OV - exec - Loop 324

0x416f80 SUB	W11, W11, #1

0x416f84 ADD	X10, X10, #8

0x416f88 SUBS	X9, X9, #1

0x416f8c B.EQ	416fac

0x416f90 LDR	D2, [X10]    [1]

0x416f94 FCMP	D2, D1

0x416f98 CCMP	W11, #0, #4, #11

0x416f9c B.EQ	416f80

0x416fa0 STR	D2, [X5, X8,SXTW #3]    [2]

0x416fa4 ADD	W8, W8, #1

0x416fa8 B	416f80

/home/hbollore/qaas/qaas-runs/169-816-7419/intel/miniqmc/build/miniqmc/src/QMCWaveFunctions/Jastrow/BsplineFunctor.h: 236 - 241

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

236:   for (int jat = 0; jat < iLimit; jat++)

237:   {

238:     real_type r = distArray[jat];

239:     // pick the distances smaller than the cutoff and avoid the reference atom

240:     if (r < cutoff_radius && iStart + jat != iat)

241:       distArrayCompressed[iCount++] = distArray[jat];

Coverage (%)	Name	Source Location	Module
►100.00+	qmcplusplus::WaveFunction::rat[...]	NewTimer.h:249	exec
○	.omp_outlined..62	miniqmc.cpp:485	exec
○	__kmp_invoke_microtask		libomp.so

Path /

Metric	Value
CQA speedup if no scalar integer	1.80
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	6.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.09
Bottlenecks	P0, P1,
Function	qmcplusplus::BsplineFunctor::evaluateV(int, int, int, double const, double) const
Source	BsplineFunctor.h:236-241
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	1.50
CQA cycles if no scalar integer	0.83
CQA cycles if FP arith vectorized	1.50
CQA cycles if fully vectorized	0.25
Front-end cycles	1.38
DIV/SQRT cycles	1.50
P0 cycles	1.50
P1 cycles	1.25
P2 cycles	1.25
P3 cycles	1.25
P4 cycles	1.25
P5 cycles	1.00
P6 cycles	1.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.83
P10 cycles	0.50
P11 cycles	0.67
P12 cycles	0.00
P13 cycles	0.00
P14 cycles	0.00
Inter-iter dependencies cycles	NA
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	11.00
Nb uops	11.00
Nb loads	NA
Nb stores	1.00
Nb stack references	0.00
FLOP/cycle	0.00
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	0.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	10.67
Bytes prefetched	0.00
Bytes loaded	8.00
Bytes stored	8.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	20.00
Vector-efficiency ratio load	25.00
Vector-efficiency ratio store	25.00
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	18.75
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	12.50

Metric	Value
CQA speedup if no scalar integer	1.80
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	6.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.09
Bottlenecks	P0, P1,
Function	qmcplusplus::BsplineFunctor::evaluateV(int, int, int, double const, double) const
Source	BsplineFunctor.h:236-241
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	1.50
CQA cycles if no scalar integer	0.83
CQA cycles if FP arith vectorized	1.50
CQA cycles if fully vectorized	0.25
Front-end cycles	1.38
DIV/SQRT cycles	1.50
P0 cycles	1.50
P1 cycles	1.25
P2 cycles	1.25
P3 cycles	1.25
P4 cycles	1.25
P5 cycles	1.00
P6 cycles	1.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.83
P10 cycles	0.50
P11 cycles	0.67
P12 cycles	0.00
P13 cycles	0.00
P14 cycles	0.00
Inter-iter dependencies cycles	NA
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	11.00
Nb uops	11.00
Nb loads	NA
Nb stores	1.00
Nb stack references	0.00
FLOP/cycle	0.00
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	0.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	10.67
Bytes prefetched	0.00
Bytes loaded	8.00
Bytes stored	8.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	20.00
Vector-efficiency ratio load	25.00
Vector-efficiency ratio store	25.00
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	18.75
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	12.50

Path /

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

Function	qmcplusplus::BsplineFunctor::evaluateV(int, int, int, double const, double) const
Source file and lines	BsplineFunctor.h:236-241
Module	exec

The loop is defined in /home/hbollore/qaas/qaas-runs/169-816-7419/intel/miniqmc/build/miniqmc/src/QMCWaveFunctions/Jastrow/BsplineFunctor.h:236-241.

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

Ignoring paths for analysis
Failed to get the number of paths
RecMII not computed since number of paths is unknown or > max_paths

gain
potential
hint
expert

Code clean check

Detected a slowdown caused by scalar integer instructions (typically used for address computation). By removing them, you can lower the cost of an iteration from 1.50 to 0.83 cycles (1.80x speedup).

Workaround

Try to reorganize arrays of structures to structures of arrays
Consider to permute loops (see vectorization gain report)

Vectorization

Your loop is not vectorized. Only 20% of vector register length is used (average across all VPU instructions). By vectorizing your loop, you can lower the cost of an iteration from 1.50 to 0.25 cycles (6.00x speedup).

Details

All VPU 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
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)

No data for this section

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

The binary loop does not contain any FP arithmetical operations. The binary loop is loading 8 bytes. The binary loop is storing 8 bytes.

General properties

nb instructions	11
loop length	44
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

1.38 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.50	1.50	1.25	1.25	1.25	1.25	1.00	1.00	0.00	0.00	0.83	0.50	0.67	0.00	0.00
cycles	1.50	1.50	1.25	1.25	1.25	1.25	1.00	1.00	0.00	0.00	0.83	0.50	0.67	0.00	0.00

Cycles executing div or sqrt instructions

NA

Cycles summary

Front-end	1.38
Overall L1	1.50

Vectorization ratios

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.50 cycles. At this rate:

5% of peak load performance is reached (5.33 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
11% of peak store performance is reached (5.33 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 416f80

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P10	P11	P12	Latency	Recip. throughput
SUB W11, W11, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
ADD X10, X10, #8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
SUBS X9, X9, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33
B.EQ 416fac <_ZNK11qmcplusplus14BsplineFunctorIdE9evaluateVEiiiPKdPd+0x6c>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50
LDR D2, [X10]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
FCMP D2, D1	1	0	0	0	0	0	0	1	0	0	0	0	2	1
CCMP W11, #0, #4, #11	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
B.EQ 416f80 <_ZNK11qmcplusplus14BsplineFunctorIdE9evaluateVEiiiPKdPd+0x40>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50
STR D2, [X5, X8,SXTW #3]	1	0	0	0	0	0	0	0.50	0.50	0.50	0.50	0	2	0.50
ADD W8, W8, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
B 416f80 <_ZNK11qmcplusplus14BsplineFunctorIdE9evaluateVEiiiPKdPd+0x40>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50

Function	qmcplusplus::BsplineFunctor::evaluateV(int, int, int, double const, double) const
Source file and lines	BsplineFunctor.h:236-241
Module	exec

The loop is defined in /home/hbollore/qaas/qaas-runs/169-816-7419/intel/miniqmc/build/miniqmc/src/QMCWaveFunctions/Jastrow/BsplineFunctor.h:236-241.

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

Ignoring paths for analysis
Failed to get the number of paths
RecMII not computed since number of paths is unknown or > max_paths

gain
potential
hint
expert

Code clean check

Detected a slowdown caused by scalar integer instructions (typically used for address computation). By removing them, you can lower the cost of an iteration from 1.50 to 0.83 cycles (1.80x speedup).

Workaround

Try to reorganize arrays of structures to structures of arrays
Consider to permute loops (see vectorization gain report)

Vectorization

Your loop is not vectorized. Only 20% of vector register length is used (average across all VPU instructions). By vectorizing your loop, you can lower the cost of an iteration from 1.50 to 0.25 cycles (6.00x speedup).

Details

All VPU 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
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)

No data for this section

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

The binary loop does not contain any FP arithmetical operations. The binary loop is loading 8 bytes. The binary loop is storing 8 bytes.

General properties

nb instructions	11
loop length	44
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

1.38 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.50	1.50	1.25	1.25	1.25	1.25	1.00	1.00	0.00	0.00	0.83	0.50	0.67	0.00	0.00
cycles	1.50	1.50	1.25	1.25	1.25	1.25	1.00	1.00	0.00	0.00	0.83	0.50	0.67	0.00	0.00

Cycles executing div or sqrt instructions

NA

Cycles summary

Front-end	1.38
Overall L1	1.50

Vectorization ratios

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.50 cycles. At this rate:

5% of peak load performance is reached (5.33 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
11% of peak store performance is reached (5.33 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 416f80

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P10	P11	P12	Latency	Recip. throughput
SUB W11, W11, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
ADD X10, X10, #8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
SUBS X9, X9, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33
B.EQ 416fac <_ZNK11qmcplusplus14BsplineFunctorIdE9evaluateVEiiiPKdPd+0x6c>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50
LDR D2, [X10]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
FCMP D2, D1	1	0	0	0	0	0	0	1	0	0	0	0	2	1
CCMP W11, #0, #4, #11	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
B.EQ 416f80 <_ZNK11qmcplusplus14BsplineFunctorIdE9evaluateVEiiiPKdPd+0x40>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50
STR D2, [X5, X8,SXTW #3]	1	0	0	0	0	0	0	0.50	0.50	0.50	0.50	0	2	0.50
ADD W8, W8, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
B 416f80 <_ZNK11qmcplusplus14BsplineFunctorIdE9evaluateVEiiiPKdPd+0x40>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50

Report Configuration

Code clean check

Workaround

Vectorization

Details

Workaround

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code

Code clean check

Workaround

Vectorization

Details

Workaround

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code