1/*
2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2009 Marek Olšák <maraeo@gmail.com>
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
23
24/* r300_emit: Functions for emitting state. */
25
26#include "util/format/u_format.h"
27#include "util/u_math.h"
28
29#include "r300_context.h"
30#include "r300_cb.h"
31#include "r300_cs.h"
32#include "r300_emit.h"
33#include "r300_fs.h"
34#include "r300_screen.h"
35#include "r300_screen_buffer.h"
36#include "r300_vs.h"
37
38void r300_emit_blend_state(struct r300_context* r300,
39                           unsigned size, void* state)
40{
41    struct r300_blend_state* blend = (struct r300_blend_state*)state;
42    struct pipe_framebuffer_state* fb =
43        (struct pipe_framebuffer_state*)r300->fb_state.state;
44    struct pipe_surface *cb;
45    CS_LOCALS(r300);
46
47    cb = fb->nr_cbufs ? r300_get_nonnull_cb(fb, 0) : NULL;
48
49    if (cb) {
50        if (cb->format == PIPE_FORMAT_R16G16B16A16_FLOAT) {
51            WRITE_CS_TABLE(blend->cb_noclamp, size);
52        } else if (cb->format == PIPE_FORMAT_R16G16B16X16_FLOAT) {
53            WRITE_CS_TABLE(blend->cb_noclamp_noalpha, size);
54        } else {
55            unsigned swz = r300_surface(cb)->colormask_swizzle;
56            WRITE_CS_TABLE(blend->cb_clamp[swz], size);
57        }
58    } else {
59        WRITE_CS_TABLE(blend->cb_no_readwrite, size);
60    }
61}
62
63void r300_emit_blend_color_state(struct r300_context* r300,
64                                 unsigned size, void* state)
65{
66    struct r300_blend_color_state* bc = (struct r300_blend_color_state*)state;
67    CS_LOCALS(r300);
68
69    WRITE_CS_TABLE(bc->cb, size);
70}
71
72void r300_emit_clip_state(struct r300_context* r300,
73                          unsigned size, void* state)
74{
75    struct r300_clip_state* clip = (struct r300_clip_state*)state;
76    CS_LOCALS(r300);
77
78    WRITE_CS_TABLE(clip->cb, size);
79}
80
81void r300_emit_dsa_state(struct r300_context* r300, unsigned size, void* state)
82{
83    struct r300_dsa_state* dsa = (struct r300_dsa_state*)state;
84    struct pipe_framebuffer_state* fb =
85        (struct pipe_framebuffer_state*)r300->fb_state.state;
86    boolean is_r500 = r300->screen->caps.is_r500;
87    CS_LOCALS(r300);
88    uint32_t alpha_func = dsa->alpha_function;
89
90    /* Choose the alpha ref value between 8-bit (FG_ALPHA_FUNC.AM_VAL) and
91     * 16-bit (FG_ALPHA_VALUE). */
92    if (is_r500 && (alpha_func & R300_FG_ALPHA_FUNC_ENABLE)) {
93        struct pipe_surface *cb = fb->nr_cbufs ? r300_get_nonnull_cb(fb, 0) : NULL;
94
95        if (cb &&
96            (cb->format == PIPE_FORMAT_R16G16B16A16_FLOAT ||
97             cb->format == PIPE_FORMAT_R16G16B16X16_FLOAT)) {
98            alpha_func |= R500_FG_ALPHA_FUNC_FP16_ENABLE;
99        } else {
100            alpha_func |= R500_FG_ALPHA_FUNC_8BIT;
101        }
102    }
103
104    /* Setup alpha-to-coverage. */
105    if (r300->alpha_to_coverage && r300->msaa_enable) {
106        /* Always set 3/6, it improves precision even for 2x and 4x MSAA. */
107        alpha_func |= R300_FG_ALPHA_FUNC_MASK_ENABLE |
108                      R300_FG_ALPHA_FUNC_CFG_3_OF_6;
109    }
110
111    BEGIN_CS(size);
112    OUT_CS_REG(R300_FG_ALPHA_FUNC, alpha_func);
113    OUT_CS_TABLE(fb->zsbuf ? &dsa->cb_begin : dsa->cb_zb_no_readwrite, size-2);
114    END_CS;
115}
116
117static void get_rc_constant_state(
118    float vec[4],
119    struct r300_context * r300,
120    struct rc_constant * constant)
121{
122    struct r300_textures_state* texstate = r300->textures_state.state;
123    struct r300_resource *tex;
124
125    assert(constant->Type == RC_CONSTANT_STATE);
126
127    /* vec should either be (0, 0, 0, 1), which should be a relatively safe
128     * RGBA or STRQ value, or it could be one of the RC_CONSTANT_STATE
129     * state factors. */
130
131    switch (constant->u.State[0]) {
132        /* Factor for converting rectangle coords to
133         * normalized coords. Should only show up on non-r500. */
134        case RC_STATE_R300_TEXRECT_FACTOR:
135            tex = r300_resource(texstate->sampler_views[constant->u.State[1]]->base.texture);
136            vec[0] = 1.0 / tex->tex.width0;
137            vec[1] = 1.0 / tex->tex.height0;
138            vec[2] = 0;
139            vec[3] = 1;
140            break;
141
142        case RC_STATE_R300_TEXSCALE_FACTOR:
143            tex = r300_resource(texstate->sampler_views[constant->u.State[1]]->base.texture);
144            /* Add a small number to the texture size to work around rounding errors in hw. */
145            vec[0] = tex->b.width0  / (tex->tex.width0  + 0.001f);
146            vec[1] = tex->b.height0 / (tex->tex.height0 + 0.001f);
147            vec[2] = tex->b.depth0  / (tex->tex.depth0  + 0.001f);
148            vec[3] = 1;
149            break;
150
151        case RC_STATE_R300_VIEWPORT_SCALE:
152            vec[0] = r300->viewport.scale[0];
153            vec[1] = r300->viewport.scale[1];
154            vec[2] = r300->viewport.scale[2];
155            vec[3] = 1;
156            break;
157
158        case RC_STATE_R300_VIEWPORT_OFFSET:
159            vec[0] = r300->viewport.translate[0];
160            vec[1] = r300->viewport.translate[1];
161            vec[2] = r300->viewport.translate[2];
162            vec[3] = 1;
163            break;
164
165        default:
166            fprintf(stderr, "r300: Implementation error: "
167                "Unknown RC_CONSTANT type %d\n", constant->u.State[0]);
168            vec[0] = 0;
169            vec[1] = 0;
170            vec[2] = 0;
171            vec[3] = 1;
172    }
173}
174
175/* Convert a normal single-precision float into the 7.16 format
176 * used by the R300 fragment shader.
