r300: Added a TODO note and some tiny cleanups to r300_emit.c.
[mesa.git] / src / mesa / drivers / dri / r300 / r300_emit.c
1 /*
2 Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved.
3
4 The Weather Channel (TM) funded Tungsten Graphics to develop the
5 initial release of the Radeon 8500 driver under the XFree86 license.
6 This notice must be preserved.
7
8 Permission is hereby granted, free of charge, to any person obtaining
9 a copy of this software and associated documentation files (the
10 "Software"), to deal in the Software without restriction, including
11 without limitation the rights to use, copy, modify, merge, publish,
12 distribute, sublicense, and/or sell copies of the Software, and to
13 permit persons to whom the Software is furnished to do so, subject to
14 the following conditions:
15
16 The above copyright notice and this permission notice (including the
17 next paragraph) shall be included in all copies or substantial
18 portions of the Software.
19
20 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
21 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
23 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
24 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
25 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
26 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27
28 **************************************************************************/
29
30 /**
31 * \file
32 *
33 * \author Keith Whitwell <keith@tungstengraphics.com>
34 */
35
36 #include "glheader.h"
37 #include "mtypes.h"
38 #include "colormac.h"
39 #include "imports.h"
40 #include "macros.h"
41 #include "image.h"
42
43 #include "swrast_setup/swrast_setup.h"
44 #include "math/m_translate.h"
45 #include "tnl/tnl.h"
46 #include "tnl/t_context.h"
47
48 #include "r300_context.h"
49 #include "radeon_ioctl.h"
50 #include "r300_state.h"
51 #include "r300_emit.h"
52 #include "r300_ioctl.h"
53
54 #ifdef USER_BUFFERS
55 #include "r300_mem.h"
56 #endif
57
58 #if SWIZZLE_X != R300_INPUT_ROUTE_SELECT_X || \
59 SWIZZLE_Y != R300_INPUT_ROUTE_SELECT_Y || \
60 SWIZZLE_Z != R300_INPUT_ROUTE_SELECT_Z || \
61 SWIZZLE_W != R300_INPUT_ROUTE_SELECT_W || \
62 SWIZZLE_ZERO != R300_INPUT_ROUTE_SELECT_ZERO || \
63 SWIZZLE_ONE != R300_INPUT_ROUTE_SELECT_ONE
64 #error Cannot change these!
65 #endif
66
67 #define DEBUG_ALL DEBUG_VERTS
68
69 #if defined(USE_X86_ASM)
70 #define COPY_DWORDS( dst, src, nr ) \
71 do { \
72 int __tmp; \
73 __asm__ __volatile__( "rep ; movsl" \
74 : "=%c" (__tmp), "=D" (dst), "=S" (__tmp) \
75 : "0" (nr), \
76 "D" ((long)dst), \
77 "S" ((long)src) ); \
78 } while (0)
79 #else
80 #define COPY_DWORDS( dst, src, nr ) \
81 do { \
82 int j; \
83 for ( j = 0 ; j < nr ; j++ ) \
84 dst[j] = ((int *)src)[j]; \
85 dst += nr; \
86 } while (0)
87 #endif
88
89 static void r300EmitVec4(GLcontext * ctx,
90 struct r300_dma_region *rvb,
91 GLvoid * data, int stride, int count)
92 {
93 int i;
94 int *out = (int *)(rvb->address + rvb->start);
95
