This also has several other bug fixes and refactorings within it that
I realized were possible while updating every where that had checked
sigma > 0 to be sigma > kEffectivelyZeroSigma.
The big things are that SkBlurPriv.h goes away and its functions are
just moved into SkGpuBlurUtils since they were only used by the GPU.
The implementations of those functions are also collected into
SkGpuBlurUtils.cpp. I removed the GrMatrixConvolution::MakeGaussian,
in favor of SkGpuBlurUtils filling in the kernel itself and then calling
the regular Make. This let me consolidate two different 1D kernel
computing functions, and remove the 1D fallback code from the 2D kernel
calculation because GaussianBlur() can detect that earlier.
The new GM, BlurSigmaSmall, originally drew incorrectly on the GPU
backend because it's small but non-zero sigma would trick the sigma > 0
checks in various places so we'd do a full 2 pass X/Y blur. However,
when the sigma was too small, the kernel was just filled with 0s so the
Y pass would effectively clear everything. While I could have just fixed
that to be a [0, 1, 0] kernel, updating the blur pipeline to compare
against integer radii seems more robust.
Change-Id: I3c41e0235a27615a9056b25e627ffedd995264bd
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328797
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
This reverts commit e16eca95f5.
Reason for revert: ASAN error on fuzzer
https://status.skia.org/logs/snBeMRUkDrwDYbnm2SAG/7ad38736-d579-4e94-bc10-87c002f3f7d6/fd7b6ea1-5d36-4612-85d1-88462a5271f7
Original change's description:
> Create a basic IRNode pooling system.
>
> Allocations are redirected by overriding `operator new` and `operator
> delete` on the IRNode class. This allows us to use our existing
> `unique_ptr` and `make_unique` calls as-is. The Pool class is simple;
> it holds a fixed number of nodes and recycles them as they are returned.
>
> A fixed pool size of 2000 nodes was chosen. That is large enough to hold
> the contents of `sksl_large` during compilation, but it can be
> overflowed by very large shaders, or if multiple programs are converted
> at the same time. Exhausting the pool is not a problem; if this happens,
> additional nodes will be allocated via the system allocator as usual.
> More elaborate schemes are possible but might not add a lot of value.
>
> Thread safety is accomplished by placing the pool in a `thread_local`
> static during a Program's creation and destruction; the pool is freed
> when the program is destroyed. One important consequence of this
> strategy is that a program must free every node that it allocated during
> its creation, or else the node will be leaked. In debug, leaking a node
> will be detected and causes a DEBUGFAIL. In release, the pool will be
> freed despite having a live node in it, and if that node is later freed,
> that pointer will be passed to the system `free` (which is likely to
> cause a crash).
>
> In this CL, iOS does not support pooling, since support for
> `thread_local` was only added on iOS 9. This is fixed in the followup
> CL, http://review.skia.org/328837, which uses pthread keys on iOS.
>
> Nanobench shows ~15% improvement:
> (last week) http://screen/5CNBhTaZApcDA8h
> (today) http://screen/8ti5Rymvf6LUs8i
>
> Change-Id: I559de73606ee1be54e5eae7f82129dc928a63e3c
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/326876
> Commit-Queue: John Stiles <johnstiles@google.com>
> Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
> Auto-Submit: John Stiles <johnstiles@google.com>
TBR=brianosman@google.com,ethannicholas@google.com,johnstiles@google.com
Change-Id: I625d95a14057727b297c0bfc5b98bcd78ad8572c
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328906
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
This reverts commit f4594d1d5b.
Reason for revert: skp gone bad
Original change's description:
> Perform bounding rect-relative calcs in full float in GrRRectBlurEffect
>
> Add GM that tests very wide/tall SkRRects with blurs
>
> Bug: chromium:1138810
> Change-Id: Ib5a2e04de50c441f57f5e4b6194c3f9829323dc9
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328383
> Commit-Queue: Brian Salomon <bsalomon@google.com>
> Reviewed-by: Michael Ludwig <michaelludwig@google.com>
TBR=bsalomon@google.com,michaelludwig@google.com
Change-Id: I3633efd4802a0a9493831ec471304333272fa87e
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: chromium:1138810
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328905
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
These changes cause fewer unnecessary temporary variables to be emitted
in the final gencode and give the optimizer less work to do.
Change-Id: Ied0e83904e2d108382666d18ab733334292e5a63
Bug: skia:10549
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328838
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Switch statements were not caught when originally implementing
http://review.skia.org/328384.
Change-Id: Iff21e5743bf6a604e13c45a736799b4929844472
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328900
Commit-Queue: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Add GM that tests very wide/tall SkRRects with blurs
Bug: chromium:1138810
Change-Id: Ib5a2e04de50c441f57f5e4b6194c3f9829323dc9
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328383
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Allocations are redirected by overriding `operator new` and `operator
delete` on the IRNode class. This allows us to use our existing
`unique_ptr` and `make_unique` calls as-is. The Pool class is simple;
it holds a fixed number of nodes and recycles them as they are returned.
