OV - exec - Loop 167

	Loop Id: 167	Module: exec	Source: advec_mom.cpp:219-221 [...]	Coverage: 4.67%

0x42ae44 SEL	Z23.D, P1, Z4.D, Z17.D

0x42ae48 ADD	X10, X10, X8

0x42ae4c SDIVR	Z23.D, P0/M, Z23.D, Z22.D

0x42ae50 ADD	Z24.D, Z5.D, Z23.D

0x42ae54 MOVPRFX	Z25, Z22

0x42ae58 MLS	Z25.D, P0/M, Z23.D, Z4.D

0x42ae5c ADD	Z23.D, Z7.D, Z23.D

0x42ae60 ADD	Z22.D, Z22.D, Z16.D

0x42ae64 SXTW	Z24.D, P0/M, Z24.D

0x42ae68 SXTW	Z23.D, P0/M, Z23.D

0x42ae6c MOVPRFX	Z28, Z1

0x42ae70 MUL	Z28.D, P0/M, Z28.D, Z24.D

0x42ae74 MOVPRFX	Z26, Z0

0x42ae78 MUL	Z26.D, P0/M, Z26.D, Z24.D

0x42ae7c MUL	Z23.D, P0/M, Z23.D, Z2.D

0x42ae80 ADD	Z25.D, Z6.D, Z25.D

0x42ae84 ADR	Z29.D, [Z19, Z25.D,SXTW #3]    [11]

0x42ae88 ADR	Z27.D, [Z18, Z25.D,SXTW #3]    [9]

0x42ae8c ADR	Z28.D, [Z29, Z28.D,LSL #3]    [13]

0x42ae90 ADR	Z29.D, [Z20, Z25.D,SXTW #3]    [1]

0x42ae94 ADR	Z26.D, [Z27, Z26.D,LSL #3]    [8]

0x42ae98 ADR	Z25.D, [Z21, Z25.D,SXTW #3]    [2]

0x42ae9c ADR	Z23.D, [Z29, Z23.D,LSL #3]    [12]

0x42aea0 LD1D	{Z27.D}, P1/Z, [V26.D]    [6]

0x42aea4 LD1D	{Z28.D}, P1/Z, [V28.D]    [10]

0x42aea8 LD1D	{Z23.D}, P1/Z, [V23.D]    [3]

0x42aeac FMLA	Z23.D, P0/M, Z27.D, Z28.D

0x42aeb0 MOVPRFX	Z27, Z2

0x42aeb4 MUL	Z27.D, P0/M, Z27.D, Z24.D

0x42aeb8 MUL	Z24.D, P0/M, Z24.D, Z3.D

0x42aebc ADR	Z27.D, [Z29, Z27.D,LSL #3]    [12]

0x42aec0 ADR	Z24.D, [Z25, Z24.D,LSL #3]    [5]

0x42aec4 LD1D	{Z27.D}, P1/Z, [V27.D]    [7]

0x42aec8 FSUB	Z23.D, Z23.D, Z27.D

0x42aecc LD1D	{Z24.D}, P1/Z, [V24.D]    [4]

0x42aed0 FDIV	Z23.D, P0/M, Z23.D, Z24.D

0x42aed4 ST1D	{Z23.D}, P1, [V26.D]    [6]

0x42aed8 WHILELO	P1.D, X10, X9

0x42aedc B.MI	42ae44

/home/eoseret/qaas/qaas_runs/178-219-7589/intel/CloverLeaf2.0-CXX/build/CloverLeaf2.0-CXX/src/omp/context.h: 69 - 69

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

69:   T &operator()(size_t i, size_t j) const { return data[i + j * sizeX]; }

/home/eoseret/qaas/qaas_runs/178-219-7589/intel/CloverLeaf2.0-CXX/build/CloverLeaf2.0-CXX/src/omp/advec_mom.cpp: 219 - 221

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

219:     for (int j = (y_min + 1); j < (y_max + 1 + 2); j++) {

220:       for (int i = (x_min + 1); i < (x_max + 1 + 2); i++) {

221:         vel1(i, j) = (vel1(i, j) * node_mass_pre(i, j) + mom_flux(i + 0, j - 1) - mom_flux(i, j)) / node_mass_post(i, j);

