c2f96af2c1215de84e1a635c4e7793df2c5f254b
[mesa.git] / src / mesa / drivers / dri / r300 / r300_vertprog.c
1 /**************************************************************************
2
3 Copyright (C) 2005 Aapo Tahkola <aet@rasterburn.org>
4 Copyright (C) 2008 Oliver McFadden <z3ro.geek@gmail.com>
5
6 All Rights Reserved.
7
8 Permission is hereby granted, free of charge, to any person obtaining a
9 copy of this software and associated documentation files (the "Software"),
10 to deal in the Software without restriction, including without limitation
11 on the rights to use, copy, modify, merge, publish, distribute, sub
12 license, and/or sell copies of the Software, and to permit persons to whom
13 the Software is furnished to do so, subject to the following conditions:
14
15 The above copyright notice and this permission notice (including the next
16 paragraph) shall be included in all copies or substantial portions of the
17 Software.
18
19 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
20 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21 FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
22 THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
23 DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
24 OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
25 USE OR OTHER DEALINGS IN THE SOFTWARE.
26
27 **************************************************************************/
28
29 /* Radeon R5xx Acceleration, Revision 1.2 */
30
31 #include "main/glheader.h"
32 #include "main/macros.h"
33 #include "main/enums.h"
34 #include "shader/program.h"
35 #include "shader/programopt.h"
36 #include "shader/prog_instruction.h"
37 #include "shader/prog_optimize.h"
38 #include "shader/prog_parameter.h"
39 #include "shader/prog_print.h"
40 #include "shader/prog_statevars.h"
41 #include "tnl/tnl.h"
42
43 #include "compiler/radeon_compiler.h"
44 #include "radeon_mesa_to_rc.h"
45 #include "r300_context.h"
46 #include "r300_fragprog_common.h"
47 #include "r300_state.h"
48
49 /**
50 * Write parameter array for the given vertex program into dst.
51 * Return the total number of components written.
52 */
53 static int r300VertexProgUpdateParams(GLcontext * ctx, struct r300_vertex_program *vp, float *dst)
54 {
55 int i;
56
57 if (vp->Base->IsNVProgram) {
58 _mesa_load_tracked_matrices(ctx);
59 } else {
60 if (vp->Base->Base.Parameters) {
61 _mesa_load_state_parameters(ctx, vp->Base->Base.Parameters);
62 }
63 }
64
65 for(i = 0; i < vp->code.constants.Count; ++i) {
66 const float * src = 0;
67 const struct rc_constant * constant = &vp->code.constants.Constants[i];
68
69 switch(constant->Type) {
70 case RC_CONSTANT_EXTERNAL:
71 if (vp->Base->IsNVProgram) {
72 src = ctx->VertexProgram.Parameters[constant->u.External];
73 } else {
74 src = vp->Base->Base.Parameters->ParameterValues[constant->u.External];
75 }
76 break;
77
78 case RC_CONSTANT_IMMEDIATE:
79 src = constant->u.Immediate;
80 break;
81 }
82
83 dst[4*i] = src[0];
84 dst[4*i + 1] = src[1];
85 dst[4*i + 2] = src[2];
86 dst[4*i + 3] = src[3];
87 }
88
89 return 4 * vp->code.constants.Count;
90 }
91
92 static GLbitfield compute_required_outputs(struct gl_vertex_program * vp, GLbitfield fpreads)
93 {
94 GLbitfield outputs = 0;
95 int i;
96
