Apple Microarchitecture Research by Dougall Johnson

M1/A14 P-core (Firestorm): Overview | Base Instructions | SIMD and FP Instructions
M1/A14 E-core (Icestorm):  Overview | Base Instructions | SIMD and FP Instructions

SQDMLSL (scalar, S)

Test 1: uops

Code:

  sqdmlsl s0, h1, h2
  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

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)
1004303310011100010007590510001000300011000
1004303310011100010007590510001000300011000
1004303310011100010007590510001000300011000
1004303310011100010007590510001000311711000
1004303310011100010007590510001000300011000
1004303310011100010007590510001000300011000
1004303310011100010007590510001000300011000
1004303310011100010007590510001000300011000
1004303310011100010007590510001000300011000
1004303310011100010007590510001000300011000

Test 2: Latency 1->1

Code:

  sqdmlsl s0, h1, h2
  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

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)
1020430033101011011000010010000300769247101312001004420230138110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030132110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020530167101231011002210010097300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030132110000100
1020430033101011011000010010000300769247101312001004420030252110000100

1000 unrolls and 10 iterations

Result (median cycles for code): 3.0033

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)
10024300331002121100002010000707689051002020100042030000111000010
10024300331002121100002010000707689051002020100002030132111000010
10024300331002121100002010000707689051002020100002030000111000010
10024300331002121100002010000707689051002020100002030000111000010
10024300331002121100002010000707689051002020100002030000111000010
10024300331002121100002010000707689051002020100002030000111000010
10024300331002121100002010000707689051002020100002030000111000010
10024300331002121100002010000707689051002020100002030000111000010
10024300331002121100002010000707689051002020100002030000111000010
10024300331002121100002010000707689051002020100002030000111000010

Test 3: Latency 1->2

Code:

  sqdmlsl s0, h0, h1
  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

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)
1020430033101011011000010010000300768905101002001000620030012110000100
1020430033101011011000010010000300768905101002001000620030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100

1000 unrolls and 10 iterations

Result (median cycles for code): 3.0033

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)
1002430033100212110000201000070768905100202010006200300001101000010
1002430033100212110000201000070768905100202010000200300001101000010
1002430033100212110000201000070768905100202010000200300001101000010
1002430033100212110000201000070768905100202010000200300001101000010
1002430033100212110000201000070768905100202010000200300001101000010
1002430033100212110000201000070768905100202010000200300001101000010
1002430033100212110000201000070768905100202010000200300001101000010
1002430033100212110000201000070768905100202010000200300001101000010
1002430033100212110000201000070768905100202010000200300001101000010
1002430033100212110000201000070768905100202010000200300001101000010

Test 4: Latency 1->3

Code:

  sqdmlsl s0, h1, h0
  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

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)
1020430033101011011000010010000300768905101002001000620030018110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100
1020430033101011011000010010000300768905101002001000420030012110000100

1000 unrolls and 10 iterations

Result (median cycles for code): 3.0033

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)
10024300331002121100002010000707689051002020100002030000111000010
10024300331002121100002010000707689051002020100002030000111000010
10024300331002121100002010000707689051002020100002030000111000010
10024300331002121100002010000707689051002020100002030000111000010
10024300331002121100002010000707689051002020100002030000111000010
10024300331002121100002010000707689051002020100002030000111000010
10024300331002121100002010000707689051002020100002030135111000010
10024300331002121100002010000707689051002020100002030000111000010
10024300331002121100002010000707689051002020100002030000111000010
10024300331002121100002010000707689051002020100002030000111000010

Test 5: throughput

Count: 8

Code:

  movi v0.16b, 0
  sqdmlsl s0, h8, h9
  movi v1.16b, 0
  sqdmlsl s1, h8, h9
  movi v2.16b, 0
  sqdmlsl s2, h8, h9
  movi v3.16b, 0
  sqdmlsl s3, h8, h9
  movi v4.16b, 0
  sqdmlsl s4, h8, h9
  movi v5.16b, 0
  sqdmlsl s5, h8, h9
  movi v6.16b, 0
  sqdmlsl s6, h8, h9
  movi v7.16b, 0
  sqdmlsl s7, h8, h9
  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

