public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc(refs/vendors/redhat/heads/gcc-8-branch)] PR libfortran/97063 - Wrong result for vector (step size is negative) * matrix
@ 2021-04-23 10:17 Jakub Jelinek
0 siblings, 0 replies; only message in thread
From: Jakub Jelinek @ 2021-04-23 10:17 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:a20e7bbbd8ba711665763c551182a34218da50d4
commit a20e7bbbd8ba711665763c551182a34218da50d4
Author: Harald Anlauf <anlauf@gmx.de>
Date: Sun Oct 18 20:15:26 2020 +0200
PR libfortran/97063 - Wrong result for vector (step size is negative) * matrix
The MATMUL intrinsic provided a wrong result for rank-1 times rank-2 array
when a negative stride was used for addressing the elements of the rank-1
array, because a check on strides was erroneously placed before the check
on the rank. Interchange order of checks.
libgfortran/ChangeLog:
* m4/matmul_internal.m4: Move check for rank-1 times rank-2 before
checks on strides for rank-2 times rank-2.
* generated/matmul_c10.c: Regenerated.
* generated/matmul_c16.c: Likewise.
* generated/matmul_c4.c: Likewise.
* generated/matmul_c8.c: Likewise.
* generated/matmul_i1.c: Likewise.
* generated/matmul_i16.c: Likewise.
* generated/matmul_i2.c: Likewise.
* generated/matmul_i4.c: Likewise.
* generated/matmul_i8.c: Likewise.
* generated/matmul_r10.c: Likewise.
* generated/matmul_r16.c: Likewise.
* generated/matmul_r4.c: Likewise.
* generated/matmul_r8.c: Likewise.
* generated/matmulavx128_c10.c: Likewise.
* generated/matmulavx128_c16.c: Likewise.
* generated/matmulavx128_c4.c: Likewise.
* generated/matmulavx128_c8.c: Likewise.
* generated/matmulavx128_i1.c: Likewise.
* generated/matmulavx128_i16.c: Likewise.
* generated/matmulavx128_i2.c: Likewise.
* generated/matmulavx128_i4.c: Likewise.
* generated/matmulavx128_i8.c: Likewise.
* generated/matmulavx128_r10.c: Likewise.
* generated/matmulavx128_r16.c: Likewise.
* generated/matmulavx128_r4.c: Likewise.
* generated/matmulavx128_r8.c: Likewise.
gcc/testsuite/ChangeLog:
* gfortran.dg/matmul_20.f90: New test.
(cherry picked from commit cd6cd6aed195b4ec7d652e8b41d60b60e174304e)
Diff:
---
gcc/testsuite/gfortran.dg/matmul_20.f90 | 47 +++++++++++
libgfortran/generated/matmul_c10.c | 140 +++++++++++++++----------------
libgfortran/generated/matmul_c16.c | 140 +++++++++++++++----------------
libgfortran/generated/matmul_c4.c | 140 +++++++++++++++----------------
libgfortran/generated/matmul_c8.c | 140 +++++++++++++++----------------
libgfortran/generated/matmul_i1.c | 140 +++++++++++++++----------------
libgfortran/generated/matmul_i16.c | 140 +++++++++++++++----------------
libgfortran/generated/matmul_i2.c | 140 +++++++++++++++----------------
libgfortran/generated/matmul_i4.c | 140 +++++++++++++++----------------
libgfortran/generated/matmul_i8.c | 140 +++++++++++++++----------------
libgfortran/generated/matmul_r10.c | 140 +++++++++++++++----------------
libgfortran/generated/matmul_r16.c | 140 +++++++++++++++----------------
libgfortran/generated/matmul_r4.c | 140 +++++++++++++++----------------
libgfortran/generated/matmul_r8.c | 140 +++++++++++++++----------------
libgfortran/generated/matmulavx128_c10.c | 56 ++++++-------
libgfortran/generated/matmulavx128_c16.c | 56 ++++++-------
libgfortran/generated/matmulavx128_c4.c | 56 ++++++-------
libgfortran/generated/matmulavx128_c8.c | 56 ++++++-------
libgfortran/generated/matmulavx128_i1.c | 56 ++++++-------
libgfortran/generated/matmulavx128_i16.c | 56 ++++++-------
libgfortran/generated/matmulavx128_i2.c | 56 ++++++-------
libgfortran/generated/matmulavx128_i4.c | 56 ++++++-------
libgfortran/generated/matmulavx128_i8.c | 56 ++++++-------
libgfortran/generated/matmulavx128_r10.c | 56 ++++++-------
libgfortran/generated/matmulavx128_r16.c | 56 ++++++-------
libgfortran/generated/matmulavx128_r4.c | 56 ++++++-------
libgfortran/generated/matmulavx128_r8.c | 56 ++++++-------
libgfortran/m4/matmul_internal.m4 | 28 +++----
28 files changed, 1335 insertions(+), 1288 deletions(-)
diff --git a/gcc/testsuite/gfortran.dg/matmul_20.f90 b/gcc/testsuite/gfortran.dg/matmul_20.f90
new file mode 100644
index 00000000000..7a211a4974d
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/matmul_20.f90
@@ -0,0 +1,47 @@
+! { dg-do run }
+! PR97063 - Wrong result for vector (step size is negative) * matrix
+
+program p
+ implicit none
+ integer, parameter :: m = 3, k = 2*m, l = k-1, n = 4
+ integer :: i, j, m1, m2, ms
+ integer :: ai(k), bi(k,n), ci(n), ci_ref(n), c1, c2
+ real :: ar(k), br(k,n), cr(n), cr_ref(n)
+
+ ai(:) = [(i,i=0,k-1)]
+ bi(:,:) = reshape ([(((5*i+j),i=0,k-1),j=0,n-1)],[k,n])
+
+ ! Parameters of subscript triplet
+ m1 = 1; m2 = l; ms = 2
+
+ ! Reference values for cross-checks: integer variant
+ c1 = dot_product (ai(m1:m2: ms), bi(m1:m2: ms,1))
+ c2 = dot_product (ai(m1:m2: ms), bi(m1:m2: ms,2))
+ ci_ref = matmul (ai(m1:m2: ms), bi(m1:m2: ms,:))
+ ci = matmul (ai(m2:m1:-ms), bi(m2:m1:-ms,:))
+
+ if (ci_ref(1) /= c1 .or. ci_ref(2) /= c2) stop 1
