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/half_float.h"
28#include "util/u_helpers.h"
29#include "util/u_math.h"
30#include "util/u_memory.h"
31#include "util/u_pack_color.h"
32#include "util/u_transfer.h"
33#include "util/u_blend.h"
34
35#include "tgsi/tgsi_parse.h"
36
37#include "pipe/p_config.h"
38
39#include "r300_cb.h"
40#include "r300_context.h"
41#include "r300_emit.h"
42#include "r300_reg.h"
43#include "r300_screen.h"
44#include "r300_screen_buffer.h"
45#include "r300_state_inlines.h"
46#include "r300_fs.h"
47#include "r300_texture.h"
48#include "r300_vs.h"
49
50/* r300_state: Functions used to initialize 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 blend_discard_conditionally(unsigned eqRGB, unsigned eqA,
174                                            unsigned dstRGB, unsigned dstA,
175                                            unsigned srcRGB, unsigned srcA)
176{
177    unsigned blend_control = 0;
178
179    /* Optimization: discard pixels which don't change the colorbuffer.
180     *
181     * The code below is non-trivial and some math is involved.
182     *
183     * Discarding pixels must be disabled when FP16 AA is enabled.
184     * This is a hardware bug. Also, this implementation wouldn't work
185     * with FP blending enabled and equation clamping disabled.
186     *
187     * Equations other than ADD are rarely used and therefore won't be
188     * optimized. */
189    if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) &&
190        (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) {
191        /* ADD: X+Y
192         * REVERSE_SUBTRACT: Y-X
193         *
194         * The idea is:
195         * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1,
196         * then CB will not be changed.
197         *
198         * Given the srcFactor and dstFactor variables, we can derive
199         * what src and dst should be equal to and discard appropriate
200         * pixels.
201         */
202        if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) {
203            blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0;
204        } else if (blend_discard_if_src_alpha_1(srcRGB, srcA,
205                                                dstRGB, dstA)) {
206            blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1;
207        } else if (blend_discard_if_src_color_0(srcRGB, srcA,
208                                                dstRGB, dstA)) {
209            blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0;
210        } else if (blend_discard_if_src_color_1(srcRGB, srcA,
211                                                dstRGB, dstA)) {
212            blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1;
213        } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA,
214                                                      dstRGB, dstA)) {
215            blend_control |=
216                R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0;
217        } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA,
218                                                      dstRGB, dstA)) {
219            blend_control |=
220                R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1;
221        }
222    }
223    return blend_control;
224}
225
226/* The hardware colormask is clunky a must be swizzled depending on the format.
227 * This was figured out by trial-and-error. */
228static unsigned bgra_cmask(unsigned mask)
229{
230    return ((mask & PIPE_MASK_R) << 2) |
231           ((mask & PIPE_MASK_B) >> 2) |
232           (mask & (PIPE_MASK_G | PIPE_MASK_A));
233}
234
235static unsigned rgba_cmask(unsigned mask)
236{
237    return mask & PIPE_MASK_RGBA;
238}
239
240static unsigned rrrr_cmask(unsigned mask)
241{
242    return (mask & PIPE_MASK_R) |
243           ((mask & PIPE_MASK_R) << 1) |
244           ((mask & PIPE_MASK_R) << 2) |
245           ((mask & PIPE_MASK_R) << 3);
246}
247
248static unsigned aaaa_cmask(unsigned mask)
249{
250    return ((mask & PIPE_MASK_A) >> 3) |
251           ((mask & PIPE_MASK_A) >> 2) |
252           ((mask & PIPE_MASK_A) >> 1) |
253           (mask & PIPE_MASK_A);
254}
255
256static unsigned grrg_cmask(unsigned mask)
257{
258    return ((mask & PIPE_MASK_R) << 1) |
259           ((mask & PIPE_MASK_R) << 2) |
260           ((mask & PIPE_MASK_G) >> 1) |
261           ((mask & PIPE_MASK_G) << 2);
262}
263
264static unsigned arra_cmask(unsigned mask)
265{
266    return ((mask & PIPE_MASK_R) << 1) |
267           ((mask & PIPE_MASK_R) << 2) |
268           ((mask & PIPE_MASK_A) >> 3) |
269           (mask & PIPE_MASK_A);
270}
271
272static unsigned blend_read_enable(unsigned eqRGB, unsigned eqA,
273                                  unsigned dstRGB, unsigned dstA,
274                                  unsigned srcRGB, unsigned srcA,
275                                  boolean src_alpha_optz)
276{
277    unsigned blend_control = 0;
278
279    /* Optimization: some operations do not require the destination color.
