J3/14-253r1 To: J3 From: John Reid & Daniel Chen Subject: Known inactive images Date: 2014 October 14 References: 14-250r2, N2027 Discussion ---------- Old proposal: The changes proposed here are based on a suggestion in my ballot. I think we need a clear definition of what is meant by a "known" failed or stopped image. And if we add the concept of a stalled image, as proposed in 14-250, I think we need a clear definition of what is meant by a "known" stalled image. New proposal: We find defining "known" failed, stalled or stopped images are misleading. However, adding a note to describe how a program can detect failed, stalled or stopped images and retrieve that information could be helpful. Additionally, the existing description of STAT_STOPPED_IMAGES in the base standard needs to account for setting a STAT argument in a call to a collective subroutine. Edits to N2027: --------------- [14:7+] Add a Note 5.10 "In addition to detecting that an image failed by having the variable in a STAT=specifier or a STAT argument of a call to a collective or atomic subroutine assigned the value STAT_FAILED_IMAGE, an image can get the indices of failed images in a specified team by invoking the intrinsic function FAILED_IMAGES. There could be a delay from when an image has failed to when a processor detects it." [14:7+] Add a Note 5.11 to the end of the new subclause 5.9: "In addition to detecting that an image stalled by having the variable in a STAT=specifier or a STAT argument of a call to a collective subroutine assigned the value STAT_STALLED_IMAGE, an image can get the indices of stalled images in a specified team by invoking the intrinsic function STALLED_IMAGES. There could be a delay from when an image has stalled to when a processor detects it." [40:15+] Add "{In 13.8.2.24 STAT_STOPPED_IMAGE, edit the paragraph as follows.} The value of the default integer scalar constant STAT_STOPPED_IMAGE is assigned to the variable specified in a STAT= specifier (6.7.4, 8.5.7) \uwave{or a STAT argument in a call to a collective subroutine} if execution of the statement with that specifier or argument requires synchronization with an image that has initiated termination of execution. This value shall be positive \sout{and different from the value of IOSTAT INQUIRE INTERNAL UNIT}." [40:15+] Add "{In 13.8.2.24 STAT_STOPPED_IMAGE, insert a new Note after paragraph 1} "In addition to detecting that an image initiated normal termination by having the variable in a STAT=specifier or a STAT argument of a call to a collective subroutine assigned the value STAT_STOPPED_IMAGE, an image can get the indices of the images that have initiated normal termination in a specified team by invoking the intrinsic function STOPPED_IMAGES. There could be a delay from when an image has initiated normal termination to when a processor detects it."