96 if (RADEON_DEBUG & DEBUG_VERTS)
97 fprintf(stderr, "%s count %d stride %d\n",
98 __FUNCTION__, count, stride);
99
100 if (stride == 4)
101 COPY_DWORDS(out, data, count);
102 else
103 for (i = 0; i < count; i++) {
104 out[0] = *(int *)data;
105 out++;
106 data += stride;
107 }
108 }
109
110 static void r300EmitVec8(GLcontext * ctx,
111 struct r300_dma_region *rvb,
112 GLvoid * data, int stride, int count)
113 {
114 int i;
115 int *out = (int *)(rvb->address + rvb->start);
116
117 if (RADEON_DEBUG & DEBUG_VERTS)
118 fprintf(stderr, "%s count %d stride %d\n",
119 __FUNCTION__, count, stride);
120
121 if (stride == 8)
122 COPY_DWORDS(out, data, count * 2);
123 else
124 for (i = 0; i < count; i++) {
125 out[0] = *(int *)data;
126 out[1] = *(int *)(data + 4);
127 out += 2;
128 data += stride;
129 }
130 }
131
132 static void r300EmitVec12(GLcontext * ctx,
133 struct r300_dma_region *rvb,
134 GLvoid * data, int stride, int count)
135 {
136 int i;
137 int *out = (int *)(rvb->address + rvb->start);
138
139 if (RADEON_DEBUG & DEBUG_VERTS)
140 fprintf(stderr, "%s count %d stride %d out %p data %p\n",
141 __FUNCTION__, count, stride, (void *)out, (void *)data);
142
143 if (stride == 12)
144 COPY_DWORDS(out, data, count * 3);
145 else
146 for (i = 0; i < count; i++) {
147 out[0] = *(int *)data;
148 out[1] = *(int *)(data + 4);
149 out[2] = *(int *)(data + 8);
150 out += 3;
151 data += stride;
152 }
153 }
154
155 static void r300EmitVec16(GLcontext * ctx,
156 struct r300_dma_region *rvb,
157 GLvoid * data, int stride, int count)
158 {
159 int i;
160 int *out = (int *)(rvb->address + rvb->start);
161
162 if (RADEON_DEBUG & DEBUG_VERTS)
163 fprintf(stderr, "%s count %d stride %d\n",
164 __FUNCTION__, count, stride);
165
166 if (stride == 16)
167 COPY_DWORDS(out, data, count * 4);
168 else
169 for (i = 0; i < count; i++) {
170 out[0] = *(int *)data;
171 out[1] = *(int *)(data + 4);
172 out[2] = *(int *)(data + 8);
173 out[3] = *(int *)(data + 12);
174 out += 4;
175 data += stride;
176 }
177 }
178
179 static void r300EmitVec(GLcontext * ctx,
180 struct r300_dma_region *rvb,
181 GLvoid * data, int size, int stride, int count)
182 {
183 r300ContextPtr rmesa = R300_CONTEXT(ctx);
184
185 if (RADEON_DEBUG & DEBUG_VERTS)
186 fprintf(stderr, "%s count %d size %d stride %d\n",
187 __FUNCTION__, count, size, stride);
188
189 /* Gets triggered when playing with future_hw_tcl_on ... */
190 //assert(!rvb->buf);
191
192 if (stride == 0) {
193 r300AllocDmaRegion(rmesa, rvb, size * 4, 4);
194 count = 1;
195 rvb->aos_offset = GET_START(rvb);
196 rvb->aos_stride = 0;
197 } else {
198 r300AllocDmaRegion(rmesa, rvb, size * count * 4, 4); /* alignment? */
199 rvb->aos_offset = GET_START(rvb);
200 rvb->aos_stride = size;
201 }
202
203 /* Emit the data
204 */
205 switch (size) {
206 case 1:
207 r300EmitVec4(ctx, rvb, data, stride, count);
208 break;
209 case 2:
210 r300EmitVec8(ctx, rvb, data, stride, count);
211 break;
212 case 3:
213 r300EmitVec12(ctx, rvb, data, stride, count);
