OV - exec - Function __libm_exp

	Function: __libm_exp_z0	Module: exec	Source: :0-0	Coverage (incl. loops): 0.01% \| (excl. loops): 0.01%

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

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

0x52ce40 ENDBR64

0x52ce44 VMOVQ	%XMM0,%RAX

0x52ce49 SHR	$0x30,%RAX

0x52ce4d AND	$0x7fff,%EAX

0x52ce52 LEA	-0x4084(%RAX),%ECX

0x52ce58 CMP	$-0x404,%ECX

0x52ce5e JB	52cefc

0x52ce64 VMOVSD	0x363ec(%RIP),%XMM1

0x52ce6c VMOVSD	0x363dc(%RIP),%XMM2

0x52ce74 VFMADD213SD	{rn-sae},%XMM1,%XMM0,%XMM2

0x52ce7a VADDSD	0x363de(%RIP),%XMM2,%XMM1

0x52ce82 VFMADD231SD	0x363dd(%RIP),%XMM1,%XMM0

0x52ce8b VFMADD231SD	0x363dc(%RIP),%XMM1,%XMM0

0x52ce94 VMOVQ	%XMM2,%RAX

0x52ce99 MOV	%EAX,%ECX

0x52ce9b AND	$0x3ffc0,%ECX

0x52cea1 VMOVD	%ECX,%XMM1

0x52cea5 VPSLLQ	$0x2e,%XMM1,%XMM1

0x52ceaa AND	$0x3f,%EAX

0x52cead SAL	$0x4,%EAX

0x52ceb0 LEA	0x39519(%RIP),%RCX

0x52ceb7 VMOVQ	(%RAX,%RCX,1),%XMM2

0x52cebc VPOR	%XMM2,%XMM1,%XMM1

0x52cec0 VMOVSD	0x363b0(%RIP),%XMM2

0x52cec8 VFMADD213SD	0x363af(%RIP),%XMM0,%XMM2

0x52ced1 VMOVSD	0x363af(%RIP),%XMM3

0x52ced9 VFMADD213SD	0x363ae(%RIP),%XMM0,%XMM3

0x52cee2 VMULSD	%XMM0,%XMM0,%XMM4

0x52cee6 VFMADD231SD	%XMM2,%XMM4,%XMM3

0x52ceeb VADDSD	0x8(%RCX,%RAX,1),%XMM0,%XMM0

0x52cef1 VFMADD231SD	%XMM3,%XMM4,%XMM0

0x52cef6 VFMADD132SD	%XMM1,%XMM1,%XMM0

0x52cefb RET

0x52cefc CMP	$0x3c7f,%EAX

0x52cf01 JA	52cf0c

0x52cf03 VADDSD	0xa135(%RIP),%XMM0,%XMM0

0x52cf0b RET

0x52cf0c JMP	5252a0

0x52cf11 NOPW	%CS:(%RAX,%RAX,1)

0x52cf1b NOPL	(%RAX,%RAX,1)

Coverage (%)	Name	Source Location	Module
►76.80+	qmcplusplus::WaveFunction::eva[...]	stl_vector.h:919	exec
○	qmcplusplus::NonLocalPP<double[...]	NonLocalPP.hpp:135	exec
○	main.extracted.107	NewTimer.h:249	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
►23.20+	main.extracted.107	refwrap.h:338	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so

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 0.01% of application time for run icx_4

Source file and lines
Module	exec

This function has 3 execution paths.

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 11% 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 3.67 to 0.36 cycles (10.06x 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)

FMA

Detected 2.67 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).

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 6.6666666666667 FP arithmetical operations:

3.67: addition or subtraction (2 inside FMA instructions)
3: multiply (2 inside FMA instructions)

The binary function is loading 32 bytes (4 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.21 FP operations per loaded or stored byte.

