1. 26 May, 2020 2 commits
  2. 25 May, 2020 1 commit
    • Ross McIlroy's avatar
      [TurboProp] Don't try to rewire unreachable blocks to end. · f34771f7
      Ross McIlroy authored
      We can't consistently rewire the successor blocks of an unreachable node to
      disconnect them from the graph when we are trying to maintain the schedule.
      Instead simply leave the code there. As a future optimization we could add a
      proper scheduled dead code elimination phase which can deal with this.
      
      As a side-effect, one of the tests sees a int64 DeadValue, so add support for that
      in the instruction selector.
      
      BUG=chromium:1083272,chromium:1083763,chromium:1084953,v8:9684
      
      Change-Id: I69a6feaeef4eae62110392e27ea848b28bccf787
      Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2209061
      Commit-Queue: Tobias Tebbi <tebbi@chromium.org>
      Reviewed-by: 's avatarTobias Tebbi <tebbi@chromium.org>
      Auto-Submit: Ross McIlroy <rmcilroy@chromium.org>
      Cr-Commit-Position: refs/heads/master@{#67953}
      f34771f7
  3. 22 May, 2020 1 commit
    • Seth Brenith's avatar
      [torque] Generate better code when using `&` operator on bitfields · 98438d86
      Seth Brenith authored
      Sometimes CSA code carefully constructs a mask to check several
      bitfields at once. Thus far, such a check has been very awkward to write
      in Torque. This change adds a way to do so, using the
      non-short-circuiting binary `&` operator. So now you can write an
      expression that depends on several bitfields from a bitfield struct,
      like `x.a == 5 & x.b & !x.c & x.d == 2` (assuming b is a one-bit value),
      and it will be reduced to a single mask and equality check. To
      demonstrate a usage of this new reduction, this change ports the trivial
      macro IsSimpleObjectMap to Torque. I manually verified that the
      generated code for the builtin SetDataProperties, which uses that macro,
      is unchanged.
      
      Bug: v8:7793
      Change-Id: I4a23e0005d738a6699ea0f2a63f9fd67b01e7026
      Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2183276
      Commit-Queue: Seth Brenith <seth.brenith@microsoft.com>
      Reviewed-by: 's avatarTobias Tebbi <tebbi@chromium.org>
      Cr-Commit-Position: refs/heads/master@{#67948}
      98438d86
  4. 21 May, 2020 3 commits
    • Ng Zhi An's avatar
      [wasm-simd][arm] Improve f64x2 lt and le codegen · d48bbd52
      Ng Zhi An authored
      From 10 to 8 instructions (each). We do this by using mi (instead of lt)
      and ls (instead of le), which check for strictly less than and greater
      than or unordered. That way we don't have to have an extra mov for NaN.
      
      Change-Id: I18ff876ac12b7097d73d6cbbb64de6c2a1148e43
      Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2208934Reviewed-by: 's avatarDeepti Gandluri <gdeepti@chromium.org>
      Commit-Queue: Zhi An Ng <zhin@chromium.org>
      Cr-Commit-Position: refs/heads/master@{#67947}
      d48bbd52
    • Ng Zhi An's avatar
      [wasm-simd] Rename anytrue and alltrue to follow proposal · 4c2f84b5
      Ng Zhi An authored
      The proposal uses the lane shape, e.g. i64x2.anytrue, and we were using
      s1x2.anytrue in our opcodes. This was a legacy naming, because we were
      trying to bitpack the booleans. Now that we aren't doing that, rename
      these to be more consistent with the proposal.
      
      This was done with a straightforward sed script, changing both cpp code
      and also some comments in mjsunit test files.
      
