OV - exec - Loop 109

	Loop Id: 109	Module: exec	Source: timestep.c:110-116	Coverage: 0.04%

0x410620 VPMOVSXDQ	(%R14,%RDX,4),%ZMM17    [2]

0x410627 VMOVUPD	(%R15),%ZMM18    [3]

0x41062d VMOVUPD	0x40(%R15),%ZMM19    [3]

0x410634 VMOVUPD	0x80(%R15),%ZMM20    [3]

0x41063b VMULPD	%ZMM20,%ZMM20,%ZMM20

0x410641 VMULPD	%ZMM19,%ZMM19,%ZMM19

0x410647 VMULPD	%ZMM18,%ZMM18,%ZMM18

0x41064d VMOVAPD	%ZMM18,%ZMM21

0x410653 VPERMT2PD	%ZMM19,%ZMM4,%ZMM21

0x410659 VPERMT2PD	%ZMM20,%ZMM5,%ZMM21

0x41065f VMOVAPD	%ZMM18,%ZMM22

0x410665 VPERMT2PD	%ZMM19,%ZMM6,%ZMM22

0x41066b VPERMT2PD	%ZMM20,%ZMM7,%ZMM22

0x410671 VADDPD	%ZMM21,%ZMM22,%ZMM21

0x410677 VPERMT2PD	%ZMM18,%ZMM8,%ZMM19

0x41067d VPERMT2PD	%ZMM20,%ZMM9,%ZMM19

0x410683 VADDPD	%ZMM19,%ZMM21,%ZMM18

0x410689 VPSLLQ	$0x4,%ZMM17,%ZMM17

0x410690 VXORPD	%XMM19,%XMM19,%XMM19

0x410696 KXNORW	%K0,%K0,%K1

0x41069a VMULPD	%ZMM3,%ZMM18,%ZMM18

0x4106a0 VGATHERQPD	0x8(%R13,%ZMM17,1),%ZMM19{%K1}    [1]

0x4106a8 VDIVPD	%ZMM19,%ZMM18,%ZMM17

0x4106ae VADDPD	%ZMM16,%ZMM17,%ZMM16

0x4106b4 ADD	$0xc0,%R15

0x4106bb ADD	$0x8,%RDX

0x4106bf CMP	%EBX,%EDX

0x4106c1 JLE	410620

/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
►97.27+	__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
○	__GI___clone3		libc.so.6
►2.21+	__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

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	1.04
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	1.02
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.19
Bottlenecks	micro-operation queue,
Function	kineticEnergy.extracted
Source	timestep.c:110-116
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	max
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	9.25
CQA cycles if no scalar integer	8.88
CQA cycles if FP arith vectorized	9.25
CQA cycles if fully vectorized	9.08
Front-end cycles	9.25
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	3.00
P7 cycles	3.00
P8 cycles	3.00
P9 cycles	3.00
P10 cycles	7.75
P11 cycles	7.75
P12 cycles	7.75
P13 cycles	7.75
P14 cycles	5.00
P15 cycles	5.00
DIV/SQRT cycles	4.00
Inter-iter dependencies cycles	2
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	28.00
Nb uops	74.00
Nb loads	5.00
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	6.92
Nb FLOP add-sub	24.00
Nb FLOP mul	32.00
Nb FLOP fma	0.00
Nb FLOP div	8.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	31.14
Bytes prefetched	0.00
Bytes loaded	288.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	2.00
Stride n	0.00
Stride unknown	0.00
Stride indirect	1.00
Vectorization ratio all	100.00
Vectorization ratio load	100.00
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	100.00
Vector-efficiency ratio all	94.57
Vector-efficiency ratio load	90.00
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	100.00
Vector-efficiency ratio add_sub	100.00
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	100.00
Vector-efficiency ratio other	93.18