280     *
281     * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled,
282     * otherwise blending gives incorrect results. It seems to be
283     * a hardware bug. */
284    if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN ||
285        eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX ||
286        dstRGB != PIPE_BLENDFACTOR_ZERO ||
287        dstA != PIPE_BLENDFACTOR_ZERO ||
288        util_blend_factor_uses_dest(srcRGB, false) ||
289        util_blend_factor_uses_dest(srcA, true)) {
290        /* Enable reading from the colorbuffer. */
291        blend_control |= R300_READ_ENABLE;
292
293        if (src_alpha_optz) {
294            /* Optimization: Depending on incoming pixels, we can
295             * conditionally disable the reading in hardware... */
296            if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN &&
297                eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) {
298                /* Disable reading if SRC_ALPHA == 0. */
299                if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
300                     dstRGB == PIPE_BLENDFACTOR_ZERO) &&
301                    (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
302                     dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
303                     dstA == PIPE_BLENDFACTOR_ZERO) &&
304                    (srcRGB != PIPE_BLENDFACTOR_DST_COLOR &&
305                     srcRGB != PIPE_BLENDFACTOR_DST_ALPHA &&
306                     srcRGB != PIPE_BLENDFACTOR_INV_DST_COLOR &&
307                     srcRGB != PIPE_BLENDFACTOR_INV_DST_ALPHA)) {
308                     blend_control |= R500_SRC_ALPHA_0_NO_READ;
309                }
310
311                /* Disable reading if SRC_ALPHA == 1. */
312                if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
313                     dstRGB == PIPE_BLENDFACTOR_ZERO) &&
314                    (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
315                     dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
316                     dstA == PIPE_BLENDFACTOR_ZERO) &&
317                    (srcRGB != PIPE_BLENDFACTOR_DST_COLOR &&
318                     srcRGB != PIPE_BLENDFACTOR_DST_ALPHA &&
319                     srcRGB != PIPE_BLENDFACTOR_INV_DST_COLOR &&
320                     srcRGB != PIPE_BLENDFACTOR_INV_DST_ALPHA)) {
321                     blend_control |= R500_SRC_ALPHA_1_NO_READ;
322                }
323            }
324        }
325    }
326    return blend_control;
327}
328
329/* Create a new blend state based on the CSO blend state.
330 *
331 * This encompasses alpha blending, logic/raster ops, and blend dithering. */
332static void* r300_create_blend_state(struct pipe_context* pipe,
333                                     const struct pipe_blend_state* state)
334{
335    struct r300_screen* r300screen = r300_screen(pipe->screen);
336    struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state);
337    uint32_t blend_control = 0;       /* R300_RB3D_CBLEND: 0x4e04 */
338    uint32_t blend_control_noclamp = 0;    /* R300_RB3D_CBLEND: 0x4e04 */
339    uint32_t blend_control_noalpha = 0;    /* R300_RB3D_CBLEND: 0x4e04 */
340    uint32_t blend_control_noalpha_noclamp = 0;    /* R300_RB3D_CBLEND: 0x4e04 */
341    uint32_t alpha_blend_control = 0; /* R300_RB3D_ABLEND: 0x4e08 */
342    uint32_t alpha_blend_control_noclamp = 0; /* R300_RB3D_ABLEND: 0x4e08 */
343    uint32_t alpha_blend_control_noalpha = 0; /* R300_RB3D_ABLEND: 0x4e08 */
344    uint32_t alpha_blend_control_noalpha_noclamp = 0; /* R300_RB3D_ABLEND: 0x4e08 */
345    uint32_t rop = 0;                 /* R300_RB3D_ROPCNTL: 0x4e18 */
346    uint32_t dither = 0;              /* R300_RB3D_DITHER_CTL: 0x4e50 */
347    int i;
348
349    const unsigned eqRGB = state->rt[0].rgb_func;
350    const unsigned srcRGB = state->rt[0].rgb_src_factor;
351    const unsigned dstRGB = state->rt[0].rgb_dst_factor;
352
353    const unsigned eqA = state->rt[0].alpha_func;
354    const unsigned srcA = state->rt[0].alpha_src_factor;
355    const unsigned dstA = state->rt[0].alpha_dst_factor;
356
357    unsigned srcRGBX = srcRGB;
358    unsigned dstRGBX = dstRGB;
359    CB_LOCALS;
360
361    blend->state = *state;
362
363    /* force DST_ALPHA to ONE where we can */
364    switch (srcRGBX) {
365    case PIPE_BLENDFACTOR_DST_ALPHA:
366        srcRGBX = PIPE_BLENDFACTOR_ONE;
367        break;
368    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
369        srcRGBX = PIPE_BLENDFACTOR_ZERO;
370        break;
371    }
372
373    switch (dstRGBX) {
374    case PIPE_BLENDFACTOR_DST_ALPHA:
375        dstRGBX = PIPE_BLENDFACTOR_ONE;
376        break;
377    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
378        dstRGBX = PIPE_BLENDFACTOR_ZERO;
379        break;
380    }
381
382    /* Get blending register values. */
383    if (state->rt[0].blend_enable) {
384        unsigned blend_eq, blend_eq_noclamp;
385
386        /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha,
387         * this is just the crappy D3D naming */
388        blend_control = blend_control_noclamp =
389            R300_ALPHA_BLEND_ENABLE |
390            ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) |
391            ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT);
392
393        blend_control_noalpha = blend_control_noalpha_noclamp =
394            R300_ALPHA_BLEND_ENABLE |
395            ( r300_translate_blend_factor(srcRGBX) << R300_SRC_BLEND_SHIFT) |
396            ( r300_translate_blend_factor(dstRGBX) << R300_DST_BLEND_SHIFT);
397
398        blend_eq = r300_translate_blend_function(eqRGB, TRUE);
399        blend_eq_noclamp = r300_translate_blend_function(eqRGB, FALSE);
400
401        blend_control |= blend_eq;
402        blend_control_noalpha |= blend_eq;
403        blend_control_noclamp |= blend_eq_noclamp;
404        blend_control_noalpha_noclamp |= blend_eq_noclamp;
405
406        /* Optimization: some operations do not require the destination color. */
407        blend_control |= blend_read_enable(eqRGB, eqA, dstRGB, dstA,
408                                           srcRGB, srcA, r300screen->caps.is_r500);
409        blend_control_noclamp |= blend_read_enable(eqRGB, eqA, dstRGB, dstA,
410                                                   srcRGB, srcA, FALSE);
411        blend_control_noalpha |= blend_read_enable(eqRGB, eqA, dstRGBX, dstA,
412                                                   srcRGBX, srcA, r300screen->caps.is_r500);
413        blend_control_noalpha_noclamp |= blend_read_enable(eqRGB, eqA, dstRGBX, dstA,
414                                                           srcRGBX, srcA, FALSE);
415
416        /* Optimization: discard pixels which don't change the colorbuffer.
