Hello world,
the attached patch reduces the stack usage by the blocked
version of matmul for cases where we don't need the full buffer.
This should improve stack usage.
Regression-tested. I also added a stress test (around 3 secs of
CPU time on my system), it will only run once due to the "dg-do run"
hack).
OK for trunk?
Thomas
2017-05-05 Thomas Koenig <tkoe...@gcc.gnu.org>
PR fortran/80602
* m4/matmul_internal.m4: 'matmul_name`: Change
t1 to a VLA of the required size.
* 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.
2017-05-05 Thomas Koenig <tkoe...@gcc.gnu.org>
PR fortran/80602
* gfortran.dg/matmul_15.f90: New test case.
Index: generated/matmul_c10.c
===================================================================
--- generated/matmul_c10.c (Revision 247566)
+++ generated/matmul_c10.c (Arbeitskopie)
@@ -286,8 +286,7 @@ matmul_c10_avx (gfc_array_c10 * const restrict ret
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_10 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_10 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -311,6 +310,17 @@ matmul_c10_avx (gfc_array_c10 * const restrict ret
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_10 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -829,8 +839,7 @@ matmul_c10_avx2 (gfc_array_c10 * const restrict re
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_10 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_10 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -854,6 +863,17 @@ matmul_c10_avx2 (gfc_array_c10 * const restrict re
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_10 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1372,8 +1392,7 @@ matmul_c10_avx512f (gfc_array_c10 * const restrict
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_10 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_10 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1397,6 +1416,17 @@ matmul_c10_avx512f (gfc_array_c10 * const restrict
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_10 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1911,8 +1941,7 @@ matmul_c10_vanilla (gfc_array_c10 * const restrict
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_10 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_10 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1936,6 +1965,17 @@ matmul_c10_vanilla (gfc_array_c10 * const restrict
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_10 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -2508,8 +2548,7 @@ matmul_c10 (gfc_array_c10 * const restrict retarra
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_10 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_10 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -2533,6 +2572,17 @@ matmul_c10 (gfc_array_c10 * const restrict retarra
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_10 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
Index: generated/matmul_c16.c
===================================================================
--- generated/matmul_c16.c (Revision 247566)
+++ generated/matmul_c16.c (Arbeitskopie)
@@ -286,8 +286,7 @@ matmul_c16_avx (gfc_array_c16 * const restrict ret
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_16 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_16 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -311,6 +310,17 @@ matmul_c16_avx (gfc_array_c16 * const restrict ret
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_16 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -829,8 +839,7 @@ matmul_c16_avx2 (gfc_array_c16 * const restrict re
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_16 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_16 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -854,6 +863,17 @@ matmul_c16_avx2 (gfc_array_c16 * const restrict re
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_16 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1372,8 +1392,7 @@ matmul_c16_avx512f (gfc_array_c16 * const restrict
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_16 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_16 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1397,6 +1416,17 @@ matmul_c16_avx512f (gfc_array_c16 * const restrict
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_16 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1911,8 +1941,7 @@ matmul_c16_vanilla (gfc_array_c16 * const restrict
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_16 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_16 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1936,6 +1965,17 @@ matmul_c16_vanilla (gfc_array_c16 * const restrict
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_16 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -2508,8 +2548,7 @@ matmul_c16 (gfc_array_c16 * const restrict retarra
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_16 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_16 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -2533,6 +2572,17 @@ matmul_c16 (gfc_array_c16 * const restrict retarra
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_16 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
Index: generated/matmul_c4.c
===================================================================
--- generated/matmul_c4.c (Revision 247566)
+++ generated/matmul_c4.c (Arbeitskopie)
@@ -286,8 +286,7 @@ matmul_c4_avx (gfc_array_c4 * const restrict retar
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_4 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_4 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -311,6 +310,17 @@ matmul_c4_avx (gfc_array_c4 * const restrict retar
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_4 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -829,8 +839,7 @@ matmul_c4_avx2 (gfc_array_c4 * const restrict reta
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_4 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_4 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -854,6 +863,17 @@ matmul_c4_avx2 (gfc_array_c4 * const restrict reta
