r300_state.c revision fa3f1348e49feeac511dbe5b22bbddc47f56ba81
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#include "draw/draw_context.h"
25
26#include "util/u_framebuffer.h"
27#include "util/u_math.h"
28#include "util/u_mm.h"
29#include "util/u_memory.h"
30#include "util/u_pack_color.h"
31#include "util/u_transfer.h"
32
33#include "tgsi/tgsi_parse.h"
34
35#include "pipe/p_config.h"
36
37#include "r300_cb.h"
38#include "r300_context.h"
39#include "r300_emit.h"
40#include "r300_reg.h"
41#include "r300_screen.h"
42#include "r300_screen_buffer.h"
43#include "r300_state_inlines.h"
44#include "r300_fs.h"
45#include "r300_texture.h"
46#include "r300_vs.h"
47#include "r300_winsys.h"
48#include "r300_hyperz.h"
49
50/* r300_state: Functions used to intialize state context by translating
51 * Gallium state objects into semi-native r300 state objects. */
52
53#define UPDATE_STATE(cso, atom) \
54    if (cso != atom.state) { \
55        atom.state = cso;    \
56        r300_mark_atom_dirty(r300, &(atom));   \
57    }
58
59static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA,
60                                            unsigned dstRGB, unsigned dstA)
61{
62    /* If the blend equation is ADD or REVERSE_SUBTRACT,
63     * SRC_ALPHA == 0, and the following state is set, the colorbuffer
64     * will not be changed.
65     * Notice that the dst factors are the src factors inverted. */
66    return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
67            srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
68            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
69           (srcA == PIPE_BLENDFACTOR_SRC_COLOR ||
70            srcA == PIPE_BLENDFACTOR_SRC_ALPHA ||
71            srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
72            srcA == PIPE_BLENDFACTOR_ZERO) &&
73           (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
74            dstRGB == PIPE_BLENDFACTOR_ONE) &&
75           (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
76            dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
77            dstA == PIPE_BLENDFACTOR_ONE);
78}
79
80static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA,
81                                            unsigned dstRGB, unsigned dstA)
82{
83    /* If the blend equation is ADD or REVERSE_SUBTRACT,
84     * SRC_ALPHA == 1, and the following state is set, the colorbuffer
85     * will not be changed.
86     * Notice that the dst factors are the src factors inverted. */
87    return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
88            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
89           (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
90            srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
91            srcA == PIPE_BLENDFACTOR_ZERO) &&
92           (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
93            dstRGB == PIPE_BLENDFACTOR_ONE) &&
94           (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
95            dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
96            dstA == PIPE_BLENDFACTOR_ONE);
97}
98
99static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA,
100                                            unsigned dstRGB, unsigned dstA)
101{
102    /* If the blend equation is ADD or REVERSE_SUBTRACT,
103     * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer
104     * will not be changed.
105     * Notice that the dst factors are the src factors inverted. */
106    return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
107            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
108           (srcA == PIPE_BLENDFACTOR_ZERO) &&
109           (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
110            dstRGB == PIPE_BLENDFACTOR_ONE) &&
111           (dstA == PIPE_BLENDFACTOR_ONE);
112}
113
114static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA,
115                                            unsigned dstRGB, unsigned dstA)
116{
117    /* If the blend equation is ADD or REVERSE_SUBTRACT,
118     * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer
119     * will not be changed.
120     * Notice that the dst factors are the src factors inverted. */
121    return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
122            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
123           (srcA == PIPE_BLENDFACTOR_ZERO) &&
124           (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
125            dstRGB == PIPE_BLENDFACTOR_ONE) &&
126           (dstA == PIPE_BLENDFACTOR_ONE);
127}
128
129static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA,
130                                                  unsigned dstRGB, unsigned dstA)
131{
132    /* If the blend equation is ADD or REVERSE_SUBTRACT,
133     * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set,
134     * the colorbuffer will not be changed.
135     * Notice that the dst factors are the src factors inverted. */
136    return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
137            srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
138            srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
139            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
140           (srcA == PIPE_BLENDFACTOR_SRC_COLOR ||
141            srcA == PIPE_BLENDFACTOR_SRC_ALPHA ||
142            srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
143            srcA == PIPE_BLENDFACTOR_ZERO) &&
144           (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
145            dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
146            dstRGB == PIPE_BLENDFACTOR_ONE) &&
147           (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
148            dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
149            dstA == PIPE_BLENDFACTOR_ONE);
150}
151
152static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA,
153                                                  unsigned dstRGB, unsigned dstA)
154{
155    /* If the blend equation is ADD or REVERSE_SUBTRACT,
156     * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set,
157     * the colorbuffer will not be changed.
158     * Notice that the dst factors are the src factors inverted. */
159    return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
160            srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
161            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
162           (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
163            srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
164            srcA == PIPE_BLENDFACTOR_ZERO) &&
165           (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
166            dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
167            dstRGB == PIPE_BLENDFACTOR_ONE) &&
168           (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
169            dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
170            dstA == PIPE_BLENDFACTOR_ONE);
171}
172
173static unsigned bgra_cmask(unsigned mask)
174{
175    /* Gallium uses RGBA color ordering while R300 expects BGRA. */
176
177    return ((mask & PIPE_MASK_R) << 2) |
178           ((mask & PIPE_MASK_B) >> 2) |
179           (mask & (PIPE_MASK_G | PIPE_MASK_A));
180}
181
182/* Create a new blend state based on the CSO blend state.
183 *
184 * This encompasses alpha blending, logic/raster ops, and blend dithering. */
185static void* r300_create_blend_state(struct pipe_context* pipe,
186                                     const struct pipe_blend_state* state)
187{
188    struct r300_screen* r300screen = r300_screen(pipe->screen);
189    struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state);
190    uint32_t blend_control = 0;       /* R300_RB3D_CBLEND: 0x4e04 */
191    uint32_t alpha_blend_control = 0; /* R300_RB3D_ABLEND: 0x4e08 */
192    uint32_t color_channel_mask = 0;  /* R300_RB3D_COLOR_CHANNEL_MASK: 0x4e0c */
193    uint32_t rop = 0;                 /* R300_RB3D_ROPCNTL: 0x4e18 */
194    uint32_t dither = 0;              /* R300_RB3D_DITHER_CTL: 0x4e50 */
195    CB_LOCALS;
196
197    if (state->rt[0].blend_enable)
198    {
199        unsigned eqRGB = state->rt[0].rgb_func;
200        unsigned srcRGB = state->rt[0].rgb_src_factor;
201        unsigned dstRGB = state->rt[0].rgb_dst_factor;
202
203        unsigned eqA = state->rt[0].alpha_func;
204        unsigned srcA = state->rt[0].alpha_src_factor;
205        unsigned dstA = state->rt[0].alpha_dst_factor;
206
207        /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha,
208         * this is just the crappy D3D naming */
209        blend_control = R300_ALPHA_BLEND_ENABLE |
210            r300_translate_blend_function(eqRGB) |
211            ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) |
212            ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT);
213
214        /* Optimization: some operations do not require the destination color.
215         *
216         * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled,
217         * otherwise blending gives incorrect results. It seems to be
218         * a hardware bug. */
219        if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN ||
220            eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX ||
221            dstRGB != PIPE_BLENDFACTOR_ZERO ||
222            dstA != PIPE_BLENDFACTOR_ZERO ||
223            srcRGB == PIPE_BLENDFACTOR_DST_COLOR ||
224            srcRGB == PIPE_BLENDFACTOR_DST_ALPHA ||
225            srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR ||
226            srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
227            srcA == PIPE_BLENDFACTOR_DST_COLOR ||
228            srcA == PIPE_BLENDFACTOR_DST_ALPHA ||
229            srcA == PIPE_BLENDFACTOR_INV_DST_COLOR ||
230            srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
231            srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) {
232            /* Enable reading from the colorbuffer. */
233            blend_control |= R300_READ_ENABLE;
234
235            if (r300screen->caps.is_r500) {
236                /* Optimization: Depending on incoming pixels, we can
237                 * conditionally disable the reading in hardware... */
238                if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN &&
239                    eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) {
240                    /* Disable reading if SRC_ALPHA == 0. */
241                    if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
242                         dstRGB == PIPE_BLENDFACTOR_ZERO) &&
243                        (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
244                         dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
245                         dstA == PIPE_BLENDFACTOR_ZERO)) {
246                         blend_control |= R500_SRC_ALPHA_0_NO_READ;
247                    }
248
249                    /* Disable reading if SRC_ALPHA == 1. */
250                    if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
251                         dstRGB == PIPE_BLENDFACTOR_ZERO) &&
252                        (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
253                         dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
254                         dstA == PIPE_BLENDFACTOR_ZERO)) {
255                         blend_control |= R500_SRC_ALPHA_1_NO_READ;
256                    }
257                }
258            }
259        }
260
261        /* Optimization: discard pixels which don't change the colorbuffer.
