Apple M1 Microarchitecture Research by Dougall Johnson

Firestorm: Overview | Base Instructions | SIMD and FP Instructions
Icestorm:  Overview | Base Instructions | SIMD and FP Instructions

FNMADD (scalar, H)

Test 1: uops

Code:

  fnmadd h0, h0, 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: 0.000

Load/store unit issues: 0.000

SIMD/FP unit issues: 1.000

retire (01)cycle (02)031e3f4e51inst issue (52)~issue fp/simd (54)~dispatch fp/simd (57)huge thing fp/simd (5b)60696b6d6edispatch stall (70)scheduler rewind (75)scheduler stall (76)~dispatch op (78)~map op fp/simd (7e)~map lookup fp/simd (81)8283pipeline redirect (84)85inst all (8c)inst neon or fp (9a)a0a1a8accfd5d6ddinst fetch restart (de)e0? fp/simd (ee)f5f6f7f8fd
100440373096134072510001000100053190804018040374037325833895100010003000403740371110011000000073116113473100040384038403840384038
100440373106134072510001000100053190814018040374037325833895100010003000403740371110011000241073116113473100040384038403840384038
100440373006134072510001000100053190804018040374037325833895100010003000403740371110011000000073116113473100040384038403840384038
100440373008234072510001000100053190814018040374037325833895100010003000403740371110011000000073116113473100040384038403840384038
1004403730276134072510001000100053190804018040374037325833895100010003000403740371110011000000073116113473100040384038403840384038
100440373006134072510001000100053190814018040374037325833895100010003000403740371110011000000073116113473100040384038403840384038
100440373006134072510001000100053190804018040374037325833895100010003000403740371110011000000073116113473100040384038403840384038
100440373006134072510001000100053190814018040374037325833895100010003000403740371110011000000073116113473100040384038403840384038
100540373006134072510001000100053190814018040374037325833895100010003000403740371110011000000073116113473100040384038403840384038
100440373006134072510001000100053190814018040374037325833895100010003000403740371110011000000073116113473100040384038403840384038

Test 2: Latency 1->2

Code:

  fnmadd h0, h0, 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): 4.0037

retire (01)cycle (02)03183a3f4e51inst issue (52)~issue int (53)~issue fp/simd (54)~dispatch int (56)~dispatch fp/simd (57)huge thing int (59)huge thing fp/simd (5b)60696b6d6edispatch stall (70)72scheduler rewind (75)scheduler stall (76)~dispatch op (78)~map op int (7c)~map op fp/simd (7e)~map lookup int (7f)~map lookup fp/simd (81)8283pipeline redirect (84)85inst all (8c)inst branch (8d)inst branch taken (90)inst b.cc (94)inst integer (97)inst neon or fp (9a)9fa1a8accfd0d5d6ddinst fetch restart (de)e0? fp/simd (ee)gpr retires (ef)f5f6f7f8fd
10204400373000061394072510100100100001001000050057069080400180400374003738108033874510100200100002003000040037400371110201100991001001000010000071203162239479100001004003840038400384003840038
102044003729900441394072510100100100001001000050057069081400180400374003738108033874510100200100002003000040037400371110201100991001001000010000071012162239479100001004003840038400384003840038
10204400373000061394072510100100100001001000050057069081400650400374003738108033874510100200100002003000040037400371110201100991001001000010000071013162239479100001004003840038400384003840038
10204400372990061394072510100100100001001000050057069080400180400374003738108033874510100200100002003000040037400371110201100991001001000010000071012162239479100001004003840038400384003840038
10204400373000061394072510100100100001001000050057069080400180400374003738108033874510100200100002003000040037400371110201100991001001000010000071012162239479100001004003840038400384003840038
102044018130100536394072510100100100001001000050057069081400180400374003738108033874510100200100002003000040037400371110201100991001001000010000071012162239479100001004003840038400384003840038
10204400373000061394072510100100100001001000050057069080400180400374003738108033874510100200100002003000040037400371110201100991001001000010000071012163239479100001004003840038400384003840038
10204400373000061394072510100100100001001000050057069081400180400374003738108733874510100200100002003000040037400371110201100991001001000010000071012160239479100001004003840038400384003840038
10204400373000061394072510100100100001001000050057069080400180400374003738108033874510100200100002003000040037400371110201100991001001000010000071012162239479100001004003840038400384003840038
10204400373000061394072510100100100001001000050057069081400180400374003738108033874510100200100002003000040037400371110201100991001001000010000071012162239479100001004003840038400384003840038

