========= INFO ========
memcpy test by Siarhei Siamashka <siarhei dot siamashka at nokia dot com>
http://sourceware.org/ml/libc-ports/2009-07/msg00000.html <- sources of the test are located there

The neon parts were removed, since not all the hardware tested below have neon

========= Marvell Sheevaplug (Kirkwood DDR2-800(400MHz)) =========
--- Running benchmarks (average case/perfect alignment case) ---

very small data test:
memcpy_arm  :  (3 bytes copy) =  101.4 MB/s /  103.5 MB/s
memcpy_arm  :  (4 bytes copy) =   90.1 MB/s /   71.6 MB/s
memcpy_arm  :  (5 bytes copy) =   96.0 MB/s /   89.6 MB/s
memcpy_arm  :  (7 bytes copy) =  106.8 MB/s /  121.7 MB/s
memcpy_arm  :  (8 bytes copy) =  118.4 MB/s /  133.7 MB/s
memcpy_arm  :  (11 bytes copy) =  146.0 MB/s /  183.2 MB/s
memcpy_arm  :  (12 bytes copy) =  156.8 MB/s /  199.9 MB/s
memcpy_arm  :  (15 bytes copy) =  186.6 MB/s /  243.1 MB/s
memcpy_arm  :  (16 bytes copy) =  192.2 MB/s /  259.4 MB/s
memcpy_arm  :  (24 bytes copy) =  258.9 MB/s /  368.7 MB/s
memcpy_arm  :  (31 bytes copy) =  303.3 MB/s /  453.3 MB/s

L1 cached data:
memcpy_arm  :  (4096 bytes copy) = 1044.2 MB/s / 2283.8 MB/s
memcpy_arm  :  (6144 bytes copy) = 1049.8 MB/s / 2289.4 MB/s

L2 cached data:
memcpy_arm  :  (65536 bytes copy) =  764.6 MB/s / 1063.0 MB/s
memcpy_arm  :  (98304 bytes copy) =  668.2 MB/s /  879.8 MB/s

SDRAM:
memcpy_arm  :  (2097152 bytes copy) =  267.7 MB/s /  293.1 MB/s
memcpy_arm  :  (3145728 bytes copy) =  252.8 MB/s /  275.6 MB/s


========= Genesi EfikaMX TO3 (i.mx515 DDR2-400(200MHz)) =========

--- Running benchmarks (average case/perfect alignment case) ---

very small data test:
memcpy_arm  :  (3 bytes copy) =  111.1 MB/s /  114.9 MB/s
memcpy_arm  :  (4 bytes copy) =   85.9 MB/s /   90.5 MB/s
memcpy_arm  :  (5 bytes copy) =   97.3 MB/s /  116.7 MB/s
memcpy_arm  :  (7 bytes copy) =  118.9 MB/s /  162.6 MB/s
memcpy_arm  :  (8 bytes copy) =  131.3 MB/s /  170.9 MB/s
memcpy_arm  :  (11 bytes copy) =  164.3 MB/s /  233.8 MB/s
memcpy_arm  :  (12 bytes copy) =  176.2 MB/s /  233.9 MB/s
memcpy_arm  :  (15 bytes copy) =  202.4 MB/s /  292.9 MB/s
memcpy_arm  :  (16 bytes copy) =  211.1 MB/s /  284.7 MB/s
memcpy_arm  :  (24 bytes copy) =  272.6 MB/s /  361.9 MB/s
memcpy_arm  :  (31 bytes copy) =  315.0 MB/s /  434.8 MB/s

L1 cached data:
memcpy_arm  :  (4096 bytes copy) = 1322.6 MB/s / 2224.1 MB/s
memcpy_arm  :  (6144 bytes copy) = 1338.0 MB/s / 2250.1 MB/s

L2 cached data:
memcpy_arm  :  (65536 bytes copy) =  919.4 MB/s / 1184.5 MB/s
memcpy_arm  :  (98304 bytes copy) =  900.8 MB/s / 1162.1 MB/s

SDRAM:
memcpy_arm  :  (2097152 bytes copy) =  212.6 MB/s /  193.4 MB/s
memcpy_arm  :  (3145728 bytes copy) =  211.7 MB/s /  223.9 MB/s

(*) 1 MB = 1000000 bytes
(*) 'memcpy_arm' - an implementation for older ARM cores from glibc-ports



