OV - exec - Loop 545

0x421638 ORR	W1, WZR, W19

0x42163c ADRP	X0,

0x421640 ADD	X0, X0, #896

0x421644 BL	40f990

0x421648 LDP	X20, X19, [SP, #128]    [2]

0x42164c LDP	X22, X21, [SP, #112]    [2]

0x421650 LDP	X24, X23, [SP, #96]    [2]

0x421654 LDP	X26, X25, [SP, #80]    [2]

0x421658 LDP	X28, X27, [SP, #64]    [2]

0x42165c LDP	X29, X30, [SP, #48]    [2]

0x421660 ADD	SP, SP, #144

0x421664 RET

0x421668 FMOV	D0, #1.0000000

0x42166c MOVN	X10, #0

0x421670 MOVZ	W11, #1

0x421674 B	42169c

0x421680 LDR	X13, [X22, X12,LSL #3]    [3]

0x421684 CMP	X13, #1

0x421688 B.LT	4216b0

0x42168c STR	X11, [X22, X12,LSL #3]    [3]

0x421690 CMP	X9, X8

0x421694 ADD	X9, X9, #1

0x421698 B.EQ	421638

0x42169c LDR	X12, [X20, X9,LSL #3]    [1]

0x4216a0 LDR	D1, [X21, X12,LSL #3]    [4]

0x4216a4 FCMP	D1, D0

0x4216a8 B.PL	421680

/home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/parcsr_ls/par_coarsen.c: 2435 - 2453

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

2435: #pragma omp parallel for private(ig, i, jS, j) HYPRE_SMP_SCHEDULE

2436: #endif

2437:      for (ig = 0; ig < graph_size; ig++) {

2438:        i = graph_array[ig];

[...]

2446:        if(measure_array[i]<1.) CF_marker[i]= F_PT;

[...]

2453:        if (CF_marker[i] > 0) CF_marker[i] = C_PT;

Coverage (%)	Name	Source Location	Module

Path /

Metric	Value
CQA speedup if no scalar integer	3.50
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	7.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.04
Bottlenecks	P10,
Function	.omp_outlined..21
Source	par_coarsen.c:2435-2438,par_coarsen.c:2446-2446,par_coarsen.c:2453-2453
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	3.50
CQA cycles if no scalar integer	1.00
CQA cycles if FP arith vectorized	3.50
CQA cycles if fully vectorized	0.50
Front-end cycles	3.38
DIV/SQRT cycles	3.00
P0 cycles	3.00
P1 cycles	2.50
P2 cycles	2.50
P3 cycles	2.50
P4 cycles	2.50
P5 cycles	1.00
P6 cycles	0.50
P7 cycles	0.25
P8 cycles	0.25
P9 cycles	3.50
P10 cycles	3.17
P11 cycles	3.33
P12 cycles	0.50
P13 cycles	0.50
P14 cycles	0.00
Inter-iter dependencies cycles	NA
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	27.00
Nb uops	27.00
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	36.57
Bytes prefetched	0.00
Bytes loaded	120.00
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	0.00
Vector-efficiency ratio all	21.88
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	20.00

Metric	Value
CQA speedup if no scalar integer	3.50
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	7.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.04
Bottlenecks	P10,
Function	.omp_outlined..21
Source	par_coarsen.c:2435-2438,par_coarsen.c:2446-2446,par_coarsen.c:2453-2453
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	3.50
CQA cycles if no scalar integer	1.00
CQA cycles if FP arith vectorized	3.50
CQA cycles if fully vectorized	0.50
Front-end cycles	3.38
DIV/SQRT cycles	3.00
P0 cycles	3.00
P1 cycles	2.50
P2 cycles	2.50
P3 cycles	2.50
P4 cycles	2.50
P5 cycles	1.00
P6 cycles	0.50
P7 cycles	0.25
P8 cycles	0.25
P9 cycles	3.50
P10 cycles	3.17
P11 cycles	3.33
P12 cycles	0.50
P13 cycles	0.50
P14 cycles	0.00
Inter-iter dependencies cycles	NA
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	27.00
Nb uops	27.00
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	36.57
Bytes prefetched	0.00
Bytes loaded	120.00
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	0.00
Vector-efficiency ratio all	21.88
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	20.00

Path /

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

Function	.omp_outlined..21
Source file and lines	par_coarsen.c:2435-2453
Module	exec

The loop is defined in /home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/parcsr_ls/par_coarsen.c:2435-2438,2446,2453.

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

Ignoring paths for analysis
Failed to get the number of paths
RecMII not computed since number of paths is unknown or > max_paths

gain
potential
hint
expert

Code clean check

Detected a slowdown caused by scalar integer instructions (typically used for address computation). By removing them, you can lower the cost of an iteration from 3.50 to 1.00 cycles (3.50x 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. Only 21% of vector register length is used (average across all VPU instructions). By vectorizing your loop, you can lower the cost of an iteration from 3.50 to 0.50 cycles (7.00x 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)

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 3.50 to 0.50 cycles (7.00x speedup).