Coverage (%)	Name	Module
►98.44+	__kmp_invoke_microtask	libomp.so
○	__kmp_invoke_task_func	libomp.so
○	__kmp_launch_thread	libomp.so
○	__kmp_launch_worker(void*)	libomp.so
○	start_thread	libc.so.6
○	thread_start	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 /

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	1.29 - 1.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	2.27
Bottlenecks	P6, P8,
Function	advec_mom_kernel(int, int, int, int, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer1D&, clover::Buffer1D&, int, int, int) [clone .omp_outlined.21]
Source	context.h:69-69,advec_mom.cpp:219-221
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	max
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	15.00 - 34.08
CQA cycles if no scalar integer	15.00 - 34.08
CQA cycles if FP arith vectorized	15.00 - 34.08
CQA cycles if fully vectorized	11.63 - 34.08
Front-end cycles	4.88
P0 cycles	0.50
P1 cycles	0.50
P2 cycles	0.50
P3 cycles	0.25
P4 cycles	1.00
P5 cycles	0.25
P6 cycles	15.00
P7 cycles	15.00
P8 cycles	3.00
P9 cycles	3.00
P10 cycles	2.17
P11 cycles	1.83
P12 cycles	2.00
P13 cycles	0.00
P14 cycles	0.00
DIV/SQRT cycles	7.87 - 34.08
Inter-iter dependencies cycles	2
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	39.00
Nb uops	39.00
Nb loads	NA
Nb stores	1.00
Nb stack references	0.00
FLOP/cycle	1.07 - 0.47
Nb FLOP add-sub	4.00
Nb FLOP mul	0.00
Nb FLOP fma	4.00
Nb FLOP div	4.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	5.67 - 12.88
Bytes prefetched	0.00
Bytes loaded	161.13
Bytes stored	32.00
Stride 0	0.00
Stride 1	0.00
Stride n	0.00
Stride unknown	9.00
Stride indirect	4.00
Vectorization ratio all	63.89
Vectorization ratio load	35.71
Vectorization ratio store	100.00
Vectorization ratio mul	100.00
Vectorization ratio add_sub	40.00
Vectorization ratio fma	100.00
Vectorization ratio div_sqrt	100.00
Vectorization ratio other	42.86
Vector-efficiency ratio all	81.25
Vector-efficiency ratio load	51.79
Vector-efficiency ratio store	100.00
Vector-efficiency ratio mul	100.00
Vector-efficiency ratio add_sub	55.00
Vector-efficiency ratio fma	100.00
Vector-efficiency ratio div_sqrt	100.00
Vector-efficiency ratio other	100.00

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	1.29 - 1.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	2.27
Bottlenecks	P6, P8,
Function	advec_mom_kernel(int, int, int, int, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer1D&, clover::Buffer1D&, int, int, int) [clone .omp_outlined.21]
Source	context.h:69-69,advec_mom.cpp:219-221
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	max
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	15.00 - 34.08
CQA cycles if no scalar integer	15.00 - 34.08
CQA cycles if FP arith vectorized	15.00 - 34.08
CQA cycles if fully vectorized	11.63 - 34.08
Front-end cycles	4.88
P0 cycles	0.50
P1 cycles	0.50
P2 cycles	0.50
P3 cycles	0.25
P4 cycles	1.00
P5 cycles	0.25
P6 cycles	15.00
P7 cycles	15.00
P8 cycles	3.00
P9 cycles	3.00
P10 cycles	2.17
P11 cycles	1.83
P12 cycles	2.00
P13 cycles	0.00
P14 cycles	0.00
DIV/SQRT cycles	7.87 - 34.08
Inter-iter dependencies cycles	2
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	39.00
Nb uops	39.00
Nb loads	NA
Nb stores	1.00
Nb stack references	0.00
FLOP/cycle	1.07 - 0.47
Nb FLOP add-sub	4.00
Nb FLOP mul	0.00
Nb FLOP fma	4.00
Nb FLOP div	4.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	5.67 - 12.88
Bytes prefetched	0.00
Bytes loaded	161.13
Bytes stored	32.00
Stride 0	0.00
Stride 1	0.00
Stride n	0.00
Stride unknown	9.00
Stride indirect	4.00
Vectorization ratio all	63.89
Vectorization ratio load	35.71
Vectorization ratio store	100.00
Vectorization ratio mul	100.00
Vectorization ratio add_sub	40.00
Vectorization ratio fma	100.00
Vectorization ratio div_sqrt	100.00
Vectorization ratio other	42.86
Vector-efficiency ratio all	81.25
Vector-efficiency ratio load	51.79
Vector-efficiency ratio store	100.00
Vector-efficiency ratio mul	100.00
Vector-efficiency ratio add_sub	55.00
Vector-efficiency ratio fma	100.00
Vector-efficiency ratio div_sqrt	100.00
Vector-efficiency ratio other	100.00

