src/common/RTMath.h

/***************************************************************************
 *                                                                         *
 *   LinuxSampler - modular, streaming capable sampler                     *
 *                                                                         *
 *   Copyright (C) 2003, 2004 by Benno Senoner and Christian Schoenebeck   *
 *   Copyright (C) 2005 Christian Schoenebeck                              *
 *                                                                         *
 *   This program is free software; you can redistribute it and/or modify  *
 *   it under the terms of the GNU General Public License as published by  *
 *   the Free Software Foundation; either version 2 of the License, or     *
 *   (at your option) any later version.                                   *
 *                                                                         *
 *   This program is distributed in the hope that it will be useful,       *
 *   but WITHOUT ANY WARRANTY; without even the implied warranty of        *
 *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         *
 *   GNU General Public License for more details.                          *
 *                                                                         *
 *   You should have received a copy of the GNU General Public License     *
 *   along with this program; if not, write to the Free Software           *
 *   Foundation, Inc., 59 Temple Place, Suite 330, Boston,                 *
 *   MA  02111-1307  USA                                                   *
 ***************************************************************************/

#ifndef __RT_MATH_H__
#define __RT_MATH_H__

#include <math.h>
#include <stdint.h>
#include "global.h"

/// Needed for calculating frequency ratio used to pitch a sample
#define TWELVEHUNDREDTH_ROOT_OF_TWO     1.000577789506555

enum implementation_t {
    CPP,
    ASM_X86_MMX_SSE
};

/** @brief Real Time Math Base Class
 *
 * Math functions for real time operation. This base class contains all
 * non-template methods.
 */
class RTMathBase {
    public:
        /**
         * Highly accurate time stamp.
         */
        typedef uint32_t time_stamp_t;

        /**
         * We read the processor's cycle count register as a reference
         * for the real time. These are of course only abstract values
         * with arbitrary time entity, but that's not a problem as long
         * as we calculate relatively.
         */
        static time_stamp_t CreateTimeStamp();

        /**
         * Calculates the frequency ratio for a pitch value given in cents
         * (assuming equal tempered scale of course, divided into 12
         * semitones per octave and 100 cents per semitone).
         *
         * Note: CONFIG_MAX_PITCH (defined in config.h) has to be defined to an
         * appropriate value, otherwise the behavior of this function is
         * undefined, but most probably if CONFIG_MAX_PITCH is too small, the
         * application will crash due to segmentation fault here.
         *
         * @param cents - pitch value in cents (+1200 cents means +1 octave)
         * @returns  frequency ratio (e.g. +2.0 for +1 octave)
         */
        inline static double CentsToFreqRatio(double Cents) {
            int   index_int   = (int) (Cents);      // integer index
            float index_fract = Cents - index_int;  // fractional part of index
            return pCentsToFreqTable[index_int] + index_fract * (pCentsToFreqTable[index_int+1] - pCentsToFreqTable[index_int]);
        }

    private:
        static float  CentsToFreqTable[CONFIG_MAX_PITCH * 1200 * 2 + 1];
        static float* pCentsToFreqTable;

        static float* InitCentsToFreqTable();
};

/** @brief Real Time Math
 *
 * This is a template which provides customized methods for the desired low
 * level implementation. The ASM_X86_MMX_SSE implementation of each method
 * for example doesn't use 387 FPU instruction. This is needed for MMX
 * algorithms which do not allow mixed MMX and 387 instructions.
 */
template<implementation_t IMPL = CPP>
class __RTMath : public RTMathBase {
    public:
        // conversion using truncate
        inline static int Int(const float a) {
            switch (IMPL) {
                case CPP: {
                    return (int) a;
                }
                #if ARCH_X86
                case ASM_X86_MMX_SSE: {
                    int ret;
                    asm (
                        "cvttss2si %1, %0  # convert to int\n\t"
                        : "=r" (ret)
                        : "m" (a)
                    );
                    return ret;
                }
                #endif // ARCH_X86
            }
        }

        //for doubles and everything else except floats
        template<class T_a> inline static int Int(const T_a a) {
            return (int) a;
        }

        inline static float Float(const int a) {
            switch (IMPL) {
                case CPP: {
                    return (float) a;
                }
                #if ARCH_X86
                case ASM_X86_MMX_SSE: {
                    float ret;
                    asm (
                        "cvtsi2ss %1, %%xmm0  # convert to float\n\t"
                        "movss    %%xmm0,%0   # output\n\t"
                        : "=m" (ret)
                        : "r" (a)
                    );
                    return ret;
                }
                #endif // ARCH_X86
            }
        }

