Re: [PATCH] nvptx: do not implicitly enable -ftoplevel-reorder

[Thomas Schwinge <thomas at codesourcery dot com>]

Hi!

On Wed, 20 Jul 2016 08:27:56 -0400, Nathan Sidwell <nathan@acm.org> wrote:
> On 07/19/16 14:34, Alexander Monakov wrote:
> > I've recently committed a middle-end patch that adds handling of undefined
> > variables (that the nvptx backend needs) under -fno-toplevel-reorder (svn rev.
> > 238371).  With that change, it's no longer necessary to implicitly enable
> > -ftoplevel-reorder in the backend, and the following patch removes that.

Yay for less special-casing!

> > Tested with nvptx-none-run, OK for trunk?
> 
> ok thanks

Hmm.  In an offloading configuration I see the following regression:

    [-PASS:-]{+FAIL:+} libgomp.oacc-c/../libgomp.oacc-c-c++-common/reduction-cplx-dbl.c -DACC_DEVICE_TYPE_nvidia=1 -DACC_MEM_SHARED=0 -foffload=nvptx-none  -O0  (test for excess errors)
    [-PASS:-]{+UNRESOLVED:+} libgomp.oacc-c/../libgomp.oacc-c-c++-common/reduction-cplx-dbl.c -DACC_DEVICE_TYPE_nvidia=1 -DACC_MEM_SHARED=0 -foffload=nvptx-none  -O0  [-execution test-]{+compilation failed to produce executable+}

    ptxas /tmp/ccmABlIh.o, line 971; error   : State space mismatch between instruction and address in instruction 'atom'
    ptxas /tmp/ccmABlIh.o, line 1008; error   : State space mismatch between instruction and address in instruction 'atom'
    ptxas /tmp/ccmABlIh.o, line 1017; error   : State space mismatch between instruction and address in instruction 'atom'
    ptxas /tmp/ccmABlIh.o, line 1035; error   : State space mismatch between instruction and address in instruction 'atom'
    ptxas /tmp/ccmABlIh.o, line 1331; error   : State space mismatch between instruction and address in instruction 'atom'
    ptxas /tmp/ccmABlIh.o, line 1373; error   : State space mismatch between instruction and address in instruction 'atom'
    ptxas /tmp/ccmABlIh.o, line 1385; error   : State space mismatch between instruction and address in instruction 'atom'
    ptxas /tmp/ccmABlIh.o, line 1408; error   : State space mismatch between instruction and address in instruction 'atom'
    ptxas /tmp/ccmABlIh.o, line 971; error   : Unknown symbol '__reduction_lock'
    ptxas /tmp/ccmABlIh.o, line 971; error   : Label expected for forward reference of '__reduction_lock'
    ptxas /tmp/ccmABlIh.o, line 1008; error   : Unknown symbol '__reduction_lock'
    ptxas /tmp/ccmABlIh.o, line 1008; error   : Label expected for forward reference of '__reduction_lock'
    ptxas /tmp/ccmABlIh.o, line 1017; error   : Unknown symbol '__reduction_lock'
    ptxas /tmp/ccmABlIh.o, line 1017; error   : Label expected for forward reference of '__reduction_lock'
    ptxas /tmp/ccmABlIh.o, line 1035; error   : Unknown symbol '__reduction_lock'
    ptxas /tmp/ccmABlIh.o, line 1035; error   : Label expected for forward reference of '__reduction_lock'
    ptxas /tmp/ccmABlIh.o, line 1331; error   : Unknown symbol '__reduction_lock'
    ptxas /tmp/ccmABlIh.o, line 1331; error   : Label expected for forward reference of '__reduction_lock'
    ptxas /tmp/ccmABlIh.o, line 1373; error   : Unknown symbol '__reduction_lock'
    ptxas /tmp/ccmABlIh.o, line 1373; error   : Label expected for forward reference of '__reduction_lock'
    ptxas /tmp/ccmABlIh.o, line 1385; error   : Unknown symbol '__reduction_lock'
    ptxas /tmp/ccmABlIh.o, line 1385; error   : Label expected for forward reference of '__reduction_lock'
    ptxas /tmp/ccmABlIh.o, line 1408; error   : Unknown symbol '__reduction_lock'
    ptxas /tmp/ccmABlIh.o, line 1408; error   : Label expected for forward reference of '__reduction_lock'
    ptxas fatal   : Ptx assembly aborted due to errors
    nvptx-as: ptxas returned 255 exit status
    mkoffload: fatal error: [...]/build-gcc/gcc/x86_64-pc-linux-gnu-accel-nvptx-none-gcc return
    ed 1 exit status
    compilation terminated.
    lto-wrapper: fatal error: [...]/build-gcc/gcc//accel/nvptx-none/mkoffload returned 1 exit s
    tatus
    compilation terminated.
    [...]/ld: lto-wrapper failed
    collect2: error: ld returned 1 exit status

Same for C++.