262         *
263         * The code below is non-trivial and some math is involved.
264         *
265         * Discarding pixels must be disabled when FP16 AA is enabled.
266         * This is a hardware bug. Also, this implementation wouldn't work
267         * with FP blending enabled and equation clamping disabled.
268         *
269         * Equations other than ADD are rarely used and therefore won't be
270         * optimized. */
271        if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) &&
272            (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) {
273            /* ADD: X+Y
274             * REVERSE_SUBTRACT: Y-X
275             *
276             * The idea is:
277             * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1,
278             * then CB will not be changed.
279             *
280             * Given the srcFactor and dstFactor variables, we can derive
281             * what src and dst should be equal to and discard appropriate
282             * pixels.
283             */
284            if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) {
285                blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0;
286            } else if (blend_discard_if_src_alpha_1(srcRGB, srcA,
287                                                    dstRGB, dstA)) {
288                blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1;
289            } else if (blend_discard_if_src_color_0(srcRGB, srcA,
290                                                    dstRGB, dstA)) {
291                blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0;
292            } else if (blend_discard_if_src_color_1(srcRGB, srcA,
293                                                    dstRGB, dstA)) {
294                blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1;
295            } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA,
296                                                          dstRGB, dstA)) {
297                blend_control |=
298                    R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0;
299            } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA,
300                                                          dstRGB, dstA)) {
301                blend_control |=
302                    R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1;
303            }
304        }
305
306        /* separate alpha */
307        if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) {
308            blend_control |= R300_SEPARATE_ALPHA_ENABLE;
309            alpha_blend_control =
310                r300_translate_blend_function(eqA) |
311                (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) |
312                (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT);
313        }
314    }
315
316    /* PIPE_LOGICOP_* don't need to be translated, fortunately. */
317    if (state->logicop_enable) {
318        rop = R300_RB3D_ROPCNTL_ROP_ENABLE |
319                (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT;
320    }
321
322    /* Color channel masks for all MRTs. */
323    color_channel_mask = bgra_cmask(state->rt[0].colormask);
324    if (r300screen->caps.is_r500 && state->independent_blend_enable) {
325        if (state->rt[1].blend_enable) {
326            color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4;
327        }
328        if (state->rt[2].blend_enable) {
329            color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8;
330        }
331        if (state->rt[3].blend_enable) {
332            color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12;
333        }
334    }
335
336    /* Neither fglrx nor classic r300 ever set this, regardless of dithering
337     * state. Since it's an optional implementation detail, we can leave it
338     * out and never dither.
339     *
340     * This could be revisited if we ever get quality or conformance hints.
341     *
342    if (state->dither) {
343        dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT |
344                        R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT;
345    }
346    */
347
348    /* Build a command buffer. */
349    BEGIN_CB(blend->cb, 8);
350    OUT_CB_REG(R300_RB3D_ROPCNTL, rop);
351    OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3);
352    OUT_CB(blend_control);
353    OUT_CB(alpha_blend_control);
354    OUT_CB(color_channel_mask);
355    OUT_CB_REG(R300_RB3D_DITHER_CTL, dither);
356    END_CB;
357
358    /* The same as above, but with no colorbuffer reads and writes. */
359    BEGIN_CB(blend->cb_no_readwrite, 8);
360    OUT_CB_REG(R300_RB3D_ROPCNTL, rop);
361    OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3);
362    OUT_CB(0);
363    OUT_CB(0);
364    OUT_CB(0);
365    OUT_CB_REG(R300_RB3D_DITHER_CTL, dither);
366    END_CB;
367
368    return (void*)blend;
369}
370
371/* Bind blend state. */
372static void r300_bind_blend_state(struct pipe_context* pipe,
373                                  void* state)
374{
375    struct r300_context* r300 = r300_context(pipe);
376
377    UPDATE_STATE(state, r300->blend_state);
378}
379
380/* Free blend state. */
381static void r300_delete_blend_state(struct pipe_context* pipe,
382                                    void* state)
383{
384    FREE(state);
385}
386
387/* Convert float to 10bit integer */
388static unsigned float_to_fixed10(float f)
389{
390    return CLAMP((unsigned)(f * 1023.9f), 0, 1023);
391}
392
393/* Set blend color.
394 * Setup both R300 and R500 registers, figure out later which one to write. */
395static void r300_set_blend_color(struct pipe_context* pipe,
396                                 const struct pipe_blend_color* color)
397{
398    struct r300_context* r300 = r300_context(pipe);
399    struct r300_blend_color_state* state =
400        (struct r300_blend_color_state*)r300->blend_color_state.state;
401    CB_LOCALS;
402
403    if (r300->screen->caps.is_r500) {
404        /* XXX if FP16 blending is enabled, we should use the FP16 format */
405        BEGIN_CB(state->cb, 3);
406        OUT_CB_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2);
407        OUT_CB(float_to_fixed10(color->color[0]) |
408               (float_to_fixed10(color->color[3]) << 16));
409        OUT_CB(float_to_fixed10(color->color[2]) |
410               (float_to_fixed10(color->color[1]) << 16));
411        END_CB;
412    } else {
413        union util_color uc;
414        util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
415
416        BEGIN_CB(state->cb, 2);
417        OUT_CB_REG(R300_RB3D_BLEND_COLOR, uc.ui);
418        END_CB;
419    }
420
421    r300_mark_atom_dirty(r300, &r300->blend_color_state);
422}
423
424static void r300_set_clip_state(struct pipe_context* pipe,
425                                const struct pipe_clip_state* state)
426{
427    struct r300_context* r300 = r300_context(pipe);
428    struct r300_clip_state *clip =
429            (struct r300_clip_state*)r300->clip_state.state;
430    CB_LOCALS;
431
432    clip->clip = *state;
433
434    if (r300->screen->caps.has_tcl) {
435        r300->clip_state.size = 2 + !!state->nr * 3 + state->nr * 4;
436
437        BEGIN_CB(clip->cb, r300->clip_state.size);
438        if (state->nr) {
439           OUT_CB_REG(R300_VAP_PVS_VECTOR_INDX_REG,
440                   (r300->screen->caps.is_r500 ?
441                    R500_PVS_UCP_START : R300_PVS_UCP_START));
442           OUT_CB_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, state->nr * 4);
443           OUT_CB_TABLE(state->ucp, state->nr * 4);
444        }
445        OUT_CB_REG(R300_VAP_CLIP_CNTL, ((1 << state->nr) - 1) |
446                   R300_PS_UCP_MODE_CLIP_AS_TRIFAN);
447        END_CB;
448
449        r300_mark_atom_dirty(r300, &r300->clip_state);
450    } else {
451        draw_set_clip_state(r300->draw, state);
452    }
453}
454
455static void
456r300_set_sample_mask(struct pipe_context *pipe,
457                     unsigned sample_mask)
458{
459}
460
461
462/* Create a new depth, stencil, and alpha state based on the CSO dsa state.
463 *
464 * This contains the depth buffer, stencil buffer, alpha test, and such.
