Created attachment 50609 [details] Fix for this regression GCC10 wrongly recognizes CPU as POWER9 if -misel is specified, even when it is configured for 32-bit processors: % cat test.c int main(void) { return 0; } % gcc10 -misel -S test.c -o test10.S % gcc9 -misel -S test.c -o test9.S % diff -u test9.S test10.S --- test.9 2021-04-16 10:53:47.116917225 +0900 +++ test.10 2021-04-16 10:53:38.075282430 +0900 @@ -1,5 +1,5 @@ .file "test.c" - .machine ppc + .machine power9 .section ".text" .align 2 .globl main ...(snip)... This is because rs6000_machine_from_flags() assumes ISEL instructions are supported only by POWER9 and successors. However, ISEL is also implemented for 32-bit processors like e500. Please apply the attached patch to fix the problem. ---- % gcc10 -v Using built-in specs. COLLECT_GCC=/build/tools.broken/bin/powerpc--netbsd-gcc COLLECT_LTO_WRAPPER=/build/tools.broken/libexec/gcc/powerpc--netbsd/10.3.0/lto-wrapper Target: powerpc--netbsd Configured with: /usr/src/tools/gcc/../../external/gpl3/gcc/dist/configure --target=powerpc--netbsd --enable-long-long --enable-threads --with-bugurl=http://www.NetBSD.org/support/send-pr.html --with-pkgversion='NetBSD nb1 20210411' --with-system-zlib --without-isl --enable-__cxa_atexit --enable-libstdcxx-time=rt --enable-libstdcxx-threads --with-diagnostics-color=auto-if-env --with-default-libstdcxx-abi=new --with-sysroot=/build/dest/evbppc-powerpc --with-mpc=/build/tools.broken --with-mpfr=/build/tools.broken --with-gmp=/build/tools.broken --disable-nls --disable-multilib --program-transform-name='s,^,powerpc--netbsd-,' --enable-languages='c c++ objc' --prefix=/build/tools.broken Thread model: posix Supported LTO compression algorithms: zlib gcc version 10.3.0 (NetBSD nb1 20210411) ---- % gcc9 -v Using built-in specs. COLLECT_GCC=/build/tools.9/bin/powerpc--netbsd-gcc COLLECT_LTO_WRAPPER=/build/tools.9/libexec/gcc/powerpc--netbsd/9.3.0/lto-wrapper Target: powerpc--netbsd Configured with: /usr/src/tools/gcc/../../external/gpl3/gcc.old/dist/configure --target=powerpc--netbsd --enable-long-long --enable-threads --with-bugurl=http://www.NetBSD.org/support/send-pr.html --with-pkgversion='NetBSD nb1 20200907' --with-system-zlib --without-isl --enable-__cxa_atexit --enable-libstdcxx-time=rt --enable-libstdcxx-threads --with-diagnostics-color=auto-if-env --with-default-libstdcxx-abi=new --with-sysroot=/build/dest/evbppc-powerpc --with-mpc=/build/tools.9 --with-mpfr=/build/tools.9 --with-gmp=/build/tools.9 --disable-nls --disable-multilib --program-transform-name='s,^,powerpc--netbsd-,' --enable-languages='c c++ objc' --prefix=/build/tools.9 Thread model: posix gcc version 9.3.0 (NetBSD nb1 20200907)
e500 support had been moved to the powerpcspe target; so assuming power9 for -misel is correct. e500mc support is still there though.