..., and in nvptx-none target testing I see the following:

C:

    [-PASS:-]{+WARNING: program timed out.+}
    {+FAIL:+} gcc.dg/large-size-array-4.c (test for excess errors)

This one times out when creating a huge *.s file:

    [...]
            .global .align 8 .u64 name[2147483649] = { 0, 0, [...]

Running with -ftoplevel-reorder (as implicitly enabled before), a *.s
file with just the preamble gets created.  Are these behaviors correct?

    [-PASS:-]{+FAIL:+} gcc.dg/pr16973.c (test for excess errors)

    [...]/source-gcc/gcc/testsuite/gcc.dg/pr16973.c:12:1: error: initializer for integer/fixed-point value is too complicated

C++:

    PASS: g++.dg/debug/dwarf2/dwarf4-typedef.C  -std=gnu++98  (test for warnings, line 9)
    FAIL: g++.dg/debug/dwarf2/dwarf4-typedef.C  -std=gnu++98 (test for excess errors)
    PASS: g++.dg/debug/dwarf2/dwarf4-typedef.C  -std=gnu++11  (test for warnings, line 9)
    [-PASS:-]{+FAIL:+} g++.dg/debug/dwarf2/dwarf4-typedef.C  -std=gnu++11 (test for excess errors)
    PASS: g++.dg/debug/dwarf2/dwarf4-typedef.C  -std=gnu++14  (test for warnings, line 9)
    [-PASS:-]{+FAIL:+} g++.dg/debug/dwarf2/dwarf4-typedef.C  -std=gnu++14 (test for excess errors)

These now all FAIL due to "sorry, unimplemented: target cannot support
nonlocal goto", which is "acceptable", but should be XFAILed.  (Though,
that has generally not yet been done for C++.)

    UNSUPPORTED: g++.dg/cpp0x/variadic98.C  -std=c++98
    [-PASS:-]{+FAIL:+} g++.dg/cpp0x/variadic98.C  -std=c++11 (test for excess errors)
    [-PASS:-]{+FAIL:+} g++.dg/cpp0x/variadic98.C  -std=c++14 (test for excess errors)

Again, "sorry, unimplemented: target cannot support nonlocal goto".

    UNSUPPORTED: g++.dg/cpp1y/nsdmi-aggr1.C  -std=c++98
    UNSUPPORTED: g++.dg/cpp1y/nsdmi-aggr1.C  -std=c++11
    [-PASS:-]{+FAIL:+} g++.dg/cpp1y/nsdmi-aggr1.C  -std=c++14 (test for excess errors)
    [-PASS:-]{+UNRESOLVED:+} g++.dg/cpp1y/nsdmi-aggr1.C  -std=c++14 [-execution test-]{+compilation failed to produce executable+}

Now fails with:

    nvptx-as: circular reference in variable initializers

    FAIL: g++.dg/other/pr47218.C  -std=gnu++98 (test for excess errors)
    [-FAIL: g++.dg/other/pr47218.C  -std=gnu++11 (internal compiler error)-]
    FAIL: g++.dg/other/pr47218.C  -std=gnu++11 (test for excess errors)
    [-FAIL: g++.dg/other/pr47218.C  -std=gnu++14 (internal compiler error)-]
    FAIL: g++.dg/other/pr47218.C  -std=gnu++14 (test for excess errors)

:-) These now all FAIL with "sorry, unimplemented: target cannot support
nonlocal goto" instead of cryptic ptxas errors before:

    ptxas pr47218.o, line 74; error   : Unimplemented feature: labels as initial values
    ptxas pr47218.o, line 339; error   : Arguments mismatch for instruction 'mov'
    ptxas pr47218.o, line 341; error   : Arguments mismatch for instruction 'mov'
    ptxas pr47218.o, line 387; fatal   : Parsing error near '.byte': syntax error
    ptxas fatal   : Ptx assembly aborted due to errors
    nvptx-as: ptxas returned 255 exit status
    xg++: internal compiler error: Segmentation fault (program as)

    [-PASS:-]{+FAIL:+} g++.dg/torture/20090329-1.C   -O0  (test for excess errors)
    PASS: g++.dg/torture/20090329-1.C   -O1  (test for excess errors)
    PASS: g++.dg/torture/20090329-1.C   -O2  (test for excess errors)
    PASS: g++.dg/torture/20090329-1.C   -O3 -fomit-frame-pointer -funroll-loops -fpeel-loops -ftracer -finline-functions  (test for excess errors)

Another "sorry, unimplemented: target cannot support nonlocal goto".

