Created attachment 38445 [details] Pre-processed source for stress-vecmath.c Hi! When building stress-ng on powerpc-linux-gnuspe, an internal compiler error occurs when compiling stress-vecmath.c: cc -g -O2 -fstack-protector-strong -Wformat -Werror=format-security -Wall -Wextra -DVERSION="0.06.01" -O2 -DHAVE_APPARMOR -DHAVE_KEYUTILS_H -DHAVE_XATTR_H -DHAVE_LIB_BSD -DHAVE_LIB_Z -DHAVE_LIB_CRYPT -DHAVE_LIB_RT -DHAVE_LIB_PTHREAD -DHAVE_FLOAT_DECIMAL -DHAVE_SECCOMP_H -DHAVE_LIB_AIO -DHAVE_SYS_CAP_H -c -o stress-vecmath.o stress-vecmath.c stress-vecmath.c: In function 'stress_vecmath': stress-vecmath.c:49:4: internal compiler error: in expand_expr_real_2, at expr.c:9348 a ^= c; \ ^ stress-vecmath.c:114:4: note: in expansion of macro 'OPS' OPS(a64, b64, c64, s64); ^ 0x103166eb expand_expr_real_2(separate_ops*, rtx_def*, machine_mode, expand_modifier) ../../src/gcc/expr.c:9348 0x1021f24b expand_gimple_stmt_1 ../../src/gcc/cfgexpand.c:3462 0x1021f24b expand_gimple_stmt ../../src/gcc/cfgexpand.c:3522 0x10222ed3 expand_gimple_basic_block ../../src/gcc/cfgexpand.c:5534 0x10224f33 execute ../../src/gcc/cfgexpand.c:6152 Please submit a full bug report, with preprocessed source if appropriate. Please include the complete backtrace with any bug report. See <file:///usr/share/doc/gcc-5/README.Bugs> for instructions. Preprocessed source stored into /tmp/ccRHm88P.out file, please attach this to your bugreport. This has been failing for quite some time. However, it seems it does not seem to be related to gcc directly as both the last successful build [1] and the last unsuccessful build [2] used both the same revision of gcc-4.9.2: r218987. Attaching the pre-processed source from the last build with gcc-5. Adrian > [1] https://buildd.debian.org/status/fetch.php?pkg=stress-ng&arch=powerpcspe&ver=0.03.12-1&stamp=1422673290 > [2] https://buildd.debian.org/status/fetch.php?pkg=stress-ng&arch=powerpcspe&ver=0.03.13-1&stamp=1422840779
https://buildd.debian.org/status/fetch.php?pkg=stress-ng&arch=powerpcspe&ver=0.06.01-1&stamp=1462574245
I can reproduce it w/ gcc 5.4.0, but cannot w/ 7.0.0-alpha20170205 snapshot.
(In reply to Arseny Solokha from comment #2) > I can reproduce it w/ gcc 5.4.0, but cannot w/ 7.0.0-alpha20170205 snapshot. We also haven't seen it in Debian since last May, see: > https://buildd.debian.org/status/logs.php?pkg=stress-ng&arch=powerpcspe stress-ng has been building fine ever since.
int mw; int j3 (int cn) { const int l7 = 0; int wz; for (wz = l7; wz < l7 + 3; ++wz) while (mw != 0) { cn ^= -(wz == l7); ++mw; } return cn; } actually leads to ICE w/ -O2 (-O3, -Ofast) -ftree-vectorize all the way from gcc-8.0.0-alpha20170507 snapshot down to 4.9: % powerpc-e500v2-linux-gnuspe-gcc-8.0.0-alpha20170507 -O2 -ftree-vectorize -c ndknofut.c ndknofut.c: In function 'j3': ndknofut.c:12:12: internal compiler error: in expand_expr_real_2, at expr.c:9069 cn ^= -(wz == l7); ~~~^~~~~~~~~~~~~~ 0x8a1b2a expand_expr_real_2(separate_ops*, rtx_def*, machine_mode, expand_modifier) /var/tmp/portage/cross-powerpc-e500v2-linux-gnuspe/gcc-8.0.0_alpha20170507/work/gcc-8-20170507/gcc/expr.c:9069 0x74e4de expand_gimple_stmt_1 /var/tmp/portage/cross-powerpc-e500v2-linux-gnuspe/gcc-8.0.0_alpha20170507/work/gcc-8-20170507/gcc/cfgexpand.c:3676 0x74e4de expand_gimple_stmt /var/tmp/portage/cross-powerpc-e500v2-linux-gnuspe/gcc-8.0.0_alpha20170507/work/gcc-8-20170507/gcc/cfgexpand.c:3737 0x7510e7 expand_gimple_basic_block /var/tmp/portage/cross-powerpc-e500v2-linux-gnuspe/gcc-8.0.0_alpha20170507/work/gcc-8-20170507/gcc/cfgexpand.c:5744 0x7576df execute /var/tmp/portage/cross-powerpc-e500v2-linux-gnuspe/gcc-8.0.0_alpha20170507/work/gcc-8-20170507/gcc/cfgexpand.c:6357
The powerpcspe backend has been deprecated in GCC 8 and removed during GCC 9 development. See corresponding mailing list threads[1,2,3] for details. [1] https://gcc.gnu.org/legacy-ml/gcc/2018-04/msg00102.html [2] https://gcc.gnu.org/legacy-ml/gcc-patches/2018-12/msg00123.html [3] https://gcc.gnu.org/pipermail/gcc/2020-May/232342.html