Related to bug 1608121.
When we use Ion and WebAssembly, we might get multiple return values
from function calls. This patch refactors the direct JS-to-Wasm-via-Ion
call path to better support multiple values.
The patch adds a safepoint at the call, as appears to be necessary in
general, and also when results are captured as MIR values. There is a
slight optimization in that void-returning Wasm functions just get a
MConstant result that's visible to the optimizer.
Differential Revision: https://phabricator.services.mozilla.com/D60228
--HG--
extra : moz-landing-system : lando
Cranelift fails this assertion and it's not clear to me that there is any restriction inside
Cranelift that would limit the set of registers we'd need to check here.
Differential Revision: https://phabricator.services.mozilla.com/D58887
--HG--
extra : moz-landing-system : lando
This commit adds support for receiving stackmaps from Cranelift
and converting to wasm::Stackmaps. [1] will change the stackmap
representation in Cranelift to be the same as in Spidermonkey.
The stack overflow/interrupt trap handler stackmap is implemented by
sharing code with Ion.
[1] https://github.com/bytecodealliance/cranelift/pull/1317
Differential Revision: https://phabricator.services.mozilla.com/D58886
--HG--
extra : moz-landing-system : lando
cranelift-wasm cannot express arbitrary control flow inside of
'global.get/set' easily. This makes it difficult to implement the pre/post write
barriers inline like the baseline compiler does.
[1] has a discussion about the ideal solution to problem, which involves a new IR
concept called templates. We will want to use that when it is ready.
In the short-term, this commit implements pre/post write barriers using slow OOL
functions and implements Wasm globals.
[1] https://github.com/bytecodealliance/cranelift/issues/1176
Differential Revision: https://phabricator.services.mozilla.com/D58885
--HG--
extra : moz-landing-system : lando
This commit translates the basic parts of reference types that does not
require any barriers. AnyRef,FuncRef,NullRef are translated to R32/R64. The
existing bulk-memory instructions are extended to operate on multiple tables,
and the new reference type instructions are implemented with instance calls.
Differential Revision: https://phabricator.services.mozilla.com/D58884
--HG--
extra : moz-landing-system : lando
This commit changes the wasm compiler feature detection bits to report that
cranelift supports reference types.
Differential Revision: https://phabricator.services.mozilla.com/D58883
--HG--
extra : moz-landing-system : lando
It used to be that some XPIDL interfaces were allowed to be non-nsISupports,
and were only used to hold a bunch of constants. It appears that we've now
had enough de-COM work happen that there are no longer any, so we can remove
support for this.
Differential Revision: https://phabricator.services.mozilla.com/D61008
--HG--
extra : moz-landing-system : lando
For builtinclass interfaces, or ones not declared in IDL, we don't call out
into scripted QueryInterface anyway, so we can return earlier, before we ever
set up the AutoEntryScript.
Differential Revision: https://phabricator.services.mozilla.com/D61005
--HG--
extra : moz-landing-system : lando
Cranelift fails this assertion and it's not clear to me that there is any restriction inside
Cranelift that would limit the set of registers we'd need to check here.
Differential Revision: https://phabricator.services.mozilla.com/D58887
--HG--
extra : moz-landing-system : lando
This commit adds support for receiving stackmaps from Cranelift
and converting to wasm::Stackmaps. [1] will change the stackmap
representation in Cranelift to be the same as in Spidermonkey.
The stack overflow/interrupt trap handler stackmap is implemented by
sharing code with Ion.
[1] https://github.com/bytecodealliance/cranelift/pull/1317
Differential Revision: https://phabricator.services.mozilla.com/D58886
--HG--
extra : moz-landing-system : lando
cranelift-wasm cannot express arbitrary control flow inside of
'global.get/set' easily. This makes it difficult to implement the pre/post write
barriers inline like the baseline compiler does.
[1] has a discussion about the ideal solution to problem, which involves a new IR
concept called templates. We will want to use that when it is ready.