214 break;
215 case 4:
216 r300EmitVec16(ctx, rvb, data, stride, count);
217 break;
218 default:
219 assert(0);
220 _mesa_exit(-1);
221 break;
222 }
223
224 }
225
226 /* TODO: explain this... */
227 #define R300_VIR0_AOS_SIZE_SHIFT 0
228 #define R300_VIR0_AOS_INPUT_SHIFT 8
229 #define R300_VIR0_AOS_STOP_SHIFT 13
230 #define R300_VIR0_AOS_TYPE_SHIFT 14
231 #define R300_VIR0_HIGH_SHIFT 16
232
233 // Pack 4 elemets in a 16 bit (aos_size first 8, input next 5, 1 stop bit(Whild gues), aos_type last 2);
234 static inline GLuint t_vir_pack(GLvector4f ** dt, int *inputs, int i)
235 {
236 GLuint dw;
237 dw = (dt[i]->size - 1) << R300_VIR0_AOS_SIZE_SHIFT;
238 dw |= inputs[i] << R300_VIR0_AOS_INPUT_SHIFT;
239 //dw |= t_type(&dt[i]) << R300_VIR0_AOS_TYPE_SHIFT;
240 return dw;
241 }
242
243 static GLuint t_vir0(uint32_t * dst, GLvector4f ** dt, int *inputs,
244 GLint * tab, GLuint nr)
245 {
246 GLuint i, dw, dwInternel;
247
248 for (i = 0; i + 1 < nr; i += 2) {
249 dw = t_vir_pack(dt, inputs, tab[i]);
250 dwInternel = t_vir_pack(dt, inputs, tab[i + 1]);
251 dw |= dwInternel << R300_VIR0_HIGH_SHIFT;
252
253 if (i + 2 == nr) {
254 dw |=
255 (1 <<
256 (R300_VIR0_AOS_STOP_SHIFT + R300_VIR0_HIGH_SHIFT));
257 }
258 dst[i >> 1] = dw; // Is the same as i/2
259 }
260
261 if (nr & 1) {
262 dw = t_vir_pack(dt, inputs, tab[nr - 1]);
263 dw |= 1 << R300_VIR0_AOS_STOP_SHIFT;
264
265 dst[nr >> 1] = dw;
266 }
267
268 return (nr + 1) >> 1; // Is the same as (nr+1)/2
269 }
270
271 static GLuint t_swizzle(int swizzle[4])
272 {
273 return (swizzle[0] << R300_INPUT_ROUTE_X_SHIFT) |
274 (swizzle[1] << R300_INPUT_ROUTE_Y_SHIFT) |
275 (swizzle[2] << R300_INPUT_ROUTE_Z_SHIFT) |
276 (swizzle[3] << R300_INPUT_ROUTE_W_SHIFT);
277 }
278
279 static GLuint t_vir1(uint32_t * dst, int swizzle[][4], GLuint nr)
280 {
281 GLuint i;
282
283 for (i = 0; i + 1 < nr; i += 2) {
284 dst[i >> 1] = t_swizzle(swizzle[i]) | R300_INPUT_ROUTE_ENABLE;
285 dst[i >> 1] |=
286 (t_swizzle(swizzle[i + 1]) | R300_INPUT_ROUTE_ENABLE)
287 << 16;
288 }
289
290 if (nr & 1)
291 dst[nr >> 1] =
292 t_swizzle(swizzle[nr - 1]) | R300_INPUT_ROUTE_ENABLE;
293
294 return (nr + 1) >> 1;
295 }
296
297 static GLuint t_vic(GLcontext * ctx, GLuint InputsRead)
298 {
299 r300ContextPtr r300 = R300_CONTEXT(ctx);
300 GLuint i, vic_1 = 0;
301
302 if (InputsRead & (1 << VERT_ATTRIB_POS))
303 vic_1 |= R300_INPUT_CNTL_POS;
304
305 if (InputsRead & (1 << VERT_ATTRIB_NORMAL))
306 vic_1 |= R300_INPUT_CNTL_NORMAL;
307
308 if (InputsRead & (1 << VERT_ATTRIB_COLOR0))
309 vic_1 |= R300_INPUT_CNTL_COLOR;
310
311 r300->state.texture.tc_count = 0;
312 for (i = 0; i < ctx->Const.MaxTextureUnits; i++)
313 if (InputsRead & (1 << (VERT_ATTRIB_TEX0 + i))) {
314 r300->state.texture.tc_count++;
315 vic_1 |= R300_INPUT_CNTL_TC0 << i;
316 }
317
318 return vic_1;
319 }
320
321 /* Emit vertex data to GART memory
322 * Route inputs to the vertex processor
323 * This function should never return R300_FALLBACK_TCL when using software tcl.
