Platform/GFX/MemoryBandwidth

From MozillaWiki
< Platform‎ | GFX
Jump to navigation Jump to search

Description

==========================================================================
== Memory latency test                                                  ==
==                                                                      ==
== Average time is measured for random memory accesses in the buffers   ==
== of different sizes. The larger is the buffer, the more significant   ==
== are relative contributions of TLB, L1/L2 cache misses and SDRAM      ==
== accesses. For extremely large buffer sizes we are expecting to see   ==
== page table walk with several requests to SDRAM for almost every      ==
== memory access (though 64MiB is not nearly large enough to experience ==
== this effect to its fullest).                                         ==
==                                                                      ==
== Note 1: All the numbers are representing extra time, which needs to  ==
==         be added to L1 cache latency. The cycle timings for L1 cache ==
==         latency can be usually found in the processor documentation. ==
== Note 2: Dual random read means that we are simultaneously performing ==
==         two independent memory accesses at a time. In the case if    ==
==         the memory subsystem can't handle multiple outstanding       ==
==         requests, dual random read has the same timings as two       ==
==         single reads performed one after another.                    ==
==========================================================================

==========================================================================
== Memory bandwidth tests                                               ==
==                                                                      ==
== Note 1: 1MB = 1000000 bytes                                          ==
== Note 2: Results for 'copy' tests show how many bytes can be          ==
==         copied per second (adding together read and writen           ==
==         bytes would have provided twice higher numbers)              ==
== Note 3: 2-pass copy means that we are using a small temporary buffer ==
==         to first fetch data into it, and only then write it to the   ==
==         destination (source -> L1 cache, L1 cache -> destination)    ==
== Note 4: If sample standard deviation exceeds 0.1%, it is shown in    ==
==         brackets                                                     ==
==========================================================================

MacBookPro


==========================================================================
== Memory bandwidth tests                                               ==
==========================================================================

 C copy backwards                                     :   9805.2 MB/s (2.5%)
 C copy                                               :   9832.3 MB/s (6.9%)
 C copy prefetched (32 bytes step)                    :   9556.3 MB/s (2.5%)
 C copy prefetched (64 bytes step)                    :   9532.4 MB/s (2.1%)
 C 2-pass copy                                        :   9445.8 MB/s (1.6%)
 C 2-pass copy prefetched (32 bytes step)             :   9120.2 MB/s (2.7%)
 C 2-pass copy prefetched (64 bytes step)             :   9102.7 MB/s (2.2%)
 C fill                                               :  16685.5 MB/s (4.3%)
 ---
 standard memcpy                                      :  11950.6 MB/s (1.8%)
 standard memset                                      :  23193.5 MB/s (1.9%)

==========================================================================
== Memory latency test                                                  ==
==========================================================================

block size : single random read / dual random read
      1024 :    0.0 ns          /     0.0 ns 
      2048 :    0.0 ns          /     0.0 ns 
      4096 :    0.0 ns          /     0.0 ns 
      8192 :    0.0 ns          /     0.0 ns 
     16384 :    0.0 ns          /     0.0 ns 
     32768 :    0.0 ns          /     0.0 ns 
     65536 :    1.0 ns          /     1.2 ns 
    131072 :    1.5 ns          /     1.7 ns 
    262144 :    1.8 ns          /     2.0 ns 
    524288 :    6.4 ns          /     8.0 ns 
   1048576 :    8.9 ns          /    10.2 ns 
   2097152 :   10.7 ns          /    11.3 ns 
   4194304 :   13.7 ns          /    14.0 ns 
   8388608 :   36.8 ns          /    45.9 ns 
  16777216 :   42.8 ns          /    47.1 ns 
  33554432 :   54.9 ns          /    49.5 ns 
  67108864 :   81.9 ns          /    94.9 ns


Flame


==========================================================================
== Memory bandwidth tests                                               ==
==========================================================================

