1848b8605Smrg/*
2848b8605Smrg * Mesa 3-D graphics library
3848b8605Smrg *
4848b8605Smrg * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
5848b8605Smrg *
6848b8605Smrg * Permission is hereby granted, free of charge, to any person obtaining a
7848b8605Smrg * copy of this software and associated documentation files (the "Software"),
8848b8605Smrg * to deal in the Software without restriction, including without limitation
9848b8605Smrg * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10848b8605Smrg * and/or sell copies of the Software, and to permit persons to whom the
11848b8605Smrg * Software is furnished to do so, subject to the following conditions:
12848b8605Smrg *
13848b8605Smrg * The above copyright notice and this permission notice shall be included
14848b8605Smrg * in all copies or substantial portions of the Software.
15848b8605Smrg *
16848b8605Smrg * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
17848b8605Smrg * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18848b8605Smrg * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19848b8605Smrg * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
20848b8605Smrg * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21848b8605Smrg * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22848b8605Smrg * OTHER DEALINGS IN THE SOFTWARE.
23848b8605Smrg *
24848b8605Smrg * Authors:
25848b8605Smrg *    Gareth Hughes
26848b8605Smrg */
27848b8605Smrg
28848b8605Smrg#include "main/glheader.h"
29848b8605Smrg#include "main/context.h"
30848b8605Smrg#include "main/macros.h"
31848b8605Smrg#include "main/imports.h"
32848b8605Smrg
33848b8605Smrg#include "m_matrix.h"
34848b8605Smrg#include "m_xform.h"
35848b8605Smrg
36848b8605Smrg#include "m_debug.h"
37848b8605Smrg#include "m_debug_util.h"
38848b8605Smrg
39848b8605Smrg#ifdef __UNIXOS2__
40848b8605Smrg/* The linker doesn't like empty files */
41848b8605Smrgstatic char dummy;
42848b8605Smrg#endif
43848b8605Smrg
44848b8605Smrg#ifdef DEBUG_MATH  /* This code only used for debugging */
45848b8605Smrg
46848b8605Smrgstatic clip_func *clip_tab[2] = {
47848b8605Smrg   _mesa_clip_tab,
48848b8605Smrg   _mesa_clip_np_tab
49848b8605Smrg};
50848b8605Smrgstatic char *cnames[2] = {
51848b8605Smrg   "_mesa_clip_tab",
52848b8605Smrg   "_mesa_clip_np_tab"
53848b8605Smrg};
54848b8605Smrg#ifdef RUN_DEBUG_BENCHMARK
55848b8605Smrgstatic char *cstrings[2] = {
56848b8605Smrg   "clip, perspective divide",
57848b8605Smrg   "clip, no divide"
58848b8605Smrg};
59848b8605Smrg#endif
60848b8605Smrg
61848b8605Smrg
62848b8605Smrg/* =============================================================
63848b8605Smrg * Reference cliptests
64848b8605Smrg */
65848b8605Smrg
66848b8605Smrgstatic GLvector4f *ref_cliptest_points4( GLvector4f *clip_vec,
67848b8605Smrg					 GLvector4f *proj_vec,
68848b8605Smrg					 GLubyte clipMask[],
69848b8605Smrg					 GLubyte *orMask,
70848b8605Smrg					 GLubyte *andMask,
71848b8605Smrg					 GLboolean viewport_z_clip )
72848b8605Smrg{