General properties

nb instructions	18
nb uops	17
loop length	96
used x86 registers	2
used mmx registers	0
used xmm registers	2.33
used ymm registers	0
used zmm registers	0
nb stack references	0
ADD-SUB / MUL ratio	2.00

Front-end

MACRO FUSION NOT POSSIBLE

micro-operation queue	2.89 cycles
front end	2.89 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11
uops	3.67	3.03	1.56	1.56	0.00	2.40	2.90	0.00	0.00	0.00	2.33	1.56
cycles	3.67	3.03	1.56	1.56	0.00	2.40	2.90	0.00	0.00	0.00	2.33	1.56

Execution ports to units layout:

P0 (256 bits): VPU, BRU, ALU, DIV/SQRT
P1 (256 bits): ALU, VPU
P2 (512 bits): load
P3 (512 bits): load
P4 (512 bits): store data
P5 (512 bits): ALU, VPU
P6: ALU, BRU
P7: store address
P8: store address
P9 (512 bits): store data
P10: ALU
P11 (512 bits): load

Cycles executing div or sqrt instructions

Cycles summary

Front-end	2.89
Dispatch	3.67
Overall L1	3.67

Vectorization ratios

INT

all	11%
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	13%

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

INT+FP

all	3%
load	0%
store	NA (no store vectorizable/vectorized instructions)
mul	0%
add-sub	0%
fma	0%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	13%

Vector efficiency ratios

INT

all	11%
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	11%

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

INT+FP

all	11%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	12%
add-sub	12%
fma	12%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	11%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each call to the function takes 3.67 cycles. At this rate:

2% of peak load performance is reached (5.58 out of 192.00 bytes loaded per cycle (GB/s @ 1GHz))

The code analyzed by CQA in that panel excludes loops and represents 0.01% of application time for run icx_4

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 13% 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 6.50 to 0.81 cycles (8.00x 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

Performance is limited by execution of FP multiply or FMA (fused multiply-add) operations (the FP multiply/FMA unit is a bottleneck). By removing all these bottlenecks, you can lower the cost of an iteration from 6.50 to 6.00 cycles (1.08x speedup).

Workaround

Reduce the number of FP multiply/FMA instructions

FMA

Detected 8 FMA (fused multiply-add) operations. Presence of both ADD/SUB and MUL operations.

Workaround

Type of elements and instruction set

11 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 19 FP arithmetical operations:

10: addition or subtraction (8 inside FMA instructions)
9: multiply (8 inside FMA instructions)

The binary function is loading 88 bytes (11 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.22 FP operations per loaded or stored byte.

General properties

nb instructions	33
nb uops	32
loop length	188
used x86 registers	2
used mmx registers	0
used xmm registers	5
used ymm registers	0
used zmm registers	0
nb stack references	0
ADD-SUB / MUL ratio	2.00

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	5.50 cycles
front end	5.50 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11
uops	6.50	6.00	4.00	4.00	0.00	4.10	4.20	0.00	0.00	0.00	4.20	4.00
cycles	6.50	6.00	4.00	4.00	0.00	4.10	4.20	0.00	0.00	0.00	4.20	4.00

Execution ports to units layout:

P0 (256 bits): VPU, BRU, ALU, DIV/SQRT
P1 (256 bits): ALU, VPU
P2 (512 bits): load
P3 (512 bits): load
P4 (512 bits): store data
P5 (512 bits): ALU, VPU
P6: ALU, BRU
P7: store address
P8: store address
P9 (512 bits): store data
P10: ALU
P11 (512 bits): load

Cycles executing div or sqrt instructions

Cycles summary

Front-end	5.50
Dispatch	6.50
Overall L1	6.50

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	40%

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

INT+FP

all	9%
load	0%
store	NA (no store vectorizable/vectorized instructions)
mul	0%
add-sub	0%
fma	0%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	40%

Vector efficiency ratios

INT

all	15%
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	16%

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

INT+FP

all	13%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	12%
add-sub	12%
fma	12%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	16%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each call to the function takes 6.50 cycles. At this rate:

7% of peak load performance is reached (13.54 out of 192.00 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: 52ce40