177 */
178uint32_t pack_float24(float f)
179{
180    union {
181        float fl;
182        uint32_t u;
183    } u;
184    float mantissa;
185    int exponent;
186    uint32_t float24 = 0;
187
188    if (f == 0.0)
189        return 0;
190
191    u.fl = f;
192
193    mantissa = frexpf(f, &exponent);
194
195    /* Handle -ve */
196    if (mantissa < 0) {
197        float24 |= (1 << 23);
198        mantissa = mantissa * -1.0;
199    }
200    /* Handle exponent, bias of 63 */
201    exponent += 62;
202    float24 |= (exponent << 16);
203    /* Kill 7 LSB of mantissa */
204    float24 |= (u.u & 0x7FFFFF) >> 7;
205
206    return float24;
207}
208
209void r300_emit_fs(struct r300_context* r300, unsigned size, void *state)
210{
211    struct r300_fragment_shader *fs = r300_fs(r300);
212    CS_LOCALS(r300);
213
214    WRITE_CS_TABLE(fs->shader->cb_code, fs->shader->cb_code_size);
215}
216
217void r300_emit_fs_constants(struct r300_context* r300, unsigned size, void *state)
218{
219    struct r300_fragment_shader *fs = r300_fs(r300);
220    struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state;
221    unsigned count = fs->shader->externals_count;
222    unsigned i, j;
223    CS_LOCALS(r300);
224
225    if (count == 0)
226        return;
227
228    BEGIN_CS(size);
229    OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X, count * 4);
230    if (buf->remap_table){
231        for (i = 0; i < count; i++) {
232            float *data = (float*)&buf->ptr[buf->remap_table[i]*4];
233            for (j = 0; j < 4; j++)
234                OUT_CS(pack_float24(data[j]));
235        }
236    } else {
237        for (i = 0; i < count; i++)
238            for (j = 0; j < 4; j++)
239                OUT_CS(pack_float24(*(float*)&buf->ptr[i*4+j]));
240    }
241
242    END_CS;
243}
244
245void r300_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, void *state)
246{
247    struct r300_fragment_shader *fs = r300_fs(r300);
248    struct rc_constant_list *constants = &fs->shader->code.constants;
249    unsigned i;
250    unsigned count = fs->shader->rc_state_count;
251    unsigned first = fs->shader->externals_count;
252    unsigned end = constants->Count;
253    unsigned j;
254    CS_LOCALS(r300);
255
256    if (count == 0)
257        return;
258
259    BEGIN_CS(size);
260    for(i = first; i < end; ++i) {
261        if (constants->Constants[i].Type == RC_CONSTANT_STATE) {
262            float data[4];
263
264            get_rc_constant_state(data, r300, &constants->Constants[i]);
265
266            OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4);
267            for (j = 0; j < 4; j++)
268                OUT_CS(pack_float24(data[j]));
269        }
270    }
271    END_CS;
272}
273
274void r500_emit_fs(struct r300_context* r300, unsigned size, void *state)
275{
276    struct r300_fragment_shader *fs = r300_fs(r300);
277    CS_LOCALS(r300);
278
279    WRITE_CS_TABLE(fs->shader->cb_code, fs->shader->cb_code_size);
280}
281
282void r500_emit_fs_constants(struct r300_context* r300, unsigned size, void *state)
283{
284    struct r300_fragment_shader *fs = r300_fs(r300);
285    struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state;
286    unsigned count = fs->shader->externals_count;
287    CS_LOCALS(r300);
288
289    if (count == 0)
290        return;
291
292    BEGIN_CS(size);
293    OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_CONST);
294    OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, count * 4);
295    if (buf->remap_table){
296        for (unsigned i = 0; i < count; i++) {
297            uint32_t *data = &buf->ptr[buf->remap_table[i]*4];
298            OUT_CS_TABLE(data, 4);
299        }
300    } else {
301        OUT_CS_TABLE(buf->ptr, count * 4);
302    }
303    END_CS;
304}
305
306void r500_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, void *state)
307{
308    struct r300_fragment_shader *fs = r300_fs(r300);
309    struct rc_constant_list *constants = &fs->shader->code.constants;
310    unsigned i;
311    unsigned count = fs->shader->rc_state_count;
312    unsigned first = fs->shader->externals_count;
313    unsigned end = constants->Count;
314    CS_LOCALS(r300);
315
316    if (count == 0)
317        return;
318
319    BEGIN_CS(size);
320    for(i = first; i < end; ++i) {
321        if (constants->Constants[i].Type == RC_CONSTANT_STATE) {
322            float data[4];
323
324            get_rc_constant_state(data, r300, &constants->Constants[i]);
325
326            OUT_CS_REG(R500_GA_US_VECTOR_INDEX,
327                       R500_GA_US_VECTOR_INDEX_TYPE_CONST |
328                       (i & R500_GA_US_VECTOR_INDEX_MASK));
329            OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, 4);
330            OUT_CS_TABLE(data, 4);
331        }
332    }
333    END_CS;
334}
335
336void r300_emit_gpu_flush(struct r300_context *r300, unsigned size, void *state)
337{
338    struct r300_gpu_flush *gpuflush = (struct r300_gpu_flush*)state;
339    struct pipe_framebuffer_state* fb =
340            (struct pipe_framebuffer_state*)r300->fb_state.state;
341    uint32_t height = fb->height;
342    uint32_t width = fb->width;
343    CS_LOCALS(r300);
344
345    if (r300->cbzb_clear) {
346        struct r300_surface *surf = r300_surface(fb->cbufs[0]);
347
348        height = surf->cbzb_height;
349        width = surf->cbzb_width;
350    }
351
352    DBG(r300, DBG_SCISSOR,
353	"r300: Scissor width: %i, height: %i, CBZB clear: %s\n",
354	width, height, r300->cbzb_clear ? "YES" : "NO");
355
356    BEGIN_CS(size);
357
358    /* Set up scissors.
359     * By writing to the SC registers, SC & US assert idle. */
360    OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2);
361    if (r300->screen->caps.is_r500) {
362        OUT_CS(0);
363        OUT_CS(((width  - 1) << R300_SCISSORS_X_SHIFT) |
364               ((height - 1) << R300_SCISSORS_Y_SHIFT));
365    } else {
366        OUT_CS((1440 << R300_SCISSORS_X_SHIFT) |
367               (1440 << R300_SCISSORS_Y_SHIFT));
368        OUT_CS(((width  + 1440-1) << R300_SCISSORS_X_SHIFT) |
369               ((height + 1440-1) << R300_SCISSORS_Y_SHIFT));
370    }
371
372    /* Flush CB & ZB caches and wait until the 3D engine is idle and clean. */
373    OUT_CS_TABLE(gpuflush->cb_flush_clean, 6);
374    END_CS;
375}
376
377void r300_emit_aa_state(struct r300_context *r300, unsigned size, void *state)
378{
379    struct r300_aa_state *aa = (struct r300_aa_state*)state;
380    CS_LOCALS(r300);
381
382    BEGIN_CS(size);
383    OUT_CS_REG(R300_GB_AA_CONFIG, aa->aa_config);
384