Path /

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

Function	advec_mom_kernel(int, int, int, int, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer1D&, clover::Buffer1D&, int, int, int) [clone .omp_outlined.21]
Source file and lines	advec_mom.cpp:219-221
Module	exec

The loop is defined in:

/home/eoseret/qaas/qaas_runs/178-219-7589/intel/CloverLeaf2.0-CXX/build/CloverLeaf2.0-CXX/src/omp/context.h:69
/home/eoseret/qaas/qaas_runs/178-219-7589/intel/CloverLeaf2.0-CXX/build/CloverLeaf2.0-CXX/src/omp/advec_mom.cpp:219-221

The related source loop is not unrolled or unrolled with no peel/tail loop.

gain
potential
hint
expert

Vectorization

Your loop is partially vectorized. 81% of vector register length is used (average across all VPU instructions).

Details

63% of VPU instructions are used in vector version (process two or more data elements in vector registers):

35% of VPU loads are used in vector version.
40% of VPU addition or subtraction instructions are used in vector version.
42% of VPU instructions that are not load, store, addition, subtraction nor multiply instructions are used in vector version.

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 34.08 to 14.00 cycles (2.43x speedup).

FMA

Detected 4 FMA (fused multiply-add) operations.

Slow data structures access

Detected data structures (typically arrays) that cannot be efficiently read/written

Details

Constant unknown stride: 9 occurrence(s)
Irregular (variable stride) or indirect: 4 occurrence(s)

Non-unit stride (uncontiguous) accesses are not efficiently using data caches

Workaround

Try to reorganize arrays of structures to structures of arrays
Consider to permute loops (see vectorization gain report)
Try to remove indirect accesses. If applicable, precompute elements out of the innermost loop.

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

The binary loop is composed of 16 FP arithmetical operations:

8: addition or subtraction (4 inside FMA instructions)
4: multiply (all inside FMA instructions)
4: divide

The binary loop is loading 161 bytes. The binary loop is storing 32 bytes.

Arithmetic intensity

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

General properties

nb instructions	39
nb uops	39
loop length	156
used w registers	0
used x registers	3
used b registers	0
used h registers	0
used s registers	0
used d registers	0
used q registers	0
used v registers	5
used z registers	22
nb stack references	0

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	4.88 cycles
front end	4.88 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	0.50	0.50	0.50	0.25	1.00	0.25	15.00	15.00	3.00	3.00	2.17	1.83	2.00	0.00	0.00
cycles	0.50	0.50	0.50	0.25	1.00	0.25	15.00	15.00	3.00	3.00	2.17	1.83	2.00	0.00	0.00

Execution ports to units layout:

P0: BRU
P1: BRU
P2: ALU
P3: ALU
P4: ALU
P5: ALU
P6 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P7 (128 bits): VPU, ALU, FP store data
P8 (128 bits): VPU, ALU, DIV/SQRT
P9 (128 bits): ALU, VPU
P10 (256 bits): store address, load
P11 (256 bits): store address, load
P12 (256 bits): load
P13 (64 bits): store data
P14 (64 bits): store data

Cycles executing div or sqrt instructions	7.87-34.08
Longest recurrence chain latency (RecMII)	2.00

Cycles summary

Front-end	4.88
Dispatch	15.00
DIV/SQRT	7.87-34.08
Data deps.	2.00
Overall L1	15.00-34.08

Vectorization ratios

INT

all	60%
load	35%
store	100%
mul	100%
add-sub	35%
fma	NA (no fma vectorizable/vectorized instructions)
other	50%

