Enable compiles with clang-cl on windows

.github/workflows/x86-windows.yml

@@ -35,23 +35,29 @@ jobs:
       with:
         submodules: recursive
 
+    - uses: ilammy/msvc-dev-cmd@v1
+      with:
+        arch: x64
+
     - name: Setup vcpkg with caching
       uses: friendlyanon/setup-vcpkg@v1
       with: 
         committish: c9f906558f9bb12ee9811d6edc98ec9255c6cda5
         path: vcpkg
 
     - name: Configure CMake
-      run: cmake -B ${{github.workspace}}/build-windows -DCMAKE_BUILD_TYPE=${{env.BUILD_TYPE}} -DCMAKE_TOOLCHAIN_FILE:STRING=${{github.workspace}}/vcpkg/scripts/buildsystems/vcpkg.cmake -DCMAKE_EXPORT_COMPILE_COMMANDS:BOOL=TRUE -A"x64" -G"Visual Studio 17 2022"
+      shell: bash
+      run: VCPKG_ROOT=./vcpkg ./toolchains/windows/cmake_configure.sh
 
     - name: Build
-      run: cmake --build ${{github.workspace}}/build-windows --config ${{env.BUILD_TYPE}}
+      shell: bash
+      run: ninja -C build-windows
 
-    - name: Copy files (Release)
+    - name: Copy files 
       shell: bash
       run: ./toolchains/windows/create_package.sh
 
-    - name: Upload files (Release)
+    - name: Upload files 
       uses: actions/upload-artifact@v3
       with:
         name: dab_radio_windows_x64

CMakeLists.txt

@@ -13,7 +13,7 @@ find_package(implot REQUIRED)
 set(viterbi_DIR ${CMAKE_CURRENT_SOURCE_DIR}/vendor/viterbi_decoder)
 find_package(viterbi CONFIG REQUIRED)
 
-if(DEFINED WIN32) 
+if(WIN32) 
     set(rtlsdr_DIR ${CMAKE_CURRENT_SOURCE_DIR}/vendor/librtlsdr)
     find_package(rtlsdr CONFIG REQUIRED)
     set(RTLSDR_LIBS rtlsdr::rtlsdr)
@@ -23,7 +23,7 @@ else()
     set(RTLSDR_LIBS PkgConfig::rtlsdr)
 endif()
 
-if(MSVC)
+if(WIN32)
     find_package(portaudio CONFIG REQUIRED)
     find_package(FFTW3f CONFIG REQUIRED)
     set(PORTAUDIO_LIBS portaudio)
@@ -39,15 +39,15 @@ endif()
 target_compile_definitions(easyloggingpp PRIVATE ELPP_THREAD_SAFE)
 
 # for posix threads
-if(NOT DEFINED WIN32) 
+if(NOT WIN32) 
     find_package(Threads REQUIRED)
     add_compile_options(-pthread) 
     link_libraries(Threads::Threads)
 endif()
 
 # simd and math compile options
 if(NOT ${CMAKE_SYSTEM_PROCESSOR} STREQUAL "aarch64")
-    if(DEFINED MSVC)
+    if(MSVC)
         add_compile_definitions(_CRT_SECURE_NO_WARNINGS)
         add_compile_definitions(_SILENCE_NONFLOATING_COMPLEX_DEPRECATION_WARNING)
         # for msvc multi processor compilation

src/ofdm/dsp/apply_pll.cpp

@@ -50,9 +50,13 @@ void create_dt_step_simd(float* arr, const size_t K, const float dt) {
     }
 }
 
+#if defined(_MSC_VER) && !defined(__clang__)
+#define IS_INTEL_SVML 1
+#endif
+
 #if defined(__SSE3__)
 
-#if !defined(_MSC_VER)
+#if !IS_INTEL_SVML
 #define USE_SSE_AUTO
 #define SSE_MATHFUN_WITH_CODE
 #include "./x86/sse_mathfun.h"
@@ -104,7 +108,7 @@ float apply_pll_sse3(
 
 #if defined(__AVX__)
 
-#if !defined(_MSC_VER)
+#if !IS_INTEL_SVML
 #include "./x86/avx_mathfun.h"
 #define _mm256_cos_ps(x) cos256_ps(x)
 #endif

src/ofdm/dsp/x86/sse_mathfun.h

@@ -200,8 +200,8 @@ _PS_CONST(cephes_log_q2, 0.693359375);
 //    _PD_CONST_TYPE(mant_mask, int, 0x7f800000);
 //    _PD_CONST_TYPE(inv_mant_mask, int, ~0x7f800000);
 
