diff --git a/ChangeLog b/ChangeLog index 1e91fcb1e..e3dcb7700 100644 --- a/ChangeLog +++ b/ChangeLog @@ -1,3 +1,18 @@ +2007-03-29 Tomas Frydrych + + * configure.ac: fixed typo + + * clutter/clutter-fixed.c: + * clutter/clutter-fixed.h: + (clutter_tani): fast implementation of tan() + (clutter_qmulx): improved-precission fixed point multiply + + * clutter/clutter-state.c: + (perspectivex): fixed point implementaiton of perspective() + (_clutter_stage_sync_viewport): + (clutter_stage_get_actor_at_pos): + use perspectivex() instead of perspective() + 2007-03-27 Matthew Allum * clutter/cogl/Makefile.am: diff --git a/clutter/clutter-fixed.c b/clutter/clutter-fixed.c index 80d3fd239..b90517f50 100644 --- a/clutter/clutter-fixed.c +++ b/clutter/clutter-fixed.c @@ -256,6 +256,124 @@ clutter_sini (ClutterAngle angle) return result; } +/* pre-computed tan table for 1st quadrant + * + * Currently contains 257 entries. + * + */ +static ClutterFixed tan_tbl [] = +{ + 0x00000000L, 0x00000192L, 0x00000324L, 0x000004b7L, + 0x00000649L, 0x000007dbL, 0x0000096eL, 0x00000b01L, + 0x00000c94L, 0x00000e27L, 0x00000fbaL, 0x0000114eL, + 0x000012e2L, 0x00001477L, 0x0000160cL, 0x000017a1L, + 0x00001937L, 0x00001acdL, 0x00001c64L, 0x00001dfbL, + 0x00001f93L, 0x0000212cL, 0x000022c5L, 0x0000245fL, + 0x000025f9L, 0x00002795L, 0x00002931L, 0x00002aceL, + 0x00002c6cL, 0x00002e0aL, 0x00002faaL, 0x0000314aL, + 0x000032ecL, 0x0000348eL, 0x00003632L, 0x000037d7L, + 0x0000397dL, 0x00003b24L, 0x00003cccL, 0x00003e75L, + 0x00004020L, 0x000041ccL, 0x00004379L, 0x00004528L, + 0x000046d8L, 0x0000488aL, 0x00004a3dL, 0x00004bf2L, + 0x00004da8L, 0x00004f60L, 0x0000511aL, 0x000052d5L, + 0x00005492L, 0x00005651L, 0x00005812L, 0x000059d5L, + 0x00005b99L, 0x00005d60L, 0x00005f28L, 0x000060f3L, + 0x000062c0L, 0x0000648fL, 0x00006660L, 0x00006834L, + 0x00006a0aL, 0x00006be2L, 0x00006dbdL, 0x00006f9aL, + 0x0000717aL, 0x0000735dL, 0x00007542L, 0x0000772aL, + 0x00007914L, 0x00007b02L, 0x00007cf2L, 0x00007ee6L, + 0x000080dcL, 0x000082d6L, 0x000084d2L, 0x000086d2L, + 0x000088d6L, 0x00008adcL, 0x00008ce7L, 0x00008ef4L, + 0x00009106L, 0x0000931bL, 0x00009534L, 0x00009750L, + 0x00009971L, 0x00009b95L, 0x00009dbeL, 0x00009febL, + 0x0000a21cL, 0x0000a452L, 0x0000a68cL, 0x0000a8caL, + 0x0000ab0eL, 0x0000ad56L, 0x0000afa3L, 0x0000b1f5L, + 0x0000b44cL, 0x0000b6a8L, 0x0000b909L, 0x0000bb70L, + 0x0000bdddL, 0x0000c04fL, 0x0000c2c7L, 0x0000c545L, + 0x0000c7c9L, 0x0000ca53L, 0x0000cce3L, 0x0000cf7aL, + 0x0000d218L, 0x0000d4bcL, 0x0000d768L, 0x0000da1aL, + 0x0000dcd4L, 0x0000df95L, 0x0000e25eL, 0x0000e52eL, + 0x0000e806L, 