OV - exec - Loop 1092

0x4658e4 ADD	X10, X10, #1

0x4658e8 CMP	X10, X8

0x4658ec B.GE	465940

0x4658f0 LDR	X4, [X12, X10,LSL #3]    [5]

0x4658f4 LDR	X19, [X13, X4,LSL #3]    [3]

0x4658f8 CMP	X19, #1

0x4658fc B.LT	4658e4

0x465900 LDR	X4, [X9, X4,LSL #3]    [6]

0x465904 CMP	X4, X18

0x465908 B.EQ	4658e4

0x46590c LDR	X19, [X24, X4,LSL #3]    [2]

0x465910 CMP	X19, X1

0x465914 B.GE	4658e4

0x465918 LDUR	X8, [X29, #480]    [1]

0x46591c STR	X8, [X24, X4,LSL #3]    [2]

0x465920 LDUR	X8, [X29, #480]    [1]

0x465924 ADD	X8, X8, #1

0x465928 STUR	X8, [X29, #480]    [1]

0x46592c LDR	X8, [X11, X28,LSL #3]    [4]

0x465930 B	4658e4

/home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c: 1743 - 1752

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

1743:                  for (jj2 = S_diag_i[i2]; jj2 < S_diag_i[i2+1]; jj2++)

1744:                  {

1745:                     i3 = S_diag_j[jj2];

1746:                     if (CF_marker[i3] > 0)

1747:                     {

1748:                        index = fine_to_coarse[i3];

1749:                        if (index != ic && S_marker[index] < jj_row_begin_diag)

1750:                        {

1751:                           S_marker[index] = num_nonzeros_diag;

1752:                           num_nonzeros_diag++;

Coverage (%)	Name	Source Location	Module

Path /

Metric	Value
CQA speedup if no scalar integer	1.00
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	1.20
Bottlenecks
Function	.omp_outlined..20#0x465590
Source	par_strength.c:1743-1752
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	1.88
CQA cycles if no scalar integer	1.88
CQA cycles if FP arith vectorized	1.88
CQA cycles if fully vectorized	0.47
Front-end cycles	1.56
DIV/SQRT cycles	1.75
P0 cycles	1.75
P1 cycles	1.13
P2 cycles	1.13
P3 cycles	1.13
P4 cycles	1.13
P5 cycles	0.00
P6 cycles	0.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	1.50
P10 cycles	1.50
P11 cycles	1.50
P12 cycles	0.25
P13 cycles	0.25
P14 cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	12.50
Nb uops	12.50
Nb loads	NA
Nb stores	0.50
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	18.00
Bytes prefetched	0.00
Bytes loaded	32.00
Bytes stored	4.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	0.00
Vectorization ratio load	NA
Vectorization ratio store	NA
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	25.00
Vector-efficiency ratio load	NA
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
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	1.00
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	1.14
Bottlenecks	P0, P1,
Function	.omp_outlined..20#0x465590
Source	par_strength.c:1743-1752
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	1.00
CQA cycles if no scalar integer	1.00
CQA cycles if FP arith vectorized	1.00
CQA cycles if fully vectorized	0.25
Front-end cycles	0.88
DIV/SQRT cycles	1.00
P0 cycles	1.00
P1 cycles	0.75
P2 cycles	0.75
P3 cycles	0.75
P4 cycles	0.75
P5 cycles	0.00
P6 cycles	0.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.67
P10 cycles	0.67
P11 cycles	0.67
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	7.00
Nb uops	7.00
Nb loads	NA
Nb stores	0.00
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	16.00
Bytes prefetched	0.00
Bytes loaded	16.00
Bytes stored	0.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	0.00
Vectorization ratio load	NA
Vectorization ratio store	NA
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	25.00
Vector-efficiency ratio load	NA
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
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	1.00
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	1.20
Bottlenecks	P0, P1,
Function	.omp_outlined..20#0x465590
Source	par_strength.c:1743-1752
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	1.50
CQA cycles if no scalar integer	1.50
CQA cycles if FP arith vectorized	1.50
CQA cycles if fully vectorized	0.38
Front-end cycles	1.25
DIV/SQRT cycles	1.50
P0 cycles	1.50
P1 cycles	1.00
P2 cycles	1.00
P3 cycles	1.00
P4 cycles	1.00
P5 cycles	0.00
P6 cycles	0.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	1.00
P10 cycles	1.00
P11 cycles	1.00
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	10.00
Nb uops	10.00
Nb loads	NA
Nb stores	0.00
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	16.00
Bytes prefetched	0.00
Bytes loaded	24.00
Bytes stored	0.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	0.00
Vectorization ratio load	NA
Vectorization ratio store	NA
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	25.00
Vector-efficiency ratio load	NA
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
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	1.00
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	1.20
Bottlenecks	P10, P11, P12,
Function	.omp_outlined..20#0x465590
Source	par_strength.c:1743-1752
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	3.00
CQA cycles if no scalar integer	3.00
CQA cycles if FP arith vectorized	3.00
CQA cycles if fully vectorized	0.75
Front-end cycles	2.50
DIV/SQRT cycles	2.50
P0 cycles	2.50
P1 cycles	1.50
P2 cycles	1.50
P3 cycles	1.50
P4 cycles	1.50
P5 cycles	0.00
P6 cycles	0.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	3.00
P10 cycles	3.00
P11 cycles	3.00
P12 cycles	1.00
P13 cycles	1.00
P14 cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	20.00
Nb uops	20.00
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	24.00
Bytes prefetched	0.00
Bytes loaded	56.00
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	NA
Vectorization ratio store	NA
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	25.00
Vector-efficiency ratio load	NA
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
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	1.00
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	1.23
Bottlenecks	P0, P1,
Function	.omp_outlined..20#0x465590
Source	par_strength.c:1743-1752
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	2.00
CQA cycles if no scalar integer	2.00
CQA cycles if FP arith vectorized	2.00
CQA cycles if fully vectorized	0.50
Front-end cycles	1.63
DIV/SQRT cycles	2.00
P0 cycles	2.00
P1 cycles	1.25
P2 cycles	1.25
P3 cycles	1.25
P4 cycles	1.25
P5 cycles	0.00
P6 cycles	0.00
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	13.00
Nb uops	13.00
Nb loads	NA
Nb stores	0.00
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	16.00
Bytes prefetched	0.00
Bytes loaded	32.00
Bytes stored	0.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	0.00
Vectorization ratio load	NA
Vectorization ratio store	NA
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	25.00
Vector-efficiency ratio load	NA
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
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..20#0x465590
Source file and lines	par_strength.c:1743-1752
Module	exec

