OV - exec - Loop 95

0x40b510 INC	%R10

0x40b513 CMP	$0x1b,%R10

0x40b517 JE	40b4c0

0x40b519 MOVSXD	(%R8,%R10,4),%RAX

0x40b51d TEST	%RAX,%RAX

0x40b520 JS	40b724

0x40b526 MOV	-0x60(%RBP),%RCX

0x40b52a MOV	(%RCX,%RAX,4),%ECX

0x40b52d TEST	%ECX,%ECX

0x40b52f JLE	40b510

0x40b531 MOV	-0x58(%RBP),%RDX

0x40b535 SAL	$0x6,%EAX

0x40b538 DEC	%ECX

0x40b53a XOR	%EBX,%EBX

0x40b53c LEA	(%RCX,%RCX,2),%RCX

0x40b540 LEA	(%RAX,%RAX,2),%RAX

0x40b544 LEA	0x18(,%RCX,8),%R14

0x40b54c MOV	0x20(%RDX),%R11

0x40b550 MOV	0x18(%R11),%R12

0x40b554 LEA	0x10(%R12,%RAX,8),%R13

0x40b559 JMP	40b56c

  (96) 0x40b560 LEA	0x1(%RBX),%RAX

  (96) 0x40b564 CMP	%R9,%RBX

  (96) 0x40b567 MOV	%RAX,%RBX

  (96) 0x40b56a JE	40b510

  (96) 0x40b56c LEA	(%RBX,%RDI,1),%RSI

  (96) 0x40b570 XOR	%R15D,%R15D

  (96) 0x40b573 LEA	(%RSI,%RSI,2),%RCX

  (96) 0x40b577 LEA	(%R12,%RCX,8),%RAX

  (96) 0x40b57b JMP	40b589

    (97) 0x40b580 ADD	$0x18,%R15

    (97) 0x40b584 CMP	%R15,%R14

    (97) 0x40b587 JE	40b560

    (97) 0x40b589 VMOVUPD	(%RAX),%XMM11

    (97) 0x40b58d VMOVSD	0x10(%RAX),%XMM12

    (97) 0x40b592 VSUBPD	-0x10(%R13,%R15,1),%XMM11,%XMM11

    (97) 0x40b599 VSUBSD	(%R13,%R15,1),%XMM12,%XMM12

    (97) 0x40b5a0 VMULPD	%XMM11,%XMM11,%XMM13

    (97) 0x40b5a5 VSHUFPD	$0x1,%XMM13,%XMM13,%XMM14

    (97) 0x40b5ab VADDSD	%XMM14,%XMM13,%XMM13

    (97) 0x40b5b0 VFMADD231SD	%XMM12,%XMM12,%XMM13

    (97) 0x40b5b5 VUCOMISD	%XMM13,%XMM2

    (97) 0x40b5ba JB	40b580

    (97) 0x40b5bc VUCOMISD	%XMM4,%XMM13

    (97) 0x40b5c0 JBE	40b580

    (97) 0x40b5c2 VDIVSD	%XMM13,%XMM6,%XMM13

    (97) 0x40b5c7 MOV	0x30(%R11),%RDX

    (97) 0x40b5cb VMULSD	%XMM13,%XMM13,%XMM14

    (97) 0x40b5d0 VMULSD	%XMM1,%XMM13,%XMM15

    (97) 0x40b5d4 VMULSD	%XMM3,%XMM13,%XMM13

    (97) 0x40b5d8 VMULSD	%XMM15,%XMM14,%XMM14

    (97) 0x40b5dd VADDSD	%XMM7,%XMM14,%XMM15

    (97) 0x40b5e1 VMULSD	%XMM14,%XMM13,%XMM13

    (97) 0x40b5e6 VFNMADD213SD	%XMM0,%XMM14,%XMM15

    (97) 0x40b5eb VMULSD	%XMM8,%XMM15,%XMM15

    (97) 0x40b5f0 VADDSD	(%RDX,%RSI,8),%XMM15,%XMM16

    (97) 0x40b5f7 VADDSD	%XMM5,%XMM15,%XMM5

    (97) 0x40b5fb VMOVSD	%XMM16,(%RDX,%RSI,8)