      Bug: v8:10506
      Change-Id: If077ed805de23520d8580d6b3b1906c80f67b94f
      Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2207915
      Commit-Queue: Zhi An Ng <zhin@chromium.org>
      Reviewed-by: 's avatarDeepti Gandluri <gdeepti@chromium.org>
      Reviewed-by: 's avatarTobias Tebbi <tebbi@chromium.org>
      Cr-Commit-Position: refs/heads/master@{#67945}
      4c2f84b5
    • Seth Brenith's avatar
      [diagnostics] Support --turbo-profiling for builtins · 18c73676
      Seth Brenith authored
      Currently, if d8 is run with the --turbo-profiling flag, it prints info
      about every TurboFan-compiled function. This info includes the number of
      times that each basic block in the function was run. It also includes
      text representations of the function's schedule and code, so that the
      person reading the output can associate counters with blocks of code.
      
      The data about each function is currently stored in a
      BasicBlockProfiler::Data instance, which is attached to a list owned by
      the singleton BasicBlockProfiler. Each Data contains an
      std::vector<uint32_t> which represents how many times each block in the
      function has executed. The generated code for each block uses a raw
      pointer into the storage of that vector to implement incrementing the
      counter.
      
      With this change, if you compile with v8_enable_builtins_profiling and
      then run with --turbo-profiling, d8 will print that same info about
      builtins too.
      
      In order to generate code that can survive being serialized to a
      snapshot and reloaded, this change uses counters in the JS heap instead
      of a std::vector outside the JS heap. The steps for instrumentation are
      as follows:
      
      1. Between scheduling and instruction selection, add code to increment
         the counter for each block. The counters array doesn't yet exist at
         this point, and allocation is disallowed, so at this point the code
         refers to a special marker value.
      2. During finalization of the code, allocate a BasicBlockProfilingData
         object on the JS heap containing data equivalent to what is stored in
         BasicBlockProfiler::Data. This includes a ByteArray that is big
         enough to store the counters for each block.
      3. Patch the reference in the BuiltinsConstantsTableBuilder so that
         instead of referring to the marker object, it now refers to this
         ByteArray. Also add the BasicBlockProfilingData object to a list that
         is attached to the heap roots so it can be easily accessed for
         printing.
      
      Because these steps include modifying the BuiltinsConstantsTableBuilder,
      this procedure is only applicable to builtins. Runtime-generated code
      still uses raw pointers into std::vector instances. In order to keep
      divergence between these code paths to a minimum, most work is done
      referring to instances of BasicBlockProfiler::Data (the C++ class), and
      functions are provided to copy back and forth between that type and
      BasicBlockProfilingData (the JS heap object).
      
      This change is intended only to make --turbo-profiling work consistently
      on more kinds of functions, but with some further work, this data could
      form the basis for:
      - code coverage info for fuzzers, and/or
      - hot-path info for profile-guided optimization.
      
      Bug: v8:10470, v8:9119
      Change-Id: Ib556a5bc3abe67cdaa2e3ee62702a2a08b11cb61
      Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2159738
      Commit-Queue: Seth Brenith <seth.brenith@microsoft.com>
      Reviewed-by: 's avatarUlan Degenbaev <ulan@chromium.org>
      Reviewed-by: 's avatarJakob Gruber <jgruber@chromium.org>
      Cr-Commit-Position: refs/heads/master@{#67944}
      18c73676
  5. 19 May, 2020 2 commits
  6. 18 May, 2020 1 commit
    • Ng Zhi An's avatar
      [wasm-simd][arm] Improve codegen for any_true · 9d1dda7e
      Ng Zhi An authored
      The codegen uses a bunch of vpmax to try and keep set bits around. The
      datatype for vpmax does not need to change for each instruction, since
      vpmax U32 will persist set bits just as well. This simplifies the
      instruction sequences for S1x8 and S1x16 anytrue.
      
      I added a test to check a special case when a f64x2 contains -0.0 (top
      bit set). A previous attempt to optimize codegen used floating point
      compare, which does not distinguish between 0.0 and -0.0. So -0.0 will
      compare equals to 0.0, and incorrect return 0 for anytrue.
      
