OV - exec - Loop 171

	Loop Id: 171	Module: exec	Source: timestep.c:110-116	Coverage: 0.05%

0x41ab90 MOVUPD	0x40(%R14),%XMM7    [4]

0x41ab96 MOVUPD	0x50(%R14),%XMM12    [4]

0x41ab9c MOVUPD	(%R14),%XMM11    [4]

0x41aba1 MOVUPD	0x10(%R14),%XMM8    [4]

0x41aba7 MOVDQU	(%R15,%RDX,4),%XMM9    [5]

0x41abad MOVUPD	0x20(%R14),%XMM13    [4]

0x41abb3 MOVUPD	0x30(%R14),%XMM10    [4]

0x41abb9 MULPD	%XMM13,%XMM13

0x41abbe MULPD	%XMM12,%XMM12

0x41abc3 MULPD	%XMM11,%XMM11

0x41abc8 MULPD	%XMM8,%XMM8

0x41abcd MULPD	%XMM7,%XMM7

0x41abd1 MULPD	%XMM10,%XMM10

0x41abd6 MOVAPD	%XMM7,%XMM14

0x41abdb MOVSD	%XMM10,%XMM14

0x41abe0 MOVAPD	%XMM8,%XMM15

0x41abe5 MOVSD	%XMM11,%XMM15

0x41abea SHUFPD	$0x1,%XMM12,%XMM10

0x41abf0 SHUFPD	$0x2,%XMM13,%XMM8

0x41abf6 SHUFPD	$0x2,%XMM12,%XMM7

0x41abfc SHUFPD	$0x1,%XMM13,%XMM11

0x41ac02 XORPD	%XMM12,%XMM12

0x41ac07 PCMPGTD	%XMM9,%XMM12

0x41ac0c PSHUFD	$-0x12,%XMM9,%XMM13

0x41ac12 ADDPD	%XMM14,%XMM10

0x41ac17 PUNPCKLDQ	%XMM12,%XMM9

0x41ac1c PXOR	%XMM12,%XMM12

0x41ac21 PCMPGTD	%XMM13,%XMM12

0x41ac26 ADDPD	%XMM15,%XMM11

0x41ac2b PUNPCKLDQ	%XMM12,%XMM13

0x41ac30 PSLLQ	$0x4,%XMM13

0x41ac36 PADDQ	%XMM4,%XMM13

0x41ac3b ADDPD	%XMM11,%XMM8

0x41ac40 PSLLQ	$0x4,%XMM9

0x41ac46 PADDQ	%XMM4,%XMM9

0x41ac4b MOVQ	%XMM9,%RAX

0x41ac50 ADDPD	%XMM10,%XMM7

0x41ac55 PSHUFD	$-0x12,%XMM9,%XMM9

0x41ac5b MOVQ	%XMM9,%RCX

0x41ac60 MOVQ	%XMM13,%R12

0x41ac65 MULPD	%XMM2,%XMM8

0x41ac6a PSHUFD	$-0x12,%XMM13,%XMM9

0x41ac70 MOVQ	%XMM9,%R10

0x41ac75 MOVSD	(%RAX),%XMM9    [2]

0x41ac7a MULPD	%XMM2,%XMM7

0x41ac7e MOVHPD	(%RCX),%XMM9    [6]

0x41ac83 MOVSD	(%R12),%XMM10    [3]

0x41ac89 DIVPD	%XMM9,%XMM8

0x41ac8e MOVHPD	(%R10),%XMM10    [1]

0x41ac93 DIVPD	%XMM10,%XMM7

0x41ac98 ADDPD	%XMM8,%XMM5

0x41ac9d ADDPD	%XMM7,%XMM6

0x41aca1 ADD	$0x60,%R14

0x41aca5 ADD	$0x4,%RDX

0x41aca9 CMP	%EBX,%EDX

0x41acab JLE	41ab90

/home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c: 110 - 116

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

110:       for (int iOff=MAXATOMS*iBox,ii=0; ii<s->boxes->nAtoms[iBox]; ii++,iOff++)

111:       {

112:          int iSpecies = s->atoms->iSpecies[iOff];

