$ gcc/xgcc -v -B gcc -O2 -S md.i Reading specs from gcc/specs Target: ia64-suse-linux Configured with: ../configure --host=ia64-suse-linux --enable-shared --enable-threads --enable-__cxa_atexit --with-system-zlib --with-system-libunwind Thread model: posix gcc version 4.0.0 20050219 (experimental) gcc/cc1 -fpreprocessed md.i -quiet -dumpbase md.i -auxbase md -O2 -version -o md.s GNU C version 4.0.0 20050219 (experimental) (ia64-suse-linux) compiled by GNU C version 4.0.0 20050219 (experimental). GGC heuristics: --param ggc-min-expand=30 --param ggc-min-heapsize=4096 md.i: In function ‘do_md’: md.i:3741: error: insn does not satisfy its constraints: (insn 5989 5988 5990 236 (set (reg:SI 262 p6) (reg:SI 17 r17)) 4 {*movsi_internal} (nil) (nil)) md.i:3741: internal compiler error: in reload_cse_simplify_operands, at postreload.c:391 Please submit a full bug report, with preprocessed source if appropriate. See <URL:http://gcc.gnu.org/bugs.html> for instructions. Broken by this change: 2005-02-06 Roger Sayle <roger@eyesopen.com> Andrew Pinski <pinskia@physics.uc.edu> Paolo Bonzini <paolo.bonzini@lu.unisi.ch> * simplify-rtx.c (simplify_relational_operation_1): Simplify (ne:SI (zero_extract:SI FOO (const_int 1) BAR) (const_int 0)) into just (zero_extract:SI FOO (const_int 1) BAR).
Created attachment 8236 [details] Testcase
This seems more likely a target bug or a latent bug.
This looks like a reload (or a target reload) issue: We have the following RTL: (insn 4097 4096 4098 236 (set (reg:BI 2598) (and:BI (ne:BI (reg/v:SI 477 [ bNotLastFrame ]) (const_int 0 [0x0])) (subreg:BI (reg:SI 2934) 0))) 66 {*cmpsi_and_0} (nil) (nil)) Note the subreg here. We exand it to: (insn 5994 4096 5995 236 (set (reg:SI 2 r2) (reg:SI 131 f3 [2934])) 4 {*movsi_internal} (nil) (nil)) (insn 5995 5994 5996 236 (set (reg:SI 262 p6) (reg:SI 2 r2)) 4 {*movsi_internal} (nil) (nil)) (insn 5996 5995 4097 236 (set (reg:SI 2 r2) (reg/v:SI 148 f20 [orig:477 bNotLastFrame ] [477])) 4 {*movsi_internal} (nil) (nil))(insn 4097 5996 4098 236 (set (reg:BI 262 p6 [2598]) (and:BI (ne:BI (reg:SI 2 r2) (const_int 0 [0x0])) (reg:BI 262 p6))) 66 {*cmpsi_and_0} (nil) (nil)) So either we have a reload issue or we have a target issue.
Subject: Bug 20097 CVSROOT: /cvs/gcc Module name: gcc Changes by: sayle@gcc.gnu.org 2005-02-23 18:20:53 Modified files: gcc : ChangeLog simplify-rtx.c Log message: PR target/20018 PR rtl-optimization/20097 * simplify-rtx.c (simplify_relational_operation_1): Avoid creating BImode SUBREGs of SImode registers which confuse the ia64 backend. Patches: http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.7571&r2=2.7572 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/simplify-rtx.c.diff?cvsroot=gcc&r1=1.229&r2=1.230
Fixed.
(In reply to comment #3) > This looks like a reload (or a target reload) issue: > We have the following RTL: > (insn 4097 4096 4098 236 (set (reg:BI 2598) > (and:BI (ne:BI (reg/v:SI 477 [ bNotLastFrame ]) > (const_int 0 [0x0])) > (subreg:BI (reg:SI 2934) 0))) 66 {*cmpsi_and_0} (nil) > (nil)) > > Note the subreg here. Right, that makes no sense, and neither does... > (insn 5995 5994 5996 236 (set (reg:SI 262 p6) > (reg:SI 2 r2)) 4 {*movsi_internal} (nil) (nil)) this. Will look into this if I have some time.
*** Bug 20018 has been marked as a duplicate of this bug. ***