re PR fortran/86704 (Segmentation fault when using matmul in combination with transpose)

2018-08-25  Thomas Koenig  <tkoenig@gcc.gnu.org>

	PR libfortran/86704
	* m4/matmul_internal.m4: Correct calculation of needed buffer size
	for arrays of shape (1,n).
	* generated/matmul_c10.c: Regenerated
	* generated/matmul_c16.c: Regenerated
	* generated/matmul_c4.c: Regenerated
	* generated/matmul_c8.c: Regenerated
	* generated/matmul_i1.c: Regenerated
	* generated/matmul_i16.c: Regenerated
	* generated/matmul_i2.c: Regenerated
	* generated/matmul_i4.c: Regenerated
	* generated/matmul_i8.c: Regenerated
	* generated/matmul_r10.c: Regenerated
	* generated/matmul_r16.c: Regenerated
	* generated/matmul_r4.c: Regenerated
	* generated/matmul_r8.c: Regenerated
	* generated/matmulavx128_c10.c: Regenerated
	* generated/matmulavx128_c16.c: Regenerated
	* generated/matmulavx128_c4.c: Regenerated
	* generated/matmulavx128_c8.c: Regenerated
	* generated/matmulavx128_i1.c: Regenerated
	* generated/matmulavx128_i16.c: Regenerated
	* generated/matmulavx128_i2.c: Regenerated
	* generated/matmulavx128_i4.c: Regenerated
	* generated/matmulavx128_i8.c: Regenerated
	* generated/matmulavx128_r10.c: Regenerated
	* generated/matmulavx128_r16.c: Regenerated
	* generated/matmulavx128_r4.c: Regenerated
	* generated/matmulavx128_r8.c: Regenerated

2018-08-25  Thomas Koenig  <tkoenig@gcc.gnu.org>

	PR libfortran/86704
	* gfortran.dg/matmul_19.f90: New test.

From-SVN: r263856

diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 13601d2de3bf2bec7366e82f4d3d8a27124d1aef..67f4cdedbb043359f6cf7b0ce29e10ceb2546a17 100644 (file)
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,8 @@
+2018-08-25  Thomas Koenig  <tkoenig@gcc.gnu.org>
+
+       PR libfortran/86704
+       * gfortran.dg/matmul_19.f90: New test.
+
 2018-08-25  Janus Weil  <janus@gcc.gnu.org>
 
        PR fortran/86545

diff --git a/gcc/testsuite/gfortran.dg/matmul_19.f90 b/gcc/testsuite/gfortran.dg/matmul_19.f90
new file mode 100644 (file)
index 0000000..c454924
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/matmul_19.f90
@@ -0,0 +1,25 @@
+! { dg-do  run }
+! { dg-options "-finline-matmul-limit=0" }
+! PR 86704 - this used to segfault.
+
+program testmaticovenasobeni
+implicit none
+  
+  character(len=10) :: line
+  write (unit=line,fmt=*) testmatmul(120,1,3)
+
+  contains
+
+   function testmatmul(m,n,o)
+     integer, intent(in) :: m,n,o
+     real    :: A(n,m),B(n,o),C(m,o)
+     logical :: testmatmul
+    
+     call random_number(A)
+     call random_number(B)
+     
+     C=matmul(transpose(A),B) 
+     testmatmul=.true.
+   end function 
+
+end program testmaticovenasobeni