In the short-term, this commit implements pre/post write barriers using slow OOL
functions and implements Wasm globals.
[1] https://github.com/bytecodealliance/cranelift/issues/1176
Differential Revision: https://phabricator.services.mozilla.com/D58885
--HG--
extra : moz-landing-system : lando
This commit translates the basic parts of reference types that does not
require any barriers. AnyRef,FuncRef,NullRef are translated to R32/R64. The
existing bulk-memory instructions are extended to operate on multiple tables,
and the new reference type instructions are implemented with instance calls.
Differential Revision: https://phabricator.services.mozilla.com/D58884
--HG--
extra : moz-landing-system : lando
This commit changes the wasm compiler feature detection bits to report that
cranelift supports reference types.
Differential Revision: https://phabricator.services.mozilla.com/D58883
--HG--
extra : moz-landing-system : lando
It used to be that some XPIDL interfaces were allowed to be non-nsISupports,
and were only used to hold a bunch of constants. It appears that we've now
had enough de-COM work happen that there are no longer any, so we can remove
support for this.
Differential Revision: https://phabricator.services.mozilla.com/D61008
--HG--
extra : moz-landing-system : lando
It used to be that some XPIDL interfaces were allowed to be non-nsISupports,
and were only used to hold a bunch of constants. It appears that we've now
had enough de-COM work happen that there are no longer any, so we can remove
support for this.
Differential Revision: https://phabricator.services.mozilla.com/D61008
--HG--
extra : moz-landing-system : lando
Cranelift fails this assertion and it's not clear to me that there is any restriction inside
Cranelift that would limit the set of registers we'd need to check here.
Differential Revision: https://phabricator.services.mozilla.com/D58887
--HG--
extra : moz-landing-system : lando
This commit adds support for receiving stackmaps from Cranelift
and converting to wasm::Stackmaps. [1] will change the stackmap
representation in Cranelift to be the same as in Spidermonkey.
The stack overflow/interrupt trap handler stackmap is implemented by
sharing code with Ion.
[1] https://github.com/bytecodealliance/cranelift/pull/1317
Differential Revision: https://phabricator.services.mozilla.com/D58886
--HG--
extra : moz-landing-system : lando
cranelift-wasm cannot express arbitrary control flow inside of
'global.get/set' easily. This makes it difficult to implement the pre/post write
barriers inline like the baseline compiler does.
[1] has a discussion about the ideal solution to problem, which involves a new IR
concept called templates. We will want to use that when it is ready.
In the short-term, this commit implements pre/post write barriers using slow OOL
functions and implements Wasm globals.
[1] https://github.com/bytecodealliance/cranelift/issues/1176
Differential Revision: https://phabricator.services.mozilla.com/D58885
--HG--
extra : moz-landing-system : lando
This commit translates the basic parts of reference types that does not
require any barriers. AnyRef,FuncRef,NullRef are translated to R32/R64. The
existing bulk-memory instructions are extended to operate on multiple tables,
and the new reference type instructions are implemented with instance calls.
Differential Revision: https://phabricator.services.mozilla.com/D58884
--HG--
extra : moz-landing-system : lando
This commit changes the wasm compiler feature detection bits to report that
cranelift supports reference types.
Differential Revision: https://phabricator.services.mozilla.com/D58883
--HG--
extra : moz-landing-system : lando
Most of the complexity of this patch is trying to preserve the same level of
assertion in the future case where a BytecodeEmitter may or may not have a
pre-allocated script.
Differential Revision: https://phabricator.services.mozilla.com/D60712
--HG--
extra : moz-landing-system : lando
The fields these methods deal with are not dead: they are just
interacted with via other means.
I found these via the code coverage report for XPCInlines.h.
Differential Revision: https://phabricator.services.mozilla.com/D60863
--HG--
extra : moz-landing-system : lando
Move allocation into a function call can prevent JIT keeps the return
value of weakRef.deref().