-    _PD_CONST_TYPE(sign_mask, uint64_t, 0x8000000000000000LL);
-    _PD_CONST_TYPE(inv_sign_mask, uint64_t, ~0x8000000000000000LL);
+    _PD_CONST_TYPE(sign_mask, uint64_t, 0x8000000000000000ULL);
+    _PD_CONST_TYPE(inv_sign_mask, uint64_t, ~0x8000000000000000ULL);
 
 #endif
 
@@ -1307,18 +1307,18 @@ static inline double scalCumSumSumSq( double *xa, int n, double *sumSQ )
 
 #    if defined(__x86_64__) || defined(x86_64) || defined(_LP64)
 //      static inline v2df _mm_abs_pd( v2df a )
-//      { _PD_CONST_TYPE(abs_mask, uint64_t, ~0x8000000000000000LL);
+//      { _PD_CONST_TYPE(abs_mask, uint64_t, ~0x8000000000000000ULL);
 //          return _mm_and_pd(a, *(v2df*)_pd_abs_mask);
 //      }
     /*!
         SSE2 'intrinsic' to take the absolute value of a
      */
     static inline v2df _mm_abs_pd(  v2df a )
-    { const static uint64_t am1[2] = {~0x8000000000000000LL,~0x8000000000000000LL};
+    { const static uint64_t am1[2] = {~0x8000000000000000ULL,~0x8000000000000000ULL};
         return _mm_and_pd(a, *((v2df*)am1) );
     }
     static inline double _mm_abs_sd( double a )
-    { const static uint64_t am2 = {~0x8000000000000000LL};
+    { const static uint64_t am2 = {~0x8000000000000000ULL};
       v2si r = _mm_and_si64( *((v2si*)&a), *((v2si*)&am2) );
         return *((double*) &r);
     }

toolchains/windows/README.md

@@ -1,4 +1,19 @@
 ## Instructions
-1. <code>./toolchains/windows/cmake_configure.bat</code>
-2. <code>./toolchains/windows/build.bat</code>
-3. <code>./build-windows/\<program\></code>
+1. ```./toolchains/windows/cmake_configure.sh```
+2. ```ninja -C build-windows```
+3. ```./build-windows/\<program\>```
+
+## Compiling with clang-cl
+Compiling with clang gives better performing binaries compared to MSVC cl.
+
+1. Setup x64 developer environment for C++ for MSVC.
+2. Make sure clang for C++ is installed through Visual Studio.
+3. Enter a bash shell.
+4. ```CC=clang CXX=clang++ ./toolchains/windows/cmake_configure.sh```
+
+## Compiling with different SIMD flags
+The default windows build compiles with AVX2 instructions. If your CPU doesn't support these then you will have to modify the scripts.
+
+```vcpkg.json``` contains ```fftw3f``` which uses the ```avx2``` feature. Change this to ```[avx, sse2, sse]```.
+
+```CMakeLists.txt``` contains ```/arch:AVX2```. Change this to ```/arch:AVX``` or remove this compiler option.

toolchains/windows/cmake_configure.sh

@@ -0,0 +1,21 @@
+#!/bin/sh
+if [[ -z "$VCPKG_ROOT" ]]; then
+    TOOLCHAIN_FILE="C:/tools/vcpkg/scripts/buildsystems/vcpkg.cmake"
+else
+    TOOLCHAIN_FILE="$VCPKG_ROOT/scripts/buildsystems/vcpkg.cmake"
+fi
+
+if [[ -z "$BUILD_OUT" ]]; then
+    BUILD_OUT="build-windows"
+fi
+
+if [[ -z "$BUILD_TYPE" ]]; then
+    BUILD_TYPE="Release"
+fi
+
+cmake .\
+ -B $BUILD_OUT\
+ -G Ninja\
+ -DCMAKE_BUILD_TYPE:STRING=$BUILD_TYPE\
+ -DCMAKE_TOOLCHAIN_FILE:STRING=$TOOLCHAIN_FILE\
+ -DCMAKE_EXPORT_COMPILE_COMMANDS:BOOL=TRUE