OV - exec - Function loadAtomsBuffer

	Function: loadAtomsBuffer	Module: exec	Source: haloExchange.c:362-393	Coverage (incl. loops): 0.04% \| (excl. loops): 0.00%

/home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/haloExchange.c: 362 - 393

--------------------------------------------------------------------------------

362: {

363:    AtomExchangeParms* parms = (AtomExchangeParms*) vparms;

364:    SimFlat* s = (SimFlat*) data;

365:    AtomMsg* buf = (AtomMsg*) charBuf;

366:

367:    real_t* pbcFactor = parms->pbcFactor[face];

368:    real3 shift;

369:    shift[0] = pbcFactor[0] * s->domain->globalExtent[0];

370:    shift[1] = pbcFactor[1] * s->domain->globalExtent[1];

371:    shift[2] = pbcFactor[2] * s->domain->globalExtent[2];

372:

373:    int nCells = parms->nCells[face];

374:    int* cellList = parms->cellList[face];

375:    int nBuf = 0;

376:    for (int iCell=0; iCell<nCells; ++iCell)

377:    {

378:       int iBox = cellList[iCell];

379:       int iOff = iBox*MAXATOMS;

380:       for (int ii=iOff; ii<iOff+s->boxes->nAtoms[iBox]; ++ii)

381:       {

382:          buf[nBuf].gid  = s->atoms->gid[ii];

383:          buf[nBuf].type = s->atoms->iSpecies[ii];

384:          buf[nBuf].rx = s->atoms->r[ii][0] + shift[0];

385:          buf[nBuf].ry = s->atoms->r[ii][1] + shift[1];

386:          buf[nBuf].rz = s->atoms->r[ii][2] + shift[2];

387:          buf[nBuf].px = s->atoms->p[ii][0];

388:          buf[nBuf].py = s->atoms->p[ii][1];

389:          buf[nBuf].pz = s->atoms->p[ii][2];

390:          ++nBuf;

391:       }

392:    }

393:    return nBuf*sizeof(AtomMsg);

0x407260 PUSH	%RBP

0x407261 MOV	%RSP,%RBP

0x407264 PUSH	%R15

0x407266 PUSH	%R14

0x407268 PUSH	%R13

0x40726a PUSH	%R12

0x40726c PUSH	%RBX

0x40726d MOV	%RCX,-0x38(%RBP)

0x407271 MOVSXD	%EDX,%RAX

0x407274 MOV	(%RDI,%RAX,4),%ECX

0x407277 MOV	%RCX,-0x40(%RBP)

0x40727b TEST	%ECX,%ECX

0x40727d JLE	407380

0x407283 MOV	0x48(%RDI,%RAX,8),%RCX

0x407288 MOV	0x10(%RSI),%R8

0x40728c VMOVUPD	0x48(%R8),%XMM0

0x407292 VMULPD	(%RCX),%XMM0,%XMM0

0x407296 MOV	0x18(%RDI,%RAX,8),%RDX

0x40729b VMOVSD	0x58(%R8),%XMM1

0x4072a1 VMULSD	0x10(%RCX),%XMM1,%XMM1

0x4072a6 MOV	%RSI,-0x30(%RBP)

0x4072aa MOV	0x18(%RSI),%RAX

0x4072ae MOV	0x78(%RAX),%R8

0x4072b2 XOR	%R9D,%R9D

0x4072b5 XOR	%EDI,%EDI

0x4072b7 JMP	4072cd

0x4072b9 NOPL	(%RAX)

(42) 0x4072c0 INC	%R9

(42) 0x4072c3 CMP	-0x40(%RBP),%R9

(42) 0x4072c7 JE	407382

(42) 0x4072cd MOVSXD	(%RDX,%R9,4),%R10

(42) 0x4072d1 CMPL	$0,(%R8,%R10,4)

