This allows us to remove 100 LOC from the inliner and is very unlikely
to affect any existing benchmark. We don't have any evidence to support
the idea that a one-iteration `for` loop with `continue`-based exits
will be any faster than a standard function call on any existing GPU.
Our fragment processors are generally written to avoid early returns,
in large part to avoid hitting this path.
This drastically impacts BlendEnum.sksl (which can no longer flatten out
a switch over every blend function in SkSL) but is otherwise a wash.
See: http://go/optimization-in-sksl-inliner suggestion 4(a)
Change-Id: I1f9c27bcd7a8de46cc4e8d0b9768d75957cf1c50
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/385377
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This reverts commit 92748af1a5.
Reason for revert: SkSLCommaSideEffects_GPU crashing on Android
Original change's description:
> Inline functions of the form 'return (expr)' only.
>
> This drastically reduces the number of functions which we allow to be
> inlined. If this change does not hurt our performance, it will allow us
> to trivially remove hundreds of LOC. All current data leads us to
> believe that it may affect the Mali 400 but is highly unlikely to change
> results on any other device in the tree.
>
> More info: http://go/optimization-in-sksl-inliner
>
> Change-Id: Ia6b706742ce5407453e0e697b6c1f9201084c0e8
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/384858
> Auto-Submit: John Stiles <johnstiles@google.com>
> Commit-Queue: John Stiles <johnstiles@google.com>
> Reviewed-by: Brian Osman <brianosman@google.com>
> Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
TBR=brianosman@google.com,ethannicholas@google.com,johnstiles@google.com
Change-Id: I6a670dacaa58fe3386ff50375ac6d1cac4fd7f2c
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/385161
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
This drastically reduces the number of functions which we allow to be
inlined. If this change does not hurt our performance, it will allow us
to trivially remove hundreds of LOC. All current data leads us to
believe that it may affect the Mali 400 but is highly unlikely to change
results on any other device in the tree.
More info: http://go/optimization-in-sksl-inliner
Change-Id: Ia6b706742ce5407453e0e697b6c1f9201084c0e8
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/384858
Auto-Submit: John Stiles <johnstiles@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
I ran into an issue in an upcoming CL which generated a particularly
ugly switch statement:
switch (x) {
default:
discard;}
So I cleaned this up, and while resolving this issue, managed to improve
a bunch of existing codegen as well. The formatting change has been
split out to a separate CL since it impacts so many golden outputs.
Change-Id: I7a6be29903c47560dcc7f6acd3ef15fd0c5c3c50
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/384179
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Change-Id: Ic2d1240ab785101365b0fd934562505fb5a3e599
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/381816
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
The optimizer now properly recognizes all types of exits from a switch
statement. Break, continue and return are all potential exits and need
to be considered when determining the exit path from the switch.
Previously, dead code elimination was hiding the effects of this bug
from us, but it meant that an optimized switch had the potential to
generate lots of worthless IR nodes which then needed to be detected and
eliminated by the CFG. In particular, this affected the enum form of
blend, causing a catastrophic amount of extra work to be done.
Change-Id: If857e38cadfc016884624ea4db25a273ad3dce5b
Bug: skia:11352
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/372958
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
When we detect a static switch, the optimizer finds the matching switch-
case and eliminates all the other switch-cases. It handles case
fall-through by scanning forward and looking for an unconditional break.
However, the inliner has an interesting quirk--it can replace `return`
statements inside of a switch with `continue` statements, since the body
of the inlined function has been wrapped with a for-loop to allow for
early exits. The optimizer does not recognize these continue statements
as exits from the switch (although they certainly qualify), so it
treats continues as fallen-through and keeps emitting switch-cases.
The dead-code elimination pass was actually doing us a favor here and
eliminating the excess code later. A flag was added to disable DCE in
order to reveal the problem in a test.
Change-Id: I8ff19fde5e32d0ab73d7c5411da40cb953a446f5
Bug: skia:11352
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/372956
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
