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

	Function: qmcplusplus::WaveFunction::ratio(qmcplusplus::ParticleSet&, int)	Module: exec	Source: WaveFunction.cpp:213-221 [...]	Coverage: 0.04%

/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/QMCWaveFunctions/WaveFunction.cpp: 213 - 221

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

213: {

214:   valT ratio = (iat < nelup ? Det_up->ratio(P, iat) : Det_dn->ratio(P, iat));

215:

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

217:   {

218:     ScopedTimer local_timer(jastrow_timers[i]);

219:     ratio *= Jastrows[i]->ratio(P, iat);

220:   }

221:   return ratio;

/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(); }

0x40ceb0 STP	D9, D8, [SP, #944]!

0x40ceb4 STP	X29, X30, [SP, #16]

0x40ceb8 STR	X23, [SP, #32]

0x40cebc STP	X22, X21, [SP, #48]

0x40cec0 STP	X20, X19, [SP, #64]

0x40cec4 ADD	X29, SP, #16

0x40cec8 LDR	W8, [X0, #52]

0x40cecc ADD	X9, X0, #8

0x40ced0 ORR	X21, XZR, X0

0x40ced4 ORR	W19, WZR, W2

0x40ced8 ORR	X20, XZR, X1

0x40cedc CMP	W8, W2

0x40cee0 CSEL	X8, X0, X9, #12

0x40cee4 LDR	X0, [X8]

0x40cee8 LDR	X8, [X0]

0x40ceec LDR	X8, [X8, #48]

0x40cef0 BLR	X8

0x40cef4 LDP	X9, X8, [X21, #16]

0x40cef8 FMOV	D8, D0

0x40cefc CMP	X8, X9

0x40cf00 B.EQ	40cf58

0x40cf04 ORR	X23, XZR, XZR

(138) 0x40cf08 LDR	X8, [X21, #88]

(138) 0x40cf0c LDR	X22, [X8, X23,LSL #3]

(138) 0x40cf10 ORR	X0, XZR, X22

(138) 0x40cf14 BL	44ed40

(138) 0x40cf18 LDR	X8, [X21, #16]

(138) 0x40cf1c LDR	X0, [X8, X23,LSL #3]

(138) 0x40cf20 LDR	X8, [X0]

(138) 0x40cf24 LDR	X8, [X8, #48]

(138) 0x40cf28 ORR	X1, XZR, X20

(138) 0x40cf2c ORR	W2, WZR, W19

(138) 0x40cf30 BLR	X8

(138) 0x40cf34 ORR	X0, XZR, X22

(138) 0x40cf38 FMOV	D9, D0

(138) 0x40cf3c BL	44ee60

(138) 0x40cf40 LDP	X9, X8, [X21, #16]

(138) 0x40cf44 FMUL	D8, D9, D8

(138) 0x40cf48 ADD	X23, X23, #1

(138) 0x40cf4c SUB	X8, X8, X9

(138) 0x40cf50 CMP	X23, X8,ASR #3

(138) 0x40cf54 B.CC	40cf08

0x40cf58 FMOV	D0, D8

0x40cf5c LDP	X20, X19, [SP, #64]

0x40cf60 LDP	X22, X21, [SP, #48]

0x40cf64 LDR	X23, [SP, #32]

0x40cf68 LDP	X29, X30, [SP, #16]

0x40cf6c LDP	D9, D8, [SP], #80

0x40cf70 RET

0x40cf74 BL	409430

0x40cf78 ORR	X19, XZR, X0

0x40cf7c ORR	X0, XZR, X22

0x40cf80 BL	44ee60

0x40cf84 ORR	X0, XZR, X19

0x40cf88 BL	403dd0

0x40cf8c BL	409430

Coverage (%)	Name	Source Location	Module
►100.00+	.omp_outlined..62	miniqmc.cpp:485	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:213-221
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/QMCWaveFunctions/WaveFunction.cpp:213-221
/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

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 5.17 to 1.83 cycles (2.82x speedup).

Workaround

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

Vectorization

Your function is poorly vectorized. Only 28% of vector register length is used (average across all VPU instructions). By fully vectorizing your function, you can lower the cost of an iteration from 5.17 to 0.91 cycles (5.70x speedup).

