OV - exec - Function __libm_round

	Function: __libm_round_y8	Module: exec	Source: :0-0	Coverage (incl. loops): 2.71% \| (excl. loops): 2.71%

  ***  This Panel is Intentionally Left Blank.  ***

It is due to a lack of debug symbols in the given object
(loop or function).

0x50de80 ENDBR64

0x50de84 PEXTRW	$0x3,%XMM0,%ECX

0x50de89 MOV	$0x7ff0,%EAX

0x50de8e MOVSD	%XMM0,%XMM1

0x50de92 MOVSD	%XMM0,%XMM2

0x50de96 AND	%ECX,%EAX

0x50de98 CMP	$0x7ff0,%EAX

0x50de9d MOVQ	0x3a3fb(%RIP),%XMM3

0x50dea5 PAND	%XMM3,%XMM2

0x50dea9 MOVSD	%XMM2,%XMM3

0x50dead ROUNDSD	$0,%XMM0,%XMM0

0x50deb3 JGE	50dede

0x50deb5 SUBSD	%XMM0,%XMM1

0x50deb9 PXOR	%XMM2,%XMM1

0x50debd MOVQ	0x3a3eb(%RIP),%XMM4

0x50dec5 POR	%XMM4,%XMM2

0x50dec9 CMPSD	$0,0x3a3ee(%RIP),%XMM1

0x50ded2 PAND	%XMM2,%XMM1

0x50ded6 ADDSD	%XMM1,%XMM0

0x50deda POR	%XMM3,%XMM0

0x50dede RET

0x50dedf NOP

Coverage (%)	Name	Source Location	Module
►12.41+	qmcplusplus::SoaDistanceTableA[...]	ParticleBConds3DSoa.h:291	exec
○	_ZN11qmcplusplus11ParticleSet6[...]	ParticleSet.cpp:242	exec
○	qmcplusplus::NonLocalPP<double[...]	NonLocalPP.hpp:134	exec
○	main.extracted.107	NewTimer.h:249	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►12.37+	qmcplusplus::SoaDistanceTableA[...]	ParticleBConds3DSoa.h:290	exec
○	_ZN11qmcplusplus11ParticleSet6[...]	ParticleSet.cpp:242	exec
○	qmcplusplus::NonLocalPP<double[...]	NonLocalPP.hpp:134	exec
○	main.extracted.107	NewTimer.h:249	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►8.46+	qmcplusplus::SoaDistanceTableA[...]	ParticleBConds3DSoa.h:289	exec
○	_ZN11qmcplusplus11ParticleSet6[...]	ParticleSet.cpp:242	exec
○	qmcplusplus::NonLocalPP<double[...]	NonLocalPP.hpp:134	exec
○	main.extracted.107	NewTimer.h:249	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►5.28+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:247	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:275	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►5.09+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:247	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:283	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►3.55+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:247	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:275	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►3.46+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:247	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:283	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►3.25+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:248	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:275	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►3.24+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:248	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:283	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►2.97+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:247	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:275	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►2.93+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:247	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:283	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►2.85+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:247	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:275	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►2.74+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:247	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:283	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►2.51+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:248	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:283	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►2.45+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:248	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:275	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►2.20+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:248	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:283	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►2.15+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:248	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:275	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►1.74+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:248	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:283	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►1.68+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:248	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:275	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►1.50+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:246	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:275	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►1.45+	void qmcplusplus::DTD_BConds<d[...]	ParticleBConds3DSoa.h:246	exec
○	qmcplusplus::SoaDistanceTableA[...]	SoaDistanceTableAAOMPTarget.h:283	exec
○	_ZN11qmcplusplus11ParticleSet2[...]	ParticleSet.cpp:343	exec
○	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		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 /

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

The code analyzed by CQA in that panel excludes loops and represents 2.71% of application time for run icx_3

Source file and lines
Module	exec

The structure of this function is probably <if then [else] end>.

The presence of multiple execution paths is typically the main/first bottleneck.
Try to simplify control inside function: ideally, try to remove all conditional expressions, for example by (if applicable):

hoisting them (moving them outside the function)
turning them into conditional moves, MIN or MAX

gain
potential
hint
expert

Vectorization

Your function is probably not vectorized. Only 14% of vector register length is used (average across all SSE/AVX instructions). By vectorizing your function, you can lower the cost of an iteration from 2.44 to 0.28 cycles (8.67x speedup).

Details

Store and arithmetical 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 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)

No data for this section

Complex instructions

Detected COMPLEX INSTRUCTIONS.

Details

These instructions generate more than one micro-operation and only one of them can be decoded during a cycle and the extra micro-operations increase pressure on execution units.

PEXTRW: 0 occurrences

Type of elements and instruction set

2 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 function (in the source code) and the binary function

The binary function is composed of 1 FP arithmetical operations:

1: addition or subtraction

The binary function is loading 16 bytes (2 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.06 FP operations per loaded or stored byte.