+ if (any (ci /= ci_ref)) stop 2
+
+ ! Real variant
+ ar = real (ai)
+ br = real (bi)
+ cr_ref = matmul (ar(m1:m2: ms), br(m1:m2: ms,:))
+ cr = matmul (ar(m2:m1:-ms), br(m2:m1:-ms,:))
+
+ if (any (cr_ref /= real (ci_ref))) stop 3
+ if (any (cr /= cr_ref )) stop 4
+
+ ! Mixed variants
+ cr_ref = matmul (ar(m1:m2: ms), bi(m1:m2: ms,:))
+ cr = matmul (ar(m2:m1:-ms), bi(m2:m1:-ms,:))
+
+ if (any (cr_ref /= real (ci_ref))) stop 5
+ if (any (cr /= cr_ref )) stop 6
+
+ cr_ref = matmul (ai(m1:m2: ms), br(m1:m2: ms,:))
+ cr = matmul (ai(m2:m1:-ms), br(m2:m1:-ms,:))
+
+ if (any (cr_ref /= real (ci_ref))) stop 7
+ if (any (cr /= cr_ref )) stop 8
+end program
diff --git a/libgfortran/generated/matmul_c10.c b/libgfortran/generated/matmul_c10.c
index 462d71e23f5..3db1d37a9c7 100644
--- a/libgfortran/generated/matmul_c10.c
+++ b/libgfortran/generated/matmul_c10.c
@@ -579,20 +579,6 @@ matmul_c10_avx (gfc_array_c10 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_10)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_10 *restrict bbase_y;
@@ -607,6 +593,20 @@ matmul_c10_avx (gfc_array_c10 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_10)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_10 *restrict abase_x;
@@ -1136,20 +1136,6 @@ matmul_c10_avx2 (gfc_array_c10 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_10)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_10 *restrict bbase_y;
@@ -1164,6 +1150,20 @@ matmul_c10_avx2 (gfc_array_c10 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_10)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_10 *restrict abase_x;
@@ -1693,20 +1693,6 @@ matmul_c10_avx512f (gfc_array_c10 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_10)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_10 *restrict bbase_y;
@@ -1721,6 +1707,20 @@ matmul_c10_avx512f (gfc_array_c10 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_10)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_10 *restrict abase_x;
@@ -2264,20 +2264,6 @@ matmul_c10_vanilla (gfc_array_c10 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_10)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_10 *restrict bbase_y;
@@ -2292,6 +2278,20 @@ matmul_c10_vanilla (gfc_array_c10 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_10)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_10 *restrict abase_x;
@@ -2895,20 +2895,6 @@ matmul_c10 (gfc_array_c10 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_10)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_10 *restrict bbase_y;
@@ -2923,6 +2909,20 @@ matmul_c10 (gfc_array_c10 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_10)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_10 *restrict abase_x;
diff --git a/libgfortran/generated/matmul_c16.c b/libgfortran/generated/matmul_c16.c
index 2062739ce1f..16637b86bd3 100644
--- a/libgfortran/generated/matmul_c16.c
+++ b/libgfortran/generated/matmul_c16.c
@@ -579,20 +579,6 @@ matmul_c16_avx (gfc_array_c16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_16 *restrict bbase_y;
@@ -607,6 +593,20 @@ matmul_c16_avx (gfc_array_c16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_16 *restrict abase_x;
@@ -1136,20 +1136,6 @@ matmul_c16_avx2 (gfc_array_c16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_16 *restrict bbase_y;
@@ -1164,6 +1150,20 @@ matmul_c16_avx2 (gfc_array_c16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_16 *restrict abase_x;
@@ -1693,20 +1693,6 @@ matmul_c16_avx512f (gfc_array_c16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_16 *restrict bbase_y;
@@ -1721,6 +1707,20 @@ matmul_c16_avx512f (gfc_array_c16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_16 *restrict abase_x;
@@ -2264,20 +2264,6 @@ matmul_c16_vanilla (gfc_array_c16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_16 *restrict bbase_y;
@@ -2292,6 +2278,20 @@ matmul_c16_vanilla (gfc_array_c16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_16 *restrict abase_x;
@@ -2895,20 +2895,6 @@ matmul_c16 (gfc_array_c16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_16 *restrict bbase_y;
@@ -2923,6 +2909,20 @@ matmul_c16 (gfc_array_c16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_16 *restrict abase_x;
diff --git a/libgfortran/generated/matmul_c4.c b/libgfortran/generated/matmul_c4.c
index 91d193dca37..40a8c9ac87c 100644
--- a/libgfortran/generated/matmul_c4.c
+++ b/libgfortran/generated/matmul_c4.c
@@ -579,20 +579,6 @@ matmul_c4_avx (gfc_array_c4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_4 *restrict bbase_y;
@@ -607,6 +593,20 @@ matmul_c4_avx (gfc_array_c4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_4 *restrict abase_x;
@@ -1136,20 +1136,6 @@ matmul_c4_avx2 (gfc_array_c4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_4 *restrict bbase_y;
@@ -1164,6 +1150,20 @@ matmul_c4_avx2 (gfc_array_c4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_4 *restrict abase_x;
@@ -1693,20 +1693,6 @@ matmul_c4_avx512f (gfc_array_c4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_4 *restrict bbase_y;
