Skip to content
Merged
Show file tree
Hide file tree
Changes from 1 commit
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Rewrite [ds]hgeqz to use FMA with Householder reflectors
  • Loading branch information
angsch committed Jul 6, 2022
commit 0a6cd431891bbb9cef57e461e023ac7f0c9e5d8e
70 changes: 37 additions & 33 deletions SRC/dhgeqz.f
Original file line number Diff line number Diff line change
Expand Up @@ -337,9 +337,9 @@ SUBROUTINE DHGEQZ( JOB, COMPQ, COMPZ, N, ILO, IHI, H, LDH, T, LDT,
$ BTOL, C, C11I, C11R, C12, C21, C22I, C22R, CL,
$ CQ, CR, CZ, ESHIFT, S, S1, S1INV, S2, SAFMAX,
$ SAFMIN, SCALE, SL, SQI, SQR, SR, SZI, SZR, T1,
$ TAU, TEMP, TEMP2, TEMPI, TEMPR, U1, U12, U12L,
$ U2, ULP, VS, W11, W12, W21, W22, WABS, WI, WR,
$ WR2
$ T2, T3, TAU, TEMP, TEMP2, TEMPI, TEMPR, U1,
$ U12, U12L, U2, ULP, VS, W11, W12, W21, W22,
$ WABS, WI, WR, WR2
* ..
* .. Local Arrays ..
DOUBLE PRECISION V( 3 )
Expand Down Expand Up @@ -1132,25 +1132,27 @@ SUBROUTINE DHGEQZ( JOB, COMPQ, COMPZ, N, ILO, IHI, H, LDH, T, LDT,
H( J+2, J-1 ) = ZERO
END IF
*
T2 = TAU*V( 2 )
T3 = TAU*V( 3 )
DO 230 JC = J, ILASTM
TEMP = TAU*( H( J, JC )+V( 2 )*H( J+1, JC )+V( 3 )*
$ H( J+2, JC ) )
H( J, JC ) = H( J, JC ) - TEMP
H( J+1, JC ) = H( J+1, JC ) - TEMP*V( 2 )
H( J+2, JC ) = H( J+2, JC ) - TEMP*V( 3 )
TEMP2 = TAU*( T( J, JC )+V( 2 )*T( J+1, JC )+V( 3 )*
$ T( J+2, JC ) )
T( J, JC ) = T( J, JC ) - TEMP2
T( J+1, JC ) = T( J+1, JC ) - TEMP2*V( 2 )
T( J+2, JC ) = T( J+2, JC ) - TEMP2*V( 3 )
TEMP = H( J, JC )+V( 2 )*H( J+1, JC )+V( 3 )*
$ H( J+2, JC )
H( J, JC ) = H( J, JC ) - TEMP*TAU
H( J+1, JC ) = H( J+1, JC ) - TEMP*T2
H( J+2, JC ) = H( J+2, JC ) - TEMP*T3
TEMP2 = T( J, JC )+V( 2 )*T( J+1, JC )+V( 3 )*
$ T( J+2, JC )
T( J, JC ) = T( J, JC ) - TEMP2*TAU
T( J+1, JC ) = T( J+1, JC ) - TEMP2*T2
T( J+2, JC ) = T( J+2, JC ) - TEMP2*T3
230 CONTINUE
IF( ILQ ) THEN
DO 240 JR = 1, N
TEMP = TAU*( Q( JR, J )+V( 2 )*Q( JR, J+1 )+V( 3 )*
$ Q( JR, J+2 ) )
Q( JR, J ) = Q( JR, J ) - TEMP
Q( JR, J+1 ) = Q( JR, J+1 ) - TEMP*V( 2 )
Q( JR, J+2 ) = Q( JR, J+2 ) - TEMP*V( 3 )
TEMP = Q( JR, J )+V( 2 )*Q( JR, J+1 )+V( 3 )*
$ Q( JR, J+2 )
Q( JR, J ) = Q( JR, J ) - TEMP*TAU
Q( JR, J+1 ) = Q( JR, J+1 ) - TEMP*T2
Q( JR, J+2 ) = Q( JR, J+2 ) - TEMP*T3
240 CONTINUE
END IF
*
Expand Down Expand Up @@ -1238,27 +1240,29 @@ SUBROUTINE DHGEQZ( JOB, COMPQ, COMPZ, N, ILO, IHI, H, LDH, T, LDT,
*
* Apply transformations from the right.
