Skip to content

V
V8
Switch branch/tag
  1. 19 Jul, 2021 1 commit
    • Jakob Gruber's avatar
      [compiler] Make JSFunction bg-serialized · 0dba97f8
      Jakob Gruber authored 
      This wraps up the transition away from kSerialized ref kinds.

Since JSFunctionRef is a complex type, we don't attempt full
consistency on the background thread. Instead, we serialize functions
on the background in a partially-racy manner, in which consistency
between different JSFunction fields is *not* guaranteed. Consistency
is later verified through a new compilation dependency kind during
finalization.

Bug: v8:7790, v8:12004
Change-Id: Ic2b78af9c9fe183c8769d323132bb304b151dc75
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2968404
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Reviewed-by: 's avatarIgor Sheludko <ishell@chromium.org>
Cr-Commit-Position: refs/heads/master@{#75789}
      0dba97f8
  3. 05 Jul, 2021 1 commit
    • Georg Neis's avatar
      [compiler] Fix a (harmless) DCHECK failure · ce08dec0
      Georg Neis authored 
      A frame state's outer frame state input can be kDeadValue. A DCHECK
did not take that into account. In release builds there was no issue
because we branch on the opcode anyways.

While fixing this bug, I'm strengthening the FrameState class such that
a FrameState node must have a kFrameState operator. I'm also
- changing the result type of outer_frame_state() from FrameState to
  Node* since it may in fact not be a kFrameState;
- removing has_outer_frame_state() because I find it unintuitive to
  have outer_frame_state() return non-NULL even when
  has_outer_frame_state() would return true.

Bug: chromium:1224758
Change-Id: I8ebed75c62e31f7eef71e2941fd18869d8a56af3
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3001356Reviewed-by: 's avatarNico Hartmann <nicohartmann@chromium.org>
Commit-Queue: Georg Neis <neis@chromium.org>
Cr-Commit-Position: refs/heads/master@{#75553}
      ce08dec0
  5. 30 Jun, 2021 1 commit
  7. 28 Jun, 2021 1 commit
  9. 25 Jun, 2021 1 commit
  11. 24 Jun, 2021 3 commits
  13. 17 Jun, 2021 1 commit
  15. 11 Jun, 2021 1 commit
  17. 07 Jun, 2021 1 commit
  19. 04 Jun, 2021 1 commit
  21. 28 May, 2021 1 commit
  23. 12 May, 2021 2 commits
  25. 11 May, 2021 1 commit
  27. 06 May, 2021 1 commit
    • Fanchen Kong's avatar
      Collect receiver to feedback for prototype.apply · 519c82ce
      Fanchen Kong authored 
      When a function is invoked by prototype.apply, it may undergo following transformation in the JSCallReducer:
	receiver.apply(this, args) ->
	this.receiver(...args) Since the new target (also the receiver of apply()) is not collected to the feedback slot, further speculative optimization on the new target is not available if the new target
is not a heapconstant.

With this CL, the receiver will be collected to the feedback instead of the target if the target is a prototype.apply. It may improve the performance of the following usecase by ~80%.

function reduceArray(func, arr, r) {
    for (var i = 0, len = arr.length; i < len; i++) {
            r = func.apply(null, r, arr[i]);
    }
    return r;
}

var a = 0; for (var i = 0; i < 10000000; i++) {
    a += reduceArray(Math.imul, [5,6,2,3,7,6,8,3,7,9,2,5,], 1);
}
console.log(a);

This CL also improves the runTime score of JetStream2/richards-wasm by ~45% in default, ~60% with --turbo-inline-js-wasm-calls.

Change-Id: I542eb8d3fcb592f4e0993af93ba1af70e89c3982
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2639813
Commit-Queue: Fanchen Kong <fanchen.kong@intel.com>
Reviewed-by: 's avatarRoss McIlroy <rmcilroy@chromium.org>
Reviewed-by: 's avatarGeorg Neis <neis@chromium.org>
Cr-Commit-Position: refs/heads/master@{#74413}
      519c82ce
  29. 30 Apr, 2021 1 commit
  31. 29 Apr, 2021 1 commit
  33. 26 Apr, 2021 3 commits
  35. 21 Apr, 2021 2 commits
  37. 17 Apr, 2021 1 commit
    • Ng Zhi An's avatar
      Revert "[wasm-simd] Canonicalize shuffles when creating TurboFan graph" · ede5e814
      Ng Zhi An authored 
      This reverts commit d16eefe0.