A fixed pool size of 2000 nodes was chosen. That is large enough to hold
the contents of `sksl_large` during compilation, but it can be
overflowed by very large shaders, or if multiple programs are converted
at the same time. Exhausting the pool is not a problem; if this happens,
additional nodes will be allocated via the system allocator as usual.
More elaborate schemes are possible but might not add a lot of value.
Thread safety is accomplished by placing the pool in a `thread_local`
static during a Program's creation and destruction; the pool is freed
when the program is destroyed. One important consequence of this
strategy is that a program must free every node that it allocated during
its creation, or else the node will be leaked. In debug, leaking a node
will be detected and causes a DEBUGFAIL. In release, the pool will be
freed despite having a live node in it, and if that node is later freed,
that pointer will be passed to the system `free` (which is likely to
cause a crash).
In this CL, iOS does not support pooling, since support for
`thread_local` was only added on iOS 9. This is fixed in the followup
CL, http://review.skia.org/328837, which uses pthread keys on iOS.
Nanobench shows ~15% improvement:
(last week) http://screen/5CNBhTaZApcDA8h
(today) http://screen/8ti5Rymvf6LUs8i
Change-Id: I559de73606ee1be54e5eae7f82129dc928a63e3c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/326876
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Prepare this class to hold either a proxyView or a vertex blob
Bug: 1108408
Change-Id: Ib6abb4da64ccc70b9e2af2546e1b071396dd42cb
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328836
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
PLS and discard-only shaders are the only time this has an impact,
and it doesn't seem like a problem to have the declaration?
Removes one use of variable reference counts, which are going to be
refactored.
Change-Id: Idb8d06087eed56070252ee02dcf907bf0d24c5a1
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328796
Reviewed-by: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Not yet used as of this CL.
Change-Id: Ic82ab5e2e2ca17fb11c16e22cfa6b7ad5ff74c77
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328657
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
This is conceptually very similar to http://review.skia.org/328384, but
the inliner doesn't use `clone()` when it clones a node.
Change-Id: I7456b63687ce2f93a7980fb101dfc97e143a378f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328817
Commit-Queue: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
After an IRNode is cloned, callers expect to be able to safely mutate
its SymbolTable, so it can't be left with a built-in one.
Change-Id: If658fd11ad580da552f9d689edeeed4c842b38c9
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328384
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Change-Id: I28b01a83960b6f7c8e715a817dc75a2408465a26
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328658
Commit-Queue: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
This new method receives its arguments in the same order as the
underlying C system call. This CL should not change behavior.
Change-Id: I6fd1e497b19d38d1133b1b8187146b98131093a8
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328624
Commit-Queue: Adam Barth <abarth@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
Auto-Submit: Adam Barth <abarth@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
Seems harmless to fall back to other SetupCrashHandler() options.
Change-Id: I2e7ce5dd72f391af88a57f4bd29831122fe784b9
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327893
Reviewed-by: Peter Foley <pefoley@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
This functionality will also be needed in TriangulatingPathOp::onPrePrepareDraws
Bug: 1108408
Change-Id: I0bf3f3c14820b946f4c4bc738c55e7a5eb2e7ced
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328556
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This has a slight ripple effect into Enum, as it was using the builtin
status as an indicator that the enum was shared with C++ code. This now
has a dedicated bool flag.
Change-Id: Id03efa902546775666acd031e6d57123e02b6c6e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328381
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This is everything except for literally removing the class.
Change-Id: I2f16caf865d1bcf9c0f267aed73313c0676a73bb
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327222
Commit-Queue: Adlai Holler <adlai@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Our Metal codegen assumes that out params are pointers, but Metal's
built-in frexp actually takes a reference for the exponent, not a
pointer. We now add in a helper function to translate.
Change-Id: I24686347d07151dd99a1ff1c43aff2b35c3181e5
Bug: skia:10762
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328387
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Commit-Queue: Jim Van Verth <jvanverth@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This loads in the known digests from Gold, starts the test
harness (which runs the GMs using puppeteer) and then uses
goldctl to upload the results to Gold when finished.
This will fail (and should not be landed) until
https://skia-review.googlesource.com/c/buildbot/+/328156
makes it into goldctl and the cipd build.
Bug: skia:10812
Change-Id: I89e5cf188d8f2adeba4ff676525d9bfbdcb46d5a
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328380
Reviewed-by: Leandro Lovisolo <lovisolo@google.com>
Reviewed-by: Eric Boren <borenet@google.com>
Change-Id: Iff707ae968b84609acde89ba94dec7a5d53c7525
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328576
Commit-Queue: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
This should fix the macOS fuzzer build.