1000 unrolls and 10 iterations

Result (median cycles for code): 4.0037

retire (01)cycle (02)03070b1e1f3f4e51inst issue (52)~issue int (53)~issue fp/simd (54)~dispatch int (56)~dispatch fp/simd (57)huge thing int (59)huge thing fp/simd (5b)60696d6edispatch stall (70)scheduler rewind (75)scheduler stall (76)~dispatch op (78)~map op int (7c)~map op fp/simd (7e)~map lookup int (7f)~map lookup fp/simd (81)8283pipeline redirect (84)85inst all (8c)inst branch (8d)inst branch taken (90)inst b.cc (94)inst integer (97)inst neon or fp (9a)9fa0a6a7a8accfd5d6ddinst fetch restart (de)e0? int output thing (e9)? fp/simd (ee)gpr retires (ef)f5f6f7f8fd
100244003729900006139407251001010100001010000505706908040018400374008438130338767100102010000203000040037400371110021109101010000100006966403163339473010000104003840038400384003840038
100244003730000005363940725100101010000101000050570690814001840037400373813033876710158201000020300004003740037111002110910101000010020236403163339473010000104003840038400384003840086
100244003730000006139407431001010100001010000505706908040018400374003738130338767100102010000203000040037400371110021109101010000100005166403163339473010000104003840038400384003840038
100244003730000006139407451001010100001010000505706908040018400374003738130338767100102010000203000040037400371110021109101010000100004536403163339473010000104007140038400384003840038
10024400373000066886139407441001010100001010000505706908040018400374003738130338767100102010000203000040037400841110021109101010000100005836403163339473010000104003840038400384003840038
100244003729900006139407251001010100001010000505706908140018400374003738130338767100102210000203000040037400371110021109101010000100004836403163339473010000104003840038400384003840038
100244008430000006139407251001010100001010000505706908040018400374003738130338767100102010000203000040037400371110021109101010000100004236403163339473010000104003840038400384003840038
100244003730000006139407251001010100001010000505706908040018400374003738130338767100102010000203000040037400371110021109101010000100004766403243339473010000104003840038400384003840038
10024400373000000613940725100101010000101000050570690804001840037400373813033876710010201000020300004003740037111002110910101000010000796403163339473010000104003840038400384003840038
100244003730000006139407251001010100001010000505706908040018400374003738130338767100102010000203000040037400371110021109101010000100004066403253339473010000104003840038400384003840038

Test 3: Latency 1->3

Code:

  fnmadd h0, h1, h0, 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): 4.0037

retire (01)cycle (02)031e3a3f4e51inst issue (52)~issue int (53)~issue fp/simd (54)~dispatch int (56)~dispatch fp/simd (57)huge thing int (59)huge thing fp/simd (5b)696d6edispatch stall (70)scheduler rewind (75)scheduler stall (76)~dispatch op (78)~map op int (7c)~map op fp/simd (7e)~map lookup int (7f)~map lookup fp/simd (81)8283pipeline redirect (84)85inst all (8c)inst branch (8d)inst branch taken (90)inst b.cc (94)inst integer (97)inst neon or fp (9a)9fa8accdcfd0d5d6ddinst fetch restart (de)e0? int output thing (e9)? fp/simd (ee)gpr retires (ef)f5f6f7f8fd
102044003729900613940725101001001000010010000500570690840018400374003738108338745101002001000020030000400374003711102011009910010010000100230710031622394790100001004003840038400384003840038
10204400373000034839407251010010010000100100005005706908400184003740037381083387451010020010000200300004003740037111020110099100100100001006200710121622394790100001004003840038400384003840038
102044003730090613940725101001001000010010000500570690840018400374003738108338745101002001000020030000400374003711102011009910010010000100100710131622394790100001004003840085400384003840038
102044003730000613940725101001001000010010000500570690840018400374003738108338745101002001000020030000400374003711102011009910010010000100660710121622394790100001004003840038400384003840038
102044003730000613940725101001001000010010000500570690840018400374003738108338745101002001000020030000400374003711102011009910010010000100100710121622394790100001004003840038400384003840038
102044003729900613940725101001001000010010000500570690840018400374003738108338745101002001000020030000400374003711102011009910010010000100000710121622394790100001004003840038400384003840038
102044003730000613940725101001001000010010000500570690840018400374003738108338745101002001000020030000400374003711102011009910010010000100200710121622394790100001004003840038400384003840038
1020440037300006139407251010010010000100100005005706908400184003740037381083387451010020010000200300004003740037111020110099100100100001002600710121622394790100001004003840038400384003840038
102044003731300613940725101001001000010010000500570690840018400374003738108338745101002001000020030000400374003711102011009910010010000100101710121622394790100001004003840038400384003840038
102044003730000613940725101001001000010010000500570690840018400374003738108338745101002001000020030000400374003711102011009910010010000100000710121622394790100001004003840038400384003840038

