OV - exec - Loop 327

0x4256e8 LDR	D0, [X16], #8    [1]

0x4256ec ORR	V24.16B, V2.16B, V2.16B

0x4256f0 ORR	V23.16B, V4.16B, V4.16B

0x4256f4 SUBS	X17, X17, #1

0x4256f8 FMUL	D0, D0, D1

0x4256fc FRINTZ	D3, D0

0x425700 FCVTZS	W18, D0

0x425704 FSUB	D0, D0, S3

0x425708 ADD	X18, X15, W18,SXTW #3

0x42570c FMUL	D3, D0, D0

0x425710 FMLA	V24.2D, V17.2D, V0.D[0]

0x425714 FMLA	V23.2D, V7.2D, V0.D[0]

0x425718 LDP	Q21, Q22, [X18]    [2]

0x42571c FMUL	D6, D3, D0

0x425720 FMLA	V24.2D, V5.2D, V3.D[0]

0x425724 FMLA	V23.2D, V16.2D, V3.D[0]

0x425728 FMLA	V24.2D, V20.2D, V6.D[0]

0x42572c FMLA	V23.2D, V19.2D, V6.D[0]

0x425730 FMUL	V0.2D, V24.2D, V22.2D

0x425734 FMLA	V0.2D, V23.2D, V21.2D

0x425738 FADDP	D0, V0.2D

0x42573c FADD	D18, D18, D0

0x425740 B.NE	4256e8

/usr/lib/gcc/aarch64-amazon-linux/14/../../../../include/c++/14/bits/stl_vector.h: 1150 - 1150

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

1150: 	return *(this->_M_impl._M_start + __n);

/home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/build/miniqmc/src/QMCWaveFunctions/Jastrow/BsplineFunctor.h: 246 - 260

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

246:   for (int jat = 0; jat < iCount; jat++)

247:   {

248:     real_type r = distArrayCompressed[jat];

249:     r *= DeltaRInv;

250:     int i         = (int)r;

251:     real_type t   = r - real_type(i);

252:     real_type tp0 = t * t * t;

253:     real_type tp1 = t * t;

254:     real_type tp2 = t;

255:

256:     real_type d1 = SplineCoefs[i + 0] * (A[0] * tp0 + A[1] * tp1 + A[2] * tp2 + A[3]);

257:     real_type d2 = SplineCoefs[i + 1] * (A[4] * tp0 + A[5] * tp1 + A[6] * tp2 + A[7]);

258:     real_type d3 = SplineCoefs[i + 2] * (A[8] * tp0 + A[9] * tp1 + A[10] * tp2 + A[11]);

259:     real_type d4 = SplineCoefs[i + 3] * (A[12] * tp0 + A[13] * tp1 + A[14] * tp2 + A[15]);

260:     d += (d1 + d2 + d3 + d4);

Coverage (%)	Name	Source Location	Module
►100.00+	qmcplusplus::WaveFunction::eva[...]	WaveFunction.cpp:272	exec
○	qmcplusplus::NonLocalPP<double[...]	NonLocalPP.hpp:126	exec
○	main.omp_outlined.62	NewTimer.h:249	exec
○	__kmp_invoke_microtask		libomp.so
○	__kmp_fork_call		libomp.so
○	__kmpc_fork_call		libomp.so
○	main	miniqmc.cpp:409	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start	new_allocator.h:172	exec

Coverage (%)	Name	Source Location	Module
►100.00+	qmcplusplus::WaveFunction::eva[...]	WaveFunction.cpp:272	exec
○	qmcplusplus::NonLocalPP<double[...]	NonLocalPP.hpp:126	exec
○	main.omp_outlined.62	NewTimer.h:249	exec
○	__kmp_invoke_microtask		libomp.so
○	__kmp_invoke_task_func		libomp.so
○	__kmp_launch_thread		libomp.so
○	__kmp_launch_worker(void*)		libomp.so
○	start_thread		libc.so.6
○	thread_start		libc.so.6

Coverage (%)	Name	Source Location	Module
►100.00+	qmcplusplus::WaveFunction::eva[...]	WaveFunction.cpp:272	exec
○	qmcplusplus::NonLocalPP<double[...]	NonLocalPP.hpp:126	exec
○	main.omp_outlined.62	NewTimer.h:249	exec
○	__kmp_invoke_microtask		libomp.so
○	__kmp_invoke_task_func		libomp.so
○	__kmp_launch_thread		libomp.so
○	__kmp_launch_worker(void*)		libomp.so
○	start_thread		libc.so.6
○	thread_start		libc.so.6

Coverage (%)	Name	Source Location	Module
►100.00+	qmcplusplus::WaveFunction::eva[...]	WaveFunction.cpp:272	exec
○	qmcplusplus::NonLocalPP<double[...]	NonLocalPP.hpp:126	exec
○	main.omp_outlined.62	NewTimer.h:249	exec
○	__kmp_invoke_microtask		libomp.so
○	__kmp_invoke_task_func		libomp.so
○	__kmp_launch_thread		libomp.so
○	__kmp_launch_worker(void*)		libomp.so
○	start_thread		libc.so.6
○	thread_start		libc.so.6

Coverage (%)	Name	Source Location	Module
►100.00+	qmcplusplus::WaveFunction::eva[...]	WaveFunction.cpp:272	exec
○	qmcplusplus::NonLocalPP<double[...]	NonLocalPP.hpp:126	exec
○	main.omp_outlined.62	NewTimer.h:249	exec
○	__kmp_invoke_microtask		libomp.so
○	__kmp_invoke_task_func		libomp.so
○	__kmp_launch_thread		libomp.so
○	__kmp_launch_worker(void*)		libomp.so
○	start_thread		libc.so.6
○	thread_start		libc.so.6

Coverage (%)	Name	Source Location	Module
►100.00+	qmcplusplus::WaveFunction::eva[...]	WaveFunction.cpp:272	exec
○	qmcplusplus::NonLocalPP<double[...]	NonLocalPP.hpp:126	exec
○	main.omp_outlined.62	NewTimer.h:249	exec
○	__kmp_invoke_microtask		libomp.so
○	__kmp_invoke_task_func		libomp.so
○	__kmp_launch_thread		libomp.so
○	__kmp_launch_worker(void*)		libomp.so
○	start_thread		libc.so.6
○	thread_start		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	1.29
CQA speedup if FP arith vectorized	1.85
CQA speedup if fully vectorized	2.48
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.57
Bottlenecks	P6, P7, P8, P9,
Function	miniqmcreference::TwoBodyJastrowRef >::evaluateRatios(qmcplusplus::VirtualParticleSet&, std::vector >&)
Source	stl_vector.h:1150-1150,BsplineFunctor.h:246-260
Source loop unroll info	multi-versionned
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	4.50
CQA cycles if no scalar integer	3.50
CQA cycles if FP arith vectorized	2.44
CQA cycles if fully vectorized	1.81
Front-end cycles	2.88
P0 cycles	0.50
P1 cycles	0.50
P2 cycles	0.75
P3 cycles	0.75
P4 cycles	0.75
P5 cycles	0.75
P6 cycles	4.50
P7 cycles	4.50
P8 cycles	4.50
P9 cycles	4.50
P10 cycles	0.67
P11 cycles	0.67
P12 cycles	0.67
P13 cycles	0.00
P14 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	2
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	23.00
Nb uops	23.00
Nb loads	NA
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	8.00
Nb FLOP add-sub	3.00
Nb FLOP mul	5.00
Nb FLOP fma	14.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	0.00
Bytes prefetched	0.00
Bytes loaded	0.00
Bytes stored	0.00
Stride 0	1.00
Stride 1	0.00
Stride n	0.00
Stride unknown	1.00
Stride indirect	0.00
Vectorization ratio all	52.38
Vectorization ratio load	50.00
Vectorization ratio store	NA
Vectorization ratio mul	25.00
Vectorization ratio add_sub	0.00
Vectorization ratio fma	100.00
Vectorization ratio div_sqrt	NA
Vectorization ratio other	50.00
Vector-efficiency ratio all	41.67
Vector-efficiency ratio load	62.50
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	31.25
Vector-efficiency ratio add_sub	31.25
Vector-efficiency ratio fma	50.00
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	37.50

Metric	Value
CQA speedup if no scalar integer	1.29
CQA speedup if FP arith vectorized	1.85
CQA speedup if fully vectorized	2.48
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.57
Bottlenecks	P6, P7, P8, P9,
Function	miniqmcreference::TwoBodyJastrowRef >::evaluateRatios(qmcplusplus::VirtualParticleSet&, std::vector >&)
Source	stl_vector.h:1150-1150,BsplineFunctor.h:246-260
Source loop unroll info	multi-versionned
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	4.50
CQA cycles if no scalar integer	3.50
CQA cycles if FP arith vectorized	2.44
CQA cycles if fully vectorized	1.81
Front-end cycles	2.88
P0 cycles	0.50
P1 cycles	0.50
P2 cycles	0.75
P3 cycles	0.75
P4 cycles	0.75
P5 cycles	0.75
P6 cycles	4.50
P7 cycles	4.50
P8 cycles	4.50
P9 cycles	4.50
P10 cycles	0.67
P11 cycles	0.67
P12 cycles	0.67
P13 cycles	0.00
P14 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	2
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	23.00
Nb uops	23.00
Nb loads	NA
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	8.00
Nb FLOP add-sub	3.00
Nb FLOP mul	5.00
Nb FLOP fma	14.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	0.00
Bytes prefetched	0.00
Bytes loaded	0.00
Bytes stored	0.00
Stride 0	1.00
Stride 1	0.00
Stride n	0.00
Stride unknown	1.00
Stride indirect	0.00
Vectorization ratio all	52.38
Vectorization ratio load	50.00
Vectorization ratio store	NA
Vectorization ratio mul	25.00
Vectorization ratio add_sub	0.00
Vectorization ratio fma	100.00
Vectorization ratio div_sqrt	NA
Vectorization ratio other	50.00
Vector-efficiency ratio all	41.67
Vector-efficiency ratio load	62.50
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	31.25
Vector-efficiency ratio add_sub	31.25
Vector-efficiency ratio fma	50.00
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	37.50

Path /

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

Function	miniqmcreference::TwoBodyJastrowRef >::evaluateRatios(qmcplusplus::VirtualParticleSet&, std::vector >&)
Source file and lines	BsplineFunctor.h:246-260
Module	exec

The loop is defined in:

/usr/lib/gcc/aarch64-amazon-linux/14/../../../../include/c++/14/bits/stl_vector.h:1150
/home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/build/miniqmc/src/QMCWaveFunctions/Jastrow/BsplineFunctor.h:246-260

The related source loop is multi-versionned.

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 4.50 to 3.50 cycles (1.29x speedup).

Workaround

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

Vectorization

Your loop is partially vectorized. Only 41% 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 4.50 to 1.81 cycles (2.48x speedup).

Details

52% 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.
0% of VPU addition or subtraction instructions are used in vector version.
25% of VPU multiply instructions are used in vector version.
50% of VPU 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

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)

FMA

Detected 14 FMA (fused multiply-add) operations. Presence of both ADD/SUB and MUL operations.

Workaround

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: 1 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)

Conversion instructions

Detected expensive conversion instructions.

Details

FCVTZS: 1 occurrences►
- /home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/build/miniqmc/src/QMCWaveFunctions/Jastrow/BsplineFunctor.h:250

Workaround

Avoid mixing data with different types. In particular, check if the type of constants is the same as array elements.

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

The binary loop is composed of 36 FP arithmetical operations:

17: addition or subtraction (14 inside FMA instructions)
19: multiply (14 inside FMA instructions)

The binary loop does not load or store any data.

General properties

nb instructions	23
nb uops	23
loop length	92
used w registers	1
used x registers	4
used b registers	0
used h registers	0
used s registers	1
used d registers	5
used q registers	2
used v registers	15
used z registers	0
nb stack references	0
ADD-SUB / MUL ratio	0.75

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	2.88 cycles
front end	2.88 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	0.50	0.50	0.75	0.75	0.75	0.75	4.50	4.50	4.50	4.50	0.67	0.67	0.67	0.00	0.00
cycles	0.50	0.50	0.75	0.75	0.75	0.75	4.50	4.50	4.50	4.50	0.67	0.67	0.67	0.00	0.00

Execution ports to units layout:

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

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	2.00

Cycles summary

Front-end	2.88
Dispatch	4.50
Data deps.	2.00
Overall L1	4.50

Vectorization ratios

INT

all	60%
load	50%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	0%
fma	NA (no fma vectorizable/vectorized instructions)
other	100%

FP

all	50%
load	NA (no load vectorizable/vectorized instructions)
store	NA (no store vectorizable/vectorized instructions)
mul	25%
add-sub	0%
fma	100%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	0%

INT+FP

all	52%
load	50%
store	NA (no store vectorizable/vectorized instructions)
mul	25%
add-sub	0%
fma	100%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	50%

Vector efficiency ratios

INT

all	50%
load	62%
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	50%

FP

all	39%
load	NA (no load vectorizable/vectorized instructions)
store	NA (no store vectorizable/vectorized instructions)
mul	31%
add-sub	33%
fma	50%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	25%

INT+FP

all	41%
load	62%
store	NA (no store vectorizable/vectorized instructions)
mul	31%
add-sub	31%
fma	50%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	37%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 4.50 cycles.

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 4256e8

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	Latency	Recip. throughput	Vectorization
LDR D0, [X16], #8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.33	0.33	0.33	6	0.33	scal (25.0%)
ORR V24.16B, V2.16B, V2.16B	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	vect (50.0%)
ORR V23.16B, V4.16B, V4.16B	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	vect (50.0%)
SUBS X17, X17, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	1	0.33	scal (25.0%)
FMUL D0, D0, D1	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	scal (25.0%)
FRINTZ D3, D0	1	0	0	0	0	0	0	0.50	0	0.50	0	0	0	0	3	0.50	scal (25.0%)
FCVTZS W18, D0	1	0	0	0	0	0	0	1	0	0	0	0	0	0	3	1	scal (25.0%)
FSUB D0, D0, S3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	scal (25.0%)
ADD X18, X15, W18,SXTW #3	1	0	0	0	0	0.50	0.50	0	0	0	0	0	0	0	2	0.50	N/A
FMUL D3, D0, D0	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	scal (25.0%)
FMLA V24.2D, V17.2D, V0.D[0]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (50.0%)
FMLA V23.2D, V7.2D, V0.D[0]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (50.0%)
LDP Q21, Q22, [X18]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.67	vect (100.0%)
FMUL D6, D3, D0	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	scal (25.0%)
FMLA V24.2D, V5.2D, V3.D[0]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (50.0%)
FMLA V23.2D, V16.2D, V3.D[0]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (50.0%)
FMLA V24.2D, V20.2D, V6.D[0]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (50.0%)
FMLA V23.2D, V19.2D, V6.D[0]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (50.0%)
FMUL V0.2D, V24.2D, V22.2D	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	vect (50.0%)
FMLA V0.2D, V23.2D, V21.2D	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (50.0%)
FADDP D0, V0.2D	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	scal (50.0%)
FADD D18, D18, D0	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	scal (25.0%)
B.NE 4256e8 <_ZN16miniqmcreference17TwoBodyJastrowRefIN11qmcplusplus14BsplineFunctorIdEEE14evaluateRatiosERNS1_18VirtualParticleSetERSt6vectorIdSaIdEE+0x3a8>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A

Function	miniqmcreference::TwoBodyJastrowRef >::evaluateRatios(qmcplusplus::VirtualParticleSet&, std::vector >&)
Source file and lines	BsplineFunctor.h:246-260
Module	exec

The loop is defined in:

/usr/lib/gcc/aarch64-amazon-linux/14/../../../../include/c++/14/bits/stl_vector.h:1150
/home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/build/miniqmc/src/QMCWaveFunctions/Jastrow/BsplineFunctor.h:246-260

The related source loop is multi-versionned.

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 4.50 to 3.50 cycles (1.29x speedup).

Workaround

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

Vectorization

Your loop is partially vectorized. Only 41% 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 4.50 to 1.81 cycles (2.48x speedup).

Details

52% 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.
0% of VPU addition or subtraction instructions are used in vector version.
25% of VPU multiply instructions are used in vector version.
50% of VPU 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

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)

FMA

Detected 14 FMA (fused multiply-add) operations. Presence of both ADD/SUB and MUL operations.

Workaround

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: 1 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)

Conversion instructions

Detected expensive conversion instructions.

Details

FCVTZS: 1 occurrences►
- /home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/build/miniqmc/src/QMCWaveFunctions/Jastrow/BsplineFunctor.h:250

Workaround

Avoid mixing data with different types. In particular, check if the type of constants is the same as array elements.

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

The binary loop is composed of 36 FP arithmetical operations:

17: addition or subtraction (14 inside FMA instructions)
19: multiply (14 inside FMA instructions)

The binary loop does not load or store any data.

General properties

nb instructions	23
nb uops	23
loop length	92
used w registers	1
used x registers	4
used b registers	0
used h registers	0
used s registers	1
used d registers	5
used q registers	2
used v registers	15
used z registers	0
nb stack references	0
ADD-SUB / MUL ratio	0.75

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	2.88 cycles
front end	2.88 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	0.50	0.50	0.75	0.75	0.75	0.75	4.50	4.50	4.50	4.50	0.67	0.67	0.67	0.00	0.00
cycles	0.50	0.50	0.75	0.75	0.75	0.75	4.50	4.50	4.50	4.50	0.67	0.67	0.67	0.00	0.00

Execution ports to units layout:

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

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	2.00

Cycles summary

Front-end	2.88
Dispatch	4.50
Data deps.	2.00
Overall L1	4.50

Vectorization ratios

INT

all	60%
load	50%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	0%
fma	NA (no fma vectorizable/vectorized instructions)
other	100%

FP

all	50%
load	NA (no load vectorizable/vectorized instructions)
store	NA (no store vectorizable/vectorized instructions)
mul	25%
add-sub	0%
fma	100%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	0%

INT+FP

all	52%
load	50%
store	NA (no store vectorizable/vectorized instructions)
mul	25%
add-sub	0%
fma	100%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	50%

Vector efficiency ratios

INT

all	50%
load	62%
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	50%

FP

all	39%
load	NA (no load vectorizable/vectorized instructions)
store	NA (no store vectorizable/vectorized instructions)
mul	31%
add-sub	33%
fma	50%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	25%

INT+FP

all	41%
load	62%
store	NA (no store vectorizable/vectorized instructions)
mul	31%
add-sub	31%
fma	50%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	37%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 4.50 cycles.

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 4256e8

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	Latency	Recip. throughput	Vectorization
LDR D0, [X16], #8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.33	0.33	0.33	6	0.33	scal (25.0%)
ORR V24.16B, V2.16B, V2.16B	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	vect (50.0%)
ORR V23.16B, V4.16B, V4.16B	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	vect (50.0%)
SUBS X17, X17, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	1	0.33	scal (25.0%)
FMUL D0, D0, D1	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	scal (25.0%)
FRINTZ D3, D0	1	0	0	0	0	0	0	0.50	0	0.50	0	0	0	0	3	0.50	scal (25.0%)
FCVTZS W18, D0	1	0	0	0	0	0	0	1	0	0	0	0	0	0	3	1	scal (25.0%)
FSUB D0, D0, S3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	scal (25.0%)
ADD X18, X15, W18,SXTW #3	1	0	0	0	0	0.50	0.50	0	0	0	0	0	0	0	2	0.50	N/A
FMUL D3, D0, D0	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	scal (25.0%)
FMLA V24.2D, V17.2D, V0.D[0]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (50.0%)
FMLA V23.2D, V7.2D, V0.D[0]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (50.0%)
LDP Q21, Q22, [X18]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.67	vect (100.0%)
FMUL D6, D3, D0	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	scal (25.0%)
FMLA V24.2D, V5.2D, V3.D[0]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (50.0%)
FMLA V23.2D, V16.2D, V3.D[0]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (50.0%)
FMLA V24.2D, V20.2D, V6.D[0]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (50.0%)
FMLA V23.2D, V19.2D, V6.D[0]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (50.0%)
FMUL V0.2D, V24.2D, V22.2D	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	3	0.25	vect (50.0%)
FMLA V0.2D, V23.2D, V21.2D	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25	vect (50.0%)
FADDP D0, V0.2D	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	scal (50.0%)
FADD D18, D18, D0	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	2	0.25	scal (25.0%)
B.NE 4256e8 <_ZN16miniqmcreference17TwoBodyJastrowRefIN11qmcplusplus14BsplineFunctorIdEEE14evaluateRatiosERNS1_18VirtualParticleSetERSt6vectorIdSaIdEE+0x3a8>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A

