OV - exec - Loop 98

0x20edd0 MOV	-0x80(%RBP),%RSI

0x20edd4 ADDL	$0x40,-0x30(%RBP)

0x20edd8 CMP	-0x70(%RBP),%RSI

0x20eddc LEA	0x1(%RSI),%RAX

0x20ede0 MOV	%RAX,%RSI

0x20ede3 JGE	20efac

0x20ede9 MOV	-0x58(%RBP),%RAX

0x20eded MOV	-0x78(%RBP),%R9

0x20edf1 MOV	%RSI,-0x80(%RBP)

0x20edf5 MOV	(%R9,%RSI,8),%R9

0x20edf9 MOV	(%RAX,%RSI,4),%EAX

0x20edfc MOV	%R9,-0x68(%RBP)

0x20ee00 TEST	%EAX,%EAX

0x20ee02 JLE	20ef80

0x20ee08 MOVSXD	-0x30(%RBP),%R9

0x20ee0c SAL	$0x6,%ESI

0x20ee0f ADD	%ESI,%EAX

0x20ee11 XOR	%ESI,%ESI

0x20ee13 CLTQ

0x20ee15 MOV	%RAX,-0xa0(%RBP)

0x20ee1c MOV	%R9,-0x90(%RBP)

0x20ee23 JMP	20ee40

  (99) 0x20ee30 MOV	-0x98(%RBP),%RSI

  (99) 0x20ee37 INC	%RSI

  (99) 0x20ee3a CMP	%RSI,-0x60(%RBP)

  (99) 0x20ee3e JE	20edd0

  (99) 0x20ee40 MOV	-0x68(%RBP),%RAX

  (99) 0x20ee44 MOV	(%RAX,%RSI,4),%EAX

  (99) 0x20ee47 TEST	%EAX,%EAX

  (99) 0x20ee49 JS	20f07c

  (99) 0x20ee4f MOV	%RSI,-0x98(%RBP)

  (99) 0x20ee56 MOV	-0x58(%RBP),%RSI

  (99) 0x20ee5a MOV	(%RSI,%RAX,4),%ESI

  (99) 0x20ee5d TEST	%ESI,%ESI

  (99) 0x20ee5f JLE	20ee30

  (99) 0x20ee61 SAL	$0x6,%EAX

  (99) 0x20ee64 ADD	%EAX,%ESI

  (99) 0x20ee66 MOVSXD	%EAX,%R11

  (99) 0x20ee69 MOVSXD	%ESI,%R13

  (99) 0x20ee6c MOV	-0x88(%RBP),%RSI

  (99) 0x20ee73 LEA	(%R11,%R11,2),%RAX

  (99) 0x20ee77 MOV	0x20(%RSI),%R9

  (99) 0x20ee7b MOV	0x18(%R9),%R14

  (99) 0x20ee7f LEA	0x10(%R14,%RAX,8),%RAX

  (99) 0x20ee84 MOV	%RAX,-0xa8(%RBP)

  (99) 0x20ee8b MOV	-0x90(%RBP),%RAX

  (99) 0x20ee92 JMP	20eeac

    (100) 0x20eea0 INC	%RAX

    (100) 0x20eea3 CMP	%RAX,-0xa0(%RBP)

