Demos of the SWAR library (#69)

Remaining comments are just nits, merging for unblock.

* VSCode interaction

* Builds with update

* Zooification of Lemire 8 bytes

* Benchmarks operative

* Implementations tested & benchmarked with Catch2 too; refactoring of swar benchmarks to share between Google benchmark and Catch2

* Draft of strlen

* Strlen implemented

* Improvement in strlen, SIMD strlen, improved CMake to support AVX in Xcode

---------

Co-authored-by: Eddie <eddie see email elsewhere>

Author: thecppzoo

benchmark/CMakeLists.txt

@@ -1,28 +1,64 @@
-cmake_minimum_required (VERSION 2.6)
+cmake_minimum_required (VERSION 3.8)
 
-set(CMAKE_CXX_FLAGS "-std=c++17 -g3 -fno-unroll-loops")
+project(ZooBenchmark)
 
-set(CMAKE_BUILD_TYPE "Release")
 
-project(ZooBenchmark)
+set(CMAKE_CXX_FLAGS "-fno-unroll-loops")
+
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_BUILD_TYPE "RelWithDebInfo")
+
+if(CMAKE_BUILD_TYPE MATCHES "Release|RelWithDebInfo")
+    if(CMAKE_CXX_COMPILER_ID MATCHES "GNU|Clang")
+        string(APPEND CMAKE_CXX_FLAGS_RELEASE " -march=native -fno-unroll-loops")
+        string(APPEND CMAKE_CXX_FLAGS_RELWITHDEBINFO " -march=native -fno-unroll-loops")
+    endif()
+endif()
+
+# Macro to set properties for Xcode targets
+macro(set_xcode_properties TARGET_NAME)
+    if(CMAKE_GENERATOR STREQUAL Xcode)
+        set_target_properties(${TARGET_NAME} PROPERTIES
+            XCODE_ATTRIBUTE_ENABLE_AVX YES
+            XCODE_ATTRIBUTE_ENABLE_AVX2 YES
+            XCODE_ATTRIBUTE_OTHER_CPLUSPLUSFLAGS "-mavx -mavx2"
+            XCODE_ATTRIBUTE_OTHER_CFLAGS "-mavx -mavx2"
+        )
+    endif()
+endmacro()
 
 include_directories(
     ../test/inc
-    ../../benchmark/include
+    dependencies/google_benchmark/include
     ../inc
     ../junkyard/inc
     ../test/third_party/Catch2/single_include
 )
 
 # build google benchmark (target: benchmark)
 # do not build tests of benchmarking lib
-# set(BENCHMARK_ENABLE_TESTING OFF CACHE BOOL "Suppressing benchmark's tests" FORCE)
+set(BENCHMARK_ENABLE_TESTING OFF CACHE BOOL "Suppressing benchmark's tests" FORCE)
+set(CMAKE_BUILD_TYPE "RelWithDebInfo" CACHE STRING "Build type" FORCE)
+add_subdirectory(dependencies/google_benchmark)
 
-# add_subdirectory(dependencies/google_benchmark)
+add_executable(
+  catch2Benchmark
+    catch2BenchmarkMain.cpp catch2Functions.cpp catch2swar-demo.cpp
+    atoi.cpp
+    egyptian.cpp
+    # RobinHood.benchmark.cpp
+)
+set_xcode_properties(catch2Benchmark)
 
+add_executable(
+    zoo-google-benchmark benchmark_main.cpp cfs.cpp cfs/cfs_utility.cpp
+)
+set_xcode_properties(zoo-google-benchmark)
 
 add_executable(
-  catch2Benchmark
-    catch2BenchmarkMain.cpp catch2Functions.cpp egyptian.cpp
-    RobinHood.benchmark.cpp
+    zoo-atoi-benchmark benchmark_main.cpp bm-swar.cpp atoi.cpp
 )
+set_xcode_properties(zoo-atoi-benchmark)
+
+target_link_libraries(zoo-google-benchmark benchmark::benchmark)
+target_link_libraries(zoo-atoi-benchmark benchmark::benchmark)