(42) 0x4072d6 JLE	4072c0

(42) 0x4072d8 MOV	%R10D,%EAX

(42) 0x4072db SAL	$0x6,%EAX

(42) 0x4072de MOV	-0x30(%RBP),%RCX

(42) 0x4072e2 MOV	0x20(%RCX),%RCX

(42) 0x4072e6 MOV	0x8(%RCX),%R11

(42) 0x4072ea MOV	0x10(%RCX),%RBX

(42) 0x4072ee MOV	0x18(%RCX),%R14

(42) 0x4072f2 MOV	0x20(%RCX),%R15

(42) 0x4072f6 MOVSXD	%EAX,%RCX

(42) 0x4072f9 MOVSXD	%EDI,%RAX

(42) 0x4072fc IMUL	$0x38,%RAX,%R13

(42) 0x407300 ADD	-0x38(%RBP),%R13

(42) 0x407304 LEA	(%RCX,%RCX,2),%RAX

(42) 0x407308 LEA	0x4(,%RAX,2),%RAX

(42) 0x407310 MOV	%RCX,%R12

(42) 0x407313 NOPW	%CS:(%RAX,%RAX,1)

  (43) 0x407320 MOV	(%R11,%R12,4),%ESI

  (43) 0x407324 MOV	%ESI,(%R13)

  (43) 0x407328 MOV	(%RBX,%R12,4),%ESI

  (43) 0x40732c MOV	%ESI,0x4(%R13)

  (43) 0x407330 VADDPD	-0x10(%R14,%RAX,4),%XMM0,%XMM2

  (43) 0x407337 VMOVUPD	%XMM2,0x8(%R13)

  (43) 0x40733d VADDSD	(%R14,%RAX,4),%XMM1,%XMM2

  (43) 0x407343 VMOVSD	%XMM2,0x18(%R13)

  (43) 0x407349 VMOVUPS	-0x10(%R15,%RAX,4),%XMM2

  (43) 0x407350 VMOVUPS	%XMM2,0x20(%R13)

  (43) 0x407356 VMOVSD	(%R15,%RAX,4),%XMM2

  (43) 0x40735c VMOVSD	%XMM2,0x30(%R13)

  (43) 0x407362 INC	%R12

  (43) 0x407365 MOVSXD	(%R8,%R10,4),%RSI

  (43) 0x407369 ADD	%RCX,%RSI

  (43) 0x40736c ADD	$0x38,%R13

  (43) 0x407370 INC	%EDI

  (43) 0x407372 ADD	$0x6,%RAX

  (43) 0x407376 CMP	%RSI,%R12

  (43) 0x407379 JL	407320

(42) 0x40737b JMP	4072c0

0x407380 XOR	%EDI,%EDI

0x407382 IMUL	$0x38,%EDI,%EAX

0x407385 POP	%RBX

0x407386 POP	%R12

0x407388 POP	%R13

0x40738a POP	%R14

0x40738c POP	%R15

0x40738e POP	%RBP

0x40738f RET

Coverage (%)	Name	Source Location	Module
►14.95+	haloExchange	haloExchange.c:289	exec
○	timestep	timestep.c:150	exec
○	main	CoMD.c:125	exec
►13.40+	haloExchange	haloExchange.c:288	exec
○	timestep	timestep.c:150	exec
○	main	CoMD.c:125	exec
►11.07+	haloExchange	haloExchange.c:289	exec
○	timestep	timestep.c:150	exec
○	main	CoMD.c:125	exec
►10.87+	haloExchange	haloExchange.c:289	exec
○	timestep	timestep.c:150	exec
○	main	CoMD.c:125	exec
►10.49+	haloExchange	haloExchange.c:288	exec
○	timestep	timestep.c:150	exec
○	main	CoMD.c:125	exec
►10.48+	haloExchange	haloExchange.c:288	exec
○	timestep	timestep.c:150	exec
○	main	CoMD.c:125	exec
►5.05+	haloExchange	haloExchange.c:289	exec
○	timestep	timestep.c:150	exec
○	main	CoMD.c:125	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start		exec
►4.85+	haloExchange	haloExchange.c:288	exec
○	timestep	timestep.c:150	exec
○	main	CoMD.c:125	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start		exec
►4.85+	haloExchange	haloExchange.c:289	exec
○	timestep	timestep.c:150	exec
○	main	CoMD.c:125	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start		exec
►4.85+	haloExchange	haloExchange.c:289	exec
○	timestep	timestep.c:150	exec
○	main	CoMD.c:125	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start		exec
►4.08+	haloExchange	haloExchange.c:288	exec
○	timestep	timestep.c:150	exec
○	main	CoMD.c:125	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start		exec
►3.88+	haloExchange	haloExchange.c:288	exec
○	timestep	timestep.c:150	exec
○	main	CoMD.c:125	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start		exec

min	med	avg	max

Percentile Index	10	20	30	40	50	60	70	80	90	100
Value

min	med	avg	max

Percentile Index	10	20	30	40	50	60	70	80	90	100
Value

Path /

Average path: Display a virtual path defined by average values of all real paths

The code analyzed by CQA in that panel excludes loops and represents 0.00% of application time for run icx_0

Source file and lines	haloExchange.c:362-393
Module	exec

The function is defined in /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/haloExchange.c:362-393.