    (100) 0x20eeaa JLE	20ee30

    (100) 0x20eeac MOV	-0xa8(%RBP),%R10

    (100) 0x20eeb3 LEA	(%RAX,%RAX,2),%R12

    (100) 0x20eeb7 MOV	%R11,%R15

    (100) 0x20eeba LEA	(%R14,%R12,8),%RSI

    (100) 0x20eebe JMP	20eecc

      (101) 0x20eec0 INC	%R15

      (101) 0x20eec3 ADD	$0x18,%R10

      (101) 0x20eec7 CMP	%R13,%R15

      (101) 0x20eeca JGE	20eea0

      (101) 0x20eecc VMOVUPD	(%RSI),%XMM8

      (101) 0x20eed0 VSUBPD	-0x10(%R10),%XMM8,%XMM8

      (101) 0x20eed6 VMOVAPD	%XMM8,%XMM9

      (101) 0x20eedb VFMADD213SD	%XMM1,%XMM8,%XMM9

      (101) 0x20eee0 VSHUFPD	$0x1,%XMM8,%XMM8,%XMM10

      (101) 0x20eee6 VFMADD213SD	%XMM9,%XMM10,%XMM10

      (101) 0x20eeeb VMOVSD	0x10(%RSI),%XMM9

      (101) 0x20eef0 VSUBSD	(%R10),%XMM9,%XMM9

      (101) 0x20eef5 VFMADD231SD	%XMM9,%XMM9,%XMM10

      (101) 0x20eefa VUCOMISD	%XMM1,%XMM10

      (101) 0x20eefe JBE	20eec0

      (101) 0x20ef00 VMOVSD	(%RCX),%XMM11

      (101) 0x20ef04 VUCOMISD	%XMM10,%XMM11

      (101) 0x20ef09 JB	20eec0

      (101) 0x20ef0b VDIVSD	%XMM10,%XMM2,%XMM10

      (101) 0x20ef10 MOV	0x30(%R9),%RBX

      (101) 0x20ef14 VMOVSD	(%RBX,%RAX,8),%XMM13

      (101) 0x20ef19 VMULSD	%XMM10,%XMM10,%XMM11

      (101) 0x20ef1e VMULSD	%XMM11,%XMM10,%XMM11

      (101) 0x20ef23 VMULSD	(%RDX),%XMM11,%XMM11

      (101) 0x20ef27 VADDSD	%XMM3,%XMM11,%XMM12

      (101) 0x20ef2b VFMSUB213SD	(%R8),%XMM11,%XMM12

      (101) 0x20ef30 VFMADD231SD	%XMM4,%XMM12,%XMM13

      (101) 0x20ef35 VFMADD231SD	%XMM12,%XMM4,%XMM0

      (101) 0x20ef3a VMOVSD	%XMM13,(%RBX,%RAX,8)

      (101) 0x20ef3f VMULSD	(%RDI),%XMM5,%XMM13

      (101) 0x20ef43 MOV	0x28(%R9),%RBX

      (101) 0x20ef47 VMULSD	%XMM13,%XMM11,%XMM13

      (101) 0x20ef4c VFMADD213SD	%XMM7,%XMM6,%XMM11

      (101) 0x20ef51 VMULSD	%XMM13,%XMM10,%XMM10

      (101) 0x20ef56 VMULSD	%XMM10,%XMM11,%XMM10

      (101) 0x20ef5b VMOVDDUP	%XMM10,%XMM11

      (101) 0x20ef60 VFNMADD213PD	(%RBX,%R12,8),%XMM8,%XMM11

      (101) 0x20ef66 VMOVUPD	%XMM11,(%RBX,%R12,8)

      (101) 0x20ef6c VFNMADD213SD	0x10(%RBX,%R12,8),%XMM10,%XMM9

      (101) 0x20ef73 VMOVSD	%XMM9,0x10(%RBX,%R12,8)

      (101) 0x20ef7a JMP	20eec0

0x20ef80 XOR	%EAX,%EAX

0x20ef82 NOPW	%CS:(%RAX,%RAX,1)

  (97) 0x20ef90 MOV	-0x68(%RBP),%RSI

  (97) 0x20ef94 CMPL	$0,(%RSI,%RAX,4)

  (97) 0x20ef98 JS	20f07c

  (97) 0x20ef9e INC	%RAX

  (97) 0x20efa1 CMP	%RAX,-0x60(%RBP)

  (97) 0x20efa5 JNE	20ef90

0x20efa7 JMP	20edd0

/home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c: 173 - 216

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

173:    for (int iBox=0; iBox<s->boxes->nLocalBoxes; iBox++)

174:    {

175:       int nIBox = s->boxes->nAtoms[iBox];

176:

177:       // loop over neighbors of iBox

178:       for (int jTmp=0; jTmp<nNbrBoxes; jTmp++)

179:       {

180:          int jBox = s->boxes->nbrBoxes[iBox][jTmp];

181:

182:          assert(jBox>=0);

183:

184:          int nJBox = s->boxes->nAtoms[jBox];

185:

186:          // loop over atoms in iBox

187:          for (int iOff=MAXATOMS*iBox; iOff<(iBox*MAXATOMS+nIBox); iOff++)

188:          {

189:

190:             // loop over atoms in jBox

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

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	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	9.60
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.25
Bottlenecks
Function	ljForce.omp_outlined.5
Source	ljForce.c:173-175,ljForce.c:178-178,ljForce.c:216-216
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	1.88
CQA cycles if no scalar integer	1.88
CQA cycles if FP arith vectorized	1.88
CQA cycles if fully vectorized	0.20
Front-end cycles	1.50
P0 cycles	0.67
P1 cycles	0.67
P2 cycles	0.67
P3 cycles	0.67
P4 cycles	0.67
P5 cycles	0.67
P6 cycles	1.88
P7 cycles	1.88
P8 cycles	1.88
P9 cycles	1.88
P10 cycles	0.00
P11 cycles	0.00
P12 cycles	0.00
P13 cycles	0.00
P14 cycles	0.00
P15 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	0
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	13.00
Nb uops	12.00
Nb loads	4.50
Nb stores	3.00
Nb stack references	5.50
FLOP/cycle	0.00
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	0.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	28.89
Bytes prefetched	0.00
Bytes loaded	30.00
Bytes stored	24.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	0.00
Vectorization ratio load	NA
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	11.98
Vector-efficiency ratio load	NA
Vector-efficiency ratio store	12.50
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	9.38

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	11.08
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.20
Bottlenecks	P6, P7, P8, P9,
Function	ljForce.omp_outlined.5
Source	ljForce.c:173-175,ljForce.c:178-178,ljForce.c:216-216
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	2.25
CQA cycles if no scalar integer	2.25
CQA cycles if FP arith vectorized	2.25
CQA cycles if fully vectorized	0.20
Front-end cycles	1.88
P0 cycles	1.00
P1 cycles	1.00
P2 cycles	1.00
P3 cycles	1.00
P4 cycles	1.00
P5 cycles	1.00
P6 cycles	2.25
P7 cycles	2.25
P8 cycles	2.25
P9 cycles	2.25
P10 cycles	0.00
P11 cycles	0.00
P12 cycles	0.00
P13 cycles	0.00
P14 cycles	0.00
P15 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	0
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	16.00
Nb uops	15.00
Nb loads	5.00
Nb stores	4.00
Nb stack references	7.00
FLOP/cycle	0.00
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	0.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	28.44
Bytes prefetched	0.00
Bytes loaded	32.00
Bytes stored	32.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	0.00
Vectorization ratio load	NA
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	11.46
Vector-efficiency ratio load	NA
Vector-efficiency ratio store	12.50
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	9.38

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	8.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.33
Bottlenecks	P6, P7, P8, P9,
Function	ljForce.omp_outlined.5
Source	ljForce.c:173-175,ljForce.c:178-178,ljForce.c:216-216
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	1.50
CQA cycles if no scalar integer	1.50
CQA cycles if FP arith vectorized	1.50
CQA cycles if fully vectorized	0.19
Front-end cycles	1.13
P0 cycles	0.33
P1 cycles	0.33
P2 cycles	0.33
P3 cycles	0.33
P4 cycles	0.33
P5 cycles	0.33
P6 cycles	1.50
P7 cycles	1.50
P8 cycles	1.50
P9 cycles	1.50
P10 cycles	0.00
P11 cycles	0.00
P12 cycles	0.00
P13 cycles	0.00
P14 cycles	0.00
P15 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	0
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	10.00
Nb uops	9.00
Nb loads	4.00
Nb stores	2.00
Nb stack references	4.00
FLOP/cycle	0.00
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	0.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	29.33
Bytes prefetched	0.00
Bytes loaded	28.00
Bytes stored	16.00
Stride 0	1.00
Stride 1	0.00
Stride n	0.00
Stride unknown	0.00
Stride indirect	1.00
Vectorization ratio all	0.00
Vectorization ratio load	NA
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	NA
Vector-efficiency ratio all	12.50
Vector-efficiency ratio load	NA
Vector-efficiency ratio store	12.50
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	NA

Path /

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

Function	ljForce.omp_outlined.5
Source file and lines	ljForce.c:173-216
Module	exec

The loop is defined in /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:173-191,197-216.

Analyzed code is defined in /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:173-175,178,216.

Warnings:
Non-innermost loop: analyzing only self part (ignoring child loops).
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 11% of vector register length is used (average across all SSE/AVX instructions). By vectorizing your loop, you can lower the cost of an iteration from 1.88 to 0.20 cycles (9.60x speedup).

Details

All SSE/AVX instructions are used in scalar version (process only one data element in vector registers). Since your execution units are vector units, only a vectorized loop can use their full power.

Workaround

Try another compiler or update/tune your current one:
- recompile with ffast-math (included in Ofast) to extend loop vectorization to FP reductions.
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)

No data for this section

Conversion instructions

Detected expensive conversion instructions.

Details

CLTQ: 0 occurrences

Workaround

Avoid mixing data with different types. In particular, check if the type of constants is the same as array elements.

Type of elements and instruction set

No instructions are processing arithmetic or math operations on FP elements. This loop is probably writing/copying data or processing integer elements.

