OV - exec - Loop 152

0x418110 MOVSD	-0x10(%R10,%R8,1),%XMM1    [2]

0x418117 MULSD	%XMM0,%XMM1

0x41811b ADDSD	-0x10(%R9,%R8,1),%XMM1    [1]

0x418122 MOVSD	%XMM1,-0x10(%R9,%R8,1)    [1]

0x418129 MOVSD	-0x8(%R10,%R8,1),%XMM1    [2]

0x418130 MULSD	%XMM0,%XMM1

0x418134 ADDSD	-0x8(%R9,%R8,1),%XMM1    [1]

0x41813b MOVSD	%XMM1,-0x8(%R9,%R8,1)    [1]

0x418142 MOVSD	(%R10,%R8,1),%XMM1    [2]

0x418148 MULSD	%XMM0,%XMM1

0x41814c ADDSD	(%R9,%R8,1),%XMM1    [1]

0x418152 MOVSD	%XMM1,(%R9,%R8,1)    [1]

0x418158 ADD	$0x18,%R8

0x41815c DEC	%EDI

0x41815e JNE	418110

/beegfs/hackathon/users/eoseret/qaas_runs/170-850-7424/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];

Path /

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	4.80
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.29
Bottlenecks	P5, P6, P7,
Function	advanceVelocity.extracted
Source	timestep.c:74-78
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	3.00
CQA cycles if no scalar integer	3.00
CQA cycles if FP arith vectorized	3.00
CQA cycles if fully vectorized	0.63
Front-end cycles	2.33
DIV/SQRT cycles	0.50
P0 cycles	0.50
P1 cycles	0.25
P2 cycles	0.25
P3 cycles	0.50
P4 cycles	3.00
P5 cycles	3.00
P6 cycles	3.00
P7 cycles	1.50
P8 cycles	1.50
P9 cycles	1.50
P10 cycles	1.50
P11 cycles	1.50
P12 cycles	1.50
P13 cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	15.00
Nb uops	14.00
Nb loads	6.00
Nb stores	3.00
Nb stack references	0.00
FLOP/cycle	2.00
Nb FLOP add-sub	3.00
Nb FLOP mul	3.00
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	24.00
Bytes prefetched	0.00
Bytes loaded	48.00
Bytes stored	24.00
Stride 0	0.00
Stride 1	2.00
Stride n	0.00
Stride unknown	0.00
Stride indirect	0.00
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	0.00
Vectorization ratio add_sub	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	NA
Vector-efficiency ratio all	12.50
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	12.50
Vector-efficiency ratio mul	12.50
Vector-efficiency ratio add_sub	12.50
Vector-efficiency ratio fma	NA
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	4.80
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.29
Bottlenecks	P5, P6, P7,
Function	advanceVelocity.extracted
Source	timestep.c:74-78
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	3.00
CQA cycles if no scalar integer	3.00
CQA cycles if FP arith vectorized	3.00
CQA cycles if fully vectorized	0.63
Front-end cycles	2.33
DIV/SQRT cycles	0.50
P0 cycles	0.50
P1 cycles	0.25
P2 cycles	0.25
P3 cycles	0.50
P4 cycles	3.00
P5 cycles	3.00
P6 cycles	3.00
P7 cycles	1.50
P8 cycles	1.50
P9 cycles	1.50
P10 cycles	1.50
P11 cycles	1.50
P12 cycles	1.50
P13 cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	15.00
Nb uops	14.00
Nb loads	6.00
Nb stores	3.00
Nb stack references	0.00
FLOP/cycle	2.00
Nb FLOP add-sub	3.00
Nb FLOP mul	3.00
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	24.00
Bytes prefetched	0.00
Bytes loaded	48.00
Bytes stored	24.00
Stride 0	0.00
Stride 1	2.00
Stride n	0.00
Stride unknown	0.00
Stride indirect	0.00
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	0.00
Vectorization ratio add_sub	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	NA
Vector-efficiency ratio all	12.50
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	12.50
Vector-efficiency ratio mul	12.50
Vector-efficiency ratio add_sub	12.50
Vector-efficiency ratio fma	NA
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.extracted
Source file and lines	timestep.c:74-78
Module	exec

The loop is defined in /beegfs/hackathon/users/eoseret/qaas_runs/170-850-7424/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:74-78.

