optimize per-pixel fog program
[mesa.git] / src / mesa / shader / programopt.c
1 /*
2 * Mesa 3-D graphics library
3 * Version: 6.5.2
4 *
5 * Copyright (C) 1999-2006 Brian Paul All Rights Reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11 * and/or sell copies of the Software, and to permit persons to whom the
12 * Software is furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included
15 * in all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
21 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
22 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
23 */
24
25 /**
26 * \file programopt.c
27 * Vertex/Fragment program optimizations and transformations for program
28 * options, etc.
29 *
30 * \author Brian Paul
31 */
32
33
34 #include "glheader.h"
35 #include "context.h"
36 #include "imports.h"
37 #include "mtypes.h"
38 #include "program.h"
39 #include "programopt.h"
40 #include "program_instruction.h"
41
42
43 /**
44 * This function inserts instructions for coordinate modelview * projection
45 * into a vertex program.
46 * May be used to implement the position_invariant option.
47 */
48 void
49 _mesa_insert_mvp_code(GLcontext *ctx, struct gl_vertex_program *vprog)
50 {
51 struct prog_instruction *newInst;
52 const GLuint origLen = vprog->Base.NumInstructions;
53 const GLuint newLen = origLen + 4;
54 GLuint i;
55
56 /*
57 * Setup state references for the modelview/projection matrix.
58 * XXX we should check if these state vars are already declared.
59 */
60 static const GLint mvpState[4][5] = {
61 { STATE_MATRIX, STATE_MVP, 0, 0, 0 }, /* state.matrix.mvp.row[0] */
62 { STATE_MATRIX, STATE_MVP, 0, 1, 1 }, /* state.matrix.mvp.row[1] */
63 { STATE_MATRIX, STATE_MVP, 0, 2, 2 }, /* state.matrix.mvp.row[2] */
64 { STATE_MATRIX, STATE_MVP, 0, 3, 3 }, /* state.matrix.mvp.row[3] */
65 };
66 GLint mvpRef[4];
67
68 for (i = 0; i < 4; i++) {
69 mvpRef[i] = _mesa_add_state_reference(vprog->Base.Parameters,
70 mvpState[i]);
71 }
72
73 /* Alloc storage for new instructions */
74 newInst = _mesa_alloc_instructions(newLen);
75 if (!newInst) {
76 _mesa_error(ctx, GL_OUT_OF_MEMORY,
77 "glProgramString(inserting position_invariant code)");
78 return;
79 }
80
81 /*
82 * Generated instructions:
83 * newInst[0] = DP4 result.position.x, mvp.row[0], vertex.position;
84 * newInst[1] = DP4 result.position.y, mvp.row[1], vertex.position;
85 * newInst[2] = DP4 result.position.z, mvp.row[2], vertex.position;
86 * newInst[3] = DP4 result.position.w, mvp.row[3], vertex.position;
87 */
88 _mesa_init_instructions(newInst, 4);
89 for (i = 0; i < 4; i++) {
90 newInst[i].Opcode = OPCODE_DP4;
91 newInst[i].DstReg.File = PROGRAM_OUTPUT;
92 newInst[i].DstReg.Index = VERT_RESULT_HPOS;
93 newInst[i].DstReg.WriteMask = (WRITEMASK_X << i);
94 newInst[i].SrcReg[0].File = PROGRAM_STATE_VAR;
95 newInst[i].SrcReg[0].Index = mvpRef[i];
96 newInst[i].SrcReg[0].Swizzle = SWIZZLE_NOOP;
97 newInst[i].SrcReg[1].File = PROGRAM_INPUT;
98 newInst[i].SrcReg[1].Index = VERT_ATTRIB_POS;
99 newInst[i].SrcReg[1].Swizzle = SWIZZLE_NOOP;
100 }
101
102 /* Append original instructions after new instructions */
103 _mesa_memcpy(newInst + 4, vprog->Base.Instructions,
104 origLen * sizeof(struct prog_instruction));
105
106 /* free old instructions */
107 _mesa_free(vprog->Base.Instructions);
108
109 /* install new instructions */
110 vprog->Base.Instructions = newInst;
111 vprog->Base.NumInstructions = newLen;
112 vprog->Base.InputsRead |= VERT_BIT_POS;
113 vprog->Base.OutputsWritten |= (1 << VERT_RESULT_HPOS);
114 }
115
116
117
118 /**
119 * Append extra instructions onto the given fragment program to implement
120 * the fog mode specified by fprog->FogOption.
121 * The fragment.fogcoord input is used to compute the fog blend factor.