 C copy backwards                                     :    337.7 MB/s (0.9%)
 C copy                                               :    999.7 MB/s (0.6%)
 C copy prefetched (32 bytes step)                    :    866.7 MB/s (0.4%)
 C copy prefetched (64 bytes step)                    :    867.4 MB/s (0.7%)
 C 2-pass copy                                        :    660.4 MB/s (0.3%)
 C 2-pass copy prefetched (32 bytes step)             :    613.3 MB/s
 C 2-pass copy prefetched (64 bytes step)             :    613.8 MB/s (0.1%)
 C fill                                               :   2501.7 MB/s (0.2%)
 ---
 standard memcpy                                      :   1017.0 MB/s (0.4%)
 standard memset                                      :   2494.4 MB/s (0.3%)
 ---
 NEON read                                            :   1817.7 MB/s (0.4%)
 NEON read prefetched (32 bytes step)                 :   1899.6 MB/s (0.3%)
 NEON read prefetched (64 bytes step)                 :   1915.5 MB/s (0.2%)
 NEON read 2 data streams                             :    550.9 MB/s (0.1%)
 NEON read 2 data streams prefetched (32 bytes step)  :   1015.1 MB/s (0.1%)
 NEON read 2 data streams prefetched (64 bytes step)  :   1087.3 MB/s (0.1%)
 NEON copy                                            :   1016.5 MB/s (0.5%)
 NEON copy prefetched (32 bytes step)                 :   1042.1 MB/s (0.3%)
 NEON copy prefetched (64 bytes step)                 :   1063.6 MB/s (0.4%)
 NEON unrolled copy                                   :    993.6 MB/s (0.3%)
 NEON unrolled copy prefetched (32 bytes step)        :    961.7 MB/s (0.3%)
 NEON unrolled copy prefetched (64 bytes step)        :   1007.5 MB/s
 NEON copy backwards                                  :   1016.1 MB/s (0.4%)
 NEON copy backwards prefetched (32 bytes step)       :   1041.8 MB/s (0.4%)
 NEON copy backwards prefetched (64 bytes step)       :   1068.9 MB/s (0.4%)
 NEON 2-pass copy                                     :    984.4 MB/s (0.4%)
 NEON 2-pass copy prefetched (32 bytes step)          :   1022.3 MB/s (0.3%)
 NEON 2-pass copy prefetched (64 bytes step)          :   1037.6 MB/s (0.5%)
 NEON unrolled 2-pass copy                            :    881.1 MB/s (0.2%)
 NEON unrolled 2-pass copy prefetched (32 bytes step) :    823.4 MB/s (0.3%)
 NEON unrolled 2-pass copy prefetched (64 bytes step) :    885.6 MB/s (0.3%)
 NEON fill                                            :   2501.0 MB/s (0.3%)
 NEON fill backwards                                  :   2498.9 MB/s (0.4%)
 VFP copy                                             :    996.2 MB/s (0.4%)
 VFP 2-pass copy                                      :    903.9 MB/s (0.3%)
 ARM fill (STRD)                                      :   2498.0 MB/s (0.5%)
 ARM fill (STM with 8 registers)                      :   2499.9 MB/s (0.2%)
 ARM fill (STM with 4 registers)                      :   2498.9 MB/s (0.4%)
 ARM copy prefetched (incr pld)                       :   1060.5 MB/s (0.4%)
 ARM copy prefetched (wrap pld)                       :   1006.0 MB/s
 ARM 2-pass copy prefetched (incr pld)                :   1005.7 MB/s (0.3%)
 ARM 2-pass copy prefetched (wrap pld)                :    984.0 MB/s (0.4%)

==========================================================================
== Memory latency test                                                  ==
==========================================================================

block size : single random read / dual random read
      1024 :    0.0 ns          /     0.0 ns 
      2048 :    0.0 ns          /     0.0 ns 
      4096 :    0.0 ns          /     0.0 ns 
      8192 :    0.0 ns          /     0.0 ns 
     16384 :    0.0 ns          /     0.0 ns 
     32768 :    0.0 ns          /     0.0 ns 
     65536 :    5.3 ns          /     9.1 ns 
    131072 :    8.2 ns          /    12.8 ns 
    262144 :    9.6 ns          /    14.3 ns 
    524288 :   12.8 ns          /    17.4 ns 
   1048576 :   69.1 ns          /   109.5 ns 
   2097152 :  104.4 ns          /   147.4 ns 
   4194304 :  122.6 ns          /   162.9 ns 
   8388608 :  133.1 ns          /   171.6 ns 
  16777216 :  140.9 ns          /   180.5 ns 
  33554432 :  148.5 ns          /   193.0 ns 
  67108864 :  167.0 ns          /   221.8 ns 

Tarako

==========================================================================
== Memory bandwidth tests                                               ==
==========================================================================