diff --git a/libgfortran/ChangeLog b/libgfortran/ChangeLog
index 60bb94c86263b303a826d90c37a252b5f4535f1c..76228e26b693b795a0406811a7187cae0849a7b7 100644 (file)
--- a/libgfortran/ChangeLog
+++ b/libgfortran/ChangeLog
@@ -1,3 +1,35 @@
+2018-08-25  Thomas Koenig  <tkoenig@gcc.gnu.org>
+
+       PR libfortran/86704
+       * m4/matmul_internal.m4: Correct calculation of needed buffer size
+       for arrays of shape (1,n).
+       * generated/matmul_c10.c: Regenerated
+       * generated/matmul_c16.c: Regenerated
+       * generated/matmul_c4.c: Regenerated
+       * generated/matmul_c8.c: Regenerated
+       * generated/matmul_i1.c: Regenerated
+       * generated/matmul_i16.c: Regenerated
+       * generated/matmul_i2.c: Regenerated
+       * generated/matmul_i4.c: Regenerated
+       * generated/matmul_i8.c: Regenerated
+       * generated/matmul_r10.c: Regenerated
+       * generated/matmul_r16.c: Regenerated
+       * generated/matmul_r4.c: Regenerated
+       * generated/matmul_r8.c: Regenerated
+       * generated/matmulavx128_c10.c: Regenerated
+       * generated/matmulavx128_c16.c: Regenerated
+       * generated/matmulavx128_c4.c: Regenerated
+       * generated/matmulavx128_c8.c: Regenerated
+       * generated/matmulavx128_i1.c: Regenerated
+       * generated/matmulavx128_i16.c: Regenerated
+       * generated/matmulavx128_i2.c: Regenerated
+       * generated/matmulavx128_i4.c: Regenerated
+       * generated/matmulavx128_i8.c: Regenerated
+       * generated/matmulavx128_r10.c: Regenerated
+       * generated/matmulavx128_r16.c: Regenerated
+       * generated/matmulavx128_r4.c: Regenerated
+       * generated/matmulavx128_r8.c: Regenerated
+
 2018-08-23  David Edelsohn  <dje.gcc@gmail.com>
 
        * async.h (ASYNC_IO): Revert _AIX test.