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

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 0.62 cycles (4.80x 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

Performance is limited by:

reading data from caches/RAM (load units are a bottleneck)
writing data to caches/RAM (the store unit is a bottleneck)

By removing all these bottlenecks, you can lower the cost of an iteration from 3.00 to 2.33 cycles (1.29x speedup).

Workaround

Read less array elements
Write less array elements
Provide more information to your compiler:
- hardcode the bounds of the corresponding 'for' loop
- use the 'restrict' C99 keyword

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).

Type of elements and instruction set

6 SSE or AVX instructions are processing arithmetic or math operations on double precision FP elements in scalar mode (one at a time).

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

The binary loop is composed of 6 FP arithmetical operations:

3: addition or subtraction
3: multiply

The binary loop is loading 48 bytes (6 double precision FP elements). The binary loop is storing 24 bytes (3 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.08 FP operations per loaded or stored byte.

Unroll opportunity

Loop is data access bound.

Workaround

Unroll your loop if trip count is significantly higher than target unroll factor and if some data references are common to consecutive iterations. This can be done manually. Or with the unroll (resp. unroll_and_jam) directive on top of the inner (resp. surrounding) loop. You can enforce an unroll factor: #pragma unroll_and_jam N, unroll_and_jam(N), unroll N or unroll(N)

General properties

nb instructions	15
nb uops	14
loop length	80
used x86 registers	4
used mmx registers	0
used xmm registers	2
used ymm registers	0
used zmm registers	0
nb stack references	0
ADD-SUB / MUL ratio	1.00

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	2.33 cycles
front end	2.33 cycles

Back-end

	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	AGU0	AGU1	AGU2	FP0	FP1	FP2	FP3	FP4	FP5
uops	0.50	0.50	0.25	0.25	0.50	3.00	3.00	3.00	1.50	1.50	1.50	1.50	1.50	1.50
cycles	0.50	0.50	0.25	0.25	0.50	3.00	3.00	3.00	1.50	1.50	1.50	1.50	1.50	1.50

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)	1.00

Cycles summary

Front-end	2.33
Dispatch	3.00
Data deps.	1.00
Overall L1	3.00

Vectorization ratios

all	0%
load	0%
store	0%
mul	0%
add-sub	0%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

Vector efficiency ratios

all	12%
load	12%
store	12%
mul	12%
add-sub	12%
fma	NA (no fma vectorizable/vectorized instructions)
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 3.00 cycles. At this rate:

16% of peak load performance is reached (16.00 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
12% of peak store performance is reached (8.00 out of 64.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 418110

Instruction	Nb FU	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	AGU0	AGU1	AGU2	FP0	FP1	FP2	FP3	FP4	FP5	Latency	Recip. throughput
MOVSD -0x10(%R10,%R8,1),%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	1	0.50
MULSD %XMM0,%XMM1	1	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50
ADDSD -0x10(%R9,%R8,1),%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0.50	0.50	0	0	3	0.50
MOVSD %XMM1,-0x10(%R9,%R8,1)	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	0.50	0.50	1	1
MOVSD -0x8(%R10,%R8,1),%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	1	0.50
MULSD %XMM0,%XMM1	1	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50
ADDSD -0x8(%R9,%R8,1),%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0.50	0.50	0	0	3	0.50
MOVSD %XMM1,-0x8(%R9,%R8,1)	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	0.50	0.50	1	1
MOVSD (%R10,%R8,1),%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	1	0.50
MULSD %XMM0,%XMM1	1	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50
ADDSD (%R9,%R8,1),%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0.50	0.50	0	0	3	0.50
MOVSD %XMM1,(%R9,%R8,1)	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	0.50	0.50	1	1
ADD $0x18,%R8	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	0	1	0.25
DEC %EDI	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	0	1	0.25
JNE 418110 <advanceVelocity.extracted+0xd0>	1	0.50	0	0	0	0.50	0	0	0	0	0	0	0	0	0	1	0.50-1

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

The loop is defined in /beegfs/hackathon/users/eoseret/qaas_runs/170-850-7424/intel/CoMD/build/CoMD/CoMD/src-openmp/timestep.c:74-78.

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

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 0.62 cycles (4.80x 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

Performance is limited by:

reading data from caches/RAM (load units are a bottleneck)
writing data to caches/RAM (the store unit is a bottleneck)

By removing all these bottlenecks, you can lower the cost of an iteration from 3.00 to 2.33 cycles (1.29x speedup).

Workaround

Read less array elements
Write less array elements
Provide more information to your compiler:
- hardcode the bounds of the corresponding 'for' loop
- use the 'restrict' C99 keyword

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).

Type of elements and instruction set

6 SSE or AVX instructions are processing arithmetic or math operations on double precision FP elements in scalar mode (one at a time).

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

The binary loop is composed of 6 FP arithmetical operations:

3: addition or subtraction
3: multiply

The binary loop is loading 48 bytes (6 double precision FP elements). The binary loop is storing 24 bytes (3 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.08 FP operations per loaded or stored byte.

Unroll opportunity

Loop is data access bound.

Workaround

Unroll your loop if trip count is significantly higher than target unroll factor and if some data references are common to consecutive iterations. This can be done manually. Or with the unroll (resp. unroll_and_jam) directive on top of the inner (resp. surrounding) loop. You can enforce an unroll factor: #pragma unroll_and_jam N, unroll_and_jam(N), unroll N or unroll(N)

General properties

nb instructions	15
nb uops	14
loop length	80
used x86 registers	4
used mmx registers	0
used xmm registers	2
used ymm registers	0
used zmm registers	0
nb stack references	0
ADD-SUB / MUL ratio	1.00

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	2.33 cycles
front end	2.33 cycles

Back-end

	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	AGU0	AGU1	AGU2	FP0	FP1	FP2	FP3	FP4	FP5
uops	0.50	0.50	0.25	0.25	0.50	3.00	3.00	3.00	1.50	1.50	1.50	1.50	1.50	1.50
cycles	0.50	0.50	0.25	0.25	0.50	3.00	3.00	3.00	1.50	1.50	1.50	1.50	1.50	1.50

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)	1.00

Cycles summary

Front-end	2.33
Dispatch	3.00
Data deps.	1.00
Overall L1	3.00

Vectorization ratios

all	0%
load	0%
store	0%
mul	0%
add-sub	0%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

Vector efficiency ratios

all	12%
load	12%
store	12%
mul	12%
add-sub	12%
fma	NA (no fma vectorizable/vectorized instructions)
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 3.00 cycles. At this rate:

16% of peak load performance is reached (16.00 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
12% of peak store performance is reached (8.00 out of 64.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 418110

Instruction	Nb FU	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	AGU0	AGU1	AGU2	FP0	FP1	FP2	FP3	FP4	FP5	Latency	Recip. throughput
MOVSD -0x10(%R10,%R8,1),%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	1	0.50
MULSD %XMM0,%XMM1	1	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50
ADDSD -0x10(%R9,%R8,1),%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0.50	0.50	0	0	3	0.50
MOVSD %XMM1,-0x10(%R9,%R8,1)	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	0.50	0.50	1	1
MOVSD -0x8(%R10,%R8,1),%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	1	0.50
MULSD %XMM0,%XMM1	1	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50
ADDSD -0x8(%R9,%R8,1),%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0.50	0.50	0	0	3	0.50
MOVSD %XMM1,-0x8(%R9,%R8,1)	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	0.50	0.50	1	1
MOVSD (%R10,%R8,1),%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	1	0.50
MULSD %XMM0,%XMM1	1	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50
ADDSD (%R9,%R8,1),%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0.50	0.50	0	0	3	0.50
MOVSD %XMM1,(%R9,%R8,1)	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	0.50	0.50	1	1
ADD $0x18,%R8	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	0	1	0.25
DEC %EDI	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	0	1	0.25
JNE 418110 <advanceVelocity.extracted+0xd0>	1	0.50	0	0	0	0.50	0	0	0	0	0	0	0	0	0	1	0.50-1

Report Configuration

Vectorization

Details

Workaround

Execution units bottlenecks

Workaround

FMA

Workaround

Type of elements and instruction set

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

Arithmetic intensity

Unroll opportunity

Workaround

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

Type of elements and instruction set

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

Arithmetic intensity

Unroll opportunity

Workaround

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code