r300_state.c revision 45d79af644349882b4990b775737405be7cd1aa2
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_math.h"
27#include "util/u_memory.h"
28#include "util/u_pack_color.h"
29
30#include "tgsi/tgsi_parse.h"
31
32#include "pipe/p_config.h"
33
34#include "r300_context.h"
35#include "r300_emit.h"
36#include "r300_reg.h"
37#include "r300_screen.h"
38#include "r300_screen_buffer.h"
39#include "r300_state.h"
40#include "r300_state_inlines.h"
41#include "r300_fs.h"
42#include "r300_texture.h"
43#include "r300_vs.h"
44#include "r300_winsys.h"
45
46/* r300_state: Functions used to intialize state context by translating
47 * Gallium state objects into semi-native r300 state objects. */
48
49#define UPDATE_STATE(cso, atom) \
50    if (cso != atom.state) { \
51        atom.state = cso;    \
52        atom.dirty = TRUE;   \
53    }
54
55static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA,
56                                            unsigned dstRGB, unsigned dstA)
57{
58    /* If the blend equation is ADD or REVERSE_SUBTRACT,
59     * SRC_ALPHA == 0, and the following state is set, the colorbuffer
60     * will not be changed.
61     * Notice that the dst factors are the src factors inverted. */
62    return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
63            srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
64            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
65           (srcA == PIPE_BLENDFACTOR_SRC_COLOR ||
66            srcA == PIPE_BLENDFACTOR_SRC_ALPHA ||
67            srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
68            srcA == PIPE_BLENDFACTOR_ZERO) &&
69           (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
70            dstRGB == PIPE_BLENDFACTOR_ONE) &&
71           (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
72            dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
73            dstA == PIPE_BLENDFACTOR_ONE);
74}
75
76static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA,
77                                            unsigned dstRGB, unsigned dstA)
78{
79    /* If the blend equation is ADD or REVERSE_SUBTRACT,
80     * SRC_ALPHA == 1, and the following state is set, the colorbuffer
81     * will not be changed.
82     * Notice that the dst factors are the src factors inverted. */
83    return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
84            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
85           (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
86            srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
87            srcA == PIPE_BLENDFACTOR_ZERO) &&
88           (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
89            dstRGB == PIPE_BLENDFACTOR_ONE) &&
90           (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
91            dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
92            dstA == PIPE_BLENDFACTOR_ONE);
93}
94
95static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA,
96                                            unsigned dstRGB, unsigned dstA)
97{
98    /* If the blend equation is ADD or REVERSE_SUBTRACT,
99     * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer
100     * will not be changed.
101     * Notice that the dst factors are the src factors inverted. */
102    return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
103            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
104           (srcA == PIPE_BLENDFACTOR_ZERO) &&
105           (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
106            dstRGB == PIPE_BLENDFACTOR_ONE) &&
107           (dstA == PIPE_BLENDFACTOR_ONE);
108}
109
110static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA,
111                                            unsigned dstRGB, unsigned dstA)
112{
113    /* If the blend equation is ADD or REVERSE_SUBTRACT,
114     * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer
115     * will not be changed.
116     * Notice that the dst factors are the src factors inverted. */
117    return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
118            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
119           (srcA == PIPE_BLENDFACTOR_ZERO) &&
120           (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
121            dstRGB == PIPE_BLENDFACTOR_ONE) &&
122           (dstA == PIPE_BLENDFACTOR_ONE);
123}
124
125static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA,
126                                                  unsigned dstRGB, unsigned dstA)
127{
128    /* If the blend equation is ADD or REVERSE_SUBTRACT,
129     * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set,
130     * the colorbuffer will not be changed.
131     * Notice that the dst factors are the src factors inverted. */
132    return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
133            srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
134            srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
135            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
136           (srcA == PIPE_BLENDFACTOR_SRC_COLOR ||
137            srcA == PIPE_BLENDFACTOR_SRC_ALPHA ||
138            srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
139            srcA == PIPE_BLENDFACTOR_ZERO) &&
140           (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
141            dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
142            dstRGB == PIPE_BLENDFACTOR_ONE) &&
143           (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
144            dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
145            dstA == PIPE_BLENDFACTOR_ONE);
146}
147
148static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA,
149                                                  unsigned dstRGB, unsigned dstA)
150{
151    /* If the blend equation is ADD or REVERSE_SUBTRACT,
152     * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set,
153     * the colorbuffer will not be changed.
154     * Notice that the dst factors are the src factors inverted. */
155    return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
156            srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
157            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
158           (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
159            srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
160            srcA == PIPE_BLENDFACTOR_ZERO) &&
161           (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
162            dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
163            dstRGB == PIPE_BLENDFACTOR_ONE) &&
164           (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
165            dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
166            dstA == PIPE_BLENDFACTOR_ONE);
167}
168
169static unsigned bgra_cmask(unsigned mask)
170{
171    /* Gallium uses RGBA color ordering while R300 expects BGRA. */
172
173    return ((mask & PIPE_MASK_R) << 2) |
174           ((mask & PIPE_MASK_B) >> 2) |
175           (mask & (PIPE_MASK_G | PIPE_MASK_A));
176}
177
178/* Create a new blend state based on the CSO blend state.
179 *
180 * This encompasses alpha blending, logic/raster ops, and blend dithering. */
181static void* r300_create_blend_state(struct pipe_context* pipe,
182                                     const struct pipe_blend_state* state)
183{
184    struct r300_screen* r300screen = r300_screen(pipe->screen);
185    struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state);
186
187    if (state->rt[0].blend_enable)
188    {
189        unsigned eqRGB = state->rt[0].rgb_func;
190        unsigned srcRGB = state->rt[0].rgb_src_factor;
191        unsigned dstRGB = state->rt[0].rgb_dst_factor;
192
193        unsigned eqA = state->rt[0].alpha_func;
194        unsigned srcA = state->rt[0].alpha_src_factor;
195        unsigned dstA = state->rt[0].alpha_dst_factor;
196
197        /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha,
198         * this is just the crappy D3D naming */
199        blend->blend_control = R300_ALPHA_BLEND_ENABLE |
200            r300_translate_blend_function(eqRGB) |
201            ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) |
202            ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT);
203
204        /* Optimization: some operations do not require the destination color.