It is not correct to check for node equality during the graph
construction phase, because we can have optimizations that will combine
same nodes. So it can happen that in wasm-compiler, the inputs to
shuffle are not the same, so we canonicalize using that knowledge that
it will not be the same, and allow indices > 15. But later we can have
optimizations that combine the 2 inputs (e.g. splat of the same
constants), and the instruction selector will see that the input nodes
are the same.

Bug: v8:11542,chromium:1199662
Change-Id: I21c175f4707708038710147f64d687d1b14c6ecc
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2829986
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Reviewed-by: 's avatarBill Budge <bbudge@chromium.org>
Cr-Commit-Position: refs/heads/master@{#74017}
      ede5e814
  39. 16 Apr, 2021 3 commits
    • Leszek Swirski's avatar
      Reland "[codegen] Add static interface descriptors" · 2871e05c
      Leszek Swirski authored 
      This is a reland of ae0752df

Reland fixes:

  * Remove UNREACHABLE() from constexpr switch, since we don't have a
    CONSTEXPR_UNREACHABLE() (it's ok, the switch is exhaustive for the
    enum anyway).
  * Fix IsRegisterArray trait to use public inheritance and size_t for
    std::array size.

Original change's description:
> [codegen] Add static interface descriptors
>
> Add a new CRTP StaticCallInterfaceDescriptor class, which provides
> static constexpr getters for a descriptor's registers, parameter counts,
> and so on. Each CallInterfaceDescriptor subclass is changed to extend
> StaticCallInterfaceDescriptor, with StaticCallInterfaceDescriptor itself
> extending CallInterfaceDescriptor to still provide a dynamic lookup
> where needed.
>
> StaticCallInterfaceDescriptor provides a couple of customisation points,
> where it reads its CRTP derived descriptor's static fields and
> functions, with default fallbacks where appropriate. With these
> customisation points, the definition of CallInterfaceDescriptor
> subclasses is simplified to:
>
>     a) Providing parameter names (as before)
>     b) Providing parameter types (as before)
>     c) Optionally setting flags (like kNoContext or kAllowVarArgs) as
>        static booleans on the class.
>     d) Optionally providing a `registers()` method that returns a
>        std::array<Register, N> of registers that may be used for
>        parameters (if not provided, this defaults to the implementation
>        specific default register set).
>
> Parameter registers (and register count) are automagically set based on
> the number of parameters and number of given registers, with extra magic
> to ignore no_reg registers (to reduce ia32 special casing). The
> CallInterfaceDescriptorData is initialized based on these static
> functions, rather than manual per-descriptor initializers.
>
> This allows us to skip loading descriptors dynamically for CallBuiltin
> in Sparkplug, and instead lets us use a bit of template magic to
> statically set up arguments for the calls. Any other users of statically
> known descriptors will also benefit, thanks to C++ picking the static
> methods over the dynamic methods on the base class when available.
>
> Because we can remove various virtual functions and trigger heavier
> inlining of constantly known values, binary size slightly decreases with
> this change.
>
> Note that torque-generated descriptors are changed to use the same magic,
> rather than having Torque-specific magic, for consistency.
>
> Bug: v8:11420
> Change-Id: Icc5e238b6313a08734feb564204a13226b450c22
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2814518
> Auto-Submit: Leszek Swirski <leszeks@chromium.org>
> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org>
> Reviewed-by: Clemens Backes <clemensb@chromium.org>
> Reviewed-by: Igor Sheludko <ishell@chromium.org>
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Commit-Queue: Clemens Backes <clemensb@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#73996}

TBR=nicohartmann@chromium.org,clemensb@chromium.org,ishell@chromium.org,clemensb@chromium.org

Bug: v8:11420
Change-Id: Icd1f6cdb3c178e74460044b1e9623139929ceba8
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2831872Reviewed-by: 's avatarLeszek Swirski <leszeks@chromium.org>
Reviewed-by: 's avatarClemens Backes <clemensb@chromium.org>
Reviewed-by: 's avatarIgor Sheludko <ishell@chromium.org>
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Cr-Commit-Position: refs/heads/master@{#74010}
      2871e05c
    • Leszek Swirski's avatar
      Revert "[codegen] Add static interface descriptors" · 5dea60d6
      Leszek Swirski authored 
      This reverts commit ae0752df.

