[Fix] fix libxsmm build issues on Mac OS (#5626)

CMakeLists.txt

@@ -202,12 +202,16 @@ endif(NOT MSVC)
 
 # Compile LIBXSMM
 if((NOT MSVC) AND USE_LIBXSMM)
+  set(LIBXSMM_COMPILER "${CMAKE_C_COMPILER}")
+  if (APPLE)
+    set(LIBXSMM_COMPILER "${CMAKE_C_COMPILER} -isysroot ${CMAKE_OSX_SYSROOT}")
+  endif()
   if(REBUILD_LIBXSMM)
-    add_custom_target(libxsmm COMMAND make realclean COMMAND make -j ECFLAGS="-Wno-error=deprecated-declarations" BLAS=0 CC=${CMAKE_C_COMPILER}
+    add_custom_target(libxsmm COMMAND make realclean COMMAND make -j ECFLAGS="-Wno-error=deprecated-declarations" BLAS=0 CC=${LIBXSMM_COMPILER}
                       WORKING_DIRECTORY ${CMAKE_SOURCE_DIR}/third_party/libxsmm
                       )
   else(REBUILD_LIBXSMM)
-    add_custom_target(libxsmm COMMAND make -j ECFLAGS="-Wno-error=deprecated-declarations" BLAS=0 CC=${CMAKE_C_COMPILER}
+    add_custom_target(libxsmm COMMAND make -j ECFLAGS="-Wno-error=deprecated-declarations" BLAS=0 CC=${LIBXSMM_COMPILER}
                       WORKING_DIRECTORY ${CMAKE_SOURCE_DIR}/third_party/libxsmm
                       )
   endif(REBUILD_LIBXSMM)

src/array/cpu/spmm_blocking_libxsmm.h

@@ -42,8 +42,12 @@ struct CSRMatrixInternal {
 };
 
 int32_t GetLLCSize() {
+#ifdef _SC_LEVEL3_CACHE_SIZE
   int32_t cache_size = sysconf(_SC_LEVEL3_CACHE_SIZE);
   if (cache_size < 0) cache_size = DGL_CPU_LLC_SIZE;
+#else
+  int32_t cache_size = DGL_CPU_LLC_SIZE;
+#endif
   return cache_size;
 }
 
@@ -292,9 +296,9 @@ inline void SpMMBlockwiseOpSum(
     const DType *E, DType *C, bool has_idx, IdType N, IdType num_M_blocks,
     IdType num_K_blocks, IdType M_block_size,
     libxsmm_meltwfunction_opreduce_vecs_idx kernel) {
-  DType(*in_matrix1)[N] = (DType(*)[N])B;
-  DType(*in_matrix2)[N] = (DType(*)[N])E;
-  DType(*output)[N] = (DType(*)[N])C;
+  const DType *in_matrix1 = B;
+  const DType *in_matrix2 = E;
+  DType *output = C;
 #pragma omp parallel
   {
     for (IdType k = 0; k < num_K_blocks; k++) {
@@ -313,13 +317,13 @@ inline void SpMMBlockwiseOpSum(
           params.n = row_end - row_start;
           params.indices = &cur_csr.indices[row_start];
           params.in_matrix = in_matrix1;
-          params.out_vec = &output[dst][0];
+          params.out_vec = &output[dst * N];
           params.scale_vals = nullptr;
           if (has_idx) {
             params.in_matrix2 = in_matrix2;
             params.indices2 = &cur_csr.data[row_start];
           } else {
-            params.in_matrix2 = &in_matrix2[row_start];
+            params.in_matrix2 = &in_matrix2[row_start * N];
           }
           kernel(&params);
         }
@@ -351,11 +355,11 @@ inline void SpMMBlockwiseOpCmp(
     const DType *E, DType *C, IdType *argB, IdType *argE, bool has_idx,
     IdType N, IdType num_M_blocks, IdType num_K_blocks, IdType M_block_size,
     libxsmm_meltwfunction_opreduce_vecs_idx kernel) {
-  DType(*in_matrix1)[N] = (DType(*)[N])B;
-  DType(*in_matrix2)[N] = (DType(*)[N])E;
-  DType(*output)[N] = (DType(*)[N])C;
-  IdType(*out_matrix1)[N] = (IdType(*)[N])argB;
-  IdType(*out_matrix2)[N] = (IdType(*)[N])argE;
+  const DType *in_matrix1 = B;
+  const DType *in_matrix2 = E;
+  DType *output = C;
+  IdType *out_matrix1 = argB;
+  IdType *out_matrix2 = argE;
 
 #pragma omp parallel
   {
@@ -375,15 +379,15 @@ inline void SpMMBlockwiseOpCmp(
           params.n = row_end - row_start;
           params.indices = &cur_csr.indices[row_start];
           params.in_matrix = in_matrix1;
-          params.out_vec = &output[dst][0];
-          params.argop_off_vec_0 = &out_matrix1[dst][0];
-          params.argop_off_vec_1 = &out_matrix2[dst][0];
+          params.out_vec = &output[dst * N];
+          params.argop_off_vec_0 = &out_matrix1[dst * N];
+          params.argop_off_vec_1 = &out_matrix2[dst * N];
           params.scale_vals = nullptr;
           if (has_idx) {
             params.in_matrix2 = in_matrix2;
             params.indices2 = &cur_csr.data[row_start];
           } else {
-            params.in_matrix2 = &in_matrix2[row_start];
+            params.in_matrix2 = &in_matrix2[row_start * N];
           }
           kernel(&params);
         }