remap: Add stereo to mono and 4-channel special case remapping
The generic matrix remapping is rather inefficient; special-case code
improves performance by 3x easily.
v4: split into s16 and float code, 4-channel remapping
v3: fix remap_mono_to_stereo_c(), use assignment
v2: use consistent array addressing
on Intel Core i7-870 @ 2.93 GHz (GCC 4.6, 64-bit):
Checking special remap (float, mono->stereo)
func: 70392 usec (avg: 703.92, min = 583, max = 1879, stddev = 295.192).
orig: 193042 usec (avg: 1930.42, min = 1457, max = 2269, stddev = 89.9045).
Checking special remap (float, mono->4-channel)
func: 118408 usec (avg: 1184.08, min = 1151, max = 1454, stddev = 57.1244).
orig: 380074 usec (avg: 3800.74, min = 3740, max = 4180, stddev = 96.3389).
Checking special remap (s16, mono->stereo)
func: 60574 usec (avg: 605.74, min = 582, max = 659, stddev = 20.7681).
orig: 188262 usec (avg: 1882.62, min = 1804, max = 2167, stddev = 79.17).
Checking special remap (s16, mono->4-channel)
func: 120331 usec (avg: 1203.31, min = 1151, max = 1429, stddev = 55.2863).
orig: 376028 usec (avg: 3760.28, min = 3609, max = 4096, stddev = 122.043).
Checking special remap (float, stereo->mono)
func: 61408 usec (avg: 614.08, min = 580, max = 867, stddev = 50.933).
orig: 186484 usec (avg: 1864.84, min = 1808, max = 2121, stddev = 65.3967).
Checking special remap (float, 4-channel->mono)
func: 118101 usec (avg: 1181.01, min = 1157, max = 1383, stddev = 36.4474).
orig: 365191 usec (avg: 3651.91, min = 3540, max = 4083, stddev = 117.509).
Checking special remap (s16, stereo->mono)
func: 82908 usec (avg: 829.08, min = 795, max = 953, stddev = 33.3409).
orig: 182565 usec (avg: 1825.65, min = 1774, max = 2117, stddev = 65.5401).
Checking special remap (s16, 4-channel->mono)
func: 132025 usec (avg: 1320.25, min = 1284, max = 1509, stddev = 47.0133).
orig: 363347 usec (avg: 3633.47, min = 3560, max = 4012, stddev = 111.259).
on ARM Cortex-A8 (TI OMAP3 DM3730 @ 1GHz) (Linaro GCC 4.6):
Checking special remap (float, mono->stereo)
func:
1213562 usec (avg: 12135.6, min = 4669, max = 16266, stddev = 2067.64).
orig:
9251927 usec (avg: 92519.3, min = 87372, max = 134216, stddev = 5965.79).
Checking special remap (float, mono->4-channel)
func:
2479550 usec (avg: 24795.5, min = 7507, max = 29358, stddev = 2690.16).
orig:
13186133 usec (avg: 131861, min = 119843, max = 263855, stddev = 27309).
Checking special remap (s16, mono->stereo)
func: 471894 usec (avg: 4718.94, min = 4058, max = 9583, stddev = 1302.7).
orig:
1673826 usec (avg: 16738.3, min = 14679, max = 31342, stddev = 2271.67).
Checking special remap (s16, mono->4-channel)
func: 869508 usec (avg: 8695.08, min = 7019, max = 19165, stddev = 1866.94).
orig:
3317020 usec (avg: 33170.2, min = 29327, max = 47577, stddev = 2029.11).
Checking special remap (float, stereo->mono)
func:
4405182 usec (avg: 44051.8, min = 41443, max = 77912, stddev = 4160.54).
orig:
13245064 usec (avg: 132451, min = 125244, max = 182282, stddev = 8543.93).
Checking special remap (float, 4-channel->mono)
func:
8607974 usec (avg: 86079.7, min = 81909, max = 116608, stddev = 4311.52).
orig:
26326036 usec (avg: 263260, min = 255097, max = 312928, stddev = 10111.5).
Checking special remap (s16, stereo->mono)
func:
1209135 usec (avg: 12091.4, min = 10742, max = 16632, stddev = 1633.88).
orig:
3081515 usec (avg: 30815.2, min = 27008, max = 50537, stddev = 3124.35).
Checking special remap (s16, 4-channel->mono)
func:
1653868 usec (avg: 16538.7, min = 14648, max = 20721, stddev = 1834.52).
orig:
6017854 usec (avg: 60178.5, min = 56061, max = 89569, stddev = 4052.86).
benchmark code will be posted as follow-up patches
Signed-off-by: Peter Meerwald <pmeerw@pmeerw.net>