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

INT+FP

all	63%
load	35%
store	100%
mul	100%
add-sub	40%
fma	100%
div/sqrt	100%
other	42%

Vector efficiency ratios

INT

all	79%
load	51%
store	100%
mul	100%
add-sub	51%
fma	NA (no fma vectorizable/vectorized instructions)
other	100%

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

INT+FP

all	81%
load	51%
store	100%
mul	100%
add-sub	55%
fma	100%
div/sqrt	100%
other	100%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 34.08 cycles. At this rate:

4% of peak load performance is reached (4.73 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
1% of peak store performance is reached (0.94 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 42ae44

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	Latency	Recip. throughput	Vectorization
SEL Z23.D, P1, Z4.D, Z17.D	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	vect (100.0%)
ADD X10, X10, X8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	1	0.25	N/A
SDIVR Z23.D, P0/M, Z23.D, Z22.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	7-20	0.87-20	vect (100.0%)
ADD Z24.D, Z5.D, Z23.D	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	vect (100.0%)
MOVPRFX Z25, Z22	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (100.0%)
MLS Z25.D, P0/M, Z23.D, Z4.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	5	2	vect (100.0%)
ADD Z23.D, Z7.D, Z23.D	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	vect (100.0%)
ADD Z22.D, Z22.D, Z16.D	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	vect (100.0%)
SXTW Z24.D, P0/M, Z24.D	1	0	0	0	0	0	0	0	1	0	0	0	0	0	2	1	vect (100.0%)
SXTW Z23.D, P0/M, Z23.D	1	0	0	0	0	0	0	0	1	0	0	0	0	0	2	1	vect (100.0%)
MOVPRFX Z28, Z1	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (100.0%)
MUL Z28.D, P0/M, Z28.D, Z24.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	5	2	vect (100.0%)
MOVPRFX Z26, Z0	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (100.0%)
MUL Z26.D, P0/M, Z26.D, Z24.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	5	2	vect (100.0%)
MUL Z23.D, P0/M, Z23.D, Z2.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	5	2	vect (100.0%)
ADD Z25.D, Z6.D, Z25.D	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	vect (100.0%)
ADR Z29.D, [Z19, Z25.D,SXTW #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
ADR Z27.D, [Z18, Z25.D,SXTW #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
ADR Z28.D, [Z29, Z28.D,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
ADR Z29.D, [Z20, Z25.D,SXTW #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
ADR Z26.D, [Z27, Z26.D,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
ADR Z25.D, [Z21, Z25.D,SXTW #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
ADR Z23.D, [Z29, Z23.D,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
LD1D {Z27.D}, P1/Z, [V26.D]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.33	0.33	0.33	9	2	vect (100.0%)
LD1D {Z28.D}, P1/Z, [V28.D]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.33	0.33	0.33	9	2	vect (100.0%)
LD1D {Z23.D}, P1/Z, [V23.D]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.33	0.33	0.33	9	2	vect (100.0%)
FMLA Z23.D, P0/M, Z27.D, Z28.D	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	4	0.50	vect (100.0%)
MOVPRFX Z27, Z2	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (100.0%)
MUL Z27.D, P0/M, Z27.D, Z24.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	5	2	vect (100.0%)
MUL Z24.D, P0/M, Z24.D, Z3.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	5	2	vect (100.0%)
ADR Z27.D, [Z29, Z27.D,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
ADR Z24.D, [Z25, Z24.D,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
LD1D {Z27.D}, P1/Z, [V27.D]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.33	0.33	0.33	9	2	vect (100.0%)
FSUB Z23.D, Z23.D, Z27.D	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	vect (100.0%)
LD1D {Z24.D}, P1/Z, [V24.D]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.33	0.33	0.33	9	2	vect (100.0%)
FDIV Z23.D, P0/M, Z23.D, Z24.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	7-15	6.99-14.08	vect (100.0%)
ST1D {Z23.D}, P1, [V26.D]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	0	6	2	vect (100.0%)
WHILELO P1.D, X10, X9	1	0	0	0	0	1	0	0	0	0	0	0	0	0	3	2	N/A
B.MI 42ae44 <_Z16advec_mom_kerneliiiiRN6clover8Buffer2DIdEES2_S2_S2_S2_S2_S2_S2_S2_S2_S2_S2_S2_RNS_8Buffer1DIdEES5_iii.omp_outlined.21+0x124>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A