73848b8605Smrg   const GLuint stride = clip_vec->stride;
74848b8605Smrg   const GLuint count = clip_vec->count;
75848b8605Smrg   const GLfloat *from = (GLfloat *)clip_vec->start;
76848b8605Smrg   GLuint c = 0;
77848b8605Smrg   GLfloat (*vProj)[4] = (GLfloat (*)[4])proj_vec->start;
78848b8605Smrg   GLubyte tmpAndMask = *andMask;
79848b8605Smrg   GLubyte tmpOrMask = *orMask;
80848b8605Smrg   GLuint i;
81848b8605Smrg   for ( i = 0 ; i < count ; i++, STRIDE_F(from, stride) ) {
82848b8605Smrg      const GLfloat cx = from[0];
83848b8605Smrg      const GLfloat cy = from[1];
84848b8605Smrg      const GLfloat cz = from[2];
85848b8605Smrg      const GLfloat cw = from[3];
86848b8605Smrg      GLubyte mask = 0;
87848b8605Smrg      if ( -cx + cw < 0 ) mask |= CLIP_RIGHT_BIT;
88848b8605Smrg      if (  cx + cw < 0 ) mask |= CLIP_LEFT_BIT;
89848b8605Smrg      if ( -cy + cw < 0 ) mask |= CLIP_TOP_BIT;
90848b8605Smrg      if (  cy + cw < 0 ) mask |= CLIP_BOTTOM_BIT;
91848b8605Smrg      if (viewport_z_clip) {
92848b8605Smrg	 if ( -cz + cw < 0 ) mask |= CLIP_FAR_BIT;
93848b8605Smrg	 if (  cz + cw < 0 ) mask |= CLIP_NEAR_BIT;
94848b8605Smrg      }
95848b8605Smrg      clipMask[i] = mask;
96848b8605Smrg      if ( mask ) {
97848b8605Smrg	 c++;
98848b8605Smrg	 tmpAndMask &= mask;
99848b8605Smrg	 tmpOrMask |= mask;
100848b8605Smrg	 vProj[i][0] = 0;
101848b8605Smrg	 vProj[i][1] = 0;
102848b8605Smrg	 vProj[i][2] = 0;
103848b8605Smrg	 vProj[i][3] = 1;
104848b8605Smrg      } else {
105848b8605Smrg	 GLfloat oow = 1.0F / cw;
106848b8605Smrg	 vProj[i][0] = cx * oow;
107848b8605Smrg	 vProj[i][1] = cy * oow;
108848b8605Smrg	 vProj[i][2] = cz * oow;
109848b8605Smrg	 vProj[i][3] = oow;
110848b8605Smrg      }
111848b8605Smrg   }
112848b8605Smrg
113848b8605Smrg   *orMask = tmpOrMask;
114848b8605Smrg   *andMask = (GLubyte) (c < count ? 0 : tmpAndMask);
115848b8605Smrg
116848b8605Smrg   proj_vec->flags |= VEC_SIZE_4;
117848b8605Smrg   proj_vec->size = 4;
118848b8605Smrg   proj_vec->count = clip_vec->count;
119848b8605Smrg   return proj_vec;
120848b8605Smrg}
121848b8605Smrg
122848b8605Smrg/* Keep these here for now, even though we don't use them...
123848b8605Smrg */
124848b8605Smrgstatic GLvector4f *ref_cliptest_points3( GLvector4f *clip_vec,
125848b8605Smrg					 GLvector4f *proj_vec,
126848b8605Smrg					 GLubyte clipMask[],
127848b8605Smrg					 GLubyte *orMask,
128848b8605Smrg					 GLubyte *andMask,
129848b8605Smrg                                         GLboolean viewport_z_clip )
130848b8605Smrg{
131848b8605Smrg   const GLuint stride = clip_vec->stride;
132848b8605Smrg   const GLuint count = clip_vec->count;
133848b8605Smrg   const GLfloat *from = (GLfloat *)clip_vec->start;
134848b8605Smrg
135848b8605Smrg   GLubyte tmpOrMask = *orMask;
136848b8605Smrg   GLubyte tmpAndMask = *andMask;
137848b8605Smrg   GLuint i;
138848b8605Smrg   for ( i = 0 ; i < count ; i++, STRIDE_F(from, stride) ) {
139848b8605Smrg      const GLfloat cx = from[0], cy = from[1], cz = from[2];
140848b8605Smrg      GLubyte mask = 0;