385    if (aa->dest) {
386        OUT_CS_REG_SEQ(R300_RB3D_AARESOLVE_OFFSET, 3);
387        OUT_CS(aa->dest->offset);
388        OUT_CS(aa->dest->pitch & R300_RB3D_AARESOLVE_PITCH_MASK);
389        OUT_CS(R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE |
390               R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE);
391        OUT_CS_RELOC(aa->dest);
392    } else {
393        OUT_CS_REG(R300_RB3D_AARESOLVE_CTL, 0);
394    }
395
396    END_CS;
397}
398
399void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
400{
401    struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)state;
402    struct r300_surface* surf;
403    unsigned i;
404    uint32_t rb3d_cctl = 0;
405
406    CS_LOCALS(r300);
407
408    BEGIN_CS(size);
409
410    if (r300->screen->caps.is_r500) {
411        rb3d_cctl = R300_RB3D_CCTL_INDEPENDENT_COLORFORMAT_ENABLE_ENABLE;
412    }
413    /* NUM_MULTIWRITES replicates COLOR[0] to all colorbuffers. */
414    if (fb->nr_cbufs && r300->fb_multiwrite) {
415        rb3d_cctl |= R300_RB3D_CCTL_NUM_MULTIWRITES(fb->nr_cbufs);
416    }
417    if (r300->cmask_in_use) {
418        rb3d_cctl |= R300_RB3D_CCTL_AA_COMPRESSION_ENABLE |
419                     R300_RB3D_CCTL_CMASK_ENABLE;
420    }
421
422    OUT_CS_REG(R300_RB3D_CCTL, rb3d_cctl);
423
424    /* Set up colorbuffers. */
425    for (i = 0; i < fb->nr_cbufs; i++) {
426        surf = r300_surface(r300_get_nonnull_cb(fb, i));
427
428        OUT_CS_REG(R300_RB3D_COLOROFFSET0 + (4 * i), surf->offset);
429        OUT_CS_RELOC(surf);
430
431        OUT_CS_REG(R300_RB3D_COLORPITCH0 + (4 * i), surf->pitch);
432        OUT_CS_RELOC(surf);
433
434        if (r300->cmask_in_use && i == 0) {
435            OUT_CS_REG(R300_RB3D_CMASK_OFFSET0, 0);
436            OUT_CS_REG(R300_RB3D_CMASK_PITCH0, surf->pitch_cmask);
437            OUT_CS_REG(R300_RB3D_COLOR_CLEAR_VALUE, r300->color_clear_value);
438            if (r300->screen->caps.is_r500) {
439                OUT_CS_REG_SEQ(R500_RB3D_COLOR_CLEAR_VALUE_AR, 2);
440                OUT_CS(r300->color_clear_value_ar);
441                OUT_CS(r300->color_clear_value_gb);
442            }
443        }
444    }
445
446    /* Set up the ZB part of the CBZB clear. */
447    if (r300->cbzb_clear) {
448        surf = r300_surface(fb->cbufs[0]);
449
450        OUT_CS_REG(R300_ZB_FORMAT, surf->cbzb_format);
451
452        OUT_CS_REG(R300_ZB_DEPTHOFFSET, surf->cbzb_midpoint_offset);
453        OUT_CS_RELOC(surf);
454
455        OUT_CS_REG(R300_ZB_DEPTHPITCH, surf->cbzb_pitch);
456        OUT_CS_RELOC(surf);
457
458        DBG(r300, DBG_CBZB,
459            "CBZB clearing cbuf %08x %08x\n", surf->cbzb_format,
460            surf->cbzb_pitch);
461    }
462    /* Set up a zbuffer. */
463    else if (fb->zsbuf) {
464        surf = r300_surface(fb->zsbuf);
465
466        OUT_CS_REG(R300_ZB_FORMAT, surf->format);
467
468        OUT_CS_REG(R300_ZB_DEPTHOFFSET, surf->offset);
469        OUT_CS_RELOC(surf);
470
471        OUT_CS_REG(R300_ZB_DEPTHPITCH, surf->pitch);
472        OUT_CS_RELOC(surf);
473
474        if (r300->hyperz_enabled) {
475            /* HiZ RAM. */
476            OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0);
477            OUT_CS_REG(R300_ZB_HIZ_PITCH, surf->pitch_hiz);
478            /* Z Mask RAM. (compressed zbuffer) */
479            OUT_CS_REG(R300_ZB_ZMASK_OFFSET, 0);
480            OUT_CS_REG(R300_ZB_ZMASK_PITCH, surf->pitch_zmask);
481        }
482    }
483
484    END_CS;
485}
486
487void r300_emit_hyperz_state(struct r300_context *r300,
488                            unsigned size, void *state)
489{
490    struct r300_hyperz_state *z = state;
491    CS_LOCALS(r300);
492
493    if (z->flush)
494        WRITE_CS_TABLE(&z->cb_flush_begin, size);
495    else
496        WRITE_CS_TABLE(&z->cb_begin, size - 2);
497}
498
499void r300_emit_hyperz_end(struct r300_context *r300)
500{
501    struct r300_hyperz_state z =
502            *(struct r300_hyperz_state*)r300->hyperz_state.state;
503
504    z.flush = 1;
505    z.zb_bw_cntl = 0;
506    z.zb_depthclearvalue = 0;
507    z.sc_hyperz = R300_SC_HYPERZ_ADJ_2;
508    z.gb_z_peq_config = 0;
509
510    r300_emit_hyperz_state(r300, r300->hyperz_state.size, &z);
511}
512
513#define R300_NIBBLES(x0, y0, x1, y1, x2, y2, d0y, d0x)  \
514    (((x0) & 0xf) | (((y0) & 0xf) << 4) |		   \
515    (((x1) & 0xf) << 8) | (((y1) & 0xf) << 12) |	   \
516    (((x2) & 0xf) << 16) | (((y2) & 0xf) << 20) |	   \
517    (((d0y) & 0xf) << 24) | (((d0x) & 0xf) << 28))
518
519static unsigned r300_get_mspos(int index, unsigned *p)
520{
521    unsigned reg, i, distx, disty, dist;
522
523    if (index == 0) {
524        /* MSPOS0 contains positions for samples 0,1,2 as (X,Y) pairs of nibbles,
525         * followed by a (Y,X) pair containing the minimum distance from the pixel
526         * edge:
527         *     X0, Y0, X1, Y1, X2, Y2, D0_Y, D0_X
528         *
529         * There is a quirk when setting D0_X. The value represents the distance
530         * from the left edge of the pixel quad to the first sample in subpixels.
531         * All values less than eight should use the actual value, but „7‟ should
532         * be used for the distance „8‟. The hardware will convert 7 into 8 internally.
533         */
534        distx = 11;
535        for (i = 0; i < 12; i += 2) {
536            if (p[i] < distx)
537                distx = p[i];
538        }
539
540        disty = 11;
541        for (i = 1; i < 12; i += 2) {
542            if (p[i] < disty)
543                disty = p[i];
544        }
545
546        if (distx == 8)
547            distx = 7;
548
549        reg = R300_NIBBLES(p[0], p[1], p[2], p[3], p[4], p[5], disty, distx);
550    } else {
551        /* MSPOS1 contains positions for samples 3,4,5 as (X,Y) pairs of nibbles,
552         * followed by the minimum distance from the pixel edge (not sure if X or Y):
553         *     X3, Y3, X4, Y4, X5, Y5, D1
554         */
555        dist = 11;
556        for (i = 0; i < 12; i++) {
557            if (p[i] < dist)
558                dist = p[i];
559        }
560
561        reg = R300_NIBBLES(p[6], p[7], p[8], p[9], p[10], p[11], dist, 0);
562    }
563    return reg;
564}
565
566void r300_emit_fb_state_pipelined(struct r300_context *r300,
567                                  unsigned size, void *state)
568{
569    /* The sample coordinates are in the range [0,11], because
570     * GB_TILE_CONFIG.SUBPIXEL is set to the 1/12 subpixel precision.
571     *
572     * Some sample coordinates reach to neighboring pixels and should not be used.