#if 0
        //for everything except ints
        template<class T_a> inline static float Float(T_a a) {
            return (float) a;
        }
#endif

        inline static float Sum(const float& a, const float& b) {
            switch (IMPL) {
                case CPP: {
                    return (a + b);
                }
                #if ARCH_X86
                case ASM_X86_MMX_SSE: {
                    float ret;
                    asm (
                        "movss    %1, %%xmm0  # load a\n\t"
                        "addss    %2, %%xmm0  # a + b\n\t"
                        "movss    %%xmm0, %0  # output\n\t"
                        : "=m" (ret)
                        : "m" (a), "m" (b)
                    );
                    return ret;
                }
                #endif // ARCH_X86
            }
        }

        template<class T_a, class T_b> inline static T_a Sum(const T_a a, const T_b b) {
            return (a + b);
        }

        inline static float Sub(const float& a, const float& b) {
            switch (IMPL) {
                case CPP: {
                    return (a - b);
                }
                #if ARCH_X86
                case ASM_X86_MMX_SSE: {
                    float ret;
                    asm (
                        "movss    %1, %%xmm0  # load a\n\t"
                        "subss    %2, %%xmm0  # a - b\n\t"
                        "movss    %%xmm0, %0  # output\n\t"
                        : "=m" (ret)
                        : "m" (a), "m" (b)
                    );
                    return ret;
                }
                #endif // ARCH_X86
            }
        }

        template<class T_a, class T_b> inline static T_a Sub(const T_a a, const T_b b) {
            return (a - b);
        }

        inline static float Mul(const float a, const float b) {
            switch (IMPL) {
                case CPP: {
                    return (a * b);
                }
                #if ARCH_X86
                case ASM_X86_MMX_SSE: {
                    float ret;
                    asm (
                        "movss    %1, %%xmm0  # load a\n\t"
                        "mulss    %2, %%xmm0  # a * b\n\t"
                        "movss    %%xmm0, %0  # output\n\t"
                        : "=m" (ret)
                        : "m" (a), "m" (b)
                    );
                    return ret;
                }
                #endif // ARCH_X86
            }
        }

        template<class T_a, class T_b> inline static T_a Mul(const T_a a, const T_b b) {
            return (a * b);
        }

        inline static float Div(const float a, const float b) {
            switch (IMPL) {
                case CPP: {
                    return (a / b);
                }
                #if ARCH_X86
                case ASM_X86_MMX_SSE: {
                    float ret;
                    asm (
                        "movss    %1, %%xmm0  # load a\n\t"
                        "divss    %2, %%xmm0  # a / b\n\t"
                        "movss    %%xmm0, %0  # output\n\t"
                        : "=m" (ret)
                        : "m" (a), "m" (b)
                    );
                    return ret;
                }
                #endif // ARCH_X86
            }
        }

        template<class T_a, class T_b> inline static T_a Div(const T_a a, const T_b b) {
            return (a / b);
        }

        inline static float Min(const float a, const float b) {
            switch (IMPL) {
                case CPP: {
                    return (b < a) ? b : a;
                }
                #if ARCH_X86
                case ASM_X86_MMX_SSE: {
                    float ret;
                    asm (
                        "movss    %1, %%xmm0  # load a\n\t"
                        "minss    %2, %%xmm0  # Minimum(a, b)\n\t"
                        "movss    %%xmm0, %0  # output\n\t"
                        : "=m" (ret)
                        : "m" (a), "m" (b)
                    );
                    return ret;
                }
                #endif // ARCH_X86
            }
        }

        template<class T_a, class T_b> inline static T_a Min(const T_a a, const T_b b) {
            return (b < a) ? b : a;
        }

        inline static float Max(const float a, const float b) {
            switch (IMPL) {
                case CPP: {
                    return (b > a) ? b : a;
                }
                #if ARCH_X86
                case ASM_X86_MMX_SSE: {
                    float ret;
                    asm (
                        "movss    %1, %%xmm0  # load a\n\t"
                        "maxss    %2, %%xmm0  # Maximum(a, b)\n\t"
                        "movss    %%xmm0, %0  # output\n\t"
                        : "=m" (ret)
                        : "m" (a), "m" (b)
                    );
                    return ret;
                }
                #endif // ARCH_X86
            }
        }

        template<class T_a, class T_b> inline static T_a Max(const T_a a, const T_b b) {
            return (b > a) ? b : a;
        }

