Merge remote branch 'origin/7.8'
[mesa.git] / src / mesa / drivers / dri / i965 / brw_wm_pass1.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 "brw_wm.h"
35
36
37 static GLuint get_tracked_mask(struct brw_wm_compile *c,
38 struct brw_wm_instruction *inst)
39 {
40 GLuint i;
41 for (i = 0; i < 4; i++) {
42 if (inst->writemask & (1<<i)) {
43 if (!inst->dst[i]->contributes_to_output) {
44 inst->writemask &= ~(1<<i);
45 inst->dst[i] = 0;
46 }
47 }
48 }
49
50 return inst->writemask;
51 }
52
53 /* Remove a reference from a value's usage chain.
54 */
55 static void unlink_ref(struct brw_wm_ref *ref)
56 {
57 struct brw_wm_value *value = ref->value;
58
59 if (ref == value->lastuse) {
60 value->lastuse = ref->prevuse;
61 }
62 else {
63 struct brw_wm_ref *i = value->lastuse;
64 while (i->prevuse != ref) i = i->prevuse;
65 i->prevuse = ref->prevuse;
66 }
67 }
68
69 static void track_arg(struct brw_wm_compile *c,
70 struct brw_wm_instruction *inst,
71 GLuint arg,
72 GLuint readmask)
73 {
74 GLuint i;
75
76 for (i = 0; i < 4; i++) {
77 struct brw_wm_ref *ref = inst->src[arg][i];
78 if (ref) {
79 if (readmask & (1<<i)) {
80 ref->value->contributes_to_output = 1;
81 }
82 else {
83 unlink_ref(ref);
84 inst->src[arg][i] = NULL;
85 }
86 }
87 }
88 }
89
90 static GLuint get_texcoord_mask( GLuint tex_idx )
91 {
92 switch (tex_idx) {
93 case TEXTURE_1D_INDEX:
94 return WRITEMASK_X;
95 case TEXTURE_2D_INDEX:
96 return WRITEMASK_XY;
97 case TEXTURE_3D_INDEX:
98 return WRITEMASK_XYZ;
99 case TEXTURE_CUBE_INDEX:
100 return WRITEMASK_XYZ;
101 case TEXTURE_RECT_INDEX:
102 return WRITEMASK_XY;
103 default: return 0;
104 }
105 }
106
107
108 /* Step two: Basically this is dead code elimination.
109 *
110 * Iterate backwards over instructions, noting which values
111 * contribute to the final result. Adjust writemasks to only
112 * calculate these values.
113 */
114 void brw_wm_pass1( struct brw_wm_compile *c )
115 {
116 GLint insn;
117
118 for (insn = c->nr_insns-1; insn >= 0; insn--) {
119 struct brw_wm_instruction *inst = &c->instruction[insn];
120 GLuint writemask;
121 GLuint read0, read1, read2;
122
123 if (inst->opcode == OPCODE_KIL) {
124 track_arg(c, inst, 0, WRITEMASK_XYZW); /* All args contribute to final */
125 continue;
126 }
127
128 if (inst->opcode == WM_FB_WRITE) {
129 track_arg(c, inst, 0, WRITEMASK_XYZW);
130 track_arg(c, inst, 1, WRITEMASK_XYZW);
131 if (c->key.source_depth_to_render_target &&
132 c->key.computes_depth)
133 track_arg(c, inst, 2, WRITEMASK_Z);
134 else
135 track_arg(c, inst, 2, 0);
136 continue;
137 }
138
139 /* Lookup all the registers which were written by this
140 * instruction and get a mask of those that contribute to the output:
141 */
142 writemask = get_tracked_mask(c, inst);
143 if (!writemask) {
144 GLuint arg;
145 for (arg = 0; arg < 3; arg++)
146 track_arg(c, inst, arg, 0);
