MIPS 24K

Qualcomm Atheros QCA9533 (MIPS 24Kc), 650MHz, 32 MB (16-bit DDR2 195MHz CL3 x 2). TP-Link TL-WR841N.

L1 Data cache = 32 KB. 32 B/line. 4-way, VIPT.
L1 Instruction cache = ? KB, 32 B/line, 4-WAY, VIPT.
DTLB L1 size = 8 items, 1 page per item.
JTLB size = 16 items, 2 pages per item.

Instruction Cache: Atag entry holds 20 bits of physical address, a valid bit, a lock bit, and an optional parity bit per way. The instruction data entry holds two instructions (64 bits) per way, as well as 6 bits of pre-decode information to speed the decode of branch and jump instructions, and 9 optional parity bits (one per data byte plus one more for the pre-decode information).

L1 Data Cache Latency = 3 cycles.
L1 Data Cache Latency = 5 cycles for access with complex address calculation (size_t n, *p; n = p[n]). MIPS ISA doesn't support complex address modes in LOAD instruction.
RAM Latency = 3 cycles + 128 ns (open RAM page, inside 512 bytes block).
RAM Latency = 3 cycles + 170 ns (random page).
DTLB miss penalty = 4 cycles, Parallel miss: 2.3 cycles per read access.
JTLB miss penalty = 42 cycles, Parallel miss: 43 cycles per read access.

4 KB pages

  32 K     3                              TLB + L1
  64 K     5 +  90 ns       2   90 ns     + 4 DTLB miss + 170 ns RAM
 128 K     7 + 130 ns       2   40 ns
 256 K    29 + 150 ns      22   20 ns     + 42 (JTLB miss)     
 512 K    39 + 160 ns      10 + 10 ns      
   1 M    44 + 164 ns       5 +  5 ns
   2 M    47 + 166 ns       3 +  2 ns               
   4 M    48 + 168 ns       1 +  2 ns               
   8 M    49 + 175 ns       1 +  7 ns     
  16 M    49 + 210 ns       1 +  35 ns    + 170 ns RAM (Page walk to RAM)

4-bytes range cross penalty = 302 cycles
L1 B/W (Parallel Random Read) = 1 cycle per one access
RAM Read B/W (Parallel Random Read) = 59 ns / cache line
RAM Read B/W (Read, 4 Bytes step) = 220 MB/s
RAM Read B/W (Read, 32 Bytes step) = 580 MB/s
RAM Read B/W (Read, 32 Bytes step, pointer-chasing) = 240 MB/s (no hardware prefetch)
RAM Write (4-32 Bytes step) = 155 MB/s

Branch misprediction penalty = 4 cycles.

Integer pipeline:

#	Stage	Name	Description
1	IF	Instruction Fetch First	I-cache tag/data arrays accessed Branch History Table accessed ITLB address translation performed EJTAG break/watch compares done
2	IS	Instruction Fetch Second	Detect I-cache hit Way select MIPS32 Branch prediction
3	RF	Register File Access	Register File access Instruction decoding/dispatch logic Bypass muxes
4	AG	Address Generation	D-cache Address Generation Bypass muxes
5	EX	Execute/Memory Access	Skewed ALU DTLB Start DCache access Branch Resolution
6	MS	Memory Access Second	Complete DCache access DCache hit detection Way select mux Load align
7	ER	Exception Resolution	Instruction completion Register file write setup Exception processing
8	WB	Writeback	Register file writeback occurs on rising edge of this cycle

MIPS16

There are some decoding or execution problems for MIPS16 code. The execution rate for sequence of simple instructions is only about 0.75 IPC (3 instructions per 4 cycles) for MIPS16 code. Why?

7-Zip Benchmark

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

LZMA Dictionary is reduced, so some rating values are not accurate.


# gcc-6 -O2 -march=24kc

7-Zip (a) 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=C.UTF-8,Utf16=on,HugeFiles=on,32 bits,1 CPU BE)

freq= 650
BE
CPU Freq:   638   639   639   639   634   636   638   636

RAM size:      28 MB,  # CPU hardware threads:   1
RAM usage:     16 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

18:        255   100    227    227  |       5558   100    450    449
19:        247   100    222    222  |       5485   100    450    449
20:        239   100    218    218  |       5413   100    448    449
----------------------------------  | ------------------------------
Avr:             100    222    222  |              100    449    449
Tot:             100    336    336


7-Zip (a) 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=C.UTF-8,Utf16=on,HugeFiles=on,32 bits,1 CPU BE)

freq= 650
BE
CPU Freq:   629   639   639   639   634   639   636   637

RAM size:      28 MB,  # CPU hardware threads:   1
Dictionary reduced to:  19
RAM usage:     16 MB,  # Benchmark threads:      1


Method           Speed Usage    R/U Rating   E/U Effec
                 KiB/s     %   MIPS   MIPS     %     %

CPU                      100    637    637    98    98
CPU                      100    637    638    98    98
CPU                      100    637    637    98    98