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_4 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1372,8 +1392,7 @@ matmul_c4_avx512f (gfc_array_c4 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_4 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_4 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1397,6 +1416,17 @@ matmul_c4_avx512f (gfc_array_c4 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_4 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1911,8 +1941,7 @@ matmul_c4_vanilla (gfc_array_c4 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_4 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_4 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1936,6 +1965,17 @@ matmul_c4_vanilla (gfc_array_c4 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_4 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -2508,8 +2548,7 @@ matmul_c4 (gfc_array_c4 * const restrict retarray,
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_4 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_4 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -2533,6 +2572,17 @@ matmul_c4 (gfc_array_c4 * const restrict retarray,
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_4 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
Index: generated/matmul_c8.c
===================================================================
--- generated/matmul_c8.c (Revision 247566)
+++ generated/matmul_c8.c (Arbeitskopie)
@@ -286,8 +286,7 @@ matmul_c8_avx (gfc_array_c8 * const restrict retar
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_8 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_8 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -311,6 +310,17 @@ matmul_c8_avx (gfc_array_c8 * const restrict retar
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_8 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -829,8 +839,7 @@ matmul_c8_avx2 (gfc_array_c8 * const restrict reta
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_8 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_8 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -854,6 +863,17 @@ matmul_c8_avx2 (gfc_array_c8 * const restrict reta
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_8 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1372,8 +1392,7 @@ matmul_c8_avx512f (gfc_array_c8 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_8 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_8 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1397,6 +1416,17 @@ matmul_c8_avx512f (gfc_array_c8 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_8 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1911,8 +1941,7 @@ matmul_c8_vanilla (gfc_array_c8 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_8 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_8 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1936,6 +1965,17 @@ matmul_c8_vanilla (gfc_array_c8 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_8 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -2508,8 +2548,7 @@ matmul_c8 (gfc_array_c8 * const restrict retarray,
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_COMPLEX_8 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_COMPLEX_8 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -2533,6 +2572,17 @@ matmul_c8 (gfc_array_c8 * const restrict retarray,
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_COMPLEX_8 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
Index: generated/matmul_i1.c
===================================================================
--- generated/matmul_i1.c (Revision 247566)
+++ generated/matmul_i1.c (Arbeitskopie)
@@ -286,8 +286,7 @@ matmul_i1_avx (gfc_array_i1 * const restrict retar
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_1 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_1 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -311,6 +310,17 @@ matmul_i1_avx (gfc_array_i1 * const restrict retar
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_1 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -829,8 +839,7 @@ matmul_i1_avx2 (gfc_array_i1 * const restrict reta
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_1 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_1 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -854,6 +863,17 @@ matmul_i1_avx2 (gfc_array_i1 * const restrict reta
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_1 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1372,8 +1392,7 @@ matmul_i1_avx512f (gfc_array_i1 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_1 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_1 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1397,6 +1416,17 @@ matmul_i1_avx512f (gfc_array_i1 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_1 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1911,8 +1941,7 @@ matmul_i1_vanilla (gfc_array_i1 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_1 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_1 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1936,6 +1965,17 @@ matmul_i1_vanilla (gfc_array_i1 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_1 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -2508,8 +2548,7 @@ matmul_i1 (gfc_array_i1 * const restrict retarray,
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_1 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_1 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -2533,6 +2572,17 @@ matmul_i1 (gfc_array_i1 * const restrict retarray,
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_1 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
Index: generated/matmul_i16.c
===================================================================
--- generated/matmul_i16.c (Revision 247566)
+++ generated/matmul_i16.c (Arbeitskopie)
@@ -286,8 +286,7 @@ matmul_i16_avx (gfc_array_i16 * const restrict ret
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_16 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_16 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -311,6 +310,17 @@ matmul_i16_avx (gfc_array_i16 * const restrict ret
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_16 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -829,8 +839,7 @@ matmul_i16_avx2 (gfc_array_i16 * const restrict re