e500mc also supports isel. See its reference manual (Sec. 3.4.5.3): https://www.nxp.com/docs/en/reference-manual/E500MCRM.pdf or source code of binutils/gas: https://sourceware.org/git/?p=binutils-gdb.git;a=blob;f=gas/config/tc-ppc.c;h=c719b408b0eb59357f6d7e7022a9d041e181c318;hb=HEAD#l3973
Indeed, several rs6000-cpus.def entries have MASK_ISEL: RS6000_CPU ("8540", PROCESSOR_PPC8540, MASK_STRICT_ALIGN | MASK_ISEL) RS6000_CPU ("8548", PROCESSOR_PPC8548, MASK_STRICT_ALIGN | MASK_ISEL) RS6000_CPU ("e500mc", PROCESSOR_PPCE500MC, MASK_PPC_GFXOPT | MASK_ISEL) RS6000_CPU ("e500mc64", PROCESSOR_PPCE500MC64, MASK_POWERPC64 | MASK_PPC_GFXOPT | MASK_ISEL) RS6000_CPU ("e5500", PROCESSOR_PPCE5500, MASK_POWERPC64 | MASK_PPC_GFXOPT | MASK_ISEL) RS6000_CPU ("e6500", PROCESSOR_PPCE6500, POWERPC_7400_MASK | MASK_POWERPC64 | MASK_MFCRF | MASK_ISEL) So perhaps we should consider MASK_ISEL as power9 thing only for flags which include power7-ish or later ISAs? --- rs6000.c.jj4 2021-04-09 10:18:15.582266372 +0200 +++ rs6000.c 2021-04-19 16:30:07.033208577 +0200 @@ -5766,6 +5766,9 @@ rs6000_machine_from_flags (void) /* Disable the flags that should never influence the .machine selection. */ flags &= ~(OPTION_MASK_PPC_GFXOPT | OPTION_MASK_PPC_GPOPT); + if ((flags & (ISA_3_1_MASKS_SERVER + & ~(ISA_2_5_MASKS_SERVER | MASK_ISEL))) == 0) + flags &= ~MASK_ISEL; if ((flags & (ISA_3_1_MASKS_SERVER & ~ISA_3_0_MASKS_SERVER)) != 0) return "power10";
(In reply to Andrew Pinski from comment #1) > e500 support had been moved to the powerpcspe target; so assuming power9 for > -misel is correct. > > e500mc support is still there though. There never *was* separate e500 support in GCC!
Created attachment 50629 [details] Proposed simpler patch A simpler patch. I'll commit this later today (if no one stops me).
LGTM.
Thank you for discussion. The proposed patch works fine for me (except for missing ``;'' at the end of the line).
The master branch has been updated by Segher Boessenkool <segher@gcc.gnu.org>: https://gcc.gnu.org/g:6156df483fa50a08f561b6c248819f2992aa380d commit r12-9-g6156df483fa50a08f561b6c248819f2992aa380d Author: Segher Boessenkool <segher@kernel.crashing.org> Date: Tue Apr 20 12:00:50 2021 +0000 rs6000: Fix cpu selection w/ isel (PR100108) There are various non-IBM CPUs with isel as well, so it is easiest if we just don't consider that flag here (it is not needed). 2021-04-20 Segher Boessenkool <segher@kernel.crashing.org> PR target/100108 * config/rs6000/rs6000.c (rs6000_machine_from_flags): Do not consider OPTION_MASK_ISEL.
The releases/gcc-11 branch has been updated by Segher Boessenkool <segher@gcc.gnu.org>: https://gcc.gnu.org/g:8bd3b0c11cb08f63d2ca41a21596d16635626661 commit r11-8264-g8bd3b0c11cb08f63d2ca41a21596d16635626661 Author: Segher Boessenkool <segher@kernel.crashing.org> Date: Tue Apr 20 12:00:50 2021 +0000 rs6000: Fix cpu selection w/ isel (PR100108) There are various non-IBM CPUs with isel as well, so it is easiest if we just don't consider that flag here (it is not needed). 2021-04-20 Segher Boessenkool <segher@kernel.crashing.org> PR target/100108 * config/rs6000/rs6000.c (rs6000_machine_from_flags): Do not consider OPTION_MASK_ISEL. (cherry picked from commit 6156df483fa50a08f561b6c248819f2992aa380d)
Fixed on trunk and 11. Will do 10 in a week or so.
Segher, let me thank you again!
The releases/gcc-10 branch has been updated by Segher Boessenkool <segher@gcc.gnu.org>: https://gcc.gnu.org/g:5f665c1ca452673e9812cd92b07bd31441c0ac5b commit r10-9814-g5f665c1ca452673e9812cd92b07bd31441c0ac5b Author: Segher Boessenkool <segher@kernel.crashing.org> Date: Tue Apr 20 12:00:50 2021 +0000 rs6000: Fix cpu selection w/ isel (PR100108) There are various non-IBM CPUs with isel as well, so it is easiest if we just don't consider that flag here (it is not needed). 2021-04-20 Segher Boessenkool <segher@kernel.crashing.org> PR target/100108 * config/rs6000/rs6000.c (rs6000_machine_from_flags): Do not consider OPTION_MASK_ISEL. (cherry picked from commit 6156df483fa50a08f561b6c248819f2992aa380d)
Fixed on GCC 10 and later. Closing.