OV - exec - Loop 221

	Loop Id: 221	Module: exec	Source: field_summary_kernel.f90:58-72	Coverage: 0.14%

0x232fe0 MOVSXD	%R14D,%RAX

0x232fe3 VXORPD	%XMM2,%XMM2,%XMM2

0x232fe7 XOR	%EDI,%EDI

0x232fe9 ADD	%R13,%RAX

0x232fec NOPL	(%RAX)

  (219) 0x232ff0 VMOVSD	(%RBP,%RDI,8),%XMM3

  (219) 0x232ff6 VMOVSD	(%R8,%RDI,8),%XMM4

  (219) 0x232ffc INC	%RDI

  (219) 0x232fff VMULSD	%XMM3,%XMM3,%XMM3

  (219) 0x233003 VMULSD	%XMM4,%XMM4,%XMM4

  (219) 0x233007 VADDSD	%XMM4,%XMM3,%XMM3

  (219) 0x23300b VMULSD	%XMM0,%XMM3,%XMM3

  (219) 0x23300f VADDSD	%XMM3,%XMM2,%XMM2

  (219) 0x233013 CMP	%RDI,%RAX

  (219) 0x233016 JNE	232ff0

0x233018 XOR	%EDI,%EDI

0x23301a NOPW	(%RAX,%RAX,1)

  (220) 0x233020 VMOVSD	(%RCX,%RDI,8),%XMM3

  (220) 0x233025 VMOVSD	(%RBX,%RDI,8),%XMM4

  (220) 0x23302a INC	%RDI

  (220) 0x23302d VMULSD	%XMM3,%XMM3,%XMM3

  (220) 0x233031 VMULSD	%XMM4,%XMM4,%XMM4

  (220) 0x233035 VADDSD	%XMM4,%XMM3,%XMM3

  (220) 0x233039 VMULSD	%XMM0,%XMM3,%XMM3

  (220) 0x23303d VADDSD	%XMM3,%XMM2,%XMM2

  (220) 0x233041 CMP	%RDI,%RAX

  (220) 0x233044 JNE	233020

0x233046 MOV	0x30(%RSP),%RDI

0x23304b LEA	(%R12,%RSI,1),%RAX

0x23304f DEC	%R13

0x233052 INC	%R14D

0x233055 ADD	$0x8,%RBP

0x233059 ADD	$0x8,%R8

0x23305d ADD	$0x8,%RBX

0x233061 ADD	$0x8,%RCX

0x233065 SUB	%R9,%RAX

0x233068 VMOVSD	(%RDI,%RAX,8),%XMM3

0x23306d VMULSD	(%R10,%RAX,8),%XMM3,%XMM4

0x233073 VMULSD	(%R11,%RAX,8),%XMM4,%XMM5

0x233079 VADDSD	%XMM3,%XMM10,%XMM10

0x23307d VMULSD	(%R15,%RAX,8),%XMM3,%XMM3

0x233083 LEA	0x1(%RSI),%RAX

0x233087 VADDSD	%XMM4,%XMM9,%XMM9

0x23308b VMULSD	%XMM1,%XMM4,%XMM4

0x23308f VMULSD	%XMM4,%XMM2,%XMM2

0x233093 VADDSD	%XMM5,%XMM7,%XMM7

0x233097 VADDSD	%XMM2,%XMM6,%XMM6

0x23309b VADDSD	%XMM3,%XMM8,%XMM8

0x23309f CMP	%RDX,%RSI

0x2330a2 MOV	%RAX,%RSI

0x2330a5 JNE	232fe0

/home/eoseret/qaas_runs_ZEN5/174-049-0948/intel/CloverLeaf1.3-FC/build/CloverLeaf1.3-FC/CloverLeaf_ref/kernels/field_summary_kernel.f90: 58 - 72

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

58:       DO j=x_min,x_max

59:         vsqrd=0.0

60:         DO kv=k,k+1

61:           DO jv=j,j+1

62:             vsqrd=vsqrd+0.25*(xvel0(jv,kv)**2+yvel0(jv,kv)**2)

63:           ENDDO

64:         ENDDO

65:         cell_vol=volume(j,k)

66:         cell_mass=cell_vol*density0(j,k)

67:         vol=vol+cell_vol

68:         mass=mass+cell_mass

69:         ie=ie+cell_mass*energy0(j,k)

70:         ke=ke+cell_mass*0.5*vsqrd

71:         press=press+cell_vol*pressure(j,k)

