[Bug middle-end/85450] ICE: invalid types in nop conversion during GIMPLE pass: ompexp

[ktkachov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85450

ktkachov at gcc dot gnu.org changed:

   What|Removed |Added

 CC||ktkachov at gcc dot gnu.org

--- Comment #11 from ktkachov at gcc dot gnu.org ---
The commit r259711 should have fixed the aarch64 ICEs

[Bug middle-end/85450] ICE: invalid types in nop conversion during GIMPLE pass: ompexp

[sch...@linux-m68k.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85450

--- Comment #10 from Andreas Schwab  ---
This also breaks building libgcc with -mabi=ilp32, and the patch in #c8 fixes
that.

[Bug middle-end/85450] ICE: invalid types in nop conversion during GIMPLE pass: ompexp

[clyon at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85450

--- Comment #9 from Christophe Lyon  ---
Created attachment 44024
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=44024=edit
vfprintf.i

Your patch proposal fixes the ICE with sysopen.i, but I still have a crash with
vfprintf.i (with -mabi=ilp32)

[Bug middle-end/85450] ICE: invalid types in nop conversion during GIMPLE pass: ompexp

[rguenther at suse dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85450

--- Comment #8 from rguenther at suse dot de  ---
On Thu, 26 Apr 2018, clyon at gcc dot gnu.org wrote:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85450
> 
> --- Comment #7 from Christophe Lyon  ---
> Created attachment 44023
>   --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=44023=edit
> Newlib's sysopen.i
> 
> I could reproduce the problem with aarch64-none-elf too, and this requires
> -mabi=ilp32.
> 
> Compiling the attached file with -c ./sysopen.i  -mabi=ilp32:
> 
> In file included from
> /home/christophe.lyon/src/GCC/sources/newlib/newlib/libc/include/sys/fcntl.h:3,
>  from
> /home/christophe.lyon/src/GCC/sources/newlib/newlib/libc/include/fcntl.h:1,
>  from
> /home/christophe.lyon/src/GCC/sources/newlib/newlib/libc/syscalls/sysopen.c:4:
> /home/christophe.lyon/src/GCC/sources/newlib/newlib/libc/include/sys/_default_fcntl.h:
> In function 'open':
> /home/christophe.lyon/src/GCC/sources/newlib/newlib/libc/include/sys/_default_fcntl.h:182:12:
> error: invalid types in nop conversion
>  extern int open _PARAMS ((const char *, int, ...));
> ^~~~
> long long int
> void *
> _19 = (long long int) _18;
> /home/christophe.lyon/src/GCC/sources/newlib/newlib/libc/include/sys/_default_fcntl.h:182:12:
> error: invalid types in nop conversion
> long long int
> void *
> _28 = (long long int) _18;
> during GIMPLE pass: lower_vaarg
> /home/christophe.lyon/src/GCC/sources/newlib/newlib/libc/include/sys/_default_fcntl.h:182:12:
> internal compiler error: verify_gimple failed
> 0xc4cd2b verify_gimple_in_cfg(function*, bool)
> 
> /home/christophe.lyon/src/GCC/sources/gcc-fsf/trunk/gcc/tree-cfg.c:5585
> 0xb03a27 execute_function_todo
> /home/christophe.lyon/src/GCC/sources/gcc-fsf/trunk/gcc/passes.c:1994
> 0xb04399 execute_todo
> /home/christophe.lyon/src/GCC/sources/gcc-fsf/trunk/gcc/passes.c:2048

Caused by aarch64_gimplify_va_arg_expr:

  /* Advance ap.__stack  */
  t = fold_convert (intDI_type_node, arg);
  t = build2 (PLUS_EXPR, TREE_TYPE (t), t,
  build_int_cst (TREE_TYPE (t), size + 7));
  t = build2 (BIT_AND_EXPR, TREE_TYPE (t), t,
  build_int_cst (TREE_TYPE (t), -8));
  t = fold_convert (TREE_TYPE (arg), t);

I'm not sure why, for ILP32, this code converts to intDI_type_node.  
Probably missing ILP32 adjustments.