Warnings:
Ignoring paths for analysis

gain
potential
hint
expert

Code clean check

Detected a slowdown caused by scalar integer instructions (typically used for address computation). By removing them, you can lower the cost of an iteration from 4.38 to 1.25 cycles (3.50x speedup).

Workaround

Try to reorganize arrays of structures to structures of arrays
Consider to permute loops (see vectorization gain report)

Vectorization

Your function is poorly vectorized. Only 14% of vector register length is used (average across all SSE/AVX instructions). By fully vectorizing your function, you can lower the cost of an iteration from 4.38 to 0.39 cycles (11.20x speedup).

Details

22% of SSE/AVX instructions are used in vector version (process two or more data elements in vector registers):

40% of SSE/AVX loads are used in vector version.
0% of SSE/AVX stores are used in vector version.
50% of SSE/AVX multiply instructions are used in vector version.
0% of SSE/AVX instructions that are not load, store, addition, subtraction nor multiply instructions are used in vector version.

Since your execution units are vector units, only a fully vectorized function can use their full power.

Workaround

Try another compiler or update/tune your current one
Make array accesses unit-stride:
- If your function streams arrays of structures (AoS), try to use structures of arrays instead (SoA): for(i) a[i].x = b[i].x; (slow, non stride 1) => for(i) a.x[i] = b.x[i]; (fast, stride 1)

Execution units bottlenecks

Found no such bottlenecks but see expert reports for more complex bottlenecks.

No data for this section

Type of elements and instruction set

1 SSE or AVX instructions are processing arithmetic or math operations on double precision FP elements in scalar mode (one at a time). 1 SSE or AVX instructions are processing arithmetic or math operations on double precision FP elements in vector mode (two at a time).

Matching between your function (in the source code) and the binary function

The binary function is composed of 3 FP arithmetical operations:

3: multiply

The binary function is loading 92 bytes (11 double precision FP elements). The binary function is storing 24 bytes (3 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.03 FP operations per loaded or stored byte.

General properties

nb instructions	36
nb uops	35
loop length	112
used x86 registers	14
used mmx registers	0
used xmm registers	2
used ymm registers	0
used zmm registers	0
nb stack references	3

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	4.38 cycles
front end	4.38 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	1.00	1.00	1.00	1.00	1.00	1.00	3.25	3.25	3.25	3.25	1.00	1.00	0.00	0.00	0.00	0.00
cycles	1.00	1.00	1.00	1.00	1.00	1.00	3.25	3.25	3.25	3.25	1.00	1.00	0.00	0.00	0.00	0.00

Execution ports to units layout:

P0: ALU
P1: ALU
P2: ALU
P3: ALU, BRU
P4: ALU, BRU
P5: ALU, BRU
P6 (512 bits): store address, load
P7 (512 bits): store address, load
P8 (512 bits): store address, load
P9 (512 bits): store address, load
P10 (512 bits): VPU, DIV/SQRT
P11 (512 bits): VPU, DIV/SQRT
P12 (512 bits): VPU
P13 (512 bits): VPU
P14 (512 bits): FP store data
P15 (512 bits): FP store data

Cycles executing div or sqrt instructions

Cycles summary

Front-end	4.38
Dispatch	3.25
Overall L1	4.38

Vectorization ratios

INT

all	0%
load	0%
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	0%

all	50%
load	50%
store	NA (no store vectorizable/vectorized instructions)
mul	50%
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

INT+FP

all	22%
load	40%
store	0%
mul	50%
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	0%

Vector efficiency ratios

INT

all	11%
load	12%
store	12%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	6%

all	18%
load	18%
store	NA (no store vectorizable/vectorized instructions)
mul	18%
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

INT+FP

all	14%
load	17%
store	12%
mul	18%
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	6%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each call to the function takes 4.38 cycles. At this rate:

8% of peak load performance is reached (21.03 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
4% of peak store performance is reached (5.49 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Performance is limited by instruction throughput (loading/decoding program instructions to execution core) (front-end is a bottleneck). By removing all these bottlenecks, you can lower the cost of an iteration from 4.38 to 3.25 cycles (1.35x speedup).

