Porting to MSVC (#28)

* Improve: better support for Windows and MSVC
* Improve: Move logging around
* Improve: CMakeLists for MSVC some more
* Fix: Missing OpenBLAS config and macros
* Make: Uniform conditions for MSVC
* Docs: Style and links
* Docs: Mention MSVC compatibility
* Docs: Notes on Chrono

---------

Co-authored-by: Ash Vardanian <[email protected]>

.vscode/settings.json

@@ -17,13 +17,15 @@
     "colsb",
     "consteval",
     "coro",
+    "cplusplus",
     "cppcoro",
     "CTRE",
     "CUDA",
     "denormal",
     "DOTPROD",
     "Dusíková",
     "Eigen",
+    "Eron",
     "excerise",
     "fconcepts",
     "Fedor",
@@ -53,6 +55,7 @@
     "Niels",
     "nlohmann",
     "NVCC",
+    "openblas",
     "openmp",
     "Ormrod",
     "Peta",

CMakeLists.txt

@@ -1,5 +1,5 @@
 # Let's used CMake 3.16+ for native sanitizers support
-cmake_minimum_required(VERSION 3.16 FATAL_ERROR) 
+cmake_minimum_required(VERSION 3.16 FATAL_ERROR)
 
 # ------------------------------------------------------------------------------
 # Project Setup
@@ -32,17 +32,25 @@ endif()
 # ------------------------------------------------------------------------------
 find_package(Threads REQUIRED)
 find_package(OpenMP REQUIRED)
-find_package(BLAS REQUIRED)
-if (BLAS_FOUND)
-    message(STATUS "BLAS found: ${BLAS_LIBRARIES}")
-else ()
-    message(FATAL_ERROR "BLAS not found")
-endif ()
-
 
 set(FETCHCONTENT_QUIET OFF)
 include(FetchContent)
 
+# Fetch and build OpenBLAS
+FetchContent_Declare(
+  OpenBLAS
+  GIT_REPOSITORY https://github.com/xianyi/OpenBLAS.git
+  GIT_TAG v0.3.29
+)
+
+# Set OpenBLAS build options
+set(NOFORTRAN ON CACHE BOOL "Disable Fortran" FORCE)
+set(BUILD_WITHOUT_LAPACK OFF CACHE BOOL "Build without LAPACK" FORCE)
+set(USE_THREAD ON CACHE BOOL "Use threading" FORCE)
+
+# Make OpenBLAS available
+FetchContent_MakeAvailable(OpenBLAS)
+
 # GTest (required by Google Benchmark)
 FetchContent_Declare(
   GoogleTest
@@ -105,7 +113,7 @@ endif()
 FetchContent_Declare(
   VictorZverovichFMT
   GIT_REPOSITORY https://github.com/fmtlib/fmt.git
-  GIT_TAG 11.1.0
+  GIT_TAG 11.1.2
 )
 FetchContent_MakeAvailable(VictorZverovichFMT)
 
@@ -189,19 +197,23 @@ add_executable(less_slow less_slow.cpp)
 set_target_properties(less_slow PROPERTIES POSITION_INDEPENDENT_CODE ON)
 
 # Conditionally add the assembly file(s)
-if(CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64")
-    set_source_files_properties(less_slow_amd64.S PROPERTIES LANGUAGE ASM)
-    target_sources(less_slow PRIVATE less_slow_amd64.S)
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")
-    set_source_files_properties(less_slow_aarch64.S PROPERTIES LANGUAGE ASM)
-    target_sources(less_slow PRIVATE less_slow_aarch64.S)
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64|amd64|AMD64|x64")
+  set_source_files_properties(less_slow_amd64.S PROPERTIES LANGUAGE ASM)
+  target_sources(less_slow PRIVATE less_slow_amd64.S)
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64|ARM64")
+  set_source_files_properties(less_slow_aarch64.S PROPERTIES LANGUAGE ASM)
+  target_sources(less_slow PRIVATE less_slow_aarch64.S)
 endif()
 
 # ------------------------------------------------------------------------------
 # Compiler Flags / Options
 # ------------------------------------------------------------------------------
-if(NOT CMAKE_SYSTEM_NAME STREQUAL "Darwin")
-  # Apple Clang doesn't support -march=native
+# Check for compiler support of `-march=native`
+if(CMAKE_CXX_COMPILER_ID MATCHES "Intel")
+  target_compile_options(less_slow PRIVATE -xHost)
+elseif(CMAKE_CXX_COMPILER_ID MATCHES "AppleClang" OR CMAKE_CXX_COMPILER_ID MATCHES "MSVC")
+# Apple's Clang and MSVC can't auto-detect the highest CPU features
+else()
   target_compile_options(less_slow PRIVATE -march=native)
 endif()
 
