|
0c77f3ad
|
2021-03-10T15:58:00
|
|
Vulkan: Implement shader buffers descriptor cache.
Implements a descriptor set cache for UBOs, SSBOs, and atomic counter
buffers. Storage Images and framebuffer fetch input attachments are
not yet included. Requires moving the buffer barrier handling into
ContextVk, similarly to how we handle the barriers for Textures.
The packed description key for the descriptors uses a "fast" vector
with a basic minimum size. For most cases of a few buffers this will
fit easily in stack memory, but for larger programs with many buffers
we fit this into heap memory. The key has a large upper bound due to
the high ES 3.2 requirements and the need to index several values such
as the offset and binding size.
We use dynamic offsets for uniform buffers when possible. This ensures
applications like Manhattan 3.1 that use sets of common buffers with
changing offsets hit the cache most of the time.
Because of resource limits we pick at compilation time whether to use
dynamic or static descriptor sets. Mostly this applies to tests that
use a large number of uniform buffers. A future implementation could
be smart and would recompile the program with heuristics to use a
minimal number of dynamic indices.
Reduces the number of descriptor set updates from ~300 -> ~30 per frame
in Manhattan 3.1 and in Asphalt 9 from 900+ to as low as 0 per frame.
Bug: angleproject:5736
Change-Id: I5c2a3881bec90d301dab15cc86c8a70e60674ad7
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2757515
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
|
efc53d0f
|
2021-04-03T21:30:15
|
|
Vulkan: Refactor to descriptor offset array.
This is in preparation for using dynamic offsets for other sets than
the default descriptors.
Bug: angleproject:5736
Change-Id: I546e688ad57a35b380f8d7f303ffc0c62773ac18
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2804813
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
|
|
4cdbdf90
|
2021-03-25T18:54:50
|
|
Vulkan: Assign XFB location first then assign varying location
Varying location is determined based on XFB location. If the transform
feedback stage is Vertex, there is no problem. But if the transform
feedback stage is GS or TES, previous implementation can cause shader
interface mismatch.
Bug: angleproject:5557
Test: KHR-GLES32.core.tessellation_shader.tessellation_control_to_tessellation_evaluation.gl_in
Change-Id: I1ecc7913a178c9e674307c528d1bdf13aabcb665
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2784713
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Mohan Maiya <m.maiya@samsung.com>
|
|
|
572fd801
|
2021-03-22T16:14:34
|
|
Vulkan: Rename Uniforms/XFB descriptor desc.
The new name is more consistent with the other names.
Refactoring change only.
Bug: angleproject:5736
Change-Id: Idc47fef29040e3a422267795c4536163a6f8eb4e
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2779953
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
c07ef602
|
2021-01-05T12:26:05
|
|
Vulkan: Move xfb buffer decl to translator in emulation path
This makes @@ XFB-DECL @@ empty on this path. Ultimately, this is
working towards removing both @@ XFB-DECL @@ and @@ XFB-OUT @@ macros
for both the emulation and extension paths, allowing the shaders to be
compiled at compile time rather than link time.
Bug: angleproject:3606
Change-Id: If16e9d92c419a04ecd3094481ed546d0708cdb43
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2611305
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
|
|
|
d654ac9b
|
2020-12-30T12:28:41
|
|
Program: Support multiple varying packings.
Instead of using a single varying packing for all program stages,
we switch to using a varying register packing for each pair of
input/output shaders. This allows several valid use cases that use
many varying to succeed. For instance Geometry Shaders have both
an input and output varying packing. With tessellation shaders the
upper bound of valid varying packings in one Program goes up even
more.
We keep multiple varying packings at once inside a new
"ProgramVaryingPacking" class. Internally the class keeps a unique
varying mapping for each input/output interface in the program.
Separable programs with "open" interfaces are handled specially.
Fixes a bug where varying counting was artificially limited for
programs with more than two shaders.
