Previously, GLSL and Metal code generators would emit a struct wherever
the type was first used in the code, regardless of where it was
originally defined or what scope the type needs to live in. This CL adds
a ProgramElement for struct definitions, so that structs will now appear
at the top-level as they were originally defined. In the case of Metal,
some special handling is also needed to handle the Globals struct
properly.
Not yet fully supported:
- No special handling for structs declared inside functions yet
- No support for structs in separate scopes with overlapping names
The severity of the remaining issues depends mostly on whether we want
to support structs inside functions in Runtime Effects.
Change-Id: Ia95d4529506cb3fa6da63f5cb548199a93e1c0c5
Bug: skia:10922, skia:10923, skia:10925, skia:10926
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/338600
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This reverts commit e81fb87bb4.
Reason for revert: checking results with less-aggressive inliner
Original change's description:
> Revert "Simplify _blend_set_color_saturation, removing an instruction."
>
> This reverts commit ed289e777c.
>
> Reason for revert: causing strange artifacts, only on Adreno
>
> Original change's description:
> > Simplify _blend_set_color_saturation, removing an instruction.
> >
> > This tightens up our intrinsics slightly; after inlining, it eliminates
> > one scratch variable. (We no longer need to copy `sda` into `hueColor`
> > as hueColor is now unchanged.)
> >
> > Change-Id: Iece5ba2fe11cde54481704a1787114a2c2a66d9b
> > Reviewed-on: https://skia-review.googlesource.com/c/skia/+/336599
> > 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>
>
> TBR=brianosman@google.com,johnstiles@google.com
>
> Change-Id: Ica506467b0a4e03d0cbe482034acfa2d9f8d2c16
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/337560
> Reviewed-by: John Stiles <johnstiles@google.com>
TBR=brianosman@google.com,johnstiles@google.com
Change-Id: Ia93263f3269c057e7eaa69ca2b05e783d18c0199
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/337944
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Previously, we'd gauge suitability for inlining by counting the nodes in
a function; past a certain limit, the function was considered "too big."
Now, we also incorporate the number of times that function is called.
So if a function is called three times, and its size is 20 nodes, it
would be considered to have an inlining cost of 60 (3 * 20) instead of
20.
This should tamp down the aggressive nature of the inliner in cases like
gaussian convolution or complicated blends, and will hopefully satisfy
Pinpoint.
No change visible in Nanobench (which doesn't test any of these sorts of
patterns, but certainly inlines things): http://screen/AwD5hkgkEfjVx4g
Change-Id: Ie5e32898245ac854adb9ddd52d87001df6a67125
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/337676
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This reverts commit ed289e777c.
Reason for revert: causing strange artifacts, only on Adreno
Original change's description:
> Simplify _blend_set_color_saturation, removing an instruction.
>
> This tightens up our intrinsics slightly; after inlining, it eliminates
> one scratch variable. (We no longer need to copy `sda` into `hueColor`
> as hueColor is now unchanged.)
>
> Change-Id: Iece5ba2fe11cde54481704a1787114a2c2a66d9b
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/336599
> 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>
TBR=brianosman@google.com,johnstiles@google.com
Change-Id: Ica506467b0a4e03d0cbe482034acfa2d9f8d2c16
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/337560
Reviewed-by: John Stiles <johnstiles@google.com>
Previously, any builtin functions would be optimized as a side-effect of
optimizing programs that used them. Now that shared elements aren't
being optimized in that way, we explicitly optimize any shared modules
when they are first created. We don't remove dead elements, but we
we do substitute settings, simplify, and inline.
Bug: skia:10905
Change-Id: I701b5e9f52fb880ef3e6f4c67694d08602f47e95
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/336440
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
This tightens up our intrinsics slightly; after inlining, it eliminates
one scratch variable. (We no longer need to copy `sda` into `hueColor`
as hueColor is now unchanged.)
Change-Id: Iece5ba2fe11cde54481704a1787114a2c2a66d9b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/336599
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 will reorder constructors with swizzles applied, such as
`half4(1, 2, 3, 4).xxyz` --> `half4(1, 1, 2, 3)`
`half4(1, colRGB).yzwx` --> `half4(colRGB.x, colRGB.y, colRGB.z, 1)`
Note that, depending on the swizzle components, some elements of the
constructor may be duplicated and others may be eliminated. The
optimizer makes sure to leave the swizzle alone if it would duplicate
anything non-trivial, or if it would eliminate anything with a side
effect.
