blob: c42beca22903af40079705c5972b1d9be5a817b5 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
|
commit b2cecde549c76cbd1c8b7d0cee2c6799936c1e7a
Author: Greg Daniel <egdaniel@google.com>
Date: Thu Jun 16 11:29:08 2022 -0400
Fix not using texture barrier on StrokeTessOp.
Previously we were overwriting the renderpassXferBarriers flag on
ProgramInfo to set it to kNone. This flag is meant to say whether or not
the entire render pass uses barriers or not. This is needed in Vulkan
because all pipelines in a render pass that has an input attachment
must bind the input attachment regardless if it is used or not. So the
pipeline must be created with a layout for an input attachment
descriptor set.
This change just removes to performance optimization to only use the
barrier on the stencil and not fill draw. This use case shouldn't
come up too often and also shouldn't be a big perf hit regardless.
The way GrAppliedClip is created/used it is hard for us to create
multiple different Pipeline objects: one for stencil and one for the
fill.
Bug: skia:13402
Change-Id: I15ce74b4d41b90d3dd4169a1f4fb77ed87c8b26d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/549898
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
diff --git a/src/gpu/ganesh/ops/StrokeTessellateOp.cpp b/src/gpu/ganesh/ops/StrokeTessellateOp.cpp
index 8f47441eb9..de8153cd0f 100644
--- a/src/gpu/ganesh/ops/StrokeTessellateOp.cpp
+++ b/src/gpu/ganesh/ops/StrokeTessellateOp.cpp
@@ -179,7 +179,12 @@ void StrokeTessellateOp::prePrepareTessellator(GrTessellationShader::ProgramArgs
fStencilProgram = GrTessellationShader::MakeProgram(args, fTessellationShader, pipeline,
&kMarkStencil);
fillStencil = &kTestAndResetStencil;
- args.fXferBarrierFlags = GrXferBarrierFlags::kNone;
+ // TODO: Currently if we have a texture barrier for a dst read it will get put in before
+ // both the stencil draw and the fill draw. In reality we only really need the barrier
+ // once to guard the reads of the color buffer in the fill from the previous writes. Maybe
+ // we can investigate how to remove one of these barriers but it is probably not something
+ // that is required a lot and thus the extra barrier shouldn't be too much of a perf hit to
+ // general Skia use.
}
fFillProgram = GrTessellationShader::MakeProgram(args, fTessellationShader, pipeline,
diff --git a/src/gpu/ganesh/vk/GrVkPipelineStateBuilder.cpp b/src/gpu/ganesh/vk/GrVkPipelineStateBuilder.cpp
index 54bc7f857a..0d61b8c4cb 100644
--- a/src/gpu/ganesh/vk/GrVkPipelineStateBuilder.cpp
+++ b/src/gpu/ganesh/vk/GrVkPipelineStateBuilder.cpp
@@ -279,6 +279,10 @@ GrVkPipelineState* GrVkPipelineStateBuilder::finalize(const GrProgramDesc& desc,
}
}
+ // The vulkan spec says that if a subpass has an input attachment, then the input attachment
+ // descriptor set must be bound to all pipelines in that subpass. This includes pipelines that
+ // don't actually use the input attachment. Thus we look at the renderPassBarriers and not just
+ // the DstProxyView barrier flags to determine if we use the input attachment.
bool usesInput = SkToBool(fProgramInfo.renderPassBarriers() & GrXferBarrierFlags::kTexture);
uint32_t layoutCount =
usesInput ? GrVkUniformHandler::kDescSetCount : (GrVkUniformHandler::kDescSetCount - 1);
|