This CL also disables GS support when we're emulating line raster
so we don't have to figure out the details on how to place the
special position varying.
Bug: angleproject:5496
Change-Id: I1f9a327c4750caef570c608d86953e9d0cc5eea3
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2606532
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
|
|
|
92e7bc89
|
2020-12-23T15:18:42
|
|
Remove ProgramLinkedResources from ProgramExecutable.
Instead of storing the entire LinkedResources struct, we can keep it
only for the duration of the linking calls. Refactoring change only.
It sets the stage for more refactoring. This change also switches the
link call to use LinkingState's ProgramLinkedResources directly to
avoid the need to copy the varying packing or use a pointer.
Bug: angleproject:4514
Bug: angleproject:5496
Change-Id: Iefea3c16a33213dc338cc54efaa7c3064ea6ae08
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2601403
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
|
093250e0
|
2020-12-23T12:07:46
|
|
Vulkan: Make shader variable info maps a class.
Instead of using a map type directly we abstract the info maps into an
encapsulated class. We can enforce a specific API set instead of using
the same API as the map class. This also cleans up a few of the APIs
related to these maps.
This change will allow future changes to the way the variables are
stored in the class without drastically changing the interface.
Bug: angleproject:3572
Bug: angleproject:4524
Change-Id: Ic1a63e1776c39f49b895a1274bae8282d7a6b9b5
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2600080
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
|
|
|
06c39376
|
2020-11-27T12:45:33
|
|
Vulkan: Minimize gl_PerVertex members
As ANGLE doesn't redeclare gl_PerVertex, glslang always defines it with:
gl_Position
gl_PointSize
gl_ClipDistance
gl_CullDistance
The unused members here contribute to varying component limits. The
last two are unlikely to be used, and the second member is rarely used
as well.
This change keeps it simple and strips the trailing inactive members,
which for all intents and purposes accurately minimizes this struct.
Bug: angleproject:5405
Change-Id: I59c22af4988a3da7b1e428913d0ea13be9031cea
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2562754
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
f12e4123
|
2020-08-25T12:01:12
|
|
Vulkan: Match descriptor pool sizes to descriptor set layouts
When a descriptor pool is created, a list of descriptor types and counts
are given to vkCreateDescriptorPool(). Later, when allocating a
descriptor set from the pool, we pass along a descriptor set layout to
vkAllocateDescriptorSets() which is used to determine how many of each
type of descriptor (i.e. binding) to allocate from the pool.
In order for our "free descriptor set" counts to be accurate for each
pool, the descriptor pools need to be created with descriptor counts
that match the descriptor set layout binding counts.
This change fixes a bug where the descriptor set layouts were created
with more bindings than the descriptor pool sizes, causing the "free
descriptor set" count to be inaccurate, leading to allocating too many
descriptor sets from a pool.
Bug: angleproject:3570
Test: VulkanDescriptorSetTest
Change-Id: I660bf02d29a1291391fb15f39e6479bf348d0f83
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2391114
Commit-Queue: Tim Van Patten <timvp@google.com>
Reviewed-by: Charlie Lao <cclao@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
ae09e889
|
2020-09-10T17:00:13
|
|
Fix mismatch issue with precision qualifiers.
GLSL allows varyings passed from one stage to another to not match in
precision (e.g. float & half-float). Vulkan doesn't allow that so adjust
those mismatches to use the higher precision.
To fix we keep track of the precision of varyings and in the Vulkan
backend if we see they are different patch up the SPIR-V to make them
match.
Bug: angleproject:3078
Change-Id: I385d31e082da46ccdd4817b6612f5f9d9cbce17c
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2337755
Commit-Queue: Courtney Goeltzenleuchter <courtneygo@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
|
|
|
c5b5cf6c
|
2020-09-10T16:58:18
|
|
Refactor to pass ProgramMergedVaryings to link impl
Follow-on CL needs the ProgramMergedVaryings in the Vulkan backend to
generate valid SPIRV.