205         *
206         * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled,
207         * otherwise blending gives incorrect results. It seems to be
208         * a hardware bug. */
209        if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN ||
210            eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX ||
211            dstRGB != PIPE_BLENDFACTOR_ZERO ||
212            dstA != PIPE_BLENDFACTOR_ZERO ||
213            srcRGB == PIPE_BLENDFACTOR_DST_COLOR ||
214            srcRGB == PIPE_BLENDFACTOR_DST_ALPHA ||
215            srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR ||
216            srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
217            srcA == PIPE_BLENDFACTOR_DST_COLOR ||
218            srcA == PIPE_BLENDFACTOR_DST_ALPHA ||
219            srcA == PIPE_BLENDFACTOR_INV_DST_COLOR ||
220            srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
221            srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) {
222            /* Enable reading from the colorbuffer. */
223            blend->blend_control |= R300_READ_ENABLE;
224
225            if (r300screen->caps.is_r500) {
226                /* Optimization: Depending on incoming pixels, we can
227                 * conditionally disable the reading in hardware... */
228                if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN &&
229                    eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) {
230                    /* Disable reading if SRC_ALPHA == 0. */
231                    if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
232                         dstRGB == PIPE_BLENDFACTOR_ZERO) &&
233                        (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
234                         dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
235                         dstA == PIPE_BLENDFACTOR_ZERO)) {
236                         blend->blend_control |= R500_SRC_ALPHA_0_NO_READ;
237                    }
238
239                    /* Disable reading if SRC_ALPHA == 1. */
240                    if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
241                         dstRGB == PIPE_BLENDFACTOR_ZERO) &&
242                        (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
243                         dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
244                         dstA == PIPE_BLENDFACTOR_ZERO)) {
245                         blend->blend_control |= R500_SRC_ALPHA_1_NO_READ;
246                    }
247                }
248            }
249        }
250
251        /* Optimization: discard pixels which don't change the colorbuffer.
252         *
253         * The code below is non-trivial and some math is involved.
254         *
255         * Discarding pixels must be disabled when FP16 AA is enabled.
256         * This is a hardware bug. Also, this implementation wouldn't work
257         * with FP blending enabled and equation clamping disabled.
258         *
259         * Equations other than ADD are rarely used and therefore won't be
260         * optimized. */
261        if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) &&
262            (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) {
263            /* ADD: X+Y
264             * REVERSE_SUBTRACT: Y-X
265             *
266             * The idea is:
267             * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1,
268             * then CB will not be changed.
269             *
270             * Given the srcFactor and dstFactor variables, we can derive
271             * what src and dst should be equal to and discard appropriate
272             * pixels.
273             */
274            if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) {
275                blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0;
276            } else if (blend_discard_if_src_alpha_1(srcRGB, srcA,
277                                                    dstRGB, dstA)) {
278                blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1;
279            } else if (blend_discard_if_src_color_0(srcRGB, srcA,
280                                                    dstRGB, dstA)) {
281                blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0;
282            } else if (blend_discard_if_src_color_1(srcRGB, srcA,
283                                                    dstRGB, dstA)) {
284                blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1;
285            } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA,
286                                                          dstRGB, dstA)) {
287                blend->blend_control |=
288                    R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0;
289            } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA,
290                                                          dstRGB, dstA)) {
291                blend->blend_control |=
292                    R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1;
293            }
294        }
295
296        /* separate alpha */
297        if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) {
298            blend->blend_control |= R300_SEPARATE_ALPHA_ENABLE;
299            blend->alpha_blend_control =
300                r300_translate_blend_function(eqA) |
301                (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) |
302                (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT);
303        }
304    }
305
306    /* PIPE_LOGICOP_* don't need to be translated, fortunately. */
307    if (state->logicop_enable) {
308        blend->rop = R300_RB3D_ROPCNTL_ROP_ENABLE |
309                (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT;
310    }
311
312    /* Color channel masks for all MRTs. */
313    blend->color_channel_mask = bgra_cmask(state->rt[0].colormask);
314    if (r300screen->caps.is_r500 && state->independent_blend_enable) {
315        if (state->rt[1].blend_enable) {
316            blend->color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4;
317        }
318        if (state->rt[2].blend_enable) {
319            blend->color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8;
320        }
321        if (state->rt[3].blend_enable) {
322            blend->color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12;
323        }
324    }
325
326    /* Neither fglrx nor classic r300 ever set this, regardless of dithering
327     * state. Since it's an optional implementation detail, we can leave it
328     * out and never dither.
329     *
330     * This could be revisited if we ever get quality or conformance hints.
331     *
332    if (state->dither) {
333        blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT |
334                        R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT;
335    }
336    */
337
338    return (void*)blend;
339}
340
341/* Bind blend state. */
342static void r300_bind_blend_state(struct pipe_context* pipe,
343                                  void* state)
344{
345    struct r300_context* r300 = r300_context(pipe);
346
347    UPDATE_STATE(state, r300->blend_state);
348}
349
350/* Free blend state. */
351static void r300_delete_blend_state(struct pipe_context* pipe,
352                                    void* state)
353{
354    FREE(state);
355}
356
357/* Convert float to 10bit integer */
358static unsigned float_to_fixed10(float f)
359{
360    return CLAMP((unsigned)(f * 1023.9f), 0, 1023);
361}
362
363/* Set blend color.
364 * Setup both R300 and R500 registers, figure out later which one to write. */
365static void r300_set_blend_color(struct pipe_context* pipe,
366                                 const struct pipe_blend_color* color)
367{
368    struct r300_context* r300 = r300_context(pipe);
369    struct r300_blend_color_state* state =
370        (struct r300_blend_color_state*)r300->blend_color_state.state;
371    union util_color uc;
372
373    util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
374    state->blend_color = uc.ui;
375
376    /* XXX if FP16 blending is enabled, we should use the FP16 format */
377    state->blend_color_red_alpha =
378        float_to_fixed10(color->color[0]) |
379        (float_to_fixed10(color->color[3]) << 16);
380    state->blend_color_green_blue =
381        float_to_fixed10(color->color[2]) |
382        (float_to_fixed10(color->color[1]) << 16);
383
384    r300->blend_color_state.size = r300->screen->caps.is_r500 ? 3 : 2;
385    r300->blend_color_state.dirty = TRUE;
386}
387
388static void r300_set_clip_state(struct pipe_context* pipe,
389                                const struct pipe_clip_state* state)
390{
391    struct r300_context* r300 = r300_context(pipe);
392
393    r300->clip = *state;
394
395    if (r300->screen->caps.has_tcl) {
396        memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state));
397        r300->clip_state.size = 29;
398    } else {
399        draw_flush(r300->draw);
400        draw_set_clip_state(r300->draw, state);
401        r300->clip_state.size = 2;
402    }
403
404    r300->clip_state.dirty = TRUE;
405}
406
407/* Create a new depth, stencil, and alpha state based on the CSO dsa state.
408 *
409 * This contains the depth buffer, stencil buffer, alpha test, and such.
