OV - exec - Function .omp_outlined..15#0x49c320

	Function: .omp_outlined..15#0x49c320	Module: exec	Source: IJMatrix_parcsr.c:2798-2812	Coverage: 0.36%

/home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/IJ_mv/IJMatrix_parcsr.c: 2798 - 2812

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

2798: #pragma omp parallel for private (i,j,j0,temp)

2799: #endif

2800:          for (i=0; i < num_rows; i++)

2801:          {

2802:    	    j0 = diag_i[i];

2803:    	    for (j=j0; j < diag_i[i+1]; j++)

2804:    	    {

2805:    	       diag_j[j] -= col_0;

2806:    	       if (diag_j[j] == i)

2807:    	       {

2808:    	          temp = diag_data[j0];

2809:    	          diag_data[j0] = diag_data[j];

2810:    	          diag_data[j] = temp;

2811:    	          diag_j[j] = diag_j[j0];

2812:    	          diag_j[j0] = i;

0x49c320 SUB	SP, SP, #112

0x49c324 STP	X29, X30, [SP, #48]

0x49c328 STP	X24, X23, [SP, #64]

0x49c32c STP	X22, X21, [SP, #80]

0x49c330 STP	X20, X19, [SP, #96]

0x49c334 ADD	X29, SP, #48

0x49c338 LDR	X8, [X2]

0x49c33c SUBS	X24, X8, #1

0x49c340 B.LT	49c3b8

0x49c344 LDR	W20, [X0]

0x49c348 LDR	X21, [X3]

0x49c34c LDR	X22, [X4]

0x49c350 ORR	X19, XZR, X5

0x49c354 MOVZ	W8, #1

0x49c358 ADRP	X0,

0x49c35c ADD	X0, X0, #3568

0x49c360 ADD	X3, SP, #20

0x49c364 LDR	X23, [X6]

0x49c368 SUB	X4, X29, #8

0x49c36c SUB	X5, X29, #16

0x49c370 ADD	X6, SP, #24

0x49c374 STP	X24, XZR, [X29, #1008]

0x49c378 STR	X8, [SP, #24]

0x49c37c STR	WZR, [SP, #20]

0x49c380 MOVZ	W2, #34

0x49c384 MOVZ	W7, #1

0x49c388 STR	X8, [SP]

0x49c38c ORR	W1, WZR, W20

0x49c390 BL	40f550

0x49c394 LDP	X8, X9, [X29, #1008]

0x49c398 CMP	X8, X24

0x49c39c CSEL	X8, X8, X24, #11

0x49c3a0 CMP	X9, X8

0x49c3a4 B.LE	49c3d0

(491) 0x49c3a8 ORR	W1, WZR, W20

(491) 0x49c3ac ADRP	X0,

(491) 0x49c3b0 ADD	X0, X0, #3592

(491) 0x49c3b4 BL	40f410

(491) 0x49c3b8 LDP	X20, X19, [SP, #96]

(491) 0x49c3bc LDP	X22, X21, [SP, #80]

(491) 0x49c3c0 LDP	X24, X23, [SP, #64]

(491) 0x49c3c4 LDP	X29, X30, [SP, #48]

(491) 0x49c3c8 ADD	SP, SP, #112

(491) 0x49c3cc RET

(491) 0x49c3d0 LDR	X10, [X21, X9,LSL #3]

(491) 0x49c3d4 B	49c3ec

0x49c3d8 HINT	#0

0x49c3dc HINT	#0

(491) 0x49c3e0 ORR	X10, XZR, X13

(491) 0x49c3e4 CMP	X11, X8

(491) 0x49c3e8 B.EQ	49c3a8

(491) 0x49c3ec ORR	X11, XZR, X9

(491) 0x49c3f0 ADD	X9, X9, #1

(491) 0x49c3f4 LDR	X13, [X21, X9,LSL #3]

(491) 0x49c3f8 CMP	X10, X13

(491) 0x49c3fc B.GE	49c3e0

(491) 0x49c400 ORR	X12, XZR, X10

(491) 0x49c404 B	49c418

  (492) 0x49c408 LDR	X13, [X21, X9,LSL #3]

  (492) 0x49c40c ADD	X12, X12, #1

  (492) 0x49c410 CMP	X12, X13

  (492) 0x49c414 B.GE	49c3e0

  (492) 0x49c418 LDR	X13, [X19]

  (492) 0x49c41c LDR	X14, [X22, X12,LSL #3]

  (492) 0x49c420 SUB	X13, X14, X13

  (492) 0x49c424 CMP	X13, X11

  (492) 0x49c428 STR	X13, [X22, X12,LSL #3]

  (492) 0x49c42c B.NE	49c408

  (492) 0x49c430 LDR	D0, [X23, X10,LSL #3]

  (492) 0x49c434 LDR	D1, [X23, X12,LSL #3]

  (492) 0x49c438 STR	D1, [X23, X10,LSL #3]

  (492) 0x49c43c STR	D0, [X23, X12,LSL #3]

  (492) 0x49c440 LDR	X13, [X22, X10,LSL #3]

  (492) 0x49c444 STR	X13, [X22, X12,LSL #3]

  (492) 0x49c448 STR	X11, [X22, X10,LSL #3]

  (492) 0x49c44c B	49c408

Coverage (%)	Name	Source Location	Module

Path /

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

Source file and lines	IJMatrix_parcsr.c:2798-2812
Module	exec

The function is defined in /home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/IJ_mv/IJMatrix_parcsr.c:2798-2812.