1000 unrolls and 10 iterations

Result (median cycles for code): 4.0037

retire (01)cycle (02)030f1e3f4e51inst issue (52)~issue int (53)~issue fp/simd (54)~dispatch int (56)~dispatch fp/simd (57)huge thing int (59)huge thing fp/simd (5b)60696d6edispatch stall (70)72scheduler rewind (75)scheduler stall (76)~dispatch op (78)~map op int (7c)~map op fp/simd (7e)~map lookup int (7f)~map lookup fp/simd (81)8283pipeline redirect (84)85inst all (8c)inst branch (8d)inst branch taken (90)inst b.cc (94)inst integer (97)inst neon or fp (9a)9fa0a8accdcfd5d6ddinst fetch restart (de)e0? fp/simd (ee)gpr retires (ef)f5f6f7f8fd
10024400372992061394072510010101000010100005057069080400184003740037381300338767100102010000203000040037400371110021109101010000100000640416333947310000104003840038400384003840038
10024400373000061394072510010101000010100005057069081400184003740037381300338767100102010000203000040037400371110021109101010000100000640316333947310000104003840038400384003840038
1002440037300001142394072510010101000010100005057069081400184003740037381300338767100102010000203000040037400371110021109101010000100000640316333947310000104003840038401224003840038
10024400373000061394072510010101000010100005057069080400184003740037381300338767100102010000203000040037400371110021109101010000100000640316333947310000104003840038400384003840038
10024400373000061394072510010101000010100005057069081400184003740037381300338767100102010000203000040037400371110021109101010000100000640316333947310000104003840038400384003840038
10024400373000061394072510010101000010100005057069081400184003740037381300338767100102010000203000040037400841110021109101010000100000640316333947310000104003840038400384003840038
10024400373000061394072510010101000010100005057069081400184003740037381300338767100102010000203000040037400371110021109101010000100000640316333947310000104003840038400384003840038
10024400373000061394072510010101000010100005057069080400184003740037381300338767100102010000203000040037400371110021109101010000100000640316333947310000104003840038400844003840038
10024400373000061394072510010101000010100005057069081400184003740037381300338767100102010000203000040037400371110021109101010000100000640316333947310000104003840038400384003840038
100244003730000156394072510010101000010100005057069080400184003740037381540338767100102010000203000040037400371110021109101010000100000640316333947310000104003840038400384003840038

Test 4: Latency 1->4

Code:

  fnmadd h0, h1, h2, h0
  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): 4.0037

retire (01)cycle (02)0304080918191e1f3a3f4e51inst issue (52)~issue int (53)~issue fp/simd (54)~dispatch int (56)~dispatch fp/simd (57)huge thing int (59)huge thing fp/simd (5b)60696d6edispatch stall (70)72scheduler rewind (75)scheduler stall (76)~dispatch op (78)~map op int (7c)~map op fp/simd (7e)~map lookup int (7f)~map lookup fp/simd (81)8283pipeline redirect (84)85inst all (8c)inst branch (8d)inst branch taken (90)inst b.cc (94)inst integer (97)inst neon or fp (9a)9fa0a1a6a7a8a9accfd0d5d6ddinst fetch restart (de)e0? int output thing (e9)? fp/simd (ee)gpr retires (ef)f5f6f7f8fd
102044003729910000000613940725101001001000010010000500570690814001840037400373810803387451010020010000200300004003740037111020110099100100100001000000000712121622394790100001004003840038400384003840038
102044003730000000000613940725101001001000010010000500570690814001840037400373810803387451010020010000200300004003740037111020110099100100100001000000000710121622394790100001004003840038400384003840038
1020440037300000001500613940725101001001000010010000500570690814001840037400373810803387451010020010000200300004003740037111020110099100100100001000000000710121622394790100001004003840038400384003840038
102044003730000000000613940725101001001000010010000500570690814001840037400373810803387451010020010000200300004003740037111020110099100100100001000000000710121622394790100001004003840038400384003840038
102044003730000000000613940725101001001000010010000500570690814001840037400373810803387451010020010000200300004003740037111020110099100100100001000000000710121622394790100001004003840038400384003840038
102044003730000000000613940725101001001000010010000500570690804001840037400373810803387451010020010000200300004003740037111020110099100100100001000000000710121622394790100001004003840038400384003840038
102044003730000000001613940725101001001000010010000500570690804001840037400373810803387451010020010000200300004003740037111020110099100100100001000000000710121622394790100001004003840038400384003840038
102044003730000000000613940725101001001000010010000500570690814001840037400373810803387451010020010000200300004003740037111020110099100100100001000000000710121622394790100001004003840038400384003840038
102044003730000000000613940725101001001000010010000500570690804001840037400373810803387451010020010000200300004003740037111020110099100100100001000000200710121622394790100001004003840038400384003840038
102044003730000000000613940725101001001000010010000500570690804001840037400373810803387451010020010000200300004003740037111020110099100100100001000000000710121622394790100001004003840038400384003840038

