Ticket #5965: freesurround-imath.patch

mythtv/libs/libmythfreesurround/el_processor.cpp

@@ -16 +16 @@
 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
 */
 
+#include "../libmyth/mythconfig.h"
 #include "el_processor.h"
 #include <cstdlib>
 #include <cstring>
@@ -31 +32 @@
 typedef FFTSample FFTComplexArray[2];
 #endif
 
+// use inline asm trig on x86_32/64
+#ifdef ARCH_X86
+#include "imath.h"
+#define SQRT isqrtf
+#define SIN isin
+#define COS icos
+#define TAN itan
+#define ASIN iasin
+#define ATAN2 iatan2f
+#else
+#define SQRT sqrt
+#define SIN sin
+#define COS cos
+#define TAN tan
+#define ASIN asin
+#define ATAN2 atan2
+#endif
 
 #ifdef USE_FFTW3
 #pragma comment (lib,"libfftw3f-3.lib")
@@ -40 +58 @@
 
 const float PI = 3.141592654;
 const float epsilon = 0.000001;
-const float center_level = 0.5*sqrt(0.5);
+const float center_level = 0.5*SQRT(0.5);
 
 // private implementation of the surround decoder
 class decoder_impl {
@@ -177 +195 @@
         unsigned int cutoff = (30*N)/srate;
         for (unsigned f=0;f<=halfN;f++) {           
             if (f<cutoff)
-                filter[5][f] = 0.5*sqrt(0.5);
+                filter[5][f] = 0.5*SQRT(0.5);
             else
                 filter[5][f] = 0.0;
         }
@@ -214 +232 @@
 
 private:
     // polar <-> cartesian coodinates conversion
-    static inline float amplitude(const float cf[2]) { return sqrt(cf[0]*cf[0] + cf[1]*cf[1]); }
-    static inline float phase(const float cf[2]) { return atan2(cf[1],cf[0]); }
+    static inline float amplitude(const float cf[2]) { return SQRT(cf[0]*cf[0] + cf[1]*cf[1]); }
+    static inline float phase(const float cf[2]) { return ATAN2(cf[1],cf[0]); }
+#ifdef ARCH_X86
+    static inline cfloat polar(float a, float p) {
+        float sc[2]; 
+        isincosf(p, sc); 
+        return cfloat(a*sc[0],a*sc[1]); 
+    }
+#else
     static inline cfloat polar(float a, float p) { return cfloat(a*cos(p),a*sin(p)); }
+#endif
     static inline float sqr(float x) { return x*x; }
     // the dreaded min/max
     static inline float min(float a, float b) { return a<b?a:b; }
@@ -377 +403 @@
     inline double get_yfs(double ampDiff, double phaseDiff) {
         double x = 1-(((1-sqr(ampDiff))*phaseDiff)/PI*2);
 #ifdef FASTER_CALC
-        double tanX = tan(x);
+        double tanX = TAN(x);
         return 0.16468622925824683 + 0.5009268347818189*x - 0.06462757726992101*x*x
             + 0.09170680403453149*x*x*x + 0.2617754892323973*tanX - 0.04180413533856156*sqr(tanX);
 #else
@@ -390 +416 @@
     inline double get_xfs(double ampDiff, double yfs) {
         double x=ampDiff,y=yfs;
 #ifdef FASTER_CALC
-        double tanX = tan(x);
-        double tanY = tan(y);
-        double asinX = asin(x);
-        double sinX = sin(x);
-        double sinY = sin(y);
+        double tanX = TAN(x);
+        double tanY = TAN(y);
+        double asinX = ASIN(x);
+        double sinX = SIN(x);
+        double sinY = SIN(y);
         double x3 = x*x*x;
         double y2 = y*y;
         double y3 = y*y2;

