OV - exec - Loop 2248

0x497168 LDR	D4, [X22, X4,LSL #3]    [4]

0x49716c FCMPE	D26, D4

0x497170 B.LS	4972c4

0x497174 LDR	X16, [X24, X4,LSL #3]    [3]

0x497178 STR	X16, [X19, X4,LSL #3]    [1]

0x49717c ADD	X4, X4, #1

0x497180 LDR	X28, [X0, #8]    [2]

0x497184 CMP	X28, X4

0x497188 B.GT	497168

0x4972c4 STR	X13, [X19, X4,LSL #3]    [1]

0x4972c8 ADD	X4, X4, #1

0x4972cc SUB	X1, X1, #1

0x4972d0 LDR	X30, [X0, #8]    [2]

0x4972d4 CMP	X30, X4

0x4972d8 B.GT	497168

/home/hbollore/qaas/qaas-runs/169-817-3176/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c: 451 - 460

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

451:                for (jA = A_diag_i[i]+1; jA < A_diag_i[i+1]; jA++)

452:                {

453:                   if (A_diag_data[jA] >= strength_threshold * row_scale)

454:                   {

455:                      S_temp_diag_j[jA] = -1;

456:                      --jS_diag;

457:                   }

458:                   else

459:                   {

460:                      S_temp_diag_j[jA] = A_diag_j[jA];

Coverage (%)	Name	Source Location	Module
►100.00+	GOMP_parallel		libomp.so
○	hypre_BoomerAMGCreateS	par_strength.c:246	exec
○	hypre_BoomerAMGSetup	par_amg_setup.c:580	exec
○	hypre_PCGSetup	pcg.c:234	exec
○	main	amg.c:398	exec
○	__libc_start_main		libc-2.31.so
○	_start	amg.c:599	exec

Path /

Metric	Value
CQA speedup if no scalar integer	1.33
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	1.08
Bottlenecks
Function	hypre_BoomerAMGCreateS._omp_fn.1
Source	par_strength.c:451-460
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	1.33
CQA cycles if no scalar integer	1.00
CQA cycles if FP arith vectorized	1.33
CQA cycles if fully vectorized	0.33
Front-end cycles	1.13
DIV/SQRT cycles	1.00
P0 cycles	1.00
P1 cycles	0.63
P2 cycles	0.63
P3 cycles	0.63
P4 cycles	0.63
P5 cycles	1.00
P6 cycles	0.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	1.33
P10 cycles	1.00
P11 cycles	1.17
P12 cycles	0.50
P13 cycles	0.50
P14 cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	9.00
Nb uops	9.00
Nb loads	NA
Nb stores	1.00
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	20.95
Bytes prefetched	0.00
Bytes loaded	20.00
Bytes stored	8.00
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	NA
Vectorization ratio mul	NA
Vectorization ratio add_sub	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	NA
Vector-efficiency ratio all	25.00
Vector-efficiency ratio load	25.00
Vector-efficiency ratio store	NA
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	NA

Metric	Value
CQA speedup if no scalar integer	1.50
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	1.13
Bottlenecks	P10,
Function	hypre_BoomerAMGCreateS._omp_fn.1
Source	par_strength.c:451-460
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	1.00
CQA cycles if FP arith vectorized	1.50
CQA cycles if fully vectorized	0.38
Front-end cycles	1.13
DIV/SQRT cycles	1.00
P0 cycles	1.00
P1 cycles	0.50
P2 cycles	0.50
P3 cycles	0.50
P4 cycles	0.50
P5 cycles	1.00
P6 cycles	0.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	1.50
P10 cycles	1.17
P11 cycles	1.33
P12 cycles	0.50
P13 cycles	0.50
P14 cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	9.00
Nb uops	9.00
Nb loads	NA
Nb stores	1.00
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	21.33
Bytes prefetched	0.00
Bytes loaded	24.00
Bytes stored	8.00
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	NA
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	NA
Vector-efficiency ratio all	25.00
Vector-efficiency ratio load	25.00
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	NA

Metric	Value
CQA speedup if no scalar integer	1.17
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	1.04
Bottlenecks	P10,
Function	hypre_BoomerAMGCreateS._omp_fn.1
Source	par_strength.c:451-460
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	1.17
CQA cycles if no scalar integer	1.00
CQA cycles if FP arith vectorized	1.17
CQA cycles if fully vectorized	0.29
Front-end cycles	1.13
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	1.00
P6 cycles	0.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	1.17
P10 cycles	0.83
P11 cycles	1.00
P12 cycles	0.50
P13 cycles	0.50
P14 cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	9.00
Nb uops	9.00
Nb loads	NA
Nb stores	1.00
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	20.57
Bytes prefetched	0.00
Bytes loaded	16.00
Bytes stored	8.00
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	NA
Vectorization ratio mul	NA
Vectorization ratio add_sub	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	NA
Vector-efficiency ratio all	25.00
Vector-efficiency ratio load	25.00
Vector-efficiency ratio store	NA
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	NA

Path /

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

Function	hypre_BoomerAMGCreateS._omp_fn.1
Source file and lines	par_strength.c:451-460
Module	exec

The loop is defined in /home/hbollore/qaas/qaas-runs/169-817-3176/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c:451-460.

