OV - K-Means scalability acfl-Ofast 100000000 - Function k_means(int, point_t*, point_t*, int*, int, int) [clone .omp

	Function: k_means(int, point_t, point_t, int*, int, int) [clone .omp_outlined]	Module: kmeans-acfl-Ofast	Source: main.cpp:113-122	Coverage (incl. loops): 100.00% \| (excl. loops): 0.00%

/home/fmusial/KMEANS_Benchmarks/kmeans/main.cpp: 113 - 122

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

113:         #pragma omp parallel for

114:         for (int i = 0; i < n; ++i) {

115:             double optimal_dist = DBL_MAX;

116:             for (int j = 0; j < k; ++j) {

117:                 double dist =

118:                     (points[i].x - centroids[j].x) * (points[i].x - centroids[j].x) +

119:                     (points[i].y - centroids[j].y) * (points[i].y - centroids[j].y);

120:                 if (dist < optimal_dist) {

121:                     optimal_dist = dist;

122:                     assignment[i] = j;

0x1c20 SUB	SP, SP, #96

0x1c24 STP	X29, X30, [SP, #32]

0x1c28 STP	X24, X23, [SP, #48]

0x1c2c STP	X22, X21, [SP, #64]

0x1c30 STP	X20, X19, [SP, #80]

0x1c34 ADD	X29, SP, #32

0x1c38 LDR	W8, [X2]

0x1c3c SUBS	W24, W8, #1

0x1c40 B.LT	1d38

0x1c44 LDR	W21, [X0]

0x1c48 ORR	X19, XZR, X6

0x1c4c ORR	X22, XZR, X5

0x1c50 ORR	X23, XZR, X4

0x1c54 ORR	X20, XZR, X3

0x1c58 MOVZ	W8, #1

0x1c5c ADRP	X0,

0x1c60 ADD	X0, X0, #3240

0x1c64 ADD	X3, SP, #16

0x1c68 SUB	X4, X29, #4

0x1c6c SUB	X5, X29, #8

0x1c70 SUB	X6, X29, #12

0x1c74 ORR	W1, WZR, W21

0x1c78 MOVZ	W2, #34

0x1c7c MOVZ	W7, #1

0x1c80 STP	W24, WZR, [X29, #504]

0x1c84 STUR	W8, [X29, #500]

0x1c88 STR	WZR, [SP, #16]

0x1c8c STR	W8, [SP]

0x1c90 BL	1460

0x1c94 LDP	W9, W8, [X29, #504]

0x1c98 SBFM	X8, X8, #0, #31

0x1c9c CMP	W9, W24

0x1ca0 CSEL	W12, W9, W24, #11

0x1ca4 CMP	W8, W12

0x1ca8 B.GT	1d28

0x1cac LDR	W9, [X20]

0x1cb0 CMP	W9, #1

0x1cb4 B.LT	1d28

0x1cb8 LDR	X10, [X23]

0x1cbc LDR	X11, [X22]

0x1cc0 ADD	W12, W12, #1

0x1cc4 MOVN	X15, #32784

0x1cc8 B	1cd8

(6) 0x1ccc ADD	X8, X8, #1

(6) 0x1cd0 CMP	W12, W8

(6) 0x1cd4 B.EQ	1d28

(6) 0x1cd8 CMP	W9, #1

(6) 0x1cdc B.LT	1ccc

(6) 0x1ce0 FMOV	D1, X15

(6) 0x1ce4 LDR	Q0, [X10, X8,LSL #4]

(6) 0x1ce8 LDR	X14, [X19]

(6) 0x1cec ORR	X13, XZR, XZR

(6) 0x1cf0 B	1d00

  (7) 0x1cf4 ADD	X13, X13, #1

  (7) 0x1cf8 CMP	X13, W9,SXTW

  (7) 0x1cfc B.GE	1ccc

  (7) 0x1d00 LDR	Q2, [X11, X13,LSL #4]