 C copy backwards                                     :    121.2 MB/s (0.2%)
 C copy                                               :    370.5 MB/s
 C copy prefetched (32 bytes step)                    :    381.3 MB/s (0.4%)
 C copy prefetched (64 bytes step)                    :    381.6 MB/s (0.7%)
 C 2-pass copy                                        :    425.6 MB/s (0.5%)
 C 2-pass copy prefetched (32 bytes step)             :    407.2 MB/s (0.5%)
 C 2-pass copy prefetched (64 bytes step)             :    406.9 MB/s (0.4%)
 C fill                                               :   1523.0 MB/s (0.6%)
 ---
 standard memcpy                                      :    363.8 MB/s (0.4%)
 standard memset                                      :   1523.4 MB/s (0.7%)
 ---
 NEON read                                            :   1817.7 MB/s (0.4%)
 NEON read prefetched (32 bytes step)                 :   1899.6 MB/s (0.3%)
 NEON read prefetched (64 bytes step)                 :   1915.5 MB/s (0.2%)
 NEON read 2 data streams                             :    550.9 MB/s (0.1%)
 NEON read 2 data streams prefetched (32 bytes step)  :   1015.1 MB/s (0.1%)
 NEON read 2 data streams prefetched (64 bytes step)  :   1087.3 MB/s (0.1%)
 NEON copy                                            :   1016.5 MB/s (0.5%)
 NEON copy prefetched (32 bytes step)                 :   1042.1 MB/s (0.3%)
 NEON copy prefetched (64 bytes step)                 :   1063.6 MB/s (0.4%)
 NEON unrolled copy                                   :    993.6 MB/s (0.3%)
 NEON unrolled copy prefetched (32 bytes step)        :    961.7 MB/s (0.3%)
 NEON unrolled copy prefetched (64 bytes step)        :   1007.5 MB/s
 NEON copy backwards                                  :   1016.1 MB/s (0.4%)
 NEON copy backwards prefetched (32 bytes step)       :   1041.8 MB/s (0.4%)
 NEON copy backwards prefetched (64 bytes step)       :   1068.9 MB/s (0.4%)
 NEON 2-pass copy                                     :    984.4 MB/s (0.4%)
 NEON 2-pass copy prefetched (32 bytes step)          :   1022.3 MB/s (0.3%)
 NEON 2-pass copy prefetched (64 bytes step)          :   1037.6 MB/s (0.5%)
 NEON unrolled 2-pass copy                            :    881.1 MB/s (0.2%)
 NEON unrolled 2-pass copy prefetched (32 bytes step) :    823.4 MB/s (0.3%)
 NEON unrolled 2-pass copy prefetched (64 bytes step) :    885.6 MB/s (0.3%)
 NEON fill                                            :   2501.0 MB/s (0.3%)
 NEON fill backwards                                  :   2498.9 MB/s (0.4%)
 VFP copy                                             :    996.2 MB/s (0.4%)
 VFP 2-pass copy                                      :    903.9 MB/s (0.3%)
 ARM fill (STRD)                                      :   2498.0 MB/s (0.5%)
 ARM fill (STM with 8 registers)                      :   2499.9 MB/s (0.2%)
 ARM fill (STM with 4 registers)                      :   2498.9 MB/s (0.4%)
 ARM copy prefetched (incr pld)                       :   1060.5 MB/s (0.4%)
 ARM copy prefetched (wrap pld)                       :   1006.0 MB/s
 ARM 2-pass copy prefetched (incr pld)                :   1005.7 MB/s (0.3%)
 ARM 2-pass copy prefetched (wrap pld)                :    984.0 MB/s (0.4%)

==========================================================================
== Memory latency test                                                  ==
==========================================================================

block size : single random read / dual random read
      1024 :    0.0 ns          /     0.0 ns 
      2048 :    0.0 ns          /     0.0 ns 
      4096 :    0.0 ns          /     0.0 ns 
      8192 :    0.0 ns          /     0.0 ns 
     16384 :    0.0 ns          /     0.0 ns 
     32768 :    0.0 ns          /     0.0 ns 
     65536 :    5.3 ns          /     9.1 ns 
    131072 :    8.2 ns          /    12.8 ns 
    262144 :    9.6 ns          /    14.3 ns 
    524288 :   12.8 ns          /    17.4 ns 
   1048576 :   69.1 ns          /   109.5 ns 
   2097152 :  104.4 ns          /   147.4 ns 
   4194304 :  122.6 ns          /   162.9 ns 
   8388608 :  133.1 ns          /   171.6 ns 
  16777216 :  140.9 ns          /   180.5 ns 
  33554432 :  148.5 ns          /   193.0 ns 
  67108864 :  167.0 ns          /   221.8 ns