    (97) 0x40b602 VMOVAPD	%XMM9,%XMM16

    (97) 0x40b608 VFMADD213SD	%XMM10,%XMM14,%XMM16

    (97) 0x40b60e MOV	0x28(%R11),%RDX

    (97) 0x40b612 VMULSD	%XMM16,%XMM13,%XMM13

    (97) 0x40b618 VMOVDDUP	%XMM13,%XMM14

    (97) 0x40b61d VFMADD213PD	(%RDX,%RCX,8),%XMM14,%XMM11

    (97) 0x40b623 VMOVUPD	%XMM11,(%RDX,%RCX,8)

    (97) 0x40b628 VFMADD213SD	0x10(%RDX,%RCX,8),%XMM13,%XMM12

    (97) 0x40b62f VMOVSD	%XMM12,0x10(%RDX,%RCX,8)

    (97) 0x40b635 JMP	40b580

/beegfs/hackathon/users/eoseret/qaas_runs_test/gmz12.benchmarkcenter.megware.com/177-374-1600/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c: 178 - 216

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

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
►97.40+	__kmp_invoke_microtask		libiomp5.so
○	__kmp_fork_call		libiomp5.so
○	__kmpc_fork_call		libiomp5.so
○	ljForce	ljForce.c:172	exec
○	timestep	timestep.c:51	exec
○	main	CoMD.c:125	exec
○1.23	__kmp_invoke_microtask		libiomp5.so

Coverage (%)	Name	Source Location	Module
►100.00+	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so

Coverage (%)	Name	Source Location	Module
►100.00+	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so

Coverage (%)	Name	Source Location	Module
►100.00+	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so

Coverage (%)	Name	Source Location	Module
►100.00+	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so

Coverage (%)	Name	Source Location	Module
►100.00+	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so

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	13.25
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.09
Bottlenecks
Function	ljForce.extracted
Source	ljForce.c:178-184,ljForce.c:191-191,ljForce.c:197-197
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	2.35
CQA cycles if no scalar integer	2.35
CQA cycles if FP arith vectorized	2.35
CQA cycles if fully vectorized	0.18
Front-end cycles	2.34
P0 cycles	2.17
P1 cycles	2.17
P2 cycles	2.17
P3 cycles	2.17
P4 cycles	2.17
P5 cycles	2.17
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	18.25
Nb uops	18.75
Nb loads	5.25
Nb stores	0.00
Nb stack references	1.75
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	14.34
Bytes prefetched	0.00
Bytes loaded	34.00
Bytes stored	0.00
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	NA
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	10.16
Vector-efficiency ratio load	NA
Vector-efficiency ratio store	NA
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	10.16

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	1.04
Bottlenecks	P0, P1, P2, P3, P4, P5,
Function	ljForce.extracted
Source	ljForce.c:178-184,ljForce.c:191-191,ljForce.c:197-197
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	1.17
CQA cycles if no scalar integer	1.17
CQA cycles if FP arith vectorized	1.17
CQA cycles if fully vectorized	0.15
Front-end cycles	1.13
P0 cycles	1.17
P1 cycles	1.17
P2 cycles	1.17
P3 cycles	1.17
P4 cycles	1.17
P5 cycles	1.17
P6 cycles	0.75
P7 cycles	0.75
P8 cycles	0.75
P9 cycles	0.75
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	3.00
Nb stores	0.00
Nb stack references	1.00
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	13.71
Bytes prefetched	0.00
Bytes loaded	16.00
Bytes stored	0.00
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	NA
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	12.50
Vector-efficiency ratio load	NA
Vector-efficiency ratio store	NA
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	12.50