        inline static float Fmodf(const float &a, const float &b) {
            switch (IMPL) {
                case CPP: {
                    return fmodf(a, b);
                }
                #if ARCH_X86
                case ASM_X86_MMX_SSE: {
                    float ret;
                    asm (
                        "movss    %1, %%xmm0  # load a\n\t"
                        "movss    %2, %%xmm1  # load b\n\t"
                        "movss    %%xmm0,%%xmm2\n\t"
                        "divss    %%xmm1, %%xmm2  # xmm2 = a / b\n\t"
                        "cvttss2si %%xmm2, %%ecx  #convert to int\n\t"
                        "cvtsi2ss %%ecx, %%xmm2  #convert back to float\n\t"
                        "mulss    %%xmm1, %%xmm2  # xmm2 = b * int(a/b)\n\t"
                        "subss    %%xmm2, %%xmm0  #sub a\n\t"
                        "movss    %%xmm0, %0  # output\n\t"
                        : "=m" (ret)
                        : "m" (a), "m" (b)
                        : "%ecx"
                    );
                    return ret;
                }
                #endif // ARCH_X86
            }
        }
};

/// convenience typedef for using the default implementation (which is CPP)
typedef __RTMath<> RTMath;

#endif // __RT_MATH_H__
1	schoenebeck	53	/***************************************************************************
2			* *
3			* LinuxSampler - modular, streaming capable sampler *
4			* *
5	schoenebeck	56	* Copyright (C) 2003, 2004 by Benno Senoner and Christian Schoenebeck *
6	schoenebeck	554	* Copyright (C) 2005 Christian Schoenebeck *
7	schoenebeck	53	* *
8			* This program is free software; you can redistribute it and/or modify *
9			* it under the terms of the GNU General Public License as published by *
10			* the Free Software Foundation; either version 2 of the License, or *
11			* (at your option) any later version. *
12			* *
13			* This program is distributed in the hope that it will be useful, *
14			* but WITHOUT ANY WARRANTY; without even the implied warranty of *
15			* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the *
16			* GNU General Public License for more details. *
17			* *
18			* You should have received a copy of the GNU General Public License *
19			* along with this program; if not, write to the Free Software *
20			* Foundation, Inc., 59 Temple Place, Suite 330, Boston, *
21			* MA 02111-1307 USA *
22			***************************************************************************/
23
24			#ifndef __RT_MATH_H__
25			#define __RT_MATH_H__
26
27			#include <math.h>
28	schoenebeck	328	#include <stdint.h>
29	schoenebeck	53	#include "global.h"
30
31			/// Needed for calculating frequency ratio used to pitch a sample
32			#define TWELVEHUNDREDTH_ROOT_OF_TWO 1.000577789506555
33
34	schoenebeck	319	enum implementation_t {
35	schoenebeck	361	CPP,
36			ASM_X86_MMX_SSE
37	schoenebeck	319	};
38
39	schoenebeck	563	/** @brief Real Time Math Base Class
40			*
41			* Math functions for real time operation. This base class contains all
42			* non-template methods.
43			*/
44	schoenebeck	319	class RTMathBase {
45	schoenebeck	53	public:
46			/**
47	schoenebeck	328	* Highly accurate time stamp.
48			*/
49			typedef uint32_t time_stamp_t;
50
51			/**
52			* We read the processor's cycle count register as a reference
53			* for the real time. These are of course only abstract values
54			* with arbitrary time entity, but that's not a problem as long
55			* as we calculate relatively.
56			*/
57			static time_stamp_t CreateTimeStamp();
58
59			/**
60	schoenebeck	53	* Calculates the frequency ratio for a pitch value given in cents
61			* (assuming equal tempered scale of course, divided into 12
62			* semitones per octave and 100 cents per semitone).
63			*