324 */
325
326 int r300EmitArrays(GLcontext * ctx)
327 {
328 r300ContextPtr rmesa = R300_CONTEXT(ctx);
329 r300ContextPtr r300 = rmesa;
330 TNLcontext *tnl = TNL_CONTEXT(ctx);
331 struct vertex_buffer *vb = &tnl->vb;
332 GLuint nr;
333 GLuint count = vb->Count;
334 GLuint i;
335 GLuint InputsRead = 0, OutputsWritten = 0;
336 int *inputs = NULL;
337 int vir_inputs[VERT_ATTRIB_MAX];
338 GLint tab[VERT_ATTRIB_MAX];
339 int swizzle[VERT_ATTRIB_MAX][4];
340
341 if (hw_tcl_on) {
342 struct r300_vertex_program *prog =
343 (struct r300_vertex_program *)
344 CURRENT_VERTEX_SHADER(ctx);
345 inputs = prog->inputs;
346 InputsRead = CURRENT_VERTEX_SHADER(ctx)->key.InputsRead;
347 OutputsWritten = CURRENT_VERTEX_SHADER(ctx)->key.OutputsWritten;
348 } else {
349 DECLARE_RENDERINPUTS(inputs_bitset);
350 inputs = r300->state.sw_tcl_inputs;
351
352 RENDERINPUTS_COPY(inputs_bitset,
353 TNL_CONTEXT(ctx)->render_inputs_bitset);
354
355 assert(RENDERINPUTS_TEST(inputs_bitset, _TNL_ATTRIB_POS));
356 InputsRead |= 1 << VERT_ATTRIB_POS;
357 OutputsWritten |= 1 << VERT_RESULT_HPOS;
358
359 assert(RENDERINPUTS_TEST(inputs_bitset, _TNL_ATTRIB_NORMAL)
360 == 0);
361
362 assert(RENDERINPUTS_TEST(inputs_bitset, _TNL_ATTRIB_COLOR0));
363 InputsRead |= 1 << VERT_ATTRIB_COLOR0;
364 OutputsWritten |= 1 << VERT_RESULT_COL0;
365
366 if (RENDERINPUTS_TEST(inputs_bitset, _TNL_ATTRIB_COLOR1)) {
367 InputsRead |= 1 << VERT_ATTRIB_COLOR1;
368 OutputsWritten |= 1 << VERT_RESULT_COL1;
369 }
370
371 for (i = 0; i < ctx->Const.MaxTextureUnits; i++)
372 if (RENDERINPUTS_TEST
373 (inputs_bitset, _TNL_ATTRIB_TEX(i))) {
374 InputsRead |= 1 << (VERT_ATTRIB_TEX0 + i);
375 OutputsWritten |= 1 << (VERT_RESULT_TEX0 + i);
376 }
377
378 for (i = 0, nr = 0; i < VERT_ATTRIB_MAX; i++)
379 if (InputsRead & (1 << i))
380 inputs[i] = nr++;
381 else
382 inputs[i] = -1;
383
384 if (!
