issue/46: Fix omp for loop iterator to ssize_t in calculate()

issue/46: Change ssize_t to ptrdiff_t for the counter in openmp loop

.github/workflows/build.yml

@@ -26,13 +26,23 @@ jobs:
     - name: check format
       run: python3 scripts/format.py --path src --check
 
+    - name: Install OpenMP
+      if: matrix.os == 'ubuntu-latest'
+      run: sudo apt-get install -y libomp-dev
+
+    - name: Install LLVM for OpenMP (Windows)
+      if: matrix.os == 'windows-latest'
+      run: |
+          choco install llvm -y
+          echo "C:\Program Files\LLVM\bin" >> $env:GITHUB_PATH
+
     - name: install xmake
       uses: xmake-io/github-action-setup-xmake@v1
       with:
         xmake-version: latest
 
     - name: configure xmake
-      run: xmake f -cv
+      run: xmake f --omp=y -cv
 
     - name: build with xmake
       run: xmake build

src/infiniop/binary/cpu/binary.h

@@ -156,10 +156,10 @@ void calculate(BinaryCpuInfo info, void *c, const void *a, const void *b, Args &
     auto a_ = reinterpret_cast<const Tdata *>(a);
     auto b_ = reinterpret_cast<const Tdata *>(b);
     auto c_ = reinterpret_cast<Tdata *>(c);
-    ssize_t data_size = info.c_data_size;
+    ptrdiff_t data_size = info.c_data_size;
 
 #pragma omp parallel for
-    for (ssize_t i = 0; i < data_size; ++i) {
+    for (ptrdiff_t i = 0; i < data_size; ++i) {
         size_t a_index = info.broadcasted ? indexToReducedOffset(i, info.ndim, info.c_strides.data(), info.a_strides.data())
                                           : indexToOffset(i, info.ndim, info.a_shape.data(), info.a_strides.data());
         size_t b_index = info.broadcasted ? indexToReducedOffset(i, info.ndim, info.c_strides.data(), info.b_strides.data())

xmake/cpu.lua

@@ -37,3 +37,8 @@ target("infinirt-cpu")
     set_languages("cxx17")
     add_files("../src/infinirt/cpu/*.cc")
 target_end()
+
+if has_config("omp") then
+    add_requires("openmp")
+    add_packages("openmp")
+end
\ No newline at end of file