465 * On the Radeon, depth and stencil buffer setup are intertwined, which is
466 * the reason for some of the strange-looking assignments across registers. */
467static void*
468        r300_create_dsa_state(struct pipe_context* pipe,
469                              const struct pipe_depth_stencil_alpha_state* state)
470{
471    struct r300_capabilities *caps = &r300_screen(pipe->screen)->caps;
472    struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state);
473    CB_LOCALS;
474
475    dsa->dsa = *state;
476
477    /* Depth test setup. - separate write mask depth for decomp flush */
478    if (state->depth.writemask) {
479        dsa->z_buffer_control |= R300_Z_WRITE_ENABLE;
480    }
481
482    if (state->depth.enabled) {
483        dsa->z_buffer_control |= R300_Z_ENABLE;
484
485        dsa->z_stencil_control |=
486            (r300_translate_depth_stencil_function(state->depth.func) <<
487                R300_Z_FUNC_SHIFT);
488    }
489
490    /* Stencil buffer setup. */
491    if (state->stencil[0].enabled) {
492        dsa->z_buffer_control |= R300_STENCIL_ENABLE;
493        dsa->z_stencil_control |=
494            (r300_translate_depth_stencil_function(state->stencil[0].func) <<
495                R300_S_FRONT_FUNC_SHIFT) |
496            (r300_translate_stencil_op(state->stencil[0].fail_op) <<
497                R300_S_FRONT_SFAIL_OP_SHIFT) |
498            (r300_translate_stencil_op(state->stencil[0].zpass_op) <<
499                R300_S_FRONT_ZPASS_OP_SHIFT) |
500            (r300_translate_stencil_op(state->stencil[0].zfail_op) <<
501                R300_S_FRONT_ZFAIL_OP_SHIFT);
502
503        dsa->stencil_ref_mask =
504                (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) |
505                (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT);
506
507        if (state->stencil[1].enabled) {
508            dsa->two_sided = TRUE;
509
510            dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK;
511            dsa->z_stencil_control |=
512            (r300_translate_depth_stencil_function(state->stencil[1].func) <<
513                R300_S_BACK_FUNC_SHIFT) |
514            (r300_translate_stencil_op(state->stencil[1].fail_op) <<
515                R300_S_BACK_SFAIL_OP_SHIFT) |
516            (r300_translate_stencil_op(state->stencil[1].zpass_op) <<
517                R300_S_BACK_ZPASS_OP_SHIFT) |
518            (r300_translate_stencil_op(state->stencil[1].zfail_op) <<
519                R300_S_BACK_ZFAIL_OP_SHIFT);
520
521            dsa->stencil_ref_bf =
522                (state->stencil[1].valuemask << R300_STENCILMASK_SHIFT) |
523                (state->stencil[1].writemask << R300_STENCILWRITEMASK_SHIFT);
524
525            if (caps->is_r500) {
526                dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK;
527            } else {
528                dsa->two_sided_stencil_ref =
529                  (state->stencil[0].valuemask != state->stencil[1].valuemask ||
530                   state->stencil[0].writemask != state->stencil[1].writemask);
531            }
532        }
533    }
534
535    /* Alpha test setup. */
536    if (state->alpha.enabled) {
537        dsa->alpha_function =
538            r300_translate_alpha_function(state->alpha.func) |
539            R300_FG_ALPHA_FUNC_ENABLE;
540
541        /* We could use 10bit alpha ref but who needs that? */
542        dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value);
543
544        if (caps->is_r500)
545            dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT;
546    }
547
548    BEGIN_CB(&dsa->cb_begin, 8);
549    OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function);
550    OUT_CB_REG_SEQ(R300_ZB_CNTL, 3);
551    OUT_CB(dsa->z_buffer_control);
552    OUT_CB(dsa->z_stencil_control);
553    OUT_CB(dsa->stencil_ref_mask);
554    OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf);
555    END_CB;
556
557    BEGIN_CB(dsa->cb_no_readwrite, 8);
558    OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function);
559    OUT_CB_REG_SEQ(R300_ZB_CNTL, 3);
560    OUT_CB(0);
561    OUT_CB(0);
562    OUT_CB(0);
563    OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, 0);
564    END_CB;
565
566    return (void*)dsa;
567}
568
569static void r300_dsa_inject_stencilref(struct r300_context *r300)
570{
571    struct r300_dsa_state *dsa =
572            (struct r300_dsa_state*)r300->dsa_state.state;
573
574    if (!dsa)
575        return;
576
577    dsa->stencil_ref_mask =
578        (dsa->stencil_ref_mask & ~R300_STENCILREF_MASK) |
579        r300->stencil_ref.ref_value[0];
580    dsa->stencil_ref_bf =
581        (dsa->stencil_ref_bf & ~R300_STENCILREF_MASK) |
582        r300->stencil_ref.ref_value[1];
583}
584
585/* Bind DSA state. */
586static void r300_bind_dsa_state(struct pipe_context* pipe,
587                                void* state)
588{
589    struct r300_context* r300 = r300_context(pipe);
590
591    if (!state) {
592        return;
593    }
594
595    UPDATE_STATE(state, r300->dsa_state);
596
597    r300_mark_atom_dirty(r300, &r300->hyperz_state); /* Will be updated before the emission. */
598    r300_dsa_inject_stencilref(r300);
599}
600
601/* Free DSA state. */
602static void r300_delete_dsa_state(struct pipe_context* pipe,
603                                  void* state)
604{
605    FREE(state);
606}
607
608static void r300_set_stencil_ref(struct pipe_context* pipe,
609                                 const struct pipe_stencil_ref* sr)
610{
611    struct r300_context* r300 = r300_context(pipe);
612
613    r300->stencil_ref = *sr;
614
615    r300_dsa_inject_stencilref(r300);
616    r300_mark_atom_dirty(r300, &r300->dsa_state);
617}
618
619static void r300_tex_set_tiling_flags(struct r300_context *r300,
620                                      struct r300_resource *tex,
621                                      unsigned level)
622{
623    /* Check if the macrotile flag needs to be changed.
624     * Skip changing the flags otherwise. */
625    if (tex->tex.macrotile[tex->surface_level] !=
626        tex->tex.macrotile[level]) {
627        r300->rws->buffer_set_tiling(tex->buf, r300->cs,
628                tex->tex.microtile, tex->tex.macrotile[level],
629                tex->tex.stride_in_bytes[0]);
630
631        tex->surface_level = level;
632    }
633}
634
635/* This switcheroo is needed just because of goddamned MACRO_SWITCH. */
636static void r300_fb_set_tiling_flags(struct r300_context *r300,
637                               const struct pipe_framebuffer_state *state)
638{
639    unsigned i;
640
641    /* Set tiling flags for new surfaces. */
642    for (i = 0; i < state->nr_cbufs; i++) {
643        r300_tex_set_tiling_flags(r300,
644                                  r300_resource(state->cbufs[i]->texture),
645                                  state->cbufs[i]->u.tex.level);
646    }
647    if (state->zsbuf) {
648        r300_tex_set_tiling_flags(r300,
649                                  r300_resource(state->zsbuf->texture),
650                                  state->zsbuf->u.tex.level);
651    }
652}
653
654static void r300_print_fb_surf_info(struct pipe_surface *surf, unsigned index,
655                                    const char *binding)
656{
657    struct pipe_resource *tex = surf->texture;
658    struct r300_resource *rtex = r300_resource(tex);
659
660    fprintf(stderr,
661            "r300:   %s[%i] Dim: %ix%i, Firstlayer: %i, "
662            "Lastlayer: %i, Level: %i, Format: %s\n"
663
664            "r300:     TEX: Macro: %s, Micro: %s, Pitch: %i, "
665            "Dim: %ix%ix%i, LastLevel: %i, Format: %s\n",
666
667            binding, index, surf->width, surf->height,
668            surf->u.tex.first_layer, surf->u.tex.last_layer, surf->u.tex.level,
669            util_format_short_name(surf->format),
670
671            rtex->tex.macrotile[0] ? "YES" : " NO",
672            rtex->tex.microtile ? "YES" : " NO",
673            rtex->tex.stride_in_pixels[0],
674            tex->width0, tex->height0, tex->depth0,
675            tex->last_level, util_format_short_name(tex->format));
676}
677
678void r300_mark_fb_state_dirty(struct r300_context *r300,
679                              enum r300_fb_state_change change)
680{
681    struct pipe_framebuffer_state *state = r300->fb_state.state;
682    boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
683
684    r300_mark_atom_dirty(r300, &r300->gpu_flush);
685    r300_mark_atom_dirty(r300, &r300->fb_state);
686
687    /* What is marked as dirty depends on the enum r300_fb_state_change. */
688    if (change == R300_CHANGED_FB_STATE) {
689        r300_mark_atom_dirty(r300, &r300->aa_state);
690    }
691
692    if (change == R300_CHANGED_FB_STATE ||
693        change == R300_CHANGED_HYPERZ_FLAG) {
694        r300_mark_atom_dirty(r300, &r300->hyperz_state);
695    }
696
697    if (change == R300_CHANGED_FB_STATE ||
698        change == R300_CHANGED_MULTIWRITE) {
699        r300_mark_atom_dirty(r300, &r300->fb_state_pipelined);
700    }
701
702    /* Now compute the fb_state atom size. */
703    r300->fb_state.size = 2 + (8 * state->nr_cbufs);
704
705    if (r300->cbzb_clear)
706        r300->fb_state.size += 10;
707    else if (state->zsbuf) {
708        r300->fb_state.size += 10;
709        if (can_hyperz)
710            r300->fb_state.size += r300->screen->caps.hiz_ram ? 8 : 4;
711    }
712
713    /* The size of the rest of atoms stays the same. */
714}
715
716static void
717r300_set_framebuffer_state(struct pipe_context* pipe,
718                           const struct pipe_framebuffer_state* state)
719{
720    struct r300_context* r300 = r300_context(pipe);
721    struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
722    struct pipe_framebuffer_state *old_state = r300->fb_state.state;
723    boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
724    unsigned max_width, max_height, i;
725    uint32_t zbuffer_bpp = 0;
726
727    if (r300->screen->caps.is_r500) {
728        max_width = max_height = 4096;
729    } else if (r300->screen->caps.is_r400) {
730        max_width = max_height = 4021;
731    } else {
732        max_width = max_height = 2560;
733    }
734
735    if (state->width > max_width || state->height > max_height) {
736        fprintf(stderr, "r300: Implementation error: Render targets are too "