I suggest to simplify the code with

Index: gcc/config/aarch64/aarch64.c
===
--- gcc/config/aarch64/aarch64.c(revision 259669)
+++ gcc/config/aarch64/aarch64.c(working copy)
@@ -12278,12 +12278,9 @@ aarch64_gimplify_va_arg_expr (tree valis
   else
 roundup = NULL;
   /* Advance ap.__stack  */
-  t = fold_convert (intDI_type_node, arg);
-  t = build2 (PLUS_EXPR, TREE_TYPE (t), t,
- build_int_cst (TREE_TYPE (t), size + 7));
+  t = fold_build_pointer_plus_hwi (arg, size + 7);
   t = build2 (BIT_AND_EXPR, TREE_TYPE (t), t,
  build_int_cst (TREE_TYPE (t), -8));
-  t = fold_convert (TREE_TYPE (arg), t);
   t = build2 (MODIFY_EXPR, TREE_TYPE (stack), unshare_expr (stack), t);
   /* String up roundup and advance.  */
   if (roundup)

which fixes the issue for me.

[Bug middle-end/85450] ICE: invalid types in nop conversion during GIMPLE pass: ompexp

[clyon at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85450

--- Comment #7 from Christophe Lyon  ---
Created attachment 44023
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=44023=edit
Newlib's sysopen.i

I could reproduce the problem with aarch64-none-elf too, and this requires
-mabi=ilp32.

Compiling the attached file with -c ./sysopen.i  -mabi=ilp32:

In file included from
/home/christophe.lyon/src/GCC/sources/newlib/newlib/libc/include/sys/fcntl.h:3,
 from
/home/christophe.lyon/src/GCC/sources/newlib/newlib/libc/include/fcntl.h:1,
 from
/home/christophe.lyon/src/GCC/sources/newlib/newlib/libc/syscalls/sysopen.c:4:
/home/christophe.lyon/src/GCC/sources/newlib/newlib/libc/include/sys/_default_fcntl.h:
In function 'open':
/home/christophe.lyon/src/GCC/sources/newlib/newlib/libc/include/sys/_default_fcntl.h:182:12:
error: invalid types in nop conversion
 extern int open _PARAMS ((const char *, int, ...));
^~~~
long long int
void *
_19 = (long long int) _18;
/home/christophe.lyon/src/GCC/sources/newlib/newlib/libc/include/sys/_default_fcntl.h:182:12:
error: invalid types in nop conversion
long long int
void *
_28 = (long long int) _18;
during GIMPLE pass: lower_vaarg
/home/christophe.lyon/src/GCC/sources/newlib/newlib/libc/include/sys/_default_fcntl.h:182:12:
internal compiler error: verify_gimple failed
0xc4cd2b verify_gimple_in_cfg(function*, bool)
/home/christophe.lyon/src/GCC/sources/gcc-fsf/trunk/gcc/tree-cfg.c:5585
0xb03a27 execute_function_todo
/home/christophe.lyon/src/GCC/sources/gcc-fsf/trunk/gcc/passes.c:1994
0xb04399 execute_todo
/home/christophe.lyon/src/GCC/sources/gcc-fsf/trunk/gcc/passes.c:2048


My GCC is configured with:
--target=aarch64-none-elf --disable-nls --without-headers --with-newlib
--disable-shared --disable-threads --disable-libssp --disable-libgomp
--disable-libmudflap --disable-libatomic --disable-libcilkrts
--without-libquadmath --disable-libquadmath --enable-languages=c --with-newlib
--disable-tls

[Bug middle-end/85450] ICE: invalid types in nop conversion during GIMPLE pass: ompexp

