issue/130: use collapse

src/infiniop/ops/gemm/cpu/gemm_cpu.cc

@@ -42,14 +42,10 @@ void calculate(
     if (info.is_transed) {
         std::swap(a, b);
     }
-#pragma omp parallel for
-    for (ptrdiff_t index = 0; index < ptrdiff_t(info.batch * info.m * info.n); ++index) {
-        size_t ind = index;
-        size_t n_ = ind % info.n;
-        ind /= info.n;
-        size_t m_ = ind % info.m;
-        ind /= info.m;
-        size_t i = ind;
+#pragma omp parallel for collapse(3)
+    for (size_t i = 0; i < info.batch; ++i) {
+        for (size_t m_ = 0; m_ < info.m; ++m_) {
+            for (size_t n_ = 0; n_ < info.n; ++n_) {
                 auto c_ = reinterpret_cast<Tdata *>(c) + i * info.c_matrix.stride + m_ * info.c_matrix.row_stride + n_ * info.c_matrix.col_stride;
                 float sum = 0;
                 for (size_t k_ = 0; k_ < info.k; ++k_) {
@@ -71,6 +67,8 @@ void calculate(
                     *c_ = beta * (*c_) + alpha * sum;
                 }
             }
+        }
+    }
 }
 
 infiniStatus_t Descriptor::calculate(