417         * It cannot be used with FP16 AA. */
418        blend_control |= blend_discard_conditionally(eqRGB, eqA, dstRGB, dstA,
419                                                     srcRGB, srcA);
420        blend_control_noalpha |= blend_discard_conditionally(eqRGB, eqA, dstRGBX, dstA,
421                                                             srcRGBX, srcA);
422
423        /* separate alpha */
424        if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) {
425            blend_control |= R300_SEPARATE_ALPHA_ENABLE;
426            blend_control_noclamp |= R300_SEPARATE_ALPHA_ENABLE;
427
428            alpha_blend_control = alpha_blend_control_noclamp =
429                (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) |
430                (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT);
431            alpha_blend_control |= r300_translate_blend_function(eqA, TRUE);
432            alpha_blend_control_noclamp |= r300_translate_blend_function(eqA, FALSE);
433        }
434        if (srcA != srcRGBX || dstA != dstRGBX || eqA != eqRGB) {
435            blend_control_noalpha |= R300_SEPARATE_ALPHA_ENABLE;
436            blend_control_noalpha_noclamp |= R300_SEPARATE_ALPHA_ENABLE;
437
438            alpha_blend_control_noalpha = alpha_blend_control_noalpha_noclamp =
439                (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) |
440                (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT);
441            alpha_blend_control_noalpha |= r300_translate_blend_function(eqA, TRUE);
442            alpha_blend_control_noalpha_noclamp |= r300_translate_blend_function(eqA, FALSE);
443        }
444    }
445
446    /* PIPE_LOGICOP_* don't need to be translated, fortunately. */
447    if (state->logicop_enable) {
448        rop = R300_RB3D_ROPCNTL_ROP_ENABLE |
449                (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT;
450    }
451
452    /* Neither fglrx nor classic r300 ever set this, regardless of dithering
453     * state. Since it's an optional implementation detail, we can leave it
454     * out and never dither.
455     *
456     * This could be revisited if we ever get quality or conformance hints.
457     *
458    if (state->dither) {
459        dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT |
460                        R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT;
461    }
462    */
463
464    /* Build a command buffer. */
465    {
466        unsigned (*func[COLORMASK_NUM_SWIZZLES])(unsigned) = {
467            bgra_cmask,
468            rgba_cmask,
469            rrrr_cmask,
470            aaaa_cmask,
471            grrg_cmask,
472            arra_cmask,
473            bgra_cmask,
474            rgba_cmask
475        };
476
477        for (i = 0; i < COLORMASK_NUM_SWIZZLES; i++) {
478            boolean has_alpha = i != COLORMASK_RGBX && i != COLORMASK_BGRX;
479
480            BEGIN_CB(blend->cb_clamp[i], 8);
481            OUT_CB_REG(R300_RB3D_ROPCNTL, rop);
482            OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3);
483            OUT_CB(has_alpha ? blend_control : blend_control_noalpha);
484            OUT_CB(has_alpha ? alpha_blend_control : alpha_blend_control_noalpha);
485            OUT_CB(func[i](state->rt[0].colormask));
486            OUT_CB_REG(R300_RB3D_DITHER_CTL, dither);
487            END_CB;
488        }
489    }
490
491    /* Build a command buffer (for RGBA16F). */
492    BEGIN_CB(blend->cb_noclamp, 8);
493    OUT_CB_REG(R300_RB3D_ROPCNTL, rop);
494    OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3);
495    OUT_CB(blend_control_noclamp);
496    OUT_CB(alpha_blend_control_noclamp);
497    OUT_CB(rgba_cmask(state->rt[0].colormask));
498    OUT_CB_REG(R300_RB3D_DITHER_CTL, dither);
499    END_CB;
500
501    /* Build a command buffer (for RGB16F). */
502    BEGIN_CB(blend->cb_noclamp_noalpha, 8);
503    OUT_CB_REG(R300_RB3D_ROPCNTL, rop);
504    OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3);
505    OUT_CB(blend_control_noalpha_noclamp);
506    OUT_CB(alpha_blend_control_noalpha_noclamp);
507    OUT_CB(rgba_cmask(state->rt[0].colormask));
508    OUT_CB_REG(R300_RB3D_DITHER_CTL, dither);
509    END_CB;
510
511    /* The same as above, but with no colorbuffer reads and writes. */
512    BEGIN_CB(blend->cb_no_readwrite, 8);
513    OUT_CB_REG(R300_RB3D_ROPCNTL, rop);
514    OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3);
515    OUT_CB(0);
516    OUT_CB(0);
517    OUT_CB(0);
518    OUT_CB_REG(R300_RB3D_DITHER_CTL, dither);
519    END_CB;
520
521    return (void*)blend;
522}
523
524/* Bind blend state. */
525static void r300_bind_blend_state(struct pipe_context* pipe,
526                                  void* state)
527{
528    struct r300_context* r300 = r300_context(pipe);
529    struct r300_blend_state *blend  = (struct r300_blend_state*)state;
530    boolean last_alpha_to_one = r300->alpha_to_one;
531    boolean last_alpha_to_coverage = r300->alpha_to_coverage;
532
533    UPDATE_STATE(state, r300->blend_state);
534
535    if (!blend)
536        return;
537
538    r300->alpha_to_one = blend->state.alpha_to_one;
539    r300->alpha_to_coverage = blend->state.alpha_to_coverage;
540
541    if (r300->alpha_to_one != last_alpha_to_one && r300->msaa_enable &&
542        r300->fs_status == FRAGMENT_SHADER_VALID) {
543        r300->fs_status = FRAGMENT_SHADER_MAYBE_DIRTY;
544    }
545
546    if (r300->alpha_to_coverage != last_alpha_to_coverage &&
547        r300->msaa_enable) {
548        r300_mark_atom_dirty(r300, &r300->dsa_state);
549    }
550}
551
552/* Free blend state. */
553static void r300_delete_blend_state(struct pipe_context* pipe,
554                                    void* state)
555{
556    FREE(state);
557}
558
559/* Convert float to 10bit integer */
560static unsigned float_to_fixed10(float f)
561{
562    return CLAMP((unsigned)(f * 1023.9f), 0, 1023);
563}
564
565/* Set blend color.