  (7) 0x1d04 FSUB	V2.2D, V0.2D, V2.2D

  (7) 0x1d08 FMUL	V2.2D, V2.2D, V2.2D

  (7) 0x1d0c FADDP	D2, V2.2D

  (7) 0x1d10 FCMP	D2, D1

  (7) 0x1d14 B.GE	1cf4

  (7) 0x1d18 FMOV	D1, D2

  (7) 0x1d1c STR	W13, [X14, X8,LSL #2]

  (7) 0x1d20 LDR	W9, [X20]

  (7) 0x1d24 B	1cf4

0x1d28 ADRP	X0,

0x1d2c ADD	X0, X0, #3264

0x1d30 ORR	W1, WZR, W21

0x1d34 BL	12f0

0x1d38 LDP	X20, X19, [SP, #80]

0x1d3c LDP	X22, X21, [SP, #64]

0x1d40 LDP	X24, X23, [SP, #48]

0x1d44 LDP	X29, X30, [SP, #32]

0x1d48 ADD	SP, SP, #96

0x1d4c RET

Coverage (%)	Name	Module
○96.03	__kmp_invoke_microtask	libomp.so
►3.97+	__kmp_invoke_microtask	libomp.so
○	Unknown function	ld-linux-aarch64.so.1

Coverage (%)	Name	Source Location	Module
○100.00	__kmp_invoke_microtask		libomp.so

Coverage (%)	Name	Source Location	Module
○100.00	__kmp_invoke_microtask		libomp.so

Coverage (%)	Name	Source Location	Module
○100.00	__kmp_invoke_microtask		libomp.so

Coverage (%)	Name	Source Location	Module
○100.00	__kmp_invoke_microtask		libomp.so

Coverage (%)	Name	Source Location	Module
○100.00	__kmp_invoke_microtask		libomp.so

Coverage (%)	Name	Source Location	Module
○100.00	__kmp_invoke_microtask		libomp.so

min	med	avg	max

Percentile Index	10	20	30	40	50	60	70	80	90	100
Value

min	med	avg	max

Percentile Index	10	20	30	40	50	60	70	80	90	100
Value

Path /

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

The code analyzed by CQA in that panel excludes loops and represents 0.00% of application time for run run_1_thread

Source file and lines	main.cpp:113-122
Module	kmeans-acfl-Ofast

The function is defined in /home/fmusial/KMEANS_Benchmarks/kmeans/main.cpp:113-122.

Warnings:
Ignoring paths for analysis

gain
potential
hint
expert

Vectorization

Your function is not vectorized. Only 24% of vector register length is used (average across all VPU instructions). By vectorizing your function, you can lower the cost of an iteration from 7.50 to 4.81 cycles (1.56x 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 7.50 to 1.00 cycles (7.50x speedup).

CALL instructions

Detected function call instructions.

Details

Calling (and then returning from) a function prevents many compiler optimizations (like vectorization), breaks control flow (which reduces pipeline performance) and executes extra instructions to save/restore the registers used inside it, which is very expensive (dozens of cycles). Consider to inline small functions.

unknown: 2 occurrences►
- /home/fmusial/KMEANS_Benchmarks/kmeans/main.cpp:113
- /home/fmusial/KMEANS_Benchmarks/kmeans/main.cpp:113

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 100 bytes. The binary function is storing 84 bytes.

General properties

nb instructions	53
nb uops	53
loop length	212
used w registers	9
used x registers	19
used b registers	0
used h registers	0
used s registers	1
used d registers	0
used q registers	0
used v registers	0
used z registers	0
nb stack references	10

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	6.63 cycles
front end	6.63 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	3.50	3.50	7.50	7.50	7.50	7.50	0.00	0.00	0.00	0.00	6.00	6.00	6.00	4.00	4.00
cycles	3.50	3.50	7.50	7.50	7.50	7.50	0.00	0.00	0.00	0.00	6.00	6.00	6.00	4.00	4.00

Execution ports to units layout:

P0: BRU
P1: BRU
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	6.63
Dispatch	7.50
Overall L1	7.50

Vectorization ratios

all	0%
load	0%
store	0%
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%

Vector efficiency ratios

all	24%
load	18%
store	32%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	20%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	22%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each call to the function takes 7.50 cycles. At this rate:

13% of peak load performance is reached (13.33 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
23% of peak store performance is reached (11.20 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the function is: 1c20