410 * On the Radeon, depth and stencil buffer setup are intertwined, which is
411 * the reason for some of the strange-looking assignments across registers. */
412static void*
413        r300_create_dsa_state(struct pipe_context* pipe,
414                              const struct pipe_depth_stencil_alpha_state* state)
415{
416    struct r300_capabilities *caps = &r300_screen(pipe->screen)->caps;
417    struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state);
418
419    /* Depth test setup. */
420    if (state->depth.enabled) {
421        dsa->z_buffer_control |= R300_Z_ENABLE;
422
423        if (state->depth.writemask) {
424            dsa->z_buffer_control |= R300_Z_WRITE_ENABLE;
425        }
426
427        dsa->z_stencil_control |=
428            (r300_translate_depth_stencil_function(state->depth.func) <<
429                R300_Z_FUNC_SHIFT);
430    }
431
432    /* Stencil buffer setup. */
433    if (state->stencil[0].enabled) {
434        dsa->z_buffer_control |= R300_STENCIL_ENABLE;
435        dsa->z_stencil_control |=
436            (r300_translate_depth_stencil_function(state->stencil[0].func) <<
437                R300_S_FRONT_FUNC_SHIFT) |
438            (r300_translate_stencil_op(state->stencil[0].fail_op) <<
439                R300_S_FRONT_SFAIL_OP_SHIFT) |
440            (r300_translate_stencil_op(state->stencil[0].zpass_op) <<
441                R300_S_FRONT_ZPASS_OP_SHIFT) |
442            (r300_translate_stencil_op(state->stencil[0].zfail_op) <<
443                R300_S_FRONT_ZFAIL_OP_SHIFT);
444
445        dsa->stencil_ref_mask =
446                (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) |
447                (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT);
448
449        if (state->stencil[1].enabled) {
450            dsa->two_sided = TRUE;
451
452            dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK;
453            dsa->z_stencil_control |=
454            (r300_translate_depth_stencil_function(state->stencil[1].func) <<
455                R300_S_BACK_FUNC_SHIFT) |
456            (r300_translate_stencil_op(state->stencil[1].fail_op) <<
457                R300_S_BACK_SFAIL_OP_SHIFT) |
458            (r300_translate_stencil_op(state->stencil[1].zpass_op) <<
459                R300_S_BACK_ZPASS_OP_SHIFT) |
460            (r300_translate_stencil_op(state->stencil[1].zfail_op) <<
461                R300_S_BACK_ZFAIL_OP_SHIFT);
462
463            dsa->stencil_ref_bf =
464                (state->stencil[1].valuemask << R300_STENCILMASK_SHIFT) |
465                (state->stencil[1].writemask << R300_STENCILWRITEMASK_SHIFT);
466
467            if (caps->is_r500) {
468                dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK;
469            } else {
470                dsa->stencil_ref_bf_fallback =
471                  (state->stencil[0].valuemask != state->stencil[1].valuemask ||
472                   state->stencil[0].writemask != state->stencil[1].writemask);
473            }
474        }
475    }
476
477    /* Alpha test setup. */
478    if (state->alpha.enabled) {
479        dsa->alpha_function =
480            r300_translate_alpha_function(state->alpha.func) |
481            R300_FG_ALPHA_FUNC_ENABLE;
482
483        /* We could use 10bit alpha ref but who needs that? */
484        dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value);
485
486        if (caps->is_r500)
487            dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT;
488    }
489
490    return (void*)dsa;
491}
492
493static void r300_update_stencil_ref_fallback_status(struct r300_context *r300)
494{
495    struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state;
496
497    if (r300->screen->caps.is_r500) {
498        return;
499    }
500
501    r300->stencil_ref_bf_fallback =
502        dsa->stencil_ref_bf_fallback ||
503        (dsa->two_sided &&
504         r300->stencil_ref.ref_value[0] != r300->stencil_ref.ref_value[1]);
505}
506
507/* Bind DSA state. */
508static void r300_bind_dsa_state(struct pipe_context* pipe,
509                                void* state)
510{
511    struct r300_context* r300 = r300_context(pipe);
512
513    if (!state) {
514        return;
515    }
516
517    UPDATE_STATE(state, r300->dsa_state);
518
519    r300_update_stencil_ref_fallback_status(r300);
520}
521
522/* Free DSA state. */
523static void r300_delete_dsa_state(struct pipe_context* pipe,
524                                  void* state)
525{
526    FREE(state);
527}
528
529static void r300_set_stencil_ref(struct pipe_context* pipe,
530                                 const struct pipe_stencil_ref* sr)
531{
532    struct r300_context* r300 = r300_context(pipe);
533
534    r300->stencil_ref = *sr;
535    r300->dsa_state.dirty = TRUE;
536
537    r300_update_stencil_ref_fallback_status(r300);
538}
539
540/* This switcheroo is needed just because of goddamned MACRO_SWITCH. */
541static void r300_fb_set_tiling_flags(struct r300_context *r300,
542                               const struct pipe_framebuffer_state *old_state,
543                               const struct pipe_framebuffer_state *new_state)
544{
545    struct r300_texture *tex;
546    unsigned i, level;
547
548    /* Set tiling flags for new surfaces. */
549    for (i = 0; i < new_state->nr_cbufs; i++) {
550        tex = r300_texture(new_state->cbufs[i]->texture);
551        level = new_state->cbufs[i]->level;
552
553        r300->rws->buffer_set_tiling(r300->rws, tex->buffer,
554                                        tex->pitch[level],
555                                        tex->microtile,
556                                        tex->mip_macrotile[level]);
557    }
558    if (new_state->zsbuf) {
559        tex = r300_texture(new_state->zsbuf->texture);
560        level = new_state->zsbuf->level;
561
562        r300->rws->buffer_set_tiling(r300->rws, tex->buffer,
563                                        tex->pitch[level],
564                                        tex->microtile,
565                                        tex->mip_macrotile[level]);
566    }
567}
568
569static void
570    r300_set_framebuffer_state(struct pipe_context* pipe,
571                               const struct pipe_framebuffer_state* state)
572{
573    struct r300_context* r300 = r300_context(pipe);
574    struct pipe_framebuffer_state *old_state = r300->fb_state.state;
575    unsigned max_width, max_height;
576    uint32_t zbuffer_bpp = 0;
577
578    if (state->nr_cbufs > 4) {
579        fprintf(stderr, "r300: Implementation error: Too many MRTs in %s, "
580            "refusing to bind framebuffer state!\n", __FUNCTION__);
581        return;
582    }
583
584    if (r300->screen->caps.is_r500) {
585        max_width = max_height = 4096;
586    } else if (r300->screen->caps.is_r400) {
587        max_width = max_height = 4021;
588    } else {
589        max_width = max_height = 2560;
590    }
591
592    if (state->width > max_width || state->height > max_height) {
593        fprintf(stderr, "r300: Implementation error: Render targets are too "
