OV - exec - Function .omp_outlined..15#0x491c00

	Function: .omp_outlined..15#0x491c00	Module: exec	Source: IJMatrix_parcsr.c:2798-2812	Coverage: 0.4%

/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;

0x491c00 SUB	SP, SP, #112

0x491c04 STP	X29, X30, [SP, #48]

0x491c08 STP	X24, X23, [SP, #64]

0x491c0c STP	X22, X21, [SP, #80]

0x491c10 STP	X20, X19, [SP, #96]

0x491c14 ADD	X29, SP, #48

0x491c18 LDR	X8, [X2]

0x491c1c SUBS	X24, X8, #1

0x491c20 B.LT	491c98

0x491c24 LDR	W20, [X0]

0x491c28 LDR	X21, [X3]

0x491c2c LDR	X22, [X4]

0x491c30 ORR	X19, XZR, X5

0x491c34 MOVZ	W8, #1

0x491c38 ADRP	X0,

0x491c3c ADD	X0, X0, #3536

0x491c40 ADD	X3, SP, #20

0x491c44 LDR	X23, [X6]

0x491c48 SUB	X4, X29, #8

0x491c4c SUB	X5, X29, #16

0x491c50 ADD	X6, SP, #24

0x491c54 STP	X24, XZR, [X29, #1008]

0x491c58 STR	X8, [SP, #24]

0x491c5c STR	WZR, [SP, #20]

0x491c60 MOVZ	W2, #34

0x491c64 MOVZ	W7, #1

0x491c68 STR	X8, [SP]

0x491c6c ORR	W1, WZR, W20

0x491c70 BL	40fa80

0x491c74 LDP	X8, X9, [X29, #1008]

0x491c78 CMP	X8, X24

0x491c7c CSEL	X8, X8, X24, #11

0x491c80 CMP	X9, X8

0x491c84 B.LE	491cb0

(463) 0x491c88 ORR	W1, WZR, W20

(463) 0x491c8c ADRP	X0,

(463) 0x491c90 ADD	X0, X0, #3560

(463) 0x491c94 BL	40f990

(463) 0x491c98 LDP	X20, X19, [SP, #96]

(463) 0x491c9c LDP	X22, X21, [SP, #80]

(463) 0x491ca0 LDP	X24, X23, [SP, #64]

(463) 0x491ca4 LDP	X29, X30, [SP, #48]

(463) 0x491ca8 ADD	SP, SP, #112

(463) 0x491cac RET

(463) 0x491cb0 LDR	X10, [X21, X9,LSL #3]

(463) 0x491cb4 B	491ccc

0x491cb8 HINT	#0

0x491cbc HINT	#0

(463) 0x491cc0 ORR	X10, XZR, X13

(463) 0x491cc4 CMP	X11, X8

(463) 0x491cc8 B.EQ	491c88

(463) 0x491ccc ORR	X11, XZR, X9

(463) 0x491cd0 ADD	X9, X9, #1

(463) 0x491cd4 LDR	X13, [X21, X9,LSL #3]

(463) 0x491cd8 CMP	X10, X13

(463) 0x491cdc B.GE	491cc0

(463) 0x491ce0 ORR	X12, XZR, X10

(463) 0x491ce4 B	491cf8

  (464) 0x491ce8 LDR	X13, [X21, X9,LSL #3]

  (464) 0x491cec ADD	X12, X12, #1

  (464) 0x491cf0 CMP	X12, X13

  (464) 0x491cf4 B.GE	491cc0

  (464) 0x491cf8 LDR	X13, [X19]

  (464) 0x491cfc LDR	X14, [X22, X12,LSL #3]

  (464) 0x491d00 SUB	X13, X14, X13

  (464) 0x491d04 CMP	X13, X11

  (464) 0x491d08 STR	X13, [X22, X12,LSL #3]

  (464) 0x491d0c B.NE	491ce8

  (464) 0x491d10 LDR	D0, [X23, X10,LSL #3]

  (464) 0x491d14 LDR	D1, [X23, X12,LSL #3]

  (464) 0x491d18 STR	D1, [X23, X10,LSL #3]

  (464) 0x491d1c STR	D0, [X23, X12,LSL #3]

  (464) 0x491d20 LDR	X13, [X22, X10,LSL #3]

  (464) 0x491d24 STR	X13, [X22, X12,LSL #3]

  (464) 0x491d28 STR	X11, [X22, X10,LSL #3]

  (464) 0x491d2c B	491ce8

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: 491c00

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 491c98 <.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, <4dec38>	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, #3536	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 40fa80 <@plt_start@+0x620>	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 491cb0 <.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: 491c00

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 491c98 <.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, <4dec38>	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, #3536	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 40fa80 <@plt_start@+0x620>	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 491cb0 <.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#0x491c00–	0.4	0.15
▼Loop 463 - IJMatrix_parcsr.c:2798-2812 - exec–	0	0
○Loop 464 - IJMatrix_parcsr.c:2803-2812 - exec	0.4	0.13

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