OV - exec - Loop 94

	Loop Id: 94	Module: exec	Source: ljForce.c:191-216 [...]	Coverage: 87.38%

0x418040 LDP	D2, D4, [X0]    [5]

0x418044 LDR	D1, [X0, #16]    [5]

0x418048 FSUB	D25, D5, S4

0x41804c FSUB	D26, D30, S2

0x418050 FSUB	D27, D31, S1

0x418054 FMUL	D24, D25, D25

0x418058 FMADD	D28, D26, D26, D24

0x41805c FMADD	D29, D27, D27, D28

0x418060 FCMPE	D29, #0

0x418064 FCCMPE	D7, D29, #1, #12

0x418068 B.GE	4180e0

0x41806c ADD	X0, X0, #24

0x418070 CMP	X0, X2

0x418074 B.NE	418040

0x4180e0 FDIV	D6, D18, D29

0x4180e4 LDP	X5, X7, [X8, #40]    [4]

0x4180e8 ADD	X0, X0, #24

0x4180ec LDR	D24, [X7, X4]    [3]

0x4180f0 ADD	X1, X5, X3

0x4180f4 FMUL	D3, D21, D6

0x4180f8 FMUL	D5, D6, D6

0x4180fc FMUL	D30, D19, D6

0x418100 FMUL	D31, D3, D5

0x418104 FMADD	D8, D31, D23, D22

0x418108 FSUB	D0, D31, S18

0x41810c FNMSUB	D2, D31, D0, D20

0x418110 FMUL	D4, D31, D8

0x418114 FMADD	D28, D2, D17, D24

0x418118 FMADD	D16, D2, D17, D16

0x41811c FMUL	D1, D4, D30

0x418120 STR	D28, [X7, X4]    [3]

0x418124 LDR	D29, [X5, X3]    [1]

0x418128 FMSUB	D26, D1, D26, D29

0x41812c STR	D26, [X5, X3]    [1]

0x418130 LDP	D6, D3, [X1, #8]    [6]

0x418134 FMSUB	D25, D1, D25, D6

0x418138 FMSUB	D27, D1, D27, D3

0x41813c STP	D25, D27, [X1, #8]    [6]

0x418140 CMP	X0, X2

0x418144 B.EQ	418078

0x418148 LDP	D30, D5, [X6]    [2]

0x41814c LDR	D31, [X6, #16]    [2]

0x418150 B	418040

/home/eoseret/qaas/qaas_runs/178-176-0594/intel/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c: 191 - 216

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

191:             for (int jOff=jBox*MAXATOMS; jOff<(jBox*MAXATOMS+nJBox); jOff++)

[...]

197:                   dr[m] = s->atoms->r[iOff][m]-s->atoms->r[jOff][m];

198:                   r2+=dr[m]*dr[m];

199:                }

200:

201:                if ( r2 <= rCut2 && r2 > 0.0)

202:                {

203:

204:                   // Important note:

205:                   // from this point on r actually refers to 1.0/r

206:                   r2 = 1.0/r2;

207:                   real_t r6 = s6 * (r2*r2*r2);

208:                   real_t eLocal = r6 * (r6 - 1.0) - eShift;

209:                   s->atoms->U[iOff] += 0.5*eLocal;

210:                   ePot += 0.5*eLocal;

211:

212:                   // different formulation to avoid sqrt computation

213:                   real_t fr = - 4.0*epsilon*r6*r2*(12.0*r6 - 6.0);

214:                   for (int m=0; m<3; m++)

215:                   {

216:                      s->atoms->f[iOff][m] -= dr[m]*fr;

Coverage (%)	Name	Source Location	Module
►98.58+	omp_fulfill_event		libgomp.so.1.0.0
○	start_thread		libc.so.6
○	thread_start		libc.so.6
►1.41+	GOMP_parallel		libgomp.so.1.0.0
○	ljForce	ljForce.c:172	exec
○	timestep	timestep.c:51	exec
○	main	CoMD.c:125	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start	CoMD.c:266	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 /