594        "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__);
595        return;
596    }
597
598    if (r300->draw) {
599        draw_flush(r300->draw);
600    }
601
602    r300->fb_state.dirty = TRUE;
603
604    /* If nr_cbufs is changed from zero to non-zero or vice versa... */
605    if (!!old_state->nr_cbufs != !!state->nr_cbufs) {
606        r300->blend_state.dirty = TRUE;
607    }
608    /* If zsbuf is set from NULL to non-NULL or vice versa.. */
609    if (!!old_state->zsbuf != !!state->zsbuf) {
610        r300->dsa_state.dirty = TRUE;
611    }
612
613    /* The tiling flags are dependent on the surface miplevel, unfortunately. */
614    r300_fb_set_tiling_flags(r300, r300->fb_state.state, state);
615
616    memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state));
617
618    r300->fb_state.size = (10 * state->nr_cbufs) + (2 * (4 - state->nr_cbufs)) +
619                          (state->zsbuf ? 10 : 0) + 11;
620
621    /* Polygon offset depends on the zbuffer bit depth. */
622    if (state->zsbuf && r300->polygon_offset_enabled) {
623        switch (util_format_get_blocksize(state->zsbuf->texture->format)) {
624            case 2:
625                zbuffer_bpp = 16;
626                break;
627            case 4:
628                zbuffer_bpp = 24;
629                break;
630        }
631
632        if (r300->zbuffer_bpp != zbuffer_bpp) {
633            r300->zbuffer_bpp = zbuffer_bpp;
634            r300->rs_state.dirty = TRUE;
635        }
636    }
637}
638
639/* Create fragment shader state. */
640static void* r300_create_fs_state(struct pipe_context* pipe,
641                                  const struct pipe_shader_state* shader)
642{
643    struct r300_fragment_shader* fs = NULL;
644
645    fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader);
646
647    /* Copy state directly into shader. */
648    fs->state = *shader;
649    fs->state.tokens = tgsi_dup_tokens(shader->tokens);
650
651    return (void*)fs;
652}
653
654void r300_mark_fs_code_dirty(struct r300_context *r300)
655{
656    struct r300_fragment_shader* fs = r300_fs(r300);
657
658    r300->fs.dirty = TRUE;
659    r300->fs_rc_constant_state.dirty = TRUE;
660    r300->fs_constants.dirty = TRUE;
661
662    if (r300->screen->caps.is_r500) {
663        r300->fs.size = r500_get_fs_atom_size(r300);
664        r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 7;
665        r300->fs_constants.size = fs->shader->externals_count * 4 + 3;
666    } else {
667        r300->fs.size = r300_get_fs_atom_size(r300);
668        r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 5;
669        r300->fs_constants.size = fs->shader->externals_count * 4 + 1;
670    }
671}
672
673/* Bind fragment shader state. */
674static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
675{
676    struct r300_context* r300 = r300_context(pipe);
677    struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
678
679    if (fs == NULL) {
680        r300->fs.state = NULL;
681        return;
682    }
683
684    r300->fs.state = fs;
685    r300_pick_fragment_shader(r300);
686    r300_mark_fs_code_dirty(r300);
687
688    r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
689
690    if (r300->vs_state.state && r300_vertex_shader_setup_wpos(r300)) {
691        r300->vap_output_state.dirty = TRUE;
692    }
693}
694
695/* Delete fragment shader state. */
696static void r300_delete_fs_state(struct pipe_context* pipe, void* shader)
697{
698    struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
699    struct r300_fragment_shader_code *tmp, *ptr = fs->first;
700
701    while (ptr) {
702        tmp = ptr;
703        ptr = ptr->next;
704        rc_constants_destroy(&tmp->code.constants);
705        FREE(tmp);
706    }
707    FREE((void*)fs->state.tokens);
708    FREE(shader);
709}
710
711static void r300_set_polygon_stipple(struct pipe_context* pipe,
712                                     const struct pipe_poly_stipple* state)
713{
714    /* XXX no idea how to set this up, but not terribly important */
715}
716
717/* Create a new rasterizer state based on the CSO rasterizer state.
718 *
719 * This is a very large chunk of state, and covers most of the graphics
720 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
721 *
722 * In a not entirely unironic sidenote, this state has nearly nothing to do
723 * with the actual block on the Radeon called the rasterizer (RS). */
724static void* r300_create_rs_state(struct pipe_context* pipe,
725                                  const struct pipe_rasterizer_state* state)
726{
727    struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state);
728    int i;
729
730    /* Copy rasterizer state for Draw. */
731    rs->rs = *state;
732
733#ifdef PIPE_ARCH_LITTLE_ENDIAN
734    rs->vap_control_status = R300_VC_NO_SWAP;
735#else
736    rs->vap_control_status = R300_VC_32BIT_SWAP;
737#endif
738
739    /* If no TCL engine is present, turn off the HW TCL. */
740    if (!r300_screen(pipe->screen)->caps.has_tcl) {
741        rs->vap_control_status |= R300_VAP_TCL_BYPASS;
742    }
743
744    rs->point_size = pack_float_16_6x(state->point_size) |
745        (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT);
746
747    rs->line_control = pack_float_16_6x(state->line_width) |
748        R300_GA_LINE_CNTL_END_TYPE_COMP;
749
750    /* Enable polygon mode */
751    if (state->fill_cw != PIPE_POLYGON_MODE_FILL ||
752        state->fill_ccw != PIPE_POLYGON_MODE_FILL) {
753        rs->polygon_mode = R300_GA_POLY_MODE_DUAL;
754    }
755
756    /* Radeons don't think in "CW/CCW", they think in "front/back". */
757    if (state->front_winding == PIPE_WINDING_CW) {
758        rs->cull_mode = R300_FRONT_FACE_CW;
759
760        /* Polygon offset */
761        if (state->offset_cw) {
762            rs->polygon_offset_enable |= R300_FRONT_ENABLE;
763        }
764        if (state->offset_ccw) {
765            rs->polygon_offset_enable |= R300_BACK_ENABLE;
766        }
767
768        /* Polygon mode */
769        if (rs->polygon_mode) {
770            rs->polygon_mode |=
771                r300_translate_polygon_mode_front(state->fill_cw);
772            rs->polygon_mode |=
773                r300_translate_polygon_mode_back(state->fill_ccw);
774        }
775    } else {
776        rs->cull_mode = R300_FRONT_FACE_CCW;
777
778        /* Polygon offset */
779        if (state->offset_ccw) {
780            rs->polygon_offset_enable |= R300_FRONT_ENABLE;
781        }
782        if (state->offset_cw) {
783            rs->polygon_offset_enable |= R300_BACK_ENABLE;
784        }
785
786        /* Polygon mode */
787        if (rs->polygon_mode) {
788            rs->polygon_mode |=
789                r300_translate_polygon_mode_front(state->fill_ccw);
790            rs->polygon_mode |=
791                r300_translate_polygon_mode_back(state->fill_cw);
792        }
793    }
794    if (state->front_winding & state->cull_mode) {
795        rs->cull_mode |= R300_CULL_FRONT;
796    }
797    if (~(state->front_winding) & state->cull_mode) {
798        rs->cull_mode |= R300_CULL_BACK;
799    }
800
