From 83ac986fc21de444c4fe53a85c620fe547f2726c Mon Sep 17 00:00:00 2001
From: Andreas Marek <amarek@rzg.mpg.de>
Date: Wed, 18 May 2016 12:29:16 +0200
Subject: [PATCH] Implement Debian patch for QR decomposition and modify it

The QR decomposition test program can now called with parameters for
matrix size, nev, and block size.
This closses issue #20
---
 Makefile.am                                   |  6 ++---
 src/elpa2.F90                                 |  2 +-
 src/elpa_qr/elpa_pdlarfb.F90                  |  1 +
 ..._real2_default_kernel_qr_decomposition.F90 | 23 ++++++++++++-------
 4 files changed, 20 insertions(+), 12 deletions(-)

diff --git a/Makefile.am b/Makefile.am
index cc10d182..f4d6be98 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -317,9 +317,9 @@ TESTS = $(check_SCRIPTS)
 	chmod +x $@
 
 ## this one does not want any arguments
-elpa2_test_real_default_kernel_qr_decomposition@SUFFIX@.sh:
-	echo '$(wrapper)./elpa2_test_real_default_kernel_qr_decomposition@SUFFIX@' > $@
-	chmod +x $@
+#elpa2_test_real_default_kernel_qr_decomposition@SUFFIX@.sh:
+#	echo '$(wrapper)./elpa2_test_real_default_kernel_qr_decomposition@SUFFIX@' > $@
+#	chmod +x $@
 
 #elpa2_test_real_default_kernel_qr_decomposition_single_precision@SUFFIX@.sh:
 #	echo '$(wrapper)./elpa2_test_real_default_kernel_qr_decomposition_single_precision@SUFFIX@' > $@
diff --git a/src/elpa2.F90 b/src/elpa2.F90
index 6a9f598e..1f28fa3e 100644
--- a/src/elpa2.F90
+++ b/src/elpa2.F90
@@ -193,7 +193,7 @@ function solve_evp_real_2stage(na, nev, a, lda, ev, q, ldq, nblk,        &
    endif
 
    if (useQRActual) then
-     if (mod(na,nblk) .ne. 0) then
+     if (mod(na,2) .ne. 0) then
        if (wantDebug) then
          write(error_unit,*) "solve_evp_real_2stage: QR-decomposition: blocksize does not fit with matrixsize"
        endif
diff --git a/src/elpa_qr/elpa_pdlarfb.F90 b/src/elpa_qr/elpa_pdlarfb.F90
index b1637ee0..1012d5f2 100644
--- a/src/elpa_qr/elpa_pdlarfb.F90
+++ b/src/elpa_qr/elpa_pdlarfb.F90
@@ -581,6 +581,7 @@ subroutine qr_tmerge_pdlarfb_1dcomm(m,mb,n,oldk,k,v,ldv,t,ldt,a,lda,baseidx,rev,
         else
             ! do not calculate parts for T merge as there is nothing to merge
 
+            mergeoffset  = 0
             updateoffset = 0
 
             tgenoffset = updateoffset + updatesize
diff --git a/test/fortran_test_programs/test_real2_default_kernel_qr_decomposition.F90 b/test/fortran_test_programs/test_real2_default_kernel_qr_decomposition.F90
index c5c665a0..3c1c30c1 100644
--- a/test/fortran_test_programs/test_real2_default_kernel_qr_decomposition.F90
+++ b/test/fortran_test_programs/test_real2_default_kernel_qr_decomposition.F90
@@ -136,19 +136,26 @@ program test_real2
 
    success = .true.
    !write_to_file = .false.
+   call read_input_parameters(na, nev, nblk, write_to_file)
 
-   if (COMMAND_ARGUMENT_COUNT() /= 0) then
-     write(error_unit,*) "This program does not support any command-line arguments"
-     stop 1
-   endif
+   !if (COMMAND_ARGUMENT_COUNT() /= 0) then
+   !  write(error_unit,*) "This program does not support any command-line arguments"
+   !  stop 1
+   !endif
 
    ! override nblk
-   nblk = 2
-   na   = 4000
-   nev  = 1500
+   !   nblk = 2
+   !   na   = 4000
+   !   nev  = 1500
+
+   ! make sure na, nbl is even
+   if (mod(nblk,2 ) .ne. 0) then
+     nblk = nblk - 1
+   endif
+
 
    ! make sure na is even
-   if (mod(na,nblk) .ne. 0) then
+   if (mod(na,2) .ne. 0) then
      na = na - 1
    endif
    ! make sure na is at least 34
-- 
GitLab