72:       ENDDO

Coverage (%)	Name	Source Location	Module

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	1.45
CQA speedup if FP arith vectorized	2.74
CQA speedup if fully vectorized	10.55
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.45
Bottlenecks	micro-operation queue,
Function	__nv_field_summary_kernel_module_field_summary_kernel__PARALLEL_F2L54_1
Source	field_summary_kernel.f90:58-58,field_summary_kernel.f90:63-72
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	3.63
CQA cycles if no scalar integer	2.50
CQA cycles if FP arith vectorized	1.32
CQA cycles if fully vectorized	0.34
Front-end cycles	3.63
P0 cycles	2.17
P1 cycles	2.17
P2 cycles	2.17
P3 cycles	2.17
P4 cycles	2.17
P5 cycles	2.17
P6 cycles	1.25
P7 cycles	1.25
P8 cycles	1.25
P9 cycles	1.25
P10 cycles	2.50
P11 cycles	2.50
P12 cycles	2.50
P13 cycles	2.50
P14 cycles	0.00
P15 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	2
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	31.00
Nb uops	29.00
Nb loads	5.00
Nb stores	0.00
Nb stack references	1.00
FLOP/cycle	2.76
Nb FLOP add-sub	5.00
Nb FLOP mul	5.00
Nb FLOP fma	0.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	11.03
Bytes prefetched	0.00
Bytes loaded	40.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	4.00
Stride n	0.00
Stride unknown	0.00
Stride indirect	0.00
Vectorization ratio all	5.56
Vectorization ratio load	0.00
Vectorization ratio store	NA
Vectorization ratio mul	0.00
Vectorization ratio add_sub	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	100.00
Vector-efficiency ratio all	12.85
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	12.50
Vector-efficiency ratio add_sub	11.93
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	25.00

Metric	Value
CQA speedup if no scalar integer	1.45
CQA speedup if FP arith vectorized	2.74
CQA speedup if fully vectorized	10.55
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.45
Bottlenecks	micro-operation queue,
Function	__nv_field_summary_kernel_module_field_summary_kernel__PARALLEL_F2L54_1
Source	field_summary_kernel.f90:58-58,field_summary_kernel.f90:63-72
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	3.63
CQA cycles if no scalar integer	2.50
CQA cycles if FP arith vectorized	1.32
CQA cycles if fully vectorized	0.34
Front-end cycles	3.63
P0 cycles	2.17
P1 cycles	2.17
P2 cycles	2.17
P3 cycles	2.17
P4 cycles	2.17
P5 cycles	2.17
P6 cycles	1.25
P7 cycles	1.25
P8 cycles	1.25
P9 cycles	1.25
P10 cycles	2.50
P11 cycles	2.50
P12 cycles	2.50
P13 cycles	2.50
P14 cycles	0.00
P15 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	2
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	31.00
Nb uops	29.00
Nb loads	5.00
Nb stores	0.00
Nb stack references	1.00
FLOP/cycle	2.76
Nb FLOP add-sub	5.00
Nb FLOP mul	5.00
Nb FLOP fma	0.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	11.03
Bytes prefetched	0.00
Bytes loaded	40.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	4.00
Stride n	0.00
Stride unknown	0.00
Stride indirect	0.00
Vectorization ratio all	5.56
Vectorization ratio load	0.00
Vectorization ratio store	NA
Vectorization ratio mul	0.00
Vectorization ratio add_sub	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	100.00
Vector-efficiency ratio all	12.85
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	12.50
Vector-efficiency ratio add_sub	11.93
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	25.00

Path /

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

Function	__nv_field_summary_kernel_module_field_summary_kernel__PARALLEL_F2L54_1
Source file and lines	field_summary_kernel.f90:58-72
Module	exec

The loop is defined in /home/eoseret/qaas_runs_ZEN5/174-049-0948/intel/CloverLeaf1.3-FC/build/CloverLeaf1.3-FC/CloverLeaf_ref/kernels/field_summary_kernel.f90:58-72.

Analyzed code is defined in /home/eoseret/qaas_runs_ZEN5/174-049-0948/intel/CloverLeaf1.3-FC/build/CloverLeaf1.3-FC/CloverLeaf_ref/kernels/field_summary_kernel.f90:58,63-72.

Warnings:
Non-innermost loop: analyzing only self part (ignoring child loops).

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 3.63 to 2.50 cycles (1.45x speedup).

Workaround

Try to reorganize arrays of structures to structures of arrays
Consider to permute loops (see vectorization gain report)
To reference allocatable arrays, use "allocatable" instead of "pointer" pointers or qualify them with the "contiguous" attribute (Fortran 2008)
For structures, limit to one indirection. For example, use a_b%c instead of a%b%c with a_b set to a%b before this loop

Vectorization

Your loop is probably not vectorized. Only 12% of vector register length is used (average across all SSE/AVX instructions). By vectorizing your loop, you can lower the cost of an iteration from 3.63 to 0.34 cycles (10.55x 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 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: Fortran storage order is column-major: do i do j a(i,j) = b(i,j) (slow, non stride 1) => do i do j a(j,i) = b(i,j) (fast, stride 1)
- If your loop streams arrays of structures (AoS), try to use structures of arrays instead (SoA): do i a(i)%x = b(i)%x (slow, non stride 1) => do i a%x(i) = b%x(i) (fast, stride 1)

Execution units bottlenecks

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

FMA

Presence of both ADD/SUB and MUL operations.