113:          real_t invMass = 0.5/s->species[iSpecies].mass;

114:          kenergy += ( s->atoms->p[iOff][0] * s->atoms->p[iOff][0] +

115:          s->atoms->p[iOff][1] * s->atoms->p[iOff][1] +

116:          s->atoms->p[iOff][2] * s->atoms->p[iOff][2] )*invMass;

Coverage (%)	Name	Source Location	Module
►91.51+	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►4.82+	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
►1.27+	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_fork_call		libiomp5.so
○	__kmpc_fork_call		libiomp5.so
○	timestep	timestep.c:120	exec
○	main	CoMD.c:125	exec
►1.27+	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6

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	1.59
CQA speedup if fully vectorized	4.38
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	NA
Bottlenecks	P10, P11, P12, P13,
Function	kineticEnergy.extracted
Source	timestep.c:110-116
Source loop unroll info	unrolled by 4
Source loop unroll confidence level	max
Unroll/vectorization loop type	main
Unroll factor	4
CQA cycles	8.75
CQA cycles if no scalar integer	8.75
CQA cycles if FP arith vectorized	5.50
CQA cycles if fully vectorized	2.00
Front-end cycles	6.88
P0 cycles	0.50
P1 cycles	0.50
P2 cycles	0.50
P3 cycles	0.50
P4 cycles	0.50
P5 cycles	0.50
P6 cycles	2.75
P7 cycles	2.75
P8 cycles	2.75
P9 cycles	2.75
P10 cycles	8.75
P11 cycles	8.75
P12 cycles	8.75
P13 cycles	8.75
P14 cycles	2.00
P15 cycles	2.00
DIV/SQRT cycles	8.00
Inter-iter dependencies cycles	2
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	56.00
Nb uops	55.00
Nb loads	11.00
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	3.66
Nb FLOP add-sub	12.00
Nb FLOP mul	16.00
Nb FLOP fma	0.00
Nb FLOP div	4.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	16.46
Bytes prefetched	0.00
Bytes loaded	144.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	2.00
Stride n	0.00
Stride unknown	4.00
Stride indirect	0.00
Vectorization ratio all	80.77
Vectorization ratio load	63.64
Vectorization ratio store	NA
Vectorization ratio mul	100.00
Vectorization ratio add_sub	100.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	100.00
Vectorization ratio other	73.91
Vector-efficiency ratio all	22.12
Vector-efficiency ratio load	20.45
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	25.00
Vector-efficiency ratio add_sub	25.00
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	25.00
Vector-efficiency ratio other	20.65

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.59
CQA speedup if fully vectorized	4.38
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	NA
Bottlenecks	P10, P11, P12, P13,
Function	kineticEnergy.extracted
Source	timestep.c:110-116
Source loop unroll info	unrolled by 4
Source loop unroll confidence level	max
Unroll/vectorization loop type	main
Unroll factor	4
CQA cycles	8.75
CQA cycles if no scalar integer	8.75
CQA cycles if FP arith vectorized	5.50
CQA cycles if fully vectorized	2.00
Front-end cycles	6.88
P0 cycles	0.50
P1 cycles	0.50
P2 cycles	0.50
P3 cycles	0.50
P4 cycles	0.50
P5 cycles	0.50
P6 cycles	2.75
P7 cycles	2.75
P8 cycles	2.75
P9 cycles	2.75
P10 cycles	8.75
P11 cycles	8.75
P12 cycles	8.75
P13 cycles	8.75
P14 cycles	2.00
P15 cycles	2.00
DIV/SQRT cycles	8.00
Inter-iter dependencies cycles	2
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	56.00
Nb uops	55.00
Nb loads	11.00
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	3.66
Nb FLOP add-sub	12.00
Nb FLOP mul	16.00
Nb FLOP fma	0.00
Nb FLOP div	4.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	16.46
Bytes prefetched	0.00
Bytes loaded	144.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	2.00
Stride n	0.00
Stride unknown	4.00
Stride indirect	0.00
Vectorization ratio all	80.77
Vectorization ratio load	63.64
Vectorization ratio store	NA
Vectorization ratio mul	100.00
Vectorization ratio add_sub	100.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	100.00
Vectorization ratio other	73.91
Vector-efficiency ratio all	22.12
Vector-efficiency ratio load	20.45
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	25.00
Vector-efficiency ratio add_sub	25.00
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	25.00
Vector-efficiency ratio other	20.65

Path /

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

Function	kineticEnergy.extracted
Source file and lines	timestep.c:110-116
Module	exec

The loop is defined in /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:110-116.