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	Latency	Recip. throughput	Vectorization
ENDBR64																N/A
VMOVQ %XMM0,%RAX	1	1	0	0	0	0	0	0	0	0	0	0	0	1	1	scal (12.5%)
SHR $0x30,%RAX	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0-2	0.50	N/A
AND $0x7fff,%EAX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1-2	0.20	N/A
LEA -0x4084(%RAX),%ECX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1-2	0.20	N/A
CMP $-0x404,%ECX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	N/A
JB 52cefc <__libm_exp_z0+0xbc>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
VMOVSD 0x363ec(%RIP),%XMM1	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD 0x363dc(%RIP),%XMM2	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VFMADD213SD {rn-sae},%XMM1,%XMM0,%XMM2	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	4	0.50	scal (12.5%)
VADDSD 0x363de(%RIP),%XMM2,%XMM1	1	0	0.50	0.33	0.33	0	0.50	0	0	0	0	0	0.33	3	0.50	scal (12.5%)
VFMADD231SD 0x363dd(%RIP),%XMM1,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0	0	0	0	0.33	4	0.50	scal (12.5%)
VFMADD231SD 0x363dc(%RIP),%XMM1,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0	0	0	0	0.33	4	0.50	scal (12.5%)
VMOVQ %XMM2,%RAX	1	1	0	0	0	0	0	0	0	0	0	0	0	1	1	scal (12.5%)
MOV %EAX,%ECX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
AND $0x3ffc0,%ECX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1-2	0.20	N/A
VMOVD %ECX,%XMM1	1	0	0	0	0	0	1	0	0	0	0	0	0	3	1	scal (6.3%)
VPSLLQ $0x2e,%XMM1,%XMM1	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	2-4	0.50	vect (25.0%)
AND $0x3f,%EAX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1-2	0.20	N/A
SAL $0x4,%EAX	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0-2	0.50	N/A
LEA 0x39519(%RIP),%RCX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
VMOVQ (%RAX,%RCX,1),%XMM2	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VPOR %XMM2,%XMM1,%XMM1	1	0.33	0.33	0	0	0	0.33	0	0	0	0	0	0	1	0.33	vect (25.0%)
VMOVSD 0x363b0(%RIP),%XMM2	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VFMADD213SD 0x363af(%RIP),%XMM0,%XMM2	1	0.50	0.50	0.33	0.33	0	0	0	0	0	0	0	0.33	4	0.50	scal (12.5%)
VMOVSD 0x363af(%RIP),%XMM3	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VFMADD213SD 0x363ae(%RIP),%XMM0,%XMM3	1	0.50	0.50	0.33	0.33	0	0	0	0	0	0	0	0.33	4	0.50	scal (12.5%)
VMULSD %XMM0,%XMM0,%XMM4	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	4	0.50	scal (12.5%)
VFMADD231SD %XMM2,%XMM4,%XMM3	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	4	0.50	scal (12.5%)
VADDSD 0x8(%RCX,%RAX,1),%XMM0,%XMM0	1	0	0.50	0.33	0.33	0	0.50	0	0	0	0	0	0.33	3	0.50	scal (12.5%)
VFMADD231SD %XMM3,%XMM4,%XMM0	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	4	0.50	scal (12.5%)
VFMADD132SD %XMM1,%XMM1,%XMM0	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	4	0.50	scal (12.5%)
RET	1	0.50	0	0.33	0.33	0	0	0.50	0	0	0	0	0.33	0	2.13	N/A

The code analyzed by CQA in that panel excludes loops and represents 0.01% of application time for run icx_4

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 2.50 to 1.50 cycles (1.67x speedup).

Workaround

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

Vectorization

Your function is not vectorized. Only 10% 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 2.50 to 0.16 cycles (16.00x speedup).

Details

All 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)

Execution units bottlenecks

Workaround

Reduce the number of FP multiply/FMA instructions

No data for this section

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 is composed of 1 FP arithmetical operations:

1: addition or subtraction

The binary function is loading 8 bytes (1 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.12 FP operations per loaded or stored byte.

General properties

nb instructions	11
nb uops	10
loop length	52
used x86 registers	2
used mmx registers	0
used xmm registers	1
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	1.67 cycles
front end	1.67 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11
uops	2.50	1.70	0.67	0.67	0.00	1.70	2.50	0.00	0.00	0.00	1.60	0.67
cycles	2.50	1.70	0.67	0.67	0.00	1.70	2.50	0.00	0.00	0.00	1.60	0.67

Execution ports to units layout:

P0 (256 bits): VPU, BRU, ALU, DIV/SQRT
P1 (256 bits): ALU, VPU
P2 (512 bits): load
P3 (512 bits): load
P4 (512 bits): store data
P5 (512 bits): ALU, VPU
P6: ALU, BRU
P7: store address
P8: store address
P9 (512 bits): store data
P10: ALU
P11 (512 bits): load

