src/common/RTMath.h

/***************************************************************************
 *                                                                         *
 *   LinuxSampler - modular, streaming capable sampler                     *
 *                                                                         *
 *   Copyright (C) 2003, 2004 by Benno Senoner and Christian Schoenebeck   *
 *   Copyright (C) 2005, 2006 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]);
        }

        /**
         * Inverse function to CentsToFreqRatio(). This function is a bit
         * slow, so it should not be called too frequently.
         */
        static double FreqRatioToCents(double FreqRatio) {
            return log(FreqRatio) / log(TWELVEHUNDREDTH_ROOT_OF_TWO);
        }

    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) {
                #if CONFIG_ASM && 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 // CONFIG_ASM && ARCH_X86
                default: {
                    return (int) a;
                }
            }
        }

        //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) {
                #if CONFIG_ASM && 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 // CONFIG_ASM && ARCH_X86
                default: {
                    return (float) a;
                }
            }
        }

#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) {
                #if CONFIG_ASM && 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 // CONFIG_ASM && ARCH_X86
                default: {
                    return (a + b);
                }
            }
        }

        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) {
                #if CONFIG_ASM && 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 // CONFIG_ASM && ARCH_X86
                default: {
                    return (a - b);
                }
            }
        }

        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) {
                #if CONFIG_ASM && 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 // CONFIG_ASM && ARCH_X86
                default: {
                    return (a * b);
                }
            }
        }

        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) {
                #if CONFIG_ASM && 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 // CONFIG_ASM && ARCH_X86
                default: {
                    return (a / b);
                }
            }
        }

        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) {
                #if CONFIG_ASM && 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 // CONFIG_ASM && ARCH_X86
                default: {
                    return std::min(a, b);
                }
            }
        }

        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) {
                #if CONFIG_ASM && 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 // CONFIG_ASM && ARCH_X86
                default: {
                    return std::max(a, b);
                }
            }
        }

        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) {
                #if CONFIG_ASM && 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 // CONFIG_ASM && ARCH_X86
                default: {
                    return fmodf(a, b);
                }
            }
        }
};