new file mythtv/libs/libmythfreesurround/imath.h

@@ +1 @@
+static inline double icos(double x) 
+{
+    double y;
+    asm ( 
+        "fldl   %1\n\t" 
+        "fcos   \n\t" 
+        :"=&t"(y):"m"(x)
+    ); 
+    return y;
+}
+
+static inline float icosf(float x) 
+{
+    float y;
+    asm ( 
+        "flds   %1\n\t" 
+        "fcos   \n\t" 
+        :"=&t"(y):"m"(x)
+    ); 
+    return y;
+}
+
+static inline double isin(double x) 
+{
+    double y;
+    asm ( 
+        "fldl   %1\n\t" 
+        "fsin   \n\t" 
+        :"=&t"(y):"m"(x)
+    ); 
+    return y;
+}
+
+static inline float isinf(float x) 
+{
+    float y;
+    asm ( 
+        "flds   %1\n\t" 
+        "fsin   \n\t" 
+        :"=&t"(y):"m"(x)
+    ); 
+    return y;
+}
+
+static inline double itan(double x) 
+{
+    double y;
+    asm ( 
+        "fldl   %1\n\t" 
+        "fptan   \n\t" 
+        "fstp   %%st(0)\n\t"
+        :"=&t"(y):"m"(x)
+    ); 
+    return y;
+}
+
+static inline float itanf(float x) 
+{
+    float y;
+    asm ( 
+        "flds   %1\n\t" 
+        "fptan   \n\t" 
+        :"=&t"(y):"m"(x)
+    ); 
+    return y;
+}
+
+static inline double *isincos(double x, double *ptr) 
+{
+    double *ptr2 = ptr + 1;
+    asm ( 
+        "fldl   %2\n\t" 
+        "fsincos \n\t" 
+        "fstpl  %0\n\t"
+        "fstpl  %1\n\t"
+        :"=m"(*ptr),"=m"(*ptr2):"m"(x)
+    ); 
+    return ptr;
+}
+
+static inline float *isincosf(float x, float *ptr) 
+{
+    float *ptr2 = ptr + 1;
+    asm ( 
+        "flds   %2\n\t" 
+        "fsincos \n\t" 
+        "fstps  %0\n\t"
+        "fstps  %1\n\t"
+        :"=m"(*ptr),"=m"(*ptr2):"m"(x)
+    ); 
+    return ptr;
+}
+
+static inline double iasin(double x)
+{
+    double y;
+    asm (
+        "fldl   %1\n\t"
+        "fld    %%st\n\t"
+        "fmul   %%st(0)\n\t"
+        "fld1   \n\t"
+        "fsubp  \n\t"
+        "fsqrt  \n\t"
+        "fpatan \n\t"
+        :"=&t"(y):"m"(x)
+    );
+    return y;
+}
+
+static inline float iasinf(float x)
+{
+    float y;
+    asm (
+        "flds   %1\n\t"
+        "fld    %%st\n\t"
+        "fmul   %%st(0)\n\t"
+        "fld1   \n\t"
+        "fsubp  \n\t"
+        "fsqrt  \n\t"
+        "fpatan \n\t"
+        :"=&t"(y):"m"(x)
+    );
+    return y;
+}
+
+static inline double iatan2(double x, double y)
+{
+    double z;
+    asm (
+        "fldl   %1\n\t"
+        "fldl   %2\n\t"
+        "fpatan \n\t"
+        :"=&t"(z):"m"(x),"m"(y)
+    );
+    return z;
+}
+
+static inline float iatan2f(float x, float y)
+{
+    float z;
+    asm (
+        "flds   %1\n\t"
+        "flds   %2\n\t"
+        "fpatan \n\t"
+        :"=&t"(z):"m"(x),"m"(y)
+    );
+    return z;
+}
+
+static inline double isqrt(double x)
+{
+    double y;
+    asm (
+        "fldl   %1\n\t"
+        "fsqrt  \n\t"
+        :"=&t"(y):"m"(x)
+    );
+    return y;
+}
+
+static inline float isqrtf(float x)
+{
+    float y;
+    asm (
+        "flds   %1\n\t"
+        "fsqrt  \n\t"
+        :"=&t"(y):"m"(x)
+    );
+    return y;
+}