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	14.61
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.10
Bottlenecks	micro-operation queue,
Function	ljForce.extracted
Source	ljForce.c:178-184,ljForce.c:191-191,ljForce.c:197-197
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	2.75
CQA cycles if no scalar integer	2.75
CQA cycles if FP arith vectorized	2.75
CQA cycles if fully vectorized	0.19
Front-end cycles	2.75
P0 cycles	2.50
P1 cycles	2.50
P2 cycles	2.50
P3 cycles	2.50
P4 cycles	2.50
P5 cycles	2.50
P6 cycles	1.75
P7 cycles	1.75
P8 cycles	1.75
P9 cycles	1.75
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	21.00
Nb uops	22.00
Nb loads	6.00
Nb stores	0.00
Nb stack references	2.00
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	14.55
Bytes prefetched	0.00
Bytes loaded	40.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	0.00
Stride n	0.00
Stride unknown	2.00
Stride indirect	0.00
Vectorization ratio all	0.00
Vectorization ratio load	NA
Vectorization ratio store	NA
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	9.38
Vector-efficiency ratio load	NA
Vector-efficiency ratio store	NA
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

Path /

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

Function	ljForce.extracted
Source file and lines	ljForce.c:178-216
Module	exec

The loop is defined in /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz12.benchmarkcenter.megware.com/177-374-1600/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:178-191,197-216.

Analyzed code is defined in /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz12.benchmarkcenter.megware.com/177-374-1600/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:178-184,191,197.

Warnings:
Non-innermost loop: analyzing only self part (ignoring child loops).
This loop has 4 execution paths.

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 10% 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 2.35 to 0.18 cycles (13.25x 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

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 34 bytes.

General properties

nb instructions	18.25
nb uops	18.75
loop length	64.50
used x86 registers	9.50
used mmx registers	0
used xmm registers	0
used ymm registers	0
used zmm registers	0
nb stack references	1.75

Front-end

MACRO FUSION NOT POSSIBLE

micro-operation queue	2.34 cycles
front end	2.34 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	2.17	2.17	2.17	2.17	2.17	2.17	1.50	1.50	1.50	1.50	0.00	0.00	0.00	0.00	0.00	0.00
cycles	2.17	2.17	2.17	2.17	2.17	2.17	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	2.34
Dispatch	2.17
Data deps.	0.00
Overall L1	2.35

Vectorization ratios

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

Vector efficiency ratios

all	10%
load	NA (no load vectorizable/vectorized instructions)
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)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	10%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 2.35 cycles. At this rate:

5% of peak load performance is reached (14.34 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))

Function	ljForce.extracted
Source file and lines	ljForce.c:178-216
Module	exec

The loop is defined in /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz12.benchmarkcenter.megware.com/177-374-1600/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:178-191,197-216.

Analyzed code is defined in /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz12.benchmarkcenter.megware.com/177-374-1600/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:178-184,191,197.

Warnings:
Non-innermost loop: analyzing only self part (ignoring child loops).
This loop has 4 execution paths.

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 1.17 to 0.15 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

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

No data for this section

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 16 bytes.

General properties

nb instructions	10
nb uops	9
loop length	33
used x86 registers	5
used mmx registers	0
used xmm registers	0
used ymm registers	0
used zmm registers	0
nb stack references	1

Front-end

ASSUMED MACRO FUSION 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	1.17	1.17	1.17	1.17	1.17	1.17	0.75	0.75	0.75	0.75	0.00	0.00	0.00	0.00	0.00	0.00
cycles	1.17	1.17	1.17	1.17	1.17	1.17	0.75	0.75	0.75	0.75	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	1.17
Data deps.	0.00
Overall L1	1.17

Vectorization ratios

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

Vector efficiency ratios

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.17 cycles. At this rate:

5% of peak load performance is reached (13.71 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 40b519