It is main loop of related source loop which is unrolled by 4 (including vectorization).

gain
potential
hint
expert

Vectorization

Your loop is partially vectorized. Only 22% of vector register length is used (average across all SSE/AVX instructions). By fully vectorizing your loop, you can lower the cost of an iteration from 8.75 to 2.00 cycles (4.38x speedup).

Details

80% of SSE/AVX instructions are used in vector version (process two or more data elements in vector registers):

63% of SSE/AVX loads are used in vector version.
73% of SSE/AVX instructions that are not load, store, addition, subtraction nor multiply instructions are used in vector version.

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

Workaround

Pass to your compiler a micro-architecture specialization option:
- use march=native
Use vector aligned instructions:
1. align your arrays on 64 bytes boundaries: replace { void *p = malloc (size); } with { void *p; posix_memalign (&p, 64, size); }.
2. inform your compiler that your arrays are vector aligned: if array 'foo' is 64 bytes-aligned, define a pointer 'p_foo' as __builtin_assume_aligned (foo, 64) and use it instead of 'foo' in the loop.

Execution units bottlenecks

Performance is limited by:

execution of FP add operations (the FP add unit is a bottleneck)
execution of FP multiply or FMA (fused multiply-add) operations (the FP multiply/FMA unit is a bottleneck)

Workaround

Reduce the number of FP add instructions
Reduce the number of FP multiply/FMA instructions

FMA

Presence of both ADD/SUB and MUL operations.

Workaround

Pass to your compiler a micro-architecture specialization option:
- use march=native
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: 4 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)

Vector unaligned load/store instructions

Detected 8 suboptimal vector unaligned load/store instructions.

Details

MOVHPD: 2 occurrences►
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:113
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:113
MOVUPD: 6 occurrences►
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:114
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:114
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:114
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:114
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:114
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:114

Workaround

Pass to your compiler a micro-architecture specialization option:
- use march=native
Use vector aligned instructions:
1. align your arrays on 64 bytes boundaries: replace { void *p = malloc (size); } with { void *p; posix_memalign (&p, 64, size); }.
2. inform your compiler that your arrays are vector aligned: if array 'foo' is 64 bytes-aligned, define a pointer 'p_foo' as __builtin_assume_aligned (foo, 64) and use it instead of 'foo' in the loop.

Type of elements and instruction set

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

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

The binary loop is composed of 32 FP arithmetical operations:

12: addition or subtraction
16: multiply
4: divide

The binary loop is loading 144 bytes (18 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.22 FP operations per loaded or stored byte.

General properties

nb instructions	56
nb uops	55
loop length	289
used x86 registers	8
used mmx registers	0
used xmm registers	13
used ymm registers	0
used zmm registers	0
nb stack references	0
ADD-SUB / MUL ratio	0.75

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	6.88 cycles
front end	6.88 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	0.50	0.50	0.50	0.50	0.50	0.50	2.75	2.75	2.75	2.75	8.75	8.75	8.75	8.75	2.00	2.00
cycles	0.50	0.50	0.50	0.50	0.50	0.50	2.75	2.75	2.75	2.75	8.75	8.75	8.75	8.75	2.00	2.00

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	8.00
Longest recurrence chain latency (RecMII)	2.00

Cycles summary

Front-end	6.88
Dispatch	8.75
DIV/SQRT	8.00
Data deps.	2.00
Overall L1	8.75

Vectorization ratios

INT

all	76%
load	100%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	100%
fma	NA (no fma vectorizable/vectorized instructions)
other	71%

all	82%
load	60%
store	NA (no store vectorizable/vectorized instructions)
mul	100%
add-sub	100%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	100%
other	77%