Function	advec_mom_kernel(int, int, int, int, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer2D&, clover::Buffer1D&, clover::Buffer1D&, int, int, int) [clone .omp_outlined.21]
Source file and lines	advec_mom.cpp:219-221
Module	exec

The loop is defined in:

/home/eoseret/qaas/qaas_runs/178-219-7589/intel/CloverLeaf2.0-CXX/build/CloverLeaf2.0-CXX/src/omp/context.h:69
/home/eoseret/qaas/qaas_runs/178-219-7589/intel/CloverLeaf2.0-CXX/build/CloverLeaf2.0-CXX/src/omp/advec_mom.cpp:219-221

The related source loop is not unrolled or unrolled with no peel/tail loop.

gain
potential
hint
expert

Vectorization

Your loop is partially vectorized. 81% of vector register length is used (average across all VPU instructions).

Details

63% of VPU instructions are used in vector version (process two or more data elements in vector registers):

35% of VPU loads are used in vector version.
40% of VPU addition or subtraction instructions are used in vector version.
42% of VPU instructions that are not load, store, addition, subtraction nor multiply instructions are used in vector version.

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 34.08 to 14.00 cycles (2.43x speedup).

FMA

Detected 4 FMA (fused multiply-add) operations.

Slow data structures access

Detected data structures (typically arrays) that cannot be efficiently read/written

Details

Constant unknown stride: 9 occurrence(s)
Irregular (variable stride) or indirect: 4 occurrence(s)

Non-unit stride (uncontiguous) accesses are not efficiently using data caches

Workaround

Try to reorganize arrays of structures to structures of arrays
Consider to permute loops (see vectorization gain report)
Try to remove indirect accesses. If applicable, precompute elements out of the innermost loop.

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

The binary loop is composed of 16 FP arithmetical operations:

8: addition or subtraction (4 inside FMA instructions)
4: multiply (all inside FMA instructions)
4: divide

The binary loop is loading 161 bytes. The binary loop is storing 32 bytes.

Arithmetic intensity

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

General properties

nb instructions	39
nb uops	39
loop length	156
used w registers	0
used x registers	3
used b registers	0
used h registers	0
used s registers	0
used d registers	0
used q registers	0
used v registers	5
used z registers	22
nb stack references	0

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	4.88 cycles
front end	4.88 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	0.50	0.50	0.50	0.25	1.00	0.25	15.00	15.00	3.00	3.00	2.17	1.83	2.00	0.00	0.00
cycles	0.50	0.50	0.50	0.25	1.00	0.25	15.00	15.00	3.00	3.00	2.17	1.83	2.00	0.00	0.00

Execution ports to units layout:

P0: BRU
P1: BRU
P2: ALU
P3: ALU
P4: ALU
P5: ALU
P6 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P7 (128 bits): VPU, ALU, FP store data
P8 (128 bits): VPU, ALU, DIV/SQRT
P9 (128 bits): ALU, VPU
P10 (256 bits): store address, load
P11 (256 bits): store address, load
P12 (256 bits): load
P13 (64 bits): store data
P14 (64 bits): store data

Cycles executing div or sqrt instructions	7.87-34.08
Longest recurrence chain latency (RecMII)	2.00

Cycles summary

Front-end	4.88
Dispatch	15.00
DIV/SQRT	7.87-34.08
Data deps.	2.00
Overall L1	15.00-34.08

Vectorization ratios

INT

all	60%
load	35%
store	100%
mul	100%
add-sub	35%
fma	NA (no fma vectorizable/vectorized instructions)
other	50%

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

INT+FP

all	63%
load	35%
store	100%
mul	100%
add-sub	40%
fma	100%
div/sqrt	100%
other	42%

Vector efficiency ratios

INT

all	79%
load	51%
store	100%
mul	100%
add-sub	51%
fma	NA (no fma vectorizable/vectorized instructions)
other	100%

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

INT+FP

all	81%
load	51%
store	100%
mul	100%
add-sub	55%
fma	100%
div/sqrt	100%
other	100%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 34.08 cycles. At this rate:

4% of peak load performance is reached (4.73 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
1% of peak store performance is reached (0.94 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 42ae44