385 (r300->radeon.radeonScreen->
386 chip_flags & RADEON_CHIPSET_TCL)) {
387 /* Fixed, apply to vir0 only */
388 memcpy(vir_inputs, inputs,
389 VERT_ATTRIB_MAX * sizeof(int));
390 inputs = vir_inputs;
391
392 if (InputsRead & VERT_ATTRIB_POS)
393 inputs[VERT_ATTRIB_POS] = 0;
394
395 if (InputsRead & (1 << VERT_ATTRIB_COLOR0))
396 inputs[VERT_ATTRIB_COLOR0] = 2;
397
398 if (InputsRead & (1 << VERT_ATTRIB_COLOR1))
399 inputs[VERT_ATTRIB_COLOR1] = 3;
400
401 for (i = VERT_ATTRIB_TEX0; i <= VERT_ATTRIB_TEX7; i++)
402 if (InputsRead & (1 << i))
403 inputs[i] = 6 + (i - VERT_ATTRIB_TEX0);
404 }
405
406 RENDERINPUTS_COPY(rmesa->state.render_inputs_bitset,
407 inputs_bitset);
408 }
409
410 assert(InputsRead);
411 assert(OutputsWritten);
412
413 for (i = 0, nr = 0; i < VERT_ATTRIB_MAX; i++)
414 if (InputsRead & (1 << i))
415 tab[nr++] = i;
416
417 if (nr > R300_MAX_AOS_ARRAYS)
418 return R300_FALLBACK_TCL;
419
420 for (i = 0; i < nr; i++) {
421 int ci;
422 int comp_size, fix, found = 0;
423
424 swizzle[i][0] = SWIZZLE_ZERO;
425 swizzle[i][1] = SWIZZLE_ZERO;
426 swizzle[i][2] = SWIZZLE_ZERO;
427 swizzle[i][3] = SWIZZLE_ONE;
428
429 for (ci = 0; ci < vb->AttribPtr[tab[i]]->size; ci++)
430 swizzle[i][ci] = ci;
431
432 if (r300IsGartMemory(rmesa, vb->AttribPtr[tab[i]]->data, 4)) {
433 if (vb->AttribPtr[tab[i]]->stride % 4)
434 return R300_FALLBACK_TCL;
435
436 rmesa->state.aos[i].address =
437 (void *)(vb->AttribPtr[tab[i]]->data);
438 rmesa->state.aos[i].start = 0;
439 rmesa->state.aos[i].aos_offset =
440 r300GartOffsetFromVirtual(rmesa,
441 vb->
442 AttribPtr[tab[i]]->data);
443 rmesa->state.aos[i].aos_stride =
444 vb->AttribPtr[tab[i]]->stride / 4;
445
446 rmesa->state.aos[i].aos_size =
447 vb->AttribPtr[tab[i]]->size;
448 } else {
449 r300EmitVec(ctx, &rmesa->state.aos[i],
450 vb->AttribPtr[tab[i]]->data,
451 vb->AttribPtr[tab[i]]->size,
452 vb->AttribPtr[tab[i]]->stride, count);
453 }
454
455 rmesa->state.aos[i].aos_size = vb->AttribPtr[tab[i]]->size;
456
457 comp_size = _mesa_sizeof_type(GL_FLOAT);
458
459 for (fix = 0; fix <= 4 - vb->AttribPtr[tab[i]]->size; fix++) {
460 if ((rmesa->state.aos[i].aos_offset -
461 comp_size * fix) % 4)
462 continue;
463
464 found = 1;
465 break;
466 }
467
468 if (found) {
469 if (fix > 0) {
470 WARN_ONCE("Feeling lucky?\n");
471 }
472
473 rmesa->state.aos[i].aos_offset -= comp_size * fix;
474
475 for (ci = 0; ci < vb->AttribPtr[tab[i]]->size; ci++)
476 swizzle[i][ci] += fix;
477 } else {
478 WARN_ONCE
479 ("Cannot handle offset %x with stride %d, comp %d\n",
480 rmesa->state.aos[i].aos_offset,
481 rmesa->state.aos[i].aos_stride,
482 vb->AttribPtr[tab[i]]->size);
483 return R300_FALLBACK_TCL;
484 }
485 }
486
487 /* setup INPUT_ROUTE */
488 R300_STATECHANGE(r300, vir[0]);