@@ -1721,6 +1707,20 @@ matmul_c4_avx512f (gfc_array_c4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_4 *restrict abase_x;
@@ -2264,20 +2264,6 @@ matmul_c4_vanilla (gfc_array_c4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_4 *restrict bbase_y;
@@ -2292,6 +2278,20 @@ matmul_c4_vanilla (gfc_array_c4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_4 *restrict abase_x;
@@ -2895,20 +2895,6 @@ matmul_c4 (gfc_array_c4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_4 *restrict bbase_y;
@@ -2923,6 +2909,20 @@ matmul_c4 (gfc_array_c4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_4 *restrict abase_x;
diff --git a/libgfortran/generated/matmul_c8.c b/libgfortran/generated/matmul_c8.c
index 425af85d1bb..75c0479f8ed 100644
--- a/libgfortran/generated/matmul_c8.c
+++ b/libgfortran/generated/matmul_c8.c
@@ -579,20 +579,6 @@ matmul_c8_avx (gfc_array_c8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_8 *restrict bbase_y;
@@ -607,6 +593,20 @@ matmul_c8_avx (gfc_array_c8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_8 *restrict abase_x;
@@ -1136,20 +1136,6 @@ matmul_c8_avx2 (gfc_array_c8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_8 *restrict bbase_y;
@@ -1164,6 +1150,20 @@ matmul_c8_avx2 (gfc_array_c8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_8 *restrict abase_x;
@@ -1693,20 +1693,6 @@ matmul_c8_avx512f (gfc_array_c8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_8 *restrict bbase_y;
@@ -1721,6 +1707,20 @@ matmul_c8_avx512f (gfc_array_c8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_8 *restrict abase_x;
@@ -2264,20 +2264,6 @@ matmul_c8_vanilla (gfc_array_c8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_8 *restrict bbase_y;
@@ -2292,6 +2278,20 @@ matmul_c8_vanilla (gfc_array_c8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_8 *restrict abase_x;
@@ -2895,20 +2895,6 @@ matmul_c8 (gfc_array_c8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_8 *restrict bbase_y;
@@ -2923,6 +2909,20 @@ matmul_c8 (gfc_array_c8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_8 *restrict abase_x;
diff --git a/libgfortran/generated/matmul_i1.c b/libgfortran/generated/matmul_i1.c
index 0c9335d4322..2550c564041 100644
--- a/libgfortran/generated/matmul_i1.c
+++ b/libgfortran/generated/matmul_i1.c
@@ -579,20 +579,6 @@ matmul_i1_avx (gfc_array_i1 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_1)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_1 *restrict bbase_y;
@@ -607,6 +593,20 @@ matmul_i1_avx (gfc_array_i1 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_1)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_1 *restrict abase_x;
@@ -1136,20 +1136,6 @@ matmul_i1_avx2 (gfc_array_i1 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_1)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_1 *restrict bbase_y;
@@ -1164,6 +1150,20 @@ matmul_i1_avx2 (gfc_array_i1 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_1)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_1 *restrict abase_x;
@@ -1693,20 +1693,6 @@ matmul_i1_avx512f (gfc_array_i1 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_1)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_1 *restrict bbase_y;
@@ -1721,6 +1707,20 @@ matmul_i1_avx512f (gfc_array_i1 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_1)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_1 *restrict abase_x;
@@ -2264,20 +2264,6 @@ matmul_i1_vanilla (gfc_array_i1 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_1)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_1 *restrict bbase_y;
@@ -2292,6 +2278,20 @@ matmul_i1_vanilla (gfc_array_i1 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_1)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_1 *restrict abase_x;
@@ -2895,20 +2895,6 @@ matmul_i1 (gfc_array_i1 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_1)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_1 *restrict bbase_y;
@@ -2923,6 +2909,20 @@ matmul_i1 (gfc_array_i1 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_1)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_1 *restrict abase_x;
diff --git a/libgfortran/generated/matmul_i16.c b/libgfortran/generated/matmul_i16.c
index b9334b3278f..907aa28c381 100644
--- a/libgfortran/generated/matmul_i16.c
+++ b/libgfortran/generated/matmul_i16.c
@@ -579,20 +579,6 @@ matmul_i16_avx (gfc_array_i16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_16 *restrict bbase_y;
@@ -607,6 +593,20 @@ matmul_i16_avx (gfc_array_i16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_16 *restrict abase_x;
@@ -1136,20 +1136,6 @@ matmul_i16_avx2 (gfc_array_i16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_16 *restrict bbase_y;
@@ -1164,6 +1150,20 @@ matmul_i16_avx2 (gfc_array_i16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_16 *restrict abase_x;
@@ -1693,20 +1693,6 @@ matmul_i16_avx512f (gfc_array_i16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_16 *restrict bbase_y;