========= Nvidia Harmony devboard (Nvidia Tegra2 DDR2-667(333MHz)) ==========
--- Running benchmarks (average case/perfect alignment case) ---

very small data test:
memcpy_arm  :  (3 bytes copy) =  200.8 MB/s /  197.3 MB/s
memcpy_arm  :  (4 bytes copy) =  111.1 MB/s /  139.3 MB/s
memcpy_arm  :  (5 bytes copy) =  123.9 MB/s /  181.5 MB/s
memcpy_arm  :  (7 bytes copy) =  146.9 MB/s /  278.0 MB/s
memcpy_arm  :  (8 bytes copy) =  162.9 MB/s /  240.5 MB/s
memcpy_arm  :  (11 bytes copy) =  209.5 MB/s /  369.3 MB/s
memcpy_arm  :  (12 bytes copy) =  223.7 MB/s /  342.2 MB/s
memcpy_arm  :  (15 bytes copy) =  268.9 MB/s /  471.8 MB/s
memcpy_arm  :  (16 bytes copy) =  280.6 MB/s /  442.2 MB/s
memcpy_arm  :  (24 bytes copy) =  380.8 MB/s /  596.9 MB/s
memcpy_arm  :  (31 bytes copy) =  440.6 MB/s /  764.2 MB/s

L1 cached data:
memcpy_arm  :  (4096 bytes copy) = 1723.1 MB/s / 3415.6 MB/s
memcpy_arm  :  (6144 bytes copy) = 1746.8 MB/s / 3456.8 MB/s

L2 cached data:
memcpy_arm  :  (65536 bytes copy) = 1600.2 MB/s / 1863.3 MB/s
memcpy_arm  :  (98304 bytes copy) = 1603.3 MB/s / 1846.3 MB/s

SDRAM:
memcpy_arm  :  (2097152 bytes copy) =  345.2 MB/s /  342.5 MB/s
memcpy_arm  :  (3145728 bytes copy) =  347.6 MB/s /  344.0 MB/s

(*) 1 MB = 1000000 bytes
(*) 'memcpy_arm' - an implementation for older ARM cores from glibc-ports

========= TI OMAP4 Pandaboard (DDR2-400(200MHz, Rev EA1)) ==========
(Note: The kernel used was a L24.9 kernel from ubuntu)

--- Running benchmarks (average case/perfect alignment case) ---

very small data test:
memcpy_arm  :  (3 bytes copy) =  122.0 MB/s /  118.8 MB/s
memcpy_arm  :  (4 bytes copy) =   69.7 MB/s /   86.3 MB/s
memcpy_arm  :  (5 bytes copy) =   79.1 MB/s /  115.1 MB/s
memcpy_arm  :  (7 bytes copy) =   91.9 MB/s /  166.7 MB/s
memcpy_arm  :  (8 bytes copy) =  100.4 MB/s /  154.9 MB/s
memcpy_arm  :  (11 bytes copy) =  131.6 MB/s /  238.2 MB/s
memcpy_arm  :  (12 bytes copy) =  140.1 MB/s /  219.7 MB/s
memcpy_arm  :  (15 bytes copy) =  167.9 MB/s /  305.5 MB/s
memcpy_arm  :  (16 bytes copy) =  175.9 MB/s /  276.7 MB/s
memcpy_arm  :  (24 bytes copy) =  234.2 MB/s /  373.5 MB/s
memcpy_arm  :  (31 bytes copy) =  271.9 MB/s /  489.2 MB/s

L1 cached data:
memcpy_arm  :  (4096 bytes copy) = 1106.2 MB/s / 2054.1 MB/s
memcpy_arm  :  (6144 bytes copy) = 1118.0 MB/s / 2078.2 MB/s

L2 cached data:
memcpy_arm  :  (65536 bytes copy) = 1019.2 MB/s / 1184.3 MB/s
memcpy_arm  :  (98304 bytes copy) = 1019.0 MB/s / 1176.9 MB/s

SDRAM:
memcpy_arm  :  (2097152 bytes copy) =  185.1 MB/s /  203.1 MB/s
memcpy_arm  :  (3145728 bytes copy) =  184.6 MB/s /  215.5 MB/s

(*) 1 MB = 1000000 bytes
(*) 'memcpy_arm' - an implementation for older ARM cores from glibc-ports