Change-Id: I470fda217ae8cf5828406b89a5696ca6aebf608d
Bug: skia:10954
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/335860
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
The optimizer can now turn the expression `half4(1).xyz` into
`half3(1)`, or `half4(1).w` into `1`. This is actually a somewhat common
case when inlining chains of fragment processors, as inputs are often
overridden to `half4(1)` or `half4(0)`. This optimization also applies
to more complex cases, e.g.:
`half2(anyFunc(sqrt(2))).yxyx` --> `half4(anyFunc(sqrt(2)))`
Since the interior of the constructor is always evaluated once in either
case, it does not actually matter what the constructor contains.
Change-Id: I8d5f358502eaa8e35d4968e74fbd6b0ce2ab6365
Bug: skia:10954
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/335818
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This prevents OOMing when given a pathological input, but is large
enough that almost all inputs should continue to compile as-is.
Change-Id: If5c46711b886ee08495bfd09af537e9dc7ea5649
Bug: skia:10945, oss-fuzz:27442
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/334838
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
In practice, the inline threshold does a good job of limiting the
blast radius here.
Change-Id: I495184116e733262ea9d84fec30885ea047ca116
Bug: skia:10945
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/334597
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@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>
This golden verifies that when the inline threshold is zero, inlining is
not performed.
Change-Id: Icad6e1faed569dd1b2469874be3b9e635ad0b9ad
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/325656
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
This reverts commit 941fc7174f.
Reason for revert: performance now seems to be roughly equal or better
(~1%) over several trials.
Nanobench: http://screen/A8e8sojaXBgbMgF
Original change's description:
> Revert "Remove inliner from IR generation stage."
>
> This reverts commit 21d7778cb5.
>
> Reason for revert: Pinpoint absolutely hates this change
>
> Original change's description:
> > Remove inliner from IR generation stage.
> >
> > There is no need to inline code during IR generation, as the optimizer
> > can now handle this.
> >
> > Change-Id: If272bfb98e945a75ec91fb4aa026e5631ac51b5b
> > Reviewed-on: https://skia-review.googlesource.com/c/skia/+/315971
> > Commit-Queue: John Stiles <johnstiles@google.com>
> > Commit-Queue: Brian Osman <brianosman@google.com>
> > Reviewed-by: Brian Osman <brianosman@google.com>
> > Auto-Submit: John Stiles <johnstiles@google.com>
>
> TBR=brianosman@google.com,ethannicholas@google.com,johnstiles@google.com
>
> Change-Id: I62c235415bcdc92a088e2a7f9c3d7dbf7e1bf669
> No-Presubmit: true
> No-Tree-Checks: true
> No-Try: true
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/317976
> Reviewed-by: John Stiles <johnstiles@google.com>
> Commit-Queue: John Stiles <johnstiles@google.com>
TBR=brianosman@google.com,ethannicholas@google.com,johnstiles@google.com
Change-Id: I6189806c678283188f4b67ee61e5886f88c2d6fc
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/324891
Reviewed-by: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This greatly improves the output from a profiler. It makes it much
easier to determine how much time is spent in searching for candidates,
versus actually inlining them.
It also improves the code readability somewhat by breaking a large
monolithic function into several smaller functions.
Change-Id: I1b3ef6ddbe46af60e673f37ded766f8077ed6b03
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/321376
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
We don't need to create a temporary variable for expressions like
`half3(x)`.
Change-Id: Ie0fa6a6dfb3d77d4372f96c676d3081f7e278852
Bug: skia:10786
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/320960
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
The following types of expression are hoisted directly into the
inlined code:
- Struct field access: `myStruct.myField`
- Swizzles: `myVector.xzy`
- Simple array indexes: `myArray[0]`
This significantly reduces the number of temporary variables generated
by the inliner.
Change-Id: Ifed226ecc87b096ec1e38752c0c38ae32bd31578
Bug: skia:10737, skia:10786
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/319919
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
These cover:
- Properly configured out-params
- Invalid/non-lvalue out-params, which currently cause an SkSL crash
- Interactions between the inliner and variable swizzles
Change-Id: I4874101236084f273e704d8717149b431d813883
Bug: skia:10753
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/319036
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
I realized that "DefaultSettings" as a name suffix was unclear, because
"Default" is a different settings mode from skslc running with
--nosettings.
In --nosettings mode, skslc uses "standalone" settings.
Change-Id: I1f5d80df0a21cec55948c4ad146169bcb34f4999
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/318210
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
This wasn't rolled back automatically by the revert at
http://review.skia.org/317976, because this unit test did not exist yet
when that CL was submitted.