Metric	Value
CQA speedup if no scalar integer	1.55
CQA speedup if FP arith vectorized	1.55
CQA speedup if fully vectorized	3.82 - 3.72
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.26
Bottlenecks
Function	ljForce._omp_fn.1
Source	ljForce.c:191-191,ljForce.c:197-216
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	max
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	4.25
CQA cycles if no scalar integer	2.75
CQA cycles if FP arith vectorized	2.75
CQA cycles if fully vectorized	1.11 - 1.14
Front-end cycles	3.38
P0 cycles	1.25
P1 cycles	1.25
P2 cycles	0.63
P3 cycles	0.63
P4 cycles	0.63
P5 cycles	0.63
P6 cycles	4.25
P7 cycles	4.25
P8 cycles	4.25
P9 cycles	4.25
P10 cycles	2.25
P11 cycles	2.08
P12 cycles	2.17
P13 cycles	0.00
P14 cycles	0.00
DIV/SQRT cycles	0.88 - 1.75
Inter-iter dependencies cycles	2
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	27.00
Nb uops	27.00
Nb loads	NA
Nb stores	1.50
Nb stack references	0.00
FLOP/cycle	4.47
Nb FLOP add-sub	3.50
Nb FLOP mul	4.00
Nb FLOP fma	5.50
Nb FLOP div	0.50
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	0.00
Bytes prefetched	0.00
Bytes loaded	0.00
Bytes stored	0.00
Stride 0	1.00
Stride 1	0.00
Stride n	1.00
Stride unknown	1.50
Stride indirect	0.00
Vectorization ratio all	11.06
Vectorization ratio load	46.43
Vectorization ratio store	33.33
Vectorization ratio mul	0.00
Vectorization ratio add_sub	0.00
Vectorization ratio fma	0.00
Vectorization ratio div_sqrt	0.00
Vectorization ratio other	0.00
Vector-efficiency ratio all	27.77
Vector-efficiency ratio load	36.61
Vector-efficiency ratio store	33.33
Vector-efficiency ratio mul	25.00
Vector-efficiency ratio add_sub	25.00
Vector-efficiency ratio fma	25.00
Vector-efficiency ratio div_sqrt	25.00
Vector-efficiency ratio other	25.00

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	3.73
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.14
Bottlenecks	P6, P7, P8, P9,
Function	ljForce._omp_fn.1
Source	ljForce.c:191-191,ljForce.c:197-216
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	max
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	2.00
CQA cycles if no scalar integer	2.00
CQA cycles if FP arith vectorized	2.00
CQA cycles if fully vectorized	0.54
Front-end cycles	1.75
P0 cycles	1.00
P1 cycles	1.00
P2 cycles	0.50
P3 cycles	0.50
P4 cycles	0.50
P5 cycles	0.50
P6 cycles	2.00
P7 cycles	2.00
P8 cycles	2.00
P9 cycles	2.00
P10 cycles	0.67
P11 cycles	0.67
P12 cycles	0.67
P13 cycles	0.00
P14 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	2
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	14.00
Nb uops	14.00
Nb loads	NA
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	4.00
Nb FLOP add-sub	3.00
Nb FLOP mul	1.00
Nb FLOP fma	2.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	0.00
Bytes prefetched	0.00
Bytes loaded	0.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	0.00
Stride n	1.00
Stride unknown	0.00
Stride indirect	0.00
Vectorization ratio all	10.00
Vectorization ratio load	50.00
Vectorization ratio store	NA
Vectorization ratio mul	0.00
Vectorization ratio add_sub	0.00
Vectorization ratio fma	0.00
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	27.50
Vector-efficiency ratio load	37.50
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	25.00
Vector-efficiency ratio add_sub	25.00
Vector-efficiency ratio fma	25.00
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	25.00