General properties

nb instructions	17
nb uops	17
loop length	74.50
used x86 registers	2
used mmx registers	0
used xmm registers	4.50
used ymm registers	0
used zmm registers	0
nb stack references	0

Front-end

MACRO FUSION NOT POSSIBLE

micro-operation queue	2.13 cycles
front end	2.13 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	0.83	0.83	0.83	0.83	0.83	0.83	0.50	0.50	0.50	0.50	2.38	2.38	2.38	2.38	0.50	0.50
cycles	0.83	0.83	0.83	0.83	0.83	0.83	0.50	0.50	0.50	0.50	2.38	2.38	2.38	2.38	0.50	0.50

Execution ports to units layout:

P0: ALU
P1: ALU
P2: ALU
P3: ALU, BRU
P4: ALU, BRU
P5: ALU, BRU
P6 (512 bits): store address, load
P7 (512 bits): store address, load
P8 (512 bits): store address, load
P9 (512 bits): store address, load
P10 (512 bits): VPU, DIV/SQRT
P11 (512 bits): VPU, DIV/SQRT
P12 (512 bits): VPU
P13 (512 bits): VPU
P14 (512 bits): FP store data
P15 (512 bits): FP store data

Cycles executing div or sqrt instructions

Cycles summary

Front-end	2.13
Dispatch	2.38
Overall L1	2.44

Vectorization ratios

INT

all	47%
load	0%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	66%

all	0%
load	0%
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%

INT+FP

all	23%
load	0%
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	31%

Vector efficiency ratios

INT

all	16%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	17%

all	12%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	12%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	12%

INT+FP

all	14%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	12%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	15%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each call to the function takes 2.44 cycles. At this rate:

2% of peak load performance is reached (6.15 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))

The code analyzed by CQA in that panel excludes loops and represents 2.71% of application time for run icx_3

Source file and lines
Module	exec

hoisting them (moving them outside the function)
turning them into conditional moves, MIN or MAX

gain
potential
hint
expert

Vectorization

Your function is probably not vectorized. Only 16% of vector register length is used (average across all SSE/AVX instructions). By vectorizing your function, you can lower the cost of an iteration from 3.25 to 0.44 cycles (7.43x speedup).

Details

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)

Execution units bottlenecks

Performance is limited by execution of FP add operations (the FP add unit is a bottleneck). By removing all these bottlenecks, you can lower the cost of an iteration from 3.25 to 2.63 cycles (1.24x speedup).

Workaround

Reduce the number of FP add instructions

No data for this section

Complex instructions

Detected COMPLEX INSTRUCTIONS.

Details

These instructions generate more than one micro-operation and only one of them can be decoded during a cycle and the extra micro-operations increase pressure on execution units.

PEXTRW: 1 occurrences►

Type of elements and instruction set

4 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 function (in the source code) and the binary function

The binary function is composed of 2 FP arithmetical operations:

2: addition or subtraction

The binary function is loading 24 bytes (3 double precision FP elements).

Arithmetic intensity

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

General properties

nb instructions	21
nb uops	21
loop length	95
used x86 registers	2
used mmx registers	0
used xmm registers	5
used ymm registers	0
used zmm registers	0
nb stack references	0

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	2.63 cycles
front end	2.63 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	0.83	0.83	0.83	0.83	0.83	0.83	0.75	0.75	0.75	0.75	3.25	3.25	3.25	3.25	0.50	0.50
cycles	0.83	0.83	0.83	0.83	0.83	0.83	0.75	0.75	0.75	0.75	3.25	3.25	3.25	3.25	0.50	0.50

Execution ports to units layout:

P0: ALU
P1: ALU
P2: ALU
P3: ALU, BRU
P4: ALU, BRU
P5: ALU, BRU
P6 (512 bits): store address, load
P7 (512 bits): store address, load
P8 (512 bits): store address, load
P9 (512 bits): store address, load
P10 (512 bits): VPU, DIV/SQRT
P11 (512 bits): VPU, DIV/SQRT
P12 (512 bits): VPU
P13 (512 bits): VPU
P14 (512 bits): FP store data
P15 (512 bits): FP store data

Cycles executing div or sqrt instructions

Cycles summary

Front-end	2.63
Dispatch	3.25
Overall L1	3.25

Vectorization ratios

INT

all	62%
load	0%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	83%

all	0%
load	0%
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%

INT+FP

all	33%
load	0%
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	45%

Vector efficiency ratios

INT

all	19%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	21%

all	12%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	12%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	12%

INT+FP

all	16%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	12%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	17%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each call to the function takes 3.25 cycles. At this rate:

2% of peak load performance is reached (7.38 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the function is: 50de80