737        "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__);
738        return;
739    }
740
741    if (old_state->zsbuf && r300->zmask_in_use && !r300->zmask_locked) {
742        /* There is a zmask in use, what are we gonna do? */
743        if (state->zsbuf) {
744            if (!pipe_surface_equal(old_state->zsbuf, state->zsbuf)) {
745                /* Decompress the currently bound zbuffer before we bind another one. */
746                r300_decompress_zmask(r300);
747            }
748        } else {
749            /* We don't bind another zbuffer, so lock the current one. */
750            r300->zmask_locked = TRUE;
751            pipe_surface_reference(&r300->locked_zbuffer, old_state->zsbuf);
752        }
753    } else if (r300->zmask_locked && r300->locked_zbuffer) {
754        /* We have a locked zbuffer now, what are we gonna do? */
755        if (state->zsbuf) {
756            if (!pipe_surface_equal(r300->locked_zbuffer, state->zsbuf)) {
757                /* We are binding some other zbuffer, so decompress the locked one,
758                 * it gets unlocked automatically. */
759                r300_decompress_zmask_locked_unsafe(r300);
760            } else {
761                /* We are binding the locked zbuffer again, so unlock it. */
762                r300->zmask_locked = FALSE;
763            }
764        }
765    }
766
767    /* If nr_cbufs is changed from zero to non-zero or vice versa... */
768    if (!!old_state->nr_cbufs != !!state->nr_cbufs) {
769        r300_mark_atom_dirty(r300, &r300->blend_state);
770    }
771    /* If zsbuf is set from NULL to non-NULL or vice versa.. */
772    if (!!old_state->zsbuf != !!state->zsbuf) {
773        r300_mark_atom_dirty(r300, &r300->dsa_state);
774    }
775
776    /* The tiling flags are dependent on the surface miplevel, unfortunately. */
777    r300_fb_set_tiling_flags(r300, state);
778
779    util_copy_framebuffer_state(r300->fb_state.state, state);
780
781    if (!r300->zmask_locked) {
782        pipe_surface_reference(&r300->locked_zbuffer, NULL);
783    }
784
785    r300_mark_fb_state_dirty(r300, R300_CHANGED_FB_STATE);
786    r300->validate_buffers = TRUE;
787
788    if (state->zsbuf) {
789        switch (util_format_get_blocksize(state->zsbuf->texture->format)) {
790        case 2:
791            zbuffer_bpp = 16;
792            break;
793        case 4:
794            zbuffer_bpp = 24;
795            break;
796        }
797
798        /* Setup Hyper-Z. */
799        if (can_hyperz) {
800            struct r300_surface *zs_surf = r300_surface(state->zsbuf);
801            struct r300_resource *tex = r300_resource(zs_surf->base.texture);
802            int level = zs_surf->base.u.tex.level;
803
804            /* work out whether we can support hiz on this buffer */
805            r300_hiz_alloc_block(r300, zs_surf);
806
807            DBG(r300, DBG_HYPERZ,
808                "hyper-z features: hiz: %d @ %08x\n", tex->hiz_mem[level] ? 1 : 0,
809                tex->hiz_mem[level] ? tex->hiz_mem[level]->ofs : 0xdeadbeef);
810        }
811
812        /* Polygon offset depends on the zbuffer bit depth. */
813        if (r300->zbuffer_bpp != zbuffer_bpp) {
814            r300->zbuffer_bpp = zbuffer_bpp;
815
816            if (r300->polygon_offset_enabled)
817                r300_mark_atom_dirty(r300, &r300->rs_state);
818        }
819    }
820
821    /* Set up AA config. */
822    if (r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0)) {
823        if (state->nr_cbufs && state->cbufs[0]->texture->nr_samples > 1) {
824            aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE;
825
826            switch (state->cbufs[0]->texture->nr_samples) {
827                case 2:
828                    aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2;
829                    break;
830                case 3:
831                    aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_3;
832                    break;
833                case 4:
834                    aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4;
835                    break;
836                case 6:
837                    aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6;
838                    break;
839            }
840        } else {
841            aa->aa_config = 0;
842        }
843    }
844
845    if (DBG_ON(r300, DBG_FB)) {
846        fprintf(stderr, "r300: set_framebuffer_state:\n");
847        for (i = 0; i < state->nr_cbufs; i++) {
848            r300_print_fb_surf_info(state->cbufs[i], i, "CB");
849        }
850        if (state->zsbuf) {
851            r300_print_fb_surf_info(state->zsbuf, 0, "ZB");
852        }
853    }
854}
855
856/* Create fragment shader state. */
857static void* r300_create_fs_state(struct pipe_context* pipe,
858                                  const struct pipe_shader_state* shader)
859{
860    struct r300_fragment_shader* fs = NULL;
861
862    fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader);
863
864    /* Copy state directly into shader. */
865    fs->state = *shader;
866    fs->state.tokens = tgsi_dup_tokens(shader->tokens);
867
868    return (void*)fs;
869}
870
871void r300_mark_fs_code_dirty(struct r300_context *r300)
872{
873    struct r300_fragment_shader* fs = r300_fs(r300);
874
875    r300_mark_atom_dirty(r300, &r300->fs);
876    r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state);
877    r300_mark_atom_dirty(r300, &r300->fs_constants);
878    r300->fs.size = fs->shader->cb_code_size;
879
880    if (r300->screen->caps.is_r500) {
881        r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 7;
882        r300->fs_constants.size = fs->shader->externals_count * 4 + 3;
883    } else {
884        r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 5;
885        r300->fs_constants.size = fs->shader->externals_count * 4 + 1;
886    }
887
888    ((struct r300_constant_buffer*)r300->fs_constants.state)->remap_table =
889            fs->shader->code.constants_remap_table;
890}
891
892/* Bind fragment shader state. */
893static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
894{
895    struct r300_context* r300 = r300_context(pipe);
896    struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
897    struct pipe_framebuffer_state *fb = r300->fb_state.state;
898    boolean last_multi_write;
899
900    if (fs == NULL) {
901        r300->fs.state = NULL;
902        return;
903    }
904
905    last_multi_write = r300_fragment_shader_writes_all(r300_fs(r300));
906
907    r300->fs.state = fs;
908    r300_pick_fragment_shader(r300);
909    r300_mark_fs_code_dirty(r300);
910
911    if (fb->nr_cbufs > 1 &&
912        last_multi_write != r300_fragment_shader_writes_all(fs)) {
913        r300_mark_fb_state_dirty(r300, R300_CHANGED_MULTIWRITE);
914    }
915
916    r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */
917}
918
919/* Delete fragment shader state. */
920static void r300_delete_fs_state(struct pipe_context* pipe, void* shader)
921{
922    struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
923    struct r300_fragment_shader_code *tmp, *ptr = fs->first;
924
925    while (ptr) {
926        tmp = ptr;
927        ptr = ptr->next;
928        rc_constants_destroy(&tmp->code.constants);
929        FREE(tmp->cb_code);
930        FREE(tmp);
931    }
932    FREE((void*)fs->state.tokens);
933    FREE(shader);
934}
935
936static void r300_set_polygon_stipple(struct pipe_context* pipe,
937                                     const struct pipe_poly_stipple* state)
938{
939    /* XXX no idea how to set this up, but not terribly important */
940}
941
942/* Create a new rasterizer state based on the CSO rasterizer state.
943 *
944 * This is a very large chunk of state, and covers most of the graphics
945 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
946 *
947 * In a not entirely unironic sidenote, this state has nearly nothing to do
948 * with the actual block on the Radeon called the rasterizer (RS). */
949static void* r300_create_rs_state(struct pipe_context* pipe,
950                                  const struct pipe_rasterizer_state* state)
951{
952    struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state);
953    float psiz;
954    uint32_t vap_control_status;    /* R300_VAP_CNTL_STATUS: 0x2140 */
955    uint32_t point_size;            /* R300_GA_POINT_SIZE: 0x421c */
956    uint32_t point_minmax;          /* R300_GA_POINT_MINMAX: 0x4230 */
957    uint32_t line_control;          /* R300_GA_LINE_CNTL: 0x4234 */
958    uint32_t polygon_offset_enable; /* R300_SU_POLY_OFFSET_ENABLE: 0x42b4 */
959    uint32_t cull_mode;             /* R300_SU_CULL_MODE: 0x42b8 */
960    uint32_t line_stipple_config;   /* R300_GA_LINE_STIPPLE_CONFIG: 0x4328 */
961    uint32_t line_stipple_value;    /* R300_GA_LINE_STIPPLE_VALUE: 0x4260 */
962    uint32_t polygon_mode;          /* R300_GA_POLY_MODE: 0x4288 */
963    uint32_t clip_rule;             /* R300_SC_CLIP_RULE: 0x43D0 */
964
965    /* Point sprites texture coordinates, 0: lower left, 1: upper right */
966    float point_texcoord_left = 0;  /* R300_GA_POINT_S0: 0x4200 */
967    float point_texcoord_bottom = 0;/* R300_GA_POINT_T0: 0x4204 */
968    float point_texcoord_right = 1; /* R300_GA_POINT_S1: 0x4208 */
969    float point_texcoord_top = 0;   /* R300_GA_POINT_T1: 0x420c */
970    CB_LOCALS;
971
972    /* Copy rasterizer state. */
973    rs->rs = *state;
974    rs->rs_draw = *state;
975
976    rs->rs.sprite_coord_enable = state->point_quad_rasterization *
977                                 state->sprite_coord_enable;
978
979    /* Override some states for Draw. */
980    rs->rs_draw.sprite_coord_enable = 0; /* We can do this in HW. */
981
982#ifdef PIPE_ARCH_LITTLE_ENDIAN
983    vap_control_status = R300_VC_NO_SWAP;
984#else
985    vap_control_status = R300_VC_32BIT_SWAP;
986#endif
987
988    /* If no TCL engine is present, turn off the HW TCL. */
989    if (!r300_screen(pipe->screen)->caps.has_tcl) {
990        vap_control_status |= R300_VAP_TCL_BYPASS;
991    }
992
993    /* Point size width and height. */
994    point_size =
995        pack_float_16_6x(state->point_size) |
996        (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT);
997
998    /* Point size clamping. */
999    if (state->point_size_per_vertex) {
1000        /* Per-vertex point size.