147 continue;
148 }
149
150 read0 = 0;
151 read1 = 0;
152 read2 = 0;
153
154 /* Mark all inputs which contribute to the marked outputs:
155 */
156 switch (inst->opcode) {
157 case OPCODE_ABS:
158 case OPCODE_FLR:
159 case OPCODE_FRC:
160 case OPCODE_MOV:
161 case OPCODE_SWZ:
162 case OPCODE_TRUNC:
163 read0 = writemask;
164 break;
165
166 case OPCODE_SUB:
167 case OPCODE_SLT:
168 case OPCODE_SLE:
169 case OPCODE_SGE:
170 case OPCODE_SGT:
171 case OPCODE_SEQ:
172 case OPCODE_SNE:
173 case OPCODE_ADD:
174 case OPCODE_MAX:
175 case OPCODE_MIN:
176 case OPCODE_MUL:
177 read0 = writemask;
178 read1 = writemask;
179 break;
180
181 case OPCODE_DDX:
182 case OPCODE_DDY:
183 read0 = writemask;
184 break;
185
186 case OPCODE_MAD:
187 case OPCODE_CMP:
188 case OPCODE_LRP:
189 read0 = writemask;
190 read1 = writemask;
191 read2 = writemask;
192 break;
193
194 case OPCODE_XPD:
195 if (writemask & WRITEMASK_X) read0 |= WRITEMASK_YZ;
196 if (writemask & WRITEMASK_Y) read0 |= WRITEMASK_XZ;
197 if (writemask & WRITEMASK_Z) read0 |= WRITEMASK_XY;
198 read1 = read0;
199 break;
200
201 case OPCODE_COS:
202 case OPCODE_EX2:
203 case OPCODE_LG2:
204 case OPCODE_RCP:
205 case OPCODE_RSQ:
206 case OPCODE_SIN:
207 case OPCODE_SCS:
208 case WM_CINTERP:
209 case WM_PIXELXY:
210 read0 = WRITEMASK_X;
211 break;
212
213 case OPCODE_POW:
214 read0 = WRITEMASK_X;
215 read1 = WRITEMASK_X;
216 break;
217
218 case OPCODE_TEX:
219 case OPCODE_TXP:
220 read0 = get_texcoord_mask(inst->tex_idx);
221
222 if (inst->tex_shadow)
223 read0 |= WRITEMASK_Z;
224 break;
225
226 case OPCODE_TXB:
227 /* Shadow ignored for txb.
228 */
229 read0 = get_texcoord_mask(inst->tex_idx) | WRITEMASK_W;
230 break;
231
232 case WM_WPOSXY:
233 read0 = writemask & WRITEMASK_XY;
234 break;
235
236 case WM_DELTAXY:
237 read0 = writemask & WRITEMASK_XY;
238 read1 = WRITEMASK_X;
239 break;
240
241 case WM_PIXELW:
242 read0 = WRITEMASK_X;
243 read1 = WRITEMASK_XY;
244 break;
245
246 case WM_LINTERP:
247 read0 = WRITEMASK_X;
248 read1 = WRITEMASK_XY;
249 break;
250
251 case WM_PINTERP:
252 read0 = WRITEMASK_X; /* interpolant */
253 read1 = WRITEMASK_XY; /* deltas */
254 read2 = WRITEMASK_W; /* pixel w */
255 break;
256
257 case OPCODE_DP3:
258 read0 = WRITEMASK_XYZ;
259 read1 = WRITEMASK_XYZ;
260 break;
261
262 case OPCODE_DPH:
263 read0 = WRITEMASK_XYZ;
264 read1 = WRITEMASK_XYZW;
265 break;
266
267 case OPCODE_DP4:
268 read0 = WRITEMASK_XYZW;
269 read1 = WRITEMASK_XYZW;
270 break;
271
272 case OPCODE_LIT:
273 read0 = WRITEMASK_XYW;
274 break;
275
276 case OPCODE_DST:
277 case WM_FRONTFACING:
278 case OPCODE_KIL_NV:
279 default:
280 break;
281 }
282
283 track_arg(c, inst, 0, read0);
284 track_arg(c, inst, 1, read1);
285 track_arg(c, inst, 2, read2);
286 }
287
288 if (INTEL_DEBUG & DEBUG_WM) {
289 brw_wm_print_program(c, "pass1");
290 }
291 }