489 ((drm_r300_cmd_header_t *) r300->hw.vir[0].cmd)->packet0.count =
490 t_vir0(&r300->hw.vir[0].cmd[R300_VIR_CNTL_0], vb->AttribPtr,
491 inputs, tab, nr);
492
493 R300_STATECHANGE(r300, vir[1]);
494 ((drm_r300_cmd_header_t *) r300->hw.vir[1].cmd)->packet0.count =
495 t_vir1(&r300->hw.vir[1].cmd[R300_VIR_CNTL_0], swizzle, nr);
496
497 /* Set up input_cntl */
498 /* I don't think this is needed for vertex buffers, but it doesn't hurt anything */
499 R300_STATECHANGE(r300, vic);
500 r300->hw.vic.cmd[R300_VIC_CNTL_0] = 0x5555; /* Hard coded value, no idea what it means */
501 r300->hw.vic.cmd[R300_VIC_CNTL_1] = t_vic(ctx, InputsRead);
502
503 /* Stage 3: VAP output */
504
505 R300_STATECHANGE(r300, vof);
506
507 r300->hw.vof.cmd[R300_VOF_CNTL_0] = 0;
508 r300->hw.vof.cmd[R300_VOF_CNTL_1] = 0;
509
510 if (OutputsWritten & (1 << VERT_RESULT_HPOS))
511 r300->hw.vof.cmd[R300_VOF_CNTL_0] |=
512 R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
513
514 if (OutputsWritten & (1 << VERT_RESULT_COL0))
515 r300->hw.vof.cmd[R300_VOF_CNTL_0] |=
516 R300_VAP_OUTPUT_VTX_FMT_0__COLOR_PRESENT;
517
518 if (OutputsWritten & (1 << VERT_RESULT_COL1))
519 r300->hw.vof.cmd[R300_VOF_CNTL_0] |=
520 R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT;
521
522 /*if(OutputsWritten & (1 << VERT_RESULT_BFC0))
523 r300->hw.vof.cmd[R300_VOF_CNTL_0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_2_PRESENT;
524
525 if(OutputsWritten & (1 << VERT_RESULT_BFC1))
526 r300->hw.vof.cmd[R300_VOF_CNTL_0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_3_PRESENT; */
527 //if(OutputsWritten & (1 << VERT_RESULT_FOGC))
528
529 if (OutputsWritten & (1 << VERT_RESULT_PSIZ))
530 r300->hw.vof.cmd[R300_VOF_CNTL_0] |=
531 R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
532
533 for (i = 0; i < ctx->Const.MaxTextureUnits; i++)
534 if (OutputsWritten & (1 << (VERT_RESULT_TEX0 + i)))
535 r300->hw.vof.cmd[R300_VOF_CNTL_1] |= (4 << (3 * i));
536
537 rmesa->state.aos_count = nr;
538
539 return R300_FALLBACK_NONE;
540 }
541
542 #ifdef USER_BUFFERS
543 void r300UseArrays(GLcontext * ctx)
544 {
545 r300ContextPtr rmesa = R300_CONTEXT(ctx);
546 int i;
547
548 if (rmesa->state.elt_dma.buf)
549 r300_mem_use(rmesa, rmesa->state.elt_dma.buf->id);
550
551 for (i = 0; i < rmesa->state.aos_count; i++) {
552 if (rmesa->state.aos[i].buf)
553 r300_mem_use(rmesa, rmesa->state.aos[i].buf->id);
554 }
555 }
556 #endif
557
558 void r300ReleaseArrays(GLcontext * ctx)
559 {
560 r300ContextPtr rmesa = R300_CONTEXT(ctx);
561 int i;
562
563 r300ReleaseDmaRegion(rmesa, &rmesa->state.elt_dma, __FUNCTION__);
564 for (i = 0; i < rmesa->state.aos_count; i++) {
565 r300ReleaseDmaRegion(rmesa, &rmesa->state.aos[i], __FUNCTION__);
566 }
567 }