573     * (e.g. Y=11)
574     *
575     * The unused samples must be set to the positions of other valid samples. */
576    static unsigned sample_locs_1x[12] = {
577        6,6,  6,6,  6,6,  6,6,  6,6,  6,6
578    };
579    static unsigned sample_locs_2x[12] = {
580        3,9,  9,3,  9,3,  9,3,  9,3,  9,3
581    };
582    static unsigned sample_locs_4x[12] = {
583        4,4,  8,8,  2,10,  10,2,  10,2,  10,2
584    };
585    static unsigned sample_locs_6x[12] = {
586        3,1,  7,3,  11,5,  1,7,  5,9,  9,10
587    };
588
589    struct pipe_framebuffer_state* fb =
590            (struct pipe_framebuffer_state*)r300->fb_state.state;
591    unsigned i, num_cbufs = fb->nr_cbufs;
592    unsigned mspos0, mspos1;
593    CS_LOCALS(r300);
594
595    /* If we use the multiwrite feature, the colorbuffers 2,3,4 must be
596     * marked as UNUSED in the US block. */
597    if (r300->fb_multiwrite) {
598        num_cbufs = MIN2(num_cbufs, 1);
599    }
600
601    BEGIN_CS(size);
602
603    /* Colorbuffer format in the US block.
604     * (must be written after unpipelined regs) */
605    OUT_CS_REG_SEQ(R300_US_OUT_FMT_0, 4);
606    for (i = 0; i < num_cbufs; i++) {
607        OUT_CS(r300_surface(r300_get_nonnull_cb(fb, i))->format);
608    }
609    for (; i < 1; i++) {
610        OUT_CS(R300_US_OUT_FMT_C4_8 |
611               R300_C0_SEL_B | R300_C1_SEL_G |
612               R300_C2_SEL_R | R300_C3_SEL_A);
613    }
614    for (; i < 4; i++) {
615        OUT_CS(R300_US_OUT_FMT_UNUSED);
616    }
617
618    /* Set sample positions. It depends on the framebuffer sample count.
619     * These are pipelined regs and as such cannot be moved to the AA state.
620     */
621    switch (r300->num_samples) {
622    default:
623        mspos0 = r300_get_mspos(0, sample_locs_1x);
624        mspos1 = r300_get_mspos(1, sample_locs_1x);
625        break;
626    case 2:
627        mspos0 = r300_get_mspos(0, sample_locs_2x);
628        mspos1 = r300_get_mspos(1, sample_locs_2x);
629        break;
630    case 4:
631        mspos0 = r300_get_mspos(0, sample_locs_4x);
632        mspos1 = r300_get_mspos(1, sample_locs_4x);
633        break;
634    case 6:
635        mspos0 = r300_get_mspos(0, sample_locs_6x);
636        mspos1 = r300_get_mspos(1, sample_locs_6x);
637        break;
638    }
639
640    OUT_CS_REG_SEQ(R300_GB_MSPOS0, 2);
641    OUT_CS(mspos0);
642    OUT_CS(mspos1);
643    END_CS;
644}
645
646void r300_emit_query_start(struct r300_context *r300, unsigned size, void*state)
647{
648    struct r300_query *query = r300->query_current;
649    CS_LOCALS(r300);
650
651    if (!query)
652	return;
653
654    BEGIN_CS(size);
655    if (r300->screen->caps.family == CHIP_RV530) {
656        OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
657    } else {
658        OUT_CS_REG(R300_SU_REG_DEST, R300_RASTER_PIPE_SELECT_ALL);
659    }
660    OUT_CS_REG(R300_ZB_ZPASS_DATA, 0);
661    END_CS;
662    query->begin_emitted = TRUE;
663}
664
665static void r300_emit_query_end_frag_pipes(struct r300_context *r300,
666                                           struct r300_query *query)
667{
668    struct r300_capabilities* caps = &r300->screen->caps;
669    uint32_t gb_pipes = r300->screen->info.r300_num_gb_pipes;
670    CS_LOCALS(r300);
671
672    assert(gb_pipes);
673
674    BEGIN_CS(6 * gb_pipes + 2);
675    /* I'm not so sure I like this switch, but it's hard to be elegant
676     * when there's so many special cases...
677     *
678     * So here's the basic idea. For each pipe, enable writes to it only,
679     * then put out the relocation for ZPASS_ADDR, taking into account a
680     * 4-byte offset for each pipe. RV380 and older are special; they have
681     * only two pipes, and the second pipe's enable is on bit 3, not bit 1,
682     * so there's a chipset cap for that. */
683    switch (gb_pipes) {
684        case 4:
685            /* pipe 3 only */
686            OUT_CS_REG(R300_SU_REG_DEST, 1 << 3);
687            OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 3) * 4);
688            OUT_CS_RELOC(r300->query_current);
689            FALLTHROUGH;
690        case 3:
691            /* pipe 2 only */
692            OUT_CS_REG(R300_SU_REG_DEST, 1 << 2);
693            OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 2) * 4);
694            OUT_CS_RELOC(r300->query_current);
695            FALLTHROUGH;
696        case 2:
697            /* pipe 1 only */
698            /* As mentioned above, accommodate RV380 and older. */
699            OUT_CS_REG(R300_SU_REG_DEST,
700                    1 << (caps->high_second_pipe ? 3 : 1));
701            OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 1) * 4);
702            OUT_CS_RELOC(r300->query_current);
703            FALLTHROUGH;
704        case 1:
705            /* pipe 0 only */
706            OUT_CS_REG(R300_SU_REG_DEST, 1 << 0);
707            OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 0) * 4);
708            OUT_CS_RELOC(r300->query_current);
709            break;
710        default:
711            fprintf(stderr, "r300: Implementation error: Chipset reports %d"
712                    " pixel pipes!\n", gb_pipes);
713            abort();
714    }
715
716    /* And, finally, reset it to normal... */
717    OUT_CS_REG(R300_SU_REG_DEST, 0xF);
718    END_CS;
719}
720
721static void rv530_emit_query_end_single_z(struct r300_context *r300,
722                                          struct r300_query *query)
723{
724    CS_LOCALS(r300);
725
726    BEGIN_CS(8);
727    OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0);
728    OUT_CS_REG(R300_ZB_ZPASS_ADDR, query->num_results * 4);
729    OUT_CS_RELOC(r300->query_current);
730    OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
731    END_CS;
732}
733
734static void rv530_emit_query_end_double_z(struct r300_context *r300,
735                                          struct r300_query *query)
736{
737    CS_LOCALS(r300);
738
739    BEGIN_CS(14);
740    OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0);
741    OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 0) * 4);
742    OUT_CS_RELOC(r300->query_current);
743    OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_1);
744    OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 1) * 4);
745    OUT_CS_RELOC(r300->query_current);
746    OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
747    END_CS;
748}
749
750void r300_emit_query_end(struct r300_context* r300)
751{
752    struct r300_capabilities *caps = &r300->screen->caps;
753    struct r300_query *query = r300->query_current;
754
755    if (!query)
756	return;
757
758    if (query->begin_emitted == FALSE)
759        return;
760
761    if (caps->family == CHIP_RV530) {
762        if (r300->screen->info.r300_num_z_pipes == 2)
763            rv530_emit_query_end_double_z(r300, query);
764        else
765            rv530_emit_query_end_single_z(r300, query);
766    } else
767        r300_emit_query_end_frag_pipes(r300, query);
768
769    query->begin_emitted = FALSE;
770    query->num_results += query->num_pipes;
771
772    /* XXX grab all the results and reset the counter. */
773    if (query->num_results >= query->buf->size / 4 - 4) {