801    if (rs->polygon_offset_enable) {
802        rs->depth_offset = state->offset_units;
803        rs->depth_scale = state->offset_scale;
804    }
805
806    if (state->line_stipple_enable) {
807        rs->line_stipple_config =
808            R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE |
809            (fui((float)state->line_stipple_factor) &
810                R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK);
811        /* XXX this might need to be scaled up */
812        rs->line_stipple_value = state->line_stipple_pattern;
813    }
814
815    if (state->flatshade) {
816        rs->color_control = R300_SHADE_MODEL_FLAT;
817    } else {
818        rs->color_control = R300_SHADE_MODEL_SMOOTH;
819    }
820
821    rs->clip_rule = state->scissor ? 0xAAAA : 0xFFFF;
822
823    /* XXX Disable point sprites until we know what's wrong with them. */
824    rs->rs.sprite_coord_enable = 0;
825
826    /* Point sprites */
827    if (rs->rs.sprite_coord_enable) {
828        rs->stuffing_enable = R300_GB_POINT_STUFF_ENABLE;
829	for (i = 0; i < 8; i++) {
830	    if (state->sprite_coord_enable & (1 << i))
831		rs->stuffing_enable |=
832		    R300_GB_TEX_STR << (R300_GB_TEX0_SOURCE_SHIFT + (i*2));
833	}
834
835        rs->point_texcoord_left = 0.0f;
836        rs->point_texcoord_right = 1.0f;
837
838        switch (state->sprite_coord_mode) {
839            case PIPE_SPRITE_COORD_UPPER_LEFT:
840                rs->point_texcoord_top = 0.0f;
841                rs->point_texcoord_bottom = 1.0f;
842                break;
843            case PIPE_SPRITE_COORD_LOWER_LEFT:
844                rs->point_texcoord_top = 1.0f;
845                rs->point_texcoord_bottom = 0.0f;
846                break;
847        }
848    }
849
850    return (void*)rs;
851}
852
853/* Bind rasterizer state. */
854static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
855{
856    struct r300_context* r300 = r300_context(pipe);
857    struct r300_rs_state* rs = (struct r300_rs_state*)state;
858    int last_sprite_coord_enable = r300->sprite_coord_enable;
859
860    if (r300->draw) {
861        draw_flush(r300->draw);
862        draw_set_rasterizer_state(r300->draw, &rs->rs, state);
863    }
864
865    if (rs) {
866        r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw;
867        r300->sprite_coord_enable = rs->rs.sprite_coord_enable;
868    } else {
869        r300->polygon_offset_enabled = FALSE;
870        r300->sprite_coord_enable = 0;
871    }
872
873    UPDATE_STATE(state, r300->rs_state);
874    r300->rs_state.size = 26 + (r300->polygon_offset_enabled ? 5 : 0);
875
876    if (last_sprite_coord_enable != r300->sprite_coord_enable) {
877        r300->rs_block_state.dirty = TRUE;
878    }
879}
880
881/* Free rasterizer state. */
882static void r300_delete_rs_state(struct pipe_context* pipe, void* state)
883{
884    FREE(state);
885}
886
887static void*
888        r300_create_sampler_state(struct pipe_context* pipe,
889                                  const struct pipe_sampler_state* state)
890{
891    struct r300_context* r300 = r300_context(pipe);
892    struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state);
893    boolean is_r500 = r300->screen->caps.is_r500;
894    int lod_bias;
895    union util_color uc;
896
897    sampler->state = *state;
898
899    sampler->filter0 |=
900        (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) |
901        (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) |
902        (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT);
903
904    sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter,
905                                                   state->mag_img_filter,
906                                                   state->min_mip_filter,
907                                                   state->max_anisotropy > 0);
908
909    sampler->filter0 |= r300_anisotropy(state->max_anisotropy);
910
911    /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
912    /* We must pass these to the merge function to clamp them properly. */
913    sampler->min_lod = MAX2((unsigned)state->min_lod, 0);
914    sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0);
915
916    lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1);
917
918    sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT;
919
920    /* This is very high quality anisotropic filtering for R5xx.
921     * It's good for benchmarking the performance of texturing but
922     * in practice we don't want to slow down the driver because it's
923     * a pretty good performance killer. Feel free to play with it. */
924    if (DBG_ON(r300, DBG_ANISOHQ) && is_r500) {
925        sampler->filter1 |= r500_anisotropy(state->max_anisotropy);
926    }
927
928    util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
929    sampler->border_color = uc.ui;
930
931    /* R500-specific fixups and optimizations */
932    if (r300->screen->caps.is_r500) {
933        sampler->filter1 |= R500_BORDER_FIX;
934    }
935
936    return (void*)sampler;
937}
938
939static void r300_bind_sampler_states(struct pipe_context* pipe,
940                                     unsigned count,
941                                     void** states)
942{
943    struct r300_context* r300 = r300_context(pipe);
944    struct r300_textures_state* state =
945        (struct r300_textures_state*)r300->textures_state.state;
946    unsigned tex_units = r300->screen->caps.num_tex_units;
947
948    if (count > tex_units) {
949        return;
950    }
951
952    memcpy(state->sampler_states, states, sizeof(void*) * count);
953    state->sampler_state_count = count;
954
955    r300->textures_state.dirty = TRUE;
956}
957
958static void r300_lacks_vertex_textures(struct pipe_context* pipe,
959                                       unsigned count,
960                                       void** states)
961{
962}
963
964static void r300_delete_sampler_state(struct pipe_context* pipe, void* state)
965{
966    FREE(state);
967}
968
969static void r300_set_fragment_sampler_views(struct pipe_context* pipe,
970                                            unsigned count,
971                                            struct pipe_sampler_view** views)
972{
973    struct r300_context* r300 = r300_context(pipe);
974    struct r300_textures_state* state =
975        (struct r300_textures_state*)r300->textures_state.state;
976    struct r300_texture *texture;
977    unsigned i;
978    unsigned tex_units = r300->screen->caps.num_tex_units;
979    boolean dirty_tex = FALSE;
980
981    if (count > tex_units) {
982        return;
983    }
984
985    for (i = 0; i < count; i++) {
986        if (&state->sampler_views[i]->base != views[i]) {
987            pipe_sampler_view_reference(
988                    (struct pipe_sampler_view**)&state->sampler_views[i],
989                    views[i]);
990
991            if (!views[i]) {
992                continue;
993            }
994
995            /* A new sampler view (= texture)... */
996            dirty_tex = TRUE;
997
998            /* Set the texrect factor in the fragment shader.