*
T2 = TAU*V(2)
T3 = TAU*V(3)
DO 260 JR = IFRSTM, MIN( J+3, ILAST )
TEMP = TAU*( H( JR, J )+V( 2 )*H( JR, J+1 )+V( 3 )*
$ H( JR, J+2 ) )
H( JR, J ) = H( JR, J ) - TEMP
H( JR, J+1 ) = H( JR, J+1 ) - TEMP*V( 2 )
H( JR, J+2 ) = H( JR, J+2 ) - TEMP*V( 3 )
TEMP = H( JR, J )+V( 2 )*H( JR, J+1 )+V( 3 )*
$ H( JR, J+2 )
H( JR, J ) = H( JR, J ) - TEMP*TAU
H( JR, J+1 ) = H( JR, J+1 ) - TEMP*T2
H( JR, J+2 ) = H( JR, J+2 ) - TEMP*T3
260 CONTINUE
DO 270 JR = IFRSTM, J + 2
TEMP = TAU*( T( JR, J )+V( 2 )*T( JR, J+1 )+V( 3 )*
$ T( JR, J+2 ) )
T( JR, J ) = T( JR, J ) - TEMP
T( JR, J+1 ) = T( JR, J+1 ) - TEMP*V( 2 )
T( JR, J+2 ) = T( JR, J+2 ) - TEMP*V( 3 )
TEMP = T( JR, J )+V( 2 )*T( JR, J+1 )+V( 3 )*
$ T( JR, J+2 )
T( JR, J ) = T( JR, J ) - TEMP*TAU
T( JR, J+1 ) = T( JR, J+1 ) - TEMP*T2
T( JR, J+2 ) = T( JR, J+2 ) - TEMP*T3
270 CONTINUE
IF( ILZ ) THEN
DO 280 JR = 1, N
TEMP = TAU*( Z( JR, J )+V( 2 )*Z( JR, J+1 )+V( 3 )*
$ Z( JR, J+2 ) )
Z( JR, J ) = Z( JR, J ) - TEMP
Z( JR, J+1 ) = Z( JR, J+1 ) - TEMP*V( 2 )
Z( JR, J+2 ) = Z( JR, J+2 ) - TEMP*V( 3 )
TEMP = Z( JR, J )+V( 2 )*Z( JR, J+1 )+V( 3 )*
$ Z( JR, J+2 )
Z( JR, J ) = Z( JR, J ) - TEMP*TAU
Z( JR, J+1 ) = Z( JR, J+1 ) - TEMP*T2
Z( JR, J+2 ) = Z( JR, J+2 ) - TEMP*T3
280 CONTINUE
END IF
T( J+1, J ) = ZERO
Expand Down
70 changes: 37 additions & 33 deletions SRC/shgeqz.f
Original file line number Diff line number Diff line change
Expand Up @@ -337,9 +337,9 @@ SUBROUTINE SHGEQZ( JOB, COMPQ, COMPZ, N, ILO, IHI, H, LDH, T, LDT,
$ BTOL, C, C11I, C11R, C12, C21, C22I, C22R, CL,
$ CQ, CR, CZ, ESHIFT, S, S1, S1INV, S2, SAFMAX,
$ SAFMIN, SCALE, SL, SQI, SQR, SR, SZI, SZR, T1,
$ TAU, TEMP, TEMP2, TEMPI, TEMPR, U1, U12, U12L,
$ U2, ULP, VS, W11, W12, W21, W22, WABS, WI, WR,
$ WR2
$ T2, T3, TAU, TEMP, TEMP2, TEMPI, TEMPR, U1,
$ U12, U12L, U2, ULP, VS, W11, W12, W21, W22,
$ WABS, WI, WR, WR2
* ..
* .. Local Arrays ..