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	Latency	Recip. throughput	Vectorization
ENDBR64																				N/A
PEXTRW $0x3,%XMM0,%ECX	2	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	7	0.50	scal (3.1%)
MOV $0x7ff0,%EAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOVSD %XMM0,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	scal (12.5%)
MOVSD %XMM0,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	scal (12.5%)
AND %ECX,%EAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
CMP $0x7ff0,%EAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOVQ 0x3a3fb(%RIP),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	4	0.50	scal (12.5%)
PAND %XMM3,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
MOVSD %XMM2,%XMM3	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	scal (12.5%)
ROUNDSD $0,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	scal (12.5%)
JGE 50dede <__libm_round_y8+0x5e>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.50-1	N/A
SUBSD %XMM0,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	scal (12.5%)
PXOR %XMM2,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
MOVQ 0x3a3eb(%RIP),%XMM4	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	4	0.50	scal (12.5%)
POR %XMM4,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
CMPSD $0,0x3a3ee(%RIP),%XMM1	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0.50	0.50	0	0	2	0.50	scal (12.5%)
PAND %XMM2,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
ADDSD %XMM1,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	scal (12.5%)
POR %XMM3,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
RET	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A

The code analyzed by CQA in that panel excludes loops and represents 2.71% of application time for run icx_3

Source file and lines
Module	exec

hoisting them (moving them outside the function)
turning them into conditional moves, MIN or MAX

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 1.63 to 1.50 cycles (1.08x speedup).

Workaround

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

Vectorization

Your function is probably not vectorized. Only 12% of vector register length is used (average across all SSE/AVX instructions). By vectorizing your function, you can lower the cost of an iteration from 1.63 to 0.12 cycles (13.00x speedup).

Details

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)

Execution units bottlenecks

Found no such bottlenecks but see expert reports for more complex bottlenecks.

No data for this section

Complex instructions

Detected COMPLEX INSTRUCTIONS.

Details

These instructions generate more than one micro-operation and only one of them can be decoded during a cycle and the extra micro-operations increase pressure on execution units.

PEXTRW: 1 occurrences►

Type of elements and instruction set

1 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 function (in the source code) and the binary function

The binary function does not contain any FP arithmetical operations. The binary function is loading 8 bytes (1 double precision FP elements).

General properties

nb instructions	13
nb uops	13
loop length	54
used x86 registers	2
used mmx registers	0
used xmm registers	4
used ymm registers	0
used zmm registers	0
nb stack references	0

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	1.63 cycles
front end	1.63 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	0.83	0.83	0.83	0.83	0.83	0.83	0.25	0.25	0.25	0.25	1.50	1.50	1.50	1.50	0.50	0.50
cycles	0.83	0.83	0.83	0.83	0.83	0.83	0.25	0.25	0.25	0.25	1.50	1.50	1.50	1.50	0.50	0.50

Execution ports to units layout:

P0: ALU
P1: ALU
P2: ALU
P3: ALU, BRU
P4: ALU, BRU
P5: ALU, BRU
P6 (512 bits): store address, load
P7 (512 bits): store address, load
P8 (512 bits): store address, load
P9 (512 bits): store address, load
P10 (512 bits): VPU, DIV/SQRT
P11 (512 bits): VPU, DIV/SQRT
P12 (512 bits): VPU
P13 (512 bits): VPU
P14 (512 bits): FP store data
P15 (512 bits): FP store data

Cycles executing div or sqrt instructions

Cycles summary

Front-end	1.63
Dispatch	1.50
Overall L1	1.63

Vectorization ratios

INT

all	33%
load	0%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	50%

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

INT+FP

all	14%
load	0%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	16%

Vector efficiency ratios

INT

all	13%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
other	14%

all	12%
load	NA (no load vectorizable/vectorized instructions)
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	12%

INT+FP

all	12%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	13%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each call to the function takes 1.63 cycles. At this rate:

1% of peak load performance is reached (4.92 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Performance is limited by instruction throughput (loading/decoding program instructions to execution core) (front-end is a bottleneck). By removing all these bottlenecks, you can lower the cost of an iteration from 1.63 to 1.50 cycles (1.08x speedup).

ASM code

In the binary file, the address of the function is: 50de80

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	Latency	Recip. throughput	Vectorization
ENDBR64																				N/A
PEXTRW $0x3,%XMM0,%ECX	2	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	7	0.50	scal (3.1%)
MOV $0x7ff0,%EAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOVSD %XMM0,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	scal (12.5%)
MOVSD %XMM0,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	scal (12.5%)
AND %ECX,%EAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
CMP $0x7ff0,%EAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOVQ 0x3a3fb(%RIP),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	4	0.50	scal (12.5%)
PAND %XMM3,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
MOVSD %XMM2,%XMM3	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	scal (12.5%)
ROUNDSD $0,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	scal (12.5%)
JGE 50dede <__libm_round_y8+0x5e>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.50-1	N/A
RET	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A

Name	Coverage (%)	Time (s)
○__libm_round_y8	2.71	4.52

Report Configuration

Vectorization

Details

Workaround

Complex instructions

Details

Type of elements and instruction set

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

Arithmetic intensity

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Vectorization

Details

Workaround

Execution units bottlenecks

Workaround

Complex instructions

Details

Type of elements and instruction set

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

Arithmetic intensity

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

Execution units bottlenecks

Complex instructions

Details

Type of elements and instruction set

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code