/// 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	829	* Copyright (C) 2005, 2006 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	829	/**
79			* Inverse function to CentsToFreqRatio(). This function is a bit
80			* slow, so it should not be called too frequently.
81			*/
82			static double FreqRatioToCents(double FreqRatio) {
83			return log(FreqRatio) / log(TWELVEHUNDREDTH_ROOT_OF_TWO);
84			}
85
86	schoenebeck	319	private:
87	schoenebeck	554	static float CentsToFreqTable[CONFIG_MAX_PITCH * 1200 * 2 + 1];
88	schoenebeck	319	static float* pCentsToFreqTable;
89
90			static float* InitCentsToFreqTable();
91			};
92
93	schoenebeck	563	/** @brief Real Time Math
94	schoenebeck	319	*
95	schoenebeck	563	* This is a template which provides customized methods for the desired low
96			* level implementation. The ASM_X86_MMX_SSE implementation of each method
97			* for example doesn't use 387 FPU instruction. This is needed for MMX
98			* algorithms which do not allow mixed MMX and 387 instructions.
99	schoenebeck	319	*/
100			template<implementation_t IMPL = CPP>
101			class __RTMath : public RTMathBase {
102			public:
103			// conversion using truncate
104			inline static int Int(const float a) {
105			switch (IMPL) {
106	schoenebeck	617	#if CONFIG_ASM && ARCH_X86
107	schoenebeck	319	case ASM_X86_MMX_SSE: {
108			int ret;
109			asm (
110			"cvttss2si %1, %0 # convert to int\n\t"
111			: "=r" (ret)
112			: "m" (a)
113			);
114			return ret;
115			}
116	schoenebeck	617	#endif // CONFIG_ASM && ARCH_X86
117	persson	685	default: {
118			return (int) a;
119			}
120	schoenebeck	319	}
121			}
122
123			//for doubles and everything else except floats
124			template<class T_a> inline static int Int(const T_a a) {
125			return (int) a;
126			}
127
128			inline static float Float(const int a) {
129			switch (IMPL) {
130	schoenebeck	617	#if CONFIG_ASM && ARCH_X86
131	schoenebeck	319	case ASM_X86_MMX_SSE: {
132			float ret;
133			asm (
134			"cvtsi2ss %1, %%xmm0 # convert to float\n\t"
135			"movss %%xmm0,%0 # output\n\t"
136			: "=m" (ret)
137			: "r" (a)
138			);
139			return ret;
140			}
141	schoenebeck	617	#endif // CONFIG_ASM && ARCH_X86
142	persson	685	default: {
143			return (float) a;
144			}
145	schoenebeck	319	}
146			}
147
148			#if 0
149			//for everything except ints
150			template<class T_a> inline static float Float(T_a a) {
151			return (float) a;
152			}
153			#endif
154
155			inline static float Sum(const float& a, const float& b) {
156			switch (IMPL) {
157	schoenebeck	617	#if CONFIG_ASM && ARCH_X86
158	schoenebeck	319	case ASM_X86_MMX_SSE: {
159			float ret;
160			asm (
161			"movss %1, %%xmm0 # load a\n\t"
162			"addss %2, %%xmm0 # a + b\n\t"
163			"movss %%xmm0, %0 # output\n\t"
164			: "=m" (ret)
165			: "m" (a), "m" (b)
166			);
167			return ret;
168			}
169	schoenebeck	617	#endif // CONFIG_ASM && ARCH_X86
170	persson	685	default: {
171			return (a + b);
172			}
173	schoenebeck	319	}
174			}
175
176			template<class T_a, class T_b> inline static T_a Sum(const T_a a, const T_b b) {
177			return (a + b);
178			}
179
180			inline static float Sub(const float& a, const float& b) {
181			switch (IMPL) {
182	schoenebeck	617	#if CONFIG_ASM && ARCH_X86
183	schoenebeck	319	case ASM_X86_MMX_SSE: {
184			float ret;
185			asm (
186			"movss %1, %%xmm0 # load a\n\t"
187			"subss %2, %%xmm0 # a - b\n\t"
188			"movss %%xmm0, %0 # output\n\t"
189			: "=m" (ret)
190			: "m" (a), "m" (b)
191			);
192			return ret;
193			}
194	schoenebeck	617	#endif // CONFIG_ASM && ARCH_X86
195	persson	685	default: {
196			return (a - b);
197			}
198	schoenebeck	319	}
199			}
200
201			template<class T_a, class T_b> inline static T_a Sub(const T_a a, const T_b b) {
202			return (a - b);
203			}
204
205			inline static float Mul(const float a, const float b) {