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_16 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_16 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -854,6 +863,17 @@ matmul_i16_avx2 (gfc_array_i16 * const restrict re
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_16 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1372,8 +1392,7 @@ matmul_i16_avx512f (gfc_array_i16 * const restrict
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_16 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_16 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1397,6 +1416,17 @@ matmul_i16_avx512f (gfc_array_i16 * const restrict
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_16 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1911,8 +1941,7 @@ matmul_i16_vanilla (gfc_array_i16 * const restrict
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_16 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_16 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1936,6 +1965,17 @@ matmul_i16_vanilla (gfc_array_i16 * const restrict
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_16 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -2508,8 +2548,7 @@ matmul_i16 (gfc_array_i16 * const restrict retarra
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_16 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_16 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -2533,6 +2572,17 @@ matmul_i16 (gfc_array_i16 * const restrict retarra
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_16 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
Index: generated/matmul_i2.c
===================================================================
--- generated/matmul_i2.c (Revision 247566)
+++ generated/matmul_i2.c (Arbeitskopie)
@@ -286,8 +286,7 @@ matmul_i2_avx (gfc_array_i2 * const restrict retar
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_2 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_2 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -311,6 +310,17 @@ matmul_i2_avx (gfc_array_i2 * const restrict retar
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_2 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -829,8 +839,7 @@ matmul_i2_avx2 (gfc_array_i2 * const restrict reta
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_2 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_2 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -854,6 +863,17 @@ matmul_i2_avx2 (gfc_array_i2 * const restrict reta
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_2 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1372,8 +1392,7 @@ matmul_i2_avx512f (gfc_array_i2 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_2 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_2 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1397,6 +1416,17 @@ matmul_i2_avx512f (gfc_array_i2 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_2 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1911,8 +1941,7 @@ matmul_i2_vanilla (gfc_array_i2 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_2 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_2 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1936,6 +1965,17 @@ matmul_i2_vanilla (gfc_array_i2 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_2 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -2508,8 +2548,7 @@ matmul_i2 (gfc_array_i2 * const restrict retarray,
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_2 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_2 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -2533,6 +2572,17 @@ matmul_i2 (gfc_array_i2 * const restrict retarray,
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_2 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
Index: generated/matmul_i4.c
===================================================================
--- generated/matmul_i4.c (Revision 247566)
+++ generated/matmul_i4.c (Arbeitskopie)
@@ -286,8 +286,7 @@ matmul_i4_avx (gfc_array_i4 * const restrict retar
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_4 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_4 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -311,6 +310,17 @@ matmul_i4_avx (gfc_array_i4 * const restrict retar
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_4 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -829,8 +839,7 @@ matmul_i4_avx2 (gfc_array_i4 * const restrict reta
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_4 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_4 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -854,6 +863,17 @@ matmul_i4_avx2 (gfc_array_i4 * const restrict reta
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_4 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1372,8 +1392,7 @@ matmul_i4_avx512f (gfc_array_i4 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_4 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_4 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1397,6 +1416,17 @@ matmul_i4_avx512f (gfc_array_i4 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_4 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1911,8 +1941,7 @@ matmul_i4_vanilla (gfc_array_i4 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_4 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_4 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1936,6 +1965,17 @@ matmul_i4_vanilla (gfc_array_i4 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_4 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -2508,8 +2548,7 @@ matmul_i4 (gfc_array_i4 * const restrict retarray,
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_4 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_4 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -2533,6 +2572,17 @@ matmul_i4 (gfc_array_i4 * const restrict retarray,
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_4 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
Index: generated/matmul_i8.c
===================================================================
--- generated/matmul_i8.c (Revision 247566)
+++ generated/matmul_i8.c (Arbeitskopie)
@@ -286,8 +286,7 @@ matmul_i8_avx (gfc_array_i8 * const restrict retar
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_8 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_8 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -311,6 +310,17 @@ matmul_i8_avx (gfc_array_i8 * const restrict retar
