OV - exec - Loop 156

  (158) 0x421560 LDR	X18, [X1]

  (158) 0x421564 LDR	D0, [X14, X18,LSL #3]

  (158) 0x421568 UBFM	X13, X18, #61, #60

  (158) 0x42156c ADD	X7, X3, X13

  (158) 0x421570 FCMPE	D0, D1

  (158) 0x421574 B.MI	4215a4

  (158) 0x421578 LDR	X30, [X3, X18,LSL #3]

  (158) 0x42157c CMP	X30, #0

  (158) 0x421580 B.LE	4215a8

  (158) 0x421584 STR	X15, [X3, X18,LSL #3]

  (158) 0x421588 ADD	X1, X1, #8

  (158) 0x42158c CMP	X8, X1

  (158) 0x421590 B.NE	421560

0x4215a4 STR	X9, [X3, X18,LSL #3]

  (157) 0x4215a8 ADD	X20, X13, #8

  (157) 0x4215ac LDR	X0, [X17, X18,LSL #3]

  (157) 0x4215b0 ADD	X21, X17, X20

  (157) 0x4215b4 LDR	X4, [X17, X20]

  (157) 0x4215b8 CMP	X0, X4

  (157) 0x4215bc B.GE	4215e4

  (162) 0x4215c0 LDR	X19, [X12, X0,LSL #3]

  (162) 0x4215c4 LDR	X2, [X3, X19,LSL #3]

  (162) 0x4215c8 CMP	X2, #0

  (162) 0x4215cc B.LE	42163c

  (162) 0x4215d0 STR	X9, [X7]

  (162) 0x4215d4 ADD	X0, X0, #1

  (162) 0x4215d8 LDR	X4, [X21]

  (162) 0x4215dc CMP	X0, X4

  (162) 0x4215e0 B.LT	4215c0

0x4215e4 LDR	X13, [X16, X18,LSL #3]

0x4215e8 ADD	X5, X16, X20

0x4215ec LDR	X30, [X16, X20]

0x4215f0 CMP	X13, X30

0x4215f4 B.GE	421588

  (160) 0x4215f8 LDR	X6, [X11, X13,LSL #3]

  (160) 0x4215fc LDR	X18, [X10, X6,LSL #3]

  (160) 0x421600 CMP	X18, #0

  (160) 0x421604 B.LE	42162c

  (161) 0x421608 STR	X9, [X7]

  (161) 0x42160c ADD	X13, X13, #1

  (161) 0x421610 LDR	X30, [X5]

  (161) 0x421614 CMP	X13, X30

  (161) 0x421618 B.GE	421588

  (161) 0x42161c LDR	X6, [X11, X13,LSL #3]

  (161) 0x421620 LDR	X18, [X10, X6,LSL #3]

  (161) 0x421624 CMP	X18, #0

  (161) 0x421628 B.GT	421608

  (160) 0x42162c ADD	X13, X13, #1

  (160) 0x421630 CMP	X13, X30

  (160) 0x421634 B.LT	4215f8

0x421638 B	421588

  (162) 0x42163c ADD	X0, X0, #1

  (162) 0x421640 CMP	X0, X4

  (162) 0x421644 B.LT	4215c0

0x421648 B	4215e4

/home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/parcsr_ls/par_coarsen.c: 2446 - 2477

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

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

[...]

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

[...]

2465: 	 for (jS = S_diag_i[i]; jS < S_diag_i[i+1]; jS++)

2466:          {

2467: 	   /* j is the column number, or the local number of the point influencing i */

2468: 	   j = S_diag_j[jS];

2469: 	   if (CF_marker[j] > 0) /* j is a C-point */

2470: 	     CF_marker[i] = F_PT;

2471: 	 }

2472: 	 /* now the external part */

2473: 	 for (jS = S_offd_i[i]; jS < S_offd_i[i+1]; jS++)

2474: 	 {

2475: 	   j = S_offd_j[jS];

2476: 	   if (CF_marker_offd[j] > 0) /* j is a C-point */

2477: 	     CF_marker[i] = F_PT;