Bug: angleproject:3078
Change-Id: Ic442a3e0bd713fec36bd6b9420f67f3b1118e5ad
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2404336
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Courtney Goeltzenleuchter <courtneygo@google.com>
|
|
8765b46a
|
2020-07-15T17:10:35
|
|
Vulkan: Make mDefaultUniformStorage per ContextVk
Right now the dynamic buffer for default uniform is per program. Most of
time the buffer is unused. That is a huge waste of memory (and these
memory are wired memory). This CL moves the mDefaultUniformStorage from
per ProgramVk to ContextVk so that we all share with each other.
Bug: b/161391337
Change-Id: I1fe8523b2b2dbc39bec3509a3432e38e34bd5713
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2274870
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
|
|
|
cc551385
|
2020-06-29T10:25:45
|
|
Vulkan: Add cache for default uniform descriptor set
Instead of free and allocate a descriptor set whenever buffer changes,
add it into the descriptorset cache. Every time you get the buffer, we
also look in the cache first to see if there is a descriptorset for that
buffer or not. If yes, we use the cached descriptor set instead of
allocate one.
Bug: b/159457348
Change-Id: I64c252dfe22db66ea8188415e1b384af371b82ab
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2273698
Commit-Queue: Charlie Lao <cclao@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
Reviewed-by: Tim Van Patten <timvp@google.com>
|
|
|
22e1f3e2
|
2020-06-26T14:26:50
|
|
Vulkan: Let all shader stages share one buffer for default uniform
Right now each shader stage has its own vk::DynamicBuffer for default
uniform storage. This is less efficient than just share one buffer. This
CL moves the storage from per shader stage into its ProgramVk object.
Bug: b/159457348
Change-Id: If47248ea23c4e48407d3b211583ae2b048d4d10f
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2265281
Commit-Queue: Charlie Lao <cclao@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
|
|
|
68e60a92
|
2020-06-26T11:10:07
|
|
Vulkan: Don't store transformed SPIR-V
Since the original SPIR-V is never store after being modified, we don't
need to keep the transformed SPIR-V either.
Bug: angleproject:4513
Test: CQ
Change-Id: I4b70cfa355a08107e6a27377f6258cbf9e4072be
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2270718
Commit-Queue: Tim Van Patten <timvp@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
774cd1b2
|
2020-06-25T09:15:24
|
|
Vulkan: Only transform SPIR-V once
Prior to this change, the SPIR-V would be transformed as part of
GlslangGetShaderSpirvCode() and then possibly another time afterward to
disable the early fragment test optmization (for monolithic programs) or
to update the descriptor set/binding values (for PPOs). By storing the
original SPIR-V in the ProgramVk and the transformed SPIR-V in the
ProgramExecutableVk, we can now remove the SPIR-V transformation from
GlslangGetShaderSpirvCode() and only perform it once in
ProgramInfo::initProgram() for monolithic programs and PPOs.
Bug: angleproject:4513
Test: CQ
Change-Id: I4e7240b2abe7c796c46cf531948dad0e37c1419a
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2267537
Commit-Queue: Tim Van Patten <timvp@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
7993f34d
|
2020-06-24T18:39:08
|
|
Vulkan: Store transformed SPIR-V in ProgramExecutableVk
The original SPIR-V is specific to a ProgramVk while the transformed
SPIR-V is specific to a ProgramExecutableVk. This CL keeps the original
SPIR-V in the ProgramVk::mOriginalShaderInfo and the transformed SPIR-V
in ProgramExecutableVk::mTransformedShaderInfo. This allows us to skip
an extra copy that was occurring during PPO link().
Bug: angleproject:4513
Test: CQ
Change-Id: I796dcd6bcfb2df08d9b8f1a5295b2821a4096f16
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2265484
Commit-Queue: Tim Van Patten <timvp@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
d10ba166
|
2020-05-19T18:00:38
|
|
Rename ProgramState::getProgramExecutable() to getExecutable()
The "Program" part of ProgramState::getProgramExecutable() is already
implied by being part of a ProgramState, so it can be removed.