1000 unrolls and 10 iterations

Result (median cycles for code): 4.0037

retire (01)cycle (02)031e1f3a3f4e51inst issue (52)~issue int (53)~issue fp/simd (54)~dispatch int (56)~dispatch fp/simd (57)huge thing int (59)huge thing fp/simd (5b)60696d6edispatch stall (70)72scheduler rewind (75)scheduler stall (76)~dispatch op (78)~map op int (7c)~map op fp/simd (7e)~map lookup int (7f)~map lookup fp/simd (81)8283pipeline redirect (84)85inst all (8c)inst branch (8d)inst branch taken (90)inst b.cc (94)inst integer (97)inst neon or fp (9a)9fa8accfd5d6ddinst fetch restart (de)e0? fp/simd (ee)gpr retires (ef)f5f6f7f8fd
100244003730000012003940725100101010000101000050570690804001840037400373813003387671001020100002030000400374003711100211091010100001000640316223947310000104003840038400384003840038
10024400373000001193940725100101010000101000050570690814001840037400373813003387671001020100002030000400374003711100211091010100001000640216223947310000104003840038400384003840038
1002440037300000613940725100101010000101000050570690814001840037400373813003387671001020100002030000400374003711100211091010100001000640216223947310000104003840038400384003840038
1002440037300000613940725100101010000101000050570690804001840037400373813003387671001020100002030000400374003711100211091010100001000640216223947310000104003840038400384003840038
100244003730000061394072510010101000010100005057069081400184003740037381300338767100102010000203000040037400371110021109101010000100141640216223947310000104003840038400384003840038
1002440037300000613940725100101010000101000050570690814001840037400373813003387671001020100002030000400374003711100211091010100001000640216223947310000104003840038400384003840038
1002440037300000613940725100101010000101000050570690804001840037400373813003387671001020100002030000400374003711100211091010100001000640216223947310000104003840038400384003840038
1002440037300000613940725100101010000101000050570690814001840037400373813003387671001020100002030000400374003711100211091010100001000640216223947310000104003840038400384003840038
1002440037300000613940725100101010000101000050570690814001840037400373813003387671001020100002030000400374003711100211091010100001000640216223947310000104003840038400384003840038
1002440037299000613940725100161010000101000050570690814001840037400373813003387671001020100002030000400374003711100211091010100001000640216223947310000104003840038400384003840038

Test 5: throughput

Count: 8

Code:

  fnmadd h0, h8, h9, h10
  fnmadd h1, h8, h9, h10
  fnmadd h2, h8, h9, h10
  fnmadd h3, h8, h9, h10
  fnmadd h4, h8, h9, h10
  fnmadd h5, h8, h9, h10
  fnmadd h6, h8, h9, h10
  fnmadd h7, h8, h9, h10
  movi v8.16b, 9
  movi v9.16b, 10
  movi v10.16b, 11

(fused SUBS/B.cc loop)

