draw: mimize cost of translate key compares, use cache universally
[mesa.git] / src / gallium / auxiliary / draw / draw_pt_post_vs.c
1 /**************************************************************************
2 *
3 * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28 #include "pipe/p_util.h"
29 #include "pipe/p_context.h"
30 #include "draw/draw_context.h"
31 #include "draw/draw_private.h"
32 #include "draw/draw_vbuf.h"
33 #include "draw/draw_vertex.h"
34 #include "draw/draw_pt.h"
35
36 struct pt_post_vs {
37 struct draw_context *draw;
38
39 boolean (*run)( struct pt_post_vs *pvs,
40 struct vertex_header *vertices,
41 unsigned count,
42 unsigned stride );
43 };
44
45
46
47 static INLINE float
48 dot4(const float *a, const float *b)
49 {
50 return (a[0]*b[0] +
51 a[1]*b[1] +
52 a[2]*b[2] +
53 a[3]*b[3]);
54 }
55
56
57
58 static INLINE unsigned
59 compute_clipmask_gl(const float *clip, /*const*/ float plane[][4], unsigned nr)
60 {
61 unsigned mask = 0x0;
62 unsigned i;
63
64 #if 0
65 debug_printf("compute clipmask %f %f %f %f\n",
66 clip[0], clip[1], clip[2], clip[3]);
67 assert(clip[3] != 0.0);
68 #endif
69
70 /* Do the hardwired planes first:
71 */
72 if (-clip[0] + clip[3] < 0) mask |= (1<<0);
73 if ( clip[0] + clip[3] < 0) mask |= (1<<1);
74 if (-clip[1] + clip[3] < 0) mask |= (1<<2);
75 if ( clip[1] + clip[3] < 0) mask |= (1<<3);
76 if ( clip[2] + clip[3] < 0) mask |= (1<<4); /* match mesa clipplane numbering - for now */
77 if (-clip[2] + clip[3] < 0) mask |= (1<<5); /* match mesa clipplane numbering - for now */
78
79 /* Followed by any remaining ones:
80 */
81 for (i = 6; i < nr; i++) {
82 if (dot4(clip, plane[i]) < 0)
83 mask |= (1<<i);
84 }
85
86 return mask;
87 }
88
89
90 /* The normal case - cliptest, rhw divide, viewport transform.
91 *
92 * Also handle identity viewport here at the expense of a few wasted
93 * instructions
94 */
95 static boolean post_vs_cliptest_viewport_gl( struct pt_post_vs *pvs,
96 struct vertex_header *vertices,
97 unsigned count,
98 unsigned stride )
99 {
100 struct vertex_header *out = vertices;
101 const float *scale = pvs->draw->viewport.scale;
102 const float *trans = pvs->draw->viewport.translate;
103 unsigned clipped = 0;
104 unsigned j;
105
106 if (0) debug_printf("%s\n");
107
108 for (j = 0; j < count; j++) {
109 out->clip[0] = out->data[0][0];
110 out->clip[1] = out->data[0][1];
111 out->clip[2] = out->data[0][2];
112 out->clip[3] = out->data[0][3];
113
114 out->vertex_id = 0xffff;
115 out->clipmask = compute_clipmask_gl(out->clip,
116 pvs->draw->plane,
117 pvs->draw->nr_planes);
118 clipped += out->clipmask;
119
120 if (out->clipmask == 0)
121 {
122 /* divide by w */
123 float w = 1.0f / out->data[0][3];
124
125 /* Viewport mapping */
126 out->data[0][0] = out->data[0][0] * w * scale[0] + trans[0];
127 out->data[0][1] = out->data[0][1] * w * scale[1] + trans[1];
128 out->data[0][2] = out->data[0][2] * w * scale[2] + trans[2];
129 out->data[0][3] = w;
130 #if 0
131 debug_printf("post viewport: %f %f %f %f\n",
132 out->data[0][0],
133 out->data[0][1],
134 out->data[0][2],
135 out->data[0][3]);
136 #endif
137 }
138
139 out = (struct vertex_header *)( (char *)out + stride );
140 }
141
142 return clipped != 0;
143 }
144
145
146
147 /* If bypass_clipping is set, skip cliptest and rhw divide.
148 */
149 static boolean post_vs_viewport( struct pt_post_vs *pvs,
150 struct vertex_header *vertices,
151 unsigned count,
152 unsigned stride )
153 {
154 struct vertex_header *out = vertices;
155 const float *scale = pvs->draw->viewport.scale;
156 const float *trans = pvs->draw->viewport.translate;
157 unsigned j;
158
159 if (0) debug_printf("%s\n", __FUNCTION__);
160 for (j = 0; j < count; j++) {
161 /* Viewport mapping only, no cliptest/rhw divide
162 */
163 out->data[0][0] = out->data[0][0] * scale[0] + trans[0];
164 out->data[0][1] = out->data[0][1] * scale[1] + trans[1];
165 out->data[0][2] = out->data[0][2] * scale[2] + trans[2];
166
167 out = (struct vertex_header *)((char *)out + stride);
168 }
169
170 return FALSE;
171 }
172
173
174 /* If bypass_clipping is set and we have an identity viewport, nothing
175 * to do.
176 */
177 static boolean post_vs_none( struct pt_post_vs *pvs,
178 struct vertex_header *vertices,
179 unsigned count,
180 unsigned stride )
181 {
182 if (0) debug_printf("%s\n", __FUNCTION__);
183 return FALSE;
184 }
185
186 boolean draw_pt_post_vs_run( struct pt_post_vs *pvs,
187 struct vertex_header *pipeline_verts,
188 unsigned count,
189 unsigned stride )
190 {
191 return pvs->run( pvs, pipeline_verts, count, stride );
192 }
193
194
195 void draw_pt_post_vs_prepare( struct pt_post_vs *pvs,
196 boolean bypass_clipping,
197 boolean identity_viewport,
198 boolean opengl )
199 {
200 if (bypass_clipping) {
201 if (identity_viewport)
202 pvs->run = post_vs_none;
203 else
204 pvs->run = post_vs_viewport;
205 }
206 else {
207 //if (opengl)
208 pvs->run = post_vs_cliptest_viewport_gl;
209 }
210 }
211
212
213 struct pt_post_vs *draw_pt_post_vs_create( struct draw_context *draw )
214 {
215 struct pt_post_vs *pvs = CALLOC_STRUCT( pt_post_vs );
216 if (!pvs)
217 return NULL;
218
219 pvs->draw = draw;
220
221 return pvs;
222 }
223
224 void draw_pt_post_vs_destroy( struct pt_post_vs *pvs )
225 {
226 FREE(pvs);
227 }