Merge branch 'master' of git+ssh://keithw@git.freedesktop.org/git/mesa/mesa into...
[mesa.git] / src / mesa / drivers / dri / i965 / brw_vs_constval.c
1 /*
2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
5
6 Permission is hereby granted, free of charge, to any person obtaining
7 a 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, sublicense, 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
16 portions of the Software.
17
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25
26 **********************************************************************/
27 /*
28 * Authors:
29 * Keith Whitwell <keith@tungstengraphics.com>
30 */
31
32
33 #include "brw_context.h"
34 #include "program.h"
35 #include "program_instruction.h"
36 #include "macros.h"
37 #include "brw_vs.h"
38
39 /* Component is active if it may diverge from [0,0,0,1]. Undef values
40 * are promoted to [0,0,0,1] for the purposes of this analysis.
41 */
42 struct tracker {
43 GLboolean twoside;
44 GLubyte active[PROGRAM_OUTPUT+1][128];
45 GLuint size_masks[4];
46 };
47
48
49 static void set_active_component( struct tracker *t,
50 GLuint file,
51 GLuint index,
52 GLubyte active )
53 {
54 switch (file) {
55 case PROGRAM_TEMPORARY:
56 case PROGRAM_INPUT:
57 case PROGRAM_OUTPUT:
58 t->active[file][index] |= active;
59
60 default:
61 break;
62 }
63 }
64
65 static void set_active( struct tracker *t,
66 struct prog_dst_register dst,
67 GLuint active )
68 {
69 set_active_component( t, dst.File, dst.Index, active & dst.WriteMask );
70 }
71
72
73 static GLubyte get_active_component( struct tracker *t,
74 GLuint file,
75 GLuint index,
76 GLuint component,
77 GLubyte swz )
78 {
79 switch (swz) {
80 case SWIZZLE_ZERO:
81 return component < 3 ? 0 : (1<<component);
82 case SWIZZLE_ONE:
83 return component == 3 ? 0 : (1<<component);
84 default:
85 switch (file) {
86 case PROGRAM_TEMPORARY:
87 case PROGRAM_INPUT:
88 case PROGRAM_OUTPUT:
89 return t->active[file][index] & (1<<component);
90 default:
91 return 1 << component;
92 }
93 }
94 }
95
96
97 static GLubyte get_active( struct tracker *t,
98 struct prog_src_register src )
99 {
100 GLuint i;
101 GLubyte active = src.NegateBase; /* NOTE! */
102
103 if (src.RelAddr)
104 return 0xf;
105
106 for (i = 0; i < 4; i++)
107 active |= get_active_component(t, src.File, src.Index, i,
108 GET_SWZ(src.Swizzle, i));
109
110 return active;
111 }
112
113 static GLubyte get_output_size( struct tracker *t,
114 GLuint idx )
115 {
116 GLubyte active = t->active[PROGRAM_OUTPUT][idx];
117 if (active & (1<<3)) return 4;
118 if (active & (1<<2)) return 3;
119 if (active & (1<<1)) return 2;
120 if (active & (1<<0)) return 1;
121 return 0;
122 }
123
124 /* Note the potential copying that occurs in the setup program:
125 */
126 static void calc_sizes( struct tracker *t )
127 {
128 GLuint i;
129
130 if (t->twoside) {
131 t->active[PROGRAM_OUTPUT][VERT_RESULT_COL0] |=
132 t->active[PROGRAM_OUTPUT][VERT_RESULT_BFC0];
133
134 t->active[PROGRAM_OUTPUT][VERT_RESULT_COL1] |=
135 t->active[PROGRAM_OUTPUT][VERT_RESULT_BFC1];
136 }
137
138 for (i = 0; i < FRAG_ATTRIB_MAX; i++) {
139 switch (get_output_size(t, i)) {
140 case 4: t->size_masks[4-1] |= 1<<i;
141 case 3: t->size_masks[3-1] |= 1<<i;
142 case 2: t->size_masks[2-1] |= 1<<i;
143 case 1: t->size_masks[1-1] |= 1<<i;
144 break;
145 }
146 }
147 }
148
149 static GLubyte szflag[4+1] = {
150 0,
151 0x1,
152 0x3,
153 0x7,
154 0xf
155 };
156
157 /* Pull a size out of the packed array:
158 */
159 static GLuint get_input_size(struct brw_context *brw,
160 GLuint attr)
161 {
162 GLuint sizes_dword = brw->vb.info.sizes[attr/16];
163 GLuint sizes_bits = (sizes_dword>>((attr%16)*2)) & 0x3;
164 return sizes_bits + 1;
165 /* return brw->vb.inputs[attr].glarray->Size; */
166 }
167
168 /* Calculate sizes of vertex program outputs. Size is the largest
169 * component index which might vary from [0,0,0,1]
170 */
171 static void calc_wm_input_sizes( struct brw_context *brw )
172 {
173 /* BRW_NEW_VERTEX_PROGRAM */
174 struct brw_vertex_program *vp =
175 (struct brw_vertex_program *)brw->vertex_program;
176 /* BRW_NEW_INPUT_DIMENSIONS */
177 struct tracker t;
178 GLuint insn;
179 GLuint i;
180
181 memset(&t, 0, sizeof(t));
182
183 /* _NEW_LIGHT */
184 if (brw->attribs.Light->Model.TwoSide)
185 t.twoside = 1;
186
187 for (i = 0; i < VERT_ATTRIB_MAX; i++)
188 if (vp->program.Base.InputsRead & (1<<i))
189 set_active_component(&t, PROGRAM_INPUT, i,
190 szflag[get_input_size(brw, i)]);
191
192 for (insn = 0; insn < vp->program.Base.NumInstructions; insn++) {
193 struct prog_instruction *inst = &vp->program.Base.Instructions[insn];
194
195 switch (inst->Opcode) {
196 case OPCODE_ARL:
197 break;
198
199 case OPCODE_MOV:
200 set_active(&t, inst->DstReg, get_active(&t, inst->SrcReg[0]));
201 break;
202
203 default:
204 set_active(&t, inst->DstReg, 0xf);
205 break;
206 }
207 }
208
209 calc_sizes(&t);
210
211 if (memcmp(brw->wm.input_size_masks, t.size_masks, sizeof(t.size_masks)) != 0) {
212 memcpy(brw->wm.input_size_masks, t.size_masks, sizeof(t.size_masks));
213 brw->state.dirty.brw |= BRW_NEW_WM_INPUT_DIMENSIONS;
214 }
215 }
216
217 const struct brw_tracked_state brw_wm_input_sizes = {
218 .dirty = {
219 .mesa = _NEW_LIGHT,
220 .brw = BRW_NEW_VERTEX_PROGRAM | BRW_NEW_INPUT_DIMENSIONS,
221 .cache = 0
222 },
223 .update = calc_wm_input_sizes
224 };
225