INT+FP

all	80%
load	63%
store	NA (no store vectorizable/vectorized instructions)
mul	100%
add-sub	100%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	100%
other	73%

Vector efficiency ratios

INT

all	20%
load	25%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	25%
fma	NA (no fma vectorizable/vectorized instructions)
other	19%

all	22%
load	20%
store	NA (no store vectorizable/vectorized instructions)
mul	25%
add-sub	25%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	25%
other	22%

INT+FP

all	22%
load	20%
store	NA (no store vectorizable/vectorized instructions)
mul	25%
add-sub	25%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	25%
other	20%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 8.75 cycles. At this rate:

6% of peak load performance is reached (16.46 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: 41ab90

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	Latency	Recip. throughput	Vectorization
MOVUPD 0x40(%R14),%XMM7	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD 0x50(%R14),%XMM12	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD (%R14),%XMM11	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD 0x10(%R14),%XMM8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
MOVDQU (%R15,%RDX,4),%XMM9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD 0x20(%R14),%XMM13	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD 0x30(%R14),%XMM10	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
MULPD %XMM13,%XMM13	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
MULPD %XMM12,%XMM12	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
MULPD %XMM11,%XMM11	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
MULPD %XMM8,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
MULPD %XMM7,%XMM7	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
MULPD %XMM10,%XMM10	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
MOVAPD %XMM7,%XMM14	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
MOVSD %XMM10,%XMM14	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	scal (12.5%)
MOVAPD %XMM8,%XMM15	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
MOVSD %XMM11,%XMM15	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	scal (12.5%)
SHUFPD $0x1,%XMM12,%XMM10	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
SHUFPD $0x2,%XMM13,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
SHUFPD $0x2,%XMM12,%XMM7	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
SHUFPD $0x1,%XMM13,%XMM11	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
XORPD %XMM12,%XMM12	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
PCMPGTD %XMM9,%XMM12	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
PSHUFD $-0x12,%XMM9,%XMM13	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
ADDPD %XMM14,%XMM10	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (25.0%)
PUNPCKLDQ %XMM12,%XMM9	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (12.5%)
PXOR %XMM12,%XMM12	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.25	vect (25.0%)
PCMPGTD %XMM13,%XMM12	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
ADDPD %XMM15,%XMM11	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (25.0%)
PUNPCKLDQ %XMM12,%XMM13	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (12.5%)
PSLLQ $0x4,%XMM13	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	1	0.50	vect (25.0%)
PADDQ %XMM4,%XMM13	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
ADDPD %XMM11,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (25.0%)
PSLLQ $0x4,%XMM9	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	1	0.50	vect (25.0%)
PADDQ %XMM4,%XMM9	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
MOVQ %XMM9,%RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	1	0.50	scal (12.5%)
ADDPD %XMM10,%XMM7	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (25.0%)
PSHUFD $-0x12,%XMM9,%XMM9	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
MOVQ %XMM9,%RCX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	1	0.50	scal (12.5%)
MOVQ %XMM13,%R12	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	1	0.50	scal (12.5%)
MULPD %XMM2,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
PSHUFD $-0x12,%XMM13,%XMM9	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
MOVQ %XMM9,%R10	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	1	0.50	scal (12.5%)
MOVSD (%RAX),%XMM9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (12.5%)
MULPD %XMM2,%XMM7	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
MOVHPD (%RCX),%XMM9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.25	0.25	0.25	0.25	0	0	1	0.50	scal (12.5%)
MOVSD (%R12),%XMM10	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (12.5%)
DIVPD %XMM9,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	13	4	vect (25.0%)
MOVHPD (%R10),%XMM10	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.25	0.25	0.25	0.25	0	0	1	0.50	scal (12.5%)
DIVPD %XMM10,%XMM7	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	13	4	vect (25.0%)
ADDPD %XMM8,%XMM5	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (25.0%)
ADDPD %XMM7,%XMM6	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (25.0%)
ADD $0x60,%R14	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
ADD $0x4,%RDX	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
CMP %EBX,%EDX	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
JLE 41ab90 <kineticEnergy.extracted+0x1b0>	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

Function	kineticEnergy.extracted
Source file and lines	timestep.c:110-116
Module	exec

gain
potential
hint
expert

Vectorization

Details

80% of SSE/AVX instructions are used in vector version (process two or more data elements in vector registers):

63% of SSE/AVX loads are used in vector version.
73% of SSE/AVX instructions that are not load, store, addition, subtraction nor multiply instructions are used in vector version.