1001         * Clamp to [0, max FB size] */
1002        psiz = pipe->screen->get_paramf(pipe->screen,
1003                                        PIPE_CAP_MAX_POINT_WIDTH);
1004        point_minmax =
1005            pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT;
1006    } else {
1007        /* We cannot disable the point-size vertex output,
1008         * so clamp it. */
1009        psiz = state->point_size;
1010        point_minmax =
1011            (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MIN_SHIFT) |
1012            (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT);
1013    }
1014
1015    /* Line control. */
1016    line_control = pack_float_16_6x(state->line_width) |
1017        R300_GA_LINE_CNTL_END_TYPE_COMP;
1018
1019    /* Enable polygon mode */
1020    polygon_mode = 0;
1021    if (state->fill_front != PIPE_POLYGON_MODE_FILL ||
1022        state->fill_back != PIPE_POLYGON_MODE_FILL) {
1023        polygon_mode = R300_GA_POLY_MODE_DUAL;
1024    }
1025
1026    /* Front face */
1027    if (state->front_ccw)
1028        cull_mode = R300_FRONT_FACE_CCW;
1029    else
1030        cull_mode = R300_FRONT_FACE_CW;
1031
1032    /* Polygon offset */
1033    polygon_offset_enable = 0;
1034    if (util_get_offset(state, state->fill_front)) {
1035       polygon_offset_enable |= R300_FRONT_ENABLE;
1036    }
1037    if (util_get_offset(state, state->fill_back)) {
1038       polygon_offset_enable |= R300_BACK_ENABLE;
1039    }
1040
1041    rs->polygon_offset_enable = polygon_offset_enable != 0;
1042
1043    /* Polygon mode */
1044    if (polygon_mode) {
1045       polygon_mode |=
1046          r300_translate_polygon_mode_front(state->fill_front);
1047       polygon_mode |=
1048          r300_translate_polygon_mode_back(state->fill_back);
1049    }
1050
1051    if (state->cull_face & PIPE_FACE_FRONT) {
1052        cull_mode |= R300_CULL_FRONT;
1053    }
1054    if (state->cull_face & PIPE_FACE_BACK) {
1055        cull_mode |= R300_CULL_BACK;
1056    }
1057
1058    if (state->line_stipple_enable) {
1059        line_stipple_config =
1060            R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE |
1061            (fui((float)state->line_stipple_factor) &
1062                R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK);
1063        /* XXX this might need to be scaled up */
1064        line_stipple_value = state->line_stipple_pattern;
1065    } else {
1066        line_stipple_config = 0;
1067        line_stipple_value = 0;
1068    }
1069
1070    if (state->flatshade) {
1071        rs->color_control = R300_SHADE_MODEL_FLAT;
1072    } else {
1073        rs->color_control = R300_SHADE_MODEL_SMOOTH;
1074    }
1075
1076    clip_rule = state->scissor ? 0xAAAA : 0xFFFF;
1077
1078    /* Point sprites coord mode */
1079    if (rs->rs.sprite_coord_enable) {
1080        switch (state->sprite_coord_mode) {
1081            case PIPE_SPRITE_COORD_UPPER_LEFT:
1082                point_texcoord_top = 0.0f;
1083                point_texcoord_bottom = 1.0f;
1084                break;
1085            case PIPE_SPRITE_COORD_LOWER_LEFT:
1086                point_texcoord_top = 1.0f;
1087                point_texcoord_bottom = 0.0f;
1088                break;
1089        }
1090    }
1091
1092    /* Build the main command buffer. */
1093    BEGIN_CB(rs->cb_main, RS_STATE_MAIN_SIZE);
1094    OUT_CB_REG(R300_VAP_CNTL_STATUS, vap_control_status);
1095    OUT_CB_REG(R300_GA_POINT_SIZE, point_size);
1096    OUT_CB_REG_SEQ(R300_GA_POINT_MINMAX, 2);
1097    OUT_CB(point_minmax);
1098    OUT_CB(line_control);
1099    OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_ENABLE, 2);
1100    OUT_CB(polygon_offset_enable);
1101    rs->cull_mode_index = 9;
1102    OUT_CB(cull_mode);
1103    OUT_CB_REG(R300_GA_LINE_STIPPLE_CONFIG, line_stipple_config);
1104    OUT_CB_REG(R300_GA_LINE_STIPPLE_VALUE, line_stipple_value);
1105    OUT_CB_REG(R300_GA_POLY_MODE, polygon_mode);
1106    OUT_CB_REG(R300_SC_CLIP_RULE, clip_rule);
1107    OUT_CB_REG_SEQ(R300_GA_POINT_S0, 4);
1108    OUT_CB_32F(point_texcoord_left);
1109    OUT_CB_32F(point_texcoord_bottom);
1110    OUT_CB_32F(point_texcoord_right);
1111    OUT_CB_32F(point_texcoord_top);
1112    END_CB;
1113
1114    /* Build the two command buffers for polygon offset setup. */
1115    if (polygon_offset_enable) {
1116        float scale = state->offset_scale * 12;
1117        float offset = state->offset_units * 4;
1118
1119        BEGIN_CB(rs->cb_poly_offset_zb16, 5);
1120        OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4);
1121        OUT_CB_32F(scale);
1122        OUT_CB_32F(offset);
1123        OUT_CB_32F(scale);
1124        OUT_CB_32F(offset);
1125        END_CB;
1126
1127        offset = state->offset_units * 2;
1128
1129        BEGIN_CB(rs->cb_poly_offset_zb24, 5);
1130        OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4);
1131        OUT_CB_32F(scale);
1132        OUT_CB_32F(offset);
1133        OUT_CB_32F(scale);
1134        OUT_CB_32F(offset);
1135        END_CB;
1136    }
1137
1138    return (void*)rs;
1139}
1140
1141/* Bind rasterizer state. */
1142static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
1143{
1144    struct r300_context* r300 = r300_context(pipe);
1145    struct r300_rs_state* rs = (struct r300_rs_state*)state;
1146    int last_sprite_coord_enable = r300->sprite_coord_enable;
1147    boolean last_two_sided_color = r300->two_sided_color;
1148
1149    if (r300->draw && rs) {
1150        draw_set_rasterizer_state(r300->draw, &rs->rs_draw, state);
1151    }
1152
1153    if (rs) {
1154        r300->polygon_offset_enabled = rs->polygon_offset_enable;
1155        r300->sprite_coord_enable = rs->rs.sprite_coord_enable;
1156        r300->two_sided_color = rs->rs.light_twoside;
1157    } else {
1158        r300->polygon_offset_enabled = FALSE;
1159        r300->sprite_coord_enable = 0;
1160        r300->two_sided_color = FALSE;
1161    }
1162
1163    UPDATE_STATE(state, r300->rs_state);
1164    r300->rs_state.size = RS_STATE_MAIN_SIZE + (r300->polygon_offset_enabled ? 5 : 0);
1165
1166    if (last_sprite_coord_enable != r300->sprite_coord_enable ||
1167        last_two_sided_color != r300->two_sided_color) {
1168        r300_mark_atom_dirty(r300, &r300->rs_block_state);
1169    }
1170}
1171
1172/* Free rasterizer state. */
1173static void r300_delete_rs_state(struct pipe_context* pipe, void* state)
1174{
1175    FREE(state);
1176}
1177
1178static void*
1179        r300_create_sampler_state(struct pipe_context* pipe,
1180                                  const struct pipe_sampler_state* state)
1181{