@@ -1721,6 +1707,20 @@ matmul_i16_avx512f (gfc_array_i16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_16 *restrict abase_x;
@@ -2264,20 +2264,6 @@ matmul_i16_vanilla (gfc_array_i16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_16 *restrict bbase_y;
@@ -2292,6 +2278,20 @@ matmul_i16_vanilla (gfc_array_i16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_16 *restrict abase_x;
@@ -2895,20 +2895,6 @@ matmul_i16 (gfc_array_i16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_16 *restrict bbase_y;
@@ -2923,6 +2909,20 @@ matmul_i16 (gfc_array_i16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_16 *restrict abase_x;
diff --git a/libgfortran/generated/matmul_i2.c b/libgfortran/generated/matmul_i2.c
index e4246e948df..f0a33ce60cf 100644
--- a/libgfortran/generated/matmul_i2.c
+++ b/libgfortran/generated/matmul_i2.c
@@ -579,20 +579,6 @@ matmul_i2_avx (gfc_array_i2 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_2)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_2 *restrict bbase_y;
@@ -607,6 +593,20 @@ matmul_i2_avx (gfc_array_i2 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_2)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_2 *restrict abase_x;
@@ -1136,20 +1136,6 @@ matmul_i2_avx2 (gfc_array_i2 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_2)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_2 *restrict bbase_y;
@@ -1164,6 +1150,20 @@ matmul_i2_avx2 (gfc_array_i2 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_2)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_2 *restrict abase_x;
@@ -1693,20 +1693,6 @@ matmul_i2_avx512f (gfc_array_i2 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_2)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_2 *restrict bbase_y;
@@ -1721,6 +1707,20 @@ matmul_i2_avx512f (gfc_array_i2 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_2)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_2 *restrict abase_x;
@@ -2264,20 +2264,6 @@ matmul_i2_vanilla (gfc_array_i2 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_2)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_2 *restrict bbase_y;
@@ -2292,6 +2278,20 @@ matmul_i2_vanilla (gfc_array_i2 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_2)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_2 *restrict abase_x;
@@ -2895,20 +2895,6 @@ matmul_i2 (gfc_array_i2 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_2)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_2 *restrict bbase_y;
@@ -2923,6 +2909,20 @@ matmul_i2 (gfc_array_i2 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_2)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_2 *restrict abase_x;
diff --git a/libgfortran/generated/matmul_i4.c b/libgfortran/generated/matmul_i4.c
index 78cf27c4fcd..b69d112db76 100644
--- a/libgfortran/generated/matmul_i4.c
+++ b/libgfortran/generated/matmul_i4.c
@@ -579,20 +579,6 @@ matmul_i4_avx (gfc_array_i4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_4 *restrict bbase_y;
@@ -607,6 +593,20 @@ matmul_i4_avx (gfc_array_i4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_4 *restrict abase_x;
@@ -1136,20 +1136,6 @@ matmul_i4_avx2 (gfc_array_i4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_4 *restrict bbase_y;
@@ -1164,6 +1150,20 @@ matmul_i4_avx2 (gfc_array_i4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_4 *restrict abase_x;
@@ -1693,20 +1693,6 @@ matmul_i4_avx512f (gfc_array_i4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_4 *restrict bbase_y;
@@ -1721,6 +1707,20 @@ matmul_i4_avx512f (gfc_array_i4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_4 *restrict abase_x;
@@ -2264,20 +2264,6 @@ matmul_i4_vanilla (gfc_array_i4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_4 *restrict bbase_y;
@@ -2292,6 +2278,20 @@ matmul_i4_vanilla (gfc_array_i4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_4 *restrict abase_x;
@@ -2895,20 +2895,6 @@ matmul_i4 (gfc_array_i4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_4 *restrict bbase_y;
@@ -2923,6 +2909,20 @@ matmul_i4 (gfc_array_i4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_4 *restrict abase_x;
diff --git a/libgfortran/generated/matmul_i8.c b/libgfortran/generated/matmul_i8.c
index cf8c401400d..3801561e02d 100644
--- a/libgfortran/generated/matmul_i8.c
+++ b/libgfortran/generated/matmul_i8.c
@@ -579,20 +579,6 @@ matmul_i8_avx (gfc_array_i8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_8 *restrict bbase_y;
@@ -607,6 +593,20 @@ matmul_i8_avx (gfc_array_i8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_8 *restrict abase_x;
@@ -1136,20 +1136,6 @@ matmul_i8_avx2 (gfc_array_i8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_8 *restrict bbase_y;
@@ -1164,6 +1150,20 @@ matmul_i8_avx2 (gfc_array_i8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_8 *restrict abase_x;
@@ -1693,20 +1693,6 @@ matmul_i8_avx512f (gfc_array_i8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_8 *restrict bbase_y;
@@ -1721,6 +1707,20 @@ matmul_i8_avx512f (gfc_array_i8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_8 *restrict abase_x;