diff --git a/libgfortran/generated/matmul_c10.c b/libgfortran/generated/matmul_c10.c
index 7cd8a6d2a0dcf5b3ec4efbf98f7eca3ca1119347..462d71e23f5aa95d656aee8771f2aa9327964406 100644 (file)
--- a/libgfortran/generated/matmul_c10.c
+++ b/libgfortran/generated/matmul_c10.c
@@ -317,8 +317,13 @@ matmul_c10_avx (gfc_array_c10 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -869,8 +874,13 @@ matmul_c10_avx2 (gfc_array_c10 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1421,8 +1431,13 @@ matmul_c10_avx512f (gfc_array_c10 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1987,8 +2002,13 @@ matmul_c10_vanilla (gfc_array_c10 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -2613,8 +2633,13 @@ matmul_c10 (gfc_array_c10 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmul_c16.c b/libgfortran/generated/matmul_c16.c
index d93099ea71e48afc8c0cf8762131c8f0bb31566e..2062739ce1f440e4dd4fc8662118198be35556e7 100644 (file)
--- a/libgfortran/generated/matmul_c16.c
+++ b/libgfortran/generated/matmul_c16.c
@@ -317,8 +317,13 @@ matmul_c16_avx (gfc_array_c16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -869,8 +874,13 @@ matmul_c16_avx2 (gfc_array_c16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1421,8 +1431,13 @@ matmul_c16_avx512f (gfc_array_c16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1987,8 +2002,13 @@ matmul_c16_vanilla (gfc_array_c16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -2613,8 +2633,13 @@ matmul_c16 (gfc_array_c16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmul_c4.c b/libgfortran/generated/matmul_c4.c
index eeef3699dcd894166f3ec153b5ff999d32f5b387..91d193dca375398157876f24d742f4b435e0beda 100644 (file)
--- a/libgfortran/generated/matmul_c4.c
+++ b/libgfortran/generated/matmul_c4.c
@@ -317,8 +317,13 @@ matmul_c4_avx (gfc_array_c4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -869,8 +874,13 @@ matmul_c4_avx2 (gfc_array_c4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1421,8 +1431,13 @@ matmul_c4_avx512f (gfc_array_c4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1987,8 +2002,13 @@ matmul_c4_vanilla (gfc_array_c4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -2613,8 +2633,13 @@ matmul_c4 (gfc_array_c4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmul_c8.c b/libgfortran/generated/matmul_c8.c
index 7a73f67193866a32f0dcf2728369f93253a36365..425af85d1bba34b4ec2887acd987ac15e0a82475 100644 (file)
--- a/libgfortran/generated/matmul_c8.c
+++ b/libgfortran/generated/matmul_c8.c
@@ -317,8 +317,13 @@ matmul_c8_avx (gfc_array_c8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -869,8 +874,13 @@ matmul_c8_avx2 (gfc_array_c8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1421,8 +1431,13 @@ matmul_c8_avx512f (gfc_array_c8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1987,8 +2002,13 @@ matmul_c8_vanilla (gfc_array_c8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -2613,8 +2633,13 @@ matmul_c8 (gfc_array_c8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmul_i1.c b/libgfortran/generated/matmul_i1.c
index f3559163bee4edf51310f16476f5a071a5ebbd73..0c9335d4322fe36965f9ee215b1da02fe520df7d 100644 (file)
--- a/libgfortran/generated/matmul_i1.c
+++ b/libgfortran/generated/matmul_i1.c
@@ -317,8 +317,13 @@ matmul_i1_avx (gfc_array_i1 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -869,8 +874,13 @@ matmul_i1_avx2 (gfc_array_i1 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1421,8 +1431,13 @@ matmul_i1_avx512f (gfc_array_i1 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1987,8 +2002,13 @@ matmul_i1_vanilla (gfc_array_i1 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -2613,8 +2633,13 @@ matmul_i1 (gfc_array_i1 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmul_i16.c b/libgfortran/generated/matmul_i16.c
index c8a08eb8e6024b7287e215682ee6dd10cbc9f05b..b9334b3278fd70777dfd501984196cb064db2988 100644 (file)
--- a/libgfortran/generated/matmul_i16.c
+++ b/libgfortran/generated/matmul_i16.c
