1 /*****************************************************************************
2
3 Downsampler2xF64Sse2.hpp
4 Author: Laurent de Soras, 2020
5
6 --- Legal stuff ---
7
8 This program is free software. It comes without any warranty, to
9 the extent permitted by applicable law. You can redistribute it
10 and/or modify it under the terms of the Do What The Fuck You Want
11 To Public License, Version 2, as published by Sam Hocevar. See
12 http://www.wtfpl.net/ for more details.
13
14 *Tab=3***********************************************************************/
15
16
17
18 #if ! defined (hiir_Downsampler2xF64Sse2_CODEHEADER_INCLUDED)
19 #define hiir_Downsampler2xF64Sse2_CODEHEADER_INCLUDED
20
21
22
23 /*\\\ INCLUDE FILES \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\*/
24
25 #include "hiir/StageProcF64Sse2.h"
26
27 #include <cassert>
28
29
30
31 namespace hiir
32 {
33
34
35
36 /*\\\ PUBLIC \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\*/
37
38
39
40 /*
41 ==============================================================================
42 Name: ctor
43 Throws: Nothing
44 ==============================================================================
45 */
46
47 template <int NC>
Downsampler2xF64Sse2()48 Downsampler2xF64Sse2 <NC>::Downsampler2xF64Sse2 ()
49 {
50 for (int i = 0; i < _nbr_stages + 1; ++i)
51 {
52 _mm_store_pd (_filter [i]._coef, _mm_setzero_pd ());
53 }
54 if (NBR_COEFS < _nbr_stages * 2)
55 {
56 _filter [_nbr_stages]._coef [0] = 1;
57 }
58
59 clear_buffers ();
60 }
61
62
63
64 /*
65 ==============================================================================
66 Name: set_coefs
67 Description:
68 Sets filter coefficients. Generate them with the PolyphaseIir2Designer
69 class.
70 Call this function before doing any processing.
71 Input parameters:
72 - coef_arr: Array of coefficients. There should be as many coefficients as
73 mentioned in the class template parameter.
74 Throws: Nothing
75 ==============================================================================
76 */
77
78 template <int NC>
set_coefs(const double coef_arr[])79 void Downsampler2xF64Sse2 <NC>::set_coefs (const double coef_arr [])
80 {
81 assert (coef_arr != nullptr);
82
83 for (int i = 0; i < NBR_COEFS; ++i)
84 {
85 const int stage = (i / _stage_width) + 1;
86 const int pos = (i ^ 1) & (_stage_width - 1);
87 _filter [stage]._coef [pos] = DataType (coef_arr [i]);
88 }
89 }
90
91
92
93 /*
94 ==============================================================================
95 Name: process_sample
96 Description:
97 Downsamples (x2) one pair of samples, to generate one output sample.
98 Input parameters:
99 - in_ptr: pointer on the two samples to decimate
100 Returns: Samplerate-reduced sample.
101 Throws: Nothing
102 ==============================================================================
103 */
104
105 template <int NC>
process_sample(const double in_ptr[2])106 double Downsampler2xF64Sse2 <NC>::process_sample (const double in_ptr [2])
107 {
108 assert (in_ptr != nullptr);
109
110 auto x = _mm_loadu_pd (in_ptr);
111 StageProcF64Sse2 <_nbr_stages>::process_sample_pos (x, &_filter [0]);
112 x = _mm_add_sd (x, _mm_shuffle_pd (x, x, 1));
113 x = _mm_mul_sd (x, _mm_set_sd (0.5f));
114
115 return _mm_cvtsd_f64 (x);
116 }
117
118
119
120 /*
121 ==============================================================================
122 Name: process_block
123 Description:
124 Downsamples (x2) a block of samples.
125 Input and output blocks may overlap, see assert() for details.
126 Input parameters:
127 - in_ptr: Input array, containing nbr_spl * 2 samples.
128 - nbr_spl: Number of samples to output, > 0
129 Output parameters:
130 - out_ptr: Array for the output samples, capacity: nbr_spl samples.
131 Throws: Nothing
132 ==============================================================================
133 */
134
135 template <int NC>
process_block(double out_ptr[],const double in_ptr[],long nbr_spl)136 void Downsampler2xF64Sse2 <NC>::process_block (double out_ptr [], const double in_ptr [], long nbr_spl)
137 {
138 assert (in_ptr != nullptr);
139 assert (out_ptr != nullptr);
140 assert (out_ptr <= in_ptr || out_ptr >= in_ptr + nbr_spl * 2);
141 assert (nbr_spl > 0);
142
143 const auto half = _mm_set1_pd (0.5f);
144 for (long pos = 0; pos < nbr_spl; ++pos)
145 {
146 auto x = _mm_loadu_pd (in_ptr + pos * 2);
147 StageProcF64Sse2 <_nbr_stages>::process_sample_pos (x, &_filter [0]);
148 x = _mm_add_sd (x, _mm_shuffle_pd (x, x, 1));
149 x = _mm_mul_sd (x, half);
150 out_ptr [pos] = _mm_cvtsd_f64 (x);
151 }
152 }
153
154
155
156 /*
157 ==============================================================================
158 Name: process_sample_split
159 Description:
160 Split (spectrum-wise) in half a pair of samples. The lower part of the
161 spectrum is a classic downsampling, equivalent to the output of
162 process_sample().
