a JPEG image processing library. Exercised with two quick running workloads. The first runs on one thread, the second on “all” threads.
Topdown profile shows the single-threaded workload as having a higher retirement rate and some backend stalls. The parallel one is flipped.
AMD metrics show not much floating point, some L2 access including misses.
elapsed 214.087
on_cpu 0.230 # 3.69 / 16 cores
utime 572.235
stime 217.040
nvcsw 46264 # 80.71%
nivcsw 11056 # 19.29%
inblock 88 # 0.41/sec
onblock 293008 # 1368.64/sec
cpu-clock 789175841055 # 789.176 seconds
task-clock 789261200257 # 789.261 seconds
page faults 78020185 # 98852.173/sec
context switches 58208 # 73.750/sec
cpu migrations 7653 # 9.696/sec
major page faults 4 # 0.005/sec
minor page faults 78020181 # 98852.168/sec
alignment faults 0 # 0.000/sec
emulation faults 0 # 0.000/sec
branches 322457994967 # 72.656 branches per 1000 inst
branch misses 15346807047 # 4.76% branch miss
conditional 231733673746 # 52.214 conditional branches per 1000 inst
indirect 6110886568 # 1.377 indirect branches per 1000 inst
cpu-cycles 3143196687512 # 0.92 GHz
instructions 4405132309118 # 1.40 IPC
slots 6291991187694 #
retiring 1536387710543 # 24.4% (29.6%)
-- ucode 3228779195 # 0.1%
-- fastpath 1533158931348 # 24.4%
frontend 929644428644 # 14.8% (17.9%)
-- latency 707692046958 # 11.2%
-- bandwidth 221952381686 # 3.5%
backend 2565494368112 # 40.8% (49.5%)
-- cpu 956203684287 # 15.2%
-- memory 1609290683825 # 25.6%
speculation 150701191697 # 2.4% ( 2.9%)
-- branch mispredict 145671010991 # 2.3%
-- pipeline restart 5030180706 # 0.1%
smt-contention 1109755007091 # 17.6% ( 0.0%)
cpu-cycles 3147769160356 # 0.91 GHz
instructions 4396937620175 # 1.40 IPC
instructions 1480153486902 # 45.732 l2 access per 1000 inst
l2 hit from l1 54965304511 # 11.34% l2 miss
l2 miss from l1 3863029248 #
l2 hit from l2 pf 8912497384 #
l3 hit from l2 pf 2337614534 #
l3 miss from l2 pf 1474996298 #
instructions 1468633262644 # 47.498 float per 1000 inst
float 512 57 # 0.000 AVX-512 per 1000 inst
float 256 602 # 0.000 AVX-256 per 1000 inst
float 128 69757460749 # 47.498 AVX-128 per 1000 inst
float MMX 0 # 0.000 MMX per 1000 inst
float scalar 0 # 0.000 scalar per 1000 inst
instructions 4410783289171 #
opcache 639131229094 # 144.902 opcache per 1000 inst
opcache miss 75966023342 # 11.9% opcache miss rate
l1 dTLB miss 15719138380 # 3.564 L1 dTLB per 1000 inst
l2 dTLB miss 696242527 # 0.158 L2 dTLB per 1000 inst
instructions 4411366720558 #
icache 147700356249 # 33.482 icache per 1000 inst
icache miss 9449915162 # 6.4% icache miss rate
l1 iTLB miss 30947776 # 0.007 L1 iTLB per 1000 inst
l2 iTLB miss 0 # 0.000 L2 iTLB per 1000 inst
tlb flush 23457 # 0.000 TLB flush per 1000 inst
Intel metrics show backend stalls are more at L1/L2 levels than all the way out to dram
elapsed 229.959
on_cpu 0.221 # 3.54 / 16 cores
utime 655.417
stime 157.900
nvcsw 47596 # 81.51%
nivcsw 10796 # 18.49%
inblock 248224 # 1079.43/sec
onblock 281728 # 1225.12/sec
cpu-clock 812664634134 # 812.665 seconds
task-clock 812731141309 # 812.731 seconds
page faults 64066346 # 78828.462/sec
context switches 59368 # 73.048/sec
cpu migrations 11485 # 14.131/sec
major page faults 1276 # 1.570/sec
minor page faults 64065070 # 78826.892/sec
alignment faults 0 # 0.000/sec
emulation faults 0 # 0.000/sec
branches 289467025331 # 66.628 branches per 1000 inst
branch misses 4791047711 # 1.66% branch miss
conditional 289467037075 # 66.628 conditional branches per 1000 inst
indirect 46740820621 # 10.759 indirect branches per 1000 inst
slots 7250543859452 #
retiring 3182672569517 # 43.9% (43.9%)
-- ucode 284462783358 # 3.9%
-- fastpath 2898209786159 # 40.0%
frontend 798789567358 # 11.0% (11.0%)
-- latency 466544490290 # 6.4%
-- bandwidth 332245077068 # 4.6%
backend 2613200296738 # 36.0% (36.0%)
-- cpu 1445575001374 # 19.9%
-- memory 1167625295364 # 16.1%
speculation 713440337672 # 9.8% ( 9.8%)
-- branch mispredict 642636577796 # 8.9%
-- pipeline restart 70803759876 # 1.0%
smt-contention 0 # 0.0% ( 0.0%)
cpu-cycles 3015898317055 # 0.82 GHz
instructions 5329185911432 # 1.77 IPC
l2 access 103439376307 # 33.427 l2 access per 1000 inst
l2 miss 23259806947 # 22.49% l2 miss
cpu-cycles 1737674083545 # 27.5% memory latency
load stalls 410906786285 # 11.1% l1 bound
l1 miss 217693607649 # 8.2% l2 bound
l2 miss 74926260792 # 2.1% l3 bound
l3 miss 38350978722 # 2.2% dram bound
store_stalls 66085442220 # 3.8% store bound