Linear solver of pressure Poisson using a point-Jacobi method. This is one of fed tests where Intel CPU performs substantially better (6046) than AMD CPU (3966). It is also a case where AMD has a variable rate. This looks like a single-threaded test.
Topdown profile shows high backend stalls, both CPU and memory. It also suggests multiple rounds of tests to find a stable point. Frontend stalls are some of the smallest of all benchmarks. Perhaps interesting to see factors here like code/data footprint sizes?
AMD metrics show a floating point program with some L2 access and misses.
elapsed 562.793
on_cpu 0.056 # 0.90 / 16 cores
utime 506.784
stime 1.067
nvcsw 2113 # 51.46%
nivcsw 1993 # 48.54%
inblock 0 # 0.00/sec
onblock 12800 # 22.74/sec
cpu-clock 507956077142 # 507.956 seconds
task-clock 507965279796 # 507.965 seconds
page faults 240790 # 474.028/sec
context switches 6723 # 13.235/sec
cpu migrations 325 # 0.640/sec
major page faults 2 # 0.004/sec
minor page faults 240788 # 474.025/sec
alignment faults 0 # 0.000/sec
emulation faults 0 # 0.000/sec
branches 78486493682 # 20.748 branches per 1000 inst
branch misses 624071097 # 0.80% branch miss
conditional 77692509960 # 20.538 conditional branches per 1000 inst
indirect 47753355 # 0.013 indirect branches per 1000 inst
cpu-cycles 2940577091356 # 0.26 GHz
instructions 4771889380353 # 1.62 IPC
slots 5884521661530 #
retiring 1586483683880 # 27.0% (27.0%)
-- ucode 14146366 # 0.0%
-- fastpath 1586469537514 # 27.0%
frontend 55781933015 # 0.9% ( 0.9%) low
-- latency 31688469264 # 0.5%
-- bandwidth 24093463751 # 0.4%
backend 4231018348818 # 71.9% (71.9%) high
-- cpu 2277772735762 # 38.7%
-- memory 1953245613056 # 33.2%
speculation 11018268385 # 0.2% ( 0.2%) low
-- branch mispredict 10153156053 # 0.2%
-- pipeline restart 865112332 # 0.0%
smt-contention 218961520 # 0.0% ( 0.0%)
cpu-cycles 2938472632454 # 0.26 GHz
instructions 4629990328053 # 1.58 IPC
instructions 1544094148954 # 88.762 l2 access per 1000 inst
l2 hit from l1 80167148026 # 20.63% l2 miss
l2 miss from l1 1331488191 #
l2 hit from l2 pf 29950113188 #
l3 hit from l2 pf 13061396001 #
l3 miss from l2 pf 13878019099 #
instructions 1543498098310 # 184.685 float per 1000 inst
float 512 69 # 0.000 AVX-512 per 1000 inst
float 256 518 # 0.000 AVX-256 per 1000 inst
float 128 285061640703 # 184.685 AVX-128 per 1000 inst
float MMX 0 # 0.000 MMX per 1000 inst
float scalar 0 # 0.000 scalar per 1000 inst
Intel metrics
elapsed 168.565
on_cpu 0.057 # 0.92 / 16 cores
utime 154.165
stime 0.438
nvcsw 1887 # 74.50%
nivcsw 646 # 25.50%
inblock 0 # 0.00/sec
onblock 1312 # 7.78/sec
cpu-clock 154630765914 # 154.631 seconds
task-clock 154634206479 # 154.634 seconds
page faults 159168 # 1029.319/sec
context switches 3205 # 20.726/sec
cpu migrations 321 # 2.076/sec
major page faults 0 # 0.000/sec