[rguenther at suse dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85450

--- Comment #6 from rguenther at suse dot de  ---
On Thu, 26 Apr 2018, clyon at gcc dot gnu.org wrote:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85450
> 
> Christophe Lyon  changed:
> 
>What|Removed |Added
> 
>  CC||clyon at gcc dot gnu.org
> 
> --- Comment #5 from Christophe Lyon  ---
> Hi Richard,
> 
> This patch (r259667) makes the newlib build fail on aarch64_be-none-elf.
> 
> I'm seeing this in my build log:
> aarch64_be-none-elf-gcc
> -B/tmp/7730760_9.tmpdir/aci-gcc-fsf/builds/gcc-fsf-gccsrc/obj-aarch64_be-none-elf/newlib/aarch64_be-none-elf/ilp32/newlib/
> -isystem
> /tmp/7730760_9.tmpdir/aci-gcc-fsf/builds/gcc-fsf-gccsrc/obj-aarch64_be-none-elf/newlib/aarch64_be-none-elf/ilp32/newlib/targ-include
> -isystem /tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/newlib/libc/include
> -B/tmp/7730760_9.tmpdir/aci-gcc-fsf/builds/gcc-fsf-gccsrc/obj-aarch64_be-none-elf/newlib/aarch64_be-none-elf/ilp32/libgloss/aarch64
> -L/tmp/7730760_9.tmpdir/aci-gcc-fsf/builds/gcc-fsf-gccsrc/obj-aarch64_be-none-elf/newlib/aarch64_be-none-elf/ilp32/libgloss/libnosys
> -L/tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/libgloss/aarch64 
> -mabi=ilp32 -DPACKAGE_NAME=\"newlib\" -DPACKAGE_TARNAME=\"newlib\"
> -DPACKAGE_VERSION=\"2.0.0\" -DPACKAGE_STRING=\"newlib\ 2.0.0\"
> -DPACKAGE_BUGREPORT=\"\" -DPACKAGE_URL=\"\" -I.
> -I/tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/newlib/libc/stdio
> -fno-builtin  -g -O2  -mabi=ilp32 -fshort-enums -DINTEGER_ONLY -c
> /tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/newlib/libc/stdio/vfprintf.c
> -o lib_a-vfiprintf.o
> In file included from
> /tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/newlib/libc/stdio/vfwscanf.c:109:
> /tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/newlib/libc/stdio/vfwscanf.c:
> In function '__svfwscanf_r':
> /tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/newlib/libc/stdio/vfwscanf.c:140:25:
> error: invalid types in nop conversion
>  #  define __SVFWSCANF_R __svfwscanf_r
>  ^
> /tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/newlib/libc/stdio/vfwscanf.c:371:8:
> note: in expansion of macro '__SVFWSCANF_R'
>  _DEFUN(__SVFWSCANF_R, (rptr, fp, fmt0, ap),
> ^
> long long int
> void *
> _360 = (long long int) _368;
> [...]
> 
> 
> 
> 
> 
> and also:
> 0xc40fa3 verify_gimple_in_cfg(function*, bool)
>
> /tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/tree-cfg.c:5585
> 0xae70ec execute_function_todo
>
> /tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/passes.c:1994
> 0xae6e85 execute_todo
>
> /tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/passes.c:2048
> 
> I'd have to reproduce the problem manually in order to share a preprocessed
> file.

Can't investigate w/o that though.  Please also say how to configure
a cross to aarch64_be-none-elf (big-endian aarch64?  please not...)

[Bug middle-end/85450] ICE: invalid types in nop conversion during GIMPLE pass: ompexp

[clyon at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85450

Christophe Lyon  changed:

   What|Removed |Added

 CC||clyon at gcc dot gnu.org

--- Comment #5 from Christophe Lyon  ---
Hi Richard,

This patch (r259667) makes the newlib build fail on aarch64_be-none-elf.

I'm seeing this in my build log:
aarch64_be-none-elf-gcc
-B/tmp/7730760_9.tmpdir/aci-gcc-fsf/builds/gcc-fsf-gccsrc/obj-aarch64_be-none-elf/newlib/aarch64_be-none-elf/ilp32/newlib/
-isystem
/tmp/7730760_9.tmpdir/aci-gcc-fsf/builds/gcc-fsf-gccsrc/obj-aarch64_be-none-elf/newlib/aarch64_be-none-elf/ilp32/newlib/targ-include
-isystem /tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/newlib/libc/include
-B/tmp/7730760_9.tmpdir/aci-gcc-fsf/builds/gcc-fsf-gccsrc/obj-aarch64_be-none-elf/newlib/aarch64_be-none-elf/ilp32/libgloss/aarch64
-L/tmp/7730760_9.tmpdir/aci-gcc-fsf/builds/gcc-fsf-gccsrc/obj-aarch64_be-none-elf/newlib/aarch64_be-none-elf/ilp32/libgloss/libnosys
-L/tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/libgloss/aarch64 
-mabi=ilp32 -DPACKAGE_NAME=\"newlib\" -DPACKAGE_TARNAME=\"newlib\"
-DPACKAGE_VERSION=\"2.0.0\" -DPACKAGE_STRING=\"newlib\ 2.0.0\"
-DPACKAGE_BUGREPORT=\"\" -DPACKAGE_URL=\"\" -I.
-I/tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/newlib/libc/stdio
-fno-builtin  -g -O2  -mabi=ilp32 -fshort-enums -DINTEGER_ONLY -c
/tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/newlib/libc/stdio/vfprintf.c
-o lib_a-vfiprintf.o
In file included from
/tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/newlib/libc/stdio/vfwscanf.c:109:
/tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/newlib/libc/stdio/vfwscanf.c:
In function '__svfwscanf_r':
/tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/newlib/libc/stdio/vfwscanf.c:140:25:
error: invalid types in nop conversion
 #  define __SVFWSCANF_R __svfwscanf_r
 ^
/tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/newlib/newlib/libc/stdio/vfwscanf.c:371:8:
note: in expansion of macro '__SVFWSCANF_R'
 _DEFUN(__SVFWSCANF_R, (rptr, fp, fmt0, ap),
^
long long int
void *
_360 = (long long int) _368;
[...]





