OV - exec - Loop 144

0x411090 LEA	0x1(%RCX),%RAX

0x411094 CMP	%R9,%RCX

0x411097 MOV	%RAX,%RCX

0x41109a JE	411020

0x41109c LEA	(%RCX,%R8,1),%RAX

0x4110a0 LEA	(%RAX,%RAX,2),%RDI

0x4110a4 LEA	(%RBX,%RDI,8),%R15

0x4110a8 MOV	%RDX,%RSI

0x4110ab MOV	%R11,%R12

0x4110ae JMP	4110b9

  (145) 0x4110b0 ADD	$0x18,%RSI

  (145) 0x4110b4 DEC	%R12

  (145) 0x4110b7 JE	411090

  (145) 0x4110b9 MOVUPD	(%R15),%XMM11

  (145) 0x4110be MOVUPD	-0x10(%RSI),%XMM12

  (145) 0x4110c4 SUBPD	%XMM12,%XMM11

  (145) 0x4110c9 MOVAPD	%XMM11,%XMM12

  (145) 0x4110ce MULPD	%XMM11,%XMM12

  (145) 0x4110d3 MOVAPD	%XMM12,%XMM13

  (145) 0x4110d8 UNPCKHPD	%XMM12,%XMM13

  (145) 0x4110dd ADDSD	%XMM12,%XMM13

  (145) 0x4110e2 MOVSD	0x10(%R15),%XMM12

  (145) 0x4110e8 SUBSD	(%RSI),%XMM12

  (145) 0x4110ed MOVAPD	%XMM12,%XMM14

  (145) 0x4110f2 MULSD	%XMM12,%XMM14

  (145) 0x4110f7 ADDSD	%XMM13,%XMM14

  (145) 0x4110fc UCOMISD	%XMM2,%XMM14

  (145) 0x411101 JA	4110b0

  (145) 0x411103 UCOMISD	%XMM4,%XMM14

  (145) 0x411108 JBE	4110b0

  (145) 0x41110a MOVAPD	%XMM6,%XMM13

  (145) 0x41110f DIVSD	%XMM14,%XMM13

  (145) 0x411114 MOVAPD	%XMM13,%XMM14

  (145) 0x411119 MULSD	%XMM13,%XMM14

  (145) 0x41111e MOVDQA	%XMM1,%XMM15

  (145) 0x411123 MULSD	%XMM13,%XMM15

  (145) 0x411128 MULSD	%XMM14,%XMM15

  (145) 0x41112d MOVAPD	%XMM15,%XMM14

  (145) 0x411132 ADDSD	%XMM7,%XMM14

  (145) 0x411137 MULSD	%XMM15,%XMM14

  (145) 0x41113c SUBSD	0x10(%RBP),%XMM14

  (145) 0x411142 MULSD	%XMM8,%XMM14

  (145) 0x411147 MOV	0x30(%R13),%R10

  (145) 0x41114b MOVSD	(%R10,%RAX,8),%XMM0

  (145) 0x411151 ADDSD	%XMM14,%XMM0

  (145) 0x411156 MOVSD	%XMM0,(%R10,%RAX,8)

  (145) 0x41115c MULSD	%XMM3,%XMM13

  (145) 0x411161 MULSD	%XMM15,%XMM13

  (145) 0x411166 MULSD	%XMM9,%XMM15

  (145) 0x41116b ADDSD	%XMM10,%XMM15

  (145) 0x411170 MULSD	%XMM15,%XMM13

  (145) 0x411175 MOV	0x28(%R13),%R10

  (145) 0x411179 MOVAPD	%XMM13,%XMM0

  (145) 0x41117e UNPCKLPD	%XMM13,%XMM0

  (145) 0x411183 MULPD	%XMM11,%XMM0

  (145) 0x411188 MOVUPD	(%R10,%RDI,8),%XMM11

  (145) 0x41118e ADDPD	%XMM0,%XMM11

  (145) 0x411193 MOVUPD	%XMM11,(%R10,%RDI,8)