ASM code

In the binary file, the address of the function is: 407260

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	Latency	Recip. throughput	Vectorization
PUSH %RBP	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
MOV %RSP,%RBP	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
PUSH %R15	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
PUSH %R14	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
PUSH %R13	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
PUSH %R12	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
PUSH %RBX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
MOV %RCX,-0x38(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	4	0.50	scal (12.5%)
MOVSXD %EDX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	1	0.17	N/A
MOV (%RDI,%RAX,4),%ECX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.25	N/A
MOV %RCX,-0x40(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	4	0.50	scal (12.5%)
TEST %ECX,%ECX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	1	0.17	N/A
JLE 407380 <loadAtomsBuffer+0x120>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	1	0.50-1	N/A
MOV 0x48(%RDI,%RAX,8),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.25	N/A
MOV 0x10(%RSI),%R8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.25	N/A
VMOVUPD 0x48(%R8),%XMM0	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.50	vect (25.0%)
VMULPD (%RCX),%XMM0,%XMM0	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	3	0.50	vect (25.0%)
MOV 0x18(%RDI,%RAX,8),%RDX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.25	scal (12.5%)
VMOVSD 0x58(%R8),%XMM1	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.50	scal (12.5%)
VMULSD 0x10(%RCX),%XMM1,%XMM1	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	3	0.50	scal (12.5%)
MOV %RSI,-0x30(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	4	0.50	scal (12.5%)
MOV 0x18(%RSI),%RAX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.25	N/A
MOV 0x78(%RAX),%R8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.25	N/A
XOR %R9D,%R9D	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.25	scal (6.3%)
XOR %EDI,%EDI	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.25	N/A
JMP 4072cd <loadAtomsBuffer+0x6d>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	1	N/A
NOPL (%RAX)	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.09	N/A
XOR %EDI,%EDI	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.25	N/A
IMUL $0x38,%EDI,%EAX	1	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	3	0.33	N/A
POP %RBX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
POP %R12	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
POP %R13	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
POP %R14	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
POP %R15	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
POP %RBP	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
RET	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	1	0.50	N/A

The code analyzed by CQA in that panel excludes loops and represents 0.00% of application time for run icx_0

Source file and lines	haloExchange.c:362-393
Module	exec

The function is defined in /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/haloExchange.c:362-393.

Warnings:
Ignoring paths for analysis

gain
potential
hint
expert

Code clean check

Detected a slowdown caused by scalar integer instructions (typically used for address computation). By removing them, you can lower the cost of an iteration from 4.38 to 1.25 cycles (3.50x speedup).

Workaround

Try to reorganize arrays of structures to structures of arrays
Consider to permute loops (see vectorization gain report)

Vectorization

Details

22% of SSE/AVX instructions are used in vector version (process two or more data elements in vector registers):

40% of SSE/AVX loads are used in vector version.
0% of SSE/AVX stores are used in vector version.
50% of SSE/AVX multiply instructions are used in vector version.
0% of SSE/AVX instructions that are not load, store, addition, subtraction nor multiply instructions are used in vector version.

Since your execution units are vector units, only a fully vectorized function can use their full power.

Workaround

Try another compiler or update/tune your current one
Make array accesses unit-stride:
- If your function streams arrays of structures (AoS), try to use structures of arrays instead (SoA): for(i) a[i].x = b[i].x; (slow, non stride 1) => for(i) a.x[i] = b.x[i]; (fast, stride 1)

Execution units bottlenecks

Found no such bottlenecks but see expert reports for more complex bottlenecks.

No data for this section

Type of elements and instruction set

Matching between your function (in the source code) and the binary function

The binary function is composed of 3 FP arithmetical operations:

3: multiply

The binary function is loading 92 bytes (11 double precision FP elements). The binary function is storing 24 bytes (3 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.03 FP operations per loaded or stored byte.