Differential Revision: https://phabricator.services.mozilla.com/D60662
--HG--
extra : moz-landing-system : lando
We need this for upcoming change which supports having [AllowShared] on
ArrayBuffer type in WebIDL.
Differential Revision: https://phabricator.services.mozilla.com/D59989
--HG--
extra : moz-landing-system : lando
Tests that involve timing are often problematic. Here we're checking that hitting the heap limit inside the five second time limit we set for skipping last ditch GCs doesn't trigger a GC, but allocating enough to hit this limit can take longer than five seconds on a busy system.
The patch makes the max heap size much smaller and so reduces the possibility that that this can happen.
Differential Revision: https://phabricator.services.mozilla.com/D60655
--HG--
extra : moz-landing-system : lando
In particular, we can't do effectful .prototype lookups in CreateThisFromIon and
elsewhere before the call instruction. This patch returns NullValue from
CreateThisFromIon in that case and changes LCallGeneric to take the slow
invoke path when it sees this value.
Differential Revision: https://phabricator.services.mozilla.com/D59507
--HG--
extra : moz-landing-system : lando
The callers are very different and have different constraints, especially in
later patches in this stack.
Differential Revision: https://phabricator.services.mozilla.com/D59505
--HG--
extra : moz-landing-system : lando
The ObjLiteral format allows for names (as atom indices) or numeric
indices to be attached to each property in the literal. The numeric case
is limited to 23 bits by the packed format in memory.
The object-literal case ({}-literal with some numeric indices) already
checks for in-range values before deciding to use the ObjLiteral
infrastructure, but the array case did not check this.
Fortunately, the indices in the array case were only used as a sanity
check for debug-mode assertions, and never for any actual indexing
operation, so no out-of-bounds behavior could result, even before this
fix. However, the sanity-check assert will fire at
object-construction-from-ObjLiteral time in cases where the ObjLiteral
mode should not have been used, because no range check was done at
ObjLiteral-construction time.
This change removes the use of indices for array-mode ObjLiterals
entirely.
Differential Revision: https://phabricator.services.mozilla.com/D60474
--HG--
extra : moz-landing-system : lando
This in turn requires fixing various other include hell / missing include
issues.
Differential Revision: https://phabricator.services.mozilla.com/D60337
--HG--
extra : moz-landing-system : lando
The ObjLiteral format allows for names (as atom indices) or numeric
indices to be attached to each property in the literal. The numeric case
is limited to 23 bits by the packed format in memory.
The object-literal case ({}-literal with some numeric indices) already
checks for in-range values before deciding to use the ObjLiteral
infrastructure, but the array case did not check this.
Fortunately, the indices in the array case were only used as a sanity
check for debug-mode assertions, and never for any actual indexing
operation, so no out-of-bounds behavior could result, even before this
fix. However, the sanity-check assert will fire at
object-construction-from-ObjLiteral time in cases where the ObjLiteral
mode should not have been used, because no range check was done at
ObjLiteral-construction time.
This change removes the use of indices for array-mode ObjLiterals
entirely.
Differential Revision: https://phabricator.services.mozilla.com/D60474
--HG--
extra : moz-landing-system : lando
This was done by:
This was done by applying:
```
diff --git a/python/mozbuild/mozbuild/code-analysis/mach_commands.py b/python/mozbuild/mozbuild/code-analysis/mach_commands.py
index 789affde7bbf..fe33c4c7d4d1 100644
--- a/python/mozbuild/mozbuild/code-analysis/mach_commands.py
+++ b/python/mozbuild/mozbuild/code-analysis/mach_commands.py
@@ -2007,7 +2007,7 @@ class StaticAnalysis(MachCommandBase):
from subprocess import Popen, PIPE, check_output, CalledProcessError
diff_process = Popen(self._get_clang_format_diff_command(commit), stdout=PIPE)
- args = [sys.executable, clang_format_diff, "-p1", "-binary=%s" % clang_format]
+ args = [sys.executable, clang_format_diff, "-p1", "-binary=%s" % clang_format, '-sort-includes']
if not output_file:
args.append("-i")
```
Then running `./mach clang-format -c <commit-hash>`
Then undoing that patch.