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

The binary loop does not contain any FP arithmetical operations. The binary loop is loading 30 bytes. The binary loop is storing 24 bytes.

General properties

nb instructions	13
nb uops	12
loop length	53.50
used x86 registers	4
used mmx registers	0
used xmm registers	0
used ymm registers	0
used zmm registers	0
nb stack references	5.50

Front-end

MACRO FUSION NOT POSSIBLE

micro-operation queue	1.50 cycles
front end	1.50 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	0.67	0.67	0.67	0.67	0.67	0.67	1.88	1.88	1.88	1.88	0.00	0.00	0.00	0.00	0.00	0.00
cycles	0.67	0.67	0.67	0.67	0.67	0.67	1.88	1.88	1.88	1.88	0.00	0.00	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	NA
Longest recurrence chain latency (RecMII)	0.00

Cycles summary

Front-end	1.50
Dispatch	1.88
Data deps.	0.00
Overall L1	1.88

Vectorization ratios

all	0%
load	NA (no load vectorizable/vectorized instructions)
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)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	0%

Vector efficiency ratios

all	11%
load	NA (no load vectorizable/vectorized instructions)
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)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	9%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.88 cycles. At this rate:

6% of peak load performance is reached (16.44 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
9% of peak store performance is reached (12.44 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Function	ljForce.omp_outlined.5
Source file and lines	ljForce.c:173-216
Module	exec

The loop is defined in /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:173-191,197-216.

Analyzed code is defined in /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:173-175,178,216.

Warnings:
Non-innermost loop: analyzing only self part (ignoring child loops).
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 11% of vector register length is used (average across all SSE/AVX instructions). By vectorizing your loop, you can lower the cost of an iteration from 2.25 to 0.20 cycles (11.08x speedup).

Details

All SSE/AVX instructions are used in scalar version (process only one data element in vector registers). Since your execution units are vector units, only a vectorized loop can use their full power.

Workaround

Try another compiler or update/tune your current one:
- recompile with ffast-math (included in Ofast) to extend loop vectorization to FP reductions.
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)

Execution units bottlenecks

Performance is limited by:

reading data from caches/RAM (load units are a bottleneck)
writing data to caches/RAM (the store unit is a bottleneck)

By removing all these bottlenecks, you can lower the cost of an iteration from 2.25 to 1.88 cycles (1.20x speedup).

Workaround

Read less array elements
Write less array elements
Provide more information to your compiler:
- hardcode the bounds of the corresponding 'for' loop
- use the 'restrict' C99 keyword

No data for this section

Conversion instructions

Detected expensive conversion instructions.

Details

CLTQ: 1 occurrences►
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:178

Workaround

Avoid mixing data with different types. In particular, check if the type of constants is the same as array elements.

Type of elements and instruction set

No instructions are processing arithmetic or math operations on FP elements. This loop is probably writing/copying data or processing integer elements.

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

The binary loop does not contain any FP arithmetical operations. The binary loop is loading 32 bytes. The binary loop is storing 32 bytes.

General properties

nb instructions	16
nb uops	15
loop length	60
used x86 registers	4
used mmx registers	0
used xmm registers	0
used ymm registers	0
used zmm registers	0
nb stack references	7

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	1.88 cycles
front end	1.88 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	2.25	2.25	2.25	2.25	0.00	0.00	0.00	0.00	0.00	0.00
cycles	1.00	1.00	1.00	1.00	1.00	1.00	2.25	2.25	2.25	2.25	0.00	0.00	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	NA
Longest recurrence chain latency (RecMII)	0.00

Cycles summary

Front-end	1.88
Dispatch	2.25
Data deps.	0.00
Overall L1	2.25

Vectorization ratios

all	0%
load	NA (no load vectorizable/vectorized instructions)
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)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	0%

Vector efficiency ratios

all	11%
load	NA (no load vectorizable/vectorized instructions)
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)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	9%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 2.25 cycles. At this rate:

5% of peak load performance is reached (14.22 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
11% of peak store performance is reached (14.22 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 20ede9