774        query->num_results = (query->buf->size / 4) / 2;
775        fprintf(stderr, "r300: Rewinding OQBO...\n");
776    }
777}
778
779void r300_emit_invariant_state(struct r300_context *r300,
780                               unsigned size, void *state)
781{
782    CS_LOCALS(r300);
783    WRITE_CS_TABLE(state, size);
784}
785
786void r300_emit_rs_state(struct r300_context* r300, unsigned size, void* state)
787{
788    struct r300_rs_state* rs = state;
789    CS_LOCALS(r300);
790
791    BEGIN_CS(size);
792    OUT_CS_TABLE(rs->cb_main, RS_STATE_MAIN_SIZE);
793    if (rs->polygon_offset_enable) {
794        if (r300->zbuffer_bpp == 16) {
795            OUT_CS_TABLE(rs->cb_poly_offset_zb16, 5);
796        } else {
797            OUT_CS_TABLE(rs->cb_poly_offset_zb24, 5);
798        }
799    }
800    END_CS;
801}
802
803void r300_emit_rs_block_state(struct r300_context* r300,
804                              unsigned size, void* state)
805{
806    struct r300_rs_block* rs = (struct r300_rs_block*)state;
807    unsigned i;
808    /* It's the same for both INST and IP tables */
809    unsigned count = (rs->inst_count & R300_RS_INST_COUNT_MASK) + 1;
810    CS_LOCALS(r300);
811
812    if (DBG_ON(r300, DBG_RS_BLOCK)) {
813        r500_dump_rs_block(rs);
814
815        fprintf(stderr, "r300: RS emit:\n");
816
817        for (i = 0; i < count; i++)
818            fprintf(stderr, "    : ip %d: 0x%08x\n", i, rs->ip[i]);
819
820        for (i = 0; i < count; i++)
821            fprintf(stderr, "    : inst %d: 0x%08x\n", i, rs->inst[i]);
822
823        fprintf(stderr, "    : count: 0x%08x inst_count: 0x%08x\n",
824            rs->count, rs->inst_count);
825    }
826
827    BEGIN_CS(size);
828    OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL, 2);
829    OUT_CS(rs->vap_vtx_state_cntl);
830    OUT_CS(rs->vap_vsm_vtx_assm);
831    OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2);
832    OUT_CS(rs->vap_out_vtx_fmt[0]);
833    OUT_CS(rs->vap_out_vtx_fmt[1]);
834    OUT_CS_REG_SEQ(R300_GB_ENABLE, 1);
835    OUT_CS(rs->gb_enable);
836
837    if (r300->screen->caps.is_r500) {
838        OUT_CS_REG_SEQ(R500_RS_IP_0, count);
839    } else {
840        OUT_CS_REG_SEQ(R300_RS_IP_0, count);
841    }
842    OUT_CS_TABLE(rs->ip, count);
843
844    OUT_CS_REG_SEQ(R300_RS_COUNT, 2);
845    OUT_CS(rs->count);
846    OUT_CS(rs->inst_count);
847
848    if (r300->screen->caps.is_r500) {
849        OUT_CS_REG_SEQ(R500_RS_INST_0, count);
850    } else {
851        OUT_CS_REG_SEQ(R300_RS_INST_0, count);
852    }
853    OUT_CS_TABLE(rs->inst, count);
854    END_CS;
855}
856
857void r300_emit_sample_mask(struct r300_context *r300,
858                           unsigned size, void *state)
859{
860    unsigned mask = (*(unsigned*)state) & ((1 << 6)-1);
861    CS_LOCALS(r300);
862
863    BEGIN_CS(size);
864    OUT_CS_REG(R300_SC_SCREENDOOR,
865               mask | (mask << 6) | (mask << 12) | (mask << 18));
866    END_CS;
867}
868
869void r300_emit_scissor_state(struct r300_context* r300,
870                             unsigned size, void* state)
871{
872    struct pipe_scissor_state* scissor = (struct pipe_scissor_state*)state;
873    CS_LOCALS(r300);
874
875    BEGIN_CS(size);
876    OUT_CS_REG_SEQ(R300_SC_CLIPRECT_TL_0, 2);
877    if (r300->screen->caps.is_r500) {
878        OUT_CS((scissor->minx << R300_CLIPRECT_X_SHIFT) |
879               (scissor->miny << R300_CLIPRECT_Y_SHIFT));
880        OUT_CS(((scissor->maxx - 1) << R300_CLIPRECT_X_SHIFT) |
881               ((scissor->maxy - 1) << R300_CLIPRECT_Y_SHIFT));
882    } else {
883        OUT_CS(((scissor->minx + 1440) << R300_CLIPRECT_X_SHIFT) |
884               ((scissor->miny + 1440) << R300_CLIPRECT_Y_SHIFT));
885        OUT_CS(((scissor->maxx + 1440-1) << R300_CLIPRECT_X_SHIFT) |
886               ((scissor->maxy + 1440-1) << R300_CLIPRECT_Y_SHIFT));
887    }
888    END_CS;
889}
890
891void r300_emit_textures_state(struct r300_context *r300,
892                              unsigned size, void *state)
893{
894    struct r300_textures_state *allstate = (struct r300_textures_state*)state;
895    struct r300_texture_sampler_state *texstate;
896    struct r300_resource *tex;
897    unsigned i;
898    boolean has_us_format = r300->screen->caps.has_us_format;
899    CS_LOCALS(r300);
900
901    BEGIN_CS(size);
902    OUT_CS_REG(R300_TX_ENABLE, allstate->tx_enable);
903
904    for (i = 0; i < allstate->count; i++) {
905        if ((1 << i) & allstate->tx_enable) {
906            texstate = &allstate->regs[i];
907            tex = r300_resource(allstate->sampler_views[i]->base.texture);
908
909            OUT_CS_REG(R300_TX_FILTER0_0 + (i * 4), texstate->filter0);
910            OUT_CS_REG(R300_TX_FILTER1_0 + (i * 4), texstate->filter1);
911            OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (i * 4),
912                       texstate->border_color);
913
914            OUT_CS_REG(R300_TX_FORMAT0_0 + (i * 4), texstate->format.format0);
915            OUT_CS_REG(R300_TX_FORMAT1_0 + (i * 4), texstate->format.format1);
916            OUT_CS_REG(R300_TX_FORMAT2_0 + (i * 4), texstate->format.format2);
917
918            OUT_CS_REG(R300_TX_OFFSET_0 + (i * 4), texstate->format.tile_config);
919            OUT_CS_RELOC(tex);
920
921            if (has_us_format) {
922                OUT_CS_REG(R500_US_FORMAT0_0 + (i * 4),
923                           texstate->format.us_format0);
924            }
925        }
926    }
927    END_CS;
928}
929
930void r300_emit_vertex_arrays(struct r300_context* r300, int offset,
931                             boolean indexed, int instance_id)
932{
933    struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
934    struct pipe_vertex_element *velem = r300->velems->velem;
935    struct r300_resource *buf;
936    int i;
937    unsigned vertex_array_count = r300->velems->count;
938    unsigned packet_size = (vertex_array_count * 3 + 1) / 2;
939    struct pipe_vertex_buffer *vb1, *vb2;
940    unsigned *hw_format_size = r300->velems->format_size;
941    unsigned size1, size2, offset1, offset2, stride1, stride2;
942    CS_LOCALS(r300);
943
944    BEGIN_CS(2 + packet_size + vertex_array_count * 2);
945    OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size);
946    OUT_CS(vertex_array_count | (!indexed ? R300_VC_FORCE_PREFETCH : 0));
947
948    if (instance_id == -1) {
949        /* Non-instanced arrays. This ignores instance_divisor and instance_id. */
950        for (i = 0; i < vertex_array_count - 1; i += 2) {
951            vb1 = &vbuf[velem[i].vertex_buffer_index];
952            vb2 = &vbuf[velem[i+1].vertex_buffer_index];
953            size1 = hw_format_size[i];
954            size2 = hw_format_size[i+1];
955
956            OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride) |
957                   R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(vb2->stride));
958            OUT_CS(vb1->buffer_offset + velem[i].src_offset   + offset * vb1->stride);
959            OUT_CS(vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride);
960        }
961
962        if (vertex_array_count & 1) {
963            vb1 = &vbuf[velem[i].vertex_buffer_index];
