This will allow us to load these inputs for unit testing in `dm`.
Change-Id: Id256ba7c30d3ec94b98048e47af44cf9efe580d5
Bug: skia:11009
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/357282
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Like _globals, it's not actually necessary to indirect through a
separate pointer at all. The output struct is now passed by reference
and the additional pointer variable is removed.
(Additionally, renamed _skGlobals back to _globals.)
Change-Id: Id089a20cb751cdaedc48462a52da78ee43783611
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/355632
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
The Inliner likes to move function bodies around; after inlining, code
can inadvertently move upwards, above ProgramElements that the code
relies on. We work around this by always emitting functions last.
Change-Id: Ie5486cc3a79a478920342fb9f578d575486fb4cf
Bug: skia:11186
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/354669
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Many of our shaders generate the same vector constant dozens of times,
e.g. Gaussian blur uses float4(1) repeatedly. This change avoids
re-emitting redundant vector constants.
Change-Id: I22a71cd8b2783fb997f52d485b49031f64ca6d96
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/350701
Auto-Submit: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Because we use `continue` for flow control handling now, we can escape
from the middle of a switch statement. This wasn't possible when we used
`break`.
This unlocks some pretty stellar optimization opportunities if the
switch value can be determined at compile time; see BlendEnum for an
example.
Change-Id: Id29be92c343c10fd604683a80c5d5bd2bd070cb0
Bug: skia:11097
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/345419
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
When we determine that a function only contains a single return
statement and it is at the top level (i.e. not inside any scopes),
there is no need to create a temporary variable and store the
result expression into a variable. Instead, we can directly replace
the function-call expression with the return-statement's expression.
Unlike my previous solution, this does not require variable
declarations to be rewritten. The no-scopes limitation makes it
slightly less effective in theory, but in practice we still get
almost all of the benefit. The no-scope limitation bites us on
structures like
@if (true) {
return x;
} else {
return y;
}
Which will optimize away the if, but leave the scope:
{
return x;
}
However, this is not a big deal; the biggest wins are single-line
helper functions like `guarded_divide` and `unpremul` which retain
the full benefit.
Change-Id: I7fbb725e65db021b9795c04c816819669815578f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/345167
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
If we aren't wrapping the inlined function body in a loop, there's no
need to add a scopeless Block; we've already got one. This doesn't
affect the final output meaningfully--it just suppresses a newline--but
it's one fewer IRNode allocation.
Change-Id: Ib7b0014e908586d8acfcf6c23520873fad31d0b7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/345163
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Previously, multiple inliner passes in a row would each apply a
separate name mangling to variable names, so names like "_25_14_3_1_pos"
were not uncommon. This change demangles the name before re-mangling it,
so we would have just "_25_pos" instead.
It's not important, but it makes things easier to read.
Change-Id: I1257222dac2a68e337f431af230ce50730cedc9b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/345116
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This reverts commit e8e4aca955.
Reason for revert: can break ES2 for-loop rules
Original change's description:
> Declare all inlined variables at the topmost scope possible.
>
> By itself, this is uninteresting and even perhaps slightly
> counterproductive (as it separates vardecl from its initializer,
> increasing LOC). However, this enables a followup CL
> (http://review.skia.org/344665) which allows single-return functions to
> be inlined without the creation of a temporary variable at all. This
> applies to the majority of fragment processors in a typical Ganesh
> hierarchy. This change will greatly reduce the number of inliner-created
> temporary copies when compiling a typical tree of FPs.
>
> Change-Id: I03423a13cf35050637dabace4a32973a08a4ed0a
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/344764
> Reviewed-by: Brian Osman <brianosman@google.com>
> Commit-Queue: John Stiles <johnstiles@google.com>
TBR=brianosman@google.com,ethannicholas@google.com,johnstiles@google.com
Change-Id: Ica01d6906bcb9cef1f49d22dda714fc9cbfa3885
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/345121
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
This reverts commit 345d72124d.
Reason for revert: can break ES2 for-loop rules
Original change's description:
> Eliminate inliner temporary variables for functions with a single exit.
>
> When we determine that a function only contains a single return
> statement, there is no need to create a temporary variable and store the
> result expression into a variable. Instead, we can directly replace the
> function-call expression with the return-statement's expression.
>
> This dramatically simplifies the final optimized output from chains of
> very simple inlined functions, which is a very common pattern for trees
> of Skia fragment processors.
>
> Change-Id: I6789064a321daf43db2e1cef4915f25ed74d6131
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/344665
> Commit-Queue: John Stiles <johnstiles@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: I60845f22159605a06047b030e2686a769121a35a
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/345120
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
When we determine that a function only contains a single return
statement, there is no need to create a temporary variable and store the
result expression into a variable. Instead, we can directly replace the
function-call expression with the return-statement's expression.