Warnings:
Ignoring paths for analysis

gain
potential
hint
expert

Vectorization

Your function is not vectorized. Only 20% of vector register length is used (average across all VPU instructions). By vectorizing your function, you can lower the cost of an iteration from 4.67 to 0.69 cycles (6.79x 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 function can use their full power.

Workaround

Try another compiler or update/tune your current one
Make array accesses unit-stride:
- If your function 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 4.67 to 0.50 cycles (9.33x speedup).

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

The binary function does not contain any FP arithmetical operations. The binary function is loading 52 bytes. The binary function is storing 100 bytes.

General properties

nb instructions	36
loop length	144
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

4.25 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.50	1.50	4.50	4.50	4.50	4.50	0.00	0.00	0.00	0.00	4.67	4.67	4.67	4.00	4.00
cycles	1.50	1.50	4.50	4.50	4.50	4.50	0.00	0.00	0.00	0.00	4.67	4.67	4.67	4.00	4.00

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

Cycles summary

Front-end	4.25
Overall L1	4.67

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	0%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each call to the function takes 4.67 cycles. At this rate:

11% of peak load performance is reached (11.14 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
44% of peak store performance is reached (21.43 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the function is: 49c320

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	Latency	Recip. throughput
SUB SP, SP, #112	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
STP X29, X30, [SP, #48]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STP X24, X23, [SP, #64]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STP X22, X21, [SP, #80]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STP X20, X19, [SP, #96]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
ADD X29, SP, #48	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
LDR X8, [X2]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
SUBS X24, X8, #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 49c3b8 <.omp_outlined..15+0x98>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
LDR W20, [X0]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
LDR X21, [X3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
LDR X22, [X4]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
ORR X19, XZR, X5	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
MOVZ W8, #1	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, <4ee358>	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, #3568	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ADD X3, SP, #20	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
LDR X23, [X6]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
SUB X4, X29, #8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
SUB X5, X29, #16	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ADD X6, SP, #24	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
STP X24, XZR, [X29, #1008]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STR X8, [SP, #24]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STR WZR, [SP, #20]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
MOVZ W2, #34	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
MOVZ W7, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
STR X8, [SP]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
ORR W1, WZR, W20	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
BL 40f550 <@plt_start@+0x160>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50
LDP X8, X9, [X29, #1008]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
CMP X8, X24	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.33
CSEL X8, X8, X24, #11	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
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
B.LE 49c3d0 <.omp_outlined..15+0xb0>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
HINT #0
HINT #0

Source file and lines	IJMatrix_parcsr.c:2798-2812
Module	exec

The function is defined in /home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/IJ_mv/IJMatrix_parcsr.c:2798-2812.

Warnings:
Ignoring paths for analysis

gain
potential
hint
expert

Vectorization

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 function can use their full power.

Workaround

Try another compiler or update/tune your current one
Make array accesses unit-stride:
- If your function 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 4.67 to 0.50 cycles (9.33x speedup).

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

The binary function does not contain any FP arithmetical operations. The binary function is loading 52 bytes. The binary function is storing 100 bytes.

General properties

nb instructions	36
loop length	144
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

4.25 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.50	1.50	4.50	4.50	4.50	4.50	0.00	0.00	0.00	0.00	4.67	4.67	4.67	4.00	4.00
cycles	1.50	1.50	4.50	4.50	4.50	4.50	0.00	0.00	0.00	0.00	4.67	4.67	4.67	4.00	4.00

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

Cycles summary

Front-end	4.25
Overall L1	4.67

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	0%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each call to the function takes 4.67 cycles. At this rate:

11% of peak load performance is reached (11.14 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
44% of peak store performance is reached (21.43 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the function is: 49c320

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	Latency	Recip. throughput
SUB SP, SP, #112	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
STP X29, X30, [SP, #48]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STP X24, X23, [SP, #64]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STP X22, X21, [SP, #80]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STP X20, X19, [SP, #96]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
ADD X29, SP, #48	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
LDR X8, [X2]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
SUBS X24, X8, #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 49c3b8 <.omp_outlined..15+0x98>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
LDR W20, [X0]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
LDR X21, [X3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
LDR X22, [X4]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
ORR X19, XZR, X5	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
MOVZ W8, #1	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, <4ee358>	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, #3568	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ADD X3, SP, #20	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
LDR X23, [X6]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33
SUB X4, X29, #8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
SUB X5, X29, #16	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
ADD X6, SP, #24	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
STP X24, XZR, [X29, #1008]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STR X8, [SP, #24]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
STR WZR, [SP, #20]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
MOVZ W2, #34	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
MOVZ W7, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
STR X8, [SP]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50
ORR W1, WZR, W20	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
BL 40f550 <@plt_start@+0x160>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50
LDP X8, X9, [X29, #1008]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1
CMP X8, X24	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.33
CSEL X8, X8, X24, #11	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25
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
B.LE 49c3d0 <.omp_outlined..15+0xb0>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50
HINT #0
HINT #0

Name	Coverage (%)	Time (s)
▼.omp_outlined..15#0x49c320–	0.36	0.14
▼Loop 491 - IJMatrix_parcsr.c:2798-2812 - exec–	0	0.01
○Loop 492 - IJMatrix_parcsr.c:2803-2812 - exec	0.36	0.12

Report Configuration

Vectorization

Details

Workaround

Expensive FP math instructions/calls

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code

Vectorization

Details

Workaround

Expensive FP math instructions/calls

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code