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

Workaround

Pass to your compiler a micro-architecture specialization option:
- use march=native
Use vector aligned instructions:
1. align your arrays on 64 bytes boundaries: replace { void *p = malloc (size); } with { void *p; posix_memalign (&p, 64, size); }.
2. inform your compiler that your arrays are vector aligned: if array 'foo' is 64 bytes-aligned, define a pointer 'p_foo' as __builtin_assume_aligned (foo, 64) and use it instead of 'foo' in the loop.

Execution units bottlenecks

Performance is limited by:

execution of FP add operations (the FP add unit is a bottleneck)
execution of FP multiply or FMA (fused multiply-add) operations (the FP multiply/FMA unit is a bottleneck)

Workaround

Reduce the number of FP add instructions
Reduce the number of FP multiply/FMA instructions

FMA

Presence of both ADD/SUB and MUL operations.

Workaround

Pass to your compiler a micro-architecture specialization option:
- use march=native
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: 4 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)

Vector unaligned load/store instructions

Detected 8 suboptimal vector unaligned load/store instructions.

Details

MOVHPD: 2 occurrences►
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:113
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:113
MOVUPD: 6 occurrences►
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:114
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:114
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:114
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:114
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:114
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:114

Workaround

Pass to your compiler a micro-architecture specialization option:
- use march=native
Use vector aligned instructions:
1. align your arrays on 64 bytes boundaries: replace { void *p = malloc (size); } with { void *p; posix_memalign (&p, 64, size); }.
2. inform your compiler that your arrays are vector aligned: if array 'foo' is 64 bytes-aligned, define a pointer 'p_foo' as __builtin_assume_aligned (foo, 64) and use it instead of 'foo' in the loop.

Type of elements and instruction set

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

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

The binary loop is composed of 32 FP arithmetical operations:

12: addition or subtraction
16: multiply
4: divide

The binary loop is loading 144 bytes (18 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.22 FP operations per loaded or stored byte.

General properties

nb instructions	56
nb uops	55
loop length	289
used x86 registers	8
used mmx registers	0
used xmm registers	13
used ymm registers	0
used zmm registers	0
nb stack references	0
ADD-SUB / MUL ratio	0.75

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	6.88 cycles
front end	6.88 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	0.50	0.50	0.50	0.50	0.50	0.50	2.75	2.75	2.75	2.75	8.75	8.75	8.75	8.75	2.00	2.00
cycles	0.50	0.50	0.50	0.50	0.50	0.50	2.75	2.75	2.75	2.75	8.75	8.75	8.75	8.75	2.00	2.00

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	8.00
Longest recurrence chain latency (RecMII)	2.00

Cycles summary

Front-end	6.88
Dispatch	8.75
DIV/SQRT	8.00
Data deps.	2.00
Overall L1	8.75

Vectorization ratios

INT

all	76%
load	100%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	100%
fma	NA (no fma vectorizable/vectorized instructions)
other	71%

all	82%
load	60%
store	NA (no store vectorizable/vectorized instructions)
mul	100%
add-sub	100%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	100%
other	77%

INT+FP

all	80%
load	63%
store	NA (no store vectorizable/vectorized instructions)
mul	100%
add-sub	100%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	100%
other	73%

Vector efficiency ratios

INT

all	20%
load	25%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	25%
fma	NA (no fma vectorizable/vectorized instructions)
other	19%

all	22%
load	20%
store	NA (no store vectorizable/vectorized instructions)
mul	25%
add-sub	25%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	25%
other	22%

INT+FP

all	22%
load	20%
store	NA (no store vectorizable/vectorized instructions)
mul	25%
add-sub	25%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	25%
other	20%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 8.75 cycles. At this rate:

6% of peak load performance is reached (16.46 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: 41ab90

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	Latency	Recip. throughput	Vectorization
MOVUPD 0x40(%R14),%XMM7	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD 0x50(%R14),%XMM12	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD (%R14),%XMM11	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD 0x10(%R14),%XMM8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
MOVDQU (%R15,%RDX,4),%XMM9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD 0x20(%R14),%XMM13	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD 0x30(%R14),%XMM10	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
MULPD %XMM13,%XMM13	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
MULPD %XMM12,%XMM12	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
MULPD %XMM11,%XMM11	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
MULPD %XMM8,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
MULPD %XMM7,%XMM7	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
MULPD %XMM10,%XMM10	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
MOVAPD %XMM7,%XMM14	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
MOVSD %XMM10,%XMM14	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	scal (12.5%)
MOVAPD %XMM8,%XMM15	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
MOVSD %XMM11,%XMM15	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	scal (12.5%)
SHUFPD $0x1,%XMM12,%XMM10	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
SHUFPD $0x2,%XMM13,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
SHUFPD $0x2,%XMM12,%XMM7	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
SHUFPD $0x1,%XMM13,%XMM11	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
XORPD %XMM12,%XMM12	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
PCMPGTD %XMM9,%XMM12	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
PSHUFD $-0x12,%XMM9,%XMM13	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
ADDPD %XMM14,%XMM10	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (25.0%)
PUNPCKLDQ %XMM12,%XMM9	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (12.5%)
PXOR %XMM12,%XMM12	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.25	vect (25.0%)
PCMPGTD %XMM13,%XMM12	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
ADDPD %XMM15,%XMM11	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (25.0%)
PUNPCKLDQ %XMM12,%XMM13	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (12.5%)
PSLLQ $0x4,%XMM13	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	1	0.50	vect (25.0%)
PADDQ %XMM4,%XMM13	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
ADDPD %XMM11,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (25.0%)
PSLLQ $0x4,%XMM9	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	1	0.50	vect (25.0%)
PADDQ %XMM4,%XMM9	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
MOVQ %XMM9,%RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	1	0.50	scal (12.5%)
ADDPD %XMM10,%XMM7	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (25.0%)
PSHUFD $-0x12,%XMM9,%XMM9	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
MOVQ %XMM9,%RCX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	1	0.50	scal (12.5%)
MOVQ %XMM13,%R12	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	1	0.50	scal (12.5%)
MULPD %XMM2,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
PSHUFD $-0x12,%XMM13,%XMM9	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
MOVQ %XMM9,%R10	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	1	0.50	scal (12.5%)
MOVSD (%RAX),%XMM9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (12.5%)
MULPD %XMM2,%XMM7	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (25.0%)
MOVHPD (%RCX),%XMM9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.25	0.25	0.25	0.25	0	0	1	0.50	scal (12.5%)
MOVSD (%R12),%XMM10	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (12.5%)
DIVPD %XMM9,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	13	4	vect (25.0%)
MOVHPD (%R10),%XMM10	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.25	0.25	0.25	0.25	0	0	1	0.50	scal (12.5%)
DIVPD %XMM10,%XMM7	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	13	4	vect (25.0%)
ADDPD %XMM8,%XMM5	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (25.0%)
ADDPD %XMM7,%XMM6	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (25.0%)
ADD $0x60,%R14	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
ADD $0x4,%RDX	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
CMP %EBX,%EDX	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
JLE 41ab90 <kineticEnergy.extracted+0x1b0>	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

Report Configuration

Vectorization

Details

Workaround

Execution units bottlenecks

Workaround

FMA

Workaround

Slow data structures access

Details

Workaround

Vector unaligned load/store instructions

Details

Workaround

Type of elements and instruction set

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

Arithmetic intensity

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code

Vectorization

Details

Workaround

Execution units bottlenecks

Workaround

FMA

Workaround

Slow data structures access

Details

Workaround

Vector unaligned load/store instructions

Details

Workaround

Type of elements and instruction set

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

Arithmetic intensity

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code