Workaround

Pass to your compiler a micro-architecture specialization option:
- Please look into your compiler manual for march=native or equivalent
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

10 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 loop (in the source code) and the binary loop

The binary loop is composed of 10 FP arithmetical operations:

5: addition or subtraction
5: multiply

The binary loop is loading 40 bytes (5 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.25 FP operations per loaded or stored byte.

General properties

nb instructions	31
nb uops	29
loop length	125
used x86 registers	16
used mmx registers	0
used xmm registers	10
used ymm registers	0
used zmm registers	0
nb stack references	1
ADD-SUB / MUL ratio	1.00

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	3.63 cycles
front end	3.63 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	2.17	2.17	2.17	2.17	2.17	2.17	1.25	1.25	1.25	1.25	2.50	2.50	2.50	2.50	0.00	0.00
cycles	2.17	2.17	2.17	2.17	2.17	2.17	1.25	1.25	1.25	1.25	2.50	2.50	2.50	2.50	0.00	0.00

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	NA
Longest recurrence chain latency (RecMII)	2.00

Cycles summary

Front-end	3.63
Dispatch	2.50
Data deps.	2.00
Overall L1	3.63

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	0%
fma	NA (no fma vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

all	8%
load	0%
store	NA (no store vectorizable/vectorized instructions)
mul	0%
add-sub	0%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	100%

INT+FP

all	5%
load	0%
store	NA (no store vectorizable/vectorized instructions)
mul	0%
add-sub	0%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	100%

Vector efficiency ratios

INT

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

all	13%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	12%
add-sub	12%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	25%

INT+FP

all	12%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	12%
add-sub	11%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	25%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 3.63 cycles. At this rate:

4% of peak load performance is reached (11.03 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 3.63 to 2.50 cycles (1.45x speedup).

ASM code

In the binary file, the address of the loop is: 232fe0

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	Latency	Recip. throughput	Vectorization
MOVSXD %R14D,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	N/A
VXORPD %XMM2,%XMM2,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
XOR %EDI,%EDI	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.25	N/A
ADD %R13,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	N/A
NOPL (%RAX)	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.09	N/A
XOR %EDI,%EDI	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.25	N/A
NOPW (%RAX,%RAX,1)	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.09	N/A
MOV 0x30(%RSP),%RDI	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	3	0.25	N/A
LEA (%R12,%RSI,1),%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.25	N/A
DEC %R13	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
INC %R14D	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	scal (6.3%)
ADD $0x8,%RBP	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
ADD $0x8,%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
ADD $0x8,%RBX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
ADD $0x8,%RCX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
SUB %R9,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSD (%RDI,%RAX,8),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	1	0.50	scal (12.5%)
VMULSD (%R10,%RAX,8),%XMM3,%XMM4	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	0	0	3	0.50	scal (12.5%)
VMULSD (%R11,%RAX,8),%XMM4,%XMM5	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	0	0	3	0.50	scal (12.5%)
VADDSD %XMM3,%XMM10,%XMM10	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	scal (12.5%)
VMULSD (%R15,%RAX,8),%XMM3,%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	0	0	3	0.50	scal (12.5%)
LEA 0x1(%RSI),%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.25	N/A
VADDSD %XMM4,%XMM9,%XMM9	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	scal (12.5%)
VMULSD %XMM1,%XMM4,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	scal (12.5%)
VMULSD %XMM4,%XMM2,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	scal (12.5%)
VADDSD %XMM5,%XMM7,%XMM7	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	scal (12.5%)
VADDSD %XMM2,%XMM6,%XMM6	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	scal (12.5%)
VADDSD %XMM3,%XMM8,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	scal (12.5%)
CMP %RDX,%RSI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV %RAX,%RSI	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
JNE 232fe0 <__nv_field_summary_kernel_module_field_summary_kernel__PARALLEL_F2L54_1_+0x2c0>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	1	0.50-1	N/A

Function	__nv_field_summary_kernel_module_field_summary_kernel__PARALLEL_F2L54_1
Source file and lines	field_summary_kernel.f90:58-72
Module	exec

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 3.63 to 2.50 cycles (1.45x speedup).

