OV - exec - Loop 116

0x40b340 LDR	D0, [X19]    [3]

0x40b344 LD3	{V1.2D, V2.2D, V3.2D}, [X1], #80    [2]

0x40b348 SUBS	X2, X2, #2

0x40b34c LD3	{V4.2D, V5.2D, V6.2D}, [X0]    [1]

0x40b350 FMLA	V4.2D, V1.2D, V0.D[0]

0x40b354 LDR	D0, [X19]    [3]

0x40b358 FMLA	V5.2D, V2.2D, V0.D[0]

0x40b35c FMLA	V6.2D, V3.2D, V0.D[0]

0x40b360 ST3	{V4.2D, V5.2D, V6.2D}, [X0], #48    [1]

0x40b364 B.NE	40b340

/home/hbollore/qaas/qaas-runs/174-108-7908/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c: 74 - 78

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

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

75:       {

76:          s->atoms->p[iOff][0] += dt*s->atoms->f[iOff][0];

77:          s->atoms->p[iOff][1] += dt*s->atoms->f[iOff][1];

78:          s->atoms->p[iOff][2] += dt*s->atoms->f[iOff][2];

Coverage (%)	Name	Source Location	Module
○100.00	__kmp_invoke_microtask		libomp.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	1.22
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.10
Bottlenecks	P12,
Function	advanceVelocity.omp_outlined
Source	timestep.c:74-78
Source loop unroll info	unrolled by 2
Source loop unroll confidence level	high
Unroll/vectorization loop type	NA
Unroll factor	2
CQA cycles	1.83
CQA cycles if no scalar integer	1.83
CQA cycles if FP arith vectorized	1.83
CQA cycles if fully vectorized	1.50
Front-end cycles	1.25
P0 cycles	0.50
P1 cycles	0.50
P2 cycles	0.42
P3 cycles	0.42
P4 cycles	0.33
P5 cycles	0.33
P6 cycles	0.25
P7 cycles	0.25
P8 cycles	1.50
P9 cycles	1.50
P10 cycles	1.50
P11 cycles	1.50
P12 cycles	1.83
P13 cycles	1.50
P14 cycles	1.67
P15 cycles	0.00
P16 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	10.00
Nb uops	10.00
Nb loads	NA
Nb stores	1.00
Nb stack references	0.00
FLOP/cycle	6.55
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	6.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	87.27
Bytes prefetched	0.00
Bytes loaded	112.00
Bytes stored	48.00
Stride 0	3.00
Stride 1	0.00
Stride n	0.00
Stride unknown	0.00
Stride indirect	0.00
Vectorization ratio all	75.00
Vectorization ratio load	50.00
Vectorization ratio store	100.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	100.00
Vectorization ratio div_sqrt	NA
Vectorization ratio other	NA
Vector-efficiency ratio all	87.50
Vector-efficiency ratio load	75.00
Vector-efficiency ratio store	100.00
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	100.00
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	NA

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	1.22
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.10
Bottlenecks	P12,
Function	advanceVelocity.omp_outlined
Source	timestep.c:74-78
Source loop unroll info	unrolled by 2
Source loop unroll confidence level	high
Unroll/vectorization loop type	NA
Unroll factor	2
CQA cycles	1.83
CQA cycles if no scalar integer	1.83
CQA cycles if FP arith vectorized	1.83
CQA cycles if fully vectorized	1.50
Front-end cycles	1.25
P0 cycles	0.50
P1 cycles	0.50
P2 cycles	0.42
P3 cycles	0.42
P4 cycles	0.33
P5 cycles	0.33
P6 cycles	0.25
P7 cycles	0.25
P8 cycles	1.50
P9 cycles	1.50
P10 cycles	1.50
P11 cycles	1.50
P12 cycles	1.83
P13 cycles	1.50
P14 cycles	1.67
P15 cycles	0.00
P16 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	10.00
Nb uops	10.00
Nb loads	NA
Nb stores	1.00
Nb stack references	0.00
FLOP/cycle	6.55
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	6.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	87.27
Bytes prefetched	0.00
Bytes loaded	112.00
Bytes stored	48.00
Stride 0	3.00
Stride 1	0.00
Stride n	0.00
Stride unknown	0.00
Stride indirect	0.00
Vectorization ratio all	75.00
Vectorization ratio load	50.00
Vectorization ratio store	100.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	100.00
Vectorization ratio div_sqrt	NA
Vectorization ratio other	NA
Vector-efficiency ratio all	87.50
Vector-efficiency ratio load	75.00
Vector-efficiency ratio store	100.00
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	100.00
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	NA