Metric	Value
CQA speedup if no scalar integer	1.86
CQA speedup if FP arith vectorized	1.86
CQA speedup if fully vectorized	3.85 - 3.72
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.30
Bottlenecks	P6, P7, P8, P9,
Function	ljForce._omp_fn.1
Source	ljForce.c:191-191,ljForce.c:197-216
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	max
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	6.50
CQA cycles if no scalar integer	3.50
CQA cycles if FP arith vectorized	3.50
CQA cycles if fully vectorized	1.69 - 1.75
Front-end cycles	5.00
P0 cycles	1.50
P1 cycles	1.50
P2 cycles	0.75
P3 cycles	0.75
P4 cycles	0.75
P5 cycles	0.75
P6 cycles	6.50
P7 cycles	6.50
P8 cycles	6.50
P9 cycles	6.50
P10 cycles	3.83
P11 cycles	3.50
P12 cycles	3.67
P13 cycles	0.00
P14 cycles	0.00
DIV/SQRT cycles	1.75 - 3.50
Inter-iter dependencies cycles	2
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	40.00
Nb uops	40.00
Nb loads	NA
Nb stores	3.00
Nb stack references	0.00
FLOP/cycle	4.62
Nb FLOP add-sub	4.00
Nb FLOP mul	7.00
Nb FLOP fma	9.00
Nb FLOP div	1.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	0.00
Bytes prefetched	0.00
Bytes loaded	0.00
Bytes stored	0.00
Stride 0	2.00
Stride 1	0.00
Stride n	1.00
Stride unknown	3.00
Stride indirect	0.00
Vectorization ratio all	12.12
Vectorization ratio load	42.86
Vectorization ratio store	33.33
Vectorization ratio mul	0.00
Vectorization ratio add_sub	0.00
Vectorization ratio fma	0.00
Vectorization ratio div_sqrt	0.00
Vectorization ratio other	0.00
Vector-efficiency ratio all	28.03
Vector-efficiency ratio load	35.71
Vector-efficiency ratio store	33.33
Vector-efficiency ratio mul	25.00
Vector-efficiency ratio add_sub	25.00
Vector-efficiency ratio fma	25.00
Vector-efficiency ratio div_sqrt	25.00
Vector-efficiency ratio other	25.00

Path /

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

Function	ljForce._omp_fn.1
Source file and lines	ljForce.c:191-216
Module	exec

The loop is defined in /home/eoseret/qaas/qaas_runs/178-176-0594/intel/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:191,197-216.

The related source loop is not unrolled or unrolled with no peel/tail loop.
The structure of this loop is probably <if then [else] end>.

The presence of multiple execution paths is typically the main/first bottleneck.
Try to simplify control inside loop: ideally, try to remove all conditional expressions, for example by (if applicable):

hoisting them (moving them outside the loop)
turning them into conditional moves, MIN or MAX

Ex: if (x<0) x=0 => x = (x<0 ? 0 : x) (or MAX(0,x) after defining the corresponding macro)

gain
potential
hint
expert

Vectorization

Your loop is not vectorized. Only 27% of vector register length is used (average across all VPU instructions). By fully vectorizing your loop, you can lower the cost of an iteration from 4.25 to 1.14 cycles (3.72x speedup).

Details

11% of VPU instructions are used in vector version (process two or more data elements in vector registers):

46% of VPU loads are used in vector version.
33% of VPU stores are used in vector version.
0% of VPU addition or subtraction instructions are used in vector version.
0% of VPU multiply instructions are used in vector version.
0% of VPU fused multiply-add instructions are used in vector version.
0% of VPU divide and square root instructions are used in vector version.
0% of VPU 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 loop can use their full power.