The loop is defined in /home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c:1743-1752.

The related source loop is not unrolled or unrolled with no peel/tail loop.
This loop has 4 execution paths.

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

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

Ex: if (x<0) x=0 => x = (x<0 ? 0 : x) (or MAX(0,x) after defining the corresponding macro)

gain
potential
hint
expert

Vectorization

Your loop is not vectorized. 4 data elements could be processed at once in vector registers. By vectorizing your loop, you can lower the cost of an iteration from 1.88 to 0.47 cycles (4.00x speedup).

Details

All VPU 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: C storage order is row-major: for(i) for(j) a[j][i] = b[j][i]; (slow, non stride 1) => for(i) for(j) a[i][j] = b[i][j]; (fast, stride 1)
- If your loop streams arrays of structures (AoS), try to use structures of arrays instead (SoA): for(i) a[i].x = b[i].x; (slow, non stride 1) => for(i) a.x[i] = b.x[i]; (fast, stride 1)

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 32 bytes. The binary loop is storing 4 bytes.

General properties

nb instructions	12.50
loop length	50
nb stack references	0

Front-end

front end

1.56 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.75	1.75	1.13	1.13	1.13	1.13	0.00	0.00	0.00	0.00	1.50	1.50	1.50	0.25	0.25
cycles	1.75	1.75	1.13	1.13	1.13	1.13	0.00	0.00	0.00	0.00	1.50	1.50	1.50	0.25	0.25

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	1.56
Data deps.	1.00
Overall L1	1.88

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%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.88 cycles. At this rate:

17% of peak load performance is reached (16.67 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
2% of peak store performance is reached (1.33 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

Function	.omp_outlined..20#0x465590
Source file and lines	par_strength.c:1743-1752
Module	exec

The loop is defined in /home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c:1743-1752.

The related source loop is not unrolled or unrolled with no peel/tail loop.
This loop has 4 execution paths.

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

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

Ex: if (x<0) x=0 => x = (x<0 ? 0 : x) (or MAX(0,x) after defining the corresponding macro)

gain
potential
hint
expert

Vectorization

Your loop is not vectorized. 4 data elements could be processed at once in vector registers. By vectorizing your loop, you can lower the cost of an iteration from 1.00 to 0.25 cycles (4.00x speedup).