@@ -213,8 +225,15 @@ if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
     -fconcepts-diagnostics-depth=10 # Needed to debug concepts
     -fopenmp # OpenMP support, also requires linking
   )
+elseif(CMAKE_CXX_COMPILER_ID MATCHES "MSVC")
+  target_compile_options(less_slow PRIVATE
+    /MP # Build with multiple processes; equivalent to `make -j` except it spans across all cores by default
+    /wd4068 # Disable the "unknown pragma" warning, as StringZilla uses many GCC and Clang pragmas
+    /Zc:__cplusplus # Make `__cplusplus` macro actually match used standard
+    /Zc:preprocessor # Use conformant preprocessor
+  )
+
 else()
-  # For other compilers (Clang, MSVC, Intel, etc.)
   target_compile_options(less_slow PRIVATE
     -Wno-deprecated-pragma
   )
@@ -230,28 +249,48 @@ if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU" OR CMAKE_CXX_COMPILER_ID MATCHES "Clang"
   )
   set_property(TARGET less_slow PROPERTY SANITIZE_ADDRESS TRUE)
   set_property(TARGET less_slow PROPERTY SANITIZE_UNDEFINED TRUE)
+elseif(CMAKE_CXX_COMPILER_ID MATCHES "MSVC")
+  target_compile_options(less_slow PRIVATE
+    $<$<CONFIG:Release>:/O2>
+    $<$<CONFIG:Release>:/Ob2>
+    $<$<CONFIG:Release>:/Oi>
+    $<$<CONFIG:Release>:/Ot>
+    $<$<CONFIG:Release>:/GL>
+  )
+  target_link_options(less_slow PRIVATE
+    $<$<CONFIG:Release>:/LTCG:incremental>
+  )
 endif()
 
 # ------------------------------------------------------------------------------
 # Link Libraries
 # ------------------------------------------------------------------------------
+# Add OpenBLAS include directory manually
+if(openblas_POPULATED)
+  target_include_directories(less_slow PRIVATE ${openblas_SOURCE_DIR})
+
+  # For config.h
+  target_include_directories(less_slow PRIVATE ${openblas_BINARY_DIR})
+endif()
+
 target_link_libraries(less_slow
   PRIVATE
-    Threads::Threads
-    benchmark
-    fmt::fmt
-    range-v3
-    cppcoro
-    unifex
-    stringzilla
-    yyjson
-    ctre
-    # There is no `absl` shortcut:
-    # https://github.com/abseil/abseil-cpp/blob/master/CMake/README.md#available-abseil-cmake-public-targets
-    absl::flat_hash_map
-    nlohmann_json::nlohmann_json
-    Eigen3::Eigen
-    ${BLAS_LIBRARIES}
-    $<$<STREQUAL:${CMAKE_SYSTEM_NAME},Linux>:TBB::tbb>
-    $<$<STREQUAL:${CMAKE_SYSTEM_NAME},Linux>:OpenMP::OpenMP_CXX>
+  Threads::Threads
+  benchmark
+  fmt::fmt
+  range-v3
+  cppcoro
+  unifex
+  stringzilla
+  yyjson
+  ctre
+  openblas
+
+  # There is no `absl` shortcut:
+  # https://github.com/abseil/abseil-cpp/blob/master/CMake/README.md#available-abseil-cmake-public-targets
+  absl::flat_hash_map
+  nlohmann_json::nlohmann_json
+  Eigen3::Eigen
+  $<$<STREQUAL:${CMAKE_SYSTEM_NAME},Linux>:TBB::tbb>
+  OpenMP::OpenMP_CXX
 )

README.md

@@ -30,14 +30,17 @@ Some of the highlights include:
 - __Is the pointer size really 64 bits__ and how to exploit [pointer-tagging](https://en.wikipedia.org/wiki/Tagged_pointer)?
 - __How many packets is [UDP](https://www.cloudflare.com/learning/ddos/glossary/user-datagram-protocol-udp/) dropping__ and how to serve web requests in [`io_uring`](https://en.wikipedia.org/wiki/Io_uring) from user-space?
 - __Scatter and Gather__ for 50% faster vectorized disjoint memory operations.
-- __How to choose between intrinsics, inline Assembly, and separate Assembly files__ for your performance-critical code?
+- __How to choose between intrinsics, inline Assembly, and separate `.S` files__ for your performance-critical code?
 - __What are Encrypted Enclaves__ and what's the latency of Intel SGX, AMD SEV, and ARM Realm? 🔜
 
 To read, jump to the [`less_slow.cpp` source file](https://github.com/ashvardanian/less_slow.cpp/blob/main/less_slow.cpp) and read the code snippets and comments.
 Follow the instructions below to run the code in your environment and compare it to the comments as you read through the source.
 
 ## Running the Benchmarks
 
+The project aims to be compatible with GCC, Clang, and MSVC compilers on Linux, MacOS, and Windows.
+That said, to cover the broadest functionality, using GCC on Linux is recommended:
+
 - If you are on Windows, it's recommended that you set up a Linux environment using [WSL](https://docs.microsoft.com/en-us/windows/wsl/install).
 - If you are on MacOS, consider using the non-native distribution of Clang from [Homebrew](https://brew.sh) or [MacPorts](https://www.macports.org).
 - If you are on Linux, make sure to install CMake and a recent version of GCC or Clang compilers to support C++20 features.