The CFG/definition map are no longer valid after replacing an expression
entirely. Swizzle-of-swizzle optimization was another case where the
optimizer would replace an expression wholesale, but failed to set the
needs-rescan flag.
Change-Id: Ida0363d738cd1d3ac2a48c824aa04065a7ca16b7
Bug: oss-fuzz:29085
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/351776
Auto-Submit: John Stiles <johnstiles@google.com>
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Change-Id: I75f907ca673ee67f5d623b032128b97833070a0b
Bug: skia:10931
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/351504
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>
Change-Id: If6b23d03b02028b51f96e97080cbd7d34cc33b8f
Bug: skia:10931
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/351503
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 flatten out expressions such as `!false` or `!true`. We
already had a similar fix-up at IR generation time which handled simple
cases, but this will catch more complicated ones like `!sk_Caps.xxxxx`
(since caps bits are only flattened out at constant propagation time).
Change-Id: I04282809d9a784266a64dbcafd097f3b0662806c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/351497
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 is not actually necessary now that constantPropagate can fully
flatten out unary negation into its constant operands. The compilation
results don't change at all.
Change-Id: I7ab55bd3720413609d799dd866e1703973cb2626
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/351202
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 fixes SPIR-V code generation when encountering nested constructors
like `float3 v4 = float3(float2(1), 1.0);` as featured in our unit test
VectorConstructors.sksl.
Change-Id: I3a0c4b466b3cb17ba50bd264f899e59c55c768ed
Bug: skia:11141
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/350032
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@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>
Previously, we had constant-value deduplication, based on the SkSL type
of the constant. However, we were still generating redundant constants,
because we would emit a separate constant for Float(n) and Half(n), or
Int(n) and Short(n), even though we generate the exact same instruction
for these constants. We now deduplicate based on the type's number-kind,
separating constant literals into three categories: floats, signed ints,
and unsigned ints. This better matches our type-handling in
getActualType.
Change-Id: I5777d4b3d567839b7aa72dc8de76908c18fc387e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/350031
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Change-Id: Ia4a1c38161046b94dc56a1a76704766f1e14aab7
Bug: skia:11131
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/350019
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Previously, the IR generator had code which could simplify conversion
constructors like `int(1.23)`. Separately, the optimizer's constant
propagation pass had its own separate implementation of these
simplifications as well.
This CL unifies the two implementations. Previously, the constant-
propagation pass version of the code only supported integer literals, so
this change also improves our code generation slightly.
Change-Id: I32c70a5f2aed210d03bef3166b1178a2d40cdabd
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/350024
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Previously `number(boolean)` casts were converted to a ternary during
IR generation, and `boolean(number)` casts caused an error.
Metal and GLSL should support this cast as written. SPIR-V needed a
little bit of logic to handle converting the boolean to a number via
OpSelect.
Change-Id: I0069781e2b5a26a25c8625ab41c2392342bfd10d
Bug: skia:11131
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/349066
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
The new unit test demonstrates load/store reordering is error-prone.
At head we're allowing loads from a given pointer to reorder later than
a store to that same pointer, and boy, that's just not sound. In the
scenario constructed by the test we reorder this swap,
x = load32 X
y = load32 Y
store32 X y
store32 Y x
using schedule() (following Op argument data dependencies) into
y = load32 Y
store32 X y
x = load32 X
store32 Y x
which moves `x = load32 X` illegally past `store X y`.
We write `y` twice instead of swapping `x` and `y`.
It's not impossible to implement that extra reordering constraint: I
think it's easiest to think about by adding implicit use edges in
schedule() from stores to prior loads of the same pointer. But that'd
be a little complicated to implement, and doesn't handle aliasing at
all, so I decided to ponder on other approaches that handle a wider
range of programs or would have a simpler implementation to reason
about. I ended up walking through this rough chain of ideas:
0) reorder using only Op argument data dependencies (HEAD)
1) don't let load(ptr) pass store(ptr) (above)
2) don't let any load pass any store (allows aliasing)
3) don't reorder any Op that touches memory
4) don't reorder any Op, period.
This CL is 4). It's certainly the easiest and cheapest implementation.
It's not clear to me that we need this scheduling, and should we find we
really want it I'll come back and work back through the list until we
find something that meets our needs.
(Hoisting of uniforms is unaffected here.)
Change-Id: I7765b1d16202e0645b11295f7e30c5e09f2b7339
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/350256
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
The code didn't take into account that x and y might be different types.
(This bug was not actually harmful; type coercion allowed the code to
compile even with the wrong type. The float would be silently splatted
into a vec and the rest of the code would work as-is.)
Change-Id: Ib76bc733f76304e451ef9197421b4bc22e29e49c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/348888
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This actually exposed a latent bug: we don't support bool(1.23) or
bool(1) casts, but these are valid in GLSL:
https://www.khronos.org/opengl/wiki/Data_Type_(GLSL)#Conversion_constructors
"to bool: A value equal to 0 or 0.0 becomes false; anything else is
true."