      Change-Id: I66013796af08a666009e6b2d774ea7ee7bdfe1ad
      Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2203113
      Commit-Queue: Zhi An Ng <zhin@chromium.org>
      Reviewed-by: 's avatarBill Budge <bbudge@chromium.org>
      Cr-Commit-Position: refs/heads/master@{#67875}
      9d1dda7e
  7. 15 May, 2020 1 commit
  8. 14 May, 2020 1 commit
  9. 13 May, 2020 4 commits
  10. 11 May, 2020 3 commits
  11. 08 May, 2020 2 commits
  12. 07 May, 2020 2 commits
  13. 06 May, 2020 4 commits
  14. 05 May, 2020 3 commits
  15. 01 May, 2020 2 commits
  16. 29 Apr, 2020 1 commit
    • Tobias Tebbi's avatar
      Reland "Reland "[turbofan][csa] optimize Smi untagging better"" · 9e9cd5df
      Tobias Tebbi authored
      This is a reland of 43b885a8
      This fixes another signed overflow in the unit test.
      
      Original change's description:
      > Reland "[turbofan][csa] optimize Smi untagging better"
      >
      > This is a reland of ff22ae80
      >
      > Original change's description:
      > > [turbofan][csa] optimize Smi untagging better
      > >
      > > - Introduce new operator variants for signed right-shifts with the
      > >   additional information that they always shift out zeros.
      > > - Use these new operators for Smi untagging.
      > > - Merge left-shifts with a preceding Smi-untagging shift.
      > > - Optimize comparisons of Smi-untagging shifts to operate on the
      > >   unshifted word.
      > > - Optimize 64bit comparisons of values expanded from 32bit to use
      > >   a 32bit comparison instead.
      > > - Change CodeStubAssembler::UntagSmi to first sign-extend and then
      > >   right-shift to enable better address computations for Smi indices.
      > >
      > > Bug: v8:9962
      > > Change-Id: If91300f365e8f01457aebf0bd43bdf88b305c460
      > > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2135734
      > > Commit-Queue: Tobias Tebbi <tebbi@chromium.org>
      > > Reviewed-by: Georg Neis <neis@chromium.org>
      > > Cr-Commit-Position: refs/heads/master@{#67378}
      >
      > Bug: v8:9962
      > Change-Id: Ieab0755806c95fb50022eb17596fb0c95f36004c
      > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2170001
      > Commit-Queue: Tobias Tebbi <tebbi@chromium.org>
      > Commit-Queue: Georg Neis <neis@chromium.org>
      > Auto-Submit: Tobias Tebbi <tebbi@chromium.org>
      > Reviewed-by: Georg Neis <neis@chromium.org>
      > Cr-Commit-Position: refs/heads/master@{#67430}
      
      Bug: v8:9962
      TBR: neis@chromium.org
      Change-Id: I79883db546bf37873b3727b8023ef688507091d9
      Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2169103
      Commit-Queue: Tobias Tebbi <tebbi@chromium.org>
      Reviewed-by: 's avatarGeorg Neis <neis@chromium.org>
      Cr-Commit-Position: refs/heads/master@{#67464}
      9e9cd5df
  17. 28 Apr, 2020 3 commits
    • Clemens Backes's avatar
      Revert "Reland "[turbofan][csa] optimize Smi untagging better"" · bef5b85d
      Clemens Backes authored
      This reverts commit 43b885a8.
      
      Reason for revert: Still fails on UBSan: https://ci.chromium.org/p/v8/builders/ci/V8%20Linux64%20UBSan/10873
      