minor page faults 159168 # 1029.319/sec
alignment faults 0 # 0.000/sec
emulation faults 0 # 0.000/sec
branches 35913794453 # 21.089 branches per 1000 inst
branch misses 88509738 # 0.25% branch miss
conditional 35913806389 # 21.089 conditional branches per 1000 inst
indirect 59414046 # 0.035 indirect branches per 1000 inst
slots 3725864821436 #
retiring 2390025479822 # 64.1% (64.1%) high
-- ucode 620664419824 # 16.7%
-- fastpath 1769361059998 # 47.5%
frontend 553782693331 # 14.9% (14.9%)
-- latency 17755344084 # 0.5%
-- bandwidth 536027349247 # 14.4%
backend 769551179018 # 20.7% (20.7%)
-- cpu 500685850178 # 13.4%
-- memory 268865328840 # 7.2%
speculation 17170761530 # 0.5% ( 0.5%) low
-- branch mispredict 12411755919 # 0.3%
-- pipeline restart 4759005611 # 0.1%
smt-contention 0 # 0.0% ( 0.0%)
cpu-cycles 648805684715 # 0.22 GHz
instructions 1890731557963 # 2.91 IPC
l2 access 101309940231 # 53.657 l2 access per 1000 inst
l2 miss 59683326414 # 58.91% l2 miss
Process overview suggests computation in the himenobmtxpa application
372 processes
12 himenobmtxpa 506.42 0.11
68 clinfo 16.48 6.32
38 vulkaninfo 0.76 1.53
6 glxinfo:gdrv0 0.11 0.07
6 glxinfo:gl0 0.11 0.06
4 vulkani:disk$0 0.08 0.17
6 php 0.07 0.15
2 glxinfo 0.05 0.04
2 glxinfo:cs0 0.05 0.04
2 glxinfo:disk$0 0.05 0.03
2 glxinfo:sh0 0.05 0.03
2 glxinfo:shlo0 0.05 0.03
2 llvmpipe-0 0.04 0.09
2 llvmpipe-1 0.04 0.09
2 llvmpipe-10 0.04 0.09
2 llvmpipe-11 0.04 0.09
2 llvmpipe-12 0.04 0.09
2 llvmpipe-13 0.04 0.09
2 llvmpipe-14 0.04 0.09
2 llvmpipe-15 0.04 0.09
2 llvmpipe-2 0.04 0.09
2 llvmpipe-3 0.04 0.09
2 llvmpipe-5 0.04 0.09
2 llvmpipe-6 0.04 0.09
2 llvmpipe-7 0.04 0.09
2 llvmpipe-8 0.04 0.09
2 llvmpipe-9 0.04 0.09
2 llvmpipe-4 0.04 0.08
6 clang 0.03 0.09
3 rocminfo 0.00 0.03
1 lspci 0.00 0.02
1 ps 0.00 0.01
82 sh 0.00 0.00
13 gcc 0.00 0.00
12 gsettings 0.00 0.00
12 himeno 0.00 0.00
8 stat 0.00 0.00
8 systemd-detect- 0.00 0.00
6 llvm-link 0.00 0.00
5 phoronix-test-s 0.00 0.00
2 cc 0.00 0.00
2 dconf worker 0.00 0.00
2 gmain 0.00 0.00
2 lscpu 0.00 0.00
2 uname 0.00 0.00
2 which 0.00 0.00
2 xset 0.00 0.00
1 date 0.00 0.00
1 dirname 0.00 0.00
1 dmesg 0.00 0.00
1 dmidecode 0.00 0.00
1 grep 0.00 0.00
1 ifconfig 0.00 0.00
1 ip 0.00 0.00
1 lsmod 0.00 0.00
1 mktemp 0.00 0.00
1 qdbus 0.00 0.00
1 readlink 0.00 0.00
1 realpath 0.00 0.00
1 sed 0.00 0.00
1 sort 0.00 0.00
1 stty 0.00 0.00
1 systemctl 0.00 0.00
1 template.sh 0.00 0.00
1 wc 0.00 0.00
1 xrandr 0.00 0.00
0 processes running
47 maximum processes
Code sizes for the application are small
mev@augusta:~/.phoronix-test-suite/installed-tests/pts/himeno-1.3.0$ size himenobmtxpa
text data bss dec hex filename
8399 676 240 9315 2463 himenobmtxpaEssentially repeated calls to the following single Jacobi function. Probably also a good case for -O3 vectorization options.
