Icestorm: SMLAL2 (by element, 4S) measurements

Code:

  smlal2 v0.4s, v1.8h, v2.h[1]

Initialization

  movi v0.16b, 1
  movi v1.16b, 2
  movi v2.16b, 3

(no loop instructions)

1000 unrolls and 1 iteration

Retires: 1.000

Issues: 1.000

Integer unit issues: 0.001

Load/store unit issues: 0.000

SIMD/FP unit issues: 1.000

Recorded non-zero counters (first 10 of 64 runs)

retire uop (01)	cycle (02)	schedule uop (52)	schedule int uop (53)	schedule simd uop (54)	dispatch simd uop (57)	ldst uops in schedulers (5b)	dispatch uop (78)	map simd uop (7e)	map simd uop inputs (81)	? int output thing (e9)	? simd retires (ee)
1004	3033	1001	1	1000	1000	75905	1000	1000	3000	1	1000
1004	3033	1001	1	1000	1000	75905	1000	1000	3000	1	1000
1004	3033	1001	1	1000	1000	75905	1000	1000	3000	1	1000
1004	3033	1001	1	1000	1000	75905	1000	1000	3000	1	1000
1004	3033	1001	1	1000	1000	75905	1000	1000	3000	1	1000
1004	3033	1001	1	1000	1000	75905	1000	1000	3000	1	1000
1004	3033	1001	1	1000	1000	75905	1000	1000	3000	1	1000
1004	3033	1001	1	1000	1000	75905	1000	1000	3000	1	1000
1004	3033	1001	1	1000	1000	75905	1000	1000	3000	1	1000
1004	3033	1001	1	1000	1000	75905	1000	1000	3000	1	1000

Test 2: Latency 1->1

Code:

  smlal2 v0.4s, v1.8h, v2.h[1]

Initialization

  movi v0.16b, 1
  movi v1.16b, 2
  movi v2.16b, 3

(fused SUBS/B.cc loop)

100 unrolls and 100 iterations

Result (median cycles for code): 3.0033

Recorded non-zero counters (first 10 of 64 runs)

retire uop (01)	cycle (02)	schedule uop (52)	schedule int uop (53)	schedule simd uop (54)	dispatch int uop (56)	dispatch simd uop (57)	int uops in schedulers (59)	ldst uops in schedulers (5b)	dispatch uop (78)	map int uop (7c)	map simd uop (7e)	map int uop inputs (7f)	map simd uop inputs (81)	? int output thing (e9)	? simd retires (ee)	? int retires (ef)
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100

1000 unrolls and 10 iterations

Result (median cycles for code): 3.0033

Recorded non-zero counters (first 10 of 64 runs)

retire uop (01)	cycle (02)	schedule uop (52)	schedule int uop (53)	schedule simd uop (54)	dispatch int uop (56)	dispatch simd uop (57)	int uops in schedulers (59)	ldst uops in schedulers (5b)	dispatch uop (78)	map int uop (7c)	map simd uop (7e)	map int uop inputs (7f)	map ldst uop inputs (80)	map simd uop inputs (81)	? int output thing (e9)	? ldst retires (ed)	? simd retires (ee)	? int retires (ef)
10024	30188	10045	21	10024	20	10099	70	770543	10119	20	10122	20	0	30354	11	0	10000	10
10024	30187	10045	21	10024	20	10099	70	770528	10119	20	10117	20	0	30351	11	0	10000	10
10024	30184	10045	21	10024	20	10099	70	771089	10152	20	10160	20	0	30483	11	0	10000	10
10024	30187	10045	21	10024	20	10099	69	770543	10119	20	10117	20	0	30000	11	0	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	0	30000	11	0	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	0	30000	11	0	10000	10
10024	30033	10021	21	10000	20	10000	70	769451	10053	20	10040	20	0	30000	11	0	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	0	30000	11	0	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	0	30000	11	0	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	0	30000	11	0	10000	10

Test 3: Latency 1->2

Code:

  smlal2 v0.4s, v0.8h, v1.h[1]

Initialization

  movi v0.16b, 1
  movi v1.16b, 2

(fused SUBS/B.cc loop)