and also:
0xc40fa3 verify_gimple_in_cfg(function*, bool)
   
/tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/tree-cfg.c:5585
0xae70ec execute_function_todo
   
/tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/passes.c:1994
0xae6e85 execute_todo
   
/tmp/7730760_9.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/passes.c:2048

I'd have to reproduce the problem manually in order to share a preprocessed
file.

[Bug middle-end/85450] ICE: invalid types in nop conversion during GIMPLE pass: ompexp

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85450

--- Comment #3 from Richard Biener  ---
Author: rguenth
Date: Thu Apr 26 07:21:42 2018
New Revision: 259667

URL: https://gcc.gnu.org/viewcvs?rev=259667=gcc=rev
Log:
2018-04-26  Richard Biener  

PR middle-end/85450
* tree-cfg.c (verify_gimple_assign_unary): Restore proper
checking of integer<->pointer conversions.
* omp-expand.c (expand_omp_for_static_nochunk): Avoid
sign-/zero-extending pointer types.
(expand_omp_for_static_chunk): Likewise.

Modified:
trunk/gcc/ChangeLog
trunk/gcc/omp-expand.c
trunk/gcc/tree-cfg.c

[Bug middle-end/85450] ICE: invalid types in nop conversion during GIMPLE pass: ompexp

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85450

Richard Biener  changed:

   What|Removed |Added

 Status|NEW |RESOLVED
 Resolution|--- |FIXED

--- Comment #4 from Richard Biener  ---
Fixed.

[Bug middle-end/85450] ICE: invalid types in nop conversion during GIMPLE pass: ompexp