@@ -317,8 +317,13 @@ matmul_i16_avx (gfc_array_i16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -869,8 +874,13 @@ matmul_i16_avx2 (gfc_array_i16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1421,8 +1431,13 @@ matmul_i16_avx512f (gfc_array_i16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1987,8 +2002,13 @@ matmul_i16_vanilla (gfc_array_i16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -2613,8 +2633,13 @@ matmul_i16 (gfc_array_i16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmul_i2.c b/libgfortran/generated/matmul_i2.c
index ae8a45978b0d91db7f5516db0fb80b366498e203..e4246e948df61d393a2ddd5536b6bce1ec20418e 100644 (file)
--- a/libgfortran/generated/matmul_i2.c
+++ b/libgfortran/generated/matmul_i2.c
@@ -317,8 +317,13 @@ matmul_i2_avx (gfc_array_i2 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -869,8 +874,13 @@ matmul_i2_avx2 (gfc_array_i2 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1421,8 +1431,13 @@ matmul_i2_avx512f (gfc_array_i2 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1987,8 +2002,13 @@ matmul_i2_vanilla (gfc_array_i2 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -2613,8 +2633,13 @@ matmul_i2 (gfc_array_i2 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmul_i4.c b/libgfortran/generated/matmul_i4.c
index 9a3822b754a31f07b20d721621a0f3b0e744e002..78cf27c4fcdbc1e93a70441c8a4f3f6b111991ee 100644 (file)
--- a/libgfortran/generated/matmul_i4.c
+++ b/libgfortran/generated/matmul_i4.c
@@ -317,8 +317,13 @@ matmul_i4_avx (gfc_array_i4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -869,8 +874,13 @@ matmul_i4_avx2 (gfc_array_i4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1421,8 +1431,13 @@ matmul_i4_avx512f (gfc_array_i4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1987,8 +2002,13 @@ matmul_i4_vanilla (gfc_array_i4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -2613,8 +2633,13 @@ matmul_i4 (gfc_array_i4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmul_i8.c b/libgfortran/generated/matmul_i8.c
index 88bfd62923fa1417b43ce456c925d21f50aeb2f0..cf8c401400dfba144fdb37113f3b26a4fe32e6bb 100644 (file)
--- a/libgfortran/generated/matmul_i8.c
+++ b/libgfortran/generated/matmul_i8.c
@@ -317,8 +317,13 @@ matmul_i8_avx (gfc_array_i8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -869,8 +874,13 @@ matmul_i8_avx2 (gfc_array_i8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1421,8 +1431,13 @@ matmul_i8_avx512f (gfc_array_i8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1987,8 +2002,13 @@ matmul_i8_vanilla (gfc_array_i8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -2613,8 +2633,13 @@ matmul_i8 (gfc_array_i8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmul_r10.c b/libgfortran/generated/matmul_r10.c
index d4f13114734fca3e288b9c375409c4d3cb228b96..e4309c80a6b4fbcb74f98646083be9df78b73dc3 100644 (file)
--- a/libgfortran/generated/matmul_r10.c
+++ b/libgfortran/generated/matmul_r10.c
@@ -317,8 +317,13 @@ matmul_r10_avx (gfc_array_r10 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -869,8 +874,13 @@ matmul_r10_avx2 (gfc_array_r10 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1421,8 +1431,13 @@ matmul_r10_avx512f (gfc_array_r10 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1987,8 +2002,13 @@ matmul_r10_vanilla (gfc_array_r10 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -2613,8 +2633,13 @@ matmul_r10 (gfc_array_r10 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmul_r16.c b/libgfortran/generated/matmul_r16.c
index f56d1c3066dd796c6fedc334a1034688133c6c71..1ab554660d7d4f09bb9bc3aa4947fc5506e65e24 100644 (file)
--- a/libgfortran/generated/matmul_r16.c
+++ b/libgfortran/generated/matmul_r16.c
@@ -317,8 +317,13 @@ matmul_r16_avx (gfc_array_r16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -869,8 +874,13 @@ matmul_r16_avx2 (gfc_array_r16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1421,8 +1431,13 @@ matmul_r16_avx512f (gfc_array_r16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1987,8 +2002,13 @@ matmul_r16_vanilla (gfc_array_r16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -2613,8 +2633,13 @@ matmul_r16 (gfc_array_r16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmul_r4.c b/libgfortran/generated/matmul_r4.c