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_8 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -829,8 +839,7 @@ matmul_i8_avx2 (gfc_array_i8 * const restrict reta
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_8 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_8 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -854,6 +863,17 @@ matmul_i8_avx2 (gfc_array_i8 * const restrict reta
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_8 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1372,8 +1392,7 @@ matmul_i8_avx512f (gfc_array_i8 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_8 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_8 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1397,6 +1416,17 @@ matmul_i8_avx512f (gfc_array_i8 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_8 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1911,8 +1941,7 @@ matmul_i8_vanilla (gfc_array_i8 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_8 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_8 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1936,6 +1965,17 @@ matmul_i8_vanilla (gfc_array_i8 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_8 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -2508,8 +2548,7 @@ matmul_i8 (gfc_array_i8 * const restrict retarray,
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_INTEGER_8 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_INTEGER_8 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -2533,6 +2572,17 @@ matmul_i8 (gfc_array_i8 * const restrict retarray,
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_INTEGER_8 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
Index: generated/matmul_r10.c
===================================================================
--- generated/matmul_r10.c (Revision 247566)
+++ generated/matmul_r10.c (Arbeitskopie)
@@ -286,8 +286,7 @@ matmul_r10_avx (gfc_array_r10 * const restrict ret
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_10 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_10 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -311,6 +310,17 @@ matmul_r10_avx (gfc_array_r10 * const restrict ret
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_10 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -829,8 +839,7 @@ matmul_r10_avx2 (gfc_array_r10 * const restrict re
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_10 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_10 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -854,6 +863,17 @@ matmul_r10_avx2 (gfc_array_r10 * const restrict re
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_10 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1372,8 +1392,7 @@ matmul_r10_avx512f (gfc_array_r10 * const restrict
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_10 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_10 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1397,6 +1416,17 @@ matmul_r10_avx512f (gfc_array_r10 * const restrict
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_10 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1911,8 +1941,7 @@ matmul_r10_vanilla (gfc_array_r10 * const restrict
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_10 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_10 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1936,6 +1965,17 @@ matmul_r10_vanilla (gfc_array_r10 * const restrict
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_10 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -2508,8 +2548,7 @@ matmul_r10 (gfc_array_r10 * const restrict retarra
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_10 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_10 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -2533,6 +2572,17 @@ matmul_r10 (gfc_array_r10 * const restrict retarra
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_10 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
Index: generated/matmul_r16.c
===================================================================
--- generated/matmul_r16.c (Revision 247566)
+++ generated/matmul_r16.c (Arbeitskopie)
@@ -286,8 +286,7 @@ matmul_r16_avx (gfc_array_r16 * const restrict ret
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_16 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_16 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -311,6 +310,17 @@ matmul_r16_avx (gfc_array_r16 * const restrict ret
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_16 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -829,8 +839,7 @@ matmul_r16_avx2 (gfc_array_r16 * const restrict re
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_16 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_16 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -854,6 +863,17 @@ matmul_r16_avx2 (gfc_array_r16 * const restrict re
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_16 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1372,8 +1392,7 @@ matmul_r16_avx512f (gfc_array_r16 * const restrict
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_16 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_16 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1397,6 +1416,17 @@ matmul_r16_avx512f (gfc_array_r16 * const restrict
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_16 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1911,8 +1941,7 @@ matmul_r16_vanilla (gfc_array_r16 * const restrict
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_16 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_16 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1936,6 +1965,17 @@ matmul_r16_vanilla (gfc_array_r16 * const restrict
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_16 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -2508,8 +2548,7 @@ matmul_r16 (gfc_array_r16 * const restrict retarra
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_16 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_16 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -2533,6 +2572,17 @@ matmul_r16 (gfc_array_r16 * const restrict retarra
