tinymembench v0.4.9 (simple benchmark for memory throughput and latency)
==========================================================================
== 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 ==
==========================================================================
C copy backwards : 2552.1 MB/s (3.5%)
C copy backwards (32 byte blocks) : 2665.1 MB/s (0.2%)
C copy backwards (64 byte blocks) : 2667.5 MB/s (2.1%)
C copy : 2302.5 MB/s (1.6%)
C copy prefetched (32 bytes step) : 2403.1 MB/s (1.6%)
C copy prefetched (64 bytes step) : 2394.5 MB/s (0.9%)
C 2-pass copy : 1960.7 MB/s
C 2-pass copy prefetched (32 bytes step) : 2154.8 MB/s
C 2-pass copy prefetched (64 bytes step) : 2142.6 MB/s (2.8%)
C fill : 11305.3 MB/s (1.7%)
C fill (shuffle within 16 byte blocks) : 11318.8 MB/s (1.5%)
C fill (shuffle within 32 byte blocks) : 11229.2 MB/s (1.5%)
C fill (shuffle within 64 byte blocks) : 11176.3 MB/s (1.2%)
---
standard memcpy : 1032.2 MB/s (1.9%)
standard memset : 8613.1 MB/s
---
NEON read : 2280.5 MB/s
NEON read prefetched (32 bytes step) : 5871.6 MB/s (1.6%)
NEON read prefetched (64 bytes step) : 5730.2 MB/s
NEON read 2 data streams : 2260.4 MB/s (5.4%)
NEON read 2 data streams prefetched (32 bytes step) : 5981.5 MB/s (1.5%)
NEON read 2 data streams prefetched (64 bytes step) : 6042.1 MB/s
NEON copy : 1243.7 MB/s (1.3%)
NEON copy prefetched (32 bytes step) : 2247.4 MB/s
NEON copy prefetched (64 bytes step) : 2250.8 MB/s
NEON unrolled copy : 1258.4 MB/s (0.7%)
NEON unrolled copy prefetched (32 bytes step) : 3037.2 MB/s (1.5%)
NEON unrolled copy prefetched (64 bytes step) : 3046.7 MB/s
NEON copy backwards : 1335.7 MB/s (2.1%)
NEON copy backwards prefetched (32 bytes step) : 2454.0 MB/s (3.3%)
NEON copy backwards prefetched (64 bytes step) : 2447.8 MB/s (1.2%)
NEON 2-pass copy : 1197.7 MB/s
NEON 2-pass copy prefetched (32 bytes step) : 2155.4 MB/s
NEON 2-pass copy prefetched (64 bytes step) : 2144.5 MB/s (1.2%)
NEON unrolled 2-pass copy : 1257.7 MB/s (1.3%)
NEON unrolled 2-pass copy prefetched (32 bytes step) : 2910.4 MB/s
NEON unrolled 2-pass copy prefetched (64 bytes step) : 2900.7 MB/s (1.5%)
NEON fill : 11270.5 MB/s (1.9%)
NEON fill backwards : 11135.2 MB/s (3.5%)
VFP copy : 1259.4 MB/s (1.3%)
VFP 2-pass copy : 1548.5 MB/s (0.2%)
ARM fill (STRD) : 8639.1 MB/s (5.1%)
ARM fill (STM with 8 registers) : 10517.5 MB/s (11.2%)
ARM fill (STM with 4 registers) : 11000.0 MB/s (16.3%)
ARM copy prefetched (incr pld) : 1959.0 MB/s (14.4%)
ARM copy prefetched (wrap pld) : 1777.1 MB/s (2.8%)
ARM 2-pass copy prefetched (incr pld) : 1874.7 MB/s (1.9%)
ARM 2-pass copy prefetched (wrap pld) : 1644.7 MB/s (2.6%)
==========================================================================
== 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. ==
==========================================================================
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 : 0.6 ns / 0.7 ns
16384 : 0.9 ns / 1.0 ns
32768 : 7.6 ns / 11.2 ns
65536 : 10.9 ns / 14.1 ns
131072 : 12.7 ns / 15.2 ns
262144 : 15.0 ns / 17.4 ns
524288 : 16.3 ns / 18.3 ns
1048576 : 29.0 ns / 44.6 ns
2097152 : 42.4 ns / 67.9 ns
4194304 : 90.0 ns / 134.8 ns
8388608 : 117.0 ns / 162.1 ns
16777216 : 132.1 ns / 175.8 ns
33554432 : 143.0 ns / 187.7 ns
67108864 : 152.7 ns / 198.2 ns