OV - exec - Loop 420

0x42c228 ADD	X3, X12, X1

0x42c22c SUB	X1, X1, #2

0x42c230 CMP	X17, X1

0x42c234 ADD	W4, W3, #1

0x42c238 SBFM	X3, X3, #0, #31

0x42c23c SBFM	X4, X4, #0, #31

0x42c240 MADD	X3, X22, X3, XZR

0x42c244 MADD	X4, X22, X4, XZR

0x42c248 LDR	D0, [X0, X4,LSL #3]    [4]

0x42c24c STR	D0, [X2, X16,LSL #3]    [2]

0x42c250 LDR	D0, [X0, X3,LSL #3]    [3]

0x42c254 STR	D0, [X2, X13,LSL #3]    [1]

0x42c258 ADD	X2, X2, X15

0x42c25c B.NE	42c228

/home/hbollore/qaas-runs/170-300-2310/intel/CloverLeafCXX/build/CloverLeafCXX/src/omp/context.h: 69 - 69

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

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

/home/hbollore/qaas-runs/170-300-2310/intel/CloverLeafCXX/build/CloverLeafCXX/src/omp/update_halo.cpp: 185 - 186

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

185:         for (int k = 0; k < depth; ++k) {

186:           field.energy1(j, y_max + 2 + k) = field.energy1(j, y_max + 1 - k);

Path /

Metric	Value
CQA speedup if no scalar integer	1.69
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	4.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.29
Bottlenecks	P2, P3, P4, P5,
Function	.omp_outlined..26
Source	context.h:69-69,update_halo.cpp:185-186
Source loop unroll info	unrolled by 5
Source loop unroll confidence level	max
Unroll/vectorization loop type	NA
Unroll factor	1
CQA cycles	2.25
CQA cycles if no scalar integer	1.33
CQA cycles if FP arith vectorized	2.25
CQA cycles if fully vectorized	0.56
Front-end cycles	1.75
DIV/SQRT cycles	0.50
P0 cycles	0.50
P1 cycles	2.25
P2 cycles	2.25
P3 cycles	2.25
P4 cycles	2.25
P5 cycles	1.00
P6 cycles	1.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	1.33
P10 cycles	1.33
P11 cycles	1.33
P12 cycles	0.00
P13 cycles	0.00
P14 cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	14.00
Nb uops	14.00
Nb loads	NA
Nb stores	2.00
Nb stack references	0.00
FLOP/cycle	0.00
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
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	14.22
Bytes prefetched	0.00
Bytes loaded	16.00
Bytes stored	16.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	25.00
Vector-efficiency ratio load	25.00
Vector-efficiency ratio store	25.00
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	25.00
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	1.69
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	4.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.29
Bottlenecks	P2, P3, P4, P5,
Function	.omp_outlined..26
Source	context.h:69-69,update_halo.cpp:185-186
Source loop unroll info	unrolled by 5
Source loop unroll confidence level	max
Unroll/vectorization loop type	NA
Unroll factor	1
CQA cycles	2.25
CQA cycles if no scalar integer	1.33
CQA cycles if FP arith vectorized	2.25
CQA cycles if fully vectorized	0.56
Front-end cycles	1.75
DIV/SQRT cycles	0.50
P0 cycles	0.50
P1 cycles	2.25
P2 cycles	2.25
P3 cycles	2.25
P4 cycles	2.25
P5 cycles	1.00
P6 cycles	1.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	1.33
P10 cycles	1.33
P11 cycles	1.33
P12 cycles	0.00
P13 cycles	0.00
P14 cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	14.00
Nb uops	14.00
Nb loads	NA
Nb stores	2.00
Nb stack references	0.00
FLOP/cycle	0.00
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
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	14.22
Bytes prefetched	0.00
Bytes loaded	16.00
Bytes stored	16.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	25.00
Vector-efficiency ratio load	25.00
Vector-efficiency ratio store	25.00
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	25.00
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	.omp_outlined..26
Source file and lines	update_halo.cpp:185-186
Module	exec

The loop is defined in:

/home/hbollore/qaas-runs/170-300-2310/intel/CloverLeafCXX/build/CloverLeafCXX/src/omp/context.h:69
/home/hbollore/qaas-runs/170-300-2310/intel/CloverLeafCXX/build/CloverLeafCXX/src/omp/update_halo.cpp:185-186

The related source loop is unrolled by 5 (including vectorization).

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 2.25 to 1.33 cycles (1.69x speedup).

Workaround

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

Unrolling/vectorization cost

This loop is peel/tail of a unrolled/vectorized loop. If its cost is not negligible compared to the main (unrolled/vectorized) loop, unrolling/vectorization is counterproductive due to low trip count.

Details

The more iterations the main loop is processing, the higher the trip count must be to amortize peel/tail overhead.

Workaround

use profile-guided optimization flags of your compiler
hardcode most frequent values of loop bounds by adding specialized paths.:
- For instance, replace for (i=0; i<n; i++) foo(i) with: switch (n) { case (4): for (i=0; i<4; i++) foo(i); break; case (6): for (i=0; i<6; i++) foo(i); break; default : for (i=0; i<n; i++) foo(i); break; }

Vectorization

Your loop is not vectorized. 4 data elements could be processed at once in vector registers.