964            size1 = hw_format_size[i];
965
966            OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride));
967            OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
968        }
969
970        for (i = 0; i < vertex_array_count; i++) {
971            buf = r300_resource(vbuf[velem[i].vertex_buffer_index].buffer.resource);
972            OUT_CS_RELOC(buf);
973        }
974    } else {
975        /* Instanced arrays. */
976        for (i = 0; i < vertex_array_count - 1; i += 2) {
977            vb1 = &vbuf[velem[i].vertex_buffer_index];
978            vb2 = &vbuf[velem[i+1].vertex_buffer_index];
979            size1 = hw_format_size[i];
980            size2 = hw_format_size[i+1];
981
982            if (velem[i].instance_divisor) {
983                stride1 = 0;
984                offset1 = vb1->buffer_offset + velem[i].src_offset +
985                          (instance_id / velem[i].instance_divisor) * vb1->stride;
986            } else {
987                stride1 = vb1->stride;
988                offset1 = vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride;
989            }
990            if (velem[i+1].instance_divisor) {
991                stride2 = 0;
992                offset2 = vb2->buffer_offset + velem[i+1].src_offset +
993                          (instance_id / velem[i+1].instance_divisor) * vb2->stride;
994            } else {
995                stride2 = vb2->stride;
996                offset2 = vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride;
997            }
998
999            OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(stride1) |
1000                   R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(stride2));
1001            OUT_CS(offset1);
1002            OUT_CS(offset2);
1003        }
1004
1005        if (vertex_array_count & 1) {
1006            vb1 = &vbuf[velem[i].vertex_buffer_index];
1007            size1 = hw_format_size[i];
1008
1009            if (velem[i].instance_divisor) {
1010                stride1 = 0;
1011                offset1 = vb1->buffer_offset + velem[i].src_offset +
1012                          (instance_id / velem[i].instance_divisor) * vb1->stride;
1013            } else {
1014                stride1 = vb1->stride;
1015                offset1 = vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride;
1016            }
1017
1018            OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(stride1));
1019            OUT_CS(offset1);
1020        }
1021
1022        for (i = 0; i < vertex_array_count; i++) {
1023            buf = r300_resource(vbuf[velem[i].vertex_buffer_index].buffer.resource);
1024            OUT_CS_RELOC(buf);
1025        }
1026    }
1027    END_CS;
1028}
1029
1030void r300_emit_vertex_arrays_swtcl(struct r300_context *r300, boolean indexed)
1031{
1032    CS_LOCALS(r300);
1033
1034    DBG(r300, DBG_SWTCL, "r300: Preparing vertex buffer %p for render, "
1035            "vertex size %d\n", r300->vbo,
1036            r300->vertex_info.size);
1037    /* Set the pointer to our vertex buffer. The emitted values are this:
1038     * PACKET3 [3D_LOAD_VBPNTR]
1039     * COUNT   [1]
1040     * FORMAT  [size | stride << 8]
1041     * OFFSET  [offset into BO]
1042     * VBPNTR  [relocated BO]
1043     */
1044    BEGIN_CS(7);
1045    OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, 3);
1046    OUT_CS(1 | (!indexed ? R300_VC_FORCE_PREFETCH : 0));
1047    OUT_CS(r300->vertex_info.size |
1048            (r300->vertex_info.size << 8));
1049    OUT_CS(r300->draw_vbo_offset);
1050    OUT_CS(0);
1051
1052    assert(r300->vbo);
1053    OUT_CS(0xc0001000); /* PKT3_NOP */
1054    OUT_CS(r300->rws->cs_lookup_buffer(&r300->cs, r300->vbo) * 4);
1055    END_CS;
1056}
1057
1058void r300_emit_vertex_stream_state(struct r300_context* r300,
1059                                   unsigned size, void* state)
1060{
1061    struct r300_vertex_stream_state *streams =
1062        (struct r300_vertex_stream_state*)state;
1063    unsigned i;
1064    CS_LOCALS(r300);
1065
1066    if (DBG_ON(r300, DBG_PSC)) {
1067        fprintf(stderr, "r300: PSC emit:\n");
1068
1069        for (i = 0; i < streams->count; i++) {
1070            fprintf(stderr, "    : prog_stream_cntl%d: 0x%08x\n", i,
1071                   streams->vap_prog_stream_cntl[i]);
1072        }
1073
1074        for (i = 0; i < streams->count; i++) {
1075            fprintf(stderr, "    : prog_stream_cntl_ext%d: 0x%08x\n", i,
1076                   streams->vap_prog_stream_cntl_ext[i]);
1077        }
1078    }
1079
1080    BEGIN_CS(size);
1081    OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0, streams->count);
1082    OUT_CS_TABLE(streams->vap_prog_stream_cntl, streams->count);
1083    OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0, streams->count);
1084    OUT_CS_TABLE(streams->vap_prog_stream_cntl_ext, streams->count);
1085    END_CS;
1086}
1087
1088void r300_emit_pvs_flush(struct r300_context* r300, unsigned size, void* state)
1089{
1090    CS_LOCALS(r300);
1091
1092    BEGIN_CS(size);
1093    OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0);
1094    END_CS;
1095}
1096
1097void r300_emit_vap_invariant_state(struct r300_context *r300,
1098                                   unsigned size, void *state)
1099{
1100    CS_LOCALS(r300);
1101    WRITE_CS_TABLE(state, size);
1102}
1103
1104void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state)
1105{
1106    struct r300_vertex_shader_code* vs = ((struct r300_vertex_shader*)state)->shader;
1107    struct r300_vertex_program_code* code = &vs->code;
1108    struct r300_screen* r300screen = r300->screen;
1109    unsigned instruction_count = code->length / 4;
1110
1111    unsigned vtx_mem_size = r300screen->caps.is_r500 ? 128 : 72;
1112    unsigned input_count = MAX2(util_bitcount(code->InputsRead), 1);
1113    unsigned output_count = MAX2(util_bitcount(code->OutputsWritten), 1);
1114    unsigned temp_count = MAX2(code->num_temporaries, 1);
1115
1116    unsigned pvs_num_slots = MIN3(vtx_mem_size / input_count,
1117                                  vtx_mem_size / output_count, 10);
1118    unsigned pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 5);
1119
1120    CS_LOCALS(r300);
1121
1122    BEGIN_CS(size);
1123
1124    /* R300_VAP_PVS_CODE_CNTL_0
1125     * R300_VAP_PVS_CONST_CNTL
1126     * R300_VAP_PVS_CODE_CNTL_1
1127     * See the r5xx docs for instructions on how to use these. */
1128    OUT_CS_REG(R300_VAP_PVS_CODE_CNTL_0, R300_PVS_FIRST_INST(0) |
1129	       R300_PVS_XYZW_VALID_INST(code->last_pos_write) |
1130	       R300_PVS_LAST_INST(instruction_count - 1));
1131    OUT_CS_REG(R300_VAP_PVS_CODE_CNTL_1, code->last_input_read);
1132
1133    OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, 0);
1134    OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, code->length);
1135    OUT_CS_TABLE(code->body.d, code->length);
1136
1137    OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(pvs_num_slots) |
1138            R300_PVS_NUM_CNTLRS(pvs_num_controllers) |
1139            R300_PVS_NUM_FPUS(r300screen->caps.num_vert_fpus) |
1140            R300_PVS_VF_MAX_VTX_NUM(12) |
1141            (r300->clip_halfz ? R300_DX_CLIP_SPACE_DEF : 0) |