566 * Setup both R300 and R500 registers, figure out later which one to write. */
567static void r300_set_blend_color(struct pipe_context* pipe,
568                                 const struct pipe_blend_color* color)
569{
570    struct r300_context* r300 = r300_context(pipe);
571    struct pipe_framebuffer_state *fb = r300->fb_state.state;
572    struct r300_blend_color_state *state =
573        (struct r300_blend_color_state*)r300->blend_color_state.state;
574    struct pipe_blend_color c;
575    struct pipe_surface *cb;
576    float tmp;
577    CB_LOCALS;
578
579    state->state = *color; /* Save it, so that we can reuse it in set_fb_state */
580    c = *color;
581    cb = fb->nr_cbufs ? r300_get_nonnull_cb(fb, 0) : NULL;
582
583    /* The blend color is dependent on the colorbuffer format. */
584    if (cb) {
585        switch (cb->format) {
586        case PIPE_FORMAT_R8_UNORM:
587        case PIPE_FORMAT_L8_UNORM:
588        case PIPE_FORMAT_I8_UNORM:
589            c.color[1] = c.color[0];
590            break;
591
592        case PIPE_FORMAT_A8_UNORM:
593            c.color[1] = c.color[3];
594            break;
595
596        case PIPE_FORMAT_R8G8_UNORM:
597            c.color[2] = c.color[1];
598            break;
599
600        case PIPE_FORMAT_L8A8_UNORM:
601        case PIPE_FORMAT_R8A8_UNORM:
602            c.color[2] = c.color[3];
603            break;
604
605        case PIPE_FORMAT_R8G8B8A8_UNORM:
606        case PIPE_FORMAT_R8G8B8X8_UNORM:
607            tmp = c.color[0];
608            c.color[0] = c.color[2];
609            c.color[2] = tmp;
610            break;
611
612        default:;
613        }
614    }
615
616    if (r300->screen->caps.is_r500) {
617        BEGIN_CB(state->cb, 3);
618        OUT_CB_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2);
619
620        switch (cb ? cb->format : 0) {
621        case PIPE_FORMAT_R16G16B16A16_FLOAT:
622        case PIPE_FORMAT_R16G16B16X16_FLOAT:
623            OUT_CB(_mesa_float_to_half(c.color[2]) |
624                   (_mesa_float_to_half(c.color[3]) << 16));
625            OUT_CB(_mesa_float_to_half(c.color[0]) |
626                   (_mesa_float_to_half(c.color[1]) << 16));
627            break;
628
629        default:
630            OUT_CB(float_to_fixed10(c.color[0]) |
631                   (float_to_fixed10(c.color[3]) << 16));
632            OUT_CB(float_to_fixed10(c.color[2]) |
633                   (float_to_fixed10(c.color[1]) << 16));
634        }
635
636        END_CB;
637    } else {
638        union util_color uc;
639        util_pack_color(c.color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
640
641        BEGIN_CB(state->cb, 2);
642        OUT_CB_REG(R300_RB3D_BLEND_COLOR, uc.ui[0]);
643        END_CB;
644    }
645
646    r300_mark_atom_dirty(r300, &r300->blend_color_state);
647}
648
649static void r300_set_clip_state(struct pipe_context* pipe,
650                                const struct pipe_clip_state* state)
651{
652    struct r300_context* r300 = r300_context(pipe);
653    struct r300_clip_state *clip =
654            (struct r300_clip_state*)r300->clip_state.state;
655    CB_LOCALS;
656
657    if (r300->screen->caps.has_tcl) {
658        BEGIN_CB(clip->cb, r300->clip_state.size);
659        OUT_CB_REG(R300_VAP_PVS_VECTOR_INDX_REG,
660                   (r300->screen->caps.is_r500 ?
661                    R500_PVS_UCP_START : R300_PVS_UCP_START));
662        OUT_CB_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, 6 * 4);
663        OUT_CB_TABLE(state->ucp, 6 * 4);
664        END_CB;
665
666        r300_mark_atom_dirty(r300, &r300->clip_state);
667    } else {
668        draw_set_clip_state(r300->draw, state);
669    }
670}
671
672/* Create a new depth, stencil, and alpha state based on the CSO dsa state.
673 *
674 * This contains the depth buffer, stencil buffer, alpha test, and such.
