1156
Change-Id: Ie4c32ad6bf6caaff0289901f30b5c6bafa2ce259
Reviewed-on: https://chromium-review.googlesource.com/529707
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
|
|
|
c564c070
|
2017-06-01T12:45:42
|
|
Pass gl::Context to impl methods instead of ContextImpl.
In some cases we might have to call back into the GL layer, passing
the Context, and if we just have a ContextImpl pointer this isn't
possible. It also removes the need for SafeGetImpl.
BUG=angleproject:2044
Change-Id: I6363e84b25648c992c25779d4c43f795aa2866d6
Reviewed-on: https://chromium-review.googlesource.com/516835
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
dd43e6cd
|
2017-03-24T14:18:49
|
|
Pass Context to VertexArray and Framebuffer syncstate.
This will enable more Vulkan-friendly idioms like clearing the
vulkan pipeline caches correctly on GL state changes immediately
because we have access to the ContextVk.
BUG=angleproject:1898
Change-Id: I16c848d8abdde8e26a38d384e565cec8548a66d0
Reviewed-on: https://chromium-review.googlesource.com/459079
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
9e54b5af
|
2016-05-25T12:57:39
|
|
Add Vulkan stubs.
Currently enabled for Windows by default.
BUG=angleproject:1319
Change-Id: I87921c579bee466465fb1e3f629bb3a40fdff659
Reviewed-on: https://chromium-review.googlesource.com/328730
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|