64	schoenebeck	554	* Note: CONFIG_MAX_PITCH (defined in config.h) has to be defined to an
65	schoenebeck	53	* appropriate value, otherwise the behavior of this function is
66	schoenebeck	554	* undefined, but most probably if CONFIG_MAX_PITCH is too small, the
67	schoenebeck	53	* application will crash due to segmentation fault here.
68			*
69			* @param cents - pitch value in cents (+1200 cents means +1 octave)
70			* @returns frequency ratio (e.g. +2.0 for +1 octave)
71			*/
72			inline static double CentsToFreqRatio(double Cents) {
73	schoenebeck	319	int index_int = (int) (Cents); // integer index
74	schoenebeck	53	float index_fract = Cents - index_int; // fractional part of index
75			return pCentsToFreqTable[index_int] + index_fract * (pCentsToFreqTable[index_int+1] - pCentsToFreqTable[index_int]);
76			}
77
78	schoenebeck	319	private:
79	schoenebeck	554	static float CentsToFreqTable[CONFIG_MAX_PITCH * 1200 * 2 + 1];
80	schoenebeck	319	static float* pCentsToFreqTable;
81
82			static float* InitCentsToFreqTable();
83			};
84
85	schoenebeck	563	/** @brief Real Time Math
86	schoenebeck	319	*
87	schoenebeck	563	* This is a template which provides customized methods for the desired low
88			* level implementation. The ASM_X86_MMX_SSE implementation of each method
89			* for example doesn't use 387 FPU instruction. This is needed for MMX
90			* algorithms which do not allow mixed MMX and 387 instructions.
91	schoenebeck	319	*/
92			template<implementation_t IMPL = CPP>
93			class __RTMath : public RTMathBase {
94			public:
95			// conversion using truncate
96			inline static int Int(const float a) {
97			switch (IMPL) {
98			case CPP: {
99			return (int) a;
100			}
101	schoenebeck	328	#if ARCH_X86
102	schoenebeck	319	case ASM_X86_MMX_SSE: {
103			int ret;
104			asm (
105			"cvttss2si %1, %0 # convert to int\n\t"
106			: "=r" (ret)
107			: "m" (a)
108			);
109			return ret;
110			}
111	schoenebeck	328	#endif // ARCH_X86
112	schoenebeck	319	}
113			}
114
115			//for doubles and everything else except floats
116			template<class T_a> inline static int Int(const T_a a) {
117			return (int) a;
118			}
119
120			inline static float Float(const int a) {
121			switch (IMPL) {
122			case CPP: {
123			return (float) a;
124			}
125	schoenebeck	328	#if ARCH_X86
126	schoenebeck	319	case ASM_X86_MMX_SSE: {
127			float ret;
128			asm (
129			"cvtsi2ss %1, %%xmm0 # convert to float\n\t"
130			"movss %%xmm0,%0 # output\n\t"
131			: "=m" (ret)
132			: "r" (a)
133			);
134			return ret;
135			}
136	schoenebeck	328	#endif // ARCH_X86
137	schoenebeck	319	}
138			}
139
140			#if 0
141			//for everything except ints
142			template<class T_a> inline static float Float(T_a a) {
143			return (float) a;
144			}
145			#endif
146
147			inline static float Sum(const float& a, const float& b) {
148			switch (IMPL) {
149			case CPP: {
150			return (a + b);
151			}
152	schoenebeck	328	#if ARCH_X86
153	schoenebeck	319	case ASM_X86_MMX_SSE: {
154			float ret;
155			asm (
156			"movss %1, %%xmm0 # load a\n\t"
157			"addss %2, %%xmm0 # a + b\n\t"
158			"movss %%xmm0, %0 # output\n\t"
159			: "=m" (ret)
160			: "m" (a), "m" (b)
161			);
162			return ret;
163			}
164	schoenebeck	328	#endif // ARCH_X86
165	schoenebeck	319	}
166			}
167
168			template<class T_a, class T_b> inline static T_a Sum(const T_a a, const T_b b) {
169			return (a + b);
170			}
171
172			inline static float Sub(const float& a, const float& b) {
173			switch (IMPL) {
174			case CPP: {
175			return (a - b);
176			}
177	schoenebeck	328	#if ARCH_X86
178	schoenebeck	319	case ASM_X86_MMX_SSE: {
179			float ret;
180			asm (
181			"movss %1, %%xmm0 # load a\n\t"
182			"subss %2, %%xmm0 # a - b\n\t"
183			"movss %%xmm0, %0 # output\n\t"
184			: "=m" (ret)
185			: "m" (a), "m" (b)
186			);
187			return ret;
188			}