100 unrolls and 100 iterations

Result (median cycles for code): 3.0033

Recorded non-zero counters (first 10 of 64 runs)

retire uop (01)	cycle (02)	schedule uop (52)	schedule int uop (53)	schedule simd uop (54)	dispatch int uop (56)	dispatch simd uop (57)	int uops in schedulers (59)	ldst uops in schedulers (5b)	dispatch uop (78)	map int uop (7c)	map simd uop (7e)	map int uop inputs (7f)	map simd uop inputs (81)	? int output thing (e9)	? simd retires (ee)	? int retires (ef)
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10006	200	30018	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	307	769247	10133	202	10046	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100

1000 unrolls and 10 iterations

Result (median cycles for code): 3.0033

Recorded non-zero counters (first 10 of 64 runs)

retire uop (01)	cycle (02)	schedule uop (52)	schedule int uop (53)	schedule simd uop (54)	dispatch int uop (56)	dispatch simd uop (57)	int uops in schedulers (59)	ldst uops in schedulers (5b)	dispatch uop (78)	map int uop (7c)	map simd uop (7e)	map int uop inputs (7f)	map simd uop inputs (81)	? int output thing (e9)	? simd retires (ee)	? int retires (ef)
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10006	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10

Test 4: Latency 1->3

Code:

  smlal2 v0.4s, v1.8h, v0.h[1]

Initialization

  movi v0.16b, 1
  movi v1.16b, 2

(fused SUBS/B.cc loop)

100 unrolls and 100 iterations

Result (median cycles for code): 3.0033

Recorded non-zero counters (first 10 of 64 runs)

retire uop (01)	cycle (02)	schedule uop (52)	schedule int uop (53)	schedule simd uop (54)	dispatch int uop (56)	dispatch simd uop (57)	int uops in schedulers (59)	ldst uops in schedulers (5b)	dispatch uop (78)	map int uop (7c)	map simd uop (7e)	map int uop inputs (7f)	map simd uop inputs (81)	? int output thing (e9)	? simd retires (ee)	? int retires (ef)
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10006	200	30018	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100
10204	30033	10101	101	10000	100	10000	300	768905	10100	200	10004	200	30012	1	10000	100

1000 unrolls and 10 iterations

Result (median cycles for code): 3.0033

Recorded non-zero counters (first 10 of 64 runs)

retire uop (01)	cycle (02)	schedule uop (52)	schedule int uop (53)	schedule simd uop (54)	dispatch int uop (56)	dispatch simd uop (57)	int uops in schedulers (59)	ldst uops in schedulers (5b)	dispatch uop (78)	map int uop (7c)	map simd uop (7e)	map int uop inputs (7f)	map simd uop inputs (81)	? int output thing (e9)	? simd retires (ee)	? int retires (ef)
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10004	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	67	769247	10051	20	10047	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10
10024	30033	10021	21	10000	20	10000	70	768905	10020	20	10000	20	30000	11	10000	10

Test 5: throughput

Count: 8

Code:

  movi v0.16b, 0
  smlal2 v0.4s, v8.8h, v9.h[1]
  movi v1.16b, 0
  smlal2 v1.4s, v8.8h, v9.h[1]
  movi v2.16b, 0
  smlal2 v2.4s, v8.8h, v9.h[1]
  movi v3.16b, 0
  smlal2 v3.4s, v8.8h, v9.h[1]
  movi v4.16b, 0
  smlal2 v4.4s, v8.8h, v9.h[1]
  movi v5.16b, 0
  smlal2 v5.4s, v8.8h, v9.h[1]
  movi v6.16b, 0
  smlal2 v6.4s, v8.8h, v9.h[1]
  movi v7.16b, 0
  smlal2 v7.4s, v8.8h, v9.h[1]

Initialization

  movi v8.16b, 9
  movi v9.16b, 10

(fused SUBS/B.cc loop)