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P10	P11	P12	P13	P14	Latency	Recip. throughput	Vectorization
SUB SP, SP, #96	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
STP X29, X30, [SP, #32]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (50.0%)
STP X24, X23, [SP, #48]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (50.0%)
STP X22, X21, [SP, #64]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (50.0%)
STP X20, X19, [SP, #80]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (50.0%)
ADD X29, SP, #32	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
LDR W8, [X2]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	scal (12.5%)
SUBS W24, W8, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33	scal (12.5%)
B.LT 1d38 <_Z7k_meansiP7point_tS0_Piii.omp_outlined+0x118>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A
LDR W21, [X0]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	scal (12.5%)
ORR X19, XZR, X6	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (25.0%)
ORR X22, XZR, X5	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
ORR X23, XZR, X4	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
ORR X20, XZR, X3	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
MOVZ W8, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (12.5%)
ADRP X0, <1fc5c>	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
ADD X0, X0, #3240	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
ADD X3, SP, #16	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (25.0%)
SUB X4, X29, #4	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (25.0%)
SUB X5, X29, #8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (25.0%)
SUB X6, X29, #12	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (25.0%)
ORR W1, WZR, W21	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (12.5%)
MOVZ W2, #34	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
MOVZ W7, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (12.5%)
STP W24, WZR, [X29, #504]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (25.0%)
STUR W8, [X29, #500]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (12.5%)
STR WZR, [SP, #16]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (12.5%)
STR W8, [SP]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (12.5%)
BL 1460 <@plt_start@+0x1a0>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	1	0.50	N/A
LDP W9, W8, [X29, #504]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	scal (25.0%)
SBFM X8, X8, #0, #31	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (100.0%)
CMP W9, W24	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33	scal (12.5%)
CSEL W12, W9, W24, #11	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (12.5%)
CMP W8, W12	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33	scal (12.5%)
B.GT 1d28 <_Z7k_meansiP7point_tS0_Piii.omp_outlined+0x108>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A
LDR W9, [X20]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	scal (12.5%)
CMP W9, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33	scal (12.5%)
B.LT 1d28 <_Z7k_meansiP7point_tS0_Piii.omp_outlined+0x108>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A
LDR X10, [X23]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	scal (25.0%)
LDR X11, [X22]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	scal (25.0%)
ADD W12, W12, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (12.5%)
MOVN X15, #32784	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (25.0%)
B 1cd8 <_Z7k_meansiP7point_tS0_Piii.omp_outlined+0xb8>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A
ADRP X0, <1fd28>	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
ADD X0, X0, #3264	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
ORR W1, WZR, W21	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (12.5%)
BL 12f0 <@plt_start@+0x30>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	1	0.50	N/A
LDP X20, X19, [SP, #80]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1	N/A
LDP X22, X21, [SP, #64]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1	N/A
LDP X24, X23, [SP, #48]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1	N/A
LDP X29, X30, [SP, #32]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1	N/A
ADD SP, SP, #96	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
RET	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A

The code analyzed by CQA in that panel excludes loops and represents 0.00% of application time for run run_1_thread

Source file and lines	main.cpp:113-122
Module	kmeans-acfl-Ofast

The function is defined in /home/fmusial/KMEANS_Benchmarks/kmeans/main.cpp:113-122.

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 7.50 to 1.00 cycles (7.50x speedup).

CALL instructions

Detected function call instructions.

Details

unknown: 2 occurrences►
- /home/fmusial/KMEANS_Benchmarks/kmeans/main.cpp:113
- /home/fmusial/KMEANS_Benchmarks/kmeans/main.cpp:113

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 100 bytes. The binary function is storing 84 bytes.