      Original change's description:
      > Reland "[turbofan][csa] optimize Smi untagging better"
      > 
      > This is a reland of ff22ae80
      > 
      > Original change's description:
      > > [turbofan][csa] optimize Smi untagging better
      > > 
      > > - Introduce new operator variants for signed right-shifts with the
      > >   additional information that they always shift out zeros.
      > > - Use these new operators for Smi untagging.
      > > - Merge left-shifts with a preceding Smi-untagging shift.
      > > - Optimize comparisons of Smi-untagging shifts to operate on the
      > >   unshifted word.
      > > - Optimize 64bit comparisons of values expanded from 32bit to use
      > >   a 32bit comparison instead.
      > > - Change CodeStubAssembler::UntagSmi to first sign-extend and then
      > >   right-shift to enable better address computations for Smi indices.
      > > 
      > > Bug: v8:9962
      > > Change-Id: If91300f365e8f01457aebf0bd43bdf88b305c460
      > > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2135734
      > > Commit-Queue: Tobias Tebbi <tebbi@chromium.org>
      > > Reviewed-by: Georg Neis <neis@chromium.org>
      > > Cr-Commit-Position: refs/heads/master@{#67378}
      > 
      > Bug: v8:9962
      > Change-Id: Ieab0755806c95fb50022eb17596fb0c95f36004c
      > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2170001
      > Commit-Queue: Tobias Tebbi <tebbi@chromium.org>
      > Commit-Queue: Georg Neis <neis@chromium.org>
      > Auto-Submit: Tobias Tebbi <tebbi@chromium.org>
      > Reviewed-by: Georg Neis <neis@chromium.org>
      > Cr-Commit-Position: refs/heads/master@{#67430}
      
      TBR=neis@chromium.org,tebbi@chromium.org
      
      Change-Id: I49e19811ebcecb846f61291bc0c4a0d8b0bc4cff
      No-Presubmit: true
      No-Tree-Checks: true
      No-Try: true
      Bug: v8:9962
      Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2168876Reviewed-by: 's avatarClemens Backes <clemensb@chromium.org>
      Commit-Queue: Clemens Backes <clemensb@chromium.org>
      Cr-Commit-Position: refs/heads/master@{#67431}
      bef5b85d
    • Tobias Tebbi's avatar
      Reland "[turbofan][csa] optimize Smi untagging better" · 43b885a8
      Tobias Tebbi authored
      This is a reland of ff22ae80
      
      Original change's description:
      > [turbofan][csa] optimize Smi untagging better
      > 
      > - Introduce new operator variants for signed right-shifts with the
      >   additional information that they always shift out zeros.
      > - Use these new operators for Smi untagging.
      > - Merge left-shifts with a preceding Smi-untagging shift.
      > - Optimize comparisons of Smi-untagging shifts to operate on the
      >   unshifted word.
      > - Optimize 64bit comparisons of values expanded from 32bit to use
      >   a 32bit comparison instead.
      > - Change CodeStubAssembler::UntagSmi to first sign-extend and then
      >   right-shift to enable better address computations for Smi indices.
      > 
      > Bug: v8:9962
      > Change-Id: If91300f365e8f01457aebf0bd43bdf88b305c460
      > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2135734
      > Commit-Queue: Tobias Tebbi <tebbi@chromium.org>
      > Reviewed-by: Georg Neis <neis@chromium.org>
      > Cr-Commit-Position: refs/heads/master@{#67378}
      
      Bug: v8:9962
      Change-Id: Ieab0755806c95fb50022eb17596fb0c95f36004c
      Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2170001
      Commit-Queue: Tobias Tebbi <tebbi@chromium.org>
      Commit-Queue: Georg Neis <neis@chromium.org>
      Auto-Submit: Tobias Tebbi <tebbi@chromium.org>
      Reviewed-by: 's avatarGeorg Neis <neis@chromium.org>
      Cr-Commit-Position: refs/heads/master@{#67430}
      43b885a8
    • Zhao Jiazhong's avatar
      [mips64][wasm-simd] Fix bugs of some wasm-simd operations. · f74d2a90
      Zhao Jiazhong authored
      Due to lack of testing environment before, there are some bugs in the
      implementations of wasm-simd on mips64 platform, this CL fix them
      according to the test on Loongson 3A4000.
      
      Change-Id: I59ab6315987fc94a06cf0bf23754f5c593879532
      Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2162416
      Commit-Queue: Zhao Jiazhong <zhaojiazhong-hf@loongson.cn>
      Reviewed-by: 's avatarZhi An Ng <zhin@chromium.org>
      Reviewed-by: 's avatarDeepti Gandluri <gdeepti@chromium.org>
      Cr-Commit-Position: refs/heads/master@{#67413}
      f74d2a90
  18. 27 Apr, 2020 1 commit
  19. 24 Apr, 2020 3 commits