Workaround

Try another compiler or update/tune your current one
Remove inter-iterations dependences from your loop and make it unit-stride:
- If your arrays have 2 or more dimensions, check whether elements are accessed contiguously and, otherwise, try to permute loops accordingly: C storage order is row-major: for(i) for(j) a[j][i] = b[j][i]; (slow, non stride 1) => for(i) for(j) a[i][j] = b[i][j]; (fast, stride 1)
- If your loop 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)

FMA

Detected 5.50 FMA (fused multiply-add) operations. Presence of both ADD/SUB and MUL operations.

Workaround

Try to change order in which elements are evaluated (using parentheses) in arithmetic expressions containing both ADD/SUB and MUL operations to enable your compiler to generate FMA instructions wherever possible. For instance a + b*c is a valid FMA (MUL then ADD). However (a+b)* c cannot be translated into an FMA (ADD then MUL).

Slow data structures access

Detected data structures (typically arrays) that cannot be efficiently read/written

Details

Constant unknown stride: 1 occurrence(s)
Constant non-unit stride: 1 occurrence(s)

Non-unit stride (uncontiguous) accesses are not efficiently using data caches

Workaround

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

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

The binary loop is composed of 19 FP arithmetical operations:

9: addition or subtraction (5 inside FMA instructions)
9.50: multiply (5 inside FMA instructions)
0.50: divide

The binary loop does not load or store any data.

General properties

nb instructions	27
nb uops	27
loop length	108
used w registers	0
used x registers	5.50
used b registers	0
used h registers	0
used s registers	3.50
used d registers	19
used q registers	0
used v registers	0
used z registers	0
nb stack references	0
ADD-SUB / MUL ratio	1.79

Front-end

MACRO FUSION NOT POSSIBLE

micro-operation queue	3.38 cycles
front end	3.38 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.25	1.25	0.63	0.63	0.63	0.63	4.25	4.25	4.25	4.25	2.25	2.08	2.17	0.00	0.00
cycles	1.25	1.25	0.63	0.63	0.63	0.63	4.25	4.25	4.25	4.25	2.25	2.08	2.17	0.00	0.00

Execution ports to units layout:

P0: BRU
P1: BRU
P2: ALU
P3: ALU
P4: ALU
P5: ALU
P6 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P7 (128 bits): VPU, ALU, FP store data
P8 (128 bits): VPU, ALU, DIV/SQRT
P9 (128 bits): ALU, VPU
P10 (256 bits): store address, load
P11 (256 bits): store address, load
P12 (256 bits): load
P13 (64 bits): store data
P14 (64 bits): store data

Cycles executing div or sqrt instructions	0.88-1.75
Longest recurrence chain latency (RecMII)	2.00

Cycles summary

Front-end	3.38
Dispatch	4.25
DIV/SQRT	0.88-1.75
Data deps.	2.00
Overall L1	4.25

Vectorization ratios

INT

all	45%
load	46%
store	33%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

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

INT+FP

all	11%
load	46%
store	33%
mul	0%
add-sub	0%
fma	0%
div/sqrt	0%
other	0%

Vector efficiency ratios

INT

all	36%
load	36%
store	33%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

all	25%
load	NA (no load vectorizable/vectorized instructions)
store	NA (no store vectorizable/vectorized instructions)
mul	25%
add-sub	25%
fma	25%
div/sqrt	25%
other	25%

INT+FP

all	27%
load	36%
store	33%
mul	25%
add-sub	25%
fma	25%
div/sqrt	25%
other	25%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 4.25 cycles.

Function	ljForce._omp_fn.1
Source file and lines	ljForce.c:191-216
Module	exec

hoisting them (moving them outside the loop)
turning them into conditional moves, MIN or MAX

Ex: if (x<0) x=0 => x = (x<0 ? 0 : x) (or MAX(0,x) after defining the corresponding macro)

gain
potential
hint
expert

Vectorization

Details

10% of VPU instructions are used in vector version (process two or more data elements in vector registers):

50% of VPU loads are used in vector version.
0% of VPU addition or subtraction instructions are used in vector version.
0% of VPU multiply instructions are used in vector version.
0% of VPU fused multiply-add instructions are used in vector version.
0% of VPU 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 loop can use their full power.