141848b8605Smrg      if ( cx >  1.0 )		mask |= CLIP_RIGHT_BIT;
142848b8605Smrg      else if ( cx < -1.0 )	mask |= CLIP_LEFT_BIT;
143848b8605Smrg      if ( cy >  1.0 )		mask |= CLIP_TOP_BIT;
144848b8605Smrg      else if ( cy < -1.0 )	mask |= CLIP_BOTTOM_BIT;
145848b8605Smrg      if (viewport_z_clip) {
146848b8605Smrg         if ( cz >  1.0 )		mask |= CLIP_FAR_BIT;
147848b8605Smrg         else if ( cz < -1.0 )	mask |= CLIP_NEAR_BIT;
148848b8605Smrg      }
149848b8605Smrg      clipMask[i] = mask;
150848b8605Smrg      tmpOrMask |= mask;
151848b8605Smrg      tmpAndMask &= mask;
152848b8605Smrg   }
153848b8605Smrg
154848b8605Smrg   *orMask = tmpOrMask;
155848b8605Smrg   *andMask = tmpAndMask;
156848b8605Smrg   return clip_vec;
157848b8605Smrg}
158848b8605Smrg
159848b8605Smrgstatic GLvector4f * ref_cliptest_points2( GLvector4f *clip_vec,
160848b8605Smrg					  GLvector4f *proj_vec,
161848b8605Smrg					  GLubyte clipMask[],
162848b8605Smrg					  GLubyte *orMask,
163848b8605Smrg					  GLubyte *andMask,
164848b8605Smrg                                          GLboolean viewport_z_clip )
165848b8605Smrg{
166848b8605Smrg   const GLuint stride = clip_vec->stride;
167848b8605Smrg   const GLuint count = clip_vec->count;
168848b8605Smrg   const GLfloat *from = (GLfloat *)clip_vec->start;
169848b8605Smrg
170848b8605Smrg   GLubyte tmpOrMask = *orMask;
171848b8605Smrg   GLubyte tmpAndMask = *andMask;
172848b8605Smrg   GLuint i;
173848b8605Smrg
174848b8605Smrg   (void) viewport_z_clip;
175848b8605Smrg
176848b8605Smrg   for ( i = 0 ; i < count ; i++, STRIDE_F(from, stride) ) {
177848b8605Smrg      const GLfloat cx = from[0], cy = from[1];
178848b8605Smrg      GLubyte mask = 0;
179848b8605Smrg      if ( cx >  1.0 )		mask |= CLIP_RIGHT_BIT;
180848b8605Smrg      else if ( cx < -1.0 )	mask |= CLIP_LEFT_BIT;
181848b8605Smrg      if ( cy >  1.0 )		mask |= CLIP_TOP_BIT;
182848b8605Smrg      else if ( cy < -1.0 )	mask |= CLIP_BOTTOM_BIT;
183848b8605Smrg      clipMask[i] = mask;
184848b8605Smrg      tmpOrMask |= mask;
185848b8605Smrg      tmpAndMask &= mask;
186848b8605Smrg   }
187848b8605Smrg
188848b8605Smrg   *orMask = tmpOrMask;
189848b8605Smrg   *andMask = tmpAndMask;
190848b8605Smrg   return clip_vec;
191848b8605Smrg}
192848b8605Smrg
193848b8605Smrgstatic clip_func ref_cliptest[5] = {
194848b8605Smrg   0,
195848b8605Smrg   0,
196848b8605Smrg   ref_cliptest_points2,
197848b8605Smrg   ref_cliptest_points3,
198848b8605Smrg   ref_cliptest_points4
199848b8605Smrg};
200848b8605Smrg
201848b8605Smrg
202848b8605Smrg/* =============================================================
203848b8605Smrg * Cliptest tests
204848b8605Smrg */
205848b8605Smrg
206848b8605SmrgALIGN16(static GLfloat, s[TEST_COUNT][4]);
207848b8605SmrgALIGN16(static GLfloat, d[TEST_COUNT][4]);
208848b8605SmrgALIGN16(static GLfloat, r[TEST_COUNT][4]);
209848b8605Smrg
210848b8605Smrg
211848b8605Smrg/**
212848b8605Smrg * Check if X, Y or Z component of the coordinate is close to W, in terms
213848b8605Smrg * of the clip test.