    [-PASS:-]{+FAIL:+} g++.dg/torture/pr42714.C   -O0  (test for excess errors)
    PASS: g++.dg/torture/pr42714.C   -O1  (test for excess errors)
    PASS: g++.dg/torture/pr42714.C   -O2  (test for excess errors)
    PASS: g++.dg/torture/pr42714.C   -O3 -g  (test for excess errors)

Another "sorry, unimplemented: target cannot support nonlocal goto".

    [-PASS:-]{+FAIL:+} g++.dg/torture/pr49039.C   -O0  (test for excess errors)
    [-PASS:-]{+UNRESOLVED:+} g++.dg/torture/pr49039.C   -O0  [-execution test-]{+compilation failed to produce executable+}
    PASS: g++.dg/torture/pr49039.C   -O1  (test for excess errors)
    PASS: g++.dg/torture/pr49039.C   -O1  execution test
    PASS: g++.dg/torture/pr49039.C   -O2  (test for excess errors)

Another "sorry, unimplemented: target cannot support nonlocal goto".

    [-PASS:-]{+FAIL:+} g++.dg/torture/pr64312.C   -O0  (test for excess errors)
    PASS: g++.dg/torture/pr64312.C   -O1  (test for excess errors)
    PASS: g++.dg/torture/pr64312.C   -O2  (test for excess errors)
    PASS: g++.dg/torture/pr64312.C   -O3 -g  (test for excess errors)

Another "sorry, unimplemented: target cannot support nonlocal goto".

    FAIL: g++.old-deja/g++.benjamin/tem01.C  -std=c++98 (test for excess errors)
    [-PASS:-]{+FAIL:+} g++.old-deja/g++.benjamin/tem01.C  -std=c++11 (test for excess errors)
    [-PASS:-]{+FAIL:+} g++.old-deja/g++.benjamin/tem01.C  -std=c++14 (test for excess errors)

Two more "sorry, unimplemented: target cannot support nonlocal goto".

    [-PASS:-]{+FAIL:+} g++.old-deja/g++.brendan/crash47.C  -std=c++98 (test for excess errors)
    [-PASS:-]{+FAIL:+} g++.old-deja/g++.brendan/crash47.C  -std=c++11 (test for excess errors)
    [-PASS:-]{+FAIL:+} g++.old-deja/g++.brendan/crash47.C  -std=c++14 (test for excess errors)

More "sorry, unimplemented: target cannot support nonlocal goto".  But
why did these PASS before?  Is this generally just a problem with -O0, or
are there any optimizations doing things differently whether
-ftoplevel-reorder is in effect or not, and we could perhaps shuffle some
things so that for nvptx we run into "sorry, unimplemented: target cannot
support nonlocal goto" less often?  That is, I'm not clear on how
-fno-toplevel-reorder interacts with "nonlocal goto".

    [-PASS:-]{+FAIL:+} g++.old-deja/g++.jason/overload33.C  -std=c++98 (test for excess errors)
    [-PASS:-]{+FAIL:+} g++.old-deja/g++.jason/overload33.C  -std=c++11 (test for excess errors)
    [-PASS:-]{+FAIL:+} g++.old-deja/g++.jason/overload33.C  -std=c++14 (test for excess errors)

More "sorry, unimplemented: target cannot support nonlocal goto"
regressions.

    PASS: g++.old-deja/g++.pt/crash55.C  -std=c++98 (test for excess errors)
    [-PASS:-]{+FAIL:+} g++.old-deja/g++.pt/crash55.C  -std=c++11 (test for excess errors)
    [-PASS:-]{+FAIL:+} g++.old-deja/g++.pt/crash55.C  -std=c++14 (test for excess errors)

These now fail with:

    ptxas crash55.o, line 61; error   : Arguments mismatch for instruction 'mov'
    ptxas crash55.o, line 63; error   : Arguments mismatch for instruction 'mov'
    ptxas crash55.o, line 80; error   : Label expected for argument 0 of instruction 'call'
    ptxas crash55.o, line 80; error   : Function '_ZN3fooIcE1dEPS0_' not declared in this scope
    ptxas crash55.o, line 80; error   : Call target not recognized
    ptxas crash55.o, line 91; fatal   : Parsing error near '.byte': syntax error
    ptxas fatal   : Ptx assembly aborted due to errors
    nvptx-as: ptxas returned 255 exit status

    [-PASS:-]{+FAIL:+} g++.old-deja/g++.pt/ttp64.C  -std=c++98 (test for excess errors)
    [-PASS:-]{+FAIL:+} g++.old-deja/g++.pt/ttp64.C  -std=c++11 (test for excess errors)
    [-PASS:-]{+FAIL:+} g++.old-deja/g++.pt/ttp64.C  -std=c++14 (test for excess errors)

More "sorry, unimplemented: target cannot support nonlocal goto"
regressions.


Grüße
 Thomas

Attachment: signature.asc
Description: PGP signature