benchmark/atoi-corpus.h

@@ -0,0 +1,135 @@
+#include "atoi.h"
+
+#include <vector>
+#include <string>
+#include <cstring>
+#include <random>
+
+struct Corpus8DecimalDigits {
+    std::vector<int> asNumbers_;
+    std::string characters_;
+
+    Corpus8DecimalDigits(std::vector<int> aNs, std::string cs):
+        asNumbers_(aNs),
+        characters_(cs)
+    {}
+
+    template<typename G>
+    static auto makeCorpus(G &generator) {
+        auto count = 1031; // 1031 is a prime number, this helps to disable in
+        // practice the branch predictor, the idea is to measure the performance
+        // of the code under measurement, not how the the unrealistic conditions
+        // of microbenchmarking help/hurt the code under measurement
+        std::string allCharacters;
+        allCharacters.resize(count * 9);
+        std::vector<int> inputs;
+        std::uniform_int_distribution<> range(0, 100*1000*1000 - 1);
+        char *base = allCharacters.data();
+        for(;;) {
+            auto input = range(generator);
+            snprintf(base, 9, "%08d", input);
+            inputs.push_back(input);
+            if(--count) { break; }
+            base += 9;
+        }
+        return Corpus8DecimalDigits(inputs, allCharacters);
+    }
+
+    struct Iterator {
+        Corpus8DecimalDigits *thy;
+        int *ip;
+        char *cp;
+
+        Iterator &operator++() {
+            ++ip;
+            cp += 9;
+            return *this;
+        }
+
+        char *operator*() {
+            return cp;
+        }
+
+        auto next() noexcept {
+            ++(*this);
+            return cp != thy->characters_.data() + thy->characters_.size();
+        }
+    };
+
+    Iterator commence() {
+        return { this, asNumbers_.data(), characters_.data() };
+    }
+};
+
+#define PARSE8BYTES_CORPUS_X_LIST \
+    X(Lemire, parse_eight_digits_swar)\
+    X(Zoo, lemire_as_zoo_swar)\
+    X(LIBC, atoi)
+
+struct CorpusStringLength {
+    std::vector<int> skips_;
+    std::string characters_;
+
+    CorpusStringLength(std::vector<int> &&skips, std::string &&cs):
+        skips_{std::move(skips)}, characters_{std::move(cs)}
+    {}
+
+    template<typename G>
+    static auto makeCorpus(G &generator) {
+        auto count = 1031; // see Corpus8DecimalDigits for why 1031
+        std::vector<int> sizes;
+        std::string allCharacters;
+        std::uniform_int_distribution<> strSize(0, 101); // again a prime
+        std::uniform_int_distribution<> characters(1, 255); // notice 0 excluded
+
+        while(count--) {
+            auto length = strSize(generator);
+            sizes.push_back(length);
+            for(auto i = length; i--; ) {
+                allCharacters.append(1, characters(generator));
+            }
+            allCharacters.append(1, '\0');
+        }
+        return CorpusStringLength(std::move(sizes), std::move(allCharacters));
+    }
+
+    struct Iterator {
+        int *skips, *sentinel;
+        char *cp;
+
+        Iterator &operator++() {
+            cp += *skips++;
+            return *this;
+        }
+
+        char *operator*() {
+            return cp;
+        }
+
+        auto next() noexcept {
+            ++(*this);
+            return sentinel != skips;
+        }
+    };
+
+    Iterator commence() {
+        return {
+            skips_.data(), skips_.data() + skips_.size(), characters_.data()
+        };
+    }
+};
+
+#define STRLEN_CORPUS_X_LIST \
+    X(LIBC_STRLEN, strlen) \
+    X(ZOO_STRLEN, zoo::c_strLength) \
+    X(ZOO_NATURAL_STRLEN, zoo::c_strLength_natural) \
+    X(ZOO_MANUAL_STRLEN, zoo::c_strLength_manualComparison) \
+    X(ZOO_AVX, zoo::avx2_strlen) \
+    X(GENERIC_GLIBC_STRLEN, STRLEN_old)
+
+#define X(Typename, FunctionToCall) \
+    struct Invoke##Typename { int operator()(const char *p) { return FunctionToCall(p); } };
+
+PARSE8BYTES_CORPUS_X_LIST
+STRLEN_CORPUS_X_LIST
+#undef X