1142            (r300screen->caps.is_r500 ? R500_TCL_STATE_OPTIMIZATION : 0));
1143
1144    /* Emit flow control instructions.  Even if there are no fc instructions,
1145     * we still need to write the registers to make sure they are cleared. */
1146    OUT_CS_REG(R300_VAP_PVS_FLOW_CNTL_OPC, code->fc_ops);
1147    if (r300screen->caps.is_r500) {
1148        OUT_CS_REG_SEQ(R500_VAP_PVS_FLOW_CNTL_ADDRS_LW_0, R300_VS_MAX_FC_OPS * 2);
1149        OUT_CS_TABLE(code->fc_op_addrs.r500, R300_VS_MAX_FC_OPS * 2);
1150    } else {
1151        OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_ADDRS_0, R300_VS_MAX_FC_OPS);
1152        OUT_CS_TABLE(code->fc_op_addrs.r300, R300_VS_MAX_FC_OPS);
1153    }
1154    OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_LOOP_INDEX_0, R300_VS_MAX_FC_OPS);
1155    OUT_CS_TABLE(code->fc_loop_index, R300_VS_MAX_FC_OPS);
1156
1157    END_CS;
1158}
1159
1160void r300_emit_vs_constants(struct r300_context* r300,
1161                            unsigned size, void *state)
1162{
1163    unsigned count = r300_vs(r300)->shader->externals_count;
1164    struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state;
1165    struct r300_vertex_shader_code *vs = r300_vs(r300)->shader;
1166    unsigned i;
1167    int imm_first = vs->externals_count;
1168    int imm_end = vs->code.constants.Count;
1169    int imm_count = vs->immediates_count;
1170    CS_LOCALS(r300);
1171
1172    BEGIN_CS(size);
1173    OUT_CS_REG(R300_VAP_PVS_CONST_CNTL,
1174               R300_PVS_CONST_BASE_OFFSET(buf->buffer_base) |
1175               R300_PVS_MAX_CONST_ADDR(MAX2(imm_end - 1, 0)));
1176    if (vs->externals_count) {
1177        OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
1178                   (r300->screen->caps.is_r500 ?
1179                   R500_PVS_CONST_START : R300_PVS_CONST_START) + buf->buffer_base);
1180        OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, count * 4);
1181        if (buf->remap_table){
1182            for (i = 0; i < count; i++) {
1183                uint32_t *data = &buf->ptr[buf->remap_table[i]*4];
1184                OUT_CS_TABLE(data, 4);
1185            }
1186        } else {
1187            OUT_CS_TABLE(buf->ptr, count * 4);
1188        }
1189    }
1190
1191    /* Emit immediates. */
1192    if (imm_count) {
1193        OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
1194                   (r300->screen->caps.is_r500 ?
1195                   R500_PVS_CONST_START : R300_PVS_CONST_START) +
1196                   buf->buffer_base + imm_first);
1197        OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, imm_count * 4);
1198        for (i = imm_first; i < imm_end; i++) {
1199            const float *data = vs->code.constants.Constants[i].u.Immediate;
1200            OUT_CS_TABLE(data, 4);
1201        }
1202    }
1203    END_CS;
1204}
1205
1206void r300_emit_viewport_state(struct r300_context* r300,
1207                              unsigned size, void* state)
1208{
1209    struct r300_viewport_state* viewport = (struct r300_viewport_state*)state;
1210    CS_LOCALS(r300);
1211
1212    BEGIN_CS(size);
1213    OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6);
1214    OUT_CS_TABLE(&viewport->xscale, 6);
1215    OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control);
1216    END_CS;
1217}
1218
1219void r300_emit_hiz_clear(struct r300_context *r300, unsigned size, void *state)
1220{
1221    struct pipe_framebuffer_state *fb =
1222        (struct pipe_framebuffer_state*)r300->fb_state.state;
1223    struct r300_resource* tex;
1224    CS_LOCALS(r300);
1225
1226    tex = r300_resource(fb->zsbuf->texture);
1227
1228    BEGIN_CS(size);
1229    OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_HIZ, 2);
1230    OUT_CS(0);
1231    OUT_CS(tex->tex.hiz_dwords[fb->zsbuf->u.tex.level]);
1232    OUT_CS(r300->hiz_clear_value);
1233    END_CS;
1234
1235    /* Mark the current zbuffer's hiz ram as in use. */
1236    r300->hiz_in_use = TRUE;
1237    r300->hiz_func = HIZ_FUNC_NONE;
1238    r300_mark_atom_dirty(r300, &r300->hyperz_state);
1239}
1240
1241void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state)
1242{
1243    struct pipe_framebuffer_state *fb =
1244        (struct pipe_framebuffer_state*)r300->fb_state.state;
1245    struct r300_resource *tex;
1246    CS_LOCALS(r300);
1247
1248    tex = r300_resource(fb->zsbuf->texture);
1249
1250    BEGIN_CS(size);
1251    OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_ZMASK, 2);
1252    OUT_CS(0);
1253    OUT_CS(tex->tex.zmask_dwords[fb->zsbuf->u.tex.level]);
1254    OUT_CS(0);
1255    END_CS;
1256
1257    /* Mark the current zbuffer's zmask as in use. */
1258    r300->zmask_in_use = TRUE;
1259    r300_mark_atom_dirty(r300, &r300->hyperz_state);
1260}
1261
1262void r300_emit_cmask_clear(struct r300_context *r300, unsigned size, void *state)
1263{
1264    struct pipe_framebuffer_state *fb =
1265        (struct pipe_framebuffer_state*)r300->fb_state.state;
1266    struct r300_resource *tex;
1267    CS_LOCALS(r300);
1268
1269    tex = r300_resource(fb->cbufs[0]->texture);
1270
1271    BEGIN_CS(size);
1272    OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_CMASK, 2);
1273    OUT_CS(0);
1274    OUT_CS(tex->tex.cmask_dwords);
1275    OUT_CS(0);
1276    END_CS;
1277
1278    /* Mark the current zbuffer's zmask as in use. */
1279    r300->cmask_in_use = TRUE;
1280    r300_mark_fb_state_dirty(r300, R300_CHANGED_CMASK_ENABLE);
1281}
1282
1283void r300_emit_ztop_state(struct r300_context* r300,
1284                          unsigned size, void* state)
1285{
1286    struct r300_ztop_state* ztop = (struct r300_ztop_state*)state;
1287    CS_LOCALS(r300);
1288
1289    BEGIN_CS(size);
1290    OUT_CS_REG(R300_ZB_ZTOP, ztop->z_buffer_top);
1291    END_CS;
1292}
1293
1294void r300_emit_texture_cache_inval(struct r300_context* r300, unsigned size, void* state)
1295{
1296    CS_LOCALS(r300);
1297
1298    BEGIN_CS(size);
1299    OUT_CS_REG(R300_TX_INVALTAGS, 0);
1300    END_CS;
1301}
1302
1303boolean r300_emit_buffer_validate(struct r300_context *r300,
1304                                  boolean do_validate_vertex_buffers,
1305                                  struct pipe_resource *index_buffer)
1306{
1307    struct pipe_framebuffer_state *fb =
1308        (struct pipe_framebuffer_state*)r300->fb_state.state;
1309    struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
1310    struct r300_textures_state *texstate =
1311        (struct r300_textures_state*)r300->textures_state.state;
1312    struct r300_resource *tex;
1313    unsigned i;
1314    boolean flushed = FALSE;
1315
1316validate:
1317    if (r300->fb_state.dirty) {
1318        /* Color buffers... */
1319        for (i = 0; i < fb->nr_cbufs; i++) {
1320            if (!fb->cbufs[i])
1321                continue;
1322            tex = r300_resource(fb->cbufs[i]->texture);
1323            assert(tex && tex->buf && "cbuf is marked, but NULL!");
1324            r300->rws->cs_add_buffer(&r300->cs, tex->buf,
1325                                    RADEON_USAGE_READWRITE | RADEON_USAGE_SYNCHRONIZED |
1326                                    (tex->b.nr_samples > 1 ?