Path /

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

Function	advanceVelocity.omp_outlined
Source file and lines	timestep.c:74-78
Module	exec

The loop is defined in /home/hbollore/qaas/qaas-runs/174-108-7908/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:74-78.

The related source loop is unrolled by 2 (including vectorization).

gain
potential
hint
expert

Vectorization

Your loop is partially vectorized. 87% of vector register length is used (average across all VPU instructions). By fully vectorizing your loop, you can lower the cost of an iteration from 1.83 to 1.50 cycles (1.22x speedup).

Details

75% of VPU instructions are used in vector version (process two or more data elements in vector registers):

50% of VPU loads are used in vector version.

Since your execution units are vector units, only a fully 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)

FMA

Detected 6 FMA (fused multiply-add) operations.

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

The binary loop is composed of 12 FP arithmetical operations:

6: addition or subtraction (all inside FMA instructions)
6: multiply (all inside FMA instructions)

The binary loop is loading 112 bytes. The binary loop is storing 48 bytes.

Arithmetic intensity

Arithmetic intensity is 0.07 FP operations per loaded or stored byte.

General properties

nb instructions	10
loop length	40
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

1.25 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	P16
uops	0.50	0.50	0.42	0.42	0.33	0.33	0.25	0.25	1.50	1.50	1.50	1.50	1.83	1.50	1.67	0.00	0.00
cycles	0.50	0.50	0.42	0.42	0.33	0.33	0.25	0.25	1.50	1.50	1.50	1.50	1.83	1.50	1.67	0.00	0.00

Execution ports to units layout:

P0: BRU
P1: BRU
P2: ALU
P3: ALU
P4: ALU
P5: ALU
P6: ALU
P7: ALU
P8 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P9 (128 bits): VPU, ALU, FP store data
P10 (128 bits): VPU, ALU, DIV/SQRT
P11 (128 bits): ALU, VPU
P12 (256 bits): store address, load
P13 (256 bits): store address, load
P14 (256 bits): load
P15 (64 bits): store data
P16 (64 bits): store data

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	1.00

Cycles summary

Front-end	1.25
Data deps.	1.00
Overall L1	1.83

Vectorization ratios

INT

all	60%
load	50%
store	100%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

FP

all	100%
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	100%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

INT+FP

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.83 cycles. At this rate:

63% of peak load performance is reached (61.09 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
54% of peak store performance is reached (26.18 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 40b340

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	Latency	Recip. throughput	Vectorization
LDR D0, [X19]	1	0	0	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33	scal (50.0%)
LD3 {V1.2D, V2.2D, V3.2D}, [X1], #80	1	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.33	0.33	0.33	8	1	vect (100.0%)
SUBS X2, X2, #2	1	0	0	0.25	0.25	0	0	0.25	0.25	0	0	0	0	0	0	0	1	0.33	N/A
LD3 {V4.2D, V5.2D, V6.2D}, [X0]	1	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.33	0.33	0.33	8	1	vect (100.0%)
FMLA V4.2D, V1.2D, V0.D[0]	1	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (100.0%)
LDR D0, [X19]	1	0	0	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33	scal (50.0%)
FMLA V5.2D, V2.2D, V0.D[0]	1	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (100.0%)
FMLA V6.2D, V3.2D, V0.D[0]	1	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (100.0%)
ST3 {V4.2D, V5.2D, V6.2D}, [X0], #48	1	0	0	0.17	0.17	0.17	0.17	0.17	0.17	0.50	0.50	0	0	0.50	0.50	0	6	3.00	vect (100.0%)
B.NE 40b340 <advanceVelocity.omp_outlined+0x1f0>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A

Function	advanceVelocity.omp_outlined
Source file and lines	timestep.c:74-78
Module	exec

The loop is defined in /home/hbollore/qaas/qaas-runs/174-108-7908/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:74-78.