Metric	Value
CQA speedup if no scalar integer	1.04
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	1.02
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.19
Bottlenecks	micro-operation queue,
Function	kineticEnergy.extracted
Source	timestep.c:110-116
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	max
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	9.25
CQA cycles if no scalar integer	8.88
CQA cycles if FP arith vectorized	9.25
CQA cycles if fully vectorized	9.08
Front-end cycles	9.25
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	3.00
P7 cycles	3.00
P8 cycles	3.00
P9 cycles	3.00
P10 cycles	7.75
P11 cycles	7.75
P12 cycles	7.75
P13 cycles	7.75
P14 cycles	5.00
P15 cycles	5.00
DIV/SQRT cycles	4.00
Inter-iter dependencies cycles	2
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	28.00
Nb uops	74.00
Nb loads	5.00
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	6.92
Nb FLOP add-sub	24.00
Nb FLOP mul	32.00
Nb FLOP fma	0.00
Nb FLOP div	8.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	31.14
Bytes prefetched	0.00
Bytes loaded	288.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	2.00
Stride n	0.00
Stride unknown	0.00
Stride indirect	1.00
Vectorization ratio all	100.00
Vectorization ratio load	100.00
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	100.00
Vector-efficiency ratio all	94.57
Vector-efficiency ratio load	90.00
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	100.00
Vector-efficiency ratio add_sub	100.00
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	100.00
Vector-efficiency ratio other	93.18

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.

The related source loop is not unrolled or unrolled with no peel/tail loop.

gain
potential
hint
expert

Vectorization

Your loop is vectorized, but using 94% register length (average across all SSE/AVX instructions). By fully vectorizing your loop, you can lower the cost of an iteration from 9.25 to 9.08 cycles (1.02x speedup).

Details

All SSE/AVX instructions are used in vector version (process two or more data elements in vector registers). 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

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

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).

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.

VGATHERQPD: 1 occurrences►
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:113

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.

Vector unaligned load/store instructions

Detected 3 optimal vector unaligned load/store instructions.

Details

VMOVUPD: 3 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

Workaround

Use vector aligned instructions:

align your arrays on 64 bytes boundaries: replace { void *p = malloc (size); } with { void *p; posix_memalign (&p, 64, size); }.
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.

Gather/scatter instructions

Detected gather/scatter instructions (typically caused by indirect accesses).

Details

VGATHERQPD: 1 occurrences►
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:113

Workaround

Try to simplify your code and/or replace indirect accesses with unit-stride ones.

Type of elements and instruction set

8 AVX-512 instructions are processing arithmetic or math operations on double precision FP elements in vector mode (eight at a time).

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

The binary loop is composed of 64 FP arithmetical operations:

24: addition or subtraction
32: multiply
8: divide

The binary loop is loading 288 bytes (36 double precision FP elements).

Arithmetic intensity

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

General properties

nb instructions	28
nb uops	74
loop length	167
used x86 registers	5
used mmx registers	0
used xmm registers	1
used ymm registers	0
used zmm registers	14
nb stack references	0
ADD-SUB / MUL ratio	0.75

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	9.25 cycles
front end	9.25 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	3.00	3.00	3.00	3.00	7.75	7.75	7.75	7.75	5.00	5.00
cycles	0.50	0.50	0.50	0.50	0.50	0.50	3.00	3.00	3.00	3.00	7.75	7.75	7.75	7.75	5.00	5.00

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	4.00
Longest recurrence chain latency (RecMII)	2.00

Cycles summary

Front-end	9.25
Dispatch	7.75
DIV/SQRT	4.00
Data deps.	2.00
Overall L1	9.25

Vectorization ratios

INT

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

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

INT+FP

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

Vector efficiency ratios

INT

all	75%
load	50%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	100%

all	96%
load	100%
store	NA (no store vectorizable/vectorized instructions)
mul	100%
add-sub	100%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	100%
other	92%

INT+FP

all	94%
load	90%
store	NA (no store vectorizable/vectorized instructions)
mul	100%
add-sub	100%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	100%
other	93%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 9.25 cycles. At this rate:

12% of peak load performance is reached (31.14 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 9.25 to 7.75 cycles (1.19x speedup).