Change-Id: Ia929a09914ffc96f081d0402d7bb05b5428f8db6
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/349977
Auto-Submit: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Change-Id: Ieb7698d357c9be05ca1f17de84215add54553f84
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/349065
Auto-Submit: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Change-Id: I2c39df532803d827d7cad876021f2ead81145f1d
Bug: skia:10902
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/349064
Auto-Submit: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Previously, the declaration didn't link back to function definition.
This makes the function appear to be undefined, which inhibits inlining
and also makes it difficult for us to validate the presence of a
definition for every called function.
Change-Id: I220ab502634cb3e1d337c23bac150af9aa6370b1
Bug: skia:10902
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/349063
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 did very little to distinguish between a built-in
intrinsic and a user-defined function whose name matches an intrinsic.
This could lead to all sorts of surprising outcomes, as our intrinsic-
rewriting code is able to make assumptions that might not hold true for
arbitrary user-defined functions.
Change-Id: I4180e0c5becdeb6a0a162534eaecfc90dda3392c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/349062
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>
Change-Id: I6c0a6192a78ce60be60a71ed75350ca1bc256d57
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/348890
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Change-Id: I954af70f545a2258babd82af0d43d509201fdc59
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/348889
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Change-Id: I4798263318c504834f23900dbb3f5d167fd17e65
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/348887
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Bug: skia:10913
Change-Id: I430e5eb3fecb0f15775db03699819194d44271b6
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/347958
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
This code was not using typeName() to emit its types, inadvertently
generating Metal code containing the `half` type.
We didn't have any unit tests which synthesized a matrix-construct
helper with half types, so Matrices.sksl was cloned into two separate
test files--MatricesFloat and MatricesHalf. These should be equivalent
except for float vs half types.
Change-Id: I19ecea994b8bc45594bb3f69e596896a3bcefe4d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/348180
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
FrExp testing was moved to the intrinsic tests as part of
http://review.skia.org/341977, and the shared versions were removed from
sksl_tests.gni at that time.
Change-Id: Ife7f3622034d97a77b60d5a98c01f71630c161d6
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/348183
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
In Metal, matrix *= matrix is not natively supported and needs to be
injected via a helper function. This helper function now properly
converts `halfNxM` types to `floatNxM` types (as Metal does not support
half types). It also returns the result by reference instead of by
value to avoid an unnecessary copy.
Matrices.sksl now includes tests for operators += -= *=. Previously we
did not have any coverage for `matrix *= matrix` at all.
Change-Id: I7dfe468ced67eaf7c2405960e8c5efe6f2acf9e4
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/348178
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>
4x4 was dividing a matrix by a scalar - this isn't allowed, multiply by
the scalar's inverse instead.
The types in the signature were derived from type.name(), which wasn't
applying the half->float re-mapping.
Finally, use raw strings so the resulting shader code isn't all crammed
on one line.
Bug: skia:10913
Change-Id: Ie28373fc138445b8c195dbd37687e4ad4504e918
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/348177
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
Previously, SKSL_INT was limited to an int32_t, so we couldn't
differentiate between -1 and 4294967295. We could paper over the
difference in some cases by relying on the expression's type, but this
was imperfect and left us unable to differentiate between an overflow
and valid results. SKSL_INT is now an int64_t; the code has been
updated to fix bugs that shook out as a result of the change.
This isn't a complete solution for overflow handling. There are still
lots of obvious places for improvement--e.g. constant folding can
easily overflow, and statements like `byte x = 1000;` are still
happily accepted.
Change-Id: I30d1f56b6f264543f3aa83046f43c2eb56d5fce4
Bug: skia:10932
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/345173
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Bug: skia:10913
Change-Id: I59f5b0fb2d015f8543b4038c2c5b18ce24c194a8
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/347956
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Previously, some types of overflow were detected, but most would assert
or silently generate invalid code. Now, the parser will properly report
an error if it encounters any integer that exceeds UINT_MAX or any float
that exceeds FLT_MAX.
This fixes test OverflowUintLiteral.sksl. Added a test for floats as
well, OverflowFloatLiteral.sksl.
OverflowIntLiteral.sksl does not fail yet, because its values are larger
than INT_MAX, not UINT_MAX. These are legal from the perspective of the
parser. This must be caught later at IR generation time.
Change-Id: Ia5a904d01427cdc9f2ab5f4174154418737835e6
Bug: skia:10932
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/347176
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This fix is overly conservative in some situations (identity conversions
among vectors with the same component type), but fixes errors in two
existing unit test cases.
Bug: skia:11116
Change-Id: If852f8591fb26817528fdc37191c49129e17d6b3
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/347053
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
This feature had devolved to just an assert, and one that isn't really
necessary - all of Ganesh is built to handle any child processor being
null. The next step is to remove nullable types entirely -- a large
amount of code.
Change-Id: I612a5867f8690400b405aa1f5c929e76cf5918fd
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/347050
Commit-Queue: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
This CL updates `compareConstant` to fail gracefully instead of
aborting if the passed-in types don't match. This lets us call
`compareConstant` without checking types first.
