diff --git a/cbits/debug.h b/cbits/debug.h --- a/cbits/debug.h +++ b/cbits/debug.h @@ -86,12 +86,12 @@ inline void hw_simd_json_print_bits_256(__m256i v) { int i = 0; - for (i = 0; i < 4; ++i) { - if (i > 0) { - printf("-"); - } - - hw_simd_json_print_bits_64(_mm256_extract_epi64(v, i)); - } + hw_simd_json_print_bits_64(_mm256_extract_epi64(v, 0)); + printf("-"); + hw_simd_json_print_bits_64(_mm256_extract_epi64(v, 1)); + printf("-"); + hw_simd_json_print_bits_64(_mm256_extract_epi64(v, 2)); + printf("-"); + hw_simd_json_print_bits_64(_mm256_extract_epi64(v, 3)); } #endif//__AVX2__ diff --git a/cbits/simd-spliced.c b/cbits/simd-spliced.c --- a/cbits/simd-spliced.c +++ b/cbits/simd-spliced.c @@ -50,7 +50,10 @@ exit(1); } - uint8_t buffer[W8_BUFFER_SIZE]; + // align stack buffer on 32 bytes + uint8_t unaligned_buffer[W8_BUFFER_SIZE + 32]; + uint8_t * buffer = unaligned_buffer + ((uintptr_t)unaligned_buffer & (uintptr_t)0x10); + fprintf(stderr, "Buffer is at address %p of size %zu\n", buffer, (size_t) W8_BUFFER_SIZE); uint8_t *bits_of_d = malloc(W32_BUFFER_SIZE); memset(bits_of_d, 0, W32_BUFFER_SIZE); uint8_t *bits_of_a = malloc(W32_BUFFER_SIZE); memset(bits_of_a, 0, W32_BUFFER_SIZE); diff --git a/hw-json-simd.cabal b/hw-json-simd.cabal --- a/hw-json-simd.cabal +++ b/hw-json-simd.cabal @@ -1,7 +1,7 @@ cabal-version: 2.2 name: hw-json-simd -version: 0.1.0.2 +version: 0.1.0.3 synopsis: SIMD-based JSON semi-indexer description: Please see the README on GitHub at category: Data @@ -9,11 +9,11 @@ bug-reports: https://github.com/haskell-works/hw-json-simd/issues author: John Ky maintainer: newhoggy@gmail.com -copyright: 2018 - 2019 John Ky +copyright: 2018-2019 John Ky license: BSD-3-Clause license-file: LICENSE build-type: Simple -tested-with: GHC == 8.6.3, GHC == 8.4.4, GHC == 8.2.2, GHC == 8.0.2 +tested-with: GHC == 8.6.3, GHC == 8.4.4, GHC == 8.2.2 extra-source-files: cbits/debug.h cbits/simd.h @@ -47,9 +47,9 @@ common base { build-depends: base >= 4 && < 5 } common bytestring { build-depends: bytestring >= 0.10.6 && < 0.11 } -common hw-prim { build-depends: hw-prim >= 0.6.2.21 && < 0.7 } +common hw-prim { build-depends: hw-prim >= 0.6.2.35 && < 0.7 } common lens { build-depends: lens >= 4 && < 5 } -common optparse-applicative { build-depends: optparse-applicative >= 0.14 && < 0.15 } +common optparse-applicative { build-depends: optparse-applicative >= 0.14 && < 0.16 } common vector { build-depends: vector >= 0.12 && < 0.13 } common semigroups { if (!impl(ghc >= 8.0.1)) { build-depends: { semigroups >= 0.8.4 && < 0.19 } } } @@ -76,12 +76,12 @@ , hw-prim , lens , vector - autogen-modules: Paths_hw_json_simd - other-modules: Paths_hw_json_simd - hs-source-dirs: src - cc-options: -mssse3 -mlzcnt -mbmi2 -mavx2 - include-dirs: cbits - build-tools: c2hs + autogen-modules: Paths_hw_json_simd + other-modules: Paths_hw_json_simd + hs-source-dirs: src + cc-options: -mssse3 -mlzcnt -mbmi2 -mavx2 + include-dirs: cbits + build-tool-depends: c2hs:c2hs exposed-modules: HaskellWorks.Data.Json.Simd.Capabilities HaskellWorks.Data.Json.Simd.Index.Simple @@ -106,11 +106,11 @@ , optparse-applicative , semigroups , vector - main-is: Main.hs - autogen-modules: Paths_hw_json_simd - hs-source-dirs: app - build-depends: hw-json-simd - ghc-options: -threaded -rtsopts -with-rtsopts=-N + main-is: Main.hs + autogen-modules: Paths_hw_json_simd + hs-source-dirs: app + build-depends: hw-json-simd + ghc-options: -threaded -rtsopts -with-rtsopts=-N other-modules: App.Commands App.Commands.Capabilities @@ -127,10 +127,10 @@ , semigroups , transformers , vector - type: exitcode-stdio-1.0 - main-is: Spec.hs - autogen-modules: Paths_hw_json_simd - other-modules: Paths_hw_json_simd - hs-source-dirs: test - ghc-options: -threaded -rtsopts -with-rtsopts=-N - build-depends: hw-json-simd + type: exitcode-stdio-1.0 + main-is: Spec.hs + autogen-modules: Paths_hw_json_simd + other-modules: Paths_hw_json_simd + hs-source-dirs: test + ghc-options: -threaded -rtsopts -with-rtsopts=-N + build-depends: hw-json-simd diff --git a/src/HaskellWorks/Data/Json/Simd/Internal/Index/Standard.hs b/src/HaskellWorks/Data/Json/Simd/Internal/Index/Standard.hs --- a/src/HaskellWorks/Data/Json/Simd/Internal/Index/Standard.hs +++ b/src/HaskellWorks/Data/Json/Simd/Internal/Index/Standard.hs @@ -9,6 +9,7 @@ import qualified Foreign as F import qualified Foreign.ForeignPtr.Unsafe as F import qualified HaskellWorks.Data.Json.Simd.Internal.Foreign as F +import qualified HaskellWorks.Foreign as F {-# ANN module ("HLint: ignore Reduce duplication" :: String) #-} {-# ANN module ("HLint: ignore Redundant do" :: String) #-} @@ -29,8 +30,7 @@ allocWorkBuffers :: Int -> IO WorkBuffers allocWorkBuffers n = do - fptr <- F.mallocForeignPtrBytes (3 * n) - let ptr = F.unsafeForeignPtrToPtr fptr + (fptr, ptr) <- F.mallocForeignPtrBytesWithAlignedCastPtr 32 (3 * n) return WorkBuffers { workBuffersF = fptr , workBuffersP = ptr `F.plusPtr` 0