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)
1602044050580110101800091008001330032019280147200800472002400391160000100
1602044010880110101800091008001330032005680113200800132002400361160000100
1602044009880109101800081008001230032005280112200800122002400361160000100
1602044008680109101800081008001230032005280112200800122002400361160000100
1602044008680109101800081008001230032005280112200800122002400361160000100
1602044010180110101800091008001330032005280112200800122002400361160000100
1602044008680109101800081008001230032005280112200800122002400361160000100
1602044008680109101800081008001230032005280112200800122002400361160000100
1602044008680109101800081008001230032005280112200800122002400361160000100
1602044008680109101800081008001230032005280112200800122002400361160000100

1000 unrolls and 10 iterations

Result (median cycles for code divided by count): 0.5050

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)
1600244410880020118000910800130300320000800102008000020240000116000010
1600244126780011118000010800000300320000800102008000020240000116000010
1600244042280011118000010800000300320000800102008000020240000116000010
1600244043980011118000010800000300320000800102008000020240000116000010
1600244038980011118000010800000300320000800102008000020240000116000010
1600244039380011118000010800000300320000800102008000020240000116000010
1600244038880011118000010800000300320000800102008000020240000116000010
1600244041180011118000010800000300320000800102008000020240000116000010
1600244042480011118000010800000300320000800102008000020240000116000010
1600244039380011118000010800000300320000800102008000020240000116000010

Test 6: throughput

Count: 16

Code:

  sqdmlsl s0, h16, h17
  sqdmlsl s1, h16, h17
  sqdmlsl s2, h16, h17
  sqdmlsl s3, h16, h17
  sqdmlsl s4, h16, h17
  sqdmlsl s5, h16, h17
  sqdmlsl s6, h16, h17
  sqdmlsl s7, h16, h17
  sqdmlsl s8, h16, h17
  sqdmlsl s9, h16, h17
  sqdmlsl s10, h16, h17
  sqdmlsl s11, h16, h17
  sqdmlsl s12, h16, h17
  sqdmlsl s13, h16, h17
  sqdmlsl s14, h16, h17
  sqdmlsl s15, h16, h17
  movi v16.16b, 17
  movi v17.16b, 18

(fused SUBS/B.cc loop)

100 unrolls and 100 iterations

Result (median cycles for code divided by count): 0.5002

retire uop (01)cycle (02)schedule uop (52)schedule int uop (53)schedule simd uop (54)schedule ldst uop (55)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)
1602048005616010510116000401001600080300064004416011020001600142004800391160000100
1602048003516010710116000601001600100300064003616010820001600122004800361160000100
1602058007016014110116004001001600520300064003616010820001600122004800391160000100
1602058007116014310116004201001600540300064003616010820001600122004800361160000100
1602048003516010510116000401001600080300064003616010820001600122004801921160000100
1602048003516010510116000401001600080300064003616010820001600122004800361160000100
1602048003516010510116000401001600080300064004416011020001600132004800391160000100
1602048003516010510116000401001600080300064004416011020001600142004800361160000100
1602048004516010710116000601001600100300064003616010820001600122004800361160000100
1602048003516010510116000401001600080300064003616010820001600122004800361160000100

1000 unrolls and 10 iterations

Result (median cycles for code divided by count): 0.5002

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)
160024801791600151116000410160008306400441600202016001420480000116000010
160024800411600111116000010160000306400001600102016000020480000116000010
160024800351600111116000010160000306400001600102016000020480000116000010
160024800351600111116000010160000306400001600102016000020480000116000010
160024800351600111116000010160000306402161600662016006320480000116000010
160024800351600111116000010160000306400001600102016000020480000116000010
160024800351600111116000010160000306400001600102016000020480000116000010
160024800371600111116000010160000306400001600102016000020480000116000010
160024800351600111116000010160000306400001600102016000020480000116000010
160024800351600111116000010160000306400001600102016000020480000116000010