@@ -2264,20 +2264,6 @@ matmul_i8_vanilla (gfc_array_i8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_8 *restrict bbase_y;
@@ -2292,6 +2278,20 @@ matmul_i8_vanilla (gfc_array_i8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_8 *restrict abase_x;
@@ -2895,20 +2895,6 @@ matmul_i8 (gfc_array_i8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_8 *restrict bbase_y;
@@ -2923,6 +2909,20 @@ matmul_i8 (gfc_array_i8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_8 *restrict abase_x;
diff --git a/libgfortran/generated/matmul_r10.c b/libgfortran/generated/matmul_r10.c
index e4309c80a6b..60740edde62 100644
--- a/libgfortran/generated/matmul_r10.c
+++ b/libgfortran/generated/matmul_r10.c
@@ -579,20 +579,6 @@ matmul_r10_avx (gfc_array_r10 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_10)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_10 *restrict bbase_y;
@@ -607,6 +593,20 @@ matmul_r10_avx (gfc_array_r10 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_10)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_10 *restrict abase_x;
@@ -1136,20 +1136,6 @@ matmul_r10_avx2 (gfc_array_r10 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_10)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_10 *restrict bbase_y;
@@ -1164,6 +1150,20 @@ matmul_r10_avx2 (gfc_array_r10 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_10)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_10 *restrict abase_x;
@@ -1693,20 +1693,6 @@ matmul_r10_avx512f (gfc_array_r10 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_10)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_10 *restrict bbase_y;
@@ -1721,6 +1707,20 @@ matmul_r10_avx512f (gfc_array_r10 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_10)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_10 *restrict abase_x;
@@ -2264,20 +2264,6 @@ matmul_r10_vanilla (gfc_array_r10 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_10)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_10 *restrict bbase_y;
@@ -2292,6 +2278,20 @@ matmul_r10_vanilla (gfc_array_r10 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_10)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_10 *restrict abase_x;
@@ -2895,20 +2895,6 @@ matmul_r10 (gfc_array_r10 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_10)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_10 *restrict bbase_y;
@@ -2923,6 +2909,20 @@ matmul_r10 (gfc_array_r10 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_10)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_10 *restrict abase_x;
diff --git a/libgfortran/generated/matmul_r16.c b/libgfortran/generated/matmul_r16.c
index 1ab554660d7..df199fb31cf 100644
--- a/libgfortran/generated/matmul_r16.c
+++ b/libgfortran/generated/matmul_r16.c
@@ -579,20 +579,6 @@ matmul_r16_avx (gfc_array_r16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_16 *restrict bbase_y;
@@ -607,6 +593,20 @@ matmul_r16_avx (gfc_array_r16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_16 *restrict abase_x;
@@ -1136,20 +1136,6 @@ matmul_r16_avx2 (gfc_array_r16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_16 *restrict bbase_y;
@@ -1164,6 +1150,20 @@ matmul_r16_avx2 (gfc_array_r16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_16 *restrict abase_x;
@@ -1693,20 +1693,6 @@ matmul_r16_avx512f (gfc_array_r16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_16 *restrict bbase_y;
@@ -1721,6 +1707,20 @@ matmul_r16_avx512f (gfc_array_r16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_16 *restrict abase_x;
@@ -2264,20 +2264,6 @@ matmul_r16_vanilla (gfc_array_r16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_16 *restrict bbase_y;
@@ -2292,6 +2278,20 @@ matmul_r16_vanilla (gfc_array_r16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_16 *restrict abase_x;
@@ -2895,20 +2895,6 @@ matmul_r16 (gfc_array_r16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_16 *restrict bbase_y;
@@ -2923,6 +2909,20 @@ matmul_r16 (gfc_array_r16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_16 *restrict abase_x;
diff --git a/libgfortran/generated/matmul_r4.c b/libgfortran/generated/matmul_r4.c
index 97dba9825b1..6aa156f525d 100644
--- a/libgfortran/generated/matmul_r4.c
+++ b/libgfortran/generated/matmul_r4.c
@@ -579,20 +579,6 @@ matmul_r4_avx (gfc_array_r4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_4 *restrict bbase_y;
@@ -607,6 +593,20 @@ matmul_r4_avx (gfc_array_r4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_4 *restrict abase_x;
@@ -1136,20 +1136,6 @@ matmul_r4_avx2 (gfc_array_r4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_4 *restrict bbase_y;
@@ -1164,6 +1150,20 @@ matmul_r4_avx2 (gfc_array_r4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_4 *restrict abase_x;
@@ -1693,20 +1693,6 @@ matmul_r4_avx512f (gfc_array_r4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_4 *restrict bbase_y;
@@ -1721,6 +1707,20 @@ matmul_r4_avx512f (gfc_array_r4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_4 *restrict abase_x;