index a5677b96969ff298f2be3cb6151feba0be184b7b..97dba9825b1f366e8b21265aaee4a92d11258f36 100644 (file)
--- a/libgfortran/generated/matmul_r4.c
+++ b/libgfortran/generated/matmul_r4.c
@@ -317,8 +317,13 @@ matmul_r4_avx (gfc_array_r4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -869,8 +874,13 @@ matmul_r4_avx2 (gfc_array_r4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1421,8 +1431,13 @@ matmul_r4_avx512f (gfc_array_r4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1987,8 +2002,13 @@ matmul_r4_vanilla (gfc_array_r4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -2613,8 +2633,13 @@ matmul_r4 (gfc_array_r4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmul_r8.c b/libgfortran/generated/matmul_r8.c
index d9acdd9d92f9a83aaa55ef20aaf9b1322a184e88..5e4c9500260fb3ffabb8da37ff965a0820c58562 100644 (file)
--- a/libgfortran/generated/matmul_r8.c
+++ b/libgfortran/generated/matmul_r8.c
@@ -317,8 +317,13 @@ matmul_r8_avx (gfc_array_r8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -869,8 +874,13 @@ matmul_r8_avx2 (gfc_array_r8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1421,8 +1431,13 @@ matmul_r8_avx512f (gfc_array_r8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -1987,8 +2002,13 @@ matmul_r8_vanilla (gfc_array_r8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -2613,8 +2633,13 @@ matmul_r8 (gfc_array_r8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmulavx128_c10.c b/libgfortran/generated/matmulavx128_c10.c
index 69ad45b334cc95a822d5716996a9808f0af9e16d..5cb0f6ad6f35472d002ea9779b388e1f7bf036ab 100644 (file)
--- a/libgfortran/generated/matmulavx128_c10.c
+++ b/libgfortran/generated/matmulavx128_c10.c
@@ -282,8 +282,13 @@ matmul_c10_avx128_fma3 (gfc_array_c10 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -835,8 +840,13 @@ matmul_c10_avx128_fma4 (gfc_array_c10 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmulavx128_c16.c b/libgfortran/generated/matmulavx128_c16.c
index 0bf8ba3f59912e92023ff9725bf6ecbc13dce38c..66272fefaf91e5c2c97cbfe8fb7f88a45dc306af 100644 (file)
--- a/libgfortran/generated/matmulavx128_c16.c
+++ b/libgfortran/generated/matmulavx128_c16.c
@@ -282,8 +282,13 @@ matmul_c16_avx128_fma3 (gfc_array_c16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -835,8 +840,13 @@ matmul_c16_avx128_fma4 (gfc_array_c16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmulavx128_c4.c b/libgfortran/generated/matmulavx128_c4.c
index 0763d5d1707a4bc6b7ec9841bc443101b69d119a..f6e06e2e88f3225bafa750a720d2b0bf329c196e 100644 (file)
--- a/libgfortran/generated/matmulavx128_c4.c
+++ b/libgfortran/generated/matmulavx128_c4.c
@@ -282,8 +282,13 @@ matmul_c4_avx128_fma3 (gfc_array_c4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -835,8 +840,13 @@ matmul_c4_avx128_fma4 (gfc_array_c4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmulavx128_c8.c b/libgfortran/generated/matmulavx128_c8.c
index d440784c285c24bea81cfbb4b917b15346671fb5..accc69c4d1a20556569fc2bb00ec28f9a96a61e8 100644 (file)
--- a/libgfortran/generated/matmulavx128_c8.c
+++ b/libgfortran/generated/matmulavx128_c8.c
@@ -282,8 +282,13 @@ matmul_c8_avx128_fma3 (gfc_array_c8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -835,8 +840,13 @@ matmul_c8_avx128_fma4 (gfc_array_c8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmulavx128_i1.c b/libgfortran/generated/matmulavx128_i1.c
index efa346145446347add7bce375b2e13e61c936edb..48b15c8074a625cac0109065e67a17561443bfbe 100644 (file)
--- a/libgfortran/generated/matmulavx128_i1.c
+++ b/libgfortran/generated/matmulavx128_i1.c
@@ -282,8 +282,13 @@ matmul_i1_avx128_fma3 (gfc_array_i1 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -835,8 +840,13 @@ matmul_i1_avx128_fma4 (gfc_array_i1 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmulavx128_i16.c b/libgfortran/generated/matmulavx128_i16.c
index 3c0e6064efe41255df6bd8e52a88e4d65dad40ca..319321eca492f872446c588b32d4814a59ecbd6b 100644 (file)
--- a/libgfortran/generated/matmulavx128_i16.c
+++ b/libgfortran/generated/matmulavx128_i16.c