Workaround

Try to reorganize arrays of structures to structures of arrays
Consider to permute loops (see vectorization gain report)
To reference allocatable arrays, use "allocatable" instead of "pointer" pointers or qualify them with the "contiguous" attribute (Fortran 2008)
For structures, limit to one indirection. For example, use a_b%c instead of a%b%c with a_b set to a%b before this loop

Vectorization

Details

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: Fortran storage order is column-major: do i do j a(i,j) = b(i,j) (slow, non stride 1) => do i do j a(j,i) = b(i,j) (fast, stride 1)
- If your loop streams arrays of structures (AoS), try to use structures of arrays instead (SoA): do i a(i)%x = b(i)%x (slow, non stride 1) => do i a%x(i) = b%x(i) (fast, stride 1)

Execution units bottlenecks

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

FMA

Presence of both ADD/SUB and MUL operations.

Workaround

Pass to your compiler a micro-architecture specialization option:
- Please look into your compiler manual for march=native or equivalent
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

10 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 loop (in the source code) and the binary loop

The binary loop is composed of 10 FP arithmetical operations:

5: addition or subtraction
5: multiply

The binary loop is loading 40 bytes (5 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.25 FP operations per loaded or stored byte.

General properties

nb instructions	31
nb uops	29
loop length	125
used x86 registers	16
used mmx registers	0
used xmm registers	10
used ymm registers	0
used zmm registers	0
nb stack references	1
ADD-SUB / MUL ratio	1.00

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	3.63 cycles
front end	3.63 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	2.17	2.17	2.17	2.17	2.17	2.17	1.25	1.25	1.25	1.25	2.50	2.50	2.50	2.50	0.00	0.00
cycles	2.17	2.17	2.17	2.17	2.17	2.17	1.25	1.25	1.25	1.25	2.50	2.50	2.50	2.50	0.00	0.00

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	NA
Longest recurrence chain latency (RecMII)	2.00

Cycles summary

Front-end	3.63
Dispatch	2.50
Data deps.	2.00
Overall L1	3.63

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	0%
fma	NA (no fma vectorizable/vectorized instructions)
other	NA (no other vectorizable/vectorized instructions)

all	8%
load	0%
store	NA (no store vectorizable/vectorized instructions)
mul	0%
add-sub	0%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	100%

INT+FP

all	5%
load	0%
store	NA (no store vectorizable/vectorized instructions)
mul	0%
add-sub	0%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	100%

Vector efficiency ratios

INT

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

all	13%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	12%
add-sub	12%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	25%

INT+FP

all	12%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	12%
add-sub	11%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	25%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 3.63 cycles. At this rate:

4% of peak load performance is reached (11.03 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))

Front-end bottlenecks

ASM code

In the binary file, the address of the loop is: 232fe0

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	Latency	Recip. throughput	Vectorization
MOVSXD %R14D,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	N/A
VXORPD %XMM2,%XMM2,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
XOR %EDI,%EDI	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.25	N/A
ADD %R13,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	N/A
NOPL (%RAX)	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.09	N/A
XOR %EDI,%EDI	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.25	N/A
NOPW (%RAX,%RAX,1)	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.09	N/A
MOV 0x30(%RSP),%RDI	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	3	0.25	N/A
LEA (%R12,%RSI,1),%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.25	N/A
DEC %R13	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
INC %R14D	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	scal (6.3%)
ADD $0x8,%RBP	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
ADD $0x8,%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
ADD $0x8,%RBX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
ADD $0x8,%RCX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
SUB %R9,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSD (%RDI,%RAX,8),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	1	0.50	scal (12.5%)
VMULSD (%R10,%RAX,8),%XMM3,%XMM4	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	0	0	3	0.50	scal (12.5%)
VMULSD (%R11,%RAX,8),%XMM4,%XMM5	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	0	0	3	0.50	scal (12.5%)
VADDSD %XMM3,%XMM10,%XMM10	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	scal (12.5%)
VMULSD (%R15,%RAX,8),%XMM3,%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	0	0	3	0.50	scal (12.5%)
LEA 0x1(%RSI),%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.25	N/A
VADDSD %XMM4,%XMM9,%XMM9	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	scal (12.5%)
VMULSD %XMM1,%XMM4,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	scal (12.5%)
VMULSD %XMM4,%XMM2,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	scal (12.5%)
VADDSD %XMM5,%XMM7,%XMM7	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	scal (12.5%)
VADDSD %XMM2,%XMM6,%XMM6	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	scal (12.5%)
VADDSD %XMM3,%XMM8,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	scal (12.5%)
CMP %RDX,%RSI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV %RAX,%RSI	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
JNE 232fe0 <__nv_field_summary_kernel_module_field_summary_kernel__PARALLEL_F2L54_1_+0x2c0>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	1	0.50-1	N/A

Report Configuration

Code clean check

Workaround

Vectorization

Details

Workaround

Execution units bottlenecks

FMA

Workaround

Type of elements and instruction set

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

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

FMA

Workaround

Type of elements and instruction set

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

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