Details

25% of VPU instructions are used in vector version (process two or more data elements in vector registers):

0% of VPU addition or subtraction instructions are used in vector version.
0% 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 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 5.17 to 2.50 cycles (2.07x 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 116 bytes. The binary function is storing 72 bytes.

General properties

nb instructions	36
loop length	144
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	4.75	4.75	4.75	4.75	0.75	0.75	0.75	0.75	5.17	4.83	5.00	2.00	2.00
cycles	3.50	3.50	4.75	4.75	4.75	4.75	0.75	0.75	0.75	0.75	5.17	4.83	5.00	2.00	2.00

Cycles executing div or sqrt instructions

Cycles summary

Front-end	4.50
Overall L1	5.17

Vectorization ratios

all	25%
load	100%
store	100%
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 5.17 cycles. At this rate:

23% of peak load performance is reached (22.45 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
29% of peak store performance is reached (13.94 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the function is: 40ceb0

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	Latency	Recip. throughput
STP D9, D8, [SP, #944]!	1	0	0	0.25	0.25	0.25	0.25	0.50	0.50	0	0	0.50	0.50	0	0	0	2	0.50
STP X29, X30, [SP, #16]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STR X23, [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 X22, X21, [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 X20, X19, [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
ADD X29, SP, #16	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 X8, [X8, #48]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
BLR X8	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
FMOV D8, D0	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	2	0.25
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 40cf58 <_ZN11qmcplusplus12WaveFunction5ratioERNS_11ParticleSetEi+0xa8>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
ORR X23, 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
FMOV D0, D8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	2	0.25
LDP X20, X19, [SP, #64]	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, #48]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDR X23, [SP, #32]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
LDP X29, X30, [SP, #16]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP D9, D8, [SP], #80	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.33	0.33	0.33	0	0	6	0.33
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, X22	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:213-221
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/QMCWaveFunctions/WaveFunction.cpp:213-221
/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

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 5.17 to 1.83 cycles (2.82x speedup).

Workaround

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

Vectorization

Details

25% of VPU instructions are used in vector version (process two or more data elements in vector registers):

0% of VPU addition or subtraction instructions are used in vector version.
0% 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 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 5.17 to 2.50 cycles (2.07x 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 116 bytes. The binary function is storing 72 bytes.

General properties

nb instructions	36
loop length	144
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	4.75	4.75	4.75	4.75	0.75	0.75	0.75	0.75	5.17	4.83	5.00	2.00	2.00
cycles	3.50	3.50	4.75	4.75	4.75	4.75	0.75	0.75	0.75	0.75	5.17	4.83	5.00	2.00	2.00

Cycles executing div or sqrt instructions

Cycles summary

Front-end	4.50
Overall L1	5.17

Vectorization ratios

all	25%
load	100%
store	100%
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 5.17 cycles. At this rate:

23% of peak load performance is reached (22.45 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
29% of peak store performance is reached (13.94 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the function is: 40ceb0

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	Latency	Recip. throughput
STP D9, D8, [SP, #944]!	1	0	0	0.25	0.25	0.25	0.25	0.50	0.50	0	0	0.50	0.50	0	0	0	2	0.50
STP X29, X30, [SP, #16]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STR X23, [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 X22, X21, [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 X20, X19, [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
ADD X29, SP, #16	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 X8, [X8, #48]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
BLR X8	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
FMOV D8, D0	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	2	0.25
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 40cf58 <_ZN11qmcplusplus12WaveFunction5ratioERNS_11ParticleSetEi+0xa8>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
ORR X23, 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
FMOV D0, D8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	2	0.25
LDP X20, X19, [SP, #64]	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, #48]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDR X23, [SP, #32]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
LDP X29, X30, [SP, #16]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP D9, D8, [SP], #80	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.33	0.33	0.33	0	0	6	0.33
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, X22	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::ratio(qmcplusplus::ParticleSet&, int)–	0.04	0.02
○Loop 138 - WaveFunction.cpp:216-219 - exec	0.04	0.02

Report Configuration

Code clean check

Workaround

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

Code clean check

Workaround

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