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	Latency	Recip. throughput	Vectorization
MOV -0x58(%RBP),%RAX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
MOV -0x78(%RBP),%R9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
MOV %RSI,-0x80(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	4	0.50	scal (12.5%)
MOV (%R9,%RSI,8),%R9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
MOV (%RAX,%RSI,4),%EAX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
MOV %R9,-0x68(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	4	0.50	scal (12.5%)
TEST %EAX,%EAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (6.3%)
JLE 20ef80 <ljForce.omp_outlined.5+0x2d0>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.50-1	N/A
MOVSXD -0x30(%RBP),%R9	1	0.17	0.17	0.17	0.17	0.17	0.17	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.25	N/A
SAL $0x6,%ESI	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
ADD %ESI,%EAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
XOR %ESI,%ESI	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.25	N/A
CLTQ																				scal (12.5%)
MOV %RAX,-0xa0(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	4	0.50	scal (12.5%)
MOV %R9,-0x90(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	4	0.50	scal (12.5%)
JMP 20ee40 <ljForce.omp_outlined.5+0x190>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	1	N/A

Function	ljForce.omp_outlined.5
Source file and lines	ljForce.c:173-216
Module	exec

The loop is defined in /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:173-191,197-216.

Analyzed code is defined in /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:173-175,178,216.

Warnings:
Non-innermost loop: analyzing only self part (ignoring child loops).
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. 8 data elements could be processed at once in vector registers. By vectorizing your loop, you can lower the cost of an iteration from 1.50 to 0.19 cycles (8.00x speedup).

Details

All SSE/AVX instructions are used in scalar version (process only one data element in vector registers). Since your execution units are vector units, only a vectorized loop can use their full power.

Workaround

Try another compiler or update/tune your current one:
- recompile with ffast-math (included in Ofast) to extend loop vectorization to FP reductions.
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)

Execution units bottlenecks

Performance is limited by:

reading data from caches/RAM (load units are a bottleneck)
writing data to caches/RAM (the store unit is a bottleneck)

By removing all these bottlenecks, you can lower the cost of an iteration from 1.50 to 1.13 cycles (1.33x speedup).

Workaround

Read less array elements
Write less array elements
Provide more information to your compiler:
- hardcode the bounds of the corresponding 'for' loop
- use the 'restrict' C99 keyword

No data for this section

Slow data structures access

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

Details

Irregular (variable stride) or indirect: 1 occurrence(s)

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

Workaround

Try to remove indirect accesses. If applicable, precompute elements out of the innermost loop.

Type of elements and instruction set

No instructions are processing arithmetic or math operations on FP elements. This loop is probably writing/copying data or processing integer elements.

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

The binary loop does not contain any FP arithmetical operations. The binary loop is loading 28 bytes. The binary loop is storing 16 bytes.

General properties

nb instructions	10
nb uops	9
loop length	47
used x86 registers	4
used mmx registers	0
used xmm registers	0
used ymm registers	0
used zmm registers	0
nb stack references	4

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	1.13 cycles
front end	1.13 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	0.33	0.33	0.33	0.33	0.33	0.33	1.50	1.50	1.50	1.50	0.00	0.00	0.00	0.00	0.00	0.00
cycles	0.33	0.33	0.33	0.33	0.33	0.33	1.50	1.50	1.50	1.50	0.00	0.00	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	NA
Longest recurrence chain latency (RecMII)	0.00

Cycles summary

Front-end	1.13
Dispatch	1.50
Data deps.	0.00
Overall L1	1.50

Vectorization ratios

all	0%
load	NA (no load vectorizable/vectorized instructions)
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)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

Vector efficiency ratios

all	12%
load	NA (no load vectorizable/vectorized instructions)
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)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.50 cycles. At this rate:

7% of peak load performance is reached (18.67 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
8% of peak store performance is reached (10.67 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 20ede9

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	Latency	Recip. throughput	Vectorization
MOV -0x58(%RBP),%RAX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV -0x78(%RBP),%R9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV %RSI,-0x80(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (12.5%)
MOV (%R9,%RSI,8),%R9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV (%RAX,%RSI,4),%EAX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV %R9,-0x68(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (12.5%)
TEST %EAX,%EAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.17	N/A
JLE 20ef80 <ljForce.omp_outlined.5+0x2d0>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.50-1	N/A
XOR %EAX,%EAX	1	0	0	0	0	0	0	0	0	0	0	0	0.25	N/A
NOPW %CS:(%RAX,%RAX,1)	0	0	0	0	0	0	0	0	0	0	0	0	0.09	N/A

Report Configuration

Vectorization

Details

Workaround

Conversion instructions

Details

Workaround

Type of elements and instruction set

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

Execution units bottlenecks

Workaround

Conversion instructions

Details

Workaround

Type of elements and instruction set

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

Vectorization

Details

Workaround

Execution units bottlenecks

Workaround

Slow data structures access

Details

Workaround

Type of elements and instruction set

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