1182    struct r300_context* r300 = r300_context(pipe);
1183    struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state);
1184    boolean is_r500 = r300->screen->caps.is_r500;
1185    int lod_bias;
1186
1187    sampler->state = *state;
1188
1189    /* r300 doesn't handle CLAMP and MIRROR_CLAMP correctly when either MAG
1190     * or MIN filter is NEAREST. Since texwrap produces same results
1191     * for CLAMP and CLAMP_TO_EDGE, we use them instead. */
1192    if (sampler->state.min_img_filter == PIPE_TEX_FILTER_NEAREST ||
1193        sampler->state.mag_img_filter == PIPE_TEX_FILTER_NEAREST) {
1194        /* Wrap S. */
1195        if (sampler->state.wrap_s == PIPE_TEX_WRAP_CLAMP)
1196            sampler->state.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1197        else if (sampler->state.wrap_s == PIPE_TEX_WRAP_MIRROR_CLAMP)
1198            sampler->state.wrap_s = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE;
1199
1200        /* Wrap T. */
1201        if (sampler->state.wrap_t == PIPE_TEX_WRAP_CLAMP)
1202            sampler->state.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1203        else if (sampler->state.wrap_t == PIPE_TEX_WRAP_MIRROR_CLAMP)
1204            sampler->state.wrap_t = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE;
1205
1206        /* Wrap R. */
1207        if (sampler->state.wrap_r == PIPE_TEX_WRAP_CLAMP)
1208            sampler->state.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1209        else if (sampler->state.wrap_r == PIPE_TEX_WRAP_MIRROR_CLAMP)
1210            sampler->state.wrap_r = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE;
1211    }
1212
1213    sampler->filter0 |=
1214        (r300_translate_wrap(sampler->state.wrap_s) << R300_TX_WRAP_S_SHIFT) |
1215        (r300_translate_wrap(sampler->state.wrap_t) << R300_TX_WRAP_T_SHIFT) |
1216        (r300_translate_wrap(sampler->state.wrap_r) << R300_TX_WRAP_R_SHIFT);
1217
1218    sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter,
1219                                                   state->mag_img_filter,
1220                                                   state->min_mip_filter,
1221                                                   state->max_anisotropy > 0);
1222
1223    sampler->filter0 |= r300_anisotropy(state->max_anisotropy);
1224
1225    /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
1226    /* We must pass these to the merge function to clamp them properly. */
1227    sampler->min_lod = (unsigned)MAX2(state->min_lod, 0);
1228    sampler->max_lod = (unsigned)MAX2(ceilf(state->max_lod), 0);
1229
1230    lod_bias = CLAMP((int)(state->lod_bias * 32 + 1), -(1 << 9), (1 << 9) - 1);
1231
1232    sampler->filter1 |= (lod_bias << R300_LOD_BIAS_SHIFT) & R300_LOD_BIAS_MASK;
1233
1234    /* This is very high quality anisotropic filtering for R5xx.
1235     * It's good for benchmarking the performance of texturing but
1236     * in practice we don't want to slow down the driver because it's
1237     * a pretty good performance killer. Feel free to play with it. */
1238    if (DBG_ON(r300, DBG_ANISOHQ) && is_r500) {
1239        sampler->filter1 |= r500_anisotropy(state->max_anisotropy);
1240    }
1241
1242    /* R500-specific fixups and optimizations */
1243    if (r300->screen->caps.is_r500) {
1244        sampler->filter1 |= R500_BORDER_FIX;
1245    }
1246
1247    return (void*)sampler;
1248}
1249
1250static void r300_bind_sampler_states(struct pipe_context* pipe,
1251                                     unsigned count,
1252                                     void** states)
1253{
1254    struct r300_context* r300 = r300_context(pipe);
1255    struct r300_textures_state* state =
1256        (struct r300_textures_state*)r300->textures_state.state;
1257    unsigned tex_units = r300->screen->caps.num_tex_units;
1258
1259    if (count > tex_units) {
1260        return;
1261    }
1262
1263    memcpy(state->sampler_states, states, sizeof(void*) * count);
1264    state->sampler_state_count = count;
1265
1266    r300_mark_atom_dirty(r300, &r300->textures_state);
1267}
1268
1269static void r300_lacks_vertex_textures(struct pipe_context* pipe,
1270                                       unsigned count,
1271                                       void** states)
1272{
1273}
1274
1275static void r300_delete_sampler_state(struct pipe_context* pipe, void* state)
1276{
1277    FREE(state);
1278}
1279
1280static uint32_t r300_assign_texture_cache_region(unsigned index, unsigned num)
1281{
1282    /* This looks like a hack, but I believe it's suppose to work like
1283     * that. To illustrate how this works, let's assume you have 5 textures.
1284     * From docs, 5 and the successive numbers are:
1285     *
1286     * FOURTH_1     = 5
1287     * FOURTH_2     = 6
1288     * FOURTH_3     = 7
1289     * EIGHTH_0     = 8
1290     * EIGHTH_1     = 9
1291     *
1292     * First 3 textures will get 3/4 of size of the cache, divived evenly
1293     * between them. The last 1/4 of the cache must be divided between
1294     * the last 2 textures, each will therefore get 1/8 of the cache.
1295     * Why not just to use "5 + texture_index" ?
1296     *
1297     * This simple trick works for all "num" <= 16.
1298     */
1299    if (num <= 1)
1300        return R300_TX_CACHE(R300_TX_CACHE_WHOLE);
1301    else
1302        return R300_TX_CACHE(num + index);
1303}
1304
1305static void r300_set_fragment_sampler_views(struct pipe_context* pipe,
1306                                            unsigned count,
1307                                            struct pipe_sampler_view** views)
1308{
1309    struct r300_context* r300 = r300_context(pipe);
1310    struct r300_textures_state* state =
1311        (struct r300_textures_state*)r300->textures_state.state;
1312    struct r300_resource *texture;
1313    unsigned i, real_num_views = 0, view_index = 0;
1314    unsigned tex_units = r300->screen->caps.num_tex_units;
1315    boolean dirty_tex = FALSE;
1316
1317    if (count > tex_units) {
1318        return;
1319    }
1320
1321    /* Calculate the real number of views. */
1322    for (i = 0; i < count; i++) {
1323        if (views[i])
1324            real_num_views++;
1325    }
1326
1327    for (i = 0; i < count; i++) {
1328        pipe_sampler_view_reference(
1329                (struct pipe_sampler_view**)&state->sampler_views[i],
1330                views[i]);
1331
1332        if (!views[i]) {
1333            continue;
1334        }
1335
1336        /* A new sampler view (= texture)... */
1337        dirty_tex = TRUE;
1338
1339        /* Set the texrect factor in the fragment shader.