General properties

nb instructions	36
nb uops	35
loop length	112
used x86 registers	14
used mmx registers	0
used xmm registers	2
used ymm registers	0
used zmm registers	0
nb stack references	3

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	4.38 cycles
front end	4.38 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	1.00	1.00	1.00	1.00	1.00	1.00	3.25	3.25	3.25	3.25	1.00	1.00	0.00	0.00	0.00	0.00
cycles	1.00	1.00	1.00	1.00	1.00	1.00	3.25	3.25	3.25	3.25	1.00	1.00	0.00	0.00	0.00	0.00

Execution ports to units layout:

P0: ALU
P1: ALU
P2: ALU
P3: ALU, BRU
P4: ALU, BRU
P5: ALU, BRU
P6 (512 bits): store address, load
P7 (512 bits): store address, load
P8 (512 bits): store address, load
P9 (512 bits): store address, load
P10 (512 bits): VPU, DIV/SQRT
P11 (512 bits): VPU, DIV/SQRT
P12 (512 bits): VPU
P13 (512 bits): VPU
P14 (512 bits): FP store data
P15 (512 bits): FP store data

Cycles executing div or sqrt instructions

Cycles summary

Front-end	4.38
Dispatch	3.25
Overall L1	4.38

Vectorization ratios

INT

all	0%
load	0%
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	0%

all	50%
load	50%
store	NA (no store vectorizable/vectorized instructions)
mul	50%
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

INT+FP

all	22%
load	40%
store	0%
mul	50%
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	0%

Vector efficiency ratios

INT

all	11%
load	12%
store	12%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	6%

all	18%
load	18%
store	NA (no store vectorizable/vectorized instructions)
mul	18%
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

INT+FP

all	14%
load	17%
store	12%
mul	18%
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	6%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each call to the function takes 4.38 cycles. At this rate:

8% of peak load performance is reached (21.03 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
4% of peak store performance is reached (5.49 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

ASM code

In the binary file, the address of the function is: 407260

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	Latency	Recip. throughput	Vectorization
PUSH %RBP	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
MOV %RSP,%RBP	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
PUSH %R15	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
PUSH %R14	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
PUSH %R13	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
PUSH %R12	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
PUSH %RBX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
MOV %RCX,-0x38(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	4	0.50	scal (12.5%)
MOVSXD %EDX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	1	0.17	N/A
MOV (%RDI,%RAX,4),%ECX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.25	N/A
MOV %RCX,-0x40(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	4	0.50	scal (12.5%)
TEST %ECX,%ECX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	1	0.17	N/A
JLE 407380 <loadAtomsBuffer+0x120>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	1	0.50-1	N/A
MOV 0x48(%RDI,%RAX,8),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.25	N/A
MOV 0x10(%RSI),%R8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.25	N/A
VMOVUPD 0x48(%R8),%XMM0	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.50	vect (25.0%)
VMULPD (%RCX),%XMM0,%XMM0	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	3	0.50	vect (25.0%)
MOV 0x18(%RDI,%RAX,8),%RDX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.25	scal (12.5%)
VMOVSD 0x58(%R8),%XMM1	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.50	scal (12.5%)
VMULSD 0x10(%RCX),%XMM1,%XMM1	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	3	0.50	scal (12.5%)
MOV %RSI,-0x30(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	4	0.50	scal (12.5%)
MOV 0x18(%RSI),%RAX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.25	N/A
MOV 0x78(%RAX),%R8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.25	N/A
XOR %R9D,%R9D	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.25	scal (6.3%)
XOR %EDI,%EDI	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.25	N/A
JMP 4072cd <loadAtomsBuffer+0x6d>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	1	N/A
NOPL (%RAX)	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.09	N/A
XOR %EDI,%EDI	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.25	N/A
IMUL $0x38,%EDI,%EAX	1	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	3	0.33	N/A
POP %RBX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
POP %R12	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
POP %R13	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
POP %R14	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
POP %R15	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
POP %RBP	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
RET	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	1	0.50	N/A

Name	Coverage (%)	Time (s)
▼loadAtomsBuffer–	0.04	0.01
▼Loop 42 - haloExchange.c:376-389 - exec–	0.01	0.04
○Loop 43 - haloExchange.c:380-389 - exec	0.03	0.28

Report Configuration

Code clean check

Workaround

Vectorization

Details

Workaround

Execution units bottlenecks

Type of elements and instruction set

Matching between your function (in the source code) and the binary function

Arithmetic intensity

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code

Code clean check

Workaround

Vectorization

Details

Workaround

Execution units bottlenecks

Type of elements and instruction set

Matching between your function (in the source code) and the binary function

Arithmetic intensity

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code