BRCM 5000

Broadcom BCM7356 (BRCM 5000 / Zephyr), 1300 MHz, 1 core, 2 threads per core, 1 GB DDR3.

L1 Data cache = 32 KB. 4-way, 32 B/line.
L1 Instruction cache = 32 KB, 4-way. 64 B/line. 3 bits per instruction (32 bits) to hold the pre-decode flags.
L2 Data cache = 256 KB. 128 B/line.

L1 Data Cache Latency = 3 cycles
L2 Data Cache Latency = 18 cycles
RAM Latency = 18 cycles + 180 ns

Fetch 4 instructions.
4K-entry Gshare-based branch history table (BHT) for each thread.
64-entry branch target buffer (BTB) for each thread.
8-entry return address stack (RAS) for each thread.
Instruction buffer: 64 entries: 32 for each thread.
2 integer execution units:
EU1: ALU, MUL, DIV, DSP
EU2: ALU, LOAD/STORE

4 KB pages

DTLB L1 size for variable-size pages = 4 pages (items) per thread
ITLB L1 size for variable-size pages = 1 page (item) per thread
4KB DTLB L1 size = 32 pages (128 KB), (32 items, 1 page per item) per thread
4KB ITLB L1 size = 32 pages (128 KB), (32 items, 1 page per item) per thread
4KB TLB L2 size = 128 pages (512 KB), (64 items, 2 pages per item) per thread

4KB DTLB L1 miss penalty = 13 cycles, Parallel miss: 13 cycles per read access.
4KB TLB L2 miss penalty = 100 cycles, Parallel miss: 100 cycles per read access.

  32 K     3                              TLB + L1
  64 K    11                8             + 15 (L2)
 128 K    15                4   
 256 K    24 +  20 ns       9   20 ns     + 13 (DTLB L1 miss)
 512 K    28 + 100 ns       4 + 80 ns     + 180 ns (RAM) 
   1 M    80 + 140 ns      52 + 40 ns     + 100 (TLB L2 miss)     
   2 M   106 + 160 ns      26 + 20 ns               
   4 M   119 + 170 ns      13 + 10 ns               
   8 M   125 + 175 ns       6 +  5 ns     
  16 M   128 + 178 ns       3 +  3 ns

4-bytes range cross penalty = 443 cycles
L1 B/W (Parallel Random Read) = 1 cycle per one access
L2 Read B/W (Read, 32 Bytes step) = 6 cycles per one access
L2 Read B/W (Read, 32 Bytes step, pointer-chasing) = 18 cycles per one access (no hardware prefetch)
L2 Read B/W (Parallel Random Read) = 4 cycle per one access
L2 Write (32 Bytes step) = 10 cycles per one access
RAM Read B/W (Parallel Random Read) = 50 ns / per cache line access (including L1 TLB miss)
RAM Read B/W (Read, 4-32 Bytes step) = 1200-1700 MB/s
RAM Read B/W (Read, 32 Bytes step, pointer-chasing) = 1360 MB/s
RAM Write (4-32 Bytes step) = 900-1000 MB/s

Pipeline

Branch misprediction penalty = 7 cycles.

12-stage pipeline:

I$
HIT?
DEC
BUF
Iss
Reg
Byp
ALU
D$
Hit?
Res
Wr

Execution speed problems

BRCM 5000 core has some performance problems with execution for some instruction sequences. For example, for repeated groups of 3 instructions:

  ; group 1
  lw   v0,0(v0)
  sll  v0,v0,0x2
  addu v0,a0,v0

  ; group 2
  lw   v0,0(v0)
  sll  v0,v0,0x2
  addu v0,a0,v0

we could expect the rate of 5 cycles per group, but the measured result was about 10 cycles per group.

7-Zip Benchmark

7z b : MIPS values are normalized with Intel Core 2 cpu.

