------- Comment #2 from paul dot richard dot thomas at cea dot fr 2006-09-14 08:45 ------- Created an attachment (id=12265) --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=12265&action=view) Patch to provide the shape of the SPREAD intrinsic + testcase
(In reply to comment #1) > Yes, it appears to be related to spread. If you comment out the > spread() in the subroutine the compiles. Additionally, if you > change "x%position(:,1:2)" to "x%position(1:3,1:2)", then the > code compiles. So, it looks like gfortran isn't interpreting the > index ranges from spread correctly if the lefthand side doesn't > have explicit ranges of indicies. Steve, That's right; neither the parameter array nor the call to SPREAD are passing the shape information to the scalarizer. Replacing the parameter by a variable permits the compilation to succeed. This is a minimal test case that illustrates the problem: real,dimension(:,:),pointer :: ptr real,dimension(1,2),parameter :: u = reshape((/ 0.25,0.75 /),(/1,2/)) ptr(:,:) = u + spread((/1.0/),2,size(u,2)) end It is fixed by the attached, as yet, unregtested patch. You will also find a dejagnuified testcase. I will do the regtest this afternoon and will submit tonight. Best regards Paul -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29060