214848b8605Smrg */
215848b8605Smrgstatic GLboolean
216848b8605Smrgxyz_close_to_w(const GLfloat c[4])
217848b8605Smrg{
218848b8605Smrg   float k = 0.0001;
219848b8605Smrg   return (fabs(c[0] - c[3]) < k ||
220848b8605Smrg           fabs(c[1] - c[3]) < k ||
221848b8605Smrg           fabs(c[2] - c[3]) < k ||
222848b8605Smrg           fabs(-c[0] - c[3]) < k ||
223848b8605Smrg           fabs(-c[1] - c[3]) < k ||
224848b8605Smrg           fabs(-c[2] - c[3]) < k);
225848b8605Smrg}
226848b8605Smrg
227848b8605Smrg
228848b8605Smrg
229848b8605Smrgstatic int test_cliptest_function( clip_func func, int np,
230848b8605Smrg				   int psize, long *cycles )
231848b8605Smrg{
232848b8605Smrg   GLvector4f source[1], dest[1], ref[1];
233848b8605Smrg   GLubyte dm[TEST_COUNT], dco, dca;
234848b8605Smrg   GLubyte rm[TEST_COUNT], rco, rca;
235848b8605Smrg   int i, j;
236848b8605Smrg#ifdef  RUN_DEBUG_BENCHMARK
237848b8605Smrg   int cycle_i;                /* the counter for the benchmarks we run */
238848b8605Smrg#endif
239848b8605Smrg   GLboolean viewport_z_clip = GL_TRUE;
240848b8605Smrg
241848b8605Smrg   (void) cycles;
242848b8605Smrg
243848b8605Smrg   if ( psize > 4 ) {
244848b8605Smrg      _mesa_problem( NULL, "test_cliptest_function called with psize > 4\n" );
245848b8605Smrg      return 0;
246848b8605Smrg   }
247848b8605Smrg
248848b8605Smrg   for ( i = 0 ; i < TEST_COUNT ; i++) {
249848b8605Smrg      ASSIGN_4V( d[i], 0.0, 0.0, 0.0, 1.0 );
250848b8605Smrg      ASSIGN_4V( s[i], 0.0, 0.0, 0.0, 1.0 );
251848b8605Smrg      for ( j = 0 ; j < psize ; j++ )
252848b8605Smrg         s[i][j] = rnd();
253848b8605Smrg   }
254848b8605Smrg
255848b8605Smrg   source->data = (GLfloat(*)[4])s;
256848b8605Smrg   source->start = (GLfloat *)s;
257848b8605Smrg   source->count = TEST_COUNT;
258848b8605Smrg   source->stride = sizeof(s[0]);
259848b8605Smrg   source->size = 4;
260848b8605Smrg   source->flags = 0;
261848b8605Smrg
262848b8605Smrg   dest->data = (GLfloat(*)[4])d;
263848b8605Smrg   dest->start = (GLfloat *)d;
264848b8605Smrg   dest->count = TEST_COUNT;
265848b8605Smrg   dest->stride = sizeof(float[4]);
266848b8605Smrg   dest->size = 0;
267848b8605Smrg   dest->flags = 0;
268848b8605Smrg
269848b8605Smrg   ref->data = (GLfloat(*)[4])r;
270848b8605Smrg   ref->start = (GLfloat *)r;
271848b8605Smrg   ref->count = TEST_COUNT;
272848b8605Smrg   ref->stride = sizeof(float[4]);
273848b8605Smrg   ref->size = 0;
274848b8605Smrg   ref->flags = 0;
275848b8605Smrg
276848b8605Smrg   dco = rco = 0;
277848b8605Smrg   dca = rca = CLIP_FRUSTUM_BITS;
278848b8605Smrg
279848b8605Smrg   ref_cliptest[psize]( source, ref, rm, &rco, &rca, viewport_z_clip );
280848b8605Smrg
281848b8605Smrg   if ( mesa_profile ) {
282848b8605Smrg      BEGIN_RACE( *cycles );
283848b8605Smrg      func( source, dest, dm, &dco, &dca, viewport_z_clip );