122 *
123 * XXX with a little work, this function could be adapted to add fog code
124 * to vertex programs too.
125 */
126 void
127 _mesa_append_fog_code(GLcontext *ctx, struct gl_fragment_program *fprog)
128 {
129 static const GLint fogPStateOpt[] = { STATE_INTERNAL,
130 STATE_FOG_PARAMS_OPTIMIZED, 0, 0, 0 };
131 static const GLint fogColorState[] = { STATE_FOG_COLOR, 0, 0, 0, 0 };
132 struct prog_instruction *newInst, *inst;
133 const GLuint origLen = fprog->Base.NumInstructions;
134 const GLuint newLen = origLen + 5;
135 GLuint i;
136 GLint fogPRefOpt, fogColorRef; /* state references */
137 GLuint colorTemp, fogFactorTemp; /* temporary registerss */
138
139 if (fprog->FogOption == GL_NONE) {
140 _mesa_problem(ctx, "_mesa_append_fog_code() called for fragment program"
141 " with FogOption == GL_NONE");
142 return;
143 }
144
145 /* Alloc storage for new instructions */
146 newInst = _mesa_alloc_instructions(newLen);
147 if (!newInst) {
148 _mesa_error(ctx, GL_OUT_OF_MEMORY,
149 "glProgramString(inserting fog_option code)");
150 return;
151 }
152
153 /* Copy orig instructions into new instruction buffer */
154 _mesa_memcpy(newInst, fprog->Base.Instructions,
155 origLen * sizeof(struct prog_instruction));
156
157 /* PARAM fogParamsRefOpt = internal optimized fog params; */
158 fogPRefOpt
159 = _mesa_add_state_reference(fprog->Base.Parameters, fogPStateOpt);
160 /* PARAM fogColorRef = state.fog.color; */
161 fogColorRef
162 = _mesa_add_state_reference(fprog->Base.Parameters, fogColorState);
163
164 /* TEMP colorTemp; */
165 colorTemp = fprog->Base.NumTemporaries++;
166 /* TEMP fogFactorTemp; */
167 fogFactorTemp = fprog->Base.NumTemporaries++;
168
169 /* Scan program to find where result.color is written */
170 inst = newInst;
171 for (i = 0; i < fprog->Base.NumInstructions; i++) {
172 if (inst->Opcode == OPCODE_END)
173 break;
174 if (inst->DstReg.File == PROGRAM_OUTPUT &&
175 inst->DstReg.Index == FRAG_RESULT_COLR) {
176 /* change the instruction to write to colorTemp w/ clamping */
177 inst->DstReg.File = PROGRAM_TEMPORARY;
178 inst->DstReg.Index = colorTemp;
179 inst->SaturateMode = SATURATE_ZERO_ONE;
180 /* don't break (may be several writes to result.color) */
181 }
182 inst++;
183 }
184 assert(inst->Opcode == OPCODE_END); /* we'll overwrite this inst */
185
186 _mesa_init_instructions(inst, 5);
187
188 /* emit instructions to compute fog blending factor */
189 if (fprog->FogOption == GL_LINEAR) {
190 /* MAD fogFactorTemp.x, fragment.fogcoord.x, fogPRefOpt.x, fogPRefOpt.y; */
191 inst->Opcode = OPCODE_MAD;
192 inst->DstReg.File = PROGRAM_TEMPORARY;
193 inst->DstReg.Index = fogFactorTemp;
194 inst->DstReg.WriteMask = WRITEMASK_X;
195 inst->SrcReg[0].File = PROGRAM_INPUT;
196 inst->SrcReg[0].Index = FRAG_ATTRIB_FOGC;
197 inst->SrcReg[0].Swizzle = SWIZZLE_X;
198 inst->SrcReg[1].File = PROGRAM_STATE_VAR;
199 inst->SrcReg[1].Index = fogPRefOpt;
200 inst->SrcReg[1].Swizzle = SWIZZLE_X;
201 inst->SrcReg[2].File = PROGRAM_STATE_VAR;
202 inst->SrcReg[2].Index = fogPRefOpt;
203 inst->SrcReg[2].Swizzle = SWIZZLE_Y;
204 inst++;
205 }
206 else {