Workaround

Try another compiler or update/tune your current one
Remove inter-iterations dependences from your loop and make it unit-stride:
- If your arrays have 2 or more dimensions, check whether elements are accessed contiguously and, otherwise, try to permute loops accordingly: C storage order is row-major: for(i) for(j) a[j][i] = b[j][i]; (slow, non stride 1) => for(i) for(j) a[i][j] = b[i][j]; (fast, stride 1)
- If your loop 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)

FMA

Detected 2 FMA (fused multiply-add) operations. Presence of both ADD/SUB and MUL operations.

Workaround

Slow data structures access

Detected data structures (typically arrays) that cannot be efficiently read/written

Details

Constant non-unit stride: 1 occurrence(s)

Non-unit stride (uncontiguous) accesses are not efficiently using data caches

Workaround

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

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

The binary loop is composed of 8 FP arithmetical operations:

5: addition or subtraction (2 inside FMA instructions)
3: multiply (2 inside FMA instructions)

The binary loop does not load or store any data.

General properties

nb instructions	14
nb uops	14
loop length	56
used w registers	0
used x registers	2
used b registers	0
used h registers	0
used s registers	3
used d registers	13
used q registers	0
used v registers	0
used z registers	0
nb stack references	0
ADD-SUB / MUL ratio	3.00

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	1.75 cycles
front end	1.75 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.00	1.00	0.50	0.50	0.50	0.50	2.00	2.00	2.00	2.00	0.67	0.67	0.67	0.00	0.00
cycles	1.00	1.00	0.50	0.50	0.50	0.50	2.00	2.00	2.00	2.00	0.67	0.67	0.67	0.00	0.00

Execution ports to units layout:

P0: BRU
P1: BRU
P2: ALU
P3: ALU
P4: ALU
P5: ALU
P6 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P7 (128 bits): VPU, ALU, FP store data
P8 (128 bits): VPU, ALU, DIV/SQRT
P9 (128 bits): ALU, VPU
P10 (256 bits): store address, load
P11 (256 bits): store address, load
P12 (256 bits): load
P13 (64 bits): store data
P14 (64 bits): store data

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	2.00

Cycles summary

Front-end	1.75
Dispatch	2.00
Data deps.	2.00
Overall L1	2.00

Vectorization ratios

INT

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

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

INT+FP

all	10%
load	50%
store	NA (no store vectorizable/vectorized instructions)
mul	0%
add-sub	0%
fma	0%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	0%

Vector efficiency ratios

INT

all	37%
load	37%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

all	25%
load	NA (no load vectorizable/vectorized instructions)
store	NA (no store vectorizable/vectorized instructions)
mul	25%
add-sub	25%
fma	25%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	25%

INT+FP

all	27%
load	37%
store	NA (no store vectorizable/vectorized instructions)
mul	25%
add-sub	25%
fma	25%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	25%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 2.00 cycles.

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 418040

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	Latency	Recip. throughput	Vectorization
LDP D2, D4, [X0]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33	vect (50.0%)
LDR D1, [X0, #16]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33	scal (25.0%)
FSUB D25, D5, S4	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	scal (25.0%)
FSUB D26, D30, S2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	scal (25.0%)
FSUB D27, D31, S1	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	scal (25.0%)
FMUL D24, D25, D25	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	scal (25.0%)
FMADD D28, D26, D26, D24	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	scal (25.0%)
FMADD D29, D27, D27, D28	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	scal (25.0%)
FCMPE D29, #0	1	0	0	0	0	0	0	1	0	0	0	0	0	0	2	1	scal (25.0%)
FCCMPE D7, D29, #1, #12	1	0	0	0	0	0	0	1	0	0	0	0	0	0	2	1	scal (25.0%)
B.GE 4180e0 <ljForce._omp_fn.1+0x1d4>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
ADD X0, X0, #24	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	1	0.25	N/A
CMP X0, X2	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	1	0.33	N/A
B.NE 418040 <ljForce._omp_fn.1+0x134>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A

Function	ljForce._omp_fn.1
Source file and lines	ljForce.c:191-216
Module	exec

hoisting them (moving them outside the loop)
turning them into conditional moves, MIN or MAX

Ex: if (x<0) x=0 => x = (x<0 ? 0 : x) (or MAX(0,x) after defining the corresponding macro)

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 6.50 to 3.50 cycles (1.86x speedup).

