OV - Compare Loops

MAQAO

options

Loops

▶attention_v2.cpp: 30 - 173.93 %

ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s	ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s
Run gcc							Run armclang
Loop Source Regions	/home/eoseret/llm-attention/attention_v2.cpp: 30-31						Loop Source Regions	/home/eoseret/llm-attention/attention_v2.cpp: 30-31
4	0.61	0.61	6.99	0	30	8.76	68	0.56	0.56	6.91	0	30	8.83
7	0.59	0.58	6.76	0	30	8.68	43	3.00	3.00	36.74	0	31.82	6.69
10	0.65	0.64	7.45	0	30	8.09	62	0.56	0.55	6.78	0	30	9.09
14	2.57	2.57	29.68	0	30	8.07	65	0.53	0.53	6.48	0	30	9.43
17	3.13	3.13	36.09	0	30	6.56	54	2.46	2.46	30.07	0	31.82	7.63

Sum on 5 analyzed binary loops (attention-gcc-native - 4, attention-gcc-native - 7, attention-gcc-native - 10, attention-gcc-native - 14, attention-gcc-native - 17)							Sum on 5 analyzed binary loops (attention-armclang-native - 68, attention-armclang-native - 43, attention-armclang-native - 62, attention-armclang-native - 65, attention-armclang-native - 54)
Analysis						Count	Analysis						Count
Data Access Issues							Data Access Issues
Presence of constant non-unit stride data access						1	Presence of constant non-unit stride data access						1
Vectorization Roadblocks							Vectorization Roadblocks
Presence of constant non-unit stride data access						1	Presence of constant non-unit stride data access						1

▶random.tcc: 333 - 0.52 %

ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s	ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s
Run gcc							Run armclang
Loop Source Regions	/home/eoseret/llm-attention/attention_v2.cpp: 163-163 /usr/include/c++/14/bits/random.tcc: 333-333 /usr/include/c++/14/bits/random.tcc: 458-466 /usr/include/c++/14/bits/random.tcc: 3367-3371						Loop Source Regions
2	0.04	0.04	0.52	0	40.63	2.83

Sum on 1 analyzed binary loop (attention-gcc-native - 2)							No Loops Overview analysis found for any assembly loop. More loops can be analyzed using option --summary-loop-count.
Analysis						Count	Analysis						Count
Loop Computation Issues
Less than 10% of the FP ADD/SUB/MUL arithmetic operations are performed using FMA						1
Presence of a large number of scalar integer instructions						1
Control Flow Issues
Presence of 2 to 4 paths						1
Data Access Issues
Presence of indirect access						1
Vectorization Roadblocks
Presence of 2 to 4 paths						1
Presence of indirect access						1

▶random.tcc: 412 - 0.48 %

ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s	ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s
Run gcc							Run armclang
Loop Source Regions	/usr/include/c++/14/bits/random.tcc: 412-417						Loop Source Regions	/usr/lib/gcc/aarch64-amazon-linux/14/../../../../include/c++/14/bits/random.tcc: 412-417
42	0.01	0.02	0.17	100	100	0	72	0.00	0.00	0.06	91.67	95.83	0
							82	0.02	0.02	0.24	95.83	97.92	0

No Loops Overview analysis found for any assembly loop. More loops can be analyzed using option --summary-loop-count.							No Loops Overview analysis found for any assembly loop. More loops can be analyzed using option --summary-loop-count.
Analysis						Count	Analysis						Count

▶attention_v2.cpp: 52 - 0.48 %

ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s	ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s
Run gcc							Run armclang
Loop Source Regions	/home/eoseret/llm-attention/attention_v2.cpp: 52-53						Loop Source Regions	/home/eoseret/llm-attention/attention_v2.cpp: 52-53
30	0.02	0.02	0.23	0	25	6.25	49	0.01	0.01	0.12	75	87.5	2.5
							37	0.00	0.00	0.06	50	62.5	29.5
							36	0.00	0.00	0.06	100	100	25