Change-Id: Ib887e74ddd32c1e576908bbe3d588205e26cdaa5
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/317978
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This reverts commit 910845fac1.
Reason for revert: IRGenerator inline change reverted
Original change's description:
> Add program-settings flag to disable the inliner.
>
> Change-Id: I6c4e7f6a2aab6710221029022a3a5f3ec323c5e2
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/317856
> Commit-Queue: John Stiles <johnstiles@google.com>
> Commit-Queue: Brian Osman <brianosman@google.com>
> Reviewed-by: Brian Osman <brianosman@google.com>
> Auto-Submit: John Stiles <johnstiles@google.com>
TBR=brianosman@google.com,ethannicholas@google.com,johnstiles@google.com
Change-Id: Ie38a29495ea8497f9db26d2603df179e696ac5ff
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/317977
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
This reverts commit 21d7778cb5.
Reason for revert: Pinpoint absolutely hates this change
Original change's description:
> Remove inliner from IR generation stage.
>
> There is no need to inline code during IR generation, as the optimizer
> can now handle this.
>
> Change-Id: If272bfb98e945a75ec91fb4aa026e5631ac51b5b
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/315971
> Commit-Queue: John Stiles <johnstiles@google.com>
> Commit-Queue: Brian Osman <brianosman@google.com>
> Reviewed-by: Brian Osman <brianosman@google.com>
> Auto-Submit: John Stiles <johnstiles@google.com>
TBR=brianosman@google.com,ethannicholas@google.com,johnstiles@google.com
Change-Id: I62c235415bcdc92a088e2a7f9c3d7dbf7e1bf669
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/317976
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
This resolves the following TODO block:
TODO(johnstiles): the skslc standalone caps bits do not enable
do-while support, so this test does not actually perform as
described; the `returny` function is not inlined at all. This will
be fixed when customizable caps-bit support is added to the golden
tests.
Change-Id: I3495e4813b9be37264a8fda978453594c1f5fa13
Bug: skia:10694
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/317859
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Change-Id: I6c4e7f6a2aab6710221029022a3a5f3ec323c5e2
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/317856
Commit-Queue: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
There is no need to inline code during IR generation, as the optimizer
can now handle this.
Change-Id: If272bfb98e945a75ec91fb4aa026e5631ac51b5b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/315971
Commit-Queue: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
`fBuiltinFMASupport` is now true on both, and
`fUsesPrecisionModifiers` is now false. Other mismatching flags exist,
but they are non-trivial to synchronize as they are tied to extension
strings.
This will help our skslc-based unit tests generate the same results as
our C++ unit tests did, but should not affect real-world results as
these defaults will all be overwritten in a non-testing scenario.
In practice, the `fUsesPrecisionModifiers` change is responsible for all
of the diffs below. The other flags did not change the results of any of
the currently-ported tests.
Change-Id: Ieb056d852b027fa87c56fd89f971a77a10a8a124
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/317204
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Our lack of proper caps-bits controls in skslc affects the outcome of
one test: "InlinerWrapsEarlyReturnsWithDoWhileBlock" does not actually
emit the do-while block because the standalone caps bits don't enable
do-while support. This will be fixed in a followup CL that adds caps-bit
support to our tests.
A few tests were renamed for consistency, a few were simplified slightly
and one test was removed because it was simply redundant (there was a
second test that covered the exact same ground as
`ForWithReturnInsideCannotBeInlined`).
Change-Id: I2e3b97cb3aea331b6d806bdb865aa78c35c7a6b9
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/316997
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Change-Id: I6097f50e7be1fa2f7772f6c454410ecbf3470eea
Bug: skia:10722
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/316977
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
The following conditions lead to the error:
- A pair of nested functions, both of which must be inlined.
- Both inlined functions create a variable with the same name.
- The outer function passes its variable to the inner function.
- The initialization of the inner variable uses the value from the outer
variable.
- The inner function does not mutate the variable, use it as an out-
parameter, or otherwise cause it to receive a temporary copy.
When all these conditions are met, both variable declarations are
inlined as-is without performing any name salting, because it's
seemingly safe to do so. The name overlap issue is not considered in the
safety checks. Inlined variable declarations are not subject to name
salting but they should be; I suspect other adversarial examples could
be crafted as well where unhandled name overlap leads to errors.
Change-Id: Ia754bee8e45c8a5c7548436594bbf04abc7a8396
Bug: skia:10722
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/316945
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