Workaround

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

Vectorization

Your loop is not vectorized. Only 28% of vector register length is used (average across all VPU instructions). By fully vectorizing your loop, you can lower the cost of an iteration from 6.50 to 1.75 cycles (3.72x speedup).

Details

12% of VPU instructions are used in vector version (process two or more data elements in vector registers):

42% of VPU loads are used in vector version.
33% of VPU stores are used in vector version.
0% of VPU addition or subtraction instructions are used in vector version.
0% of VPU multiply instructions are used in vector version.
0% of VPU fused multiply-add instructions are used in vector version.
0% of VPU divide and square root instructions are used in vector version.
0% of VPU 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 loop can use their full power.

Workaround

Try another compiler or update/tune your current one
Remove inter-iterations dependences from your loop and make it unit-stride:
- If your arrays have 2 or more dimensions, check whether elements are accessed contiguously and, otherwise, try to permute loops accordingly: C storage order is row-major: for(i) for(j) a[j][i] = b[j][i]; (slow, non stride 1) => for(i) for(j) a[i][j] = b[i][j]; (fast, stride 1)
- If your loop 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)

FMA

Detected 9 FMA (fused multiply-add) operations. Presence of both ADD/SUB and MUL operations.

Workaround

Slow data structures access

Detected data structures (typically arrays) that cannot be efficiently read/written

Details

Constant unknown stride: 3 occurrence(s)
Constant non-unit stride: 1 occurrence(s)

Non-unit stride (uncontiguous) accesses are not efficiently using data caches

Workaround

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

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

The binary loop is composed of 30 FP arithmetical operations:

13: addition or subtraction (9 inside FMA instructions)
16: multiply (9 inside FMA instructions)
1: divide

The binary loop does not load or store any data.

General properties

nb instructions	40
nb uops	40
loop length	160
used w registers	0
used x registers	9
used b registers	0
used h registers	0
used s registers	4
used d registers	25
used q registers	0
used v registers	0
used z registers	0
nb stack references	0
ADD-SUB / MUL ratio	0.57

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	5.00 cycles
front end	5.00 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.50	1.50	0.75	0.75	0.75	0.75	6.50	6.50	6.50	6.50	3.83	3.50	3.67	0.00	0.00
cycles	1.50	1.50	0.75	0.75	0.75	0.75	6.50	6.50	6.50	6.50	3.83	3.50	3.67	0.00	0.00

Execution ports to units layout:

P0: BRU
P1: BRU
P2: ALU
P3: ALU
P4: ALU
P5: ALU
P6 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P7 (128 bits): VPU, ALU, FP store data
P8 (128 bits): VPU, ALU, DIV/SQRT
P9 (128 bits): ALU, VPU
P10 (256 bits): store address, load
P11 (256 bits): store address, load
P12 (256 bits): load
P13 (64 bits): store data
P14 (64 bits): store data

Cycles executing div or sqrt instructions	1.75-3.50
Longest recurrence chain latency (RecMII)	2.00

Cycles summary

Front-end	5.00
Dispatch	6.50
DIV/SQRT	1.75-3.50
Data deps.	2.00
Overall L1	6.50

Vectorization ratios

INT

all	40%
load	42%
store	33%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

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

INT+FP

all	12%
load	42%
store	33%
mul	0%
add-sub	0%
fma	0%
div/sqrt	0%
other	0%

Vector efficiency ratios

INT

all	35%
load	35%
store	33%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

all	25%
load	NA (no load vectorizable/vectorized instructions)
store	NA (no store vectorizable/vectorized instructions)
mul	25%
add-sub	25%
fma	25%
div/sqrt	25%
other	25%

INT+FP

all	28%
load	35%
store	33%
mul	25%
add-sub	25%
fma	25%
div/sqrt	25%
other	25%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 6.50 cycles.