100 unrolls and 100 iterations

Result (median cycles for code divided by count): 0.2505

retire (01)cycle (02)03080b1e3f51inst issue (52)~issue int (53)~issue fp/simd (54)~dispatch int (56)~dispatch fp/simd (57)huge thing int (59)huge thing fp/simd (5b)60696d6edispatch stall (70)scheduler rewind (75)scheduler stall (76)~dispatch op (78)~map op int (7c)~map op fp/simd (7e)~map lookup int (7f)~map lookup fp/simd (81)8283pipeline redirect (84)85inst all (8c)inst branch (8d)inst branch taken (90)inst b.cc (94)inst integer (97)inst neon or fp (9a)9fc5cfd5d6ddinst fetch restart (de)e0? fp/simd (ee)gpr retires (ef)f5f6f7f8fd
8020420059150000422580100100800001008000050064000012002120040200409973399988010020080000200240000200402004011802011009910010080000100051101161120037800001002004120094201042009520041
8020420040150000422580100100800001008000050064000002002120040200409973399988010020080105200240315200402004011802011009910010080000100051101161120037800001002004120041200412004120041
8020420040150000422580100100800001008000050064000012002120040200409973399988010020080000200240000200402004011802011009910010080000100051101161120037800001002004120041200412004120041
8020420040150000422580100100800001008000050064000012002120040200409973399988010020080000200240000200402004011802011009910010080000100051101161120037800001002004120041200412004120041
80204200401510006122580100100800001008000050064000002002120040200409973399988010020080000200240000200402004011802011009910010080000100051101161120037800001002004120041200412004120041
80204200401500018422580100100800001008000050064000002002120040200409973399988010020080000200240000200402004011802011009910010080000100051101161120037800001002004120041200412004120041
8020420040150000422580100100800001008010650064000002002120040200409973399988010020080000200240000200402004011802011009910010080000100051101161120037800001002004120041200412004120041
80204200401500012422580100100800001008000050064000002002120040200409973399988010020080000200240000200402004011802011009910010080000100051101161120037800001002004120041200412004120041
802042004015000159422580100100800001008000050064000002002120040200409973399988010020080000200240000200402004011802011009910010080000100051101161120037800001002004120041200412004120041
8020420040150000422580100100800001008000050064000002002120040200409973399988010020080000200240000200402004011802011009910010080000100051101161120037800001002004120041200412004120041

1000 unrolls and 10 iterations

Result (median cycles for code divided by count): 0.2505

retire (01)cycle (02)0308090b18191e1f3a3f51inst issue (52)~issue int (53)~issue fp/simd (54)~dispatch int (56)~dispatch fp/simd (57)huge thing int (59)huge thing fp/simd (5b)60696d6edispatch stall (70)scheduler rewind (75)scheduler stall (76)~dispatch op (78)~map op int (7c)~map op fp/simd (7e)~map lookup int (7f)~map lookup fp/simd (81)8283pipeline redirect (84)85inst all (8c)inst branch (8d)inst branch taken (90)inst b.cc (94)inst integer (97)inst neon or fp (9a)9fa0a8a9accfd5d6ddinst fetch restart (de)e0? fp/simd (ee)gpr retires (ef)f5f6f7f8fd
80024200501500000012004125800101080000108000050640000120021200402004099963100208001020800002024000020040200401180021109101080000100000502017166172003780000102004120041200412004120041
800242004015000000270041258001010800001080000506400001200212004020040999631002080010208000020240000200402004011800211091010800001000005020171617222003780000102004120041200412004120041
8002420040150000000004125800101080000108000050640000120021201152004099963100208001020800002024000020040200401180021109101080000100000502017168172003780000102004120041200412004120041
8002420040150000000004125800101080000108000050640000120021200402004099963100208001020800002024000020040200401180021109101080000100000502017161762003780000102004120041200412004120041
8002420040150000000004125800101080000108000050640000120021200402004099963100208001020800002024000020040200401180021109101080000100000502017161762003780000102004120041200412004120041
800242004015000000000412580010108000010800005064000012002120040200409996310020800102080000202400002004020040118002110910108000010000050206161572003780000102004120041200412004120041
8002420040150000009004125800101080000108000050640000120021200402004099963100208001020800002024000020040200401180021109101080000100000502017587182003780000102004120041200412004120041
800242004015000000186004125800101080000108000050640000120021200402004099963100208001020800002024000020040200401180021109101080000100000502017165162003780000102004120041200412004120041
80024200401500000027004125800101080000108000050640000120021200402004099963100208001020800002024000020040200401180021109101080000100000502017166172003780000102004120041200412004120041
800242004015000000000412580010108000010800005064000012002120040200409996310020800102080000202400002004020040118002110910108000010000050207161862003780000102004120041200412004120041