ASM code

In the binary file, the address of the loop is: 410620

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	Latency	Recip. throughput	Vectorization
VPMOVSXDQ (%R14,%RDX,4),%ZMM17	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0.50	0.50	0	0	0	4	0.50	vect (50.0%)
VMOVUPD (%R15),%ZMM18	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 (100.0%)
VMOVUPD 0x40(%R15),%ZMM19	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 (100.0%)
VMOVUPD 0x80(%R15),%ZMM20	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 (100.0%)
VMULPD %ZMM20,%ZMM20,%ZMM20	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (100.0%)
VMULPD %ZMM19,%ZMM19,%ZMM19	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (100.0%)
VMULPD %ZMM18,%ZMM18,%ZMM18	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (100.0%)
VMOVAPD %ZMM18,%ZMM21	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (100.0%)
VPERMT2PD %ZMM19,%ZMM4,%ZMM21	1	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	5	0.50	vect (100.0%)
VPERMT2PD %ZMM20,%ZMM5,%ZMM21	1	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	5	0.50	vect (100.0%)
VMOVAPD %ZMM18,%ZMM22	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (100.0%)
VPERMT2PD %ZMM19,%ZMM6,%ZMM22	1	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	5	0.50	vect (100.0%)
VPERMT2PD %ZMM20,%ZMM7,%ZMM22	1	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	5	0.50	vect (100.0%)
VADDPD %ZMM21,%ZMM22,%ZMM21	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (100.0%)
VPERMT2PD %ZMM18,%ZMM8,%ZMM19	1	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	5	0.50	vect (100.0%)
VPERMT2PD %ZMM20,%ZMM9,%ZMM19	1	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	5	0.50	vect (100.0%)
VADDPD %ZMM19,%ZMM21,%ZMM18	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (100.0%)
VPSLLQ $0x4,%ZMM17,%ZMM17	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	1	0.50	vect (100.0%)
VXORPD %XMM19,%XMM19,%XMM19	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
KXNORW %K0,%K0,%K1	1	0	0	0	0	0	0	0	0	0	0	0.50	0	0	0.50	0	0	1	0.50	N/A
VMULPD %ZMM3,%ZMM18,%ZMM18	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (100.0%)
VGATHERQPD 0x8(%R13,%ZMM17,1),%ZMM19{%K1}	48	0	0	0	0	0	0	2	2	2	2	2.50	4	4.50	3	5	5	0-16	8.94	vect (100.0%)
VDIVPD %ZMM19,%ZMM18,%ZMM17	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	13	4	vect (100.0%)
VADDPD %ZMM16,%ZMM17,%ZMM16	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (100.0%)
ADD $0xc0,%R15	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 $0x8,%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 410620 <kineticEnergy.extracted+0x300>	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

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

The related source loop is not unrolled or unrolled with no peel/tail loop.

gain
potential
hint
expert

Vectorization

Details

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

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

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).

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.

VGATHERQPD: 1 occurrences►
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:113

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.

Vector unaligned load/store instructions

Detected 3 optimal vector unaligned load/store instructions.

Details

VMOVUPD: 3 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

Workaround

Use vector aligned instructions:

align your arrays on 64 bytes boundaries: replace { void *p = malloc (size); } with { void *p; posix_memalign (&p, 64, size); }.
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.

Gather/scatter instructions

Detected gather/scatter instructions (typically caused by indirect accesses).

Details

VGATHERQPD: 1 occurrences►
- /home/eoseret/qaas_runs_ZEN5/174-043-3878/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:113

Workaround

Try to simplify your code and/or replace indirect accesses with unit-stride ones.

Type of elements and instruction set

8 AVX-512 instructions are processing arithmetic or math operations on double precision FP elements in vector mode (eight at a time).

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

The binary loop is composed of 64 FP arithmetical operations:

24: addition or subtraction
32: multiply
8: divide

The binary loop is loading 288 bytes (36 double precision FP elements).

