X-Git-Url: http://git.sourceforge.jp/view?p=pf3gnuchains%2Fgcc-fork.git;a=blobdiff_plain;f=libgfortran%2Fgenerated%2Fmatmul_c8.c;h=65cc0a52c4b49736e8cf5bcf667d3d5a8f7683a3;hp=f22719df505a38e73b45c28e42bd1506d18d1e29;hb=4e8e57b0ce67551ca61b7883e73586ba805f0a61;hpb=9c384dcabc464328ebe0e4da1da978416882e344

diff --git a/libgfortran/generated/matmul_c8.c b/libgfortran/generated/matmul_c8.c
index f22719df505..65cc0a52c4b 100644
--- a/libgfortran/generated/matmul_c8.c
+++ b/libgfortran/generated/matmul_c8.c
@@ -36,6 +36,16 @@ Boston, MA 02110-1301, USA.  */
 
 #if defined (HAVE_GFC_COMPLEX_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_COMPLEX_8 *, const GFC_COMPLEX_8 *,
+                          const int *, const GFC_COMPLEX_8 *, const int *,
+                          const GFC_COMPLEX_8 *, GFC_COMPLEX_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:
@@ -56,18 +66,24 @@ Boston, MA 02110-1301, USA.  */
        DO I=1,M
          S = 0
          DO K=1,COUNT
-           S = S+A(I,K)+B(K,J)
+           S = S+A(I,K)*B(K,J)
          C(I,J) = S
    ENDIF
 */
 
+/* If try_blas is set to a nonzero value, then the matmul function will
+   see if there is a way to perform the matrix multiplication by a call
+   to the BLAS gemm function.  */
+
 extern void matmul_c8 (gfc_array_c8 * const restrict retarray, 
-	gfc_array_c8 * const restrict a, gfc_array_c8 * const restrict b);
+	gfc_array_c8 * const restrict a, gfc_array_c8 * const restrict b, int try_blas,
+	int blas_limit, blas_call gemm);
 export_proto(matmul_c8);
 
 void
 matmul_c8 (gfc_array_c8 * const restrict retarray, 
-	gfc_array_c8 * const restrict a, gfc_array_c8 * const restrict b)
+	gfc_array_c8 * const restrict a, gfc_array_c8 * const restrict b, int try_blas,
+	int blas_limit, blas_call gemm)
 {
   const GFC_COMPLEX_8 * restrict abase;
   const GFC_COMPLEX_8 * restrict bbase;
@@ -177,6 +193,31 @@ matmul_c8 (gfc_array_c8 * const restrict retarray,
   bbase = b->data;
   dest = retarray->data;
 
+
+  /* Now that everything is set up, we're performing the multiplication
+     itself.  */
+
+#define POW3(x) (((float) (x)) * ((float) (x)) * ((float) (x)))
+
+  if (try_blas && rxstride == 1 && (axstride == 1 || aystride == 1)
+      && (bxstride == 1 || bystride == 1)
+      && (((float) xcount) * ((float) ycount) * ((float) count)
+          > POW3(blas_limit)))
+  {
+    const int m = xcount, n = ycount, k = count, ldc = rystride;
+    const GFC_COMPLEX_8 one = 1, zero = 0;
+    const int lda = (axstride == 1) ? aystride : axstride,
+              ldb = (bxstride == 1) ? bystride : bxstride;
+
+    if (lda > 0 && ldb > 0 && ldc > 0 && m > 1 && n > 1 && k > 1)
+      {
+        assert (gemm != NULL);
+        gemm (axstride == 1 ? "N" : "T", bxstride == 1 ? "N" : "T", &m, &n, &k,
+              &one, abase, &lda, bbase, &ldb, &zero, dest, &ldc, 1, 1);
+        return;
+      }
+  }
+
   if (rxstride == 1 && axstride == 1 && bxstride == 1)
     {
       const GFC_COMPLEX_8 * restrict bbase_y;