General properties

nb instructions	53
nb uops	53
loop length	212
used w registers	9
used x registers	19
used b registers	0
used h registers	0
used s registers	1
used d registers	0
used q registers	0
used v registers	0
used z registers	0
nb stack references	10

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	6.63 cycles
front end	6.63 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	3.50	3.50	7.50	7.50	7.50	7.50	0.00	0.00	0.00	0.00	6.00	6.00	6.00	4.00	4.00
cycles	3.50	3.50	7.50	7.50	7.50	7.50	0.00	0.00	0.00	0.00	6.00	6.00	6.00	4.00	4.00

Execution ports to units layout:

P0: BRU
P1: BRU
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	6.63
Dispatch	7.50
Overall L1	7.50

Vectorization ratios

all	0%
load	0%
store	0%
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%

Vector efficiency ratios

all	24%
load	18%
store	32%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	20%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	22%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each call to the function takes 7.50 cycles. At this rate:

13% of peak load performance is reached (13.33 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
23% of peak store performance is reached (11.20 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the function is: 1c20

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P10	P11	P12	P13	P14	Latency	Recip. throughput	Vectorization
SUB SP, SP, #96	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
STP X29, X30, [SP, #32]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (50.0%)
STP X24, X23, [SP, #48]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (50.0%)
STP X22, X21, [SP, #64]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (50.0%)
STP X20, X19, [SP, #80]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (50.0%)
ADD X29, SP, #32	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
LDR W8, [X2]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	scal (12.5%)
SUBS W24, W8, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33	scal (12.5%)
B.LT 1d38 <_Z7k_meansiP7point_tS0_Piii.omp_outlined+0x118>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A
LDR W21, [X0]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	scal (12.5%)
ORR X19, XZR, X6	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (25.0%)
ORR X22, XZR, X5	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
ORR X23, XZR, X4	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
ORR X20, XZR, X3	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
MOVZ W8, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (12.5%)
ADRP X0, <1fc5c>	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
ADD X0, X0, #3240	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
ADD X3, SP, #16	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (25.0%)
SUB X4, X29, #4	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (25.0%)
SUB X5, X29, #8	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (25.0%)
SUB X6, X29, #12	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (25.0%)
ORR W1, WZR, W21	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (12.5%)
MOVZ W2, #34	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
MOVZ W7, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (12.5%)
STP W24, WZR, [X29, #504]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (25.0%)
STUR W8, [X29, #500]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (12.5%)
STR WZR, [SP, #16]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (12.5%)
STR W8, [SP]	1	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (12.5%)
BL 1460 <@plt_start@+0x1a0>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	1	0.50	N/A
LDP W9, W8, [X29, #504]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	scal (25.0%)
SBFM X8, X8, #0, #31	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (100.0%)
CMP W9, W24	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33	scal (12.5%)
CSEL W12, W9, W24, #11	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (12.5%)
CMP W8, W12	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33	scal (12.5%)
B.GT 1d28 <_Z7k_meansiP7point_tS0_Piii.omp_outlined+0x108>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A
LDR W9, [X20]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	scal (12.5%)
CMP W9, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.33	scal (12.5%)
B.LT 1d28 <_Z7k_meansiP7point_tS0_Piii.omp_outlined+0x108>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A
LDR X10, [X23]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	scal (25.0%)
LDR X11, [X22]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	scal (25.0%)
ADD W12, W12, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (12.5%)
MOVN X15, #32784	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (25.0%)
B 1cd8 <_Z7k_meansiP7point_tS0_Piii.omp_outlined+0xb8>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A
ADRP X0, <1fd28>	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
ADD X0, X0, #3264	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
ORR W1, WZR, W21	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	scal (12.5%)
BL 12f0 <@plt_start@+0x30>	1	0.50	0.50	0.50	0.50	0	0	0	0	0	0	0	1	0.50	N/A
LDP X20, X19, [SP, #80]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1	N/A
LDP X22, X21, [SP, #64]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1	N/A
LDP X24, X23, [SP, #48]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1	N/A
LDP X29, X30, [SP, #32]	1	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	1	N/A
ADD SP, SP, #96	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	1	0.25	N/A
RET	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A

