Reasons for doing this:
(1) Bug 1501316 becomes easier to fix.
(2) JSOP_TABLESWITCH is no longer a variable-length bytecode op so we can get
rid of js::GetVariableBytecodeLength.
Depends on D11018
Differential Revision: https://phabricator.services.mozilla.com/D11019
--HG--
extra : moz-landing-system : lando
With this change we no longer have to handle the offset == 0 case everywhere.
Differential Revision: https://phabricator.services.mozilla.com/D11018
--HG--
extra : moz-landing-system : lando
The baseline compiler should always call builtins assuming the hardfp
calling convention on ARM, but it would actually take the native
calling convention on the device into account and sometimes use the
softfp convention.
The reason the baseline compiler should always use hardfp is that the
builtin thunks already convert hardfp->softfp along the call path and
softfp->hardfp along the return path, if this is necessary, to allow
wasm to call builtins using the hardfp convention always.
It is possible that the situation was different when the baseline
compiler was written and that the bug is the result of subsequent
changes to the thunk layer, but this is not known precisely.
There's a driveby fix here to simplify the logic around determining
hardfp vs softfp for the system ABI; UseHardFpABI() is now always
available and does the right thing, we don't need the #ifdef nest we
had previously.
--HG--
extra : rebase_source : 5b82e0d25cad05f9064e859282dd9d886da1e672
extra : amend_source : 13b9627ddda0127c7e34fe6b01f3d252f5bfba42
This makes it easier to read these flags from JIT code. The patch also splits
them in MutableFlags and ImmutableFlags, this should let us simplify XDR and
CopyScript in the future.
Differential Revision: https://phabricator.services.mozilla.com/D10735
--HG--
extra : moz-landing-system : lando
Also uses JS_DefinePropertyById instead of JS_SetPropertyById because it's more natural.
Differential Revision: https://phabricator.services.mozilla.com/D11094
--HG--
extra : moz-landing-system : lando
This change reduces the binary size on macOS x64 by around 50KB.
Here's a diff of the impact on the code generated for Attr_Binding::get_specified
in the Mac build. It's a bit hard to read because %r12 and %rbx swap their
function, but what happens in this method is that "movq %r12, %rcx" goes
away, and the two instructions "leal 0x1(%r12) %eax" and
"movl %eax, 0x10(%rbx)" turn into an "incl 0x10(%r12)".
So the old code was preserving the original value of profilingStack->stackPointer
in a register, and then using it later to compute the incremented stackPointer.
The new code uses an "incl" instruction for the stackPointer increment and
doesn't worry that the stackPointer value might have changed since the stack
size check at the start of the function. (It can't have changed.)
before: %rbx has the ProfilingStack*, %r12 has profilingStack->stackPointer
after: %r12 has the ProfilingStack*, %rbx has profilingStack->stackPointer
@@ -3,37 +3,35 @@
movq %rsp, %rbp
pushq %r15
pushq %r14
pushq %r12
pushq %rbx
subq $0x10, %rsp
movq %rcx, %r14
movq %rdx, %r15
- movq 0x80(%rdi), %rbx
- movq %rbx, -40(%rbp)
- testq %rbx, %rbx
+ movq 0x80(%rdi), %r12
+ movq %r12, -40(%rbp)
+ testq %r12, %r12
je loc_xxxxx
- movl 0x10(%rbx), %r12d
- cmpl (%rbx), %r12d
+ movl 0x10(%r12), %ebx
+ cmpl (%r12), %ebx
jae loc_xxxxx
- movq 0x8(%rbx), %rax
- movq %r12, %rcx
- shlq $0x5, %rcx
- leaq aAttr, %rdx ; "Attr"
- movq %rdx, (%rax,%rcx)
- leaq aSpecified, %rdx ; "specified"
- movq %rdx, 0x8(%rax,%rcx)
- leaq -40(%rbp), %rdx
- movq %rdx, 0x10(%rax,%rcx)
- movl $0x3a1, 0x1c(%rax,%rcx)
- leal 0x1(%r12), %eax
- movl %eax, 0x10(%rbx)
+ movq 0x8(%r12), %rax
+ shlq $0x5, %rbx
+ leaq aAttr, %rcx ; "Attr"
+ movq %rcx, (%rax,%rbx)
+ leaq aSpecified, %rcx ; "specified"
+ movq %rcx, 0x8(%rax,%rbx)
+ leaq -40(%rbp), %rcx
+ movq %rcx, 0x10(%rax,%rbx)
+ movl $0x3a1, 0x1c(%rax,%rbx)
+ incl 0x10(%r12)
movq %r15, %rdi
call __ZNK7mozilla3dom4Attr9SpecifiedEv ; mozilla::dom::Attr::Specified() const
movzxl %al, %eax
movabsq $0xfff9000000000000, %rcx
orq %rax, %rcx
movq %rcx, (%r14)
movq -40(%rbp), %rax
@@ -47,11 +45,11 @@
popq %rbx
popq %r12
popq %r14
popq %r15
popq %rbp
ret
; endp
- movq %rbx, %rdi
+ movq %r12, %rdi
call __ZN14ProfilingStack18ensureCapacitySlowEv ; ProfilingStack::ensureCapacitySlow()
jmp loc_xxxxx
Depends on D9205
Differential Revision: https://phabricator.services.mozilla.com/D9206
--HG--
extra : moz-landing-system : lando
These flags will be used by WebIDL APIs in an upcoming patch.
Depends on D9199
Differential Revision: https://phabricator.services.mozilla.com/D9203
--HG--
extra : moz-landing-system : lando
They were not displayed in the UI, and the instructions to initialize the line
field of a stack frame increased code size unnecessarily.