Details

All VPU 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: C storage order is row-major: for(i) for(j) a[j][i] = b[j][i]; (slow, non stride 1) => for(i) for(j) a[i][j] = b[i][j]; (fast, stride 1)
- If your loop streams arrays of structures (AoS), try to use structures of arrays instead (SoA): for(i) a[i].x = b[i].x; (slow, non stride 1) => for(i) a.x[i] = b.x[i]; (fast, stride 1)

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.

Unroll opportunity

Loop body is too small to efficiently use resources.

Workaround

Unroll your loop if trip count is significantly higher than target unroll factor. This can be done manually.

General properties

nb instructions	7
loop length	28
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

0.88 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.00	1.00	0.75	0.75	0.75	0.75	0.00	0.00	0.00	0.00	0.67	0.67	0.67	0.00	0.00
cycles	1.00	1.00	0.75	0.75	0.75	0.75	0.00	0.00	0.00	0.00	0.67	0.67	0.67	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	0.88
Data deps.	1.00
Overall L1	1.00

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%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.00 cycles. At this rate:

16% of peak load performance is reached (16.00 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 4658f0

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P10	P11	P12	Latency	Recip. throughput
ADD X10, X10, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.25
CMP X10, X8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33
B.GE 465940 <.omp_outlined..20+0x3b0>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50
LDR X4, [X12, X10,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
LDR X19, [X13, X4,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
CMP X19, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33
B.LT 4658e4 <.omp_outlined..20+0x354>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50

Function	.omp_outlined..20#0x465590
Source file and lines	par_strength.c:1743-1752
Module	exec

The loop is defined in /home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c:1743-1752.

The related source loop is not unrolled or unrolled with no peel/tail loop.
This loop has 4 execution paths.

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

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

Ex: if (x<0) x=0 => x = (x<0 ? 0 : x) (or MAX(0,x) after defining the corresponding macro)

gain
potential
hint
expert

Vectorization

Your loop is not vectorized. 4 data elements could be processed at once in vector registers. By vectorizing your loop, you can lower the cost of an iteration from 1.50 to 0.37 cycles (4.00x speedup).

Details

All VPU 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: C storage order is row-major: for(i) for(j) a[j][i] = b[j][i]; (slow, non stride 1) => for(i) for(j) a[i][j] = b[i][j]; (fast, stride 1)
- If your loop streams arrays of structures (AoS), try to use structures of arrays instead (SoA): for(i) a[i].x = b[i].x; (slow, non stride 1) => for(i) a.x[i] = b.x[i]; (fast, stride 1)

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 24 bytes.

General properties

nb instructions	10
loop length	40
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

1.25 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.50	1.50	1.00	1.00	1.00	1.00	0.00	0.00	0.00	0.00	1.00	1.00	1.00	0.00	0.00
cycles	1.50	1.50	1.00	1.00	1.00	1.00	0.00	0.00	0.00	0.00	1.00	1.00	1.00	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	1.25
Data deps.	1.00
Overall L1	1.50

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%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.50 cycles. At this rate:

16% of peak load performance is reached (16.00 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 4658f0

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P10	P11	P12	Latency	Recip. throughput
ADD X10, X10, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.25
CMP X10, X8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33
B.GE 465940 <.omp_outlined..20+0x3b0>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50
LDR X4, [X12, X10,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
LDR X19, [X13, X4,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
CMP X19, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33
B.LT 4658e4 <.omp_outlined..20+0x354>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50
LDR X4, [X9, X4,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
CMP X4, X18	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33
B.EQ 4658e4 <.omp_outlined..20+0x354>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50

Function	.omp_outlined..20#0x465590
Source file and lines	par_strength.c:1743-1752
Module	exec

The loop is defined in /home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c:1743-1752.

The related source loop is not unrolled or unrolled with no peel/tail loop.
This loop has 4 execution paths.

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

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

Ex: if (x<0) x=0 => x = (x<0 ? 0 : x) (or MAX(0,x) after defining the corresponding macro)

gain
potential
hint
expert

Vectorization

Your loop is not vectorized. 4 data elements could be processed at once in vector registers. By vectorizing your loop, you can lower the cost of an iteration from 3.00 to 0.75 cycles (4.00x speedup).