▶Scalability Runs Description

Run run_1_thread	Number processes: 1Number nodes: 1Run Command: <executable> kmeans/input/100000000.in 1000 100000000 50 25MPI Command: Dataset: Run Directory: /home/fmusial/KMEANS_BenchmarksOMP_PROC_BIND: closeOMP_NUM_THREADS: 1OMP_PLACES: cores
Run run_2_threads	Number processes: 1Number nodes: 1Run Command: <executable> kmeans/input/100000000.in 1000 100000000 50 25MPI Command: Dataset: Run Directory: /home/fmusial/KMEANS_BenchmarksOMP_PLACES: coresOMP_PROC_BIND: closeOMP_NUM_THREADS: 2
Run run_4_threads	Number processes: 1Number nodes: 1Run Command: <executable> kmeans/input/100000000.in 1000 100000000 50 25MPI Command: Dataset: Run Directory: /home/fmusial/KMEANS_BenchmarksOMP_PLACES: coresOMP_PROC_BIND: closeOMP_NUM_THREADS: 4
Run run_8_threads	Number processes: 1Number nodes: 1Run Command: <executable> kmeans/input/100000000.in 1000 100000000 50 25MPI Command: Dataset: Run Directory: /home/fmusial/KMEANS_BenchmarksOMP_PLACES: coresOMP_PROC_BIND: closeOMP_NUM_THREADS: 8
Run run_16_threads	Number processes: 1Number nodes: 1Run Command: <executable> kmeans/input/100000000.in 1000 100000000 50 25MPI Command: Dataset: Run Directory: /home/fmusial/KMEANS_BenchmarksOMP_PLACES: coresOMP_PROC_BIND: closeOMP_NUM_THREADS: 16
Run run_32_threads	Number processes: 1Number nodes: 1Run Command: <executable> kmeans/input/100000000.in 1000 100000000 50 25MPI Command: Dataset: Run Directory: /home/fmusial/KMEANS_BenchmarksOMP_PROC_BIND: closeOMP_PLACES: coresOMP_NUM_THREADS: 32
Run run_64_threads	Number processes: 1Number nodes: 1Run Command: <executable> kmeans/input/100000000.in 1000 100000000 50 25MPI Command: Dataset: Run Directory: /home/fmusial/KMEANS_BenchmarksOMP_PROC_BIND: closeOMP_PLACES: coresOMP_NUM_THREADS: 64

(run_1_thread) Efficiency	(run_1_thread) Potential Speed-Up (%)	(run_2_threads) Efficiency	(run_2_threads) Potential Speed-Up (%)	(run_4_threads) Efficiency	(run_4_threads) Potential Speed-Up (%)	(run_8_threads) Efficiency	(run_8_threads) Potential Speed-Up (%)	(run_16_threads) Efficiency	(run_16_threads) Potential Speed-Up (%)	(run_32_threads) Efficiency	(run_32_threads) Potential Speed-Up (%)	(run_64_threads) Efficiency	(run_64_threads) Potential Speed-Up (%)
1	0	1	0.11	0.99	0.66	0.99	0.52	0.99	0.48	1.02	0	2.06	0

Speed-Up and Efficiency

▶Detailed Speed-Up and Efficiency

Run	Number of threads	Efficiency (ideal is 1)	Speedup	Ideal Speedup	Time (s)	Coverage (%)
run_1_thread	1	1	1	1	121.62497711182	99.995895385742
run_2_threads	2	1	2	2	61.504989624023	98.940155029297
run_4_threads	4	0.99	3.97	4	31.490001678467	97.068244934082
run_8_threads	8	0.99	7.96	8	15.800000190735	95.803146362305
run_16_threads	16	0.99	15.92	16	8.0499982833862	94.265769958496
run_32_threads	32	1.02	32.79	32	3.9849996566772	87.882949829102
run_64_threads	64	2.06	132	64	1.3700000047684	34.190166473389

Name	Coverage (%)	Time (s)
▼k_means(int, point_t, point_t, int*, int, int) [clone .omp_outlined]–	100.00	121.62
▼Loop 6 - main.cpp:114-122 - kmeans-acfl-Ofast–	0.94	1.14
○Loop 7 - main.cpp:116-122 - kmeans-acfl-Ofast	99.06	120.48

Report Configuration

Vectorization

Details

Workaround

Expensive FP math instructions/calls

CALL instructions

Details

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code

Vectorization

Details

Workaround

Expensive FP math instructions/calls

CALL instructions

Details

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code

▶Scalability Runs Description

▼Columns Filter

Speed-Up and Efficiency

▶Detailed Speed-Up and Efficiency