Cycles executing div or sqrt instructions

Cycles summary

Front-end	1.67
Dispatch	2.50
Overall L1	2.50

Vectorization ratios

INT

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)
other	0%

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	NA (no other vectorizable/vectorized instructions)

INT+FP

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%

Vector efficiency ratios

INT

all	9%
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)
other	9%

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	NA (no other vectorizable/vectorized instructions)

INT+FP

all	10%
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	9%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each call to the function takes 2.50 cycles. At this rate:

1% of peak load performance is reached (3.20 out of 192.00 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: 52ce40

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	Latency	Recip. throughput	Vectorization
ENDBR64																N/A
VMOVQ %XMM0,%RAX	1	1	0	0	0	0	0	0	0	0	0	0	0	1	1	scal (12.5%)
SHR $0x30,%RAX	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0-2	0.50	N/A
AND $0x7fff,%EAX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1-2	0.20	N/A
LEA -0x4084(%RAX),%ECX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1-2	0.20	N/A
CMP $-0x404,%ECX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (6.3%)
JB 52cefc <__libm_exp_z0+0xbc>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
CMP $0x3c7f,%EAX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	N/A
JA 52cf0c <__libm_exp_z0+0xcc>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
VADDSD 0xa135(%RIP),%XMM0,%XMM0	1	0	0.50	0.33	0.33	0	0.50	0	0	0	0	0	0.33	3	0.50	scal (12.5%)
RET	1	0.50	0	0.33	0.33	0	0	0.50	0	0	0	0	0.33	0	2.13	N/A

The code analyzed by CQA in that panel excludes loops and represents 0.01% of application time for run icx_4

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 2.00 to 1.50 cycles (1.33x speedup).

Workaround

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

Vectorization

Your function is not vectorized. Only 9% 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 2.00 to 0.12 cycles (16.00x 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

Type of elements and instruction set

No instructions are processing arithmetic or math operations on FP elements. This function is probably writing/copying data or processing integer elements.

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 does not load or store any data.

General properties

nb instructions	10
nb uops	9
loop length	48
used x86 registers	2
used mmx registers	0
used xmm registers	1
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	1.50 cycles
front end	1.50 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11
uops	2.00	1.40	0.00	0.00	0.00	1.40	2.00	0.00	0.00	0.00	1.20	0.00
cycles	2.00	1.40	0.00	0.00	0.00	1.40	2.00	0.00	0.00	0.00	1.20	0.00

Execution ports to units layout:

P0 (256 bits): VPU, BRU, ALU, DIV/SQRT
P1 (256 bits): ALU, VPU
P2 (512 bits): load
P3 (512 bits): load
P4 (512 bits): store data
P5 (512 bits): ALU, VPU
P6: ALU, BRU
P7: store address
P8: store address
P9 (512 bits): store data
P10: ALU
P11 (512 bits): load

Cycles executing div or sqrt instructions

Cycles summary

Front-end	1.50
Dispatch	2.00
Overall L1	2.00

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	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%

Vector efficiency ratios

all	9%
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	9%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each call to the function takes 2.00 cycles.

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the function is: 52ce40

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	Latency	Recip. throughput	Vectorization
ENDBR64																N/A
VMOVQ %XMM0,%RAX	1	1	0	0	0	0	0	0	0	0	0	0	0	1	1	scal (12.5%)
SHR $0x30,%RAX	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0-2	0.50	N/A
AND $0x7fff,%EAX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1-2	0.20	N/A
LEA -0x4084(%RAX),%ECX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1-2	0.20	N/A
CMP $-0x404,%ECX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (6.3%)
JB 52cefc <__libm_exp_z0+0xbc>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
CMP $0x3c7f,%EAX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	N/A
JA 52cf0c <__libm_exp_z0+0xcc>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
JMP 5252a0 <__libm_exp_e7>	1	0	0	0	0	0	0	0	0	0	0	0	0	0	2.08	N/A

Name	Coverage (%)	Time (s)
○__libm_exp_z0	0.01	0.01

Report Configuration

Vectorization

Details

Workaround

FMA

Workaround

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

FMA

Workaround

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

Workaround

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

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