207 ASSERT(fprog->FogOption == GL_EXP || fprog->FogOption == GL_EXP2);
208 /* fogPRefOpt.z = d/ln(2), fogPRefOpt.w = d/sqrt(ln(2) */
209 /* EXP: MUL fogFactorTemp.x, fogPRefOpt.z, fragment.fogcoord.x; */
210 /* EXP2: MUL fogFactorTemp.x, fogPRefOpt.w, fragment.fogcoord.x; */
211 inst->Opcode = OPCODE_MUL;
212 inst->DstReg.File = PROGRAM_TEMPORARY;
213 inst->DstReg.Index = fogFactorTemp;
214 inst->DstReg.WriteMask = WRITEMASK_X;
215 inst->SrcReg[0].File = PROGRAM_STATE_VAR;
216 inst->SrcReg[0].Index = fogPRefOpt;
217 inst->SrcReg[0].Swizzle
218 = (fprog->FogOption == GL_EXP) ? SWIZZLE_Z : SWIZZLE_W;
219 inst->SrcReg[1].File = PROGRAM_INPUT;
220 inst->SrcReg[1].Index = FRAG_ATTRIB_FOGC;
221 inst->SrcReg[1].Swizzle = SWIZZLE_X;
222 inst++;
223 if (fprog->FogOption == GL_EXP2) {
224 /* MUL fogFactorTemp.x, fogFactorTemp.x, fogFactorTemp.x; */
225 inst->Opcode = OPCODE_MUL;
226 inst->DstReg.File = PROGRAM_TEMPORARY;
227 inst->DstReg.Index = fogFactorTemp;
228 inst->DstReg.WriteMask = WRITEMASK_X;
229 inst->SrcReg[0].File = PROGRAM_TEMPORARY;
230 inst->SrcReg[0].Index = fogFactorTemp;
231 inst->SrcReg[0].Swizzle = SWIZZLE_X;
232 inst->SrcReg[1].File = PROGRAM_TEMPORARY;
233 inst->SrcReg[1].Index = fogFactorTemp;
234 inst->SrcReg[1].Swizzle = SWIZZLE_X;
235 inst++;
236 }
237 /* EX2_SAT fogFactorTemp.x, -fogFactorTemp.x; */
238 inst->Opcode = OPCODE_EX2;
239 inst->DstReg.File = PROGRAM_TEMPORARY;
240 inst->DstReg.Index = fogFactorTemp;
241 inst->DstReg.WriteMask = WRITEMASK_X;
242 inst->SrcReg[0].File = PROGRAM_TEMPORARY;
243 inst->SrcReg[0].Index = fogFactorTemp;
244 inst->SrcReg[0].NegateBase = GL_TRUE;
245 inst->SrcReg[0].Swizzle = SWIZZLE_X;
246 inst->SaturateMode = SATURATE_ZERO_ONE;
247 inst++;
248 }
249 /* LRP result.color.xyz, fogFactorTemp.xxxx, colorTemp, fogColorRef; */
250 inst->Opcode = OPCODE_LRP;
251 inst->DstReg.File = PROGRAM_OUTPUT;
252 inst->DstReg.Index = FRAG_RESULT_COLR;
253 inst->DstReg.WriteMask = WRITEMASK_XYZ;
254 inst->SrcReg[0].File = PROGRAM_TEMPORARY;
255 inst->SrcReg[0].Index = fogFactorTemp;
256 inst->SrcReg[0].Swizzle
257 = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X);
258 inst->SrcReg[1].File = PROGRAM_TEMPORARY;
259 inst->SrcReg[1].Index = colorTemp;
260 inst->SrcReg[1].Swizzle = SWIZZLE_NOOP;
261 inst->SrcReg[2].File = PROGRAM_STATE_VAR;
262 inst->SrcReg[2].Index = fogColorRef;
263 inst->SrcReg[2].Swizzle = SWIZZLE_NOOP;
264 inst++;
265 /* MOV result.color.w, colorTemp.x; # copy alpha */
266 inst->Opcode = OPCODE_MOV;
267 inst->DstReg.File = PROGRAM_OUTPUT;
268 inst->DstReg.Index = FRAG_RESULT_COLR;
269 inst->DstReg.WriteMask = WRITEMASK_W;
270 inst->SrcReg[0].File = PROGRAM_TEMPORARY;
271 inst->SrcReg[0].Index = colorTemp;
272 inst->SrcReg[0].Swizzle = SWIZZLE_NOOP;
273 inst++;
274 /* END; */
275 inst->Opcode = OPCODE_END;
276 inst++;
277
278 /* free old instructions */
279 _mesa_free(fprog->Base.Instructions);
280
281 /* install new instructions */
282 fprog->Base.Instructions = newInst;
283 fprog->Base.NumInstructions = inst - newInst;
284 fprog->Base.InputsRead |= FRAG_BIT_FOGC;
285 /* XXX do this? fprog->FogOption = GL_NONE; */
286 }