675 * On the Radeon, depth and stencil buffer setup are intertwined, which is
676 * the reason for some of the strange-looking assignments across registers. */
677static void* r300_create_dsa_state(struct pipe_context* pipe,
678                          const struct pipe_depth_stencil_alpha_state* state)
679{
680    boolean is_r500 = r300_screen(pipe->screen)->caps.is_r500;
681    struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state);
682    CB_LOCALS;
683    uint32_t alpha_value_fp16 = 0;
684    uint32_t z_buffer_control = 0;
685    uint32_t z_stencil_control = 0;
686    uint32_t stencil_ref_mask = 0;
687    uint32_t stencil_ref_bf = 0;
688
689    dsa->dsa = *state;
690
691    /* Depth test setup. - separate write mask depth for decomp flush */
692    if (state->depth_writemask) {
693        z_buffer_control |= R300_Z_WRITE_ENABLE;
694    }
695
696    if (state->depth_enabled) {
697        z_buffer_control |= R300_Z_ENABLE;
698
699        z_stencil_control |=
700            (r300_translate_depth_stencil_function(state->depth_func) <<
701                R300_Z_FUNC_SHIFT);
702    }
703
704    /* Stencil buffer setup. */
705    if (state->stencil[0].enabled) {
706        z_buffer_control |= R300_STENCIL_ENABLE;
707        z_stencil_control |=
708            (r300_translate_depth_stencil_function(state->stencil[0].func) <<
709                R300_S_FRONT_FUNC_SHIFT) |
710            (r300_translate_stencil_op(state->stencil[0].fail_op) <<
711                R300_S_FRONT_SFAIL_OP_SHIFT) |
712            (r300_translate_stencil_op(state->stencil[0].zpass_op) <<
713                R300_S_FRONT_ZPASS_OP_SHIFT) |
714            (r300_translate_stencil_op(state->stencil[0].zfail_op) <<
715                R300_S_FRONT_ZFAIL_OP_SHIFT);
716
717        stencil_ref_mask =
718                (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) |
719                (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT);
720
721        if (state->stencil[1].enabled) {
722            dsa->two_sided = TRUE;
723
724            z_buffer_control |= R300_STENCIL_FRONT_BACK;
725            z_stencil_control |=
726            (r300_translate_depth_stencil_function(state->stencil[1].func) <<
727                R300_S_BACK_FUNC_SHIFT) |
728            (r300_translate_stencil_op(state->stencil[1].fail_op) <<
729                R300_S_BACK_SFAIL_OP_SHIFT) |
730            (r300_translate_stencil_op(state->stencil[1].zpass_op) <<
731                R300_S_BACK_ZPASS_OP_SHIFT) |
732            (r300_translate_stencil_op(state->stencil[1].zfail_op) <<
733                R300_S_BACK_ZFAIL_OP_SHIFT);
734
735            stencil_ref_bf =
736                (state->stencil[1].valuemask << R300_STENCILMASK_SHIFT) |
737                (state->stencil[1].writemask << R300_STENCILWRITEMASK_SHIFT);
738
739            if (is_r500) {
740                z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK;
741            } else {
742                dsa->two_sided_stencil_ref =
743                  (state->stencil[0].valuemask != state->stencil[1].valuemask ||
744                   state->stencil[0].writemask != state->stencil[1].writemask);
745            }
746        }
747    }
748
749    /* Alpha test setup. */
750    if (state->alpha_enabled) {
751        dsa->alpha_function =
752            r300_translate_alpha_function(state->alpha_func) |
753            R300_FG_ALPHA_FUNC_ENABLE;
754
755        dsa->alpha_function |= float_to_ubyte(state->alpha_ref_value);
756        alpha_value_fp16 = _mesa_float_to_half(state->alpha_ref_value);
757    }
758
759    BEGIN_CB(&dsa->cb_begin, 8);
760    OUT_CB_REG_SEQ(R300_ZB_CNTL, 3);
761    OUT_CB(z_buffer_control);
762    OUT_CB(z_stencil_control);
763    OUT_CB(stencil_ref_mask);
764    OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, stencil_ref_bf);
765    OUT_CB_REG(R500_FG_ALPHA_VALUE, alpha_value_fp16);
766    END_CB;
767
768    BEGIN_CB(dsa->cb_zb_no_readwrite, 8);
769    OUT_CB_REG_SEQ(R300_ZB_CNTL, 3);
770    OUT_CB(0);
771    OUT_CB(0);
772    OUT_CB(0);
773    OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, 0);
774    OUT_CB_REG(R500_FG_ALPHA_VALUE, alpha_value_fp16);
775    END_CB;
776
777    return (void*)dsa;
778}
779
780static void r300_dsa_inject_stencilref(struct r300_context *r300)
781{
782    struct r300_dsa_state *dsa =
783            (struct r300_dsa_state*)r300->dsa_state.state;
784
785    if (!dsa)
786        return;
787
788    dsa->stencil_ref_mask =
789        (dsa->stencil_ref_mask & ~R300_STENCILREF_MASK) |
790        r300->stencil_ref.ref_value[0];
791    dsa->stencil_ref_bf =
792        (dsa->stencil_ref_bf & ~R300_STENCILREF_MASK) |
793        r300->stencil_ref.ref_value[1];
794}
795
796/* Bind DSA state. */
797static void r300_bind_dsa_state(struct pipe_context* pipe,
798                                void* state)
799{
800    struct r300_context* r300 = r300_context(pipe);
801
802    if (!state) {
803        return;
804    }
805
806    UPDATE_STATE(state, r300->dsa_state);
807
808    r300_mark_atom_dirty(r300, &r300->hyperz_state); /* Will be updated before the emission. */
809    r300_dsa_inject_stencilref(r300);
810}
811
812/* Free DSA state. */
813static void r300_delete_dsa_state(struct pipe_context* pipe,
814                                  void* state)
815{
816    FREE(state);
817}
818
819static void r300_set_stencil_ref(struct pipe_context* pipe,