LZMA:x1            677   100    248    248    38    38
                  5426   100    438    438    67    67
LZMA:x5:mt1        248   100    310    310    48    48
                  5508   100    439    439    68    67
LZMA:x5:mt2        251   100    314    314    48    48
                  5508   100    438    439    67    67
Deflate:x1        1452   100    184    184    28    28
                 11208   100    348    348    54    54
Deflate:x5         760   100    293    293    45    45
                 11226   100    348    348    54    54
Deflate:x7         324   100    360    360    55    55
                 11280   100    350    350    54    54
Deflate64:x5       689   100    298    298    46    46
                 11057   100    344    344    53    53
BZip2:x1           348   100    211    211    32    32
                  1806   100    196    196    30    30
BZip2:x5           317   100    265    265    41    41
                  1635   100    316    316    49    49
BZip2:x5:mt2       316   100    264    264    41    41
                  1638   100    316    316    49    49
BZip2:x7           105   100    273    273    42    42
                  1646   100    317    318    49    49
PPMD:x1            243   100    252    252    39    39
                   206   100    243    243    37    37
PPMD:x5            183   100    310    310    48    48
                   163   100    307    307    47    47
Delta:4          24107   100    148    148    23    23
                 23787   100    146    146    22    22
BCJ              36209   100    148    148    23    23
                 37010   100    152    152    23    23
AES256CBC:1       5988   100    147    147    23    23
                  6105   100    150    150    23    23
AES256CBC:2 

CRC32:1          50659   100    369    369    57    57
CRC32:4          64931   100    145    145    22    22
CRC32:8          68413   100     93     93    14    14
CRC64            53828   100    110    110    17    17
SHA256           12855   100    262    262    40    40
SHA1             25031   100    234    234    36    36
BLAKE2sp         14221   100    313    313    48    48

CPU                      100    638    637    98    98
------------------------------------------------------
Tot:                     100    321    321    49    49




# gcc-6 -O2 -march=24kc -mips16

7-Zip (a) 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=C.UTF-8,Utf16=on,HugeFiles=on,32 bits,1 CPU BE)

freq= 650
BE
CPU Freq:   469   469   464   469   469   468   467

RAM size:      28 MB,  # CPU hardware threads:   1
RAM usage:     16 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

18:        160   100    143    143  |       2398   100    194    194
19:        157   100    141    141  |       2381   100    195    195
20:        153   100    139    140  |       2367   100    196    196
----------------------------------  | ------------------------------
Avr:             100    141    141  |              100    195    195
Tot:             100    168    168


7-Zip (a) 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=C.UTF-8,Utf16=on,HugeFiles=on,32 bits,1 CPU BE)

freq= 650
BE
CPU Freq:   469   464   469   469   467   468   468

RAM size:      28 MB,  # CPU hardware threads:   1
Dictionary reduced to:  19
RAM usage:     16 MB,  # Benchmark threads:      1


Method           Speed Usage    R/U Rating   E/U Effec
                 KiB/s     %   MIPS   MIPS     %     %

CPU                      100    468    468    72    72
CPU                      100    468    468    72    72
CPU                      100    469    468    72    72

LZMA:x1            460   100    168    168    26    26
                  2324   100    188    188    29    29
LZMA:x5:mt1        157   100    197    197    30    30
                  2388   100    190    190    29    29
LZMA:x5:mt2        171   100    214    214    33    33
                  2389   100    190    190    29    29
Deflate:x1        1136   100    144    144    22    22
                  6164   100    191    191    29    29
Deflate:x5         493   100    190    190    29    29
                  6171   100    192    191    29    29
Deflate:x7         189   100    210    210    32    32
                  6204   100    192    192    30    30
Deflate64:x5       459   100    199    198    31    31
                  6161   100    192    192    29    29
BZip2:x1           234   100    142    142    22    22
                  1113   100    121    121    19    19
BZip2:x5           217   100    182    182    28    28
                  1027   100    198    198    30    30
BZip2:x5:mt2       217   100    181    181    28    28
                  1024   100    198    198    30    30
BZip2:x7            63   100    165    165    25    25
                  1033   100    199    199    31    31
PPMD:x1            145   100    150    150    23    23
                   122   100    145    145    22    22
PPMD:x5            122   100    208    208    32    32
                   107   100    201    201    31    31
Delta:4          15731   100     97     97    15    15
                 16164   100     99     99    15    15
BCJ              23122   100     95     95    15    15
                 23446   100     96     96    15    15
AES256CBC:1       3236   100     80     80    12    12
                  2905   100     71     71    11    11
AES256CBC:2 

CRC32:1          30741   100    224    224    34    34
CRC32:4          34787   100     78     78    12    12
CRC32:8          42049   100     57     57     9     9
CRC64            28638   100     59     59     9     9
SHA256            4488   100     92     92    14    14
SHA1             10408   100     97     97    15    15
BLAKE2sp          5431   100    120    119    18    18

CPU                      100    468    468    72    72
------------------------------------------------------
Tot:                     100    177    177    27    27

Links

MIPS Classic Processor Cores : MIPS32 24K