Reason for revert: Predictably, constexpr issues on non-clang compilers.

Original change's description:
> [codegen] Add static interface descriptors
>
> Add a new CRTP StaticCallInterfaceDescriptor class, which provides
> static constexpr getters for a descriptor's registers, parameter counts,
> and so on. Each CallInterfaceDescriptor subclass is changed to extend
> StaticCallInterfaceDescriptor, with StaticCallInterfaceDescriptor itself
> extending CallInterfaceDescriptor to still provide a dynamic lookup
> where needed.
>
> StaticCallInterfaceDescriptor provides a couple of customisation points,
> where it reads its CRTP derived descriptor's static fields and
> functions, with default fallbacks where appropriate. With these
> customisation points, the definition of CallInterfaceDescriptor
> subclasses is simplified to:
>
>     a) Providing parameter names (as before)
>     b) Providing parameter types (as before)
>     c) Optionally setting flags (like kNoContext or kAllowVarArgs) as
>        static booleans on the class.
>     d) Optionally providing a `registers()` method that returns a
>        std::array<Register, N> of registers that may be used for
>        parameters (if not provided, this defaults to the implementation
>        specific default register set).
>
> Parameter registers (and register count) are automagically set based on
> the number of parameters and number of given registers, with extra magic
> to ignore no_reg registers (to reduce ia32 special casing). The
> CallInterfaceDescriptorData is initialized based on these static
> functions, rather than manual per-descriptor initializers.
>
> This allows us to skip loading descriptors dynamically for CallBuiltin
> in Sparkplug, and instead lets us use a bit of template magic to
> statically set up arguments for the calls. Any other users of statically
> known descriptors will also benefit, thanks to C++ picking the static
> methods over the dynamic methods on the base class when available.
>
> Because we can remove various virtual functions and trigger heavier
> inlining of constantly known values, binary size slightly decreases with
> this change.
>
> Note that torque-generated descriptors are changed to use the same magic,
> rather than having Torque-specific magic, for consistency.
>
> Bug: v8:11420
> Change-Id: Icc5e238b6313a08734feb564204a13226b450c22
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2814518
> Auto-Submit: Leszek Swirski <leszeks@chromium.org>
> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org>
> Reviewed-by: Clemens Backes <clemensb@chromium.org>
> Reviewed-by: Igor Sheludko <ishell@chromium.org>
> Reviewed-by: Jakob Gruber <jgruber@chromium.org>
> Commit-Queue: Clemens Backes <clemensb@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#73996}

Bug: v8:11420
Change-Id: Ie5469c9253fc140590ac30b72db6eb1d93f86806
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2831485
Auto-Submit: Leszek Swirski <leszeks@chromium.org>
Commit-Queue: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Cr-Commit-Position: refs/heads/master@{#74000}
      5dea60d6
    • Leszek Swirski's avatar
      [codegen] Add static interface descriptors · ae0752df
      Leszek Swirski authored 
      Add a new CRTP StaticCallInterfaceDescriptor class, which provides
static constexpr getters for a descriptor's registers, parameter counts,
and so on. Each CallInterfaceDescriptor subclass is changed to extend
StaticCallInterfaceDescriptor, with StaticCallInterfaceDescriptor itself
extending CallInterfaceDescriptor to still provide a dynamic lookup
where needed.

StaticCallInterfaceDescriptor provides a couple of customisation points,
where it reads its CRTP derived descriptor's static fields and
functions, with default fallbacks where appropriate. With these
customisation points, the definition of CallInterfaceDescriptor
subclasses is simplified to:

    a) Providing parameter names (as before)
    b) Providing parameter types (as before)
    c) Optionally setting flags (like kNoContext or kAllowVarArgs) as
       static booleans on the class.
    d) Optionally providing a `registers()` method that returns a
       std::array<Register, N> of registers that may be used for
       parameters (if not provided, this defaults to the implementation
       specific default register set).