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 418040

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	Latency	Recip. throughput	Vectorization
LDP D2, D4, [X0]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33	vect (50.0%)
LDR D1, [X0, #16]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33	scal (25.0%)
FSUB D25, D5, S4	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	scal (25.0%)
FSUB D26, D30, S2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	scal (25.0%)
FSUB D27, D31, S1	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	scal (25.0%)
FMUL D24, D25, D25	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	scal (25.0%)
FMADD D28, D26, D26, D24	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	scal (25.0%)
FMADD D29, D27, D27, D28	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	scal (25.0%)
FCMPE D29, #0	1	0	0	0	0	0	0	1	0	0	0	0	0	0	2	1	scal (25.0%)
FCCMPE D7, D29, #1, #12	1	0	0	0	0	0	0	1	0	0	0	0	0	0	2	1	scal (25.0%)
B.GE 4180e0 <ljForce._omp_fn.1+0x1d4>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
FDIV D6, D18, D29	1	0	0	0	0	0	0	0.50	0	0.50	0	0	0	0	7-15	1.75-3.50	scal (25.0%)
LDP X5, X7, [X8, #40]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	1	N/A
ADD X0, X0, #24	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	1	0.25	N/A
LDR D24, [X7, X4]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33	scal (25.0%)
ADD X1, X5, X3	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	1	0.25	N/A
FMUL D3, D21, D6	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	scal (25.0%)
FMUL D5, D6, D6	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	scal (25.0%)
FMUL D30, D19, D6	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	scal (25.0%)
FMUL D31, D3, D5	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	scal (25.0%)
FMADD D8, D31, D23, D22	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	scal (25.0%)
FSUB D0, D31, S18	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	scal (25.0%)
FNMSUB D2, D31, D0, D20	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	scal (25.0%)
FMUL D4, D31, D8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	scal (25.0%)
FMADD D28, D2, D17, D24	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	scal (25.0%)
FMADD D16, D2, D17, D16	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	scal (25.0%)
FMUL D1, D4, D30	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	scal (25.0%)
STR D28, [X7, X4]	1	0	0	0	0	0	0	0.50	0.50	0	0	0.50	0.50	0	2	0.50	scal (25.0%)
LDR D29, [X5, X3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33	scal (25.0%)
FMSUB D26, D1, D26, D29	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	scal (25.0%)
STR D26, [X5, X3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0.50	0.50	0	2	0.50	scal (25.0%)
LDP D6, D3, [X1, #8]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33	vect (50.0%)
FMSUB D25, D1, D25, D6	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	scal (25.0%)
FMSUB D27, D1, D27, D3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	scal (25.0%)
STP D25, D27, [X1, #8]	1	0	0	0	0	0	0	0.50	0.50	0	0	0.50	0.50	0	2	0.50	vect (50.0%)
CMP X0, X2	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	1	0.33	N/A
B.EQ 418078 <ljForce._omp_fn.1+0x16c>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
LDP D30, D5, [X6]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33	vect (50.0%)
LDR D31, [X6, #16]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33	scal (25.0%)
B 418040 <ljForce._omp_fn.1+0x134>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A

Report Configuration

Vectorization

Details

Workaround

FMA

Workaround

Slow data structures access

Details

Workaround

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Vectorization

Details

Workaround

FMA

Workaround

Slow data structures access

Details

Workaround

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

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

FMA

Workaround

Slow data structures access

Details

Workaround

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code