Details

All VPU instructions are used in scalar version (process only one data element in vector registers).

No data for this section

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

The binary loop does not contain any FP arithmetical operations. The binary loop is loading 16 bytes. The binary loop is storing 16 bytes.

General properties

nb instructions	14
loop length	56
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

1.75 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	0.50	0.50	2.25	2.25	2.25	2.25	1.00	1.00	0.00	0.00	1.33	1.33	1.33	0.00	0.00
cycles	0.50	0.50	2.25	2.25	2.25	2.25	1.00	1.00	0.00	0.00	1.33	1.33	1.33	0.00	0.00

Execution ports to units layout:

P0:
P1:
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	NA
Longest recurrence chain latency (RecMII)	1.00

Cycles summary

Front-end	1.75
Data deps.	1.00
Overall L1	2.25

Vectorization ratios

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 2.25 cycles. At this rate:

7% of peak load performance is reached (7.11 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
14% of peak store performance is reached (7.11 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 42c228

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P10	P11	P12	Latency	Recip. throughput
ADD X3, X12, X1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
SUB X1, X1, #2	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
CMP X17, X1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33
ADD W4, W3, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
SBFM X3, X3, #0, #31	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
SBFM X4, X4, #0, #31	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
MADD X3, X22, X3, XZR	1	0	0	0	0	1	0	0	0	0	0	0	2	1
MADD X4, X22, X4, XZR	1	0	0	0	0	1	0	0	0	0	0	0	2	1
LDR D0, [X0, X4,LSL #3]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
STR D0, [X2, X16,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0.50	0.50	0	2	0.50
LDR D0, [X0, X3,LSL #3]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
STR D0, [X2, X13,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0.50	0.50	0	2	0.50
ADD X2, X2, X15	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
B.NE 42c228 <.omp_outlined..26+0x148>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50

Function	.omp_outlined..26
Source file and lines	update_halo.cpp:185-186
Module	exec

The loop is defined in:

/home/hbollore/qaas-runs/170-300-2310/intel/CloverLeafCXX/build/CloverLeafCXX/src/omp/context.h:69
/home/hbollore/qaas-runs/170-300-2310/intel/CloverLeafCXX/build/CloverLeafCXX/src/omp/update_halo.cpp:185-186

The related source loop is unrolled by 5 (including vectorization).

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 2.25 to 1.33 cycles (1.69x speedup).

Workaround

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

Unrolling/vectorization cost

This loop is peel/tail of a unrolled/vectorized loop. If its cost is not negligible compared to the main (unrolled/vectorized) loop, unrolling/vectorization is counterproductive due to low trip count.

Details

The more iterations the main loop is processing, the higher the trip count must be to amortize peel/tail overhead.

Workaround

use profile-guided optimization flags of your compiler
hardcode most frequent values of loop bounds by adding specialized paths.:
- For instance, replace for (i=0; i<n; i++) foo(i) with: switch (n) { case (4): for (i=0; i<4; i++) foo(i); break; case (6): for (i=0; i<6; i++) foo(i); break; default : for (i=0; i<n; i++) foo(i); break; }

Vectorization

Your loop is not vectorized. 4 data elements could be processed at once in vector registers.

Details

All VPU instructions are used in scalar version (process only one data element in vector registers).

No data for this section

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

The binary loop does not contain any FP arithmetical operations. The binary loop is loading 16 bytes. The binary loop is storing 16 bytes.

General properties

nb instructions	14
loop length	56
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

1.75 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	0.50	0.50	2.25	2.25	2.25	2.25	1.00	1.00	0.00	0.00	1.33	1.33	1.33	0.00	0.00
cycles	0.50	0.50	2.25	2.25	2.25	2.25	1.00	1.00	0.00	0.00	1.33	1.33	1.33	0.00	0.00

Execution ports to units layout:

P0:
P1:
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	NA
Longest recurrence chain latency (RecMII)	1.00

Cycles summary

Front-end	1.75
Data deps.	1.00
Overall L1	2.25

Vectorization ratios

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 2.25 cycles. At this rate:

7% of peak load performance is reached (7.11 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
14% of peak store performance is reached (7.11 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 42c228

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P10	P11	P12	Latency	Recip. throughput
ADD X3, X12, X1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
SUB X1, X1, #2	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
CMP X17, X1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33
ADD W4, W3, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
SBFM X3, X3, #0, #31	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
SBFM X4, X4, #0, #31	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
MADD X3, X22, X3, XZR	1	0	0	0	0	1	0	0	0	0	0	0	2	1
MADD X4, X22, X4, XZR	1	0	0	0	0	1	0	0	0	0	0	0	2	1
LDR D0, [X0, X4,LSL #3]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
STR D0, [X2, X16,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0.50	0.50	0	2	0.50
LDR D0, [X0, X3,LSL #3]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
STR D0, [X2, X13,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0.50	0.50	0	2	0.50
ADD X2, X2, X15	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
B.NE 42c228 <.omp_outlined..26+0x148>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50

Report Configuration

Code clean check

Workaround

Unrolling/vectorization cost

Details

Workaround

Vectorization

Details

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code

Code clean check

Workaround

Unrolling/vectorization cost

Details

Workaround

Vectorization

Details

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code