1. 06 Jun, 2022 1 commit
  2. 22 Mar, 2021 1 commit
  3. 16 Mar, 2021 1 commit
  4. 15 Dec, 2020 1 commit
    • Zhi An Ng's avatar
      [x64][wasm-simd] Pattern match 32x4 rotate · 7c98abdb
      Zhi An Ng authored
      Code like:
      
        x = wasm_v32x4_shuffle(x, x, 1, 2, 3, 0);
      
      is currently matched by S8x16Concat, which lowers to two instructions:
      
        movapd xmm_dst, xmm_src
        palignr xmm_dst, xmm_src, 0x4
      
      There is a special case after a S8x16Concat is matched:.
      
      - is_swizzle, the inputs are the same
      - it is a 32x4 shuffle (offset % 4 == 0)
      
      Which can have a better codegen:
      
      - (dst == src) shufps dst, src, 0b00111001
      - (dst != src) pshufd dst, src, 0b00111001
      
      Add a new simd shuffle matcher which will match 32x4 rotate, and
      construct the appropriate indices referring to the 32x4 elements.
      
      pshufd for the given example. However, this matching happens after
      S8x16Concat, so we get the palignr first. We could move the pattern
      matching cases around, but it will lead to some cases where
      where it would have matched a S8x16Concat, but now matches a
      S32x4shuffle instead, leading to worse codegen.
      
      Note: we also pattern match on 32x4Swizzle, which correctly generates
      Change-Id: Ie3aca53bbc06826be2cf49632de4c24ec73d0a9a
      Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2589062Reviewed-by: 's avatarBill Budge <bbudge@chromium.org>
      Commit-Queue: Zhi An Ng <zhin@chromium.org>
      Cr-Commit-Position: refs/heads/master@{#71754}
      7c98abdb
  5. 04 Aug, 2020 1 commit
  6. 31 Jul, 2020 2 commits
  7. 30 Jul, 2020 2 commits
  8. 24 Jul, 2020 1 commit
    • Ng Zhi An's avatar
      [wasm-simd] Move shuffle pattern matching out of instruction-selector · 6bd75498
      Ng Zhi An authored
      These functions match on specific patterns of shuffle that have more
      optimized implementations. Moving them out of instruction-selector
      allows us to reuse them in Liftoff. Most of these pattern matching
      functions do not depend on InstructionSelector, since they work on byte
      arrays. (The only one is CanonicalizeShuffle, which swaps node inputs.)
      
      This is only the first pass of moving those functions out. In particular
      we can clean things up more by moving the tests out of
      instruction-selector as well. Those will come in follow-up changes.
      
      Bug: v8:10696
      Change-Id: I4a4333cd8c0259875a672179e72d34dad5f7a008
      Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2308057
      Commit-Queue: Zhi An Ng <zhin@chromium.org>
      Reviewed-by: 's avatarBill Budge <bbudge@chromium.org>
      Cr-Commit-Position: refs/heads/master@{#69051}
      6bd75498