========= TI OMAP4 Pandaboard (DDR2-800(400MHz, Rev A1, provided by Gustavo Sverzut)) ==========
(Note: The kernel used was a L24.9 kernel from ubuntu)

--- Running benchmarks (average case/perfect alignment case) ---

very small data test:
memcpy_arm  :  (3 bytes copy) =  121.9 MB/s /  118.8 MB/s
memcpy_arm  :  (4 bytes copy) =   71.0 MB/s /   86.2 MB/s
memcpy_arm  :  (5 bytes copy) =   78.8 MB/s /  114.8 MB/s
memcpy_arm  :  (7 bytes copy) =   90.7 MB/s /  166.1 MB/s
memcpy_arm  :  (8 bytes copy) =  101.0 MB/s /  154.3 MB/s
memcpy_arm  :  (11 bytes copy) =  131.3 MB/s /  235.3 MB/s
memcpy_arm  :  (12 bytes copy) =  139.4 MB/s /  219.6 MB/s
memcpy_arm  :  (15 bytes copy) =  166.8 MB/s /  302.1 MB/s
memcpy_arm  :  (16 bytes copy) =  173.3 MB/s /  275.1 MB/s
memcpy_arm  :  (24 bytes copy) =  232.4 MB/s /  373.4 MB/s
memcpy_arm  :  (31 bytes copy) =  272.9 MB/s /  488.3 MB/s

L1 cached data:
memcpy_arm  :  (4096 bytes copy) = 1101.2 MB/s / 2048.0 MB/s
memcpy_arm  :  (6144 bytes copy) = 1112.7 MB/s / 2071.8 MB/s

L2 cached data:
memcpy_arm  :  (65536 bytes copy) = 1017.4 MB/s / 1185.7 MB/s
memcpy_arm  :  (98304 bytes copy) = 1016.1 MB/s / 1174.1 MB/s

SDRAM:
memcpy_arm  :  (2097152 bytes copy) =  255.3 MB/s /  282.5 MB/s
memcpy_arm  :  (3145728 bytes copy) =  255.1 MB/s /  284.7 MB/s

(*) 1 MB = 1000000 bytes
(*) 'memcpy_arm' - an implementation for older ARM cores from glibc-ports


========= Trimslice Dev-Kit (Nvidia Tegra2 DDR2-667(333MHz)) =========

--- Running benchmarks (average case/perfect alignment case) ---

very small data test:
memcpy_arm  :  (3 bytes copy) =  186.7 MB/s /  182.2 MB/s
memcpy_arm  :  (4 bytes copy) =  107.4 MB/s /  131.8 MB/s
memcpy_arm  :  (5 bytes copy) =  123.5 MB/s /  176.3 MB/s
memcpy_arm  :  (7 bytes copy) =  141.3 MB/s /  255.3 MB/s
memcpy_arm  :  (8 bytes copy) =  157.0 MB/s /  237.9 MB/s
memcpy_arm  :  (11 bytes copy) =  199.6 MB/s /  361.9 MB/s
memcpy_arm  :  (12 bytes copy) =  213.7 MB/s /  336.6 MB/s
memcpy_arm  :  (15 bytes copy) =  254.8 MB/s /  464.2 MB/s
memcpy_arm  :  (16 bytes copy) =  267.8 MB/s /  424.9 MB/s
memcpy_arm  :  (24 bytes copy) =  354.8 MB/s /  573.7 MB/s
memcpy_arm  :  (31 bytes copy) =  415.0 MB/s /  748.2 MB/s

L1 cached data:
memcpy_arm  :  (4096 bytes copy) = 1684.2 MB/s / 3152.9 MB/s
memcpy_arm  :  (6144 bytes copy) = 1705.6 MB/s / 3183.2 MB/s

L2 cached data:
memcpy_arm  :  (65536 bytes copy) = 1510.3 MB/s / 1719.2 MB/s
memcpy_arm  :  (98304 bytes copy) = 1511.7 MB/s / 1695.4 MB/s

SDRAM:
memcpy_arm  :  (2097152 bytes copy) =  364.6 MB/s /  376.6 MB/s
memcpy_arm  :  (3145728 bytes copy) =  382.5 MB/s /  397.5 MB/s

(*) 1 MB = 1000000 bytes
(*) 'memcpy_arm' - an implementation for older ARM cores from glibc-ports