1340             * Needed for RECT and NPOT fallback. */
1341        texture = r300_resource(views[i]->texture);
1342        if (texture->tex.is_npot) {
1343            r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state);
1344        }
1345
1346        state->sampler_views[i]->texcache_region =
1347                r300_assign_texture_cache_region(view_index, real_num_views);
1348        view_index++;
1349    }
1350
1351    for (i = count; i < tex_units; i++) {
1352        if (state->sampler_views[i]) {
1353            pipe_sampler_view_reference(
1354                    (struct pipe_sampler_view**)&state->sampler_views[i],
1355                    NULL);
1356        }
1357    }
1358
1359    state->sampler_view_count = count;
1360
1361    r300_mark_atom_dirty(r300, &r300->textures_state);
1362    r300->validate_buffers = TRUE;
1363
1364    if (dirty_tex) {
1365        r300_mark_atom_dirty(r300, &r300->texture_cache_inval);
1366    }
1367}
1368
1369static struct pipe_sampler_view *
1370r300_create_sampler_view(struct pipe_context *pipe,
1371                         struct pipe_resource *texture,
1372                         const struct pipe_sampler_view *templ)
1373{
1374    struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view);
1375    struct r300_resource *tex = r300_resource(texture);
1376    boolean is_r500 = r300_screen(pipe->screen)->caps.is_r500;
1377    boolean dxtc_swizzle = r300_screen(pipe->screen)->caps.dxtc_swizzle;
1378
1379    if (view) {
1380        view->base = *templ;
1381        view->base.reference.count = 1;
1382        view->base.context = pipe;
1383        view->base.texture = NULL;
1384        pipe_resource_reference(&view->base.texture, texture);
1385
1386        view->swizzle[0] = templ->swizzle_r;
1387        view->swizzle[1] = templ->swizzle_g;
1388        view->swizzle[2] = templ->swizzle_b;
1389        view->swizzle[3] = templ->swizzle_a;
1390
1391        view->format = tex->tx_format;
1392        view->format.format1 |= r300_translate_texformat(templ->format,
1393                                                         view->swizzle,
1394                                                         is_r500,
1395                                                         dxtc_swizzle);
1396        if (is_r500) {
1397            view->format.format2 |= r500_tx_format_msb_bit(templ->format);
1398        }
1399    }
1400
1401    return (struct pipe_sampler_view*)view;
1402}
1403
1404static void
1405r300_sampler_view_destroy(struct pipe_context *pipe,
1406                          struct pipe_sampler_view *view)
1407{
1408   pipe_resource_reference(&view->texture, NULL);
1409   FREE(view);
1410}
1411
1412static void r300_set_scissor_state(struct pipe_context* pipe,
1413                                   const struct pipe_scissor_state* state)
1414{
1415    struct r300_context* r300 = r300_context(pipe);
1416
1417    memcpy(r300->scissor_state.state, state,
1418        sizeof(struct pipe_scissor_state));
1419
1420    r300_mark_atom_dirty(r300, &r300->scissor_state);
1421}
1422
1423static void r300_set_viewport_state(struct pipe_context* pipe,
1424                                    const struct pipe_viewport_state* state)
1425{
1426    struct r300_context* r300 = r300_context(pipe);
1427    struct r300_viewport_state* viewport =
1428        (struct r300_viewport_state*)r300->viewport_state.state;
1429
1430    r300->viewport = *state;
1431
1432    if (r300->draw) {
1433        draw_set_viewport_state(r300->draw, state);
1434        viewport->vte_control = R300_VTX_XY_FMT | R300_VTX_Z_FMT;
1435        return;
1436    }
1437
1438    /* Do the transform in HW. */
1439    viewport->vte_control = R300_VTX_W0_FMT;
1440
1441    if (state->scale[0] != 1.0f) {
1442        viewport->xscale = state->scale[0];
1443        viewport->vte_control |= R300_VPORT_X_SCALE_ENA;
1444    }
1445    if (state->scale[1] != 1.0f) {
1446        viewport->yscale = state->scale[1];
1447        viewport->vte_control |= R300_VPORT_Y_SCALE_ENA;
1448    }
1449    if (state->scale[2] != 1.0f) {
1450        viewport->zscale = state->scale[2];
1451        viewport->vte_control |= R300_VPORT_Z_SCALE_ENA;
1452    }
1453    if (state->translate[0] != 0.0f) {
1454        viewport->xoffset = state->translate[0];
1455        viewport->vte_control |= R300_VPORT_X_OFFSET_ENA;
1456    }
1457    if (state->translate[1] != 0.0f) {
1458        viewport->yoffset = state->translate[1];
1459        viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA;
1460    }
1461    if (state->translate[2] != 0.0f) {
1462        viewport->zoffset = state->translate[2];
1463        viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA;
1464    }
1465
1466    r300_mark_atom_dirty(r300, &r300->viewport_state);
1467    if (r300->fs.state && r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) {
1468        r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state);
1469    }
1470}
1471
1472static void r300_set_vertex_buffers(struct pipe_context* pipe,
1473                                    unsigned count,
1474                                    const struct pipe_vertex_buffer* buffers)
1475{
1476    struct r300_context* r300 = r300_context(pipe);
1477    unsigned i;
1478    struct pipe_vertex_buffer dummy_vb = {0};
1479
1480    /* There must be at least one vertex buffer set, otherwise it locks up. */
1481    if (!count) {
1482        dummy_vb.buffer = r300->dummy_vb;
1483        buffers = &dummy_vb;
1484        count = 1;
1485    }
1486
1487    u_vbuf_mgr_set_vertex_buffers(r300->vbuf_mgr, count, buffers);
1488
1489    if (r300->screen->caps.has_tcl) {
1490        /* HW TCL. */
1491        for (i = 0; i < count; i++) {
1492            if (buffers[i].buffer &&
1493		!r300_resource(buffers[i].buffer)->b.user_ptr) {
1494                r300->validate_buffers = TRUE;
1495            }
1496        }
1497        r300->vertex_arrays_dirty = TRUE;
1498    } else {
1499        /* SW TCL. */
1500        draw_set_vertex_buffers(r300->draw, count, buffers);
1501    }
1502}
1503
1504static void r300_set_index_buffer(struct pipe_context* pipe,
1505                                  const struct pipe_index_buffer *ib)
1506{
1507    struct r300_context* r300 = r300_context(pipe);
1508
1509    if (ib && ib->buffer) {
1510        assert(ib->offset % ib->index_size == 0);
1511
1512        pipe_resource_reference(&r300->index_buffer.buffer, ib->buffer);
1513        memcpy(&r300->index_buffer, ib, sizeof(r300->index_buffer));
1514        r300->index_buffer.offset /= r300->index_buffer.index_size;
1515
1516        if (r300->screen->caps.has_tcl &&
1517            !r300_resource(ib->buffer)->b.user_ptr) {
1518            r300->validate_buffers = TRUE;
1519            r300->upload_ib_validated = FALSE;
1520        }
1521    }
1522    else {
1523        pipe_resource_reference(&r300->index_buffer.buffer, NULL);
1524        memset(&r300->index_buffer, 0, sizeof(r300->index_buffer));
1525    }
1526
1527    if (!r300->screen->caps.has_tcl) {
1528        draw_set_index_buffer(r300->draw, ib);
1529    }
1530}
1531
1532/* Initialize the PSC tables. */
1533static void r300_vertex_psc(struct r300_vertex_element_state *velems)
1534{
1535    struct r300_vertex_stream_state *vstream = &velems->vertex_stream;
1536    uint16_t type, swizzle;
1537    enum pipe_format format;
1538    unsigned i;
1539
1540    if (velems->count > 16) {
1541        fprintf(stderr, "r300: More than 16 vertex elements are not supported,"
1542                " requested %i, using 16.\n", velems->count);
1543        velems->count = 16;
1544    }
1545
1546    /* Vertex shaders have no semantics on their inputs,
1547     * so PSC should just route stuff based on the vertex elements,
1548     * and not on attrib information. */
1549    for (i = 0; i < velems->count; i++) {
1550        format = velems->velem[i].src_format;
1551
1552        type = r300_translate_vertex_data_type(format);
1553        if (type == R300_INVALID_FORMAT) {
1554            fprintf(stderr, "r300: Bad vertex format %s.\n",
1555                    util_format_short_name(format));
1556            assert(0);
1557            abort();
1558        }
1559
1560        type |= i << R300_DST_VEC_LOC_SHIFT;
1561        swizzle = r300_translate_vertex_data_swizzle(format);
1562
1563        if (i & 1) {
1564            vstream->vap_prog_stream_cntl[i >> 1] |= type << 16;
1565            vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16;
1566        } else {
1567            vstream->vap_prog_stream_cntl[i >> 1] |= type;
1568            vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
1569        }
1570    }
1571
1572    /* Set the last vector in the PSC. */
1573    if (i) {
1574        i -= 1;
1575    }
1576    vstream->vap_prog_stream_cntl[i >> 1] |=
1577        (R300_LAST_VEC << (i & 1 ? 16 : 0));
1578
1579    vstream->count = (i >> 1) + 1;
1580}
1581
1582static void* r300_create_vertex_elements_state(struct pipe_context* pipe,
1583                                               unsigned count,
1584                                               const struct pipe_vertex_element* attribs)
1585{
1586    struct r300_context *r300 = r300_context(pipe);
1587    struct r300_vertex_element_state *velems;
1588    unsigned i;
1589    struct pipe_vertex_element dummy_attrib = {0};
1590
1591    /* R300 Programmable Stream Control (PSC) doesn't support 0 vertex elements. */
1592    if (!count) {
1593        dummy_attrib.src_format = PIPE_FORMAT_R8G8B8A8_UNORM;
1594        attribs = &dummy_attrib;
1595        count = 1;
1596    }
1597
1598    assert(count <= PIPE_MAX_ATTRIBS);
1599    velems = CALLOC_STRUCT(r300_vertex_element_state);
1600    if (!velems)
1601        return NULL;
1602
1603    velems->count = count;
1604    velems->vmgr_elements =
1605        u_vbuf_mgr_create_vertex_elements(r300->vbuf_mgr, count, attribs,
1606                                          velems->velem);
1607
1608    if (r300_screen(pipe->screen)->caps.has_tcl) {
1609        /* Setup PSC.
