wrf 29x slowdown vs native due to 185M uninits

[derekbruening]

From [email protected] on April 17, 2013 23:32:49

on ancalagon:

cd /extsw/spec2006/benchspec/CPU2006/481.wrf/run/run_base_ref_amd64-m32-gcc42-nn.0000

/usr/bin/time /work/drmemory/git/build_x86_rel/bin/drmemory.pl -replace_malloc -dr_ops "-msgbox_mask 12 -prof_pcs" -pause_at_assert -dr /work/dr/git/exports -- ./wrf_base.amd64-m32-gcc42-nn > rsl.out.0000

Dr.M ERRORS FOUND:
Dr.M 1 unique, 2 total unaddressable access(es)
Dr.M 60 unique, 185522998 total uninitialized access(es)
Dr.M 0 unique, 0 total invalid heap argument(s)
Dr.M 0 unique, 0 total warning(s)
Dr.M 2 unique, 2 total, 331000 byte(s) of leak(s)
Dr.M 8 unique, 8 total, 188000 byte(s) of possible leak(s)
Dr.M ERRORS IGNORED:
Dr.M 2268 still-reachable allocation(s)
Dr.M (re-run with "-show_reachable" for details)
Dr.M Details: /work/drmemory/git/build_x86_rel/logs/DrMemory-wrf_base.amd64-m32-gcc42-nn.18787.000/results.txt
22675.12user 23.30system 6:19:17elapsed 99%CPU (0avgtext+0avgdata 974348maxresident)k

head pcs_; IFS=$'\n'; for i in grep '^pc._UNKNOWN' pcs\* | sed 's/pc=//' | sed 's/#=//' | awk '{printf "%s %s\n",$2, $1}'; do addr2line -f -e /work/drmemory/git/build_x86_rel/bin/bin32/release/libdrmemory.soecho $i | awk '{print $2}'| xargs echo $i; done | sort -n > RES-basename pcs_; for i in RES-pcsamples.0.; do echo $i; cat $i | perl -e 'while(<>){if(/^(\d+)\s+\S+\s+(\S+)/){$hits{$2}+=$1;}else{die"bad:$";}}foreach $k (sort {$hits{$a} <=> $hits{$b}} keys %hits){printf "%5d %s\n", $hits{$k}, $k;}' | tail -12; done

ITIMER distribution (2267893):
0.0% of time in APPLICATION (3)
9.4% of time in INTERPRETER (212391)
0.3% of time in DISPATCH (6606)
2.7% of time in SYSCALL HANDLER (60259)
0.2% of time in INDIRECT BRANCH LOOKUP (5102)
11.4% of time in FRAGMENT CACHE (258170)
76.1% of time in UNKNOWN (1725362)
RES-pcsamples.0.18787.html
19029 is_retaddr
32965 syscall_get_param
44365 report_in_suppressed_module
48299 get_shadow_table
102790 shadow_get_dword
106817 address_to_frame
117464 bitmapx2_byte
127081 print_address_common
164888 callstack_module_unload
419084 text_contains_any_string

looks like it's the uninits!

DUPLICATE ERROR COUNTS:
Error # 12: 89,552,910
Error # 14: 67,681,030
Error # 18: 2888730
Error # 20: 2031787
Error # 30: 11,865,070
Error # 31: 9331328

grep 'Error # ' /work/drmemory/git/build_x86_rel/logs/DrMemory-wrf_base.amd64-m32-gcc42-nn.18787.000/results.txt | awk '{s+=$NF} END {print s}'
185,522,985

top ones:

Error #12: UNINITIALIZED READ: reading 0xfffbad90-0xfffbad94 4 byte(s)
#0 __module_sf_noahlsm_MOD_sstep
#1 __module_sf_noahlsm_MOD_smflx
#2 __module_sf_noahlsm_MOD_nopac
#3 __module_sf_noahlsm_MOD_sflx
#4 __module_sf_noahlsm_MOD_lsm
#5 _module_surface_driver_MOD_surface_driver
#6 solve_em
Note: @0:00:53.532 in thread 18787
Note: instruction: fld 0xfffffffc(%ebx,%eax,4) -> %st0

Error #14: UNINITIALIZED READ: reading 0xfffbae30-0xfffbae34 4 byte(s)
#0 __module_sf_noahlsm_MOD_hstep
#1 __module_sf_noahlsm_MOD_shflx
#2 __module_sf_noahlsm_MOD_nopac
#3 __module_sf_noahlsm_MOD_sflx
#4 __module_sf_noahlsm_MOD_lsm
#5 _module_surface_driver_MOD_surface_driver
#6 solve_em
Note: @0:00:53.555 in thread 18787
Note: instruction: fld 0xfffffffc(%ebx,%eax,4) -> %st0

Error #18: UNINITIALIZED READ: reading 0xfffbad94-0xfffbad98 4 byte(s)
#0 __module_sf_noahlsm_MOD_sstep
#1 __module_sf_noahlsm_MOD_smflx
#2 _module_surface_driver_MOD_surface_driver
#3 solve_em
Note: @0:00:53.594 in thread 18787
Note: instruction: fld 0xfffffffc(%ebx,%eax,4) -> %st0

Error #20: UNINITIALIZED READ: reading 0xfffbae34-0xfffbae38 4 byte(s)
#0 __module_sf_noahlsm_MOD_hstep
#1 __module_sf_noahlsm_MOD_shflx
#2 __module_sf_noahlsm_MOD_sflx
#3 __module_sf_noahlsm_MOD_lsm
#4 _module_surface_driver_MOD_surface_driver
#5 solve_em
Note: @0:00:53.611 in thread 18787
Note: instruction: fld 0xfffffffc(%ebx,%eax,4) -> %st0

Error #30: UNINITIALIZED READ: reading 0xfffb9268-0xfffb926c 4 byte(s)
#0 __module_sf_noahlsm_MOD_hstep
#1 header_from_ptr
#2 __module_sf_noahlsm_MOD_shflx
#3 header_from_ptr
#4 next_chunk_forward
#5 __module_sf_noahlsm_MOD_nopac
#6 __module_sf_noahlsm_MOD_sflx
#7 __module_sf_noahlsm_MOD_lsm
#8 _module_surface_driver_MOD_surface_driver
#9 solve_em
Note: @0:03:13.993 in thread 18787
Note: instruction: fld 0xfffffffc(%ebx,%eax,4) -> %st0

Error #31: UNINITIALIZED READ: reading 0xfffb9260-0xfffb9264 4 byte(s)
#0 __module_sf_noahlsm_MOD_hstep
#1 __module_sf_noahlsm_MOD_shflx
#2 header_from_ptr
#3 next_chunk_forward
#4 __module_sf_noahlsm_MOD_nopac
#5 __module_sf_noahlsm_MOD_sflx
#6 __module_sf_noahlsm_MOD_lsm
#7 _module_surface_driver_MOD_surface_driver
#8 solve_em
Note: @0:03:14.005 in thread 18787
Note: instruction: fld 0xfffffffc(%ebx,%eax,4) -> %st0

they're all floats: is this issue #471 ?

valgrind takes 2:54:24 and reports about 100K uninits: none of the fld ones here are among them.

Original issue: http://code.google.com/p/drmemory/issues/detail?id=1188