100 unrolls and 100 iterations

Result (median cycles for code divided by count): 0.5011

Recorded non-zero counters (first 10 of 64 runs)

retire uop (01)	cycle (02)	schedule uop (52)	schedule int uop (53)	schedule simd uop (54)	dispatch int uop (56)	dispatch simd uop (57)	dispatch ldst uop (58)	int uops in schedulers (59)	simd uops in schedulers (5a)	ldst uops in schedulers (5b)	dispatch uop (78)	map int uop (7c)	map ldst uop (7d)	map simd uop (7e)	map int uop inputs (7f)	map simd uop inputs (81)	? int output thing (e9)	? simd retires (ee)	? int retires (ef)
160204	40479	80109	101	80008	100	80012	0	300	0	320056	80113	200	0	80013	200	240039	1	160000	100
160204	40107	80109	101	80008	100	80012	0	300	0	320052	80112	200	0	80012	200	240036	1	160000	100
160204	40086	80109	101	80008	100	80012	0	300	0	320052	80112	200	0	80012	200	240036	1	160000	100
160204	40086	80109	101	80008	100	80012	0	300	0	320052	80112	200	0	80012	200	240036	1	160000	100
160204	40086	80109	101	80008	100	80012	0	300	0	320052	80112	200	0	80012	200	240036	1	160000	100
160204	40086	80109	101	80008	100	80012	0	300	0	320052	80112	200	0	80012	200	240036	1	160000	100
160204	40086	80109	101	80008	100	80012	0	300	0	320052	80112	200	0	80012	200	240036	1	160000	100
160204	40086	80109	101	80008	100	80012	0	300	0	320052	80112	200	0	80012	200	240036	1	160000	100
160204	40086	80109	101	80008	100	80012	0	300	0	320052	80112	200	0	80012	200	240036	1	160000	100
160204	40086	80109	101	80008	100	80012	0	300	0	320052	80112	200	0	80012	200	240036	1	160000	100

1000 unrolls and 10 iterations

Result (median cycles for code divided by count): 0.5052

Recorded non-zero counters (first 10 of 64 runs)

retire uop (01)	cycle (02)	schedule uop (52)	schedule int uop (53)	schedule simd uop (54)	dispatch int uop (56)	dispatch simd uop (57)	int uops in schedulers (59)	ldst uops in schedulers (5b)	dispatch uop (78)	map int uop (7c)	map simd uop (7e)	map int uop inputs (7f)	map simd uop inputs (81)	? int output thing (e9)	? simd retires (ee)	? int retires (ef)
160024	43741	80020	11	80009	10	80013	30	320000	80010	20	80000	20	240000	1	160000	10
160024	41149	80011	11	80000	10	80000	30	320000	80010	20	80000	20	240000	1	160000	10
160024	40431	80011	11	80000	10	80000	30	320000	80010	20	80000	20	240000	1	160000	10
160024	40361	80011	11	80000	10	80000	30	320000	80010	20	80000	20	240000	1	160000	10
160024	40361	80011	11	80000	10	80000	30	320000	80010	20	80000	20	240000	1	160000	10
160024	40400	80011	11	80000	10	80000	30	320000	80010	20	80000	20	240000	1	160000	10
160024	40407	80011	11	80000	10	80000	30	320000	80010	20	80000	20	240000	1	160000	10
160024	41209	80011	11	80000	10	80000	30	320000	80010	20	80000	20	240000	1	160000	10
160024	40411	80011	11	80000	10	80000	30	320000	80010	20	80000	20	240000	1	160000	10
160024	40421	80011	11	80000	10	80000	30	320056	80023	20	80013	20	240039	1	160000	10

Test 6: throughput

Count: 12

Code:

  smlal2 v0.4s, v12.8h, v13.h[1]
  smlal2 v1.4s, v12.8h, v13.h[1]
  smlal2 v2.4s, v12.8h, v13.h[1]
  smlal2 v3.4s, v12.8h, v13.h[1]
  smlal2 v4.4s, v12.8h, v13.h[1]
  smlal2 v5.4s, v12.8h, v13.h[1]
  smlal2 v6.4s, v12.8h, v13.h[1]
  smlal2 v7.4s, v12.8h, v13.h[1]
  smlal2 v8.4s, v12.8h, v13.h[1]
  smlal2 v9.4s, v12.8h, v13.h[1]
  smlal2 v10.4s, v12.8h, v13.h[1]
  smlal2 v11.4s, v12.8h, v13.h[1]