### Linux-3.13 ## gcc-7 -O2 -march=mips32 -mtune=1004kc 7-Zip (a) 16.04 : Copyright (c) 1999-2016 Igor Pavlov : 2016-10-04 p7zip Version 16.04 (locale=C,Utf16=off,HugeFiles=on,32 bits,1 CPU LE) LE CPU Freq: 1275 1278 1276 1278 1277 1275 1277 1277 1277 RAM size: 559 MB, # CPU hardware threads: 1 RAM usage: 435 MB, # Benchmark threads: 1 Compressing | Decompressing Dict Speed Usage R/U Rating | Speed Usage R/U Rating KiB/s % MIPS MIPS | KiB/s % MIPS MIPS 22: 445 100 433 433 | 9854 100 840 841 23: 425 100 434 434 | 9620 100 832 833 24: 410 100 441 441 | 9411 100 827 826 25: 397 100 454 454 | 9077 100 809 808 ---------------------------------- | ------------------------------ Avr: 100 441 440 | 100 827 827 Tot: 100 634 634 7-Zip (a) 16.04 : Copyright (c) 1999-2016 Igor Pavlov : 2016-10-04 p7zip Version 16.04 (locale=C,Utf16=off,HugeFiles=on,32 bits,1 CPU LE) 7-Zip (a) 16.04 : Copyright (c) 1999-2016 Igor Pavlov : 2016-10-04 p7zip Version 16.04 (locale=C,Utf16=off,HugeFiles=on,32 bits,2 CPUs LE) LE CPU Freq: 1268 1270 1272 1271 1271 1271 1271 1271 1271 RAM size: 559 MB, # CPU hardware threads: 2 RAM usage: 441 MB, # Benchmark threads: 2 Compressing | Decompressing Dict Speed Usage R/U Rating | Speed Usage R/U Rating KiB/s % MIPS MIPS | KiB/s % MIPS MIPS 22: 639 161 386 622 | 14934 197 647 1275 23: 621 162 391 633 | 14817 199 644 1283 24: 615 167 397 661 | 14340 197 639 1259 25: 602 169 408 688 | 13865 197 628 1234 ---------------------------------- | ------------------------------ Avr: 165 395 651 | 197 639 1263 Tot: 181 517 957 7-Zip (a) 16.04 : Copyright (c) 1999-2016 Igor Pavlov : 2016-10-04 p7zip Version 16.04 (locale=C,Utf16=off,HugeFiles=on,32 bits,2 CPUs LE) LE CPU Freq: 1275 1277 1277 1274 1276 1277 1275 1277 1275 RAM size: 559 MB, # CPU hardware threads: 2 RAM usage: 450 MB, # Benchmark threads: 4 Compressing | Decompressing Dict Speed Usage R/U Rating | Speed Usage R/U Rating KiB/s % MIPS MIPS | KiB/s % MIPS MIPS 22: 724 197 357 705 | 14951 200 638 1276 23: 711 198 367 725 | 14660 200 636 1269 24: 693 198 376 745 | 14349 200 631 1260 ---------------------------------- | ------------------------------ Avr: 198 367 725 | 200 635 1268 Tot: 199 501 996

7z b -mm=* : MIPS and Effectiveness values are normalized with AMD K8 cpu.