REAL V( 3 )
Expand Down Expand Up @@ -1132,25 +1132,27 @@ SUBROUTINE SHGEQZ( JOB, COMPQ, COMPZ, N, ILO, IHI, H, LDH, T, LDT,
H( J+2, J-1 ) = ZERO
END IF
*
T2 = TAU * V( 2 )
T3 = TAU * V( 3 )
DO 230 JC = J, ILASTM
TEMP = TAU*( H( J, JC )+V( 2 )*H( J+1, JC )+V( 3 )*
$ H( J+2, JC ) )
H( J, JC ) = H( J, JC ) - TEMP
H( J+1, JC ) = H( J+1, JC ) - TEMP*V( 2 )
H( J+2, JC ) = H( J+2, JC ) - TEMP*V( 3 )
TEMP2 = TAU*( T( J, JC )+V( 2 )*T( J+1, JC )+V( 3 )*
$ T( J+2, JC ) )
T( J, JC ) = T( J, JC ) - TEMP2
T( J+1, JC ) = T( J+1, JC ) - TEMP2*V( 2 )
T( J+2, JC ) = T( J+2, JC ) - TEMP2*V( 3 )
TEMP = H( J, JC )+V( 2 )*H( J+1, JC )+V( 3 )*
$ H( J+2, JC )
H( J, JC ) = H( J, JC ) - TEMP*TAU
H( J+1, JC ) = H( J+1, JC ) - TEMP*T2
H( J+2, JC ) = H( J+2, JC ) - TEMP*T3
TEMP2 = T( J, JC )+V( 2 )*T( J+1, JC )+V( 3 )*
$ T( J+2, JC )
T( J, JC ) = T( J, JC ) - TEMP2*TAU
T( J+1, JC ) = T( J+1, JC ) - TEMP2*T2
T( J+2, JC ) = T( J+2, JC ) - TEMP2*T3
230 CONTINUE
IF( ILQ ) THEN
DO 240 JR = 1, N
TEMP = TAU*( Q( JR, J )+V( 2 )*Q( JR, J+1 )+V( 3 )*
$ Q( JR, J+2 ) )
Q( JR, J ) = Q( JR, J ) - TEMP
Q( JR, J+1 ) = Q( JR, J+1 ) - TEMP*V( 2 )
Q( JR, J+2 ) = Q( JR, J+2 ) - TEMP*V( 3 )
TEMP = Q( JR, J )+V( 2 )*Q( JR, J+1 )+V( 3 )*
$ Q( JR, J+2 )
Q( JR, J ) = Q( JR, J ) - TEMP*TAU
Q( JR, J+1 ) = Q( JR, J+1 ) - TEMP*T2
Q( JR, J+2 ) = Q( JR, J+2 ) - TEMP*T3
240 CONTINUE
END IF
*
Expand Down Expand Up @@ -1238,27 +1240,29 @@ SUBROUTINE SHGEQZ( JOB, COMPQ, COMPZ, N, ILO, IHI, H, LDH, T, LDT,
*
* Apply transformations from the right.
*
T2 = TAU*V( 2 )
T3 = TAU*V( 3 )
DO 260 JR = IFRSTM, MIN( J+3, ILAST )
TEMP = TAU*( H( JR, J )+V( 2 )*H( JR, J+1 )+V( 3 )*
$ H( JR, J+2 ) )
H( JR, J ) = H( JR, J ) - TEMP
H( JR, J+1 ) = H( JR, J+1 ) - TEMP*V( 2 )
H( JR, J+2 ) = H( JR, J+2 ) - TEMP*V( 3 )
TEMP = H( JR, J )+V( 2 )*H( JR, J+1 )+V( 3 )*
$ H( JR, J+2 )
H( JR, J ) = H( JR, J ) - TEMP*TAU
H( JR, J+1 ) = H( JR, J+1 ) - TEMP*T2
H( JR, J+2 ) = H( JR, J+2 ) - TEMP*T3
260 CONTINUE
DO 270 JR = IFRSTM, J + 2
TEMP = TAU*( T( JR, J )+V( 2 )*T( JR, J+1 )+V( 3 )*
$ T( JR, J+2 ) )
T( JR, J ) = T( JR, J ) - TEMP
T( JR, J+1 ) = T( JR, J+1 ) - TEMP*V( 2 )
T( JR, J+2 ) = T( JR, J+2 ) - TEMP*V( 3 )
TEMP = T( JR, J )+V( 2 )*T( JR, J+1 )+V( 3 )*
$ T( JR, J+2 )
T( JR, J ) = T( JR, J ) - TEMP*TAU
T( JR, J+1 ) = T( JR, J+1 ) - TEMP*T2
T( JR, J+2 ) = T( JR, J+2 ) - TEMP*T3
270 CONTINUE
IF( ILZ ) THEN
DO 280 JR = 1, N
TEMP = TAU*( Z( JR, J )+V( 2 )*Z( JR, J+1 )+V( 3 )*
$ Z( JR, J+2 ) )
Z( JR, J ) = Z( JR, J ) - TEMP
Z( JR, J+1 ) = Z( JR, J+1 ) - TEMP*V( 2 )
Z( JR, J+2 ) = Z( JR, J+2 ) - TEMP*V( 3 )
TEMP = Z( JR, J )+V( 2 )*Z( JR, J+1 )+V( 3 )*
$ Z( JR, J+2 )
Z( JR, J ) = Z( JR, J ) - TEMP*TAU
Z( JR, J+1 ) = Z( JR, J+1 ) - TEMP*T2
Z( JR, J+2 ) = Z( JR, J+2 ) - TEMP*T3
280 CONTINUE
END IF
T( J+1, J ) = ZERO
Expand Down