https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111624
Bug ID: 111624
Summary: compile-time and memory-hog compiling RISC-V
insn-emit.cc
Product: gcc
Version: 13.2.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: rguenth at gcc dot gnu.org
Target Milestone: ---
When building a cross to riscv64-linux building insn-emit.cc (of riscv trunk)
takes ages and uses many gigabytes of memory at the default optimization
level of -O2 -g.
The code generated by genemit leaves to be desired, esp. when doing
handle_overloaded_code_for (switch statements would be much better,
esp. at -O0 I would guess). This puts a heavy toll on jump threading.
A host compiler built from the GCC 13 branch head takes
349.97user 24.74system 6:14.79elapsed 99%CPU (0avgtext+0avgdata
8777668maxresident)k
0inputs+1300720outputs (0major+2940488minor)pagefaults 0swaps
callgraph ipa passes : 23.03 ( 7%)
integration : 88.11 ( 26%)
tree CFG cleanup : 74.58 ( 22%)
tree CCP : 50.46 ( 15%)
TOTAL : 339.79