1327                                        RADEON_PRIO_COLOR_BUFFER_MSAA :
1328                                        RADEON_PRIO_COLOR_BUFFER),
1329                                    r300_surface(fb->cbufs[i])->domain);
1330        }
1331        /* ...depth buffer... */
1332        if (fb->zsbuf) {
1333            tex = r300_resource(fb->zsbuf->texture);
1334            assert(tex && tex->buf && "zsbuf is marked, but NULL!");
1335            r300->rws->cs_add_buffer(&r300->cs, tex->buf,
1336                                    RADEON_USAGE_READWRITE | RADEON_USAGE_SYNCHRONIZED |
1337                                    (tex->b.nr_samples > 1 ?
1338                                        RADEON_PRIO_DEPTH_BUFFER_MSAA :
1339                                        RADEON_PRIO_DEPTH_BUFFER),
1340                                    r300_surface(fb->zsbuf)->domain);
1341        }
1342    }
1343    /* The AA resolve buffer. */
1344    if (r300->aa_state.dirty) {
1345        if (aa->dest) {
1346            r300->rws->cs_add_buffer(&r300->cs, aa->dest->buf,
1347                                    RADEON_USAGE_WRITE | RADEON_USAGE_SYNCHRONIZED |
1348                                    RADEON_PRIO_COLOR_BUFFER,
1349                                    aa->dest->domain);
1350        }
1351    }
1352    if (r300->textures_state.dirty) {
1353        /* ...textures... */
1354        for (i = 0; i < texstate->count; i++) {
1355            if (!(texstate->tx_enable & (1U << i))) {
1356                continue;
1357            }
1358
1359            tex = r300_resource(texstate->sampler_views[i]->base.texture);
1360            r300->rws->cs_add_buffer(&r300->cs, tex->buf,
1361                                     RADEON_USAGE_READ | RADEON_USAGE_SYNCHRONIZED |
1362                                     RADEON_PRIO_SAMPLER_TEXTURE,
1363                                    tex->domain);
1364        }
1365    }
1366    /* ...occlusion query buffer... */
1367    if (r300->query_current)
1368        r300->rws->cs_add_buffer(&r300->cs, r300->query_current->buf,
1369                                 RADEON_USAGE_WRITE | RADEON_USAGE_SYNCHRONIZED |
1370                                 RADEON_PRIO_QUERY,
1371                                 RADEON_DOMAIN_GTT);
1372    /* ...vertex buffer for SWTCL path... */
1373    if (r300->vbo)
1374        r300->rws->cs_add_buffer(&r300->cs, r300->vbo,
1375                                 RADEON_USAGE_READ | RADEON_USAGE_SYNCHRONIZED |
1376                                 RADEON_PRIO_VERTEX_BUFFER,
1377                                 RADEON_DOMAIN_GTT);
1378    /* ...vertex buffers for HWTCL path... */
1379    if (do_validate_vertex_buffers && r300->vertex_arrays_dirty) {
1380        struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
1381        struct pipe_vertex_buffer *last = r300->vertex_buffer +
1382                                      r300->nr_vertex_buffers;
1383        struct pipe_resource *buf;
1384
1385        for (; vbuf != last; vbuf++) {
1386            buf = vbuf->buffer.resource;
1387            if (!buf)
1388                continue;
1389
1390            r300->rws->cs_add_buffer(&r300->cs, r300_resource(buf)->buf,
1391                                    RADEON_USAGE_READ | RADEON_USAGE_SYNCHRONIZED |
1392                                    RADEON_PRIO_SAMPLER_BUFFER,
1393                                    r300_resource(buf)->domain);
1394        }
1395    }
1396    /* ...and index buffer for HWTCL path. */
1397    if (index_buffer)
1398        r300->rws->cs_add_buffer(&r300->cs, r300_resource(index_buffer)->buf,
1399                                RADEON_USAGE_READ | RADEON_USAGE_SYNCHRONIZED |
1400                                RADEON_PRIO_INDEX_BUFFER,
1401                                r300_resource(index_buffer)->domain);
1402
1403    /* Now do the validation (flush is called inside cs_validate on failure). */
1404    if (!r300->rws->cs_validate(&r300->cs)) {
1405        /* Ooops, an infinite loop, give up. */
1406        if (flushed)
1407            return FALSE;
1408
1409        flushed = TRUE;
1410        goto validate;
1411    }
1412
1413    return TRUE;
1414}
1415
1416unsigned r300_get_num_dirty_dwords(struct r300_context *r300)
1417{
1418    struct r300_atom* atom;
1419    unsigned dwords = 0;
1420
1421    foreach_dirty_atom(r300, atom) {
1422        if (atom->dirty) {
1423            dwords += atom->size;
1424        }
1425    }
1426
1427    /* let's reserve some more, just in case */
1428    dwords += 32;
1429
1430    return dwords;
1431}
1432
1433unsigned r300_get_num_cs_end_dwords(struct r300_context *r300)
1434{
1435    unsigned dwords = 0;
1436
1437    /* Emitted in flush. */
1438    dwords += 26; /* emit_query_end */
1439    dwords += r300->hyperz_state.size + 2; /* emit_hyperz_end + zcache flush */
1440    if (r300->screen->caps.is_r500)
1441        dwords += 2; /* emit_index_bias */
1442    dwords += 3; /* MSPOS */
1443
1444    return dwords;
1445}
1446
1447/* Emit all dirty state. */
1448void r300_emit_dirty_state(struct r300_context* r300)
1449{
1450    struct r300_atom *atom;
1451
1452    foreach_dirty_atom(r300, atom) {
1453        if (atom->dirty) {
1454            atom->emit(r300, atom->size, atom->state);
1455            atom->dirty = FALSE;
1456        }
1457    }
1458
1459    r300->first_dirty = NULL;
1460    r300->last_dirty = NULL;
1461    r300->dirty_hw++;
1462}
1463