Bug: angleproject:3570
Change-Id: I35ffb2af81196fa7f189f7d3a37158f5f1951141
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2209317
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Tim Van Patten <timvp@google.com>
|
|
|
fecd1afc
|
2020-05-04T10:24:57
|
|
Vulkan: Move necessary members from ProgramState to ProgramExecutable
ProgramPipeline's need to be able to link Programs before drawing, even
if the Program's previous linkProgram() failed. To work towards this,
some ProgramState members are being moved to ProgramExecutable so they
can be saved when linkProgram() succeeds and not overwritten by any
subsequent linkProgram() attempts that may fail.
This also allows the second half of this change, which is to pass in the
Program's ProgramExecutable to GlslangAssignLocations() and
GlslangWrapperVk::TransformSpirV() so the values saved from the last
successful linkProgram() are used.
Bug: angleproject:4514
Test: CQ
Change-Id: I68aa429be76c0c6e1b886be09a12200217fcc7ab
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2181448
Commit-Queue: Tim Van Patten <timvp@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
|
0e1798d6
|
2020-04-20T11:25:32
|
|
Vulkan: PPO: Allocate uniform descriptor set if list is empty
When a Program is used to draw, the buffers to hold uniforms are
allocated as part of updateUniforms(), and then if a buffer is
allocated, a corresponding descriptor set is allocated for each shader
stage with uniforms. However, if that Program is later bound to a PPO,
the PPO's descriptor set list is cleared as part of creating the
pipeline layout. This means the uniform descriptor sets need to be
re-allocated but using the previously allocated buffers so the
'anyNewBufferAllocated' flag isn't enough - we also need to check if
the PPO's list of descriptor sets is empty.
Bug: angleproject:3570
Test: KHR-GLES31.core.sepshaderobjs.StateInteraction
Change-Id: Id8df587b4f5552012590a29d120b6899bbdf677b
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2156936
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
Reviewed-by: Charlie Lao <cclao@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Tim Van Patten <timvp@google.com>
|
|
|
dee4d7a5
|
2020-04-10T10:22:56
|
|
Vulkan: Early fragment tests optimization
Checks if early fragment tests as an optimization is feasible and enable
it if we can. In the link time, if context state diagrees with
optimization (in rare case), then remove the
ExecutionModeEarlyFragmentTests sprv op code.
Bug: angleproject:4508
Change-Id: Ifbb06c0ffb050a9f3ddb16ab50362e908b4b9cf6
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2136490
Commit-Queue: Charlie Lao <cclao@google.com>
Reviewed-by: Tim Van Patten <timvp@google.com>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
|
|
|
405f8e7b
|
2020-02-24T17:38:10
|
|
Vulkan: Support Program Pipeline Objects
Add support for PPOs to the Vulkan back end.
Bug: angleproject:3570
Change-Id: I5403456929847c185467b008d810f31ecfcb60cc
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2072652
Commit-Queue: Tim Van Patten <timvp@google.com>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
|
|
|
b980c563
|
2018-11-27T11:34:27
|
|
Reformat all cpp and h files.
This applies git cl format --full to all ANGLE sources.
Bug: angleproject:2986
Change-Id: Ib504e618c1589332a37e97696cdc3515d739308f
Reviewed-on: https://chromium-review.googlesource.com/c/1351367
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
a336b90f
|
2017-08-02T16:05:21
|
|
ES31: Impl program pipeline object management entries for GL backend.
The program pipeline object management entries are:
GenProgramPipelines
DeleteProgramPipelines
BindProgramPipeline
IsProgramPipeline
BUG:angleproject:2123
Change-Id: I114d054b90caf2ee3f9befef7439552a1c309bc4
Reviewed-on: https://chromium-review.googlesource.com/629978
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
|