  (145) 0x411199 MULSD	%XMM12,%XMM13

  (145) 0x41119e ADDSD	0x10(%R10,%RDI,8),%XMM13

  (145) 0x4111a5 MOVSD	%XMM13,0x10(%R10,%RDI,8)

  (145) 0x4111ac ADDSD	%XMM14,%XMM5

  (145) 0x4111b1 JMP	4110b0

/beegfs/hackathon/users/eoseret/qaas_runs/170-850-7424/intel/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c: 187 - 216

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

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;

Path /

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	1.80
Bottlenecks
Function	ljForce.extracted
Source	ljForce.c:187-187,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	3.00
CQA cycles if no scalar integer	3.00
CQA cycles if FP arith vectorized	3.00
CQA cycles if fully vectorized	0.38
Front-end cycles	1.67
DIV/SQRT cycles	1.50
P0 cycles	1.50
P1 cycles	1.50
P2 cycles	1.50
P3 cycles	1.00
P4 cycles	0.00
P5 cycles	0.00
P6 cycles	0.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	0.00
P11 cycles	0.00
P12 cycles	0.00
P13 cycles	0.00
Inter-iter dependencies cycles	3
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	10.00
Nb uops	10.00
Nb loads	0.00
Nb stores	0.00
Nb stack references	0.00
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	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	1.00
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	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	8.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.80
Bottlenecks
Function	ljForce.extracted
Source	ljForce.c:187-187,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	3.00
CQA cycles if no scalar integer	3.00
CQA cycles if FP arith vectorized	3.00
CQA cycles if fully vectorized	0.38
Front-end cycles	1.67
DIV/SQRT cycles	1.50
P0 cycles	1.50
P1 cycles	1.50
P2 cycles	1.50
P3 cycles	1.00
P4 cycles	0.00
P5 cycles	0.00
P6 cycles	0.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	0.00
P11 cycles	0.00
P12 cycles	0.00
P13 cycles	0.00
Inter-iter dependencies cycles	3
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	10.00
Nb uops	10.00
Nb loads	0.00
Nb stores	0.00
Nb stack references	0.00
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	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	1.00
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	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

Path /

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

Function	ljForce.extracted
Source file and lines	ljForce.c:187-216
Module	exec

The loop is defined in /beegfs/hackathon/users/eoseret/qaas_runs/170-850-7424/intel/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:187-191,197-216.

Analyzed code is defined in /beegfs/hackathon/users/eoseret/qaas_runs/170-850-7424/intel/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:187,197.

Warnings:
Non-innermost loop: analyzing only self part (ignoring child loops).
This loop has 3 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 3.00 to 0.37 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

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)

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 does not load or store any data.

General properties

nb instructions	10
nb uops	10
loop length	32
used x86 registers	11
used mmx registers	0
used xmm registers	0
used ymm registers	0
used zmm registers	0
nb stack references	0

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	1.67 cycles
front end	1.67 cycles

Back-end

	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	AGU0	AGU1	AGU2	FP0	FP1	FP2	FP3	FP4	FP5
uops	1.50	1.50	1.50	1.50	1.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00
cycles	1.50	1.50	1.50	1.50	1.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00

Execution ports to units layout:

ALU0/BRU0: ALU
ALU1: ALU
ALU2: ALU
ALU3: ALU
BRU1:
AGU0 (256 bits): store address, load
AGU1 (256 bits): store address, load
AGU2 (256 bits): store address, load
FP0 (256 bits): VPU, DIV/SQRT
FP1 (256 bits): VPU, DIV/SQRT
FP2 (256 bits): VPU
FP3 (256 bits): VPU
FP4 (256 bits): FP store data
FP5 (256 bits): FP store data

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	3.00

Cycles summary

Front-end	1.67
Dispatch	1.50
Data deps.	3.00
Overall L1	3.00

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 3.00 cycles. At this rate:

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 41109c

Instruction	Nb FU	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	Latency	Recip. throughput
LEA 0x1(%RCX),%RAX	1	0.25	0.25	0.25	0.25	0	1	0.25
CMP %R9,%RCX	1	0.25	0.25	0.25	0.25	0	1	0.25
MOV %RAX,%RCX	1	0	0	0	0	0	0	0.17
JE 411020 <ljForce.extracted+0x1f0>	1	0.50	0	0	0	0.50	1	0.50-1
LEA (%RCX,%R8,1),%RAX	1	0.25	0.25	0.25	0.25	0	1	0.25
LEA (%RAX,%RAX,2),%RDI	1	0.25	0.25	0.25	0.25	0	1	0.25
LEA (%RBX,%RDI,8),%R15	1	0.25	0.25	0.25	0.25	0	1	0.25
MOV %RDX,%RSI	1	0	0	0	0	0	0	0.17
MOV %R11,%R12	1	0	0	0	0	0	0	0.17
JMP 4110b9 <ljForce.extracted+0x289>	1	0.50	0	0	0	0.50	0	1

Function	ljForce.extracted
Source file and lines	ljForce.c:187-216
Module	exec

The loop is defined in /beegfs/hackathon/users/eoseret/qaas_runs/170-850-7424/intel/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:187-191,197-216.

Analyzed code is defined in /beegfs/hackathon/users/eoseret/qaas_runs/170-850-7424/intel/CoMD/build/CoMD/CoMD/src-openmp/ljForce.c:187,197.

Warnings:
Non-innermost loop: analyzing only self part (ignoring child loops).
This loop has 3 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 3.00 to 0.37 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

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)

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 does not load or store any data.

General properties

nb instructions	10
nb uops	10
loop length	32
used x86 registers	11
used mmx registers	0
used xmm registers	0
used ymm registers	0
used zmm registers	0
nb stack references	0

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	1.67 cycles
front end	1.67 cycles

Back-end

	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	AGU0	AGU1	AGU2	FP0	FP1	FP2	FP3	FP4	FP5
uops	1.50	1.50	1.50	1.50	1.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00
cycles	1.50	1.50	1.50	1.50	1.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00

Execution ports to units layout:

ALU0/BRU0: ALU
ALU1: ALU
ALU2: ALU
ALU3: ALU
BRU1:
AGU0 (256 bits): store address, load
AGU1 (256 bits): store address, load
AGU2 (256 bits): store address, load
FP0 (256 bits): VPU, DIV/SQRT
FP1 (256 bits): VPU, DIV/SQRT
FP2 (256 bits): VPU
FP3 (256 bits): VPU
FP4 (256 bits): FP store data
FP5 (256 bits): FP store data

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	3.00

Cycles summary

Front-end	1.67
Dispatch	1.50
Data deps.	3.00
Overall L1	3.00

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 3.00 cycles. At this rate:

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 41109c

Instruction	Nb FU	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	Latency	Recip. throughput
LEA 0x1(%RCX),%RAX	1	0.25	0.25	0.25	0.25	0	1	0.25
CMP %R9,%RCX	1	0.25	0.25	0.25	0.25	0	1	0.25
MOV %RAX,%RCX	1	0	0	0	0	0	0	0.17
JE 411020 <ljForce.extracted+0x1f0>	1	0.50	0	0	0	0.50	1	0.50-1
LEA (%RCX,%R8,1),%RAX	1	0.25	0.25	0.25	0.25	0	1	0.25
LEA (%RAX,%RAX,2),%RDI	1	0.25	0.25	0.25	0.25	0	1	0.25
LEA (%RBX,%RDI,8),%R15	1	0.25	0.25	0.25	0.25	0	1	0.25
MOV %RDX,%RSI	1	0	0	0	0	0	0	0.17
MOV %R11,%R12	1	0	0	0	0	0	0	0.17
JMP 4110b9 <ljForce.extracted+0x289>	1	0.50	0	0	0	0.50	0	1

Report Configuration

Vectorization

Details

Workaround

Execution units bottlenecks

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

Vectorization

Details

Workaround

Execution units bottlenecks

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