No Loops Overview analysis found for any assembly loop. More loops can be analyzed using option --summary-loop-count.							No Loops Overview analysis found for any assembly loop. More loops can be analyzed using option --summary-loop-count.
Analysis						Count	Analysis						Count

▶attention_v2.cpp: 55 - 0.46 %

ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s	ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s
Run gcc							Run armclang
Loop Source Regions	/home/eoseret/llm-attention/attention_v2.cpp: 55-56						Loop Source Regions
31	0.04	0.04	0.46	0	25	4.38

Sum on 1 analyzed binary loop (attention-gcc-native - 31)							No Loops Overview analysis found for any assembly loop. More loops can be analyzed using option --summary-loop-count.
Analysis						Count	Analysis						Count
Loop Computation Issues
Presence of expensive FP instructions						1
Less than 10% of the FP ADD/SUB/MUL arithmetic operations are performed using FMA						1
Presence of a large number of scalar integer instructions						1
Control Flow Issues
Presence of calls						1
Vectorization Roadblocks
Presence of calls						1

▶random.tcc: 404 - 0.30 %

ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s	ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s
Run gcc							Run armclang
Loop Source Regions	/usr/include/c++/14/bits/random.tcc: 404-409						Loop Source Regions	/usr/lib/gcc/aarch64-amazon-linux/14/../../../../include/c++/14/bits/random.tcc: 404-409
41	0.01	0.01	0.12	100	100	0.38	81	0.01	0.01	0.18	97.87	98.94	0.25

No Loops Overview analysis found for any assembly loop. More loops can be analyzed using option --summary-loop-count.							No Loops Overview analysis found for any assembly loop. More loops can be analyzed using option --summary-loop-count.
Analysis						Count	Analysis						Count

▶attention_v2.cpp: 237 - 0.29 %

ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s	ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s
Run gcc							Run armclang
Loop Source Regions	/usr/include/c++/14/bits/stl_vector.h: 1128-1128 /home/eoseret/llm-attention/attention_v2.cpp: 237-238						Loop Source Regions
22	0.03	0.02	0.29	14.29	28.57	0

No Loops Overview analysis found for any assembly loop. More loops can be analyzed using option --summary-loop-count.							No Loops Overview analysis found for any assembly loop. More loops can be analyzed using option --summary-loop-count.
Analysis						Count	Analysis						Count

▶random.tcc: 458 - 0.29 %

ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s	ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s
Run gcc							Run armclang
Loop Source Regions	/home/eoseret/llm-attention/attention_v2.cpp: 164-167 /usr/include/c++/14/bits/random.tcc: 458-466 /usr/include/c++/14/bits/random.tcc: 3367-3374						Loop Source Regions
24	0.03	0.02	0.29	0	41.18	4.9

Sum on 1 analyzed binary loop (attention-gcc-native - 24)							No Loops Overview analysis found for any assembly loop. More loops can be analyzed using option --summary-loop-count.
Analysis						Count	Analysis						Count
Loop Computation Issues
Less than 10% of the FP ADD/SUB/MUL arithmetic operations are performed using FMA						1
Presence of a large number of scalar integer instructions						1
Control Flow Issues
Presence of calls						1
Presence of more than 4 paths						1
Vectorization Roadblocks
Presence of calls						1
Presence of more than 4 paths						1

▶attention_v2.cpp: 47 - 0.18 %

ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s	ASM Loop ID	Max Time Over Threads (s)	Time w.r.t. Wall Time (s)	Cov (%)	Vect. Ratio (%)	Vector Length Use (%)	GFLOP/s
Run gcc							Run armclang
Loop Source Regions	/home/eoseret/llm-attention/attention_v2.cpp: 47-48						Loop Source Regions	/home/eoseret/llm-attention/attention_v2.cpp: 47-48
29	0.00	0.00	0.06	100	100	2.25	51	0.00	0.00	0.06	100	100	2
32	0.00	0.00	0.06	100	100	4.25

No Loops Overview analysis found for any assembly loop. More loops can be analyzed using option --summary-loop-count.							No Loops Overview analysis found for any assembly loop. More loops can be analyzed using option --summary-loop-count.
Analysis						Count	Analysis						Count

×