0x0000eae7L, 0x0000edd0L, 0x0000f0c1L, + 0x0000f3bbL, 0x0000f6bfL, 0x0000f9cbL, 0x0000fce1L, + 0x00010000L, 0x00010329L, 0x0001065dL, 0x0001099aL, + 0x00010ce3L, 0x00011036L, 0x00011394L, 0x000116feL, + 0x00011a74L, 0x00011df6L, 0x00012184L, 0x0001251fL, + 0x000128c6L, 0x00012c7cL, 0x0001303fL, 0x00013410L, + 0x000137f0L, 0x00013bdfL, 0x00013fddL, 0x000143ebL, + 0x00014809L, 0x00014c37L, 0x00015077L, 0x000154c9L, + 0x0001592dL, 0x00015da4L, 0x0001622eL, 0x000166ccL, + 0x00016b7eL, 0x00017045L, 0x00017523L, 0x00017a17L, + 0x00017f22L, 0x00018444L, 0x00018980L, 0x00018ed5L, + 0x00019445L, 0x000199cfL, 0x00019f76L, 0x0001a53aL, + 0x0001ab1cL, 0x0001b11dL, 0x0001b73fL, 0x0001bd82L, + 0x0001c3e7L, 0x0001ca71L, 0x0001d11fL, 0x0001d7f4L, + 0x0001def1L, 0x0001e618L, 0x0001ed6aL, 0x0001f4e8L, + 0x0001fc96L, 0x00020473L, 0x00020c84L, 0x000214c9L, + 0x00021d44L, 0x000225f9L, 0x00022ee9L, 0x00023818L, + 0x00024187L, 0x00024b3aL, 0x00025534L, 0x00025f78L, + 0x00026a0aL, 0x000274edL, 0x00028026L, 0x00028bb8L, + 0x000297a8L, 0x0002a3fbL, 0x0002b0b5L, 0x0002bdddL, + 0x0002cb79L, 0x0002d98eL, 0x0002e823L, 0x0002f740L, + 0x000306ecL, 0x00031730L, 0x00032816L, 0x000339a6L, + 0x00034bebL, 0x00035ef2L, 0x000372c6L, 0x00038776L, + 0x00039d11L, 0x0003b3a6L, 0x0003cb48L, 0x0003e40aL, + 0x0003fe02L, 0x00041949L, 0x000435f7L, 0x0004542bL, + 0x00047405L, 0x000495a9L, 0x0004b940L, 0x0004def6L, + 0x00050700L, 0x00053196L, 0x00055ef9L, 0x00058f75L, + 0x0005c35dL, 0x0005fb14L, 0x00063709L, 0x000677c0L, + 0x0006bdd0L, 0x000709ecL, 0x00075ce6L, 0x0007b7bbL, + 0x00081b98L, 0x000889e9L, 0x0009046eL, 0x00098d4dL, + 0x000a2736L, 0x000ad593L, 0x000b9cc6L, 0x000c828aL, + 0x000d8e82L, 0x000ecb1bL, 0x001046eaL, 0x00121703L, + 0x00145b00L, 0x0017448dL, 0x001b2672L, 0x002095afL, + 0x0028bc49L, 0x0036519aL, 0x00517bb6L, 0x00a2f8fdL, + 0x46d3eab2L, +}; + +/** + * clutter_tani: + * @angle: a #ClutterAngle + * + * Very fast fixed point implementation of tan function. + * + * ClutterAngle is an integer such that 1024 represents + * full circle. + * + * Return value: #ClutterFixed sine value. + * + * Since: 0.3 + */ +ClutterFixed +clutter_tani (ClutterAngle angle) +{ + int sign = 1; + ClutterFixed result; + + /* reduce negative angle to positive + sign */ + if (angle < 0) + { + sign = 1 + ~sign; + angle = 1 + ~angle; + } + + /* reduce to <0, pi) */ + angle &= 0x1ff; + + /* reduce to first quadrant and sign */ + if (angle > 256) + { + sign = 1 + ~sign; + angle = 512 - angle; + } + + result = tan_tbl[angle]; + + if (sign < 0) + result = (1 + ~result); + + return result; +} + ClutterFixed sqrt_tbl [] = { 0x00000000L, 0x00010000L, 0x00016A0AL, 0x0001BB68L, @@ -515,6 +633,46 @@ clutter_sqrti (gint number) return (number * flt2.i + 0x1e3c68) >> 22; } +/** + * clutter_fixed_qmulx: + * @op1: #ClutterFixed + * @op2: #ClutterFixed + * + * Return value: #ClutterFixed. + * + * Multiplies two fixed values using 64bit arithmetic; this provides + * significantly better precission than the #CLUTTER_FIXED_MUL macro, + * but at performance cost (about 2.7 times slowdown on ARMv5e, and 2 times + * on x86). + * + * Since: 0.3 + */ +ClutterFixed +clutter_qmulx (ClutterFixed op1, ClutterFixed op2) +{ +#ifdef __arm__ + /* This provides about 12% speedeup on the gcc -O2 optimised + * C version + * + * Based on code found in the following thread: + * http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/2006-August/014405.html + */ + int res_low, res_hi; + + __asm__ ("smull %0, %1, %2, %3 \n" + "mov %0, %0, lsr %4 \n" + "add %1, %0, %1, lsl %5 \n" + : "=r"(res_hi), "=r"(res_low)\ + : "r"(op1), "r"(op2), "i"(CFX_Q), "i"(32-CFX_Q)); + + return (ClutterFixed) res_low; +#else + long long r = (long long) op1 * (long long) op2; + + return (unsigned int)(r >> CFX_Q); +#endif +} + /* */ const double _magic = 68719476736.0*1.5; @@ -575,3 +733,4 @@ _clutter_double_to_int (double val) } #undef _CFX_MAN + diff --git a/clutter/clutter-fixed.h b/clutter/clutter-fixed.h index 2eca15689..ed08d4198 100644 --- a/clutter/clutter-fixed.h +++ b/clutter/clutter-fixed.h @@ -140,9 +140,14 @@ typedef gint32 ClutterAngle; /* angle such that 1024 == 2*PI */ #define CFX_DIV CLUTTER_FIXED_DIV /* Fixed point math routines */ +extern inline +ClutterFixed clutter_qmulx (ClutterFixed op1, ClutterFixed op2); + ClutterFixed clutter_sinx (ClutterFixed angle); ClutterFixed clutter_sini (ClutterAngle angle); +ClutterFixed clutter_tani (ClutterAngle angle); + /* convenience macros for the cos functions */ /** diff --git a/clutter/clutter-stage.c b/clutter/clutter-stage.c index 296a83dd8..3e30d7c0c 100644 --- a/clutter/clutter-stage.c +++ b/clutter/clutter-stage.c @@ -48,6 +48,8 @@ #include +#define FIXED_PERSPECTIVE + G_DEFINE_ABSTRACT_TYPE (ClutterStage, clutter_stage, CLUTTER_TYPE_GROUP); #define CLUTTER_STAGE_GET_PRIVATE(obj) \ @@ -622,6 +624,11 @@ clutter_stage_snapshot (ClutterStage *stage, return NULL; } +#ifndef FIXED_PERSPECTIVE +/* + * Original floating point implementaiton of the perspective function, + * retained for reference purposes + */ static inline void frustum (GLfloat left, GLfloat right, @@ -658,13 +665,60 @@ perspective (GLfloat fovy, { GLfloat xmin, xmax, ymin, ymax; - ymax = zNear * tan (fovy * M_PI / 360.0); + ymax = zNear * tan (fovy * M_PI / 180.0); ymin = -ymax; xmin = ymin * aspect; xmax = ymax * aspect; + printf ("%f, %f, %f, %f\n", xmin, xmax, ymin, ymax); + frustum (xmin, xmax, ymin, ymax, zNear, zFar); } +#else + +/* + * Fixed point implementation of the perspective function + */ +static inline void +perspectivex (ClutterAngle fovy, + ClutterFixed aspect, + ClutterFixed zNear, + ClutterFixed zFar) +{ + ClutterFixed xmax, ymax; + ClutterFixed x, y, c, d; + GLfloat m[16]; + + memset (&m[0], 0, sizeof (m)); + + /* + * Based on the original algorithm in perspective(): + * + * 1) xmin = -xmax => xmax + xmin == 0 && xmax - xmin == 2 * xmax + * same true for y, hence: a == 0 && b == 0; + * + * 2) When working with small numbers, we can are loosing significant + * precision, hence we use clutter_qmulx() here, not the fast macro. + */ + ymax = clutter_qmulx (zNear, clutter_tani (fovy)); + xmax = clutter_qmulx (ymax, aspect); + + x = CFX_DIV (zNear, xmax); + y = CFX_DIV (zNear, ymax); + c = CFX_DIV (-(zFar + zNear), ( zFar - zNear)); + d = CFX_DIV (-(clutter_qmulx (2*zFar, zNear)), (zFar - zNear)); + +#define M(row,col) m[col*4+row] + M(0,0) = CLUTTER_FIXED_TO_FLOAT (x); + M(1,1) = CLUTTER_FIXED_TO_FLOAT (y); + M(2,2) = CLUTTER_FIXED_TO_FLOAT (c); + M(2,3) = CLUTTER_FIXED_TO_FLOAT (d); + M(3,2) = -1.0F; +#undef M + + glMultMatrixf (m); +} +#endif void _clutter_stage_sync_viewport (ClutterStage *stage) @@ -683,8 +737,15 @@ _clutter_stage_sync_viewport (ClutterStage *stage) glMatrixMode (GL_PROJECTION); glLoadIdentity (); - - perspective (60.0f, 1.0f, 0.1f, 100.0f); + +#ifndef FIXED_PERSPECTIVE + perspective (30.0f, 1.0f, 0.1f, 100.0f); +#else + perspectivex (85, /* 30 degrees */ + CFX_ONE, + CLUTTER_FLOAT_TO_FIXED (0.1), + CLUTTER_FLOAT_TO_FIXED (100.0)); +#endif glMatrixMode (GL_MODELVIEW); glLoadIdentity (); @@ -740,7 +801,14 @@ clutter_stage_get_actor_at_pos (ClutterStage *stage, (view[3] - 2 * (y - view[1])), 0); glScalef (view[2], -view[3], 1.0); - perspective (60.0f, 1.0f, 0.1f, 100.0f); +#ifndef FIXED_PERSPECTIVE + perspective (30.0f, 1.0f, 0.1f, 100.0f); +#else + perspectivex (85, /* 30 degrees */ + CFX_ONE, + CLUTTER_FLOAT_TO_FIXED (0.1), + CLUTTER_FLOAT_TO_FIXED (100.0)); +#endif glMatrixMode (GL_MODELVIEW); diff --git a/configure.ac b/configure.ac index b22008815..d0ec5c4d6 100644 --- a/configure.ac +++ b/configure.ac @@ -106,7 +106,7 @@ case $clutterbackend in glx) CLUTTER_FLAVOUR="glx" - CLUTTER_GOGL="gl" + CLUTTER_COGL="gl" AC_DEFINE([HAVE_CLUTTER_GLX], 1, [Have the GLX backend]) AC_CHECK_HEADERS([GL/gl.h GL/glx.h],,