float
jacobi(int nn, Matrix* a,Matrix* b,Matrix* c,
Matrix* p,Matrix* bnd,Matrix* wrk1,Matrix* wrk2)
{
int i,j,k,n,imax,jmax,kmax;
float gosa,s0,ss;
imax= p->mrows-1;
jmax= p->mcols-1;
kmax= p->mdeps-1;
for(n=0 ; n<nn ; n++){
gosa = 0.0;
for(i=1 ; i<imax; i++)
for(j=1 ; j<jmax ; j++)
for(k=1 ; k<kmax ; k++){
s0= a->m[0][i][j][k]*p->m[0][i+1][j][k]
+ a->m[1][i][j][k]*p->m[0][i][j+1][k]
+ a->m[2][i][j][k]*p->m[0][i][j][k+1]
+ b->m[0][i][j][k]
*( p->m[0][i+1][j+1][k] - p->m[0][i+1][j-1][k]
- p->m[0][i-1][j+1][k] + p->m[0][i-1][j-1][k] )
+ b->m[1][i][j][k]
*( p->m[0][i][j+1][k+1] - p->m[0][i][j-1][k+1]
- p->m[0][i][j+1][k-1] + p->m[0][i][j-1][k-1] )
+ b->m[2][i][j][k]
*( p->m[0][i+1][j][k+1] - p->m[0][i-1][j][k+1]
- p->m[0][i+1][j][k-1] + p->m[0][i-1][j][k-1] )
+ c->m[0][i][j][k] * p->m[0][i-1][j][ k]
+ c->m[1][i][j][k] * p->m[0][i][j-1][k]
+ c->m[2][i][j][k] * p->m[0][i][j][k-1]
+ wrk1->m[0][i][j][k];
ss= (s0*a->m[3][i][j][k] - p->m[0][i][j][k])*bnd->m[0][i][j][k];
gosa+= ss*ss;
wrk2->m[0][i][j][k]= p->m[0][i][j][k] + omega*ss;
}
for(i=1 ; i<imax ; i++)
for(j=1 ; j<jmax ; j++)
for(k=1 ; k<kmax ; k++)
p->m[0][i][j][k]= wrk2->m[0][i][j][k];
} /* end n loop */
return(gosa);
}
Computation is repeated runs of this application
1175981) himeno cpu=14 start=8.11 finish=49.58
1175982) himenobmtxpa cpu=7 start=8.11 finish=49.58
1175985) himeno cpu=14 start=53.59 finish=95.91
1175986) himenobmtxpa cpu=15 start=53.59 finish=95.91
1175987) himeno cpu=6 start=99.91 finish=142.20
1175988) himenobmtxpa cpu=15 start=99.92 finish=142.20
1175989) himeno cpu=14 start=146.21 finish=188.63
1175990) himenobmtxpa cpu=7 start=146.21 finish=188.63
1175992) himeno cpu=6 start=192.63 finish=235.00
1175993) himenobmtxpa cpu=15 start=192.63 finish=235.00
1175994) himeno cpu=6 start=239.01 finish=281.36
1175995) himenobmtxpa cpu=15 start=239.01 finish=281.35
1175998) himeno cpu=6 start=285.36 finish=327.68
1175999) himenobmtxpa cpu=15 start=285.36 finish=327.68
1176001) himeno cpu=11 start=331.68 finish=373.93
1176002) himenobmtxpa cpu=12 start=331.69 finish=373.93
1176003) himeno cpu=3 start=377.94 finish=420.20
1176004) himenobmtxpa cpu=12 start=377.94 finish=420.20
1176005) himeno cpu=11 start=424.21 finish=466.42
1176006) himenobmtxpa cpu=4 start=424.21 finish=466.42
1176010) himeno cpu=11 start=470.43 finish=512.50
1176011) himenobmtxpa cpu=4 start=470.43 finish=512.50
1176020) himeno cpu=11 start=516.50 finish=558.76
1176021) himenobmtxpa cpu=12 start=516.51 finish=558.76