820                                 const struct pipe_stencil_ref sr)
821{
822    struct r300_context* r300 = r300_context(pipe);
823
824    r300->stencil_ref = sr;
825
826    r300_dsa_inject_stencilref(r300);
827    r300_mark_atom_dirty(r300, &r300->dsa_state);
828}
829
830static void r300_print_fb_surf_info(struct pipe_surface *surf, unsigned index,
831                                    const char *binding)
832{
833    struct pipe_resource *tex = surf->texture;
834    struct r300_resource *rtex = r300_resource(tex);
835
836    fprintf(stderr,
837            "r300:   %s[%i] Dim: %ix%i, Firstlayer: %i, "
838            "Lastlayer: %i, Level: %i, Format: %s\n"
839
840            "r300:     TEX: Macro: %s, Micro: %s, "
841            "Dim: %ix%ix%i, LastLevel: %i, Format: %s\n",
842
843            binding, index, surf->width, surf->height,
844            surf->u.tex.first_layer, surf->u.tex.last_layer, surf->u.tex.level,
845            util_format_short_name(surf->format),
846
847            rtex->tex.macrotile[0] ? "YES" : " NO",
848            rtex->tex.microtile ? "YES" : " NO",
849            tex->width0, tex->height0, tex->depth0,
850            tex->last_level, util_format_short_name(surf->format));
851}
852
853void r300_mark_fb_state_dirty(struct r300_context *r300,
854                              enum r300_fb_state_change change)
855{
856    struct pipe_framebuffer_state *state = r300->fb_state.state;
857
858    r300_mark_atom_dirty(r300, &r300->gpu_flush);
859    r300_mark_atom_dirty(r300, &r300->fb_state);
860
861    /* What is marked as dirty depends on the enum r300_fb_state_change. */
862    if (change == R300_CHANGED_FB_STATE) {
863        r300_mark_atom_dirty(r300, &r300->aa_state);
864        r300_mark_atom_dirty(r300, &r300->dsa_state); /* for AlphaRef */
865        r300_set_blend_color(&r300->context, r300->blend_color_state.state);
866    }
867
868    if (change == R300_CHANGED_FB_STATE ||
869        change == R300_CHANGED_HYPERZ_FLAG) {
870        r300_mark_atom_dirty(r300, &r300->hyperz_state);
871    }
872
873    if (change == R300_CHANGED_FB_STATE ||
874        change == R300_CHANGED_MULTIWRITE) {
875        r300_mark_atom_dirty(r300, &r300->fb_state_pipelined);
876    }
877
878    /* Now compute the fb_state atom size. */
879    r300->fb_state.size = 2 + (8 * state->nr_cbufs);
880
881    if (r300->cbzb_clear)
882        r300->fb_state.size += 10;
883    else if (state->zsbuf) {
884        r300->fb_state.size += 10;
885        if (r300->hyperz_enabled)
886            r300->fb_state.size += 8;
887    }
888
889    if (r300->cmask_in_use) {
890        r300->fb_state.size += 6;
891        if (r300->screen->caps.is_r500 && r300->screen->info.drm_minor >= 29) {
892            r300->fb_state.size += 3;
893        }
894    }
895
896    /* The size of the rest of atoms stays the same. */
897}
898
899static void
900r300_set_framebuffer_state(struct pipe_context* pipe,
901                           const struct pipe_framebuffer_state* state)
902{
903    struct r300_context* r300 = r300_context(pipe);
904    struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
905    struct pipe_framebuffer_state *current_state = r300->fb_state.state;
906    unsigned max_width, max_height, i;
907    uint32_t zbuffer_bpp = 0;
908    boolean unlock_zbuffer = FALSE;
909
910    if (r300->screen->caps.is_r500) {
911        max_width = max_height = 4096;
912    } else if (r300->screen->caps.is_r400) {
913        max_width = max_height = 4021;
914    } else {
915        max_width = max_height = 2560;
916    }
917
918    if (state->width > max_width || state->height > max_height) {
919        fprintf(stderr, "r300: Implementation error: Render targets are too "
920        "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__);
921        return;
922    }
923
924    if (current_state->zsbuf && r300->zmask_in_use && !r300->locked_zbuffer) {
925        /* There is a zmask in use, what are we gonna do? */
926        if (state->zsbuf) {
927            if (!pipe_surface_equal(current_state->zsbuf, state->zsbuf)) {
928                /* Decompress the currently bound zbuffer before we bind another one. */
929                r300_decompress_zmask(r300);
930                r300->hiz_in_use = FALSE;
931            }
932        } else {
933            /* We don't bind another zbuffer, so lock the current one. */
934            pipe_surface_reference(&r300->locked_zbuffer, current_state->zsbuf);
935        }
936    } else if (r300->locked_zbuffer) {
937        /* We have a locked zbuffer now, what are we gonna do? */
938        if (state->zsbuf) {
939            if (!pipe_surface_equal(r300->locked_zbuffer, state->zsbuf)) {
940                /* We are binding some other zbuffer, so decompress the locked one,
941                 * it gets unlocked automatically. */
942                r300_decompress_zmask_locked_unsafe(r300);
943                r300->hiz_in_use = FALSE;
944            } else {
945                /* We are binding the locked zbuffer again, so unlock it. */
946                unlock_zbuffer = TRUE;
947            }
948        }
949    }
950    assert(state->zsbuf || (r300->locked_zbuffer && !unlock_zbuffer) || !r300->zmask_in_use);
951
952    /* If zsbuf is set from NULL to non-NULL or vice versa.. */
953    if (!!current_state->zsbuf != !!state->zsbuf) {