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	Latency	Recip. throughput	Vectorization
SEL Z23.D, P1, Z4.D, Z17.D	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	vect (100.0%)
ADD X10, X10, X8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	1	0.25	N/A
SDIVR Z23.D, P0/M, Z23.D, Z22.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	7-20	0.87-20	vect (100.0%)
ADD Z24.D, Z5.D, Z23.D	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	vect (100.0%)
MOVPRFX Z25, Z22	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (100.0%)
MLS Z25.D, P0/M, Z23.D, Z4.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	5	2	vect (100.0%)
ADD Z23.D, Z7.D, Z23.D	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	vect (100.0%)
ADD Z22.D, Z22.D, Z16.D	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	vect (100.0%)
SXTW Z24.D, P0/M, Z24.D	1	0	0	0	0	0	0	0	1	0	0	0	0	0	2	1	vect (100.0%)
SXTW Z23.D, P0/M, Z23.D	1	0	0	0	0	0	0	0	1	0	0	0	0	0	2	1	vect (100.0%)
MOVPRFX Z28, Z1	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (100.0%)
MUL Z28.D, P0/M, Z28.D, Z24.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	5	2	vect (100.0%)
MOVPRFX Z26, Z0	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (100.0%)
MUL Z26.D, P0/M, Z26.D, Z24.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	5	2	vect (100.0%)
MUL Z23.D, P0/M, Z23.D, Z2.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	5	2	vect (100.0%)
ADD Z25.D, Z6.D, Z25.D	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	vect (100.0%)
ADR Z29.D, [Z19, Z25.D,SXTW #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
ADR Z27.D, [Z18, Z25.D,SXTW #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
ADR Z28.D, [Z29, Z28.D,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
ADR Z29.D, [Z20, Z25.D,SXTW #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
ADR Z26.D, [Z27, Z26.D,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
ADR Z25.D, [Z21, Z25.D,SXTW #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
ADR Z23.D, [Z29, Z23.D,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
LD1D {Z27.D}, P1/Z, [V26.D]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.33	0.33	0.33	9	2	vect (100.0%)
LD1D {Z28.D}, P1/Z, [V28.D]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.33	0.33	0.33	9	2	vect (100.0%)
LD1D {Z23.D}, P1/Z, [V23.D]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.33	0.33	0.33	9	2	vect (100.0%)
FMLA Z23.D, P0/M, Z27.D, Z28.D	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	4	0.50	vect (100.0%)
MOVPRFX Z27, Z2	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (100.0%)
MUL Z27.D, P0/M, Z27.D, Z24.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	5	2	vect (100.0%)
MUL Z24.D, P0/M, Z24.D, Z3.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	5	2	vect (100.0%)
ADR Z27.D, [Z29, Z27.D,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
ADR Z24.D, [Z25, Z24.D,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	scal (25.0%)
LD1D {Z27.D}, P1/Z, [V27.D]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.33	0.33	0.33	9	2	vect (100.0%)
FSUB Z23.D, Z23.D, Z27.D	1	0	0	0	0	0	0	0.50	0.50	0	0	0	0	0	2	0.50	vect (100.0%)
LD1D {Z24.D}, P1/Z, [V24.D]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.33	0.33	0.33	9	2	vect (100.0%)
FDIV Z23.D, P0/M, Z23.D, Z24.D	1	0	0	0	0	0	0	1	0	0	0	0	0	0	7-15	6.99-14.08	vect (100.0%)
ST1D {Z23.D}, P1, [V26.D]	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	0	6	2	vect (100.0%)
WHILELO P1.D, X10, X9	1	0	0	0	0	1	0	0	0	0	0	0	0	0	3	2	N/A
B.MI 42ae44 <_Z16advec_mom_kerneliiiiRN6clover8Buffer2DIdEES2_S2_S2_S2_S2_S2_S2_S2_S2_S2_S2_S2_RNS_8Buffer1DIdEES5_iii.omp_outlined.21+0x124>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A

Report Configuration

Vectorization

Details

Expensive FP math instructions/calls

FMA

Slow data structures access

Details

Workaround

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

Vectorization

Details

Expensive FP math instructions/calls

FMA

Slow data structures access

Details

Workaround

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