@@ -2264,20 +2264,6 @@ matmul_r4_vanilla (gfc_array_r4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_4 *restrict bbase_y;
@@ -2292,6 +2278,20 @@ matmul_r4_vanilla (gfc_array_r4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_4 *restrict abase_x;
@@ -2895,20 +2895,6 @@ matmul_r4 (gfc_array_r4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_4 *restrict bbase_y;
@@ -2923,6 +2909,20 @@ matmul_r4 (gfc_array_r4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_4 *restrict abase_x;
diff --git a/libgfortran/generated/matmul_r8.c b/libgfortran/generated/matmul_r8.c
index 5e4c9500260..03928c4cee9 100644
--- a/libgfortran/generated/matmul_r8.c
+++ b/libgfortran/generated/matmul_r8.c
@@ -579,20 +579,6 @@ matmul_r8_avx (gfc_array_r8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_8 *restrict bbase_y;
@@ -607,6 +593,20 @@ matmul_r8_avx (gfc_array_r8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_8 *restrict abase_x;
@@ -1136,20 +1136,6 @@ matmul_r8_avx2 (gfc_array_r8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_8 *restrict bbase_y;
@@ -1164,6 +1150,20 @@ matmul_r8_avx2 (gfc_array_r8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_8 *restrict abase_x;
@@ -1693,20 +1693,6 @@ matmul_r8_avx512f (gfc_array_r8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_8 *restrict bbase_y;
@@ -1721,6 +1707,20 @@ matmul_r8_avx512f (gfc_array_r8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_8 *restrict abase_x;
@@ -2264,20 +2264,6 @@ matmul_r8_vanilla (gfc_array_r8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_8 *restrict bbase_y;
@@ -2292,6 +2278,20 @@ matmul_r8_vanilla (gfc_array_r8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_8 *restrict abase_x;
@@ -2895,20 +2895,6 @@ matmul_r8 (gfc_array_r8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_8 *restrict bbase_y;
@@ -2923,6 +2909,20 @@ matmul_r8 (gfc_array_r8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_8 *restrict abase_x;
diff --git a/libgfortran/generated/matmulavx128_c10.c b/libgfortran/generated/matmulavx128_c10.c
index 5cb0f6ad6f3..9c11629f234 100644
--- a/libgfortran/generated/matmulavx128_c10.c
+++ b/libgfortran/generated/matmulavx128_c10.c
@@ -544,20 +544,6 @@ matmul_c10_avx128_fma3 (gfc_array_c10 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_10)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_10 *restrict bbase_y;
@@ -572,6 +558,20 @@ matmul_c10_avx128_fma3 (gfc_array_c10 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_10)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_10 *restrict abase_x;
@@ -1102,20 +1102,6 @@ matmul_c10_avx128_fma4 (gfc_array_c10 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_10)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_10 *restrict bbase_y;
@@ -1130,6 +1116,20 @@ matmul_c10_avx128_fma4 (gfc_array_c10 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_10)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_10 *restrict abase_x;
diff --git a/libgfortran/generated/matmulavx128_c16.c b/libgfortran/generated/matmulavx128_c16.c
index 66272fefaf9..5a3f3f52bf1 100644
--- a/libgfortran/generated/matmulavx128_c16.c
+++ b/libgfortran/generated/matmulavx128_c16.c
@@ -544,20 +544,6 @@ matmul_c16_avx128_fma3 (gfc_array_c16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_16 *restrict bbase_y;
@@ -572,6 +558,20 @@ matmul_c16_avx128_fma3 (gfc_array_c16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_16 *restrict abase_x;
@@ -1102,20 +1102,6 @@ matmul_c16_avx128_fma4 (gfc_array_c16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_16 *restrict bbase_y;
@@ -1130,6 +1116,20 @@ matmul_c16_avx128_fma4 (gfc_array_c16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_16 *restrict abase_x;
diff --git a/libgfortran/generated/matmulavx128_c4.c b/libgfortran/generated/matmulavx128_c4.c
index f6e06e2e88f..c12bae3fafe 100644
--- a/libgfortran/generated/matmulavx128_c4.c
+++ b/libgfortran/generated/matmulavx128_c4.c
@@ -544,20 +544,6 @@ matmul_c4_avx128_fma3 (gfc_array_c4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_4 *restrict bbase_y;
@@ -572,6 +558,20 @@ matmul_c4_avx128_fma3 (gfc_array_c4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_4 *restrict abase_x;
@@ -1102,20 +1102,6 @@ matmul_c4_avx128_fma4 (gfc_array_c4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_4 *restrict bbase_y;
@@ -1130,6 +1116,20 @@ matmul_c4_avx128_fma4 (gfc_array_c4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_4 *restrict abase_x;
diff --git a/libgfortran/generated/matmulavx128_c8.c b/libgfortran/generated/matmulavx128_c8.c
index accc69c4d1a..b011ca12b2a 100644
--- a/libgfortran/generated/matmulavx128_c8.c
+++ b/libgfortran/generated/matmulavx128_c8.c
@@ -544,20 +544,6 @@ matmul_c8_avx128_fma3 (gfc_array_c8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_8 *restrict bbase_y;
@@ -572,6 +558,20 @@ matmul_c8_avx128_fma3 (gfc_array_c8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_8 *restrict abase_x;