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	Latency	Recip. throughput	Vectorization
INC %R10	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.17	N/A
CMP $0x1b,%R10	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.17	scal (12.5%)
JE 40b4c0 <ljForce.extracted+0x170>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0.33-1	N/A
MOVSXD (%R8,%R10,4),%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0.25	0.25	0.25	0.25	1	0.25	N/A
TEST %RAX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.17	scal (12.5%)
JS 40b724 <ljForce.extracted+0x3d4>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0.33-1	N/A
MOV -0x60(%RBP),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV (%RCX,%RAX,4),%ECX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
TEST %ECX,%ECX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.17	N/A
JLE 40b510 <ljForce.extracted+0x1c0>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0.33-1	N/A

Function	ljForce.extracted
Source file and lines	ljForce.c:178-216
Module	exec

The loop is defined in /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz12.benchmarkcenter.megware.com/177-374-1600/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:178-191,197-216.

Analyzed code is defined in /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz12.benchmarkcenter.megware.com/177-374-1600/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:178-184,191,197.

Warnings:
Non-innermost loop: analyzing only self part (ignoring child loops).
This loop has 4 execution paths.

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 9% 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 2.75 to 0.19 cycles (14.61x 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

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

No data for this section

Complex instructions

Detected COMPLEX INSTRUCTIONS.

Details

These instructions generate more than one micro-operation and only one of them can be decoded during a cycle and the extra micro-operations increase pressure on execution units.

LEA: 1 occurrences►
- /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz12.benchmarkcenter.megware.com/177-374-1600/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:197

Slow data structures access

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

Details

Constant unknown stride: 2 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)

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 40 bytes.

General properties

nb instructions	21
nb uops	22
loop length	75
used x86 registers	11
used mmx registers	0
used xmm registers	0
used ymm registers	0
used zmm registers	0
nb stack references	2

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	2.75 cycles
front end	2.75 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	2.50	2.50	2.50	2.50	2.50	2.50	1.75	1.75	1.75	1.75	0.00	0.00	0.00	0.00	0.00	0.00
cycles	2.50	2.50	2.50	2.50	2.50	2.50	1.75	1.75	1.75	1.75	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	2.75
Dispatch	2.50
Data deps.	0.00
Overall L1	2.75

Vectorization ratios

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

Vector efficiency ratios

all	9%
load	NA (no load vectorizable/vectorized instructions)
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)
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 2.75 cycles. At this rate:

5% of peak load performance is reached (14.55 out of 256.00 bytes loaded 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 2.75 to 2.50 cycles (1.10x speedup).

ASM code

In the binary file, the address of the loop is: 40b519

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	Latency	Recip. throughput	Vectorization
INC %R10	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.17	N/A
CMP $0x1b,%R10	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.17	scal (12.5%)
JE 40b4c0 <ljForce.extracted+0x170>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0.33-1	N/A
MOVSXD (%R8,%R10,4),%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0.25	0.25	0.25	0.25	1	0.25	N/A
TEST %RAX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.17	scal (12.5%)
JS 40b724 <ljForce.extracted+0x3d4>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0.33-1	N/A
MOV -0x60(%RBP),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV (%RCX,%RAX,4),%ECX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
TEST %ECX,%ECX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.17	scal (6.3%)
JLE 40b510 <ljForce.extracted+0x1c0>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0.33-1	N/A
MOV -0x58(%RBP),%RDX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
SAL $0x6,%EAX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.33	N/A
DEC %ECX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.17	N/A
XOR %EBX,%EBX	1	0	0	0	0	0	0	0	0	0	0	0	0.17	scal (6.3%)
LEA (%RCX,%RCX,2),%RCX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.17	N/A
LEA (%RAX,%RAX,2),%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.17	N/A
LEA 0x18(,%RCX,8),%R14	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.17	N/A
MOV 0x20(%RDX),%R11	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV 0x18(%R11),%R12	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
LEA 0x10(%R12,%RAX,8),%R13	2	0	0	0.50	0.50	0	0	0.25	0.25	0.25	0.25	2	0.50	N/A
JMP 40b56c <ljForce.extracted+0x21c>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0.50	N/A

