-/* This is a C version of the following fortran pseudo-code. The key
- point is the loop order -- we access all arrays column-first, which
- improves the performance enough to boost galgel spec score by 50%.
+
+#if defined (HAVE_GFC_REAL_8)
+
+/* Prototype for the BLAS ?gemm subroutine, a pointer to which can be
+ passed to us by the front-end, in which case we'll call it for large
+ matrices. */
+
+typedef void (*blas_call)(const char *, const char *, const int *, const int *,
+ const int *, const GFC_REAL_8 *, const GFC_REAL_8 *,
+ const int *, const GFC_REAL_8 *, const int *,
+ const GFC_REAL_8 *, GFC_REAL_8 *, const int *,
+ int, int);
+
+/* The order of loops is different in the case of plain matrix
+ multiplication C=MATMUL(A,B), and in the frequent special case where
+ the argument A is the temporary result of a TRANSPOSE intrinsic:
+ C=MATMUL(TRANSPOSE(A),B). Transposed temporaries are detected by
+ looking at their strides.
+
+ The equivalent Fortran pseudo-code is: