OV - exec - Function qmcplusplus::WaveFunction::evalGrad(qmcplusplus::ParticleSet&, int)

	Function: qmcplusplus::WaveFunction::evalGrad(qmcplusplus::ParticleSet&, int)	Module: exec	Source: WaveFunction.cpp:188-197 [...]	Coverage: 0.01%

/usr/lib/gcc/aarch64-linux-gnu/11/../../../../include/c++/11/bits/stl_vector.h: 919 - 1046

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

919:       { return size_type(this->_M_impl._M_finish - this->_M_impl._M_start); }

[...]

1046: 	return *(this->_M_impl._M_start + __n);

/home/hbollore/qaas/qaas-runs/169-816-8620/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h: 94 - 94

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

94:     (const_cast<T1&>(a) += b);

/home/hbollore/qaas/qaas-runs/169-816-8620/intel/miniqmc/build/miniqmc/src/QMCWaveFunctions/WaveFunction.cpp: 188 - 197

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

188: {

189:   posT grad_iat = (iat < nelup ? Det_up->evalGrad(P, iat) : Det_dn->evalGrad(P, iat));

190:

191:   for (size_t i = 0; i < Jastrows.size(); i++)

192:   {

193:     ScopedTimer local_timer(jastrow_timers[i]);

194:     grad_iat += Jastrows[i]->evalGrad(P, iat);

195:   }

196:   return grad_iat;

197: }

/usr/lib/gcc/aarch64-linux-gnu/11/../../../../include/c++/11/bits/refwrap.h: 338 - 338

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

338:       { return *_M_data; }

/home/hbollore/qaas/qaas-runs/169-816-8620/intel/miniqmc/build/miniqmc/src/Utilities/NewTimer.h: 242 - 249

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

242:   ScopeGuard(TIMER& t) : timer(t) { timer.start(); }

[...]

249:   ~ScopeGuard() { timer.stop(); }

0x40ccc0 SUB	SP, SP, #96

0x40ccc4 STP	X29, X30, [SP, #32]

0x40ccc8 STP	X24, X23, [SP, #48]

0x40cccc STP	X22, X21, [SP, #64]

0x40ccd0 STP	X20, X19, [SP, #80]

0x40ccd4 ADD	X29, SP, #32

0x40ccd8 ORR	X22, XZR, X8

0x40ccdc LDR	W8, [X0, #52]

0x40cce0 ADD	X9, X0, #8

0x40cce4 ORR	X21, XZR, X0

0x40cce8 ORR	W19, WZR, W2

0x40ccec ORR	X20, XZR, X1

0x40ccf0 CMP	W8, W2

0x40ccf4 CSEL	X8, X0, X9, #12

0x40ccf8 LDR	X0, [X8]

0x40ccfc LDR	X8, [X0]

0x40cd00 LDR	X9, [X8, #24]

0x40cd04 ORR	X8, XZR, X22

0x40cd08 BLR	X9

0x40cd0c LDP	X9, X8, [X21, #16]

0x40cd10 CMP	X8, X9

0x40cd14 B.EQ	40cd8c

0x40cd18 ORR	X24, XZR, XZR

0x40cd1c HINT	#0

(136) 0x40cd20 LDR	X8, [X21, #88]

(136) 0x40cd24 LDR	X23, [X8, X24,LSL #3]

(136) 0x40cd28 ORR	X0, XZR, X23

(136) 0x40cd2c BL	44ed40

(136) 0x40cd30 LDR	X8, [X21, #16]

(136) 0x40cd34 LDR	X0, [X8, X24,LSL #3]

(136) 0x40cd38 LDR	X8, [X0]

(136) 0x40cd3c LDR	X9, [X8, #24]

(136) 0x40cd40 ORR	X1, XZR, X20

(136) 0x40cd44 ORR	W2, WZR, W19

(136) 0x40cd48 ADD	X8, SP, #0

(136) 0x40cd4c BLR	X9

(136) 0x40cd50 LDR	Q0, [SP]

(136) 0x40cd54 LDR	Q1, [X22]

(136) 0x40cd58 FADD	V0.2D, V1.2D, V0.2D

(136) 0x40cd5c LDR	D1, [X22, #16]

(136) 0x40cd60 STR	Q0, [X22]

(136) 0x40cd64 LDR	D0, [SP, #16]

(136) 0x40cd68 FADD	D0, D1, D0

(136) 0x40cd6c STR	D0, [X22, #16]

(136) 0x40cd70 ORR	X0, XZR, X23

(136) 0x40cd74 BL	44ee60

(136) 0x40cd78 LDP	X9, X8, [X21, #16]

(136) 0x40cd7c ADD	X24, X24, #1

(136) 0x40cd80 SUB	X8, X8, X9

(136) 0x40cd84 CMP	X24, X8,ASR #3

(136) 0x40cd88 B.CC	40cd20

0x40cd8c LDP	X20, X19, [SP, #80]

0x40cd90 LDP	X22, X21, [SP, #64]

0x40cd94 LDP	X24, X23, [SP, #48]

0x40cd98 LDP	X29, X30, [SP, #32]

0x40cd9c ADD	SP, SP, #96

0x40cda0 RET

0x40cda4 BL	409430

0x40cda8 ORR	X19, XZR, X0

0x40cdac ORR	X0, XZR, X23

0x40cdb0 BL	44ee60

0x40cdb4 ORR	X0, XZR, X19

0x40cdb8 BL	403dd0

0x40cdbc BL	409430

Coverage (%)	Name	Source Location	Module
►100.00+	.omp_outlined..62	miniqmc.cpp:428	exec
○	__kmp_invoke_microtask		libomp.so

Path /

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

Source file and lines	WaveFunction.cpp:188-197
Module	exec

The function is defined in:

/usr/lib/gcc/aarch64-linux-gnu/11/../../../../include/c++/11/bits/stl_vector.h:919,1046
/home/hbollore/qaas/qaas-runs/169-816-8620/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:94
/home/hbollore/qaas/qaas-runs/169-816-8620/intel/miniqmc/build/miniqmc/src/QMCWaveFunctions/WaveFunction.cpp:188-197
/usr/lib/gcc/aarch64-linux-gnu/11/../../../../include/c++/11/bits/refwrap.h:338
/home/hbollore/qaas/qaas-runs/169-816-8620/intel/miniqmc/build/miniqmc/src/Utilities/NewTimer.h:242,249

Warnings:
Ignoring paths for analysis

gain
potential
hint
expert

Vectorization

Your function is not vectorized. Only 22% of vector register length is used (average across all VPU instructions). By vectorizing your function, you can lower the cost of an iteration from 5.25 to 0.87 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 function can use their full power.

Workaround

Try another compiler or update/tune your current one
Make array accesses unit-stride:
- If your function 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)

Expensive FP math instructions/calls

Detected performance impact from expensive FP math instructions/calls. By removing/reexpressing them, you can lower the cost of an iteration from 5.25 to 2.50 cycles (2.10x speedup).

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

The binary function does not contain any FP arithmetical operations. The binary function is loading 108 bytes. The binary function is storing 64 bytes.

General properties

nb instructions	37
loop length	148
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

4.50 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	3.50	3.50	5.25	5.25	5.25	5.25	0.00	0.00	0.00	0.00	4.33	4.33	4.33	2.00	2.00
cycles	3.50	3.50	5.25	5.25	5.25	5.25	0.00	0.00	0.00	0.00	4.33	4.33	4.33	2.00	2.00

Cycles executing div or sqrt instructions

Cycles summary

Front-end	4.50
Overall L1	5.25

Vectorization ratios

all	0%
load	NA (no load vectorizable/vectorized instructions)
store	NA (no store vectorizable/vectorized instructions)
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 call to the function takes 5.25 cycles. At this rate:

21% of peak load performance is reached (20.57 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
25% of peak store performance is reached (12.19 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the function is: 40ccc0

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	Latency	Recip. throughput
SUB SP, SP, #96	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
STP X29, X30, [SP, #32]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STP X24, X23, [SP, #48]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STP X22, X21, [SP, #64]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STP X20, X19, [SP, #80]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
ADD X29, SP, #32	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ORR X22, XZR, X8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
LDR W8, [X0, #52]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
ADD X9, X0, #8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ORR X21, XZR, X0	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ORR W19, WZR, W2	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ORR X20, XZR, X1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
CMP W8, W2	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.33
CSEL X8, X0, X9, #12	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
LDR X0, [X8]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
LDR X8, [X0]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
LDR X9, [X8, #24]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
ORR X8, XZR, X22	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
BLR X9	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50
LDP X9, X8, [X21, #16]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
CMP X8, X9	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.33
B.EQ 40cd8c <_ZN11qmcplusplus12WaveFunction8evalGradERNS_11ParticleSetEi+0xcc>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
ORR X24, XZR, XZR	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
HINT #0
LDP X20, X19, [SP, #80]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X22, X21, [SP, #64]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X24, X23, [SP, #48]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X29, X30, [SP, #32]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
ADD SP, SP, #96	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
RET	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
BL 409430 <__clang_call_terminate>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50
ORR X19, XZR, X0	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ORR X0, XZR, X23	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
BL 44ee60 <_ZN11qmcplusplus9TimerTypeINS_8CPUClockEE4stopEv>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50
ORR X0, XZR, X19	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
BL 403dd0 <@plt_start@+0x3d0>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50
BL 409430 <__clang_call_terminate>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50

Source file and lines	WaveFunction.cpp:188-197
Module	exec

The function is defined in:

/usr/lib/gcc/aarch64-linux-gnu/11/../../../../include/c++/11/bits/stl_vector.h:919,1046
/home/hbollore/qaas/qaas-runs/169-816-8620/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:94
/home/hbollore/qaas/qaas-runs/169-816-8620/intel/miniqmc/build/miniqmc/src/QMCWaveFunctions/WaveFunction.cpp:188-197
/usr/lib/gcc/aarch64-linux-gnu/11/../../../../include/c++/11/bits/refwrap.h:338
/home/hbollore/qaas/qaas-runs/169-816-8620/intel/miniqmc/build/miniqmc/src/Utilities/NewTimer.h:242,249

Warnings:
Ignoring paths for analysis

gain
potential
hint
expert

Vectorization

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 function can use their full power.

Workaround

Try another compiler or update/tune your current one
Make array accesses unit-stride:
- If your function 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)

Expensive FP math instructions/calls

Detected performance impact from expensive FP math instructions/calls. By removing/reexpressing them, you can lower the cost of an iteration from 5.25 to 2.50 cycles (2.10x speedup).

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

The binary function does not contain any FP arithmetical operations. The binary function is loading 108 bytes. The binary function is storing 64 bytes.

General properties

nb instructions	37
loop length	148
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

4.50 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	3.50	3.50	5.25	5.25	5.25	5.25	0.00	0.00	0.00	0.00	4.33	4.33	4.33	2.00	2.00
cycles	3.50	3.50	5.25	5.25	5.25	5.25	0.00	0.00	0.00	0.00	4.33	4.33	4.33	2.00	2.00

Cycles executing div or sqrt instructions

Cycles summary

Front-end	4.50
Overall L1	5.25

Vectorization ratios

all	0%
load	NA (no load vectorizable/vectorized instructions)
store	NA (no store vectorizable/vectorized instructions)
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 call to the function takes 5.25 cycles. At this rate:

21% of peak load performance is reached (20.57 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
25% of peak store performance is reached (12.19 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the function is: 40ccc0

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	Latency	Recip. throughput
SUB SP, SP, #96	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
STP X29, X30, [SP, #32]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STP X24, X23, [SP, #48]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STP X22, X21, [SP, #64]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STP X20, X19, [SP, #80]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
ADD X29, SP, #32	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ORR X22, XZR, X8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
LDR W8, [X0, #52]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
ADD X9, X0, #8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ORR X21, XZR, X0	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ORR W19, WZR, W2	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ORR X20, XZR, X1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
CMP W8, W2	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.33
CSEL X8, X0, X9, #12	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
LDR X0, [X8]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
LDR X8, [X0]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
LDR X9, [X8, #24]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
ORR X8, XZR, X22	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
BLR X9	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50
LDP X9, X8, [X21, #16]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
CMP X8, X9	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.33
B.EQ 40cd8c <_ZN11qmcplusplus12WaveFunction8evalGradERNS_11ParticleSetEi+0xcc>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
ORR X24, XZR, XZR	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
HINT #0
LDP X20, X19, [SP, #80]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X22, X21, [SP, #64]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X24, X23, [SP, #48]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X29, X30, [SP, #32]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
ADD SP, SP, #96	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
RET	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
BL 409430 <__clang_call_terminate>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50
ORR X19, XZR, X0	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ORR X0, XZR, X23	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
BL 44ee60 <_ZN11qmcplusplus9TimerTypeINS_8CPUClockEE4stopEv>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50
ORR X0, XZR, X19	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
BL 403dd0 <@plt_start@+0x3d0>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50
BL 409430 <__clang_call_terminate>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50

Name	Coverage (%)	Time (s)
▼qmcplusplus::WaveFunction::evalGrad(qmcplusplus::ParticleSet&, int)–	0.01	0
○Loop 136 - WaveFunction.cpp:191-194 - exec	0.01	0

Report Configuration

Vectorization

Details

Workaround

Expensive FP math instructions/calls

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code

Vectorization

Details

Workaround

Expensive FP math instructions/calls

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code