@@ -1102,20 +1102,6 @@ matmul_c8_avx128_fma4 (gfc_array_c8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_COMPLEX_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_COMPLEX_8 *restrict bbase_y;
@@ -1130,6 +1116,20 @@ matmul_c8_avx128_fma4 (gfc_array_c8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_COMPLEX_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_COMPLEX_8 *restrict abase_x;
diff --git a/libgfortran/generated/matmulavx128_i1.c b/libgfortran/generated/matmulavx128_i1.c
index 48b15c8074a..025bf06b371 100644
--- a/libgfortran/generated/matmulavx128_i1.c
+++ b/libgfortran/generated/matmulavx128_i1.c
@@ -544,20 +544,6 @@ matmul_i1_avx128_fma3 (gfc_array_i1 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_1)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_1 *restrict bbase_y;
@@ -572,6 +558,20 @@ matmul_i1_avx128_fma3 (gfc_array_i1 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_1)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_1 *restrict abase_x;
@@ -1102,20 +1102,6 @@ matmul_i1_avx128_fma4 (gfc_array_i1 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_1)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_1 *restrict bbase_y;
@@ -1130,6 +1116,20 @@ matmul_i1_avx128_fma4 (gfc_array_i1 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_1)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_1 *restrict abase_x;
diff --git a/libgfortran/generated/matmulavx128_i16.c b/libgfortran/generated/matmulavx128_i16.c
index 319321eca49..9452245fab1 100644
--- a/libgfortran/generated/matmulavx128_i16.c
+++ b/libgfortran/generated/matmulavx128_i16.c
@@ -544,20 +544,6 @@ matmul_i16_avx128_fma3 (gfc_array_i16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_16 *restrict bbase_y;
@@ -572,6 +558,20 @@ matmul_i16_avx128_fma3 (gfc_array_i16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_16 *restrict abase_x;
@@ -1102,20 +1102,6 @@ matmul_i16_avx128_fma4 (gfc_array_i16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_16 *restrict bbase_y;
@@ -1130,6 +1116,20 @@ matmul_i16_avx128_fma4 (gfc_array_i16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_16 *restrict abase_x;
diff --git a/libgfortran/generated/matmulavx128_i2.c b/libgfortran/generated/matmulavx128_i2.c
index 4d8945b10a2..60929bc654a 100644
--- a/libgfortran/generated/matmulavx128_i2.c
+++ b/libgfortran/generated/matmulavx128_i2.c
@@ -544,20 +544,6 @@ matmul_i2_avx128_fma3 (gfc_array_i2 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_2)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_2 *restrict bbase_y;
@@ -572,6 +558,20 @@ matmul_i2_avx128_fma3 (gfc_array_i2 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_2)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_2 *restrict abase_x;
@@ -1102,20 +1102,6 @@ matmul_i2_avx128_fma4 (gfc_array_i2 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_2)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_2 *restrict bbase_y;
@@ -1130,6 +1116,20 @@ matmul_i2_avx128_fma4 (gfc_array_i2 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_2)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_2 *restrict abase_x;
diff --git a/libgfortran/generated/matmulavx128_i4.c b/libgfortran/generated/matmulavx128_i4.c
index acaa00a30bb..7e6dfcff2bc 100644
--- a/libgfortran/generated/matmulavx128_i4.c
+++ b/libgfortran/generated/matmulavx128_i4.c
@@ -544,20 +544,6 @@ matmul_i4_avx128_fma3 (gfc_array_i4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_4 *restrict bbase_y;
@@ -572,6 +558,20 @@ matmul_i4_avx128_fma3 (gfc_array_i4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_4 *restrict abase_x;
@@ -1102,20 +1102,6 @@ matmul_i4_avx128_fma4 (gfc_array_i4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_4 *restrict bbase_y;
@@ -1130,6 +1116,20 @@ matmul_i4_avx128_fma4 (gfc_array_i4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_4 *restrict abase_x;
diff --git a/libgfortran/generated/matmulavx128_i8.c b/libgfortran/generated/matmulavx128_i8.c
index 56e85167a3f..b6071e898b9 100644
--- a/libgfortran/generated/matmulavx128_i8.c
+++ b/libgfortran/generated/matmulavx128_i8.c
@@ -544,20 +544,6 @@ matmul_i8_avx128_fma3 (gfc_array_i8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_8 *restrict bbase_y;
@@ -572,6 +558,20 @@ matmul_i8_avx128_fma3 (gfc_array_i8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_8 *restrict abase_x;
@@ -1102,20 +1102,6 @@ matmul_i8_avx128_fma4 (gfc_array_i8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_INTEGER_8 *restrict bbase_y;
@@ -1130,6 +1116,20 @@ matmul_i8_avx128_fma4 (gfc_array_i8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_INTEGER_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_INTEGER_8 *restrict abase_x;
diff --git a/libgfortran/generated/matmulavx128_r10.c b/libgfortran/generated/matmulavx128_r10.c
index 880c9d921b2..3fabc362fb4 100644
--- a/libgfortran/generated/matmulavx128_r10.c
+++ b/libgfortran/generated/matmulavx128_r10.c