954        r300_mark_atom_dirty(r300, &r300->dsa_state);
955    }
956
957    util_copy_framebuffer_state(r300->fb_state.state, state);
958
959    /* Remove trailing NULL colorbuffers. */
960    while (current_state->nr_cbufs && !current_state->cbufs[current_state->nr_cbufs-1])
961        current_state->nr_cbufs--;
962
963    /* Set whether CMASK can be used. */
964    r300->cmask_in_use =
965        state->nr_cbufs == 1 && state->cbufs[0] &&
966        r300->screen->cmask_resource == state->cbufs[0]->texture;
967
968    /* Need to reset clamping or colormask. */
969    r300_mark_atom_dirty(r300, &r300->blend_state);
970
971    /* Re-swizzle the blend color. */
972    r300_set_blend_color(pipe, &((struct r300_blend_color_state*)r300->blend_color_state.state)->state);
973
974    if (unlock_zbuffer) {
975        pipe_surface_reference(&r300->locked_zbuffer, NULL);
976    }
977
978    r300_mark_fb_state_dirty(r300, R300_CHANGED_FB_STATE);
979
980    if (state->zsbuf) {
981        switch (util_format_get_blocksize(state->zsbuf->format)) {
982        case 2:
983            zbuffer_bpp = 16;
984            break;
985        case 4:
986            zbuffer_bpp = 24;
987            break;
988        }
989
990        /* Polygon offset depends on the zbuffer bit depth. */
991        if (r300->zbuffer_bpp != zbuffer_bpp) {
992            r300->zbuffer_bpp = zbuffer_bpp;
993
994            if (r300->polygon_offset_enabled)
995                r300_mark_atom_dirty(r300, &r300->rs_state);
996        }
997    }
998
999    r300->num_samples = util_framebuffer_get_num_samples(state);
1000
1001    /* Set up AA config. */
1002    if (r300->num_samples > 1) {
1003        switch (r300->num_samples) {
1004        case 2:
1005            aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE |
1006                            R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2;
1007            break;
1008        case 4:
1009            aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE |
1010                            R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4;
1011            break;
1012        case 6:
1013            aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE |
1014                            R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6;
1015            break;
1016        }
1017    } else {
1018        aa->aa_config = 0;
1019    }
1020
1021    if (DBG_ON(r300, DBG_FB)) {
1022        fprintf(stderr, "r300: set_framebuffer_state:\n");
1023        for (i = 0; i < state->nr_cbufs; i++) {
1024            if (state->cbufs[i])
1025                r300_print_fb_surf_info(state->cbufs[i], i, "CB");
1026        }
1027        if (state->zsbuf) {
1028            r300_print_fb_surf_info(state->zsbuf, 0, "ZB");
1029        }
1030    }
1031}
1032
1033/* Create fragment shader state. */
1034static void* r300_create_fs_state(struct pipe_context* pipe,
1035                                  const struct pipe_shader_state* shader)
1036{
1037    struct r300_fragment_shader* fs = NULL;
1038
1039    fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader);
1040
1041    /* Copy state directly into shader. */
1042    fs->state = *shader;
1043    fs->state.tokens = tgsi_dup_tokens(shader->tokens);
1044
1045    return (void*)fs;
1046}
1047
1048void r300_mark_fs_code_dirty(struct r300_context *r300)
1049{
1050    struct r300_fragment_shader* fs = r300_fs(r300);
1051
1052    r300_mark_atom_dirty(r300, &r300->fs);
1053    r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state);
1054    r300_mark_atom_dirty(r300, &r300->fs_constants);
1055    r300->fs.size = fs->shader->cb_code_size;
1056
1057    if (r300->screen->caps.is_r500) {
1058        r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 7;
1059        r300->fs_constants.size = fs->shader->externals_count * 4 + 3;
1060    } else {
1061        r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 5;
1062        r300->fs_constants.size = fs->shader->externals_count * 4 + 1;
1063    }
1064
1065    ((struct r300_constant_buffer*)r300->fs_constants.state)->remap_table =
1066            fs->shader->code.constants_remap_table;
1067}
1068
1069/* Bind fragment shader state. */
1070static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
1071{
1072    struct r300_context* r300 = r300_context(pipe);
1073    struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
1074
1075    if (!fs) {
1076        r300->fs.state = NULL;
1077        return;
1078    }
1079
1080    r300->fs.state = fs;
1081    r300->fs_status = FRAGMENT_SHADER_DIRTY;
1082
1083    r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */
1084}
1085
1086/* Delete fragment shader state. */
1087static void r300_delete_fs_state(struct pipe_context* pipe, void* shader)
1088{
1089    struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
1090    struct r300_fragment_shader_code *tmp, *ptr = fs->first;
1091
1092    while (ptr) {
1093        tmp = ptr;
1094        ptr = ptr->next;
1095        rc_constants_destroy(&tmp->code.constants);
1096        FREE(tmp->cb_code);
1097        FREE(tmp);
1098    }
1099    FREE((void*)fs->state.tokens);
1100    FREE(shader);
1101}
1102
1103static void r300_set_polygon_stipple(struct pipe_context* pipe,
1104                                     const struct pipe_poly_stipple* state)
1105{
1106}
1107
1108/* Create a new rasterizer state based on the CSO rasterizer state.
