Update lang/pocl to 0.14 and build with llvm40 by default

[rezny]

Summary

With the release of pocl 0.14 it is now compatible with llvm39 and llvm40 so with this update the port now defaults to 40 and will follow MESA_LLVM_VER if set. The port switches to using CMake because that is what this release supports; there are no autotools files in 0.14. Also, i386 has been unblocked as it appears to work and upstream appears to support even non-x86. Maintainer timed out in PR 218332.

Test Plan

Poudriere 11.0 amd64/i386 OK

[rezny]

Trim obsolete USES=libtool:keepla

[rezny]

Update based on feedback in the PR.

[jbeich]

lang/pocl/Makefile

37

Did you forget to drop no longer used DEBUG option?

lang/pocl/files/patch-lib_CL_devices_cpuinfo.c

20–21

Prepend defined(__DragonFly__) || as hw.clockrate is also supported there. OpenBSD can use HW_CPUSPEED instead but out of scope for now.

pocl-0.13 builds fine on DragonFly without patches. Let's keep pocl-0.14 the same way:
https://github.com/DragonFlyBSD/DeltaPorts/tree/master/ports/lang/pocl
http://muscles.dragonflybsd.org/synth/logs/lang___pocl.log

35

Replace with #else that calls sysconf(_SC_NPROCESSORS_ONLN) instead. All BSDs (and OS X) support hw.ncpu. However, on NetBSD hw.ncpu is different from hw.ncpuonline.

https://github.com/DragonFlyBSD/DragonFlyBSD/blob/6f1d2f41495d/lib/libc/gen/sysconf.c#L629
https://github.com/jsonn/src/blob/105a2837c72d/lib/libc/gen/sysconf.c#L370

[jbeich]

lang/pocl/files/patch-lib_CL_devices_cpuinfo.c

72

Maybe fill device->long_name via HW_MODEL sysctl on all BSDs.

[rezny]

Revise patch-lib_CL_devices_cpuinfo.c

lang/pocl/Makefile

37

Cmake responds to the DEBUG option without any extra setting. Try the option and notice it builds Debug instead of Release configuration.

lang/pocl/files/patch-lib_CL_devices_cpuinfo.c

20–21

I had just quickly changed preprocessor directives initially. Going back and looking at the code within, there's probably a more accurate answer under dev.cpu than what hw.clockrate offers but I do not know how portable that would be.

[jbeich]

I'm a bit biased against DEBUG but the rest looks fine.

lang/pocl/Makefile

37

Does it provide anything more than symbols (e.g. assertions, statistics, verbosity, slow path)? Otherwise, WITH_DEBUG=1 facilitiy is generic for every port that respects CFLAGS/STRIP. To get useful stacktraces one still needs to build dependencies with symbols. And default flags may not be enough e.g., crashes may disappear without -O level or be confusing without -fsanitize=address.

[rezny]

In D10508#219733, @jbeich wrote:

I'm a bit biased against DEBUG but the rest looks fine.

I had left DEBUG in place because it was there and was doing something but if it shouldn't be there then DEBUG can easily go; WITH_DEBUG=1 does the same thing as setting the option.

[rezny]

Improve platform support as suggested and drop DEBUG.

[jbeich]

lang/pocl/files/patch-lib_CL_devices_cpuinfo.c

8

Windows and Solaris lack <sys/sysctl.h>, so when upstreaming you may need include guards.

# CMakeLists.txt
include(CheckIncludeFiles)
check_include_files("sys/types.h;sys/sysctl.h" HAVE_SYS_SYSCTL_H)

/* config.h.in.cmake */
#cmakedefine HAVE_SYS_SYSCTL_H

/* lib/CL/devices/cpuinfo.c */
#ifdef HAVE_SYS_SYSCTL_H
#include <sys/types.h>
#include <sys/sysctl.h>
#endif

30

value string seems to lack terminating NUL character.

$ cc -g -fsanitize=address test.c
$ ./a.out
=================================================================
==17273==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x60f0000000ed at pc 0x00000042d769 bp 0x7fffffffde30 sp 0x7fffffffd5d8
READ of size 174 at 0x60f0000000ed thread T0
    #0 0x42d768 in __interceptor_strlen (/home/foo/a.out+0x42d768)
    #1 0x80125a314 in vsscanf_l /usr/src/lib/libc/stdio/vsscanf.c:69:23
    #2 0x80125a314 in vsscanf /usr/src/lib/libc/stdio/vsscanf.c:77
    #3 0x43d24d in sscanf (/home/foo/a.out+0x43d24d)
    #4 0x499bba in pocl_cpuinfo_detect_max_clock_frequency /home/foo/test.c:26:7
    #5 0x499cfa in main /home/foo/test.c:40:18
    #6 0x40d1af in _start /usr/src/lib/csu/amd64/crt1.c:72:7

0x60f0000000ed is located 0 bytes to the right of 173-byte region [0x60f000000040,0x60f0000000ed)
allocated by thread T0 here:
    #0 0x48555c in malloc (/home/foo/a.out+0x48555c)
    #1 0x499b5e in pocl_cpuinfo_detect_max_clock_frequency /home/foo/test.c:23:23
    #2 0x499cfa in main /home/foo/test.c:40:18
    #3 0x40d1af in _start /usr/src/lib/csu/amd64/crt1.c:72:7
    #4 0x8006ddfff  (<unknown module>)

SUMMARY: AddressSanitizer: heap-buffer-overflow (/home/foo/a.out+0x42d768) in __interceptor_strlen
Shadow bytes around the buggy address:
  0x4c1dffffffc0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x4c1dffffffd0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x4c1dffffffe0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x4c1dfffffff0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x4c1e00000000: fa fa fa fa fa fa fa fa 00 00 00 00 00 00 00 00
=>0x4c1e00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00[05]fa fa
  0x4c1e00000020: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
  0x4c1e00000030: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
  0x4c1e00000040: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
  0x4c1e00000050: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
  0x4c1e00000060: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
Shadow byte legend (one shadow byte represents 8 application bytes):
  Addressable:           00
  Partially addressable: 01 02 03 04 05 06 07
  Heap left redzone:       fa
  Freed heap region:       fd
  Stack left redzone:      f1
  Stack mid redzone:       f2
  Stack right redzone:     f3
  Stack after return:      f5
  Stack use after scope:   f8
  Global redzone:          f9
  Global init order:       f6
  Poisoned by user:        f7
  Container overflow:      fc
  Array cookie:            ac
  Intra object redzone:    bb
  ASan internal:           fe
  Left alloca redzone:     ca
  Right alloca redzone:    cb
==17273==ABORTING

79

#elif defined(HW_MODEL) and/or #elif defined(HAVE_SYS_SYSCTL_H) (see another comment) maybe less assuming about portability.

[jbeich]

lang/pocl/files/patch-lib_CL_devices_cpuinfo.c

8

If you do this move #ifdef HAVE_SYS_SYSCTL_H block after #include "config.h". ;)

[rezny]

lang/pocl/files/patch-lib_CL_devices_cpuinfo.c

8

I noticed the MSVC check but this file looks to be rather Linux specific so I had not thought about portability beyond there, not that I had put any thought to upstreaming, but easy enough to add those checks.

30

Well, that answers that question... sysctl(3) doesn't specifically address the NUL terminator but implies by example that it would be provided and accounted for in the size. On the other hand, I have seen manual NUL termination of strings from sysctl() often enough to be suspicious of the simplistic example.

[rezny]

Terminate strings from sysctl and add header include guard for portability.

[swills]

Approved, looks good