The related source loop is not unrolled or unrolled with no peel/tail loop.
The structure of this loop is probably <if then [else] end>.

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 1.33 to 0.33 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 20 bytes. The binary loop is storing 8 bytes.

General properties

nb instructions	9
loop length	36
nb stack references	0

Front-end

front end

1.13 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.00	1.00	0.63	0.63	0.63	0.63	1.00	0.00	0.00	0.00	1.33	1.00	1.17	0.50	0.50
cycles	1.00	1.00	0.63	0.63	0.63	0.63	1.00	0.00	0.00	0.00	1.33	1.00	1.17	0.50	0.50

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	1.00

Cycles summary

Front-end	1.13
Data deps.	1.00
Overall L1	1.33

Vectorization ratios

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.33 cycles. At this rate:

15% of peak load performance is reached (14.86 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
12% of peak store performance is reached (6.10 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

Function	hypre_BoomerAMGCreateS._omp_fn.1
Source file and lines	par_strength.c:451-460
Module	exec

The loop is defined in /home/hbollore/qaas/qaas-runs/169-817-3176/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c:451-460.

The related source loop is not unrolled or unrolled with no peel/tail loop.
The structure of this loop is probably <if then [else] end>.

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

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 1.50 to 1.00 cycles (1.50x speedup).

Workaround

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

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. The binary loop is storing 8 bytes.

General properties

nb instructions	9
loop length	36
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

1.13 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.00	1.00	0.50	0.50	0.50	0.50	1.00	0.00	0.00	0.00	1.50	1.17	1.33	0.50	0.50
cycles	1.00	1.00	0.50	0.50	0.50	0.50	1.00	0.00	0.00	0.00	1.50	1.17	1.33	0.50	0.50

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	1.00

Cycles summary

Front-end	1.13
Data deps.	1.00
Overall L1	1.50

Vectorization ratios

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

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))
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: 497168

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P10	P11	P12	P13	P14	Latency	Recip. throughput
LDR D4, [X22, X4,LSL #3]	1	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	6	0.33
FCMPE D26, D4	1	0	0	0	0	0	0	1	0	0	0	0	0	2	1
B.LS 4972c4 <hypre_BoomerAMGCreateS._omp_fn.1+0x9a4>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	1	0.50
LDR X16, [X24, X4,LSL #3]	1	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
STR X16, [X19, X4,LSL #3]	1	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
ADD X4, X4, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.25
LDR X28, [X0, #8]	1	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
CMP X28, X4	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.33
B.GT 497168 <hypre_BoomerAMGCreateS._omp_fn.1+0x848>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	1	0.50

Function	hypre_BoomerAMGCreateS._omp_fn.1
Source file and lines	par_strength.c:451-460
Module	exec

The loop is defined in /home/hbollore/qaas/qaas-runs/169-817-3176/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c:451-460.

The related source loop is not unrolled or unrolled with no peel/tail loop.
The structure of this loop is probably <if then [else] end>.

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

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 1.17 to 1.00 cycles (1.17x speedup).

Workaround

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

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 1.17 to 0.29 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. The binary loop is storing 8 bytes.

General properties

nb instructions	9
loop length	36
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

1.13 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	1.00	0.00	0.00	0.00	1.17	0.83	1.00	0.50	0.50
cycles	1.00	1.00	0.75	0.75	0.75	0.75	1.00	0.00	0.00	0.00	1.17	0.83	1.00	0.50	0.50

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	1.00

Cycles summary

Front-end	1.13
Data deps.	1.00
Overall L1	1.17

Vectorization ratios

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.17 cycles. At this rate:

14% of peak load performance is reached (13.71 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
14% of peak store performance is reached (6.86 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 497168

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P10	P11	P12	P13	P14	Latency	Recip. throughput
LDR D4, [X22, X4,LSL #3]	1	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	6	0.33
FCMPE D26, D4	1	0	0	0	0	0	0	1	0	0	0	0	0	2	1
B.LS 4972c4 <hypre_BoomerAMGCreateS._omp_fn.1+0x9a4>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	1	0.50
STR X13, [X19, X4,LSL #3]	1	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
ADD X4, X4, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.25
SUB X1, X1, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.25
LDR X30, [X0, #8]	1	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
CMP X30, X4	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.33
B.GT 497168 <hypre_BoomerAMGCreateS._omp_fn.1+0x848>	1	0.50	0.50	0	0	0	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

Code clean check

Workaround

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

Code clean check

Workaround

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