The related source loop is unrolled by 2 (including vectorization).

gain
potential
hint
expert

Vectorization

Your loop is partially vectorized. 87% of vector register length is used (average across all VPU instructions). By fully vectorizing your loop, you can lower the cost of an iteration from 1.83 to 1.50 cycles (1.22x speedup).

Details

75% of VPU instructions are used in vector version (process two or more data elements in vector registers):

50% of VPU loads are used in vector version.

Since your execution units are vector units, only a fully 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)

FMA

Detected 6 FMA (fused multiply-add) operations.

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

The binary loop is composed of 12 FP arithmetical operations:

6: addition or subtraction (all inside FMA instructions)
6: multiply (all inside FMA instructions)

The binary loop is loading 112 bytes. The binary loop is storing 48 bytes.

Arithmetic intensity

Arithmetic intensity is 0.07 FP operations per loaded or stored byte.

General properties

nb instructions	10
loop length	40
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

1.25 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	P16
uops	0.50	0.50	0.42	0.42	0.33	0.33	0.25	0.25	1.50	1.50	1.50	1.50	1.83	1.50	1.67	0.00	0.00
cycles	0.50	0.50	0.42	0.42	0.33	0.33	0.25	0.25	1.50	1.50	1.50	1.50	1.83	1.50	1.67	0.00	0.00

Execution ports to units layout:

P0: BRU
P1: BRU
P2: ALU
P3: ALU
P4: ALU
P5: ALU
P6: ALU
P7: ALU
P8 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P9 (128 bits): VPU, ALU, FP store data
P10 (128 bits): VPU, ALU, DIV/SQRT
P11 (128 bits): ALU, VPU
P12 (256 bits): store address, load
P13 (256 bits): store address, load
P14 (256 bits): load
P15 (64 bits): store data
P16 (64 bits): store data

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	1.00

Cycles summary

Front-end	1.25
Data deps.	1.00
Overall L1	1.83

Vectorization ratios

INT

all	60%
load	50%
store	100%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

FP

all	100%
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	100%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

INT+FP

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.83 cycles. At this rate:

63% of peak load performance is reached (61.09 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
54% of peak store performance is reached (26.18 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 40b340

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	Latency	Recip. throughput	Vectorization
LDR D0, [X19]	1	0	0	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33	scal (50.0%)
LD3 {V1.2D, V2.2D, V3.2D}, [X1], #80	1	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.33	0.33	0.33	8	1	vect (100.0%)
SUBS X2, X2, #2	1	0	0	0.25	0.25	0	0	0.25	0.25	0	0	0	0	0	0	0	1	0.33	N/A
LD3 {V4.2D, V5.2D, V6.2D}, [X0]	1	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.33	0.33	0.33	8	1	vect (100.0%)
FMLA V4.2D, V1.2D, V0.D[0]	1	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (100.0%)
LDR D0, [X19]	1	0	0	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33	scal (50.0%)
FMLA V5.2D, V2.2D, V0.D[0]	1	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (100.0%)
FMLA V6.2D, V3.2D, V0.D[0]	1	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (100.0%)
ST3 {V4.2D, V5.2D, V6.2D}, [X0], #48	1	0	0	0.17	0.17	0.17	0.17	0.17	0.17	0.50	0.50	0	0	0.50	0.50	0	6	3.00	vect (100.0%)
B.NE 40b340 <advanceVelocity.omp_outlined+0x1f0>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A

Report Configuration

Vectorization

Details

Workaround

FMA

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

Arithmetic intensity

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code

Vectorization

Details

Workaround

FMA

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

Arithmetic intensity

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code