▶Scalability Runs Description

Run 1x1	Number processes: 1Number nodes: NARun Command: <executable> -g "4 2 2" -bMPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/run/oneview_runs/multicore/armclang/oneview_run_1782144418OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_NUM_THREADS: 1OMP_PLACES: threads
Run 1x2	Number processes: 1Run Command: <executable> -g "4 2 2" -bMPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/run/oneview_runs/multicore/armclang/oneview_run_1782144418OMP_NUM_THREADS: 2OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x4	Number processes: 1Run Command: <executable> -g "4 2 2" -bMPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/run/oneview_runs/multicore/armclang/oneview_run_1782144418OMP_NUM_THREADS: 4OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x8	Number processes: 1Run Command: <executable> -g "4 2 2" -bMPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/run/oneview_runs/multicore/armclang/oneview_run_1782144418OMP_NUM_THREADS: 8OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x16	Number processes: 1Run Command: <executable> -g "4 2 2" -bMPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/run/oneview_runs/multicore/armclang/oneview_run_1782144418OMP_NUM_THREADS: 16OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x24	Number processes: 1Run Command: <executable> -g "4 2 2" -bMPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/run/oneview_runs/multicore/armclang/oneview_run_1782144418OMP_NUM_THREADS: 24OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x32	Number processes: 1Run Command: <executable> -g "4 2 2" -bMPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/run/oneview_runs/multicore/armclang/oneview_run_1782144418OMP_NUM_THREADS: 32OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x40	Number processes: 1Run Command: <executable> -g "4 2 2" -bMPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/run/oneview_runs/multicore/armclang/oneview_run_1782144418OMP_NUM_THREADS: 40OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x48	Number processes: 1Run Command: <executable> -g "4 2 2" -bMPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/run/oneview_runs/multicore/armclang/oneview_run_1782144418OMP_NUM_THREADS: 48OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x56	Number processes: 1Run Command: <executable> -g "4 2 2" -bMPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/run/oneview_runs/multicore/armclang/oneview_run_1782144418OMP_NUM_THREADS: 56OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x64	Number processes: 1Run Command: <executable> -g "4 2 2" -bMPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-212-9071/intel/miniqmc/run/oneview_runs/multicore/armclang/oneview_run_1782144418OMP_NUM_THREADS: 64OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads

(1x1) Efficiency	(1x1) Potential Speed-Up (%)	(1x2) Efficiency	(1x2) Potential Speed-Up (%)	(1x4) Efficiency	(1x4) Potential Speed-Up (%)	(1x8) Efficiency	(1x8) Potential Speed-Up (%)	(1x16) Efficiency	(1x16) Potential Speed-Up (%)	(1x24) Efficiency	(1x24) Potential Speed-Up (%)	(1x32) Efficiency	(1x32) Potential Speed-Up (%)	(1x40) Efficiency	(1x40) Potential Speed-Up (%)	(1x48) Efficiency	(1x48) Potential Speed-Up (%)	(1x56) Efficiency	(1x56) Potential Speed-Up (%)	(1x64) Efficiency	(1x64) Potential Speed-Up (%)
1	0			2	0	3.99	0	7.96	0			7.97	0	13.25	0

Speed-Up and Efficiency

▶Detailed Speed-Up and Efficiency

Run	Number of threads	Efficiency (ideal is 1)	Speedup	Ideal Speedup	Time (s)	Coverage (%)
1x1	1	1	1	1	0.0049999998882413	0.011381772346795
1x2
1x4	2	2	2	4	0.0049999998882413	0.0055294591002166
1x8	2	3.99	3.99	8	0.0050000003539026	0.0025704009458423
1x16	1	7.96	7.96	16	0.0099999997764826	0.00097898172680289
1x24
1x32	2	7.97	7.97	32	0.0099999997764826	0.00067890627542511
1x40	1	13.25	13.25	40	0.015000000596046	0.00035227087209933
1x48
1x56
1x64

Report Configuration

Code clean check

Workaround

Vectorization

Details

Workaround

FMA

Workaround

Slow data structures access

Details

Workaround

Conversion instructions

Details

Workaround

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code

Code clean check

Workaround

Vectorization

Details

Workaround

FMA

Workaround

Slow data structures access

Details

Workaround

Conversion instructions

Details

Workaround

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code

▶Scalability Runs Description

▼Columns Filter

Speed-Up and Efficiency

▶Detailed Speed-Up and Efficiency