highway support attempt

Author: Gillgamesh

CMakeLists.txt

@@ -8,14 +8,15 @@ set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_CXX_EXTENSIONS ON)
 set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/lib)
 set(CMAKE_INTERPROCEDURAL_OPTIMIZATION TRUE)
+set(CMAKE_COLOR_DIAGNOSTICS ON)
 
 
 
 # Make the default build type Release. If user or another
 # project sets a different value than use that
 if(NOT CMAKE_BUILD_TYPE)
-  message(STATUS "Setting build type to default -- Release")
-  set(CMAKE_BUILD_TYPE "Release" CACHE
+  message(STATUS "Setting build type to default -- Debug")
+  set(CMAKE_BUILD_TYPE "Debug" CACHE
       STRING "Choose the type of build." FORCE)
 endif()
 message(STATUS "GraphZeppelin Build Type: ${CMAKE_BUILD_TYPE}")
@@ -51,6 +52,7 @@ else()
   message (STATUS "GraphZeppelin building executables")
 endif()
 
+
 # Get GutterTree Project
 FetchContent_Declare(
   GutterTree
@@ -67,6 +69,22 @@ FetchContent_Declare(
   GIT_TAG         main
 )
 
+FetchContent_Declare(
+  highway
+
+  GIT_REPOSITORY https://github.com/google/highway.git
+  GIT_TAG        1.2.0
+)
+
+set(HWY_ENABLE_EXAMPLES OFF CACHE INTERNAL "Disable highway examples")
+set(HWY_ENABLE_TESTS OFF CACHE INTERNAL "Disable highway tests")
+
+# Get google highway
+FetchContent_MakeAvailable(highway)
+
+# Ensure highway target is explicitly added
+add_library(highway INTERFACE IMPORTED)
+
 if (BUILD_BENCH)
   # Get Google Benchmark
   FetchContent_Declare(
@@ -81,7 +99,7 @@ if (BUILD_BENCH)
 endif()
 
 
-FetchContent_MakeAvailable(GutterTree StreamingUtilities )
+FetchContent_MakeAvailable(GutterTree StreamingUtilities)
 
 # AVAILABLE COMPILATION DEFINITIONS:
 # VERIFY_SAMPLES_F   Use a deterministic connected-components
@@ -103,8 +121,8 @@ add_library(GraphZeppelin
   src/sketch.cpp
   src/recovery.cpp
   src/util.cpp)
-add_dependencies(GraphZeppelin GutterTree StreamingUtilities)
-target_link_libraries(GraphZeppelin PUBLIC xxhash GutterTree StreamingUtilities )
+add_dependencies(GraphZeppelin GutterTree StreamingUtilities highway)
+target_link_libraries(GraphZeppelin PUBLIC xxhash GutterTree StreamingUtilities highway)
 target_include_directories(GraphZeppelin PUBLIC include/)
 target_compile_options(GraphZeppelin PUBLIC -fopenmp)
 target_link_options(GraphZeppelin PUBLIC -fopenmp)
@@ -119,8 +137,8 @@ add_library(GraphZeppelinVerifyCC
   src/recovery.cpp
   src/util.cpp
   test/util/graph_verifier.cpp)
-add_dependencies(GraphZeppelinVerifyCC GutterTree StreamingUtilities)
-target_link_libraries(GraphZeppelinVerifyCC PUBLIC xxhash GutterTree StreamingUtilities )
+add_dependencies(GraphZeppelinVerifyCC GutterTree StreamingUtilities highway)
+target_link_libraries(GraphZeppelinVerifyCC PUBLIC xxhash GutterTree StreamingUtilities highway )
 target_include_directories(GraphZeppelinVerifyCC PUBLIC include/ include/test/)
 target_compile_options(GraphZeppelinVerifyCC PUBLIC -fopenmp)
 target_link_options(GraphZeppelinVerifyCC PUBLIC -fopenmp)

include/sketch.h

@@ -8,11 +8,17 @@
 #include <cmath>
 #include <mutex>
 
+#include <hwy/highway.h>
+#include <hwy/aligned_allocator.h>
+
 #include "util.h"
 #include "bucket.h"
 
 #include "bucket_buffer.h"
 
+
+// FOR VECTORIZING CONTIGUOUS XOR
+
 // enum SerialType {
 //   FULL,
 //   RANGE,

include/util.h

@@ -1,9 +1,41 @@
 #pragma once
 #include <string>
 #include <tuple>
+#include <iostream>
 
 #include "types.h"
 
+#include <hwy/highway.h>
+
+
+HWY_BEFORE_NAMESPACE();
+namespace hwy {
+namespace HWY_NAMESPACE {
+static inline void simd_xor(uint32_t* dst, const uint32_t* src, size_t count) {
+  using namespace hwy;
+  const ScalableTag<uint32_t> d;
+  for (size_t i = 0; i < count; i += Lanes(d)) {
+    auto v_dst = LoadU(d, dst + i);
+    auto v_src = LoadU(d, src + i);
+    auto v_xor = Xor(v_dst, v_src);
+    StoreU(v_xor, d, dst + i);
+  }
+}
+static inline void simd_xor_aligned(uint32_t* dst, const uint32_t* src, size_t count) {
+  using namespace hwy;
+  const ScalableTag<uint32_t> d;
+  for (size_t i = 0; i < count; i += Lanes(d)) {
+    auto v_dst = Load(d, dst + i);
+    auto v_src = Load(d, src + i);
+    auto v_xor = Xor(v_dst, v_src);
+    Store(v_xor, d, dst + i);
+  }
+}
+} // namespace HWY_NAMESPACE
+} // namespace hwy
+HWY_AFTER_NAMESPACE();
+
+
 /**
  * Cast a double to unsigned long long with epsilon adjustment.
  * @param d         the double to cast.