Parameter registers (and register count) are automagically set based on
the number of parameters and number of given registers, with extra magic
to ignore no_reg registers (to reduce ia32 special casing). The
CallInterfaceDescriptorData is initialized based on these static
functions, rather than manual per-descriptor initializers.

This allows us to skip loading descriptors dynamically for CallBuiltin
in Sparkplug, and instead lets us use a bit of template magic to
statically set up arguments for the calls. Any other users of statically
known descriptors will also benefit, thanks to C++ picking the static
methods over the dynamic methods on the base class when available.

Because we can remove various virtual functions and trigger heavier
inlining of constantly known values, binary size slightly decreases with
this change.

Note that torque-generated descriptors are changed to use the same magic,
rather than having Torque-specific magic, for consistency.

Bug: v8:11420
Change-Id: Icc5e238b6313a08734feb564204a13226b450c22
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2814518
Auto-Submit: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: 's avatarNico Hartmann <nicohartmann@chromium.org>
Reviewed-by: 's avatarClemens Backes <clemensb@chromium.org>
Reviewed-by: 's avatarIgor Sheludko <ishell@chromium.org>
Reviewed-by: 's avatarJakob Gruber <jgruber@chromium.org>
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/master@{#73996}
      ae0752df
  41. 14 Apr, 2021 1 commit
  43. 13 Apr, 2021 1 commit
  45. 01 Apr, 2021 1 commit
  47. 29 Mar, 2021 1 commit
  49. 25 Mar, 2021 2 commits
  51. 24 Mar, 2021 1 commit
  53. 23 Mar, 2021 1 commit
    • Manos Koukoutos's avatar
      [turbofan] Introduce LoadImmutable, use it in wasm compiler · f6ee9ed0
      Manos Koukoutos authored 
      LoadImmutable represents a load from a position in memory that is known
to be immutable, e.g. an immutable IsolateRoot or an immutable field of
a WasmInstanceObject. Because the returned value cannot change through
the execution of a function, LoadImmutable is a pure operator and does
not have effect or control edges.
This will allow more aggressive optimizations of loads of fields of
the Isolate and Instance that are known to be immutable.
Requires that the memory in question has been initialized at function
start even through inlining.

Note: We may reconsider this approach once we have escape analysis for
wasm, and replace it with immutable load/initialize operators that live
inside the effect chain and are less restriced.

Bug: v8:11510
Change-Id: I5e8e4f27d7008f39f01175ffa95a9c531ba63e66
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2775568Reviewed-by: 's avatarAndreas Haas <ahaas@chromium.org>
Reviewed-by: 's avatarGeorg Neis <neis@chromium.org>
Commit-Queue: Manos Koukoutos <manoskouk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#73594}
      f6ee9ed0
  55. 22 Mar, 2021 1 commit
  57. 18 Mar, 2021 1 commit
  59. 17 Mar, 2021 1 commit
    • Ng Zhi An's avatar
      [wasm-simd] Canonicalize shuffles when creating TurboFan graph · d16eefe0
      Ng Zhi An authored 
      We currently canonicalize shuffles in the architecture specific
instruction selector. This has the drawback that if we want to pattern
match on nodes that have a shuffle as input, they need to individually
canonicalize the shuffle. There can also be a subtle bug if we
canonicalize the same shuffle node twice (see bug for details).

This moves the canonicalization to "construction time", in
wasm-compiler, when building the graph. As such, any pattern matches in
instruction-selector will only need to deal with canonicalized shuffles.

We introduce a new kind of parameter for shuffle nodes,
ShuffleParameter, to store the 16 bytes plus a bool indicating if this
is a swizzle. A swizzle essentially: inputs to the shuffle are the same
or all indices only touch 1 input. We calculate this when
canonicalizing, so store this bit of information inside of the node's
parameter.

We update the tests in x64 to handle special cases where, even though
the node's inputs are not swapped (due to canonicalization), they need
to be swapped for the specific instruction selected (e.g. palignr). The
test data also contains canonicalized shuffles, so we have to manually
canonicalize them.

Bug: v8:11542
Change-Id: I4e78082267bd03d6caedf43d68d81ef3f5f364a8
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2762420Reviewed-by: 's avatarBill Budge <bbudge@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#73495}
      d16eefe0
  61. 15 Mar, 2021 1 commit