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 120 bytes. The binary loop is storing 8 bytes.

General properties

nb instructions	27
loop length	108
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

3.38 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	3.00	3.00	2.50	2.50	2.50	2.50	1.00	0.50	0.25	0.25	3.50	3.17	3.33	0.50	0.50
cycles	3.00	3.00	2.50	2.50	2.50	2.50	1.00	0.50	0.25	0.25	3.50	3.17	3.33	0.50	0.50

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

Cycles summary

Front-end	3.38
Overall L1	3.50

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	0%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 3.50 cycles. At this rate:

35% of peak load performance is reached (34.29 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
4% of peak store performance is reached (2.29 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 421690

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	Latency	Recip. throughput
ORR W1, WZR, W19	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ADRP X0, <4db63c>	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ADD X0, X0, #896	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
BL 40f990 <@plt_start@+0x530>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50
LDP X20, X19, [SP, #128]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X22, X21, [SP, #112]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X24, X23, [SP, #96]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X26, X25, [SP, #80]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X28, X27, [SP, #64]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X29, X30, [SP, #48]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
ADD SP, SP, #144	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
RET	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
FMOV D0, #1.0000000	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	2	0.25
MOVN X10, #0	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
MOVZ W11, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
B 42169c <.omp_outlined..21+0x10c>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
LDR X13, [X22, X12,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
CMP X13, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.33
B.LT 4216b0 <.omp_outlined..21+0x120>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
STR X11, [X22, X12,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
CMP X9, X8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.33
ADD X9, X9, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
B.EQ 421638 <.omp_outlined..21+0xa8>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
LDR X12, [X20, X9,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
LDR D1, [X21, X12,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	6	0.33
FCMP D1, D0	1	0	0	0	0	0	0	1	0	0	0	0	0	0	0	0	2	1
B.PL 421680 <.omp_outlined..21+0xf0>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50

Function	.omp_outlined..21
Source file and lines	par_coarsen.c:2435-2453
Module	exec

The loop is defined in /home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/parcsr_ls/par_coarsen.c:2435-2438,2446,2453.

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

Ignoring paths for analysis
Failed to get the number of paths
RecMII not computed since number of paths is unknown or > max_paths

gain
potential
hint
expert

Code clean check

Detected a slowdown caused by scalar integer instructions (typically used for address computation). By removing them, you can lower the cost of an iteration from 3.50 to 1.00 cycles (3.50x 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. Only 21% of vector register length is used (average across all VPU instructions). By vectorizing your loop, you can lower the cost of an iteration from 3.50 to 0.50 cycles (7.00x 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)

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 3.50 to 0.50 cycles (7.00x speedup).

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 120 bytes. The binary loop is storing 8 bytes.

General properties

nb instructions	27
loop length	108
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

3.38 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	3.00	3.00	2.50	2.50	2.50	2.50	1.00	0.50	0.25	0.25	3.50	3.17	3.33	0.50	0.50
cycles	3.00	3.00	2.50	2.50	2.50	2.50	1.00	0.50	0.25	0.25	3.50	3.17	3.33	0.50	0.50

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

Cycles summary

Front-end	3.38
Overall L1	3.50

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	0%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 3.50 cycles. At this rate:

35% of peak load performance is reached (34.29 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
4% of peak store performance is reached (2.29 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 421690

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	Latency	Recip. throughput
ORR W1, WZR, W19	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ADRP X0, <4db63c>	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ADD X0, X0, #896	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
BL 40f990 <@plt_start@+0x530>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50
LDP X20, X19, [SP, #128]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X22, X21, [SP, #112]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X24, X23, [SP, #96]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X26, X25, [SP, #80]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X28, X27, [SP, #64]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
LDP X29, X30, [SP, #48]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
ADD SP, SP, #144	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
RET	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
FMOV D0, #1.0000000	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	2	0.25
MOVN X10, #0	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
MOVZ W11, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
B 42169c <.omp_outlined..21+0x10c>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
LDR X13, [X22, X12,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
CMP X13, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.33
B.LT 4216b0 <.omp_outlined..21+0x120>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
STR X11, [X22, X12,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
CMP X9, X8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.33
ADD X9, X9, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
B.EQ 421638 <.omp_outlined..21+0xa8>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
LDR X12, [X20, X9,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
LDR D1, [X21, X12,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	6	0.33
FCMP D1, D0	1	0	0	0	0	0	0	1	0	0	0	0	0	0	0	0	2	1
B.PL 421680 <.omp_outlined..21+0xf0>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50

Report Configuration

Code clean check

Workaround

Vectorization

Details

Workaround

Expensive FP math instructions/calls

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

Expensive FP math instructions/calls

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