189	schoenebeck	328	#endif // ARCH_X86
190	schoenebeck	319	}
191			}
192
193			template<class T_a, class T_b> inline static T_a Sub(const T_a a, const T_b b) {
194			return (a - b);
195			}
196
197			inline static float Mul(const float a, const float b) {
198			switch (IMPL) {
199			case CPP: {
200			return (a * b);
201			}
202	schoenebeck	328	#if ARCH_X86
203	schoenebeck	319	case ASM_X86_MMX_SSE: {
204			float ret;
205			asm (
206			"movss %1, %%xmm0 # load a\n\t"
207			"mulss %2, %%xmm0 # a * b\n\t"
208			"movss %%xmm0, %0 # output\n\t"
209			: "=m" (ret)
210			: "m" (a), "m" (b)
211			);
212			return ret;
213			}
214	schoenebeck	328	#endif // ARCH_X86
215	schoenebeck	319	}
216			}
217
218			template<class T_a, class T_b> inline static T_a Mul(const T_a a, const T_b b) {
219			return (a * b);
220			}
221
222			inline static float Div(const float a, const float b) {
223			switch (IMPL) {
224			case CPP: {
225			return (a / b);
226			}
227	schoenebeck	328	#if ARCH_X86
228	schoenebeck	319	case ASM_X86_MMX_SSE: {
229			float ret;
230			asm (
231			"movss %1, %%xmm0 # load a\n\t"
232			"divss %2, %%xmm0 # a / b\n\t"
233			"movss %%xmm0, %0 # output\n\t"
234			: "=m" (ret)
235			: "m" (a), "m" (b)
236			);
237			return ret;
238			}
239	schoenebeck	328	#endif // ARCH_X86
240	schoenebeck	319	}
241			}
242
243			template<class T_a, class T_b> inline static T_a Div(const T_a a, const T_b b) {
244			return (a / b);
245			}
246
247			inline static float Min(const float a, const float b) {
248			switch (IMPL) {
249			case CPP: {
250			return (b < a) ? b : a;
251			}
252	schoenebeck	328	#if ARCH_X86
253	schoenebeck	319	case ASM_X86_MMX_SSE: {
254			float ret;
255			asm (
256			"movss %1, %%xmm0 # load a\n\t"
257			"minss %2, %%xmm0 # Minimum(a, b)\n\t"
258			"movss %%xmm0, %0 # output\n\t"
259			: "=m" (ret)
260			: "m" (a), "m" (b)
261			);
262			return ret;
263			}
264	schoenebeck	328	#endif // ARCH_X86
265	schoenebeck	319	}
266			}
267
268			template<class T_a, class T_b> inline static T_a Min(const T_a a, const T_b b) {
269	schoenebeck	53	return (b < a) ? b : a;
270			}
271
272	schoenebeck	319	inline static float Max(const float a, const float b) {
273			switch (IMPL) {
274			case CPP: {
275			return (b > a) ? b : a;
276			}
277	schoenebeck	328	#if ARCH_X86
278	schoenebeck	319	case ASM_X86_MMX_SSE: {
279			float ret;
280			asm (
281			"movss %1, %%xmm0 # load a\n\t"
282			"maxss %2, %%xmm0 # Maximum(a, b)\n\t"
283			"movss %%xmm0, %0 # output\n\t"
284			: "=m" (ret)
285			: "m" (a), "m" (b)
286			);
287			return ret;
288			}
289	schoenebeck	328	#endif // ARCH_X86
290	schoenebeck	319	}
291			}
292
293			template<class T_a, class T_b> inline static T_a Max(const T_a a, const T_b b) {
294	schoenebeck	53	return (b > a) ? b : a;
295			}
296
297	schoenebeck	319	inline static float Fmodf(const float &a, const float &b) {
298			switch (IMPL) {
299			case CPP: {
300			return fmodf(a, b);
301			}
302	schoenebeck	328	#if ARCH_X86
303	schoenebeck	319	case ASM_X86_MMX_SSE: {
304			float ret;
305			asm (
306			"movss %1, %%xmm0 # load a\n\t"
307			"movss %2, %%xmm1 # load b\n\t"
308			"movss %%xmm0,%%xmm2\n\t"
309			"divss %%xmm1, %%xmm2 # xmm2 = a / b\n\t"
310			"cvttss2si %%xmm2, %%ecx #convert to int\n\t"
311			"cvtsi2ss %%ecx, %%xmm2 #convert back to float\n\t"
312			"mulss %%xmm1, %%xmm2 # xmm2 = b * int(a/b)\n\t"
313			"subss %%xmm2, %%xmm0 #sub a\n\t"
314			"movss %%xmm0, %0 # output\n\t"
315			: "=m" (ret)
316			: "m" (a), "m" (b)
317			: "%ecx"
318			);
319			return ret;
320			}
321	schoenebeck	328	#endif // ARCH_X86
322	schoenebeck	319	}
323			}
324	schoenebeck	53	};
325
326	schoenebeck	319	/// convenience typedef for using the default implementation (which is CPP)
327			typedef __RTMath<> RTMath;
328
329	schoenebeck	53	#endif // __RT_MATH_H__