This patch moves us away from the accumulator-based bytecode format to
one with explicit source and destination registers.
The new format has multiple benefits:
- ~25% faster on the Kraken and Octane benchmarks :^)
- Fewer instructions to accomplish the same thing
- Much easier for humans to read(!)
Because this change requires a fundamental shift in how bytecode is
generated, it is quite comprehensive.
Main implementation mechanism: generate_bytecode() virtual function now
takes an optional "preferred dst" operand, which allows callers to
communicate when they have an operand that would be optimal for the
result to go into. It also returns an optional "actual dst" operand,
which is where the completion value (if any) of the AST node is stored
after the node has "executed".
One thing of note that's new: because instructions can now take locals
as operands, this means we got rid of the GetLocal instruction.
A side-effect of that is we have to think about the temporal deadzone
(TDZ) a bit differently for locals (GetLocal would previously check
for empty values and interpret that as a TDZ access and throw).
We now insert special ThrowIfTDZ instructions in places where a local
access may be in the TDZ, to maintain the correct behavior.
There are a number of progressions and regressions from this test:
A number of async generator tests have been accidentally fixed while
converting the implementation to the new bytecode format. It didn't
seem useful to preserve bugs in the original code when converting it.
Some "does eval() return the correct completion value" tests have
regressed, in particular ones related to propagating the appropriate
completion after control flow statements like continue and break.
These are all fairly obscure issues, and I believe we can continue
working on them separately.
The net test262 result is a progression though. :^)
This is pure prep work for refactoring the bytecode to use more operands
instead of only registers.
generate_bytecode() virtuals now return an Optional<Operand>, and the
idea is to return an Operand referring to the value produced by this
AST node.
They also take an Optional<Operand> "preferred_dst" input. This is
intended to communicate the caller's preference for an output operand,
if any. This will be used to elide temporaries when we can store the
result directly in a local, for example.
The JIT compiler was an interesting experiment, but ultimately the
security & complexity cost of doing arbitrary code generation at runtime
is far too high.
In subsequent commits, the bytecode format will change drastically, and
instead of rewriting the JIT to fit the new bytecode, this patch simply
removes the JIT instead.
Other engines, JavaScriptCore in particular, have already proven that
it's possible to handle the vast majority of contemporary web content
with an interpreter. They are currently ~5x faster than us on benchmarks
when running without a JIT. We need to catch up to them before
considering performance techniques with a heavy security cost.
When an <input type=image> button is clicked, we now send the (x,y)
coordinates of the click event (relative to the image) along with the
form submission data.
Regarding the text test, we can currently only test this feature with
dialogs. The headless-browser test infrastructure cannot yet handle the
resulting navigation that would occur if we were to test with normal
form submission.
HTMLFormElement::elements is not the correct filter for submittable
elements. It includes non-submittable elements (HTMLObjectElement) and
also excludes submittable elements (HTMLInputElements in the "image"
type state, "for historical reasons").
This implements enough to represent <input type=image> with its loaded
source image (or fallback to its alt text, if applicable). This does not
implement acquring coordinates from user-activated click events on the
image.
They currently assume the DOM node is an HTMLImageElement with respect
to handling the alt attribute. The HTMLInputElement will require the
same behavior.
Previously, when constructing an XML document, the default namespace
was the empty string. This led to XML documents having empty xmlns
attributes when serialized.
Previously, CDATASection nodes were being serialized as if they were
text, which meant they were missing their start and end delimiters.
Occurrences of the '&', '<' and '>' characters were also being replaced
with their entity names.
The DOM specification states that: "Unless stated otherwise, a
document’s [...] type is 'xml'".
Previously, calls to `Document::document_type()` were returning the
incorrect value for non-HTML documents.
Our current CLUT code is pretty slow. A one-element cache can make
it quite a bit faster for images that have long runs of a single
color, such as illustrations.
It also seems to help with photos some (in `0000711.pdf` page 1) --
I suppose even them have enough repeating pixels for it to be worth
it.
Some numbers, with `pdf --render-bench`:
`0000711.pdf --page 1` (high-res photo)
before: 2.9s
after: 2.43s
`0000277.pdf --page 19` (my nemesis PDF, large-ish illustration)
before: 2.17s
after: 0.58s (!)
`0000502.pdf --page 2` (wat hoe dat)
before: 0.66s
after: 0.27s
`0000521.pdf --page 10 ` (japanese)
before: 0.52s
after: 0.29s
`0000364.pdf --page 1` (4 min test case)
before: 0.48s
after: 0.19s
Thanks to that last one, reduces the time for
`time Meta/test_pdf.py ~/Downloads/0000` from 4m22s to 1m28s.
Helps quite a bit with #23157 (but high-res photos are still too
slow).
This reverts commit a4b2e5b27b.
This was just plain wrong, I remember it making sense and fixing
something but that was probably due to local changes. It should never
have landed on master, my bad.
This allows the user to pass very specific values instead of a u64 value
that is later cast to the underlying type.
The change also adds support for passing v128 values:
- v(i64.const 4) (splat 4)
- v128.const 0x4 (just a few bits specificed, everything else = 0)
This function was called over and over in `manage_extra_channels()`,
even if the result depends only on the metadata. Instead, we now call it
once and store the result.
Of my 1000 test files, 73 have stream Type0 truetype fonts with stream
CIDToGIDMaps. This makes that work.
(With this patch, the number of files in my 1000 test files complaining
"Font is missing Name" increases from 41 to 75, so a bit under half of
the fonts using stream CIDToGIDMaps also have no 'name' table. So that's
next.)
Increases files without issues from 652 to 681.
Previously, during the start of the game BoardView component tried to
get the font from the database passing empty string as a parameter.
It caused a debug message informing user that the lookup failed.
Painting command executors are defined within the "Painting" namespace,
allowing us to remove this prefix from their names.
This commit performs the following renamings:
- Painting::PaintingCommandExecutor to Painting::CommandExecutor
- Painting::PaintingCommandExecutorCPU to Painting::CommandExecutorCPU
- Painting::PaintingCommandExecutorGPU to Painting::CommandExecutorGPU
Separating the recorder list from the painter will allow us to save it
for later execution without carrying along the painter's state. This
will be useful once we have a separate thread for executing painting
commands, to which we will have to transfer commands from the main
thread.
Preparation for https://github.com/SerenityOS/serenity/pull/23108