163 The higher part is the complementary signal: original filter response
164 is flipped from left to right, becoming a high-pass filter with the same
165 cutoff frequency. This signal is then critically sampled (decimation by 2),
166 flipping the spectrum: Fs/4...Fs/2 becomes Fs/4...0.
167 Input parameters:
168 - in_ptr: pointer on the pair of input samples
169 Output parameters:
170 - low: output sample, lower part of the spectrum (downsampling)
171 - high: output sample, higher part of the spectrum.
172 Throws: Nothing
173 ==============================================================================
174 */
175
176 template <int NC>
process_sample_split(double & low,double & high,const double in_ptr[2])177 void Downsampler2xF64Sse2 <NC>::process_sample_split (double &low, double &high, const double in_ptr [2])
178 {
179 assert (in_ptr != nullptr);
180
181 auto x = _mm_loadu_pd (in_ptr);
182 StageProcF64Sse2 <_nbr_stages>::process_sample_pos (x, &_filter [0]);
183 x = _mm_mul_pd (x, _mm_set1_pd (0.5f));
184 const auto xr = _mm_shuffle_pd (x, x, 1);
185 low = _mm_cvtsd_f64 (_mm_add_sd (xr, x));
186 high = _mm_cvtsd_f64 (_mm_sub_sd (xr, x));
187 }
188
189
190
191 /*
192 ==============================================================================
193 Name: process_block_split
194 Description:
195 Split (spectrum-wise) in half a block of samples. The lower part of the
196 spectrum is a classic downsampling, equivalent to the output of
197 process_block().
198 The higher part is the complementary signal: original filter response
199 is flipped from left to right, becoming a high-pass filter with the same
200 cutoff frequency. This signal is then critically sampled (decimation by 2),
201 flipping the spectrum: Fs/4...Fs/2 becomes Fs/4...0.
202 Input and output blocks may overlap, see assert() for details.
203 Input parameters:
204 - in_ptr: Input array, containing nbr_spl * 2 samples.
205 - nbr_spl: Number of samples for each output, > 0
206 Output parameters:
207 - out_l_ptr: Array for the output samples, lower part of the spectrum
208 (downsampling). Capacity: nbr_spl samples.
209 - out_h_ptr: Array for the output samples, higher part of the spectrum.
210 Capacity: nbr_spl samples.
211 Throws: Nothing
212 ==============================================================================
213 */
214
215 template <int NC>
process_block_split(double out_l_ptr[],double out_h_ptr[],const double in_ptr[],long nbr_spl)216 void Downsampler2xF64Sse2 <NC>::process_block_split (double out_l_ptr [], double out_h_ptr [], const double in_ptr [], long nbr_spl)
217 {
218 assert (in_ptr != nullptr);
219 assert (out_l_ptr != nullptr);
220 assert (out_l_ptr <= in_ptr || out_l_ptr >= in_ptr + nbr_spl * 2);
221 assert (out_h_ptr != nullptr);
222 assert (out_h_ptr <= in_ptr || out_h_ptr >= in_ptr + nbr_spl * 2);
223 assert (out_h_ptr != out_l_ptr);
224 assert (nbr_spl > 0);
225
226 const auto half = _mm_set1_pd (0.5f);
227 for (long pos = 0; pos < nbr_spl; ++pos)
228 {
229 auto x = _mm_loadu_pd (in_ptr + pos * 2);
230 StageProcF64Sse2 <_nbr_stages>::process_sample_pos (x, &_filter [0]);
231 x = _mm_mul_pd (x, half);
232 const auto xr = _mm_shuffle_pd (x, x, 1);
233 out_l_ptr [pos] = _mm_cvtsd_f64 (_mm_add_sd (xr, x));
234 out_h_ptr [pos] = _mm_cvtsd_f64 (_mm_sub_sd (xr, x));
235 }
236 }
237
238
239
240 /*
241 ==============================================================================
242 Name: clear_buffers
243 Description:
244 Clears filter memory, as if it processed silence since an infinite amount
245 of time.
246 Throws: Nothing
247 ==============================================================================
248 */
249
250 template <int NC>
clear_buffers()251 void Downsampler2xF64Sse2 <NC>::clear_buffers ()
252 {
253 for (int i = 0; i < _nbr_stages + 1; ++i)
254 {
255 _mm_store_pd (_filter [i]._mem, _mm_setzero_pd ());
256 }
257 }
258
259
260
261 /*\\\ PROTECTED \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\*/
262
263
264
265 /*\\\ PRIVATE \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\*/
266
267
268
269 } // namespace hiir
270
271
272
273 #endif // hiir_Downsampler2xF64Sse2_CODEHEADER_INCLUDED
274
275
276
277 /*\\\ EOF \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\*/
278