Bug: chromium:1139725
Change-Id: I14090da4ee7d9d0a6e515b05c23c0a1e50ca4e5f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328385
Auto-Submit: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Florin Malita <fmalita@chromium.org>
Commit-Queue: Michael Ludwig <michaelludwig@google.com>
For posterity, here's my initial, wrong thinking:
If we squint at "return foo" and read it as "result = foo; goto
end_of_program", and then remember we can always skip forward jumps (and
where's further forward than end of program?), early returns turn out to
work just like a store.
The reason this is wrong is that by the time we reach a final return,
the entire mask stack has been popped back down to its original default
ffffffff (active) state. But that return shouldn't override any prior
returns. So that scheme isn't quite right.
Instead we accumulate the result by disabling updates to lanes that have
already returned. By the time we're done, all lanes should have hit
_some_ active return, now asserted.
Bug: skia:10852
Change-Id: I27b05f04a60ff4a5f2fe5f59bf398c3f7224a41b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327457
Commit-Queue: Mike Klein <mtklein@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
We will need this functionality in TriangulatingPathOp::onPrePrepareDraws as well as at the
current locations
Bug: 1108408
Change-Id: I2a65f07f47a549531d84dbf2afd82ad9d9b35225
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328536
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This is useful because we can clone FunctionDefinitions without cloning
the matching FunctionDeclaration. The FunctionDeclaration will remain a
builtin, but the definition should be a malleable clone.
Change-Id: Icfc1e0855fb8fcd6914a5d657f5098986fcf19ea
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328396
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This picks up a new version of the CIPD deps and a new helper for
writing task drivers.
Bug: skia:10812
Change-Id: I8b9b57acd4d8eee9cdea86008da1f3039af0cdc9
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328496
Reviewed-by: Eric Boren <borenet@google.com>
Commit-Queue: Kevin Lubick <kjlubick@google.com>
Using `return` instead of assigning to the output color removes an
unnecessary temporary variable from the output.
Change-Id: Ica31e290f8745a7309ae32c7148516d2189308ea
Bug: skia:10549
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328386
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
The JS starts a web server and then uses puppeteer to load the html
page. The page will load the compiled JS+WASM to run the GMs,
calculate the md5 hash of the pixels using the exact same algorithm
as DM and then get those encoded as a PNG. The HTML then makes
a post request with that image data (assuming it's not an image we
have seen before) and keeps all the test results in a simple
JSON array.
The JSON array will be read in by a task driver in a followup CL.
Bug: skia:10812
Change-Id: I90e8ed42bd96ae11fa75ce84ef2ee1a97ab93d5f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328378
Reviewed-by: Leandro Lovisolo <lovisolo@google.com>
We just make a global set of known png digests and expect that the test
harness will load this with the appropriate data.
Bug: skia:10812
Change-Id: I8e1fc987d36cc57386167410514803f8c3b90a69
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/328379
Reviewed-by: Chris Dalton <csmartdalton@google.com>
This adds the budgeted parameter to the GrAttachment ctors. Currently
we only have stencil and msaa attachments which are always budgeted
but this will soon change as more things get added.
Along the same lines this fixes the gpu memory size calculate on
render target. The msaa attachment was getting double counted in
the RT and the attachment itself.
Bug: skia:10727
Change-Id: I3520de9627eadaa4074f7425df509a6c1ccbe07f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327337
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
Bug: skia:10419
Change-Id: I74429ceb1fc03e68872c74eb62f64675e1bd55cc
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327476
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Was missing this virtual so when we were registering attachments in the
cache we were never setting a scratch key on them.
Bug: skia:10727
Change-Id: Id505fcea17ed83b9f9f33739ded518adbcc2acb7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/326440
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
This causes a ~4% regression on sksl_large, but some of that
can be bought back in two ways:
1) Removing (now unnecessary) cloning of program elements
2) Hoisting the new analysis passes, with (nontrivial)
logic to update/maintain the call counts as we edit IR.
Also, this fixes bugs where we were emitting functions that
had "calls" from no-longer called functions.
Bug: skia:10776
Change-Id: I4f8c29957be2e4233a883c9a1125f363b82ee40c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327198
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
Previously, we'd add UnresolvedFunction symbols wherever the original
FunctionDeclaration was found, but that can alter shared SymbolTables
(e.g. the root table) and the node will not be cleaned up when the
program is deleted. The pooling system expects that any node created as
part of a Program will go away when the Program does.
This change puts the UnresolvedFunction symbol as close to the
FunctionDeclaration as possible without changing builtin symbol tables.
Change-Id: I4c80cfba734047f11b50c86829d95d8e393c3060
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327921
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