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85450

--- Comment #2 from Jakub Jelinek  ---
There are 4 spots, unsigned_type_for breaks for-5.c though.
There are several other spots that do use signed_type_for instead, so following
works for the testsuite:
--- gcc/omp-expand.c.jj 2018-04-16 20:35:14.663558320 +0200
+++ gcc/omp-expand.c2018-04-19 17:14:13.267424165 +0200
@@ -3501,7 +3501,12 @@ expand_omp_for_static_nochunk (struct om
   t = fold_convert (itype, s0);
   t = fold_build2 (MULT_EXPR, itype, t, step);
   if (POINTER_TYPE_P (type))
-t = fold_build_pointer_plus (n1, t);
+{
+  t = fold_build_pointer_plus (n1, t);
+  if (!POINTER_TYPE_P (TREE_TYPE (startvar))
+ && TYPE_PRECISION (TREE_TYPE (startvar)) > TYPE_PRECISION (type))
+   t = fold_convert (signed_type_for (type), t);
+}
   else
 t = fold_build2 (PLUS_EXPR, type, t, n1);
   t = fold_convert (TREE_TYPE (startvar), t);
@@ -3515,7 +3520,12 @@ expand_omp_for_static_nochunk (struct om
   t = fold_convert (itype, e0);
   t = fold_build2 (MULT_EXPR, itype, t, step);
   if (POINTER_TYPE_P (type))
-t = fold_build_pointer_plus (n1, t);
+{
+  t = fold_build_pointer_plus (n1, t);
+  if (!POINTER_TYPE_P (TREE_TYPE (startvar))
+ && TYPE_PRECISION (TREE_TYPE (startvar)) > TYPE_PRECISION (type))
+   t = fold_convert (signed_type_for (type), t);
+}
   else
 t = fold_build2 (PLUS_EXPR, type, t, n1);
   t = fold_convert (TREE_TYPE (startvar), t);
@@ -4000,7 +4010,12 @@ expand_omp_for_static_chunk (struct omp_
   t = fold_convert (itype, s0);
   t = fold_build2 (MULT_EXPR, itype, t, step);
   if (POINTER_TYPE_P (type))
-t = fold_build_pointer_plus (n1, t);
+{
+  t = fold_build_pointer_plus (n1, t);
+  if (!POINTER_TYPE_P (TREE_TYPE (startvar))
+ && TYPE_PRECISION (TREE_TYPE (startvar)) > TYPE_PRECISION (type))
+   t = fold_convert (signed_type_for (type), t);
+}
   else
 t = fold_build2 (PLUS_EXPR, type, t, n1);
   t = fold_convert (TREE_TYPE (startvar), t);
@@ -4014,7 +4029,12 @@ expand_omp_for_static_chunk (struct omp_
   t = fold_convert (itype, e0);
   t = fold_build2 (MULT_EXPR, itype, t, step);
   if (POINTER_TYPE_P (type))
-t = fold_build_pointer_plus (n1, t);
+{
+  t = fold_build_pointer_plus (n1, t);
+  if (!POINTER_TYPE_P (TREE_TYPE (startvar))
+ && TYPE_PRECISION (TREE_TYPE (startvar)) > TYPE_PRECISION (type))
+   t = fold_convert (signed_type_for (type), t);
+}
   else
 t = fold_build2 (PLUS_EXPR, type, t, n1);
   t = fold_convert (TREE_TYPE (startvar), t);

Though, I think it would be better to replace most of the signed_type_for
related to pointer types to unsigned_type_for.  Not something I have time for
right now though.

[Bug middle-end/85450] ICE: invalid types in nop conversion during GIMPLE pass: ompexp

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85450

Richard Biener  changed:

   What|Removed |Added

 Status|UNCONFIRMED |NEW
   Last reconfirmed||2018-04-18
 Ever confirmed|0   |1

--- Comment #1 from Richard Biener  ---
For for-3.c it is:

static void
expand_omp_for_static_nochunk (struct omp_region *region,
   struct omp_for_data *fd,
   gimple *inner_stmt)
{ 
...
  if (POINTER_TYPE_P (type))
t = fold_build_pointer_plus (n1, t); 
  else
t = fold_build2 (PLUS_EXPR, type, t, n1);
  t = fold_convert (TREE_TYPE (startvar), t);

where a fix could be to do

Index: omp-expand.c
===
--- omp-expand.c(revision 259457)
+++ omp-expand.c(working copy)
@@ -3501,7 +3501,10 @@ expand_omp_for_static_nochunk (struct om
   t = fold_convert (itype, s0);
   t = fold_build2 (MULT_EXPR, itype, t, step);
   if (POINTER_TYPE_P (type))
-t = fold_build_pointer_plus (n1, t);
+{
+  t = fold_build_pointer_plus (n1, t);
+  t = fold_convert (unsigned_type_for (TREE_TYPE (n1)), t);
+}
   else
 t = fold_build2 (PLUS_EXPR, type, t, n1);
   t = fold_convert (TREE_TYPE (startvar), t);

but there are likely a few "copies" of the code.  The for-3.c testcase
also exercises the following spot:

@@ -3515,7 +3518,10 @@ expand_omp_for_static_nochunk (struct om
   t = fold_convert (itype, e0);
   t = fold_build2 (MULT_EXPR, itype, t, step);
   if (POINTER_TYPE_P (type))
-t = fold_build_pointer_plus (n1, t);
+{
+  t = fold_build_pointer_plus (n1, t);
+  t = fold_convert (unsigned_type_for (TREE_TYPE (n1)), t);
+}
   else
 t = fold_build2 (PLUS_EXPR, type, t, n1);
   t = fold_convert (TREE_TYPE (startvar), t);

and that's still not enough to fix it fully...

Note that I think even for GENERIC using fold_convert the original way
is wrong.  Given fold_convert is "interesting" we might consider using
POINTERS_EXTEND_UNSIGNED there... (ok, I didn't suggest that).