999             * Needed for RECT and NPOT fallback. */
1000            texture = r300_texture(views[i]->texture);
1001            if (texture->uses_pitch) {
1002                r300->fs_rc_constant_state.dirty = TRUE;
1003            }
1004        }
1005    }
1006
1007    for (i = count; i < tex_units; i++) {
1008        if (state->sampler_views[i]) {
1009            pipe_sampler_view_reference(
1010                    (struct pipe_sampler_view**)&state->sampler_views[i],
1011                    NULL);
1012        }
1013    }
1014
1015    state->sampler_view_count = count;
1016
1017    r300->textures_state.dirty = TRUE;
1018
1019    if (dirty_tex) {
1020        r300->texture_cache_inval.dirty = TRUE;
1021    }
1022}
1023
1024static struct pipe_sampler_view *
1025r300_create_sampler_view(struct pipe_context *pipe,
1026                         struct pipe_resource *texture,
1027                         const struct pipe_sampler_view *templ)
1028{
1029    struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view);
1030    struct r300_texture *tex = r300_texture(texture);
1031    unsigned char swizzle[4];
1032
1033    if (view) {
1034        view->base = *templ;
1035        view->base.reference.count = 1;
1036        view->base.context = pipe;
1037        view->base.texture = NULL;
1038        pipe_resource_reference(&view->base.texture, texture);
1039
1040        swizzle[0] = templ->swizzle_r;
1041        swizzle[1] = templ->swizzle_g;
1042        swizzle[2] = templ->swizzle_b;
1043        swizzle[3] = templ->swizzle_a;
1044
1045        view->format = tex->tx_format;
1046        view->format.format1 |= r300_translate_texformat(templ->format,
1047                                                         swizzle);
1048        if (r300_screen(pipe->screen)->caps.is_r500) {
1049            view->format.format2 |= r500_tx_format_msb_bit(templ->format);
1050        }
1051    }
1052
1053    return (struct pipe_sampler_view*)view;
1054}
1055
1056static void
1057r300_sampler_view_destroy(struct pipe_context *pipe,
1058                          struct pipe_sampler_view *view)
1059{
1060   pipe_resource_reference(&view->texture, NULL);
1061   FREE(view);
1062}
1063
1064static void r300_set_scissor_state(struct pipe_context* pipe,
1065                                   const struct pipe_scissor_state* state)
1066{
1067    struct r300_context* r300 = r300_context(pipe);
1068
1069    memcpy(r300->scissor_state.state, state,
1070        sizeof(struct pipe_scissor_state));
1071
1072    r300->scissor_state.dirty = TRUE;
1073}
1074
1075static void r300_set_viewport_state(struct pipe_context* pipe,
1076                                    const struct pipe_viewport_state* state)
1077{
1078    struct r300_context* r300 = r300_context(pipe);
1079    struct r300_viewport_state* viewport =
1080        (struct r300_viewport_state*)r300->viewport_state.state;
1081
1082    r300->viewport = *state;
1083
1084    /* Do the transform in HW. */
1085    viewport->vte_control = R300_VTX_W0_FMT;
1086
1087    if (state->scale[0] != 1.0f) {
1088        viewport->xscale = state->scale[0];
1089        viewport->vte_control |= R300_VPORT_X_SCALE_ENA;
1090    }
1091    if (state->scale[1] != 1.0f) {
1092        viewport->yscale = state->scale[1];
1093        viewport->vte_control |= R300_VPORT_Y_SCALE_ENA;
1094    }
1095    if (state->scale[2] != 1.0f) {
1096        viewport->zscale = state->scale[2];
1097        viewport->vte_control |= R300_VPORT_Z_SCALE_ENA;
1098    }
1099    if (state->translate[0] != 0.0f) {
1100        viewport->xoffset = state->translate[0];
1101        viewport->vte_control |= R300_VPORT_X_OFFSET_ENA;
1102    }
1103    if (state->translate[1] != 0.0f) {
1104        viewport->yoffset = state->translate[1];
1105        viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA;
1106    }
1107    if (state->translate[2] != 0.0f) {
1108        viewport->zoffset = state->translate[2];
1109        viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA;
1110    }
1111
1112    r300->viewport_state.dirty = TRUE;
1113    if (r300->fs.state && r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) {
1114        r300->fs_rc_constant_state.dirty = TRUE;
1115    }
1116}
1117
1118static void r300_set_vertex_buffers(struct pipe_context* pipe,
1119                                    unsigned count,
1120                                    const struct pipe_vertex_buffer* buffers)
1121{
1122    struct r300_context* r300 = r300_context(pipe);
1123    struct pipe_vertex_buffer *vbo;
1124    unsigned i, max_index = (1 << 24) - 1;
1125    boolean any_user_buffer = FALSE;
1126
1127    if (count == r300->vertex_buffer_count &&
1128        memcmp(r300->vertex_buffer, buffers,
1129            sizeof(struct pipe_vertex_buffer) * count) == 0) {
1130        return;
1131    }
1132
1133    /* Check if the stride is aligned to the size of DWORD. */
1134    for (i = 0; i < count; i++) {
1135        if (buffers[i].buffer) {
1136            if (buffers[i].stride % 4 != 0) {
1137                // XXX Shouldn't we align the buffer?
1138                fprintf(stderr, "r300: set_vertex_buffers: "
1139                        "Unaligned buffer stride %i isn't supported.\n",
1140                        buffers[i].stride);
1141                abort();
1142            }
1143        }
1144    }
1145
1146    for (i = 0; i < count; i++) {
1147        /* Why, yes, I AM casting away constness. How did you know? */
1148        vbo = (struct pipe_vertex_buffer*)&buffers[i];
1149
1150        /* Reference our buffer. */
1151        pipe_resource_reference(&r300->vertex_buffer[i].buffer, vbo->buffer);
1152
1153        /* Skip NULL buffers */
1154        if (!buffers[i].buffer) {
1155            continue;
1156        }
1157
1158        if (r300_buffer_is_user_buffer(vbo->buffer)) {
1159            any_user_buffer = TRUE;
1160        }
1161
1162        if (vbo->max_index == ~0) {
1163	    /* if no VBO stride then only one vertex value so max index is 1 */
1164	    /* should think about converting to VS constants like svga does */
1165	    if (!vbo->stride)
1166		vbo->max_index = 1;
1167 	    else
1168            	vbo->max_index =
1169               		 (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride;
1170        }
1171
1172        max_index = MIN2(vbo->max_index, max_index);
1173    }
1174
1175    for (; i < r300->vertex_buffer_count; i++) {
1176        /* Dereference any old buffers. */
1177        pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL);
1178    }
1179
1180    memcpy(r300->vertex_buffer, buffers,
1181        sizeof(struct pipe_vertex_buffer) * count);
1182
1183    r300->vertex_buffer_count = count;
1184    r300->vertex_buffer_max_index = max_index;
1185    r300->any_user_vbs = any_user_buffer;
1186
1187    if (r300->draw) {
1188        draw_flush(r300->draw);
1189        draw_set_vertex_buffers(r300->draw, count, buffers);
1190    }
1191}
1192
1193/* Update the PSC tables. */
1194static void r300_vertex_psc(struct r300_vertex_element_state *velems)
1195{
1196    struct r300_vertex_stream_state *vstream = &velems->vertex_stream;
1197    uint16_t type, swizzle;
1198    enum pipe_format format;
1199    unsigned i;
1200
1201    if (velems->count > 16) {
1202        fprintf(stderr, "r300: More than 16 vertex elements are not supported,"
1203                " requested %i, using 16.\n", velems->count);
1204        velems->count = 16;
1205    }
1206
1207    /* Vertex shaders have no semantics on their inputs,
1208     * so PSC should just route stuff based on the vertex elements,
1209     * and not on attrib information. */
1210    for (i = 0; i < velems->count; i++) {
1211        format = velems->velem[i].src_format;
1212
1213        type = r300_translate_vertex_data_type(format) |
1214            (i << R300_DST_VEC_LOC_SHIFT);
1215        swizzle = r300_translate_vertex_data_swizzle(format);
1216
1217        if (i & 1) {
1218            vstream->vap_prog_stream_cntl[i >> 1] |= type << 16;
1219            vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16;
1220        } else {
1221            vstream->vap_prog_stream_cntl[i >> 1] |= type;
1222            vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
1223        }
1224    }
1225
1226    /* Set the last vector in the PSC. */
1227    if (i) {
1228        i -= 1;
1229    }
1230    vstream->vap_prog_stream_cntl[i >> 1] |=
1231        (R300_LAST_VEC << (i & 1 ? 16 : 0));
1232
1233    vstream->count = (i >> 1) + 1;
1234}
1235
1236static void* r300_create_vertex_elements_state(struct pipe_context* pipe,
1237                                               unsigned count,
1238                                               const struct pipe_vertex_element* attribs)
1239{
1240    struct r300_vertex_element_state *velems;
1241    unsigned i, size;
1242    enum pipe_format *format;
1243
1244    assert(count <= PIPE_MAX_ATTRIBS);
1245    velems = CALLOC_STRUCT(r300_vertex_element_state);
1246    if (velems != NULL) {
1247        velems->count = count;
1248        memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count);
1249
1250        if (r300_screen(pipe->screen)->caps.has_tcl) {
1251            r300_vertex_psc(velems);
1252
1253            /* Check if the format is aligned to the size of DWORD.