284848b8605Smrg      END_RACE( *cycles );
285848b8605Smrg   }
286848b8605Smrg   else {
287848b8605Smrg      func( source, dest, dm, &dco, &dca, viewport_z_clip );
288848b8605Smrg   }
289848b8605Smrg
290848b8605Smrg   if ( dco != rco ) {
291848b8605Smrg      printf( "\n-----------------------------\n" );
292848b8605Smrg      printf( "dco = 0x%02x   rco = 0x%02x\n", dco, rco );
293848b8605Smrg      return 0;
294848b8605Smrg   }
295848b8605Smrg   if ( dca != rca ) {
296848b8605Smrg      printf( "\n-----------------------------\n" );
297848b8605Smrg      printf( "dca = 0x%02x   rca = 0x%02x\n", dca, rca );
298848b8605Smrg      return 0;
299848b8605Smrg   }
300848b8605Smrg   for ( i = 0 ; i < TEST_COUNT ; i++ ) {
301848b8605Smrg      if ( dm[i] != rm[i] ) {
302848b8605Smrg         GLfloat *c = source->start;
303848b8605Smrg         STRIDE_F(c, source->stride * i);
304848b8605Smrg         if (psize == 4 && xyz_close_to_w(c)) {
305848b8605Smrg            /* The coordinate is very close to the clip plane.  The clipmask
306848b8605Smrg             * may vary depending on code path, but that's OK.
307848b8605Smrg             */
308848b8605Smrg            continue;
309848b8605Smrg         }
310848b8605Smrg	 printf( "\n-----------------------------\n" );
311848b8605Smrg	 printf( "mask[%d] = 0x%02x   ref mask[%d] = 0x%02x\n", i, dm[i], i,rm[i] );
312848b8605Smrg         printf(" coord = %f, %f, %f, %f\n",
313848b8605Smrg                c[0], c[1], c[2], c[3]);
314848b8605Smrg	 return 0;
315848b8605Smrg      }
316848b8605Smrg   }
317848b8605Smrg
318848b8605Smrg   /* Only verify output on projected points4 case.  FIXME: Do we need
319848b8605Smrg    * to test other cases?
320848b8605Smrg    */
321848b8605Smrg   if ( np || psize < 4 )
322848b8605Smrg      return 1;
323848b8605Smrg
324848b8605Smrg   for ( i = 0 ; i < TEST_COUNT ; i++ ) {
325848b8605Smrg      for ( j = 0 ; j < 4 ; j++ ) {
326848b8605Smrg         if ( significand_match( d[i][j], r[i][j] ) < REQUIRED_PRECISION ) {
327848b8605Smrg            printf( "\n-----------------------------\n" );
328848b8605Smrg            printf( "(i = %i, j = %i)  dm = 0x%02x   rm = 0x%02x\n",
329848b8605Smrg		    i, j, dm[i], rm[i] );
330848b8605Smrg            printf( "%f \t %f \t [diff = %e - %i bit missed]\n",
331848b8605Smrg		    d[i][0], r[i][0], r[i][0]-d[i][0],
332848b8605Smrg		    MAX_PRECISION - significand_match( d[i][0], r[i][0] ) );
333848b8605Smrg            printf( "%f \t %f \t [diff = %e - %i bit missed]\n",
334848b8605Smrg		    d[i][1], r[i][1], r[i][1]-d[i][1],
335848b8605Smrg		    MAX_PRECISION - significand_match( d[i][1], r[i][1] ) );
336848b8605Smrg            printf( "%f \t %f \t [diff = %e - %i bit missed]\n",
337848b8605Smrg		    d[i][2], r[i][2], r[i][2]-d[i][2],
338848b8605Smrg		    MAX_PRECISION - significand_match( d[i][2], r[i][2] ) );