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_16 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
Index: generated/matmul_r4.c
===================================================================
--- generated/matmul_r4.c (Revision 247566)
+++ generated/matmul_r4.c (Arbeitskopie)
@@ -286,8 +286,7 @@ matmul_r4_avx (gfc_array_r4 * const restrict retar
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_4 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_4 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -311,6 +310,17 @@ matmul_r4_avx (gfc_array_r4 * const restrict retar
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_4 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -829,8 +839,7 @@ matmul_r4_avx2 (gfc_array_r4 * const restrict reta
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_4 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_4 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -854,6 +863,17 @@ matmul_r4_avx2 (gfc_array_r4 * const restrict reta
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_4 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1372,8 +1392,7 @@ matmul_r4_avx512f (gfc_array_r4 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_4 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_4 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1397,6 +1416,17 @@ matmul_r4_avx512f (gfc_array_r4 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_4 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1911,8 +1941,7 @@ matmul_r4_vanilla (gfc_array_r4 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_4 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_4 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1936,6 +1965,17 @@ matmul_r4_vanilla (gfc_array_r4 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_4 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -2508,8 +2548,7 @@ matmul_r4 (gfc_array_r4 * const restrict retarray,
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_4 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_4 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -2533,6 +2572,17 @@ matmul_r4 (gfc_array_r4 * const restrict retarray,
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_4 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
Index: generated/matmul_r8.c
===================================================================
--- generated/matmul_r8.c (Revision 247566)
+++ generated/matmul_r8.c (Arbeitskopie)
@@ -286,8 +286,7 @@ matmul_r8_avx (gfc_array_r8 * const restrict retar
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_8 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_8 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -311,6 +310,17 @@ matmul_r8_avx (gfc_array_r8 * const restrict retar
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_8 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -829,8 +839,7 @@ matmul_r8_avx2 (gfc_array_r8 * const restrict reta
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_8 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_8 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -854,6 +863,17 @@ matmul_r8_avx2 (gfc_array_r8 * const restrict reta
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_8 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1372,8 +1392,7 @@ matmul_r8_avx512f (gfc_array_r8 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_8 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_8 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1397,6 +1416,17 @@ matmul_r8_avx512f (gfc_array_r8 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_8 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -1911,8 +1941,7 @@ matmul_r8_vanilla (gfc_array_r8 * const restrict r
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_8 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_8 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -1936,6 +1965,17 @@ matmul_r8_vanilla (gfc_array_r8 * const restrict r
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_8 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
@@ -2508,8 +2548,7 @@ matmul_r8 (gfc_array_r8 * const restrict retarray,
i1, i2, i3, i4, i5, i6;
/* Local variables */
- GFC_REAL_8 t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ GFC_REAL_8 f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -2533,6 +2572,17 @@ matmul_r8 (gfc_array_r8 * const restrict retarray,
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ GFC_REAL_8 t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
Index: m4/matmul_internal.m4
===================================================================
--- m4/matmul_internal.m4 (Revision 247566)
+++ m4/matmul_internal.m4 (Arbeitskopie)
@@ -202,8 +202,7 @@ sinclude(`matmul_asm_'rtype_code`.m4')dnl
i1, i2, i3, i4, i5, i6;
/* Local variables */
- 'rtype_name` t1[65536], /* was [256][256] */
- f11, f12, f21, f22, f31, f32, f41, f42,
+ 'rtype_name` f11, f12, f21, f22, f31, f32, f41, f42,
f13, f14, f23, f24, f33, f34, f43, f44;
index_type i, j, l, ii, jj, ll;
index_type isec, jsec, lsec, uisec, ujsec, ulsec;
@@ -227,6 +226,17 @@ sinclude(`matmul_asm_'rtype_code`.m4')dnl
if (m == 0 || n == 0 || k == 0)
return;
+ /* Adjust size of t1 to what is needed. */
+ index_type t1_dim;
+ t1_dim = (a_dim1-1) * 256 + b_dim1;
+ if (t1_dim > 65536)
+ t1_dim = 65536;
+
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wvla"
+ 'rtype_name` t1[t1_dim]; /* was [256][256] */
+#pragma GCC diagnostic pop
+
/* Empty c first. */
for (j=1; j<=n; j++)
for (i=1; i<=m; i++)
! { dg-do run }
! { dg-options "-finline-matmul-limit=0" }
! Stress-test the matmul blocking code with sizes close to or
! equal to powers ot two.
program main
implicit none
integer, dimension(*), parameter :: nn = &
& [2,3,4,5, 7,8,9, 15,16,17, 31,32,33, 63,64,65, &
127 ,228,129, 255,256,257];
integer, parameter :: s = size(nn)
real, dimension(:,:),allocatable :: a, b, c
integer :: i1, i2, i3
integer :: nx, ny, count
real :: sm
sm = 0.0
do i1=1, s
nx = nn(i1)
do i2=1,s
ny = nn(i2)
do i3=1,s
count = nn(i3)
allocate (a(nx,ny), b(ny,count), c(nx,count))
call random_number(a)
call random_number(b)
c = matmul(a,b)
sm = sm + sum(c)
deallocate(a,b,c)
end do
end do
end do
end program main