freq= 1305 LE CPU Freq: 1275 1278 1275 1277 1269 1277 1277 1277 1276 RAM size: 559 MB, # CPU hardware threads: 1 RAM usage: 225 MB, # Benchmark threads: 1 Method Speed Usage R/U Rating E/U Effec KiB/s % MIPS MIPS % % CPU 100 1278 1277 98 98 CPU 100 1278 1277 98 98 CPU 100 1275 1277 98 98 LZMA:x1 1208 100 442 442 34 34 10267 100 837 836 64 64 LZMA:x5:mt1 413 100 517 517 40 40 9373 100 791 791 61 61 LZMA:x5:mt2 446 100 558 558 43 43 9433 100 797 796 61 61 Deflate:x1 3125 100 397 397 30 30 20641 100 641 641 49 49 Deflate:x5 1405 100 541 541 41 41 20675 100 641 642 49 49 Deflate:x7 524 100 581 581 45 45 20821 100 646 646 50 50 Deflate64:x5 1173 100 507 507 39 39 20168 100 631 631 48 48 BZip2:x1 792 100 479 479 37 37 3775 100 409 409 31 31 BZip2:x5 529 100 441 441 34 34 2267 100 445 445 34 34 BZip2:x5:mt2 525 100 439 439 34 34 2246 100 441 441 34 34 BZip2:x7 192 100 499 499 38 38 2269 100 446 445 34 34 PPMD:x1 343 100 356 356 27 27 232 100 274 273 21 21 PPMD:x5 240 100 408 408 31 31 175 100 329 329 25 25 Delta:4 49706 100 305 305 23 23 48613 100 299 299 23 23 BCJ 150007 100 615 614 47 47 146250 100 599 599 46 46 AES256CBC:1 17152 100 422 422 32 32 18353 100 451 451 35 35 AES256CBC:2 CRC32:1 127344 100 927 927 71 71 CRC32:4 157844 100 352 352 27 27 CRC32:8 228949 100 310 310 24 24 CRC64 128548 100 263 263 20 20 SHA256 21119 100 431 431 33 33 SHA1 51108 100 478 478 37 37 BLAKE2sp 29747 100 655 654 50 50 CPU 100 1278 1278 98 98 ------------------------------------------------------ Tot: 100 573 573 44 44 7-Zip (a) 16.04 : Copyright (c) 1999-2016 Igor Pavlov : 2016-10-04 p7zip Version 16.04 (locale=C,Utf16=off,HugeFiles=on,32 bits,2 CPUs LE) freq= 1305 LE CPU Freq: 1266 1271 1271 1271 1271 1270 1271 1270 1270 RAM size: 559 MB, # CPU hardware threads: 2 RAM usage: 450 MB, # Benchmark threads: 2 Method Speed Usage R/U Rating E/U Effec KiB/s % MIPS MIPS % % CPU 199 1245 2478 95 190 CPU 199 1245 2478 95 190 CPU 199 1244 2478 95 190 LZMA:x1 1736 196 324 635 25 49 15320 197 633 1248 49 96 LZMA:x5:mt1 627 196 399 783 31 60 14430 199 610 1217 47 93 LZMA:x5:mt2 688 198 434 860 33 66 14417 199 610 1216 47 93 Deflate:x1 4308 196 280 547 21 42 32657 197 516 1015 40 78 Deflate:x5 2098 195 413 808 32 62 32701 197 516 1015 40 78 Deflate:x7 887 196 502 983 38 75 32928 197 520 1022 40 78 Deflate64:x5 1781 196 393 770 30 59 32582 196 520 1019 40 78 BZip2:x1 1124 196 348 680 27 52 6014 196 332 652 25 50 BZip2:x5 824 196 351 688 27 53 3785 195 380 743 29 57 BZip2:x5:mt2 818 198 346 683 26 52 3802 198 376 746 29 57 BZip2:x7 292 197 386 759 30 58 3829 196 384 751 29 58 PPMD:x1 724 196 382 749 29 57 511 195 308 602 24 46 PPMD:x5 487 196 421 826 32 63 365 199 345 684 26 52 Delta:4 84996 196 266 522 20 40 75326 197 235 463 18 35 BCJ 211153 197 439 865 34 66 212298 198 439 870 34 67 AES256CBC:1 22275 197 278 547 21 42 23356 197 291 574 22 44 AES256CBC:2 CRC32:1 157406 199 577 1146 44 88 CRC32:4 285672 196 325 638 25 49 CRC32:8 300184 197 206 407 16 31 CRC64 221614 197 230 454 18 35 SHA256 29874 197 309 609 24 47 SHA1 70338 197 334 658 26 50 BLAKE2sp 41839 197 467 920 36 71 CPU 197 1244 2456 95 188 ------------------------------------------------------ Tot: 197 448 886 34 68

Links

BRCM 5000 at EE Times

Broadcom Shows off New Cpu (Linley Gwennap)