97 #define ADD_OUTPUT(fp_attr, vp_result) \
98 do { \
99 if (fpreads & (1 << (fp_attr))) \
100 outputs |= (1 << (vp_result)); \
101 } while (0)
102
103 ADD_OUTPUT(FRAG_ATTRIB_COL0, VERT_RESULT_COL0);
104 ADD_OUTPUT(FRAG_ATTRIB_COL1, VERT_RESULT_COL1);
105
106 for (i = 0; i <= 7; ++i) {
107 ADD_OUTPUT(FRAG_ATTRIB_TEX0 + i, VERT_RESULT_TEX0 + i);
108 }
109
110 #undef ADD_OUTPUT
111
112 if ((fpreads & (1 << FRAG_ATTRIB_COL0)) &&
113 (vp->Base.OutputsWritten & (1 << VERT_RESULT_BFC0)))
114 outputs |= 1 << VERT_RESULT_BFC0;
115 if ((fpreads & (1 << FRAG_ATTRIB_COL1)) &&
116 (vp->Base.OutputsWritten & (1 << VERT_RESULT_BFC1)))
117 outputs |= 1 << VERT_RESULT_BFC1;
118
119 outputs |= 1 << VERT_RESULT_HPOS;
120 if (vp->Base.OutputsWritten & (1 << VERT_RESULT_PSIZ))
121 outputs |= 1 << VERT_RESULT_PSIZ;
122
123 return outputs;
124 }
125
126
127 static void t_inputs_outputs(struct r300_vertex_program_compiler * c)
128 {
129 int i;
130 int cur_reg;
131 GLuint OutputsWritten, InputsRead;
132
133 OutputsWritten = c->Base.Program.OutputsWritten;
134 InputsRead = c->Base.Program.InputsRead;
135
136 cur_reg = -1;
137 for (i = 0; i < VERT_ATTRIB_MAX; i++) {
138 if (InputsRead & (1 << i))
139 c->code->inputs[i] = ++cur_reg;
140 else
141 c->code->inputs[i] = -1;
142 }
143
144 cur_reg = 0;
145 for (i = 0; i < VERT_RESULT_MAX; i++)
146 c->code->outputs[i] = -1;
147
148 assert(OutputsWritten & (1 << VERT_RESULT_HPOS));
149
150 if (OutputsWritten & (1 << VERT_RESULT_HPOS)) {
151 c->code->outputs[VERT_RESULT_HPOS] = cur_reg++;
152 }
153
154 if (OutputsWritten & (1 << VERT_RESULT_PSIZ)) {
155 c->code->outputs[VERT_RESULT_PSIZ] = cur_reg++;
156 }
157
158 /* If we're writing back facing colors we need to send
159 * four colors to make front/back face colors selection work.
160 * If the vertex program doesn't write all 4 colors, lets
161 * pretend it does by skipping output index reg so the colors
162 * get written into appropriate output vectors.
163 */
164 if (OutputsWritten & (1 << VERT_RESULT_COL0)) {
165 c->code->outputs[VERT_RESULT_COL0] = cur_reg++;
166 } else if (OutputsWritten & (1 << VERT_RESULT_BFC0) ||
167 OutputsWritten & (1 << VERT_RESULT_BFC1)) {
168 cur_reg++;
169 }
170
171 if (OutputsWritten & (1 << VERT_RESULT_COL1)) {
172 c->code->outputs[VERT_RESULT_COL1] = cur_reg++;
173 } else if (OutputsWritten & (1 << VERT_RESULT_BFC0) ||
174 OutputsWritten & (1 << VERT_RESULT_BFC1)) {
175 cur_reg++;
176 }
177
178 if (OutputsWritten & (1 << VERT_RESULT_BFC0)) {
179 c->code->outputs[VERT_RESULT_BFC0] = cur_reg++;
180 } else if (OutputsWritten & (1 << VERT_RESULT_BFC1)) {
181 cur_reg++;
182 }
183
184 if (OutputsWritten & (1 << VERT_RESULT_BFC1)) {
185 c->code->outputs[VERT_RESULT_BFC1] = cur_reg++;
186 } else if (OutputsWritten & (1 << VERT_RESULT_BFC0)) {
187 cur_reg++;
188 }
189
190 for (i = VERT_RESULT_TEX0; i <= VERT_RESULT_TEX7; i++) {
191 if (OutputsWritten & (1 << i)) {
192 c->code->outputs[i] = cur_reg++;
193 }
194 }
195
196 if (OutputsWritten & (1 << VERT_RESULT_FOGC)) {
197 c->code->outputs[VERT_RESULT_FOGC] = cur_reg++;
198 }
199 }
200
201 /**
202 * The NV_vertex_program spec mandates that all registers be
203 * initialized to zero. We do this here unconditionally.
204 *
205 * \note We rely on dead-code elimination in the compiler.