1254             * We only care about the blocksizes of the formats since
1255             * swizzles are already set up. */
1256            for (i = 0; i < count; i++) {
1257                format = &velems->velem[i].src_format;
1258
1259                /* Replace some formats with their aligned counterparts,
1260                 * this is OK because we check for aligned strides too. */
1261                switch (*format) {
1262                    /* Align to RGBA8. */
1263                    case PIPE_FORMAT_R8_UNORM:
1264                    case PIPE_FORMAT_R8G8_UNORM:
1265                    case PIPE_FORMAT_R8G8B8_UNORM:
1266                        *format = PIPE_FORMAT_R8G8B8A8_UNORM;
1267                        continue;
1268                    case PIPE_FORMAT_R8_SNORM:
1269                    case PIPE_FORMAT_R8G8_SNORM:
1270                    case PIPE_FORMAT_R8G8B8_SNORM:
1271                        *format = PIPE_FORMAT_R8G8B8A8_SNORM;
1272                        continue;
1273                    case PIPE_FORMAT_R8_USCALED:
1274                    case PIPE_FORMAT_R8G8_USCALED:
1275                    case PIPE_FORMAT_R8G8B8_USCALED:
1276                        *format = PIPE_FORMAT_R8G8B8A8_USCALED;
1277                        continue;
1278                    case PIPE_FORMAT_R8_SSCALED:
1279                    case PIPE_FORMAT_R8G8_SSCALED:
1280                    case PIPE_FORMAT_R8G8B8_SSCALED:
1281                        *format = PIPE_FORMAT_R8G8B8A8_SSCALED;
1282                        continue;
1283
1284                    /* Align to RG16. */
1285                    case PIPE_FORMAT_R16_UNORM:
1286                        *format = PIPE_FORMAT_R16G16_UNORM;
1287                        continue;
1288                    case PIPE_FORMAT_R16_SNORM:
1289                        *format = PIPE_FORMAT_R16G16_SNORM;
1290                        continue;
1291                    case PIPE_FORMAT_R16_USCALED:
1292                        *format = PIPE_FORMAT_R16G16_USCALED;
1293                        continue;
1294                    case PIPE_FORMAT_R16_SSCALED:
1295                        *format = PIPE_FORMAT_R16G16_SSCALED;
1296                        continue;
1297                    case PIPE_FORMAT_R16_FLOAT:
1298                        *format = PIPE_FORMAT_R16G16_FLOAT;
1299                        continue;
1300
1301                    /* Align to RGBA16. */
1302                    case PIPE_FORMAT_R16G16B16_UNORM:
1303                        *format = PIPE_FORMAT_R16G16B16A16_UNORM;
1304                        continue;
1305                    case PIPE_FORMAT_R16G16B16_SNORM:
1306                        *format = PIPE_FORMAT_R16G16B16A16_SNORM;
1307                        continue;
1308                    case PIPE_FORMAT_R16G16B16_USCALED:
1309                        *format = PIPE_FORMAT_R16G16B16A16_USCALED;
1310                        continue;
1311                    case PIPE_FORMAT_R16G16B16_SSCALED:
1312                        *format = PIPE_FORMAT_R16G16B16A16_SSCALED;
1313                        continue;
1314                    case PIPE_FORMAT_R16G16B16_FLOAT:
1315                        *format = PIPE_FORMAT_R16G16B16A16_FLOAT;
1316                        continue;
1317
1318                    default:;
1319                }
1320
1321                size = util_format_get_blocksize(*format);
1322
1323                if (size % 4 != 0) {
1324                    /* XXX Shouldn't we align the format? */
1325                    fprintf(stderr, "r300_create_vertex_elements_state: "
1326                            "Unaligned format %s:%i isn't supported\n",
1327                            util_format_name(*format), size);
1328                    assert(0);
1329                    abort();
1330                }
1331            }
1332
1333        }
1334    }
1335    return velems;
1336}
1337
1338static void r300_bind_vertex_elements_state(struct pipe_context *pipe,
1339                                            void *state)
1340{
1341    struct r300_context *r300 = r300_context(pipe);
1342    struct r300_vertex_element_state *velems = state;
1343
1344    if (velems == NULL) {
1345        return;
1346    }
1347
1348    r300->velems = velems;
1349
1350    if (r300->draw) {
1351        draw_flush(r300->draw);
1352        draw_set_vertex_elements(r300->draw, velems->count, velems->velem);
1353    }
1354
1355    UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state);
1356    r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2;
1357}
1358
1359static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state)
1360{
1361   FREE(state);
1362}
1363
1364static void* r300_create_vs_state(struct pipe_context* pipe,
1365                                  const struct pipe_shader_state* shader)
1366{
1367    struct r300_context* r300 = r300_context(pipe);
1368
1369    struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader);
1370
1371    /* Copy state directly into shader. */
1372    vs->state = *shader;
1373    vs->state.tokens = tgsi_dup_tokens(shader->tokens);
1374
1375    if (r300->screen->caps.has_tcl) {
1376        r300_translate_vertex_shader(r300, vs, vs->state.tokens);
1377    } else {
1378        vs->draw_vs = draw_create_vertex_shader(r300->draw, shader);
1379    }
1380
1381    return vs;
1382}
1383
1384static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
1385{
1386    struct r300_context* r300 = r300_context(pipe);
1387    struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
1388
1389    if (vs == NULL) {
1390        r300->vs_state.state = NULL;
1391        return;
1392    }
1393    if (vs == r300->vs_state.state) {
1394        return;
1395    }
1396    r300->vs_state.state = vs;
1397
1398    // VS output mapping for HWTCL or stream mapping for SWTCL to the RS block
1399    if (r300->fs.state) {
1400        r300_vertex_shader_setup_wpos(r300);
1401    }
1402    memcpy(r300->vap_output_state.state, &vs->vap_out,
1403           sizeof(struct r300_vap_output_state));
1404    r300->vap_output_state.dirty = TRUE;
1405
1406    /* The majority of the RS block bits is dependent on the vertex shader. */