This change reduces the binary size on Linux x64 by around 100KB.
Here's a diff of the impact on the code generated for Attr_Binding::get_specified
in the Mac build:
@@ -20,17 +20,16 @@
movq 0x8(%rbx), %rax
movq %r12, %rcx
shlq $0x5, %rcx
leaq aGetAttrspecifi, %rdx ; "get Attr.specified"
movq %rdx, (%rax,%rcx)
movq $0x0, 0x8(%rax,%rcx)
leaq -40(%rbp), %rdx
movq %rdx, 0x10(%rax,%rcx)
- movl $0x106, 0x18(%rax,%rcx)
movl $0x1c, 0x1c(%rax,%rcx)
leal 0x1(%r12), %eax
movl %eax, 0x10(%rbx)
movq %r15, %rdi
call __ZNK7mozilla3dom4Attr9SpecifiedEv ; mozilla::dom::Attr::Specified() const
movzxl %al, %eax
movabsq $0xfff9000000000000, %rcx
Depends on D9193
Differential Revision: https://phabricator.services.mozilla.com/D9195
--HG--
extra : moz-landing-system : lando
This eliminates a few instructions from each inlined instance of
AutoProfilerLabel because we no longer need to handle allocation failure in the
inlined code.
I think this allocation should be fine to make infallible: The allocation size
is limited by the thread's stack depth, and we only hit this code path when the
stack is the deepest it's ever been during the thread's life time.
This change reduces the binary size on Linux x64 by around 100KB.
Here's a diff of the impact on the code generated for Attr_Binding::get_specified
in the Mac build, it really just eliminates one test and one jump at the very end
of the method:
@@ -9,30 +9,29 @@
movq %rcx, %r14
movq %rdx, %r15
movq 0x80(%rdi), %rbx
movq %rbx, -40(%rbp)
testq %rbx, %rbx
je loc_xxxxx
movl 0x10(%rbx), %r12d
- cmpl %r12d, (%rbx)
- jbe loc_xxxxx
+ cmpl (%rbx), %r12d
+ jae loc_xxxxx
movq 0x8(%rbx), %rax
movq %r12, %rcx
shlq $0x5, %rcx
leaq aGetAttrspecifi, %rdx ; "get Attr.specified"
movq %rdx, (%rax,%rcx)
movq $0x0, 0x8(%rax,%rcx)
leaq -40(%rbp), %rdx
movq %rdx, 0x10(%rax,%rcx)
movl $0x106, 0x18(%rax,%rcx)
movl $0x1c, 0x1c(%rax,%rcx)
-
leal 0x1(%r12), %eax
movl %eax, 0x10(%rbx)
movq %r15, %rdi
call __ZNK7mozilla3dom4Attr9SpecifiedEv ; mozilla::dom::Attr::Specified() const
movzxl %al, %eax
movabsq $0xfff9000000000000, %rcx
orq %rax, %rcx
@@ -50,12 +49,9 @@
popq %r14
popq %r15
popq %rbp
ret
; endp
movq %rbx, %rdi
call __ZN14ProfilingStack18ensureCapacitySlowEv ; ProfilingStack::ensureCapacitySlow()
- testb %al, %al
- jne loc_xxxxx
-
jmp loc_xxxxx
Depends on D9192
Differential Revision: https://phabricator.services.mozilla.com/D9193
--HG--
extra : moz-landing-system : lando
This eliminates a few instructions from every profiler label and saves code size.
We have around 9000 WebIDL constructors + methods + getters + setters which all
have an inlined instance of this code.
This change reduces the binary size on Linux x64 by around 160KB.
Here's a diff of the impact on the code generated for Attr_Binding::get_specified
in the Mac build:
movq %rsp, %rbp
pushq %r15
pushq %r14
pushq %r12
pushq %rbx
subq $0x10, %rsp
movq %rcx, %r14
movq %rdx, %r15
- movq __ZN7mozilla8profiler6detail12RacyFeatures18sActiveAndFeaturesE@GOT, %rax ; __ZN7mozilla8profiler6detail12RacyFeatures18sActiveAndFeaturesE@GOT
- movl (%rax), %eax
- testl %eax, %eax
- js loc_xxxxx
-
- movq $0x0, -40(%rbp)
- jmp loc_xxxxx
-
- movq 0x78(%rdi), %rbx
+ movq 0x80(%rdi), %rbx
movq %rbx, -40(%rbp)
testq %rbx, %rbx
je loc_xxxxx
movl 0x10(%rbx), %r12d
cmpl %r12d, (%rbx)
jbe loc_xxxxx
Differential Revision: https://phabricator.services.mozilla.com/D9192
--HG--
extra : moz-landing-system : lando
- Add JOF_ARGC, JOF_RESUMEINDEX instead of bare integer tags
- Fix some inconsistencies of JOF_ type and opcode size
- Reorganize JOF_ type list
Differential Revision: https://phabricator.services.mozilla.com/D10567
--HG--
extra : moz-landing-system : lando
Gurzau Raul
Jan de Mooij
Benjamin Bouvier
Lars T Hansen
Jeff Walden
arthur.iakab
Iain Ireland
Csoregi Natalia
Tooru Fujisawa
Ashley Hauck
Till Schneidereit
Luke Wagner
Kartikaya Gupta
Nicolas B. Pierron
André Bargull
Ehsan Akhgari
Markus Stange
Eric Faust
shindli
Kyle Machulis
Narcis Beleuzu
vinoth
Matthew Gaudet
Ted Campbell
Jonathan Kingston