@@ -544,20 +544,6 @@ matmul_r10_avx128_fma3 (gfc_array_r10 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_10)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_10 *restrict bbase_y;
@@ -572,6 +558,20 @@ matmul_r10_avx128_fma3 (gfc_array_r10 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_10)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_10 *restrict abase_x;
@@ -1102,20 +1102,6 @@ matmul_r10_avx128_fma4 (gfc_array_r10 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_10)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_10 *restrict bbase_y;
@@ -1130,6 +1116,20 @@ matmul_r10_avx128_fma4 (gfc_array_r10 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_10)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_10 *restrict abase_x;
diff --git a/libgfortran/generated/matmulavx128_r16.c b/libgfortran/generated/matmulavx128_r16.c
index 328e251a3a1..b53ce280610 100644
--- a/libgfortran/generated/matmulavx128_r16.c
+++ b/libgfortran/generated/matmulavx128_r16.c
@@ -544,20 +544,6 @@ matmul_r16_avx128_fma3 (gfc_array_r16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_16 *restrict bbase_y;
@@ -572,6 +558,20 @@ matmul_r16_avx128_fma3 (gfc_array_r16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_16 *restrict abase_x;
@@ -1102,20 +1102,6 @@ matmul_r16_avx128_fma4 (gfc_array_r16 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_16)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_16 *restrict bbase_y;
@@ -1130,6 +1116,20 @@ matmul_r16_avx128_fma4 (gfc_array_r16 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_16)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_16 *restrict abase_x;
diff --git a/libgfortran/generated/matmulavx128_r4.c b/libgfortran/generated/matmulavx128_r4.c
index 013a1804a11..ac1e1a35e5a 100644
--- a/libgfortran/generated/matmulavx128_r4.c
+++ b/libgfortran/generated/matmulavx128_r4.c
@@ -544,20 +544,6 @@ matmul_r4_avx128_fma3 (gfc_array_r4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_4 *restrict bbase_y;
@@ -572,6 +558,20 @@ matmul_r4_avx128_fma3 (gfc_array_r4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_4 *restrict abase_x;
@@ -1102,20 +1102,6 @@ matmul_r4_avx128_fma4 (gfc_array_r4 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_4)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_4 *restrict bbase_y;
@@ -1130,6 +1116,20 @@ matmul_r4_avx128_fma4 (gfc_array_r4 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_4)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_4 *restrict abase_x;
diff --git a/libgfortran/generated/matmulavx128_r8.c b/libgfortran/generated/matmulavx128_r8.c
index 4da59f9e69c..416604af13b 100644
--- a/libgfortran/generated/matmulavx128_r8.c
+++ b/libgfortran/generated/matmulavx128_r8.c
@@ -544,20 +544,6 @@ matmul_r8_avx128_fma3 (gfc_array_r8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_8 *restrict bbase_y;
@@ -572,6 +558,20 @@ matmul_r8_avx128_fma3 (gfc_array_r8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_8 *restrict abase_x;
@@ -1102,20 +1102,6 @@ matmul_r8_avx128_fma4 (gfc_array_r8 * const restrict retarray,
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = (GFC_REAL_8)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const GFC_REAL_8 *restrict bbase_y;
@@ -1130,6 +1116,20 @@ matmul_r8_avx128_fma4 (gfc_array_r8 * const restrict retarray,
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = (GFC_REAL_8)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const GFC_REAL_8 *restrict abase_x;
diff --git a/libgfortran/m4/matmul_internal.m4 b/libgfortran/m4/matmul_internal.m4
index 2020e8a50df..4c8237a66e6 100644
--- a/libgfortran/m4/matmul_internal.m4
+++ b/libgfortran/m4/matmul_internal.m4
@@ -495,20 +495,6 @@ sinclude(`matmul_asm_'rtype_code`.m4')dnl
}
}
}
- else if (axstride < aystride)
- {
- for (y = 0; y < ycount; y++)
- for (x = 0; x < xcount; x++)
- dest[x*rxstride + y*rystride] = ('rtype_name`)0;
-
- for (y = 0; y < ycount; y++)
- for (n = 0; n < count; n++)
- for (x = 0; x < xcount; x++)
- /* dest[x,y] += a[x,n] * b[n,y] */
- dest[x*rxstride + y*rystride] +=
- abase[x*axstride + n*aystride] *
- bbase[n*bxstride + y*bystride];
- }
else if (GFC_DESCRIPTOR_RANK (a) == 1)
{
const 'rtype_name` *restrict bbase_y;
@@ -523,6 +509,20 @@ sinclude(`matmul_asm_'rtype_code`.m4')dnl
dest[y*rxstride] = s;
}
}
+ else if (axstride < aystride)
+ {
+ for (y = 0; y < ycount; y++)
+ for (x = 0; x < xcount; x++)
+ dest[x*rxstride + y*rystride] = ('rtype_name`)0;
+
+ for (y = 0; y < ycount; y++)
+ for (n = 0; n < count; n++)
+ for (x = 0; x < xcount; x++)
+ /* dest[x,y] += a[x,n] * b[n,y] */
+ dest[x*rxstride + y*rystride] +=
+ abase[x*axstride + n*aystride] *
+ bbase[n*bxstride + y*bystride];
+ }
else
{
const 'rtype_name` *restrict abase_x;
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2021-04-23 10:17 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-04-23 10:17 [gcc(refs/vendors/redhat/heads/gcc-8-branch)] PR libfortran/97063 - Wrong result for vector (step size is negative) * matrix Jakub Jelinek
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).