1407    r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
1408
1409    if (r300->screen->caps.has_tcl) {
1410        r300->vs_state.dirty = TRUE;
1411        r300->vs_state.size =
1412                vs->code.length + 9 +
1413                (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0);
1414
1415        if (vs->externals_count) {
1416            r300->vs_constants.dirty = TRUE;
1417            r300->vs_constants.size = vs->externals_count * 4 + 3;
1418        } else {
1419            r300->vs_constants.size = 0;
1420        }
1421
1422        r300->pvs_flush.dirty = TRUE;
1423    } else {
1424        draw_flush(r300->draw);
1425        draw_bind_vertex_shader(r300->draw,
1426                (struct draw_vertex_shader*)vs->draw_vs);
1427    }
1428}
1429
1430static void r300_delete_vs_state(struct pipe_context* pipe, void* shader)
1431{
1432    struct r300_context* r300 = r300_context(pipe);
1433    struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
1434
1435    if (r300->screen->caps.has_tcl) {
1436        rc_constants_destroy(&vs->code.constants);
1437    } else {
1438        draw_delete_vertex_shader(r300->draw,
1439                (struct draw_vertex_shader*)vs->draw_vs);
1440    }
1441
1442    FREE((void*)vs->state.tokens);
1443    FREE(shader);
1444}
1445
1446static void r300_set_constant_buffer(struct pipe_context *pipe,
1447                                     uint shader, uint index,
1448                                     struct pipe_resource *buf)
1449{
1450    struct r300_context* r300 = r300_context(pipe);
1451    struct r300_constant_buffer *cbuf;
1452    struct pipe_transfer *tr;
1453    void *mapped;
1454    int max_size = 0;
1455
1456    switch (shader) {
1457        case PIPE_SHADER_VERTEX:
1458            cbuf = (struct r300_constant_buffer*)r300->vs_constants.state;
1459            max_size = 256;
1460            break;
1461        case PIPE_SHADER_FRAGMENT:
1462            cbuf = (struct r300_constant_buffer*)r300->fs_constants.state;
1463            if (r300->screen->caps.is_r500) {
1464                max_size = 256;
1465            } else {
1466                max_size = 32;
1467            }
1468            break;
1469        default:
1470            assert(0);
1471            return;
1472    }
1473
1474    if (buf == NULL || buf->width0 == 0 ||
1475        (mapped = pipe_buffer_map(pipe, buf, PIPE_TRANSFER_READ, &tr)) == NULL)
1476    {
1477        cbuf->count = 0;
1478        return;
1479    }
1480
1481    assert((buf->width0 % 4 * sizeof(float)) == 0);
1482
1483    /* Check the size of the constant buffer. */
1484    /* XXX Subtract immediates and RC_STATE_* variables. */
1485    if (buf->width0 > (sizeof(float) * 4 * max_size)) {
1486        fprintf(stderr, "r300: Max size of the constant buffer is "
1487                      "%i*4 floats.\n", max_size);
1488        abort();
1489    }
1490
1491    memcpy(cbuf->constants, mapped, buf->width0);
1492    cbuf->count = buf->width0 / (4 * sizeof(float));
1493    pipe_buffer_unmap(pipe, buf, tr);
1494
1495    if (shader == PIPE_SHADER_VERTEX) {
1496        if (r300->screen->caps.has_tcl) {
1497            if (r300->vs_constants.size) {
1498                r300->vs_constants.dirty = TRUE;
1499            }
1500            r300->pvs_flush.dirty = TRUE;
1501        } else if (r300->draw) {
1502            draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX,
1503                0, cbuf->constants,
1504                buf->width0);
1505        }
1506    } else if (shader == PIPE_SHADER_FRAGMENT) {
1507        r300->fs_constants.dirty = TRUE;
1508    }
1509}
1510
1511void r300_init_state_functions(struct r300_context* r300)
1512{
1513    r300->context.create_blend_state = r300_create_blend_state;
1514    r300->context.bind_blend_state = r300_bind_blend_state;
1515    r300->context.delete_blend_state = r300_delete_blend_state;
1516
1517    r300->context.set_blend_color = r300_set_blend_color;
1518
1519    r300->context.set_clip_state = r300_set_clip_state;
1520
1521    r300->context.set_constant_buffer = r300_set_constant_buffer;
1522
1523    r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state;
1524    r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state;
1525    r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state;
1526
1527    r300->context.set_stencil_ref = r300_set_stencil_ref;
1528
1529    r300->context.set_framebuffer_state = r300_set_framebuffer_state;
1530
1531    r300->context.create_fs_state = r300_create_fs_state;
1532    r300->context.bind_fs_state = r300_bind_fs_state;
1533    r300->context.delete_fs_state = r300_delete_fs_state;
1534
1535    r300->context.set_polygon_stipple = r300_set_polygon_stipple;
1536
1537    r300->context.create_rasterizer_state = r300_create_rs_state;
1538    r300->context.bind_rasterizer_state = r300_bind_rs_state;
1539    r300->context.delete_rasterizer_state = r300_delete_rs_state;
1540
1541    r300->context.create_sampler_state = r300_create_sampler_state;
1542    r300->context.bind_fragment_sampler_states = r300_bind_sampler_states;
1543    r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures;
1544    r300->context.delete_sampler_state = r300_delete_sampler_state;
1545
1546    r300->context.set_fragment_sampler_views = r300_set_fragment_sampler_views;
1547    r300->context.create_sampler_view = r300_create_sampler_view;
1548    r300->context.sampler_view_destroy = r300_sampler_view_destroy;
1549
1550    r300->context.set_scissor_state = r300_set_scissor_state;
1551
1552    r300->context.set_viewport_state = r300_set_viewport_state;
1553
1554    r300->context.set_vertex_buffers = r300_set_vertex_buffers;
1555
1556    r300->context.create_vertex_elements_state = r300_create_vertex_elements_state;
1557    r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state;
1558    r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state;
1559
1560    r300->context.create_vs_state = r300_create_vs_state;
1561    r300->context.bind_vs_state = r300_bind_vs_state;
1562    r300->context.delete_vs_state = r300_delete_vs_state;
1563}
1564