Initialization

  movi v12.16b, 13
  movi v13.16b, 14

(fused SUBS/B.cc loop)

100 unrolls and 100 iterations

Result (median cycles for code divided by count): 0.5003

Recorded non-zero counters (first 10 of 64 runs)

retire uop (01)	cycle (02)	schedule uop (52)	schedule int uop (53)	schedule simd uop (54)	dispatch int uop (56)	dispatch simd uop (57)	int uops in schedulers (59)	ldst uops in schedulers (5b)	dispatch uop (78)	map int uop (7c)	map simd uop (7e)	map int uop inputs (7f)	map simd uop inputs (81)	? int output thing (e9)	? simd retires (ee)	? int retires (ef)
120204	60060	120105	101	120004	100	120008	300	480044	120110	200	120014	200	360042	1	120000	100
120204	60035	120107	101	120006	100	120010	300	480036	120108	200	120012	200	360036	1	120000	100
120204	60035	120105	101	120004	100	120008	300	480036	120108	200	120012	200	360036	1	120000	100
120204	60035	120105	101	120004	100	120008	300	480036	120108	200	120012	200	360036	1	120000	100
120204	60035	120105	101	120004	100	120008	300	480036	120108	200	120012	200	360036	1	120000	100
120204	60035	120105	101	120004	100	120008	300	480036	120108	200	120012	200	360192	1	120000	100
120204	60035	120105	101	120004	100	120008	300	480036	120108	200	120012	200	360036	1	120000	100
120204	60035	120105	101	120004	100	120008	300	480036	120108	200	120012	200	360036	1	120000	100
120204	60035	120105	101	120004	100	120008	300	480036	120108	200	120012	200	360036	1	120000	100
120204	60035	120105	101	120004	100	120008	300	480036	120108	200	120012	200	360036	1	120000	100

1000 unrolls and 10 iterations

Result (median cycles for code divided by count): 0.5003

Recorded non-zero counters (first 10 of 64 runs)

retire uop (01)	cycle (02)	schedule uop (52)	schedule int uop (53)	schedule simd uop (54)	dispatch int uop (56)	dispatch simd uop (57)	int uops in schedulers (59)	ldst uops in schedulers (5b)	dispatch uop (78)	map int uop (7c)	map simd uop (7e)	map int uop inputs (7f)	map simd uop inputs (81)	? int output thing (e9)	? simd retires (ee)	? int retires (ef)
120024	60086	120017	11	120006	10	120010	30	480044	120020	20	120014	20	360042	1	120000	10
120024	60035	120011	11	120000	10	120000	30	480000	120010	20	120000	20	360000	1	120000	10
120024	60035	120011	11	120000	10	120000	30	480000	120010	20	120000	20	360000	1	120000	10
120024	60035	120011	11	120000	10	120000	30	480000	120010	20	120000	20	360000	1	120000	10
120024	60035	120011	11	120000	10	120000	30	480000	120010	20	120000	20	360000	1	120000	10
120024	60035	120011	11	120000	10	120000	30	480000	120010	20	120000	20	360000	1	120000	10
120024	60035	120011	11	120000	10	120000	30	480000	120010	20	120000	20	360000	1	120000	10
120024	60035	120011	11	120000	10	120000	30	480000	120010	20	120000	20	360000	1	120000	10
120024	60035	120011	11	120000	10	120000	30	480000	120010	20	120000	20	360000	1	120000	10
120024	60035	120011	11	120000	10	120000	30	480216	120066	20	120065	20	360000	1	120000	10

SMLAL2 (by element, 4S)

Test 1: uops

1000 unrolls and 1 iteration

Test 2: Latency 1->1

100 unrolls and 100 iterations

1000 unrolls and 10 iterations

Test 3: Latency 1->2

100 unrolls and 100 iterations

1000 unrolls and 10 iterations

Test 4: Latency 1->3

100 unrolls and 100 iterations

1000 unrolls and 10 iterations

Test 5: throughput

100 unrolls and 100 iterations

1000 unrolls and 10 iterations

Test 6: throughput

100 unrolls and 100 iterations

1000 unrolls and 10 iterations