Then running check_spidermonkey_style.py --fixup
Then running `./mach clang-format`
I had to fix four things:
* I needed to move <utility> back down in GuardObjects.h because I was hitting
obscure problems with our system include wrappers like this:
0:03.94 /usr/include/stdlib.h:550:14: error: exception specification in declaration does not match previous declaration
0:03.94 extern void *realloc (void *__ptr, size_t __size)
0:03.94 ^
0:03.94 /home/emilio/src/moz/gecko-2/obj-debug/dist/include/malloc_decls.h:53:1: note: previous declaration is here
0:03.94 MALLOC_DECL(realloc, void*, void*, size_t)
0:03.94 ^
0:03.94 /home/emilio/src/moz/gecko-2/obj-debug/dist/include/mozilla/mozalloc.h:22:32: note: expanded from macro 'MALLOC_DECL'
0:03.94 MOZ_MEMORY_API return_type name##_impl(__VA_ARGS__);
0:03.94 ^
0:03.94 <scratch space>:178:1: note: expanded from here
0:03.94 realloc_impl
0:03.94 ^
0:03.94 /home/emilio/src/moz/gecko-2/obj-debug/dist/include/mozmemory_wrap.h:142:41: note: expanded from macro 'realloc_impl'
0:03.94 #define realloc_impl mozmem_malloc_impl(realloc)
Which I really didn't feel like digging into.
* I had to restore the order of TrustOverrideUtils.h and related files in nss
because the .inc files depend on TrustOverrideUtils.h being included earlier.
* I had to add a missing include to RollingNumber.h
* Also had to partially restore include order in JsepSessionImpl.cpp to avoid
some -WError issues due to some static inline functions being defined in a
header but not used in the rest of the compilation unit.
Differential Revision: https://phabricator.services.mozilla.com/D60327
--HG--
extra : moz-landing-system : lando
rg -l 'mozilla/Move.h' | xargs sed -i 's/#include "mozilla\/Move.h"/#include <utility>/g'
Further manual fixups and cleanups to the include order incoming.
Differential Revision: https://phabricator.services.mozilla.com/D60323
--HG--
extra : moz-landing-system : lando
Now mfbt/Move.h is empty except for that excellent comment about move
semantics... Should we put it somewhere else and delete the header as a
follow-up? Or just delete the header and carry on?
Differential Revision: https://phabricator.services.mozilla.com/D60297
--HG--
extra : moz-landing-system : lando
By cleaning up the register set APIs very slightly we can simplify the wrappers and
make space for meaningful assertions.
Differential Revision: https://phabricator.services.mozilla.com/D59673
--HG--
extra : moz-landing-system : lando
Create a clearer distinction between the register's Encoding, which is
its hardware name, and its Code, which is a dense encoding of
bitwidth+Encoding along with a distinguished Invalid value. These
concepts exist already but it gets out of hand when the FloatRegister
uses a Code to encode the Encoding.
Make FloatRegister contain separate fields for bitwidth, encoding, and
validity, as it does on other platforms.
Add assertions on validity of inputs and on the validity of the
FloatRegister for some operations. And tidy up some, and rearrange
the file to mirror the x86 file as much as possible.
Expand the register name table so that it covers the possible range of
Code and so that we won't reference the table OOB.
Differential Revision: https://phabricator.services.mozilla.com/D59912
--HG--
extra : moz-landing-system : lando
By cleaning up the register set APIs very slightly we can simplify the wrappers and
make space for meaningful assertions.
Differential Revision: https://phabricator.services.mozilla.com/D59673
--HG--
extra : moz-landing-system : lando