@@ -282,8 +282,13 @@ matmul_i16_avx128_fma3 (gfc_array_i16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -835,8 +840,13 @@ matmul_i16_avx128_fma4 (gfc_array_i16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmulavx128_i2.c b/libgfortran/generated/matmulavx128_i2.c
index 1da7f25189f07e3399da7e15a2ec4b5631c12b24..4d8945b10a24538709750c23459658ee315a4e23 100644 (file)
--- a/libgfortran/generated/matmulavx128_i2.c
+++ b/libgfortran/generated/matmulavx128_i2.c
@@ -282,8 +282,13 @@ matmul_i2_avx128_fma3 (gfc_array_i2 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -835,8 +840,13 @@ matmul_i2_avx128_fma4 (gfc_array_i2 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmulavx128_i4.c b/libgfortran/generated/matmulavx128_i4.c
index ca7de54b9b81005f57760814f6100fb37eb7d0b0..acaa00a30bbef0611dcbab07ba2a4a27c87208da 100644 (file)
--- a/libgfortran/generated/matmulavx128_i4.c
+++ b/libgfortran/generated/matmulavx128_i4.c
@@ -282,8 +282,13 @@ matmul_i4_avx128_fma3 (gfc_array_i4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -835,8 +840,13 @@ matmul_i4_avx128_fma4 (gfc_array_i4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmulavx128_i8.c b/libgfortran/generated/matmulavx128_i8.c
index 731e55d2f630463cc4eceea7f679a2647a7cbf0a..56e85167a3f909e98f3a4cd8fafd42bdd42ce00c 100644 (file)
--- a/libgfortran/generated/matmulavx128_i8.c
+++ b/libgfortran/generated/matmulavx128_i8.c
@@ -282,8 +282,13 @@ matmul_i8_avx128_fma3 (gfc_array_i8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -835,8 +840,13 @@ matmul_i8_avx128_fma4 (gfc_array_i8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmulavx128_r10.c b/libgfortran/generated/matmulavx128_r10.c
index 190e41cbb7138f2a3fcfc3d3174d51a30b996fa6..880c9d921b2b20c6bdf24f239c76c1de5f414c06 100644 (file)
--- a/libgfortran/generated/matmulavx128_r10.c
+++ b/libgfortran/generated/matmulavx128_r10.c
@@ -282,8 +282,13 @@ matmul_r10_avx128_fma3 (gfc_array_r10 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -835,8 +840,13 @@ matmul_r10_avx128_fma4 (gfc_array_r10 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmulavx128_r16.c b/libgfortran/generated/matmulavx128_r16.c
index 2994ed311fe437aacbbb570abe5eb553d3b8517e..328e251a3a11769df77d9e40b2a4e2338719287e 100644 (file)
--- a/libgfortran/generated/matmulavx128_r16.c
+++ b/libgfortran/generated/matmulavx128_r16.c
@@ -282,8 +282,13 @@ matmul_r16_avx128_fma3 (gfc_array_r16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -835,8 +840,13 @@ matmul_r16_avx128_fma4 (gfc_array_r16 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmulavx128_r4.c b/libgfortran/generated/matmulavx128_r4.c
index 4f18b5006d8a4a9ee0a5006e39e9662e62c24112..013a1804a11d5ba3ce8bfabeff3745edbfd8c573 100644 (file)
--- a/libgfortran/generated/matmulavx128_r4.c
+++ b/libgfortran/generated/matmulavx128_r4.c
@@ -282,8 +282,13 @@ matmul_r4_avx128_fma3 (gfc_array_r4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -835,8 +840,13 @@ matmul_r4_avx128_fma4 (gfc_array_r4 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/generated/matmulavx128_r8.c b/libgfortran/generated/matmulavx128_r8.c
index 443d7426f4ffc80ca398d0160e9131a027e902b8..4da59f9e69c7b0605f2c2b61b36fd1feb6626913 100644 (file)
--- a/libgfortran/generated/matmulavx128_r8.c
+++ b/libgfortran/generated/matmulavx128_r8.c
@@ -282,8 +282,13 @@ matmul_r8_avx128_fma3 (gfc_array_r8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 
@@ -835,8 +840,13 @@ matmul_r8_avx128_fma4 (gfc_array_r8 * const restrict retarray,
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;
 

diff --git a/libgfortran/m4/matmul_internal.m4 b/libgfortran/m4/matmul_internal.m4
index 6860f606122ef97afdd80936e393248fdfd2da5a..2020e8a50dfe20e7c7fb491f361aed5025eec20b 100644 (file)
--- a/libgfortran/m4/matmul_internal.m4
+++ b/libgfortran/m4/matmul_internal.m4
@@ -233,8 +233,13 @@ sinclude(`matmul_asm_'rtype_code`.m4')dnl
        return;
 
       /* Adjust size of t1 to what is needed.  */
-      index_type t1_dim;
-      t1_dim = (a_dim1 - (ycount > 1)) * 256 + b_dim1;
+      index_type t1_dim, a_sz;
+      if (aystride == 1)
+        a_sz = rystride;
+      else
+        a_sz = a_dim1;
+
+      t1_dim = a_sz * 256 + b_dim1;
       if (t1_dim > 65536)
        t1_dim = 65536;