Change-Id: Id2acdbdf700e64bcb24825cdad2c0e000992e8cb
Bug: oss-fuzz:28904
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/347038
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
These are GLSL-isms that weren't really implemented - each one was a
"generic" type that only resolved to a single underlying type. We've
got along just fine without them for years, so update our sample()
declarations to take the actual underlying type. (Note that we had
worked around this by declaring an integer version of sample where
necessary, so we can presumably keep doing that in the future).
Change-Id: I4c46a2fa0c1f19e6278298c8005a2760329e7abf
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/347040
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
Nullable fragment processors still exist, but they're handled
transparently by sample() within C++, so there's no need for .fp files
to ever do these tests manually.
Change-Id: Idf2bc58505207560553066c0126a2a036c5d970b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/347039
Commit-Queue: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
Opaque types can no longer be copied via assignment or construction, and
various restrictions originally applied to the "fragmentProcessor" type
have been extended to cover opaque types in general.
Change-Id: I55ab7aefd1e6ef277e56a9408b430e1de5ba12ca
Bug: skia:11027
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/346264
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 intrinsic was previously lacking a unit test, and wasn't actually
implemented in Metal or SPIR-V. Fortunately it's trivial to add.
Change-Id: I68bbdc58376b579c7f3f0ae5f49323b389c2b8c4
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/346263
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, a return statement inside a scoped Block would always result
in the return expression being assigned to a temporary variable instead
of replacing the function-call-expression directly. This was done
because there might be variables inside the Block; these would have
fallen out of scope when the expression is migrated to the call site,
resulting in an invalid expression.
We aren't actually examining the return expression so we don't know if
it uses variables from an inner scope at all. (Inspecting the return
expression for variable usage is certainly possible! But it's a fair
amount of code and complexity for a small payoff.)
However, we can very easily get most of the benefit here without paying
for the complexity. In this CL we now look for variable declarations
inside of scoped Blocks. If the code doesn't add any vardecls into
scoped Blocks, there's no risk of scope problems, and we don't need to
use a temp-var to store our return expressions. If any vardecls are
added, we go back to using a temp-var as before.
Change-Id: I4c81400dad2f33db06a1c18eb671ba2140232006
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/346499
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Also renamed $matH to $hmat, to match $hvec convention. Runtime effects
will only support square matrices (like ES2), so this lets us declare
intrinsics like matrixCompMult correctly (and differently) for public
vs. private usage.
Bug: skia:11093
Change-Id: I457d83e4c5e09f8e01e7b8acb116c39ff17e52c3
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/346656
Auto-Submit: Brian Osman <brianosman@google.com>
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: John Stiles <johnstiles@google.com>
Includes a handful of test cases to exercise the system
Change-Id: I98e73a8bca063f475d2ddb51778e395697392ddb
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/346637
Reviewed-by: John Stiles <johnstiles@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
We have a handful of tests that demonstrate this behavior indirectly,
but lacked a focused test.
Change-Id: I895cc4e3bebf30721ed649244e42bf170cc6ec06
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/346497
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>
We need to rescan after optimizing away expressions that might exist
in the CFG/definition map, since we are rebuilding them from scratch and
not just stripping off excess parts from them.
Change-Id: I843a2ea3fc38428e7c0bd0e2bf7a7d41101345e3
Bug: oss-fuzz:28794
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/344972
Commit-Queue: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
This reverts commit 4129b6b65f.
Reason for revert: WASM breakage: https://task-driver.skia.org/td/UBRwnWYfbc5IwUWqtFMv
Original change's description:
> Revert "Reland "Reland "Reland "Revert "Initial land of SkSL DSL."""""
>
> This reverts commit 346dd53ac0.
>
> Change-Id: I93bb18438cc6c2ad43d058d6c3f95bcc65d0cea9
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/343916
> Commit-Queue: Ethan Nicholas <ethannicholas@google.com>
> Reviewed-by: John Stiles <johnstiles@google.com>
TBR=brianosman@google.com,ethannicholas@google.com,johnstiles@google.com
# Not skipping CQ checks because this is a reland.
Change-Id: If05145cf9d9c51f4c76fe523f6050a670b5da669
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/345169
Reviewed-by: Ethan Nicholas <ethannicholas@google.com>
Commit-Queue: John Stiles <johnstiles@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>
do-while loops aren't compatible with GLSL ES2. For-loops which run
only one time should work exactly the same for our purposes. We expect
such a loop to be unrolled by every driver, so it shouldn't come at any
performance cost.
Change-Id: Ia8de5fcab8128c34da97eaeaf81f91ad1ac36ce4
Bug: skia:11097
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/345159
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Fortunately, this had an existing opcode, so it was easy to add to our
intrinsics list, and the rest automatically worked.
Change-Id: Idcd5a2c46d6bf10c05c702faba4280a270c54929
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/345398
Commit-Queue: John Stiles <johnstiles@google.com>
Auto-Submit: John Stiles <johnstiles@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
We are still missing an implementation for Metal and SPIR-V, but at
least it's correct on GLSL now.
Change-Id: I5b365384eaefacb00faf6af7bda9b690cba00de5
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/345397
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>