206 */
207 static void initialize_NV_registers(struct radeon_compiler * compiler)
208 {
209 unsigned int reg;
210 struct rc_instruction * inst;
211
212 for(reg = 0; reg < 12; ++reg) {
213 inst = rc_insert_new_instruction(compiler, &compiler->Program.Instructions);
214 inst->U.I.Opcode = RC_OPCODE_MOV;
215 inst->U.I.DstReg.File = RC_FILE_TEMPORARY;
216 inst->U.I.DstReg.Index = reg;
217 inst->U.I.SrcReg[0].File = RC_FILE_NONE;
218 inst->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_0000;
219 }
220
221 inst = rc_insert_new_instruction(compiler, &compiler->Program.Instructions);
222 inst->U.I.Opcode = RC_OPCODE_ARL;
223 inst->U.I.DstReg.File = RC_FILE_ADDRESS;
224 inst->U.I.DstReg.Index = 0;
225 inst->U.I.DstReg.WriteMask = WRITEMASK_X;
226 inst->U.I.SrcReg[0].File = RC_FILE_NONE;
227 inst->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_0000;
228 }
229
230 static struct r300_vertex_program *build_program(GLcontext *ctx,
231 struct r300_vertex_program_key *wanted_key,
232 const struct gl_vertex_program *mesa_vp)
233 {
234 struct r300_vertex_program *vp;
235 struct r300_vertex_program_compiler compiler;
236
237 vp = _mesa_calloc(sizeof(*vp));
238 vp->Base = (struct gl_vertex_program *) _mesa_clone_program(ctx, &mesa_vp->Base);
239 _mesa_memcpy(&vp->key, wanted_key, sizeof(vp->key));
240
241 rc_init(&compiler.Base);
242 compiler.Base.Debug = (RADEON_DEBUG & RADEON_VERTS) ? GL_TRUE : GL_FALSE;
243
244 compiler.code = &vp->code;
245 compiler.RequiredOutputs = compute_required_outputs(vp->Base, vp->key.FpReads);
246 compiler.SetHwInputOutput = &t_inputs_outputs;
247
248 if (compiler.Base.Debug) {
249 fprintf(stderr, "Initial vertex program:\n");
250 _mesa_print_program(&vp->Base->Base);
251 fflush(stderr);
252 }
253
254 if (mesa_vp->IsPositionInvariant) {
255 _mesa_insert_mvp_code(ctx, vp->Base);
256 }
257
258 radeon_mesa_to_rc_program(&compiler.Base, &vp->Base->Base);
259
260 if (mesa_vp->IsNVProgram)
261 initialize_NV_registers(&compiler.Base);
262
263 rc_move_output(&compiler.Base, VERT_RESULT_PSIZ, VERT_RESULT_PSIZ, WRITEMASK_X);
264
265 if (vp->key.WPosAttr != FRAG_ATTRIB_MAX) {
266 rc_copy_output(&compiler.Base,
267 VERT_RESULT_HPOS,
268 vp->key.WPosAttr - FRAG_ATTRIB_TEX0 + VERT_RESULT_TEX0);
269 }
270
271 if (vp->key.FogAttr != FRAG_ATTRIB_MAX) {
272 rc_move_output(&compiler.Base,
273 VERT_RESULT_FOGC,
274 vp->key.FogAttr - FRAG_ATTRIB_TEX0 + VERT_RESULT_TEX0, WRITEMASK_X);
275 }
276
277 r3xx_compile_vertex_program(&compiler);
278
279 if (vp->code.constants.Count > ctx->Const.VertexProgram.MaxParameters) {
280 rc_error(&compiler.Base, "Program exceeds constant buffer size limit\n");
281 }
282
283 vp->error = compiler.Base.Error;
284
285 vp->Base->Base.InputsRead = vp->code.InputsRead;
286 vp->Base->Base.OutputsWritten = vp->code.OutputsWritten;
287
288 rc_destroy(&compiler.Base);
289
290 return vp;
291 }
292
293 struct r300_vertex_program * r300SelectAndTranslateVertexShader(GLcontext *ctx)
294 {
295 r300ContextPtr r300 = R300_CONTEXT(ctx);
296 struct r300_vertex_program_key wanted_key = { 0 };
297 struct r300_vertex_program_cont *vpc;
298 struct r300_vertex_program *vp;
299
300 vpc = (struct r300_vertex_program_cont *)ctx->VertexProgram._Current;
301
302 if (!r300->selected_fp) {
303 /* This can happen when GetProgramiv is called to check
304 * whether the program runs natively.
305 *
306 * To be honest, this is not a very good solution,
307 * but solving the problem of reporting good values
308 * for those queries is tough anyway considering that
309 * we recompile vertex programs based on the precise
310 * fragment program that is in use.