▶Scalability Runs Description

Run 8x1	Number processes: 8Number nodes: NANumber processes per node: 8Run Command: <executable> -x 200 -y 200 -z 200 --xproc=2 --yproc=2 --zproc=2MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=24 --rank-by fill --report-bindings Dataset: Run Directory: /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz12.benchmarkcenter.megware.com/177-374-1600/CoMD/run/oneview_runs/multicore/icx_1/oneview_run_1773746383OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threadsOMP_NUM_THREADS: 1
Run 8x8	Number processes: 8Number processes per node: 8Run Command: <executable> -x 200 -y 200 -z 200 --xproc=2 --yproc=2 --zproc=2MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=24 --rank-by fill --report-bindings Dataset: Run Directory: /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz12.benchmarkcenter.megware.com/177-374-1600/CoMD/run/oneview_runs/multicore/icx_1/oneview_run_1773746383OMP_NUM_THREADS: 8OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 8x12	Number processes: 8Number processes per node: 8Run Command: <executable> -x 200 -y 200 -z 200 --xproc=2 --yproc=2 --zproc=2MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=24 --rank-by fill --report-bindings Dataset: Run Directory: /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz12.benchmarkcenter.megware.com/177-374-1600/CoMD/run/oneview_runs/multicore/icx_1/oneview_run_1773746383OMP_NUM_THREADS: 12OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 8x16	Number processes: 8Number processes per node: 8Run Command: <executable> -x 200 -y 200 -z 200 --xproc=2 --yproc=2 --zproc=2MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=24 --rank-by fill --report-bindings Dataset: Run Directory: /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz12.benchmarkcenter.megware.com/177-374-1600/CoMD/run/oneview_runs/multicore/icx_1/oneview_run_1773746383OMP_NUM_THREADS: 16OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 8x20	Number processes: 8Number processes per node: 8Run Command: <executable> -x 200 -y 200 -z 200 --xproc=2 --yproc=2 --zproc=2MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=24 --rank-by fill --report-bindings Dataset: Run Directory: /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz12.benchmarkcenter.megware.com/177-374-1600/CoMD/run/oneview_runs/multicore/icx_1/oneview_run_1773746383OMP_NUM_THREADS: 20OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 8x24	Number processes: 8Number processes per node: 8Run Command: <executable> -x 200 -y 200 -z 200 --xproc=2 --yproc=2 --zproc=2MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=24 --rank-by fill --report-bindings Dataset: Run Directory: /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz12.benchmarkcenter.megware.com/177-374-1600/CoMD/run/oneview_runs/multicore/icx_1/oneview_run_1773746383OMP_NUM_THREADS: 24OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads

(8x1) Efficiency	(8x1) Potential Speed-Up (%)	(8x8) Efficiency	(8x8) Potential Speed-Up (%)	(8x12) Efficiency	(8x12) Potential Speed-Up (%)	(8x16) Efficiency	(8x16) Potential Speed-Up (%)	(8x20) Efficiency	(8x20) Potential Speed-Up (%)	(8x24) Efficiency	(8x24) Potential Speed-Up (%)
1	0	1.56	0	1.23	0	1.02	0	0.92	0.02	0.87	0.03

Speed-Up and Efficiency

▶Detailed Speed-Up and Efficiency

Run	Number of threads	Efficiency (ideal is 1)	Speedup	Ideal Speedup	Time (s)	Coverage (%)
8x1	8	1	1	1	1.4699997901917	0.34232252836227
8x8	40	1.56	7.78	5	0.25499999523163	0.31013137102127
8x12	72	1.23	11.09	9	0.19499999284744	0.28797498345375
8x16	104	1.02	13.32	13	0.15500000119209	0.28841710090637
8x20	136	0.92	15.67	17	0.19000001251698	0.26327008008957
8x24	167	0.87	18.34	21	0.12000000476837	0.24554267525673

Report Configuration

Vectorization

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

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

Complex instructions

Details

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

▶Scalability Runs Description

▼Columns Filter

Speed-Up and Efficiency

▶Detailed Speed-Up and Efficiency