Lancelot Six
1c850ca80d
ROCm programs can load a high number of compute kernels on GPU devices, especially if lazy code-object loading have been disabled. Each code object containing such program is loaded once for each device available, and each instance is reported by GDB as an individual shared library. We came across situations where the number of shared libraries opened by GDB gets higher than the allowed number of opened files for the process. Increasing the opened files limit works around the problem, but there is a better way this patch proposes to follow. Under the hood, the GPU code objects are embedded inside the host application binary and shared library binaries. GDB currently opens the underlying file once for each shared library it sees. That means that the same file is re-opened every time a code object is loaded on a GPU. This patch proposes to only open each underlying file once. This is done by implementing a reference counting mechanism so the underlying file is opened when the underlying file first needs to be opened, and closed when the last BFD using the underlying file is closed. On a program where GDB used to open about 1500 files to load all shared libraries, this patch makes it so only 54 opened file descriptors are needed. I have tested this patch on downstream ROCgdb's full testsuite and upstream GDB testsuite with no regression. Approved-By: Pedro Alves <pedro@palves.net>
README for GNU development tools This directory contains various GNU compilers, assemblers, linkers, debuggers, etc., plus their support routines, definitions, and documentation. If you are receiving this as part of a GDB release, see the file gdb/README. If with a binutils release, see binutils/README; if with a libg++ release, see libg++/README, etc. That'll give you info about this package -- supported targets, how to use it, how to report bugs, etc. It is now possible to automatically configure and build a variety of tools with one command. To build all of the tools contained herein, run the ``configure'' script here, e.g.: ./configure make To install them (by default in /usr/local/bin, /usr/local/lib, etc), then do: make install (If the configure script can't determine your type of computer, give it the name as an argument, for instance ``./configure sun4''. You can use the script ``config.sub'' to test whether a name is recognized; if it is, config.sub translates it to a triplet specifying CPU, vendor, and OS.) If you have more than one compiler on your system, it is often best to explicitly set CC in the environment before running configure, and to also set CC when running make. For example (assuming sh/bash/ksh): CC=gcc ./configure make A similar example using csh: setenv CC gcc ./configure make Much of the code and documentation enclosed is copyright by the Free Software Foundation, Inc. See the file COPYING or COPYING.LIB in the various directories, for a description of the GNU General Public License terms under which you can copy the files. REPORTING BUGS: Again, see gdb/README, binutils/README, etc., for info on where and how to report problems.