311 */
312 r300SelectAndTranslateFragmentShader(ctx);
313 }
314
315 wanted_key.FpReads = r300->selected_fp->InputsRead;
316 wanted_key.FogAttr = r300->selected_fp->fog_attr;
317 wanted_key.WPosAttr = r300->selected_fp->wpos_attr;
318
319 for (vp = vpc->progs; vp; vp = vp->next) {
320 if (_mesa_memcmp(&vp->key, &wanted_key, sizeof(wanted_key))
321 == 0) {
322 return r300->selected_vp = vp;
323 }
324 }
325
326 vp = build_program(ctx, &wanted_key, &vpc->mesa_program);
327 vp->next = vpc->progs;
328 vpc->progs = vp;
329
330 return r300->selected_vp = vp;
331 }
332
333 #define bump_vpu_count(ptr, new_count) do { \
334 drm_r300_cmd_header_t* _p=((drm_r300_cmd_header_t*)(ptr)); \
335 int _nc=(new_count)/4; \
336 if(_nc>_p->vpu.count)_p->vpu.count=_nc; \
337 } while(0)
338
339 static void r300EmitVertexProgram(r300ContextPtr r300, int dest, struct r300_vertex_program_code *code)
340 {
341 int i;
342
343 assert((code->length > 0) && (code->length % 4 == 0));
344
345 R300_STATECHANGE( r300, vap_flush );
346
347 switch ((dest >> 8) & 0xf) {
348 case 0:
349 R300_STATECHANGE(r300, vpi);
350 for (i = 0; i < code->length; i++)
351 r300->hw.vpi.cmd[R300_VPI_INSTR_0 + i + 4 * (dest & 0xff)] = (code->body.d[i]);
352 bump_vpu_count(r300->hw.vpi.cmd, code->length + 4 * (dest & 0xff));
353 break;
354 case 2:
355 R300_STATECHANGE(r300, vpp);
356 for (i = 0; i < code->length; i++)
357 r300->hw.vpp.cmd[R300_VPP_PARAM_0 + i + 4 * (dest & 0xff)] = (code->body.d[i]);
358 bump_vpu_count(r300->hw.vpp.cmd, code->length + 4 * (dest & 0xff));
359 break;
360 case 4:
361 R300_STATECHANGE(r300, vps);
362 for (i = 0; i < code->length; i++)
363 r300->hw.vps.cmd[1 + i + 4 * (dest & 0xff)] = (code->body.d[i]);
364 bump_vpu_count(r300->hw.vps.cmd, code->length + 4 * (dest & 0xff));
365 break;
366 default:
367 fprintf(stderr, "%s:%s don't know how to handle dest %04x\n", __FILE__, __FUNCTION__, dest);
368 _mesa_exit(-1);
369 }
370 }
371
372 void r300SetupVertexProgram(r300ContextPtr rmesa)
373 {
374 GLcontext *ctx = rmesa->radeon.glCtx;
375 struct r300_vertex_program *prog = rmesa->selected_vp;
376 int inst_count = 0;
377 int param_count = 0;
378
379 /* Reset state, in case we don't use something */
380 ((drm_r300_cmd_header_t *) rmesa->hw.vpp.cmd)->vpu.count = 0;
381 ((drm_r300_cmd_header_t *) rmesa->hw.vpi.cmd)->vpu.count = 0;
382 ((drm_r300_cmd_header_t *) rmesa->hw.vps.cmd)->vpu.count = 0;
383
384 R300_STATECHANGE(rmesa, vap_flush);
385 R300_STATECHANGE(rmesa, vpp);
386 param_count = r300VertexProgUpdateParams(ctx, prog, (float *)&rmesa->hw.vpp.cmd[R300_VPP_PARAM_0]);
387 bump_vpu_count(rmesa->hw.vpp.cmd, param_count);
388 param_count /= 4;
389
390 r300EmitVertexProgram(rmesa, R300_PVS_CODE_START, &(prog->code));
391 inst_count = (prog->code.length / 4) - 1;
392
393 r300VapCntl(rmesa, _mesa_bitcount(prog->code.InputsRead),
394 _mesa_bitcount(prog->code.OutputsWritten), prog->code.num_temporaries);
395
396 R300_STATECHANGE(rmesa, pvs);
397 rmesa->hw.pvs.cmd[R300_PVS_CNTL_1] = (0 << R300_PVS_FIRST_INST_SHIFT) | (inst_count << R300_PVS_XYZW_VALID_INST_SHIFT) |
398 (inst_count << R300_PVS_LAST_INST_SHIFT);
399
400 rmesa->hw.pvs.cmd[R300_PVS_CNTL_2] = (0 << R300_PVS_CONST_BASE_OFFSET_SHIFT) | (param_count << R300_PVS_MAX_CONST_ADDR_SHIFT);
401 rmesa->hw.pvs.cmd[R300_PVS_CNTL_3] = (inst_count << R300_PVS_LAST_VTX_SRC_INST_SHIFT);
402 }