fdurastante
/
psblas3
mirror of https://github.com/sfilippone/psblas3.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Tweak SIMD directives for CSR.

Browse Source
newmatdist
sfilippone 1 year ago
parent 110a66e7d0
commit 922f7dd278
4 changed files with 48 additions and 48 deletions
Show Stats Download Patch File Download Diff File

24
base/serial/impl/psb_c_csr_impl.F90
Unescape Escape View File

@ -152,7 +152,7 @@ contains
!$omp parallel do private(i,j, acc) schedule(static) !$omp parallel do private(i,j, acc) schedule(static)
do i=1,m do i=1,m
acc = czero acc = czero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -164,7 +164,7 @@ contains
!$omp parallel do private(i,j, acc) !$omp parallel do private(i,j, acc)
do i=1,m do i=1,m
acc = czero acc = czero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -176,7 +176,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = czero acc = czero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -192,7 +192,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = czero acc = czero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -204,7 +204,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = czero acc = czero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -216,7 +216,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = czero acc = czero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -231,7 +231,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = czero acc = czero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -243,7 +243,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = czero acc = czero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -255,7 +255,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = czero acc = czero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -270,7 +270,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = czero acc = czero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -282,7 +282,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = czero acc = czero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -294,7 +294,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = czero acc = czero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo

24
base/serial/impl/psb_d_csr_impl.F90
Unescape Escape View File

@ -152,7 +152,7 @@ contains
!$omp parallel do private(i,j, acc) schedule(static) !$omp parallel do private(i,j, acc) schedule(static)
do i=1,m do i=1,m
acc = dzero acc = dzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -164,7 +164,7 @@ contains
!$omp parallel do private(i,j, acc) !$omp parallel do private(i,j, acc)
do i=1,m do i=1,m
acc = dzero acc = dzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -176,7 +176,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = dzero acc = dzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -192,7 +192,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = dzero acc = dzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -204,7 +204,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = dzero acc = dzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -216,7 +216,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = dzero acc = dzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -231,7 +231,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = dzero acc = dzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -243,7 +243,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = dzero acc = dzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -255,7 +255,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = dzero acc = dzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -270,7 +270,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = dzero acc = dzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -282,7 +282,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = dzero acc = dzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -294,7 +294,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = dzero acc = dzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo

24
base/serial/impl/psb_s_csr_impl.F90
Unescape Escape View File

@ -152,7 +152,7 @@ contains
!$omp parallel do private(i,j, acc) schedule(static) !$omp parallel do private(i,j, acc) schedule(static)
do i=1,m do i=1,m
acc = szero acc = szero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -164,7 +164,7 @@ contains
!$omp parallel do private(i,j, acc) !$omp parallel do private(i,j, acc)
do i=1,m do i=1,m
acc = szero acc = szero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -176,7 +176,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = szero acc = szero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -192,7 +192,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = szero acc = szero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -204,7 +204,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = szero acc = szero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -216,7 +216,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = szero acc = szero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -231,7 +231,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = szero acc = szero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -243,7 +243,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = szero acc = szero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -255,7 +255,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = szero acc = szero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -270,7 +270,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = szero acc = szero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -282,7 +282,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = szero acc = szero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -294,7 +294,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = szero acc = szero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo

24
base/serial/impl/psb_z_csr_impl.F90
Unescape Escape View File

@ -152,7 +152,7 @@ contains
!$omp parallel do private(i,j, acc) schedule(static) !$omp parallel do private(i,j, acc) schedule(static)
do i=1,m do i=1,m
acc = zzero acc = zzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -164,7 +164,7 @@ contains
!$omp parallel do private(i,j, acc) !$omp parallel do private(i,j, acc)
do i=1,m do i=1,m
acc = zzero acc = zzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -176,7 +176,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = zzero acc = zzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -192,7 +192,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = zzero acc = zzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -204,7 +204,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = zzero acc = zzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -216,7 +216,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = zzero acc = zzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -231,7 +231,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = zzero acc = zzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -243,7 +243,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = zzero acc = zzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -255,7 +255,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = zzero acc = zzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -270,7 +270,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = zzero acc = zzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -282,7 +282,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = zzero acc = zzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo
@ -294,7 +294,7 @@ contains
!$omp parallel do private(i,j,acc) !$omp parallel do private(i,j,acc)
do i=1,m do i=1,m
acc = zzero acc = zzero
!$omp simd !$omp simd reduction(+:acc)
do j=irp(i), irp(i+1)-1 do j=irp(i), irp(i+1)-1
acc = acc + val(j) * x(ja(j)) acc = acc + val(j) * x(ja(j))
enddo enddo

Write Preview
Loading…
Cancel
Save