Notes

[Sciroccogti]

Notes

[Sciroccogti]

Benchmark

128*64 multiply 64*1 for 4096000 times

Library	CPU	int64 Time	int32 Time
XSIMD	Xeon 6230	2.019s
MIPP	Xeon 6230	1.538s	1.098s
MIPP sequential	Xeon 6230	5.868s
pureAVX	Xeon 6230	0.324s
CPU	Xeon 6230	0.367s
XSIMD	Ryzen 3700X	7.165s
MIPP	Ryzen 3700X	5.977s	2.620s
MIPP sequential	Ryzen 3700X	9.022s
pureAVX	Ryzen 3700X	0.584s
CPU	Ryzen 3700X	0.459s

128*64 multiply 64*1 for 4096000 times but skip results heavier than 22

Library	CPU	Time
MIPP	Xeon 6230	1.153s
MIPP sequential	Xeon 6230	2.413s
CPU	Xeon 6230	0.192s
MIPP	Ryzen 3700X	2.423s
MIPP sequential	Ryzen 3700X	3.535s
CPU	Ryzen 3700X	0.163s

64*1 multiply 64*1 for 128*4096000 times

Library	CPU	Time
MIPP	Ryzen 3700X	0.355s
CPU	Ryzen 3700X	0.170s

[Sciroccogti]

Binary LDPC Benchmark

NMS

using 10GBPS-ETHERNET_1723_2048.alist and NMS, iter_max = 30, FEcount = 100, comparing with aff3ct aff3ct's result

the result in the url just didn't use the specified matrix! Damn!

factor	SNR	BER	FER	aff3ct BER	aff3ct FER
0.5	0.00	1.15e-01	1.00e+00	1.04e-01	1.00e+00
0.5	0.50	1.07e-01	1.00e+00	9.05e-02	1.00e+00
0.5	1.00	9.58e-02	1.00e+00	7.90e-02	1.00e+00
0.5	1.50	8.35e-02	1.00e+00	6.67e-02	1.00e+00
0.5	2.00	7.04e-02	1.00e+00	5.53e-02	1.00e-01
0.5	2.50	6.02e-02	1.00e+00	4.37e-02	1.00e-01
0.5	3.00	4.18e-02	9.90e-01	2.79e-02	8.96e-01
0.5	3.50	1.05e-02	3.06e-01	4.49e-03	1.79e-01
0.5	4.00	2.50e-04	1.18e-02	3.31e-05	1.50e-03

SPA

using 10GBPS-ETHERNET_1723_2048.alist and SPA, iter_max = 100, FEcount = 100, comparing with aff3ct

factor	SNR	BER	FER	aff3ct BER	aff3ct FER
0.5	0.00	9.75e-02	1.00e+00	1.01e-01	1.00e+00
0.5	0.50	8.43e-02	1.00e+00	8.81e-02	1.00e+00
0.5	1.00	7.17e-02	1.00e+00	7.63e-02	1.00e+00
0.5	1.50	6.00e-02	1.00e+00	6.42e-02	1.00e+00
0.5	2.00	4.99e-02	1.00e+00	5.24e-02	1.00e-01
0.5	2.50	4.18e-02	1.00e+00	4.15e-02	1.00e-01
0.5	3.00	3.36e-02	1.00e+00	2.67e-02	8.49e-01

[Sciroccogti]

Binary LDPC Multicore

• 1 Thread: 126s every 100 FE
• 15 Thread: 12.84s every 100 FE (in fact 115 FE)

[Sciroccogti]

NB-LDPC Benchmark

EMS

using LDPC_N96_K48_GF256_d1_exp.alist, n_max = 3, n_c = 2, factor = 1, BPSK, iter_max = 10, FEcount = 100, comparing with NBLDPC

SNR	BER	SER	FER	NBLDPC BER	NBLDPC SER	NBLDPC FER
0.0		9.82e-01	9.90e+00	3.02e-01	9.52e-01	1.00e+00
0.5		9.64e-01	9.80e+00	2.94e-01	9.42e-01	9.93e-01
1.0		9.77e-01	9.90e-01	2.83e-01	9.25e-01	9.80e-01
1.5		8.72e-01	8.85e-01	2.68e-01	8.96e-01	9.58e-01
2.0		8.85e-01	8.93e-01	2.49e-01	8.29e-01	8.86e-01
2.5		8.48e-01	8.55e-01	2.15e-01	7.33e-01	7.89e-01
3.0		7.08e-01	7.14e-01	1.77e-01	6.02e-01	6.53e-01
3.5		3.10e-01	3.11e-01	1.23e-01	4.25e-01	4.69e-01
4.0		2.98e-01	2.99e-01	8.41e-02	2.84e-01	3.15e-01
4.5		1.38e-01	1.39e-01	4.40e-02	1.45e-01	1.71e-01
5.0		7.70e-02	7.76e-02	2.01e-02	6.85e-02	7.74e-02
5.5		2.15e-02	2.20e-02	6.19e-03	2.17e-02	2.43e-02
6.0		9.96e-03	9.99e-03	1.81e-03	6.43e-03	7.75e-03
6.5		4.86e-03	4.93e-03	5.30e-04	1.76e-03	2.18e-03
7.0		4.97e-04	5.07e-04	9.03e-05	3.02e-04	3.95e-04

[Sciroccogti]

15dd21f is 2s every 100 frames, and I manually optimized to 0.32s after 4c4e882, but gcc got it to 0.07s in 3b8143d !

Holy!