206			switch (IMPL) {
207	schoenebeck	617	#if CONFIG_ASM && ARCH_X86
208	schoenebeck	319	case ASM_X86_MMX_SSE: {
209			float ret;
210			asm (
211			"movss %1, %%xmm0 # load a\n\t"
212			"mulss %2, %%xmm0 # a * b\n\t"
213			"movss %%xmm0, %0 # output\n\t"
214			: "=m" (ret)
215			: "m" (a), "m" (b)
216			);
217			return ret;
218			}
219	schoenebeck	617	#endif // CONFIG_ASM && ARCH_X86
220	persson	685	default: {
221			return (a * b);
222			}
223	schoenebeck	319	}
224			}
225
226			template<class T_a, class T_b> inline static T_a Mul(const T_a a, const T_b b) {
227			return (a * b);
228			}
229
230			inline static float Div(const float a, const float b) {
231			switch (IMPL) {
232	schoenebeck	617	#if CONFIG_ASM && ARCH_X86
233	schoenebeck	319	case ASM_X86_MMX_SSE: {
234			float ret;
235			asm (
236			"movss %1, %%xmm0 # load a\n\t"
237			"divss %2, %%xmm0 # a / b\n\t"
238			"movss %%xmm0, %0 # output\n\t"
239			: "=m" (ret)
240			: "m" (a), "m" (b)
241			);
242			return ret;
243			}
244	schoenebeck	617	#endif // CONFIG_ASM && ARCH_X86
245	persson	685	default: {
246			return (a / b);
247			}
248	schoenebeck	319	}
249			}
250
251			template<class T_a, class T_b> inline static T_a Div(const T_a a, const T_b b) {
252			return (a / b);
253			}
254
255			inline static float Min(const float a, const float b) {
256			switch (IMPL) {
257	schoenebeck	617	#if CONFIG_ASM && ARCH_X86
258	schoenebeck	319	case ASM_X86_MMX_SSE: {
259			float ret;
260			asm (
261			"movss %1, %%xmm0 # load a\n\t"
262			"minss %2, %%xmm0 # Minimum(a, b)\n\t"
263			"movss %%xmm0, %0 # output\n\t"
264			: "=m" (ret)
265			: "m" (a), "m" (b)
266			);
267			return ret;
268			}
269	schoenebeck	617	#endif // CONFIG_ASM && ARCH_X86
270	persson	685	default: {
271	wylder	818	return std::min(a, b);
272	persson	685	}
273	schoenebeck	319	}
274			}
275
276			template<class T_a, class T_b> inline static T_a Min(const T_a a, const T_b b) {
277	schoenebeck	53	return (b < a) ? b : a;
278			}
279
280	schoenebeck	319	inline static float Max(const float a, const float b) {
281			switch (IMPL) {
282	schoenebeck	617	#if CONFIG_ASM && ARCH_X86
283	schoenebeck	319	case ASM_X86_MMX_SSE: {
284			float ret;
285			asm (
286			"movss %1, %%xmm0 # load a\n\t"
287			"maxss %2, %%xmm0 # Maximum(a, b)\n\t"
288			"movss %%xmm0, %0 # output\n\t"
289			: "=m" (ret)
290			: "m" (a), "m" (b)
291			);
292			return ret;
293			}
294	schoenebeck	617	#endif // CONFIG_ASM && ARCH_X86
295	persson	685	default: {
296	wylder	818	return std::max(a, b);
297	persson	685	}
298	schoenebeck	319	}
299			}
300
301			template<class T_a, class T_b> inline static T_a Max(const T_a a, const T_b b) {
302	schoenebeck	53	return (b > a) ? b : a;
303			}
304
305	schoenebeck	319	inline static float Fmodf(const float &a, const float &b) {
306			switch (IMPL) {
307	schoenebeck	617	#if CONFIG_ASM && ARCH_X86
308	schoenebeck	319	case ASM_X86_MMX_SSE: {
309			float ret;
310			asm (
311			"movss %1, %%xmm0 # load a\n\t"
312			"movss %2, %%xmm1 # load b\n\t"
313			"movss %%xmm0,%%xmm2\n\t"
314			"divss %%xmm1, %%xmm2 # xmm2 = a / b\n\t"
315			"cvttss2si %%xmm2, %%ecx #convert to int\n\t"
316			"cvtsi2ss %%ecx, %%xmm2 #convert back to float\n\t"
317			"mulss %%xmm1, %%xmm2 # xmm2 = b * int(a/b)\n\t"
318			"subss %%xmm2, %%xmm0 #sub a\n\t"
319			"movss %%xmm0, %0 # output\n\t"
320			: "=m" (ret)
321			: "m" (a), "m" (b)
322			: "%ecx"
323			);
324			return ret;
325			}
326	schoenebeck	617	#endif // CONFIG_ASM && ARCH_X86
327	persson	685	default: {
328			return fmodf(a, b);
329			}
330	schoenebeck	319	}
331			}
332	schoenebeck	53	};
333
334	schoenebeck	319	/// convenience typedef for using the default implementation (which is CPP)
335			typedef __RTMath<> RTMath;
336
337	schoenebeck	53	#endif // __RT_MATH_H__