339848b8605Smrg            printf( "%f \t %f \t [diff = %e - %i bit missed]\n",
340848b8605Smrg		    d[i][3], r[i][3], r[i][3]-d[i][3],
341848b8605Smrg		    MAX_PRECISION - significand_match( d[i][3], r[i][3] ) );
342848b8605Smrg            return 0;
343848b8605Smrg         }
344848b8605Smrg      }
345848b8605Smrg   }
346848b8605Smrg
347848b8605Smrg   return 1;
348848b8605Smrg}
349848b8605Smrg
350848b8605Smrgvoid _math_test_all_cliptest_functions( char *description )
351848b8605Smrg{
352848b8605Smrg   int np, psize;
353848b8605Smrg   long benchmark_tab[2][4];
354848b8605Smrg   static int first_time = 1;
355848b8605Smrg
356848b8605Smrg   if ( first_time ) {
357848b8605Smrg      first_time = 0;
358b8e80941Smrg      mesa_profile = getenv( "MESA_PROFILE" );
359848b8605Smrg   }
360848b8605Smrg
361848b8605Smrg#ifdef RUN_DEBUG_BENCHMARK
362848b8605Smrg   if ( mesa_profile ) {
363848b8605Smrg      if ( !counter_overhead ) {
364848b8605Smrg	 INIT_COUNTER();
365848b8605Smrg	 printf( "counter overhead: %ld cycles\n\n", counter_overhead );
366848b8605Smrg      }
367848b8605Smrg      printf( "cliptest results after hooking in %s functions:\n", description );
368848b8605Smrg   }
369848b8605Smrg#endif
370848b8605Smrg
371848b8605Smrg#ifdef RUN_DEBUG_BENCHMARK
372848b8605Smrg   if ( mesa_profile ) {
373848b8605Smrg      printf( "\n\t" );
374848b8605Smrg      for ( psize = 2 ; psize <= 4 ; psize++ ) {
375848b8605Smrg	 printf( " p%d\t", psize );
376848b8605Smrg      }
377848b8605Smrg      printf( "\n--------------------------------------------------------\n\t" );
378848b8605Smrg   }
379848b8605Smrg#endif
380848b8605Smrg
381848b8605Smrg   for ( np = 0 ; np < 2 ; np++ ) {
382848b8605Smrg      for ( psize = 2 ; psize <= 4 ; psize++ ) {
383848b8605Smrg	 clip_func func = clip_tab[np][psize];
384848b8605Smrg	 long *cycles = &(benchmark_tab[np][psize-1]);
385848b8605Smrg
386848b8605Smrg	 if ( test_cliptest_function( func, np, psize, cycles ) == 0 ) {
387848b8605Smrg	    char buf[100];
388848b8605Smrg	    sprintf( buf, "%s[%d] failed test (%s)",
389848b8605Smrg		     cnames[np], psize, description );
390848b8605Smrg	    _mesa_problem( NULL, "%s", buf );
391848b8605Smrg	 }
392848b8605Smrg#ifdef RUN_DEBUG_BENCHMARK
393848b8605Smrg	 if ( mesa_profile )
394848b8605Smrg	    printf( " %li\t", benchmark_tab[np][psize-1] );
395848b8605Smrg#endif
396848b8605Smrg      }
397848b8605Smrg#ifdef RUN_DEBUG_BENCHMARK
398848b8605Smrg      if ( mesa_profile )
399848b8605Smrg	 printf( " | [%s]\n\t", cstrings[np] );
400848b8605Smrg#endif
401848b8605Smrg   }
402848b8605Smrg#ifdef RUN_DEBUG_BENCHMARK
403848b8605Smrg   if ( mesa_profile )
404848b8605Smrg      printf( "\n" );
405848b8605Smrg#endif
406848b8605Smrg}
407848b8605Smrg
408848b8605Smrg
409848b8605Smrg#endif /* DEBUG_MATH */
410