1610         * The unused components will be replaced by (..., 0, 1). */
1611        r300_vertex_psc(velems);
1612
1613        for (i = 0; i < count; i++) {
1614            velems->format_size[i] =
1615                align(util_format_get_blocksize(velems->velem[i].src_format), 4);
1616            velems->vertex_size_dwords += velems->format_size[i] / 4;
1617        }
1618    }
1619
1620    return velems;
1621}
1622
1623static void r300_bind_vertex_elements_state(struct pipe_context *pipe,
1624                                            void *state)
1625{
1626    struct r300_context *r300 = r300_context(pipe);
1627    struct r300_vertex_element_state *velems = state;
1628
1629    if (velems == NULL) {
1630        return;
1631    }
1632
1633    r300->velems = velems;
1634
1635    u_vbuf_mgr_bind_vertex_elements(r300->vbuf_mgr, state, velems->vmgr_elements);
1636
1637    if (r300->draw) {
1638        draw_set_vertex_elements(r300->draw, velems->count, velems->velem);
1639        return;
1640    }
1641
1642    UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state);
1643    r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2;
1644    r300->vertex_arrays_dirty = TRUE;
1645}
1646
1647static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state)
1648{
1649    struct r300_context *r300 = r300_context(pipe);
1650    struct r300_vertex_element_state *velems = state;
1651
1652    u_vbuf_mgr_destroy_vertex_elements(r300->vbuf_mgr, velems->vmgr_elements);
1653    FREE(state);
1654}
1655
1656static void* r300_create_vs_state(struct pipe_context* pipe,
1657                                  const struct pipe_shader_state* shader)
1658{
1659    struct r300_context* r300 = r300_context(pipe);
1660    struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader);
1661
1662    /* Copy state directly into shader. */
1663    vs->state = *shader;
1664    vs->state.tokens = tgsi_dup_tokens(shader->tokens);
1665
1666    if (r300->screen->caps.has_tcl) {
1667        r300_init_vs_outputs(vs);
1668        r300_translate_vertex_shader(r300, vs);
1669    } else {
1670        r300_draw_init_vertex_shader(r300->draw, vs);
1671    }
1672
1673    return vs;
1674}
1675
1676static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
1677{
1678    struct r300_context* r300 = r300_context(pipe);
1679    struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
1680
1681    if (vs == NULL) {
1682        r300->vs_state.state = NULL;
1683        return;
1684    }
1685    if (vs == r300->vs_state.state) {
1686        return;
1687    }
1688    r300->vs_state.state = vs;
1689
1690    /* The majority of the RS block bits is dependent on the vertex shader. */
1691    r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */
1692
1693    if (r300->screen->caps.has_tcl) {
1694        unsigned fc_op_dwords = r300->screen->caps.is_r500 ? 3 : 2;
1695        r300_mark_atom_dirty(r300, &r300->vs_state);
1696        r300->vs_state.size =
1697                vs->code.length + 9 +
1698        (vs->code.num_fc_ops ? vs->code.num_fc_ops * fc_op_dwords + 4 : 0);
1699
1700        r300_mark_atom_dirty(r300, &r300->vs_constants);
1701        r300->vs_constants.size =
1702                2 +
1703                (vs->externals_count ? vs->externals_count * 4 + 3 : 0) +
1704                (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0);
1705
1706        ((struct r300_constant_buffer*)r300->vs_constants.state)->remap_table =
1707                vs->code.constants_remap_table;
1708
1709        r300_mark_atom_dirty(r300, &r300->pvs_flush);
1710    } else {
1711        draw_bind_vertex_shader(r300->draw,
1712                (struct draw_vertex_shader*)vs->draw_vs);
1713    }
1714}
1715
1716static void r300_delete_vs_state(struct pipe_context* pipe, void* shader)
1717{
1718    struct r300_context* r300 = r300_context(pipe);
1719    struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
1720
1721    if (r300->screen->caps.has_tcl) {
1722        rc_constants_destroy(&vs->code.constants);
1723        if (vs->code.constants_remap_table)
1724            FREE(vs->code.constants_remap_table);
1725    } else {
1726        draw_delete_vertex_shader(r300->draw,
1727                (struct draw_vertex_shader*)vs->draw_vs);
1728    }
1729
1730    FREE((void*)vs->state.tokens);
1731    FREE(shader);
1732}
1733
1734static void r300_set_constant_buffer(struct pipe_context *pipe,
1735                                     uint shader, uint index,
1736                                     struct pipe_resource *buf)
1737{
1738    struct r300_context* r300 = r300_context(pipe);
1739    struct r300_constant_buffer *cbuf;
1740    struct r300_resource *rbuf = r300_resource(buf);
1741    uint32_t *mapped;
1742
1743    switch (shader) {
1744        case PIPE_SHADER_VERTEX:
1745            cbuf = (struct r300_constant_buffer*)r300->vs_constants.state;
1746            break;
1747        case PIPE_SHADER_FRAGMENT:
1748            cbuf = (struct r300_constant_buffer*)r300->fs_constants.state;
1749            break;
1750        default:
1751            return;
1752    }
1753
1754    if (buf == NULL || buf->width0 == 0)
1755        return;
1756
1757    if (rbuf->b.user_ptr)
1758        mapped = (uint32_t*)rbuf->b.user_ptr;
1759    else if (rbuf->constant_buffer)
1760        mapped = (uint32_t*)rbuf->constant_buffer;
1761    else
1762        return;
1763
1764    if (shader == PIPE_SHADER_FRAGMENT ||
1765        (shader == PIPE_SHADER_VERTEX && r300->screen->caps.has_tcl)) {
1766        cbuf->ptr = mapped;
1767    }
1768
1769    if (shader == PIPE_SHADER_VERTEX) {
1770        if (r300->screen->caps.has_tcl) {
1771            struct r300_vertex_shader *vs =
1772                    (struct r300_vertex_shader*)r300->vs_state.state;
1773
1774            if (!vs) {
1775                cbuf->buffer_base = 0;
1776                return;
1777            }
1778
1779            cbuf->buffer_base = r300->vs_const_base;
1780            r300->vs_const_base += vs->code.constants.Count;
1781            if (r300->vs_const_base > R500_MAX_PVS_CONST_VECS) {
1782                r300->vs_const_base = vs->code.constants.Count;
1783                cbuf->buffer_base = 0;
1784                r300_mark_atom_dirty(r300, &r300->pvs_flush);
1785            }
1786            r300_mark_atom_dirty(r300, &r300->vs_constants);
1787        } else if (r300->draw) {
1788            draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX,
1789                0, mapped, buf->width0);
1790        }
1791    } else if (shader == PIPE_SHADER_FRAGMENT) {
1792        r300_mark_atom_dirty(r300, &r300->fs_constants);
1793    }
1794}
1795
1796void r300_init_state_functions(struct r300_context* r300)
1797{
1798    r300->context.create_blend_state = r300_create_blend_state;
1799    r300->context.bind_blend_state = r300_bind_blend_state;
1800    r300->context.delete_blend_state = r300_delete_blend_state;
1801
1802    r300->context.set_blend_color = r300_set_blend_color;
1803
1804    r300->context.set_clip_state = r300_set_clip_state;
1805    r300->context.set_sample_mask = r300_set_sample_mask;
1806
1807    r300->context.set_constant_buffer = r300_set_constant_buffer;
1808
1809    r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state;
1810    r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state;
1811    r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state;
1812
1813    r300->context.set_stencil_ref = r300_set_stencil_ref;
1814
1815    r300->context.set_framebuffer_state = r300_set_framebuffer_state;
1816
1817    r300->context.create_fs_state = r300_create_fs_state;
1818    r300->context.bind_fs_state = r300_bind_fs_state;
1819    r300->context.delete_fs_state = r300_delete_fs_state;
1820
1821    r300->context.set_polygon_stipple = r300_set_polygon_stipple;
1822
1823    r300->context.create_rasterizer_state = r300_create_rs_state;
1824    r300->context.bind_rasterizer_state = r300_bind_rs_state;
1825    r300->context.delete_rasterizer_state = r300_delete_rs_state;
1826
1827    r300->context.create_sampler_state = r300_create_sampler_state;
1828    r300->context.bind_fragment_sampler_states = r300_bind_sampler_states;
1829    r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures;
1830    r300->context.delete_sampler_state = r300_delete_sampler_state;
1831
1832    r300->context.set_fragment_sampler_views = r300_set_fragment_sampler_views;
1833    r300->context.create_sampler_view = r300_create_sampler_view;
1834    r300->context.sampler_view_destroy = r300_sampler_view_destroy;
1835
1836    r300->context.set_scissor_state = r300_set_scissor_state;
1837
1838    r300->context.set_viewport_state = r300_set_viewport_state;
1839
1840    r300->context.set_vertex_buffers = r300_set_vertex_buffers;
1841    r300->context.set_index_buffer = r300_set_index_buffer;
1842    r300->context.redefine_user_buffer = u_default_redefine_user_buffer;
1843
1844    r300->context.create_vertex_elements_state = r300_create_vertex_elements_state;
1845    r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state;
1846    r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state;
1847
1848    r300->context.create_vs_state = r300_create_vs_state;
1849    r300->context.bind_vs_state = r300_bind_vs_state;
1850    r300->context.delete_vs_state = r300_delete_vs_state;
1851}
1852