This dramatically simplifies the final optimized output from chains of
very simple inlined functions, which is a very common pattern for trees
of Skia fragment processors.
Change-Id: I6789064a321daf43db2e1cef4915f25ed74d6131
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/344665
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
By itself, this is uninteresting and even perhaps slightly
counterproductive (as it separates vardecl from its initializer,
increasing LOC). However, this enables a followup CL
(http://review.skia.org/344665) which allows single-return functions to
be inlined without the creation of a temporary variable at all. This
applies to the majority of fragment processors in a typical Ganesh
hierarchy. This change will greatly reduce the number of inliner-created
temporary copies when compiling a typical tree of FPs.
Change-Id: I03423a13cf35050637dabace4a32973a08a4ed0a
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/344764
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
The additional scopes were harmless, but didn't really add any value.
Originally they were used to tightly scope inlined variables, but we now
mangle inlined variable names.
Change-Id: I7b35e737598036c0b6d3d9f71cbcd4a53d609ce9
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/344757
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
SPIR-V previously didn't know what to think when it encountered a Type
with a typeKind of kEnum, and would abort. These are now treated as
32-bit signed integers.
Metal previously emitted the SkSL enum typename, which is meaningless to
Metal since we do not emit the enum itself anywhere. Metal now emits
"int" for an enum-typed variable.
(GLSL already correctly emits "int" for enum types.)
Change-Id: I05975a2a399f9c4a22c00c90be0dccacd99d793b
Bug: skia:11003
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/338856
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: 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>
We now have SPIR-V golden outputs for `blend` and `shared` tests.
This exposes a handful of SPIR-V limitations for us to address.
Change-Id: Ie5278889b8a61432403d06231b17765885bee0ac
Bug: skia:10694
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/337182
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>
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 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 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 CL is conceptually a revert of http://review.skia.org/320258,
although the code has changed shape a bit since that CL was landed.
This fix was too aggressive, and can lead to functions being dead-
stripped while they still have an active reference.
Change-Id: I6ce8b0ad9cc2a42e8be8cb10d3a8219149eca6aa
Bug: skia:10776
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/325462
Commit-Queue: John Stiles <johnstiles@google.com>
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
This removes VarDeclarationsStatement entirely. VarDeclaration instances
appear directly as statements in Programs. SkSL that declares multiple
variables in a single declaration is transformed to represent that as a
series of VarDeclaration statements.
Similarly, global variable declarations are represented by
GlobalVarDeclaration program elements, one per variable.
Bug: skia:10806
Change-Id: Idd8a2d971a8217733ed57f0dd2249d62f2f0e9c5
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/323102
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
Just a typo fix.
Change-Id: I2fe1f6ae1c99d7f20a4fa5f49eefea514e224652
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/321977
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>
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>
This allows dead-stripping to properly optimize away unreferenced clones
of intrinsic functions, and allows the inliner to detect intrinsic
functions that are only called once (which can generally always be
inlined without penalty).
Change-Id: I0cf034d880ae5d52f4cc0f93de6e2c7aad34e975
Bug: skia:10776
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/320258
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>
Previously, we copied intrinsic functions in a totally arbitrary order;
it used an unordered_set of pointers, so it could be affected by
switching standard libraries OR by malloc nondeterminism. (Surprisingly,
it was fairly consistent in practice on OS X/Linux.) This CL sorts the
intrinsic functions into a consistent order before copying them.
Change-Id: If90342bb77a9ae237a3ce91be3a9847311a722c4
Bug: skia:10749
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/318700
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Ideally the inliner would be smart enough to avoid creating a temporary
at all just for a swizzle, but a good first step is to create fewer of
them.
Change-Id: Icd6f86c294237488f7923dc787bb64a5f99bd0ac
Bug: skia:10737
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/318213
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Early returns can cause the inliner to generate suboptimal code. We
control our built-ins, so let's avoid them where we can.
Patterns like this challenge the inliner:
if (x) return y;
return z;
But this can be replaced by equivalent code that inlines better:
return x ? y : z;
Or, if a ternary can't be used, this also does a better job:
if (x) return y;
else return z;
In several cases, this allows the inliner to avoid generating a
do-while(false) block for control flow.
Change-Id: I921c929122297c40476ff15b4da631fc1581e308
Bug: skia:10737
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/318211
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@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>
Several blend functions generate a surprisingly large amount of code,
and appear to have opportunities for further optimization. At any rate,
if we make compiler changes that would affect the output of a blend
function, I think we would want to see the changes reflected in our
golden outputs.
Change-Id: Iff612dcd4bad8824b5e6e97413ffce19e5ea1c0e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/318336
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