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	1.29
Bottlenecks	P0, P1,
Function	hypre_BoomerAMGCoarsenPMIS._omp_fn.7
Source	par_coarsen.c:2446-2446,par_coarsen.c:2465-2465,par_coarsen.c:2473-2473
Source loop unroll info	NA
Source loop unroll confidence level	NA
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	1.00
DIV/SQRT cycles	1.50
P0 cycles	1.50
P1 cycles	0.50
P2 cycles	0.50
P3 cycles	0.50
P4 cycles	0.50
P5 cycles	0.00
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	NA
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	8.00
Nb uops	8.00
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	16.00
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	NA
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	NA
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	1.00
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	1.29
Bottlenecks	P0, P1,
Function	hypre_BoomerAMGCoarsenPMIS._omp_fn.7
Source	par_coarsen.c:2446-2446,par_coarsen.c:2465-2465,par_coarsen.c:2473-2473
Source loop unroll info	NA
Source loop unroll confidence level	NA
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	1.00
DIV/SQRT cycles	1.50
P0 cycles	1.50
P1 cycles	0.50
P2 cycles	0.50
P3 cycles	0.50
P4 cycles	0.50
P5 cycles	0.00
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	NA
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	8.00
Nb uops	8.00
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	16.00
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	NA
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	NA
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_BoomerAMGCoarsenPMIS._omp_fn.7
Source file and lines	par_coarsen.c:2446-2477
Module	exec

The loop is defined in /home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/parcsr_ls/par_coarsen.c:2446,2453,2465-2477.

Analyzed code is defined in /home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/parcsr_ls/par_coarsen.c:2446,2465,2473.

Warnings:

Non-innermost loop: analyzing only self part (ignoring child loops).
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

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:
- recompile with fassociative-math (included in Ofast or ffast-math) to extend loop vectorization to FP reductions.
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	8
loop length	32
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

1.00 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.50	1.50	0.50	0.50	0.50	0.50	0.00	0.00	0.00	0.00	1.17	0.83	1.00	0.50	0.50
cycles	1.50	1.50	0.50	0.50	0.50	0.50	0.00	0.00	0.00	0.00	1.17	0.83	1.00	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	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	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 1.50 cycles. At this rate:

11% of peak load performance is reached (10.67 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: 4215a4

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P10	P11	P12	P13	P14	Latency	Recip. throughput
STR X9, [X3, X18,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
LDR X13, [X16, X18,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
ADD X5, X16, X20	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
LDR X30, [X16, X20]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
CMP X13, X30	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33
B.GE 421588 <hypre_BoomerAMGCoarsenPMIS._omp_fn.7+0x98>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50
B 421588 <hypre_BoomerAMGCoarsenPMIS._omp_fn.7+0x98>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50
B 4215e4 <hypre_BoomerAMGCoarsenPMIS._omp_fn.7+0xf4>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50

Function	hypre_BoomerAMGCoarsenPMIS._omp_fn.7
Source file and lines	par_coarsen.c:2446-2477
Module	exec

The loop is defined in /home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/parcsr_ls/par_coarsen.c:2446,2453,2465-2477.

Analyzed code is defined in /home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/parcsr_ls/par_coarsen.c:2446,2465,2473.

Warnings:

Non-innermost loop: analyzing only self part (ignoring child loops).
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

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:
- recompile with fassociative-math (included in Ofast or ffast-math) to extend loop vectorization to FP reductions.
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	8
loop length	32
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

1.00 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.50	1.50	0.50	0.50	0.50	0.50	0.00	0.00	0.00	0.00	1.17	0.83	1.00	0.50	0.50
cycles	1.50	1.50	0.50	0.50	0.50	0.50	0.00	0.00	0.00	0.00	1.17	0.83	1.00	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	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	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 1.50 cycles. At this rate:

11% of peak load performance is reached (10.67 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: 4215a4

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P10	P11	P12	P13	P14	Latency	Recip. throughput
STR X9, [X3, X18,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
LDR X13, [X16, X18,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
ADD X5, X16, X20	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25
LDR X30, [X16, X20]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
CMP X13, X30	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33
B.GE 421588 <hypre_BoomerAMGCoarsenPMIS._omp_fn.7+0x98>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50
B 421588 <hypre_BoomerAMGCoarsenPMIS._omp_fn.7+0x98>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50
B 4215e4 <hypre_BoomerAMGCoarsenPMIS._omp_fn.7+0xf4>	1	0.50	0.50	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

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