Arithmetic intensity

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

General properties

nb instructions	28
nb uops	74
loop length	167
used x86 registers	5
used mmx registers	0
used xmm registers	1
used ymm registers	0
used zmm registers	14
nb stack references	0
ADD-SUB / MUL ratio	0.75

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	9.25 cycles
front end	9.25 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	3.00	3.00	3.00	3.00	7.75	7.75	7.75	7.75	5.00	5.00
cycles	0.50	0.50	0.50	0.50	0.50	0.50	3.00	3.00	3.00	3.00	7.75	7.75	7.75	7.75	5.00	5.00

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	4.00
Longest recurrence chain latency (RecMII)	2.00

Cycles summary

Front-end	9.25
Dispatch	7.75
DIV/SQRT	4.00
Data deps.	2.00
Overall L1	9.25

Vectorization ratios

INT

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

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

INT+FP

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

Vector efficiency ratios

INT

all	75%
load	50%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	100%

all	96%
load	100%
store	NA (no store vectorizable/vectorized instructions)
mul	100%
add-sub	100%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	100%
other	92%

INT+FP

all	94%
load	90%
store	NA (no store vectorizable/vectorized instructions)
mul	100%
add-sub	100%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	100%
other	93%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 9.25 cycles. At this rate:

12% of peak load performance is reached (31.14 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))

Front-end bottlenecks

ASM code

In the binary file, the address of the loop is: 410620

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	Latency	Recip. throughput	Vectorization
VPMOVSXDQ (%R14,%RDX,4),%ZMM17	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0.50	0.50	0	0	0	4	0.50	vect (50.0%)
VMOVUPD (%R15),%ZMM18	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 (100.0%)
VMOVUPD 0x40(%R15),%ZMM19	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 (100.0%)
VMOVUPD 0x80(%R15),%ZMM20	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 (100.0%)
VMULPD %ZMM20,%ZMM20,%ZMM20	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (100.0%)
VMULPD %ZMM19,%ZMM19,%ZMM19	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (100.0%)
VMULPD %ZMM18,%ZMM18,%ZMM18	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (100.0%)
VMOVAPD %ZMM18,%ZMM21	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (100.0%)
VPERMT2PD %ZMM19,%ZMM4,%ZMM21	1	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	5	0.50	vect (100.0%)
VPERMT2PD %ZMM20,%ZMM5,%ZMM21	1	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	5	0.50	vect (100.0%)
VMOVAPD %ZMM18,%ZMM22	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (100.0%)
VPERMT2PD %ZMM19,%ZMM6,%ZMM22	1	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	5	0.50	vect (100.0%)
VPERMT2PD %ZMM20,%ZMM7,%ZMM22	1	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	5	0.50	vect (100.0%)
VADDPD %ZMM21,%ZMM22,%ZMM21	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (100.0%)
VPERMT2PD %ZMM18,%ZMM8,%ZMM19	1	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	5	0.50	vect (100.0%)
VPERMT2PD %ZMM20,%ZMM9,%ZMM19	1	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	5	0.50	vect (100.0%)
VADDPD %ZMM19,%ZMM21,%ZMM18	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (100.0%)
VPSLLQ $0x4,%ZMM17,%ZMM17	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	1	0.50	vect (100.0%)
VXORPD %XMM19,%XMM19,%XMM19	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
KXNORW %K0,%K0,%K1	1	0	0	0	0	0	0	0	0	0	0	0.50	0	0	0.50	0	0	1	0.50	N/A
VMULPD %ZMM3,%ZMM18,%ZMM18	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	vect (100.0%)
VGATHERQPD 0x8(%R13,%ZMM17,1),%ZMM19{%K1}	48	0	0	0	0	0	0	2	2	2	2	2.50	4	4.50	3	5	5	0-16	8.94	vect (100.0%)
VDIVPD %ZMM19,%ZMM18,%ZMM17	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	13	4	vect (100.0%)
VADDPD %ZMM16,%ZMM17,%ZMM16	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (100.0%)
ADD $0xc0,%R15	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 $0x8,%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 410620 <kineticEnergy.extracted+0x300>	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

FMA

Workaround

Complex instructions

Details

Slow data structures access

Details

Workaround

Vector unaligned load/store instructions

Details

Workaround

Gather/scatter 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

FMA

Workaround

Complex instructions

Details

Slow data structures access

Details

Workaround

Vector unaligned load/store instructions

Details

Workaround

Gather/scatter 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