Details

All VPU 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: C storage order is row-major: for(i) for(j) a[j][i] = b[j][i]; (slow, non stride 1) => for(i) for(j) a[i][j] = b[i][j]; (fast, stride 1)
- If your loop streams arrays of structures (AoS), try to use structures of arrays instead (SoA): for(i) a[i].x = b[i].x; (slow, non stride 1) => for(i) a.x[i] = b.x[i]; (fast, stride 1)

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 56 bytes. The binary loop is storing 16 bytes.

General properties

nb instructions	20
loop length	80
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

2.50 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	2.50	2.50	1.50	1.50	1.50	1.50	0.00	0.00	0.00	0.00	3.00	3.00	3.00	1.00	1.00
cycles	2.50	2.50	1.50	1.50	1.50	1.50	0.00	0.00	0.00	0.00	3.00	3.00	3.00	1.00	1.00

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	2.50
Data deps.	1.00
Overall L1	3.00

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%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 3.00 cycles. At this rate:

19% of peak load performance is reached (18.67 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
11% of peak store performance is reached (5.33 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 4658f0

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P10	P11	P12	P13	P14	Latency	Recip. throughput
ADD X10, X10, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
CMP X10, X8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33
B.GE 465940 <.omp_outlined..20+0x3b0>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50
LDR X4, [X12, X10,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
LDR X19, [X13, X4,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
CMP X19, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33
B.LT 4658e4 <.omp_outlined..20+0x354>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50
LDR X4, [X9, X4,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
CMP X4, X18	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33
B.EQ 4658e4 <.omp_outlined..20+0x354>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50
LDR X19, [X24, X4,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
CMP X19, X1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33
B.GE 4658e4 <.omp_outlined..20+0x354>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50
LDUR X8, [X29, #480]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
STR X8, [X24, X4,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
LDUR X8, [X29, #480]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
ADD X8, X8, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
STUR X8, [X29, #480]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
LDR X8, [X11, X28,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
B 4658e4 <.omp_outlined..20+0x354>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50

Function	.omp_outlined..20#0x465590
Source file and lines	par_strength.c:1743-1752
Module	exec

The loop is defined in /home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c:1743-1752.

The related source loop is not unrolled or unrolled with no peel/tail loop.
This loop has 4 execution paths.

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

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

Ex: if (x<0) x=0 => x = (x<0 ? 0 : x) (or MAX(0,x) after defining the corresponding macro)

gain
potential
hint
expert

Vectorization

Your loop is not vectorized. 4 data elements could be processed at once in vector registers. By vectorizing your loop, you can lower the cost of an iteration from 2.00 to 0.50 cycles (4.00x speedup).

Details

All VPU 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: C storage order is row-major: for(i) for(j) a[j][i] = b[j][i]; (slow, non stride 1) => for(i) for(j) a[i][j] = b[i][j]; (fast, stride 1)
- If your loop streams arrays of structures (AoS), try to use structures of arrays instead (SoA): for(i) a[i].x = b[i].x; (slow, non stride 1) => for(i) a.x[i] = b.x[i]; (fast, stride 1)

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 32 bytes.

General properties

nb instructions	13
loop length	52
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

1.63 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	2.00	2.00	1.25	1.25	1.25	1.25	0.00	0.00	0.00	0.00	1.33	1.33	1.33	0.00	0.00
cycles	2.00	2.00	1.25	1.25	1.25	1.25	0.00	0.00	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	1.63
Data deps.	1.00
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%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 2.00 cycles. At this rate:

16% of peak load performance is reached (16.00 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 4658f0

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P10	P11	P12	Latency	Recip. throughput
ADD X10, X10, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.25
CMP X10, X8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33
B.GE 465940 <.omp_outlined..20+0x3b0>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50
LDR X4, [X12, X10,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
LDR X19, [X13, X4,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
CMP X19, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33
B.LT 4658e4 <.omp_outlined..20+0x354>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50
LDR X4, [X9, X4,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
CMP X4, X18	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33
B.EQ 4658e4 <.omp_outlined..20+0x354>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50
LDR X19, [X24, X4,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
CMP X19, X1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33
B.GE 4658e4 <.omp_outlined..20+0x354>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50

Report Configuration

Vectorization

Details

Workaround

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

Vectorization

Details

Workaround

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

Unroll opportunity

Workaround

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code

Vectorization

Details

Workaround

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

Vectorization

Details

Workaround

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

Vectorization

Details

Workaround

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