1109 *
1110 * This is a very large chunk of state, and covers most of the graphics
1111 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
1112 *
1113 * In a not entirely unironic sidenote, this state has nearly nothing to do
1114 * with the actual block on the Radeon called the rasterizer (RS). */
1115static void* r300_create_rs_state(struct pipe_context* pipe,
1116                                  const struct pipe_rasterizer_state* state)
1117{
1118    struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state);
1119    uint32_t vap_control_status;    /* R300_VAP_CNTL_STATUS: 0x2140 */
1120    uint32_t vap_clip_cntl;         /* R300_VAP_CLIP_CNTL: 0x221C */
1121    uint32_t point_size;            /* R300_GA_POINT_SIZE: 0x421c */
1122    uint32_t point_minmax;          /* R300_GA_POINT_MINMAX: 0x4230 */
1123    uint32_t line_control;          /* R300_GA_LINE_CNTL: 0x4234 */
1124    uint32_t polygon_offset_enable; /* R300_SU_POLY_OFFSET_ENABLE: 0x42b4 */
1125    uint32_t cull_mode;             /* R300_SU_CULL_MODE: 0x42b8 */
1126    uint32_t line_stipple_config;   /* R300_GA_LINE_STIPPLE_CONFIG: 0x4328 */
1127    uint32_t line_stipple_value;    /* R300_GA_LINE_STIPPLE_VALUE: 0x4260 */
1128    uint32_t polygon_mode;          /* R300_GA_POLY_MODE: 0x4288 */
1129    uint32_t clip_rule;             /* R300_SC_CLIP_RULE: 0x43D0 */
1130    uint32_t round_mode;            /* R300_GA_ROUND_MODE: 0x428c */
1131
1132    /* Point sprites texture coordinates, 0: lower left, 1: upper right */
1133    float point_texcoord_left = 0;  /* R300_GA_POINT_S0: 0x4200 */
1134    float point_texcoord_bottom = 0;/* R300_GA_POINT_T0: 0x4204 */
1135    float point_texcoord_right = 1; /* R300_GA_POINT_S1: 0x4208 */
1136    float point_texcoord_top = 0;   /* R300_GA_POINT_T1: 0x420c */
1137    boolean vclamp = !r300_context(pipe)->screen->caps.is_r500;
1138    CB_LOCALS;
1139
1140    /* Copy rasterizer state. */
1141    rs->rs = *state;
1142    rs->rs_draw = *state;
1143
1144    rs->rs.sprite_coord_enable = state->point_quad_rasterization *
1145                                 state->sprite_coord_enable;
1146
1147    /* Override some states for Draw. */
1148    rs->rs_draw.sprite_coord_enable = 0; /* We can do this in HW. */
1149    rs->rs_draw.offset_point = 0;
1150    rs->rs_draw.offset_line = 0;
1151    rs->rs_draw.offset_tri = 0;
1152    rs->rs_draw.offset_clamp = 0;
1153
1154#if UTIL_ARCH_LITTLE_ENDIAN
1155    vap_control_status = R300_VC_NO_SWAP;
1156#else
1157    vap_control_status = R300_VC_32BIT_SWAP;
1158#endif
1159
1160    /* If no TCL engine is present, turn off the HW TCL. */
1161    if (!r300_screen(pipe->screen)->caps.has_tcl) {
1162        vap_control_status |= R300_VAP_TCL_BYPASS;
1163    }
1164
1165    /* Point size width and height. */
1166    point_size =
1167        pack_float_16_6x(state->point_size) |
1168        (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT);
1169
1170    /* Point size clamping. */
1171    if (state->point_size_per_vertex) {
1172        /* Per-vertex point size.
1173         * Clamp to [0, max FB size] */
1174        float min_psiz = util_get_min_point_size(state);
1175        float max_psiz = pipe->screen->get_paramf(pipe->screen,
1176                                        PIPE_CAPF_MAX_POINT_WIDTH);
1177        point_minmax =
1178            (pack_float_16_6x(min_psiz) << R300_GA_POINT_MINMAX_MIN_SHIFT) |
1179            (pack_float_16_6x(max_psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT);
1180    } else {
1181        /* We cannot disable the point-size vertex output,
1182         * so clamp it. */
1183        float psiz = state->point_size;
1184        point_minmax =
1185            (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MIN_SHIFT) |
1186            (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT);
1187    }
1188
1189    /* Line control. */
1190    line_control = pack_float_16_6x(state->line_width) |
1191        R300_GA_LINE_CNTL_END_TYPE_COMP;
1192
1193    /* Enable polygon mode */
1194    polygon_mode = 0;
1195    if (state->fill_front != PIPE_POLYGON_MODE_FILL ||
1196        state->fill_back != PIPE_POLYGON_MODE_FILL) {
1197        polygon_mode = R300_GA_POLY_MODE_DUAL;
1198    }
1199
1200    /* Front face */
1201    if (state->front_ccw)
1202        cull_mode = R300_FRONT_FACE_CCW;
1203    else
1204        cull_mode = R300_FRONT_FACE_CW;
1205
1206    /* Polygon offset */
1207    polygon_offset_enable = 0;
1208    if (util_get_offset(state, state->fill_front)) {
1209       polygon_offset_enable |= R300_FRONT_ENABLE;
1210    }
1211    if (util_get_offset(state, state->fill_back)) {
1212       polygon_offset_enable |= R300_BACK_ENABLE;
1213    }
1214
1215    rs->polygon_offset_enable = polygon_offset_enable != 0;
1216
1217    /* Polygon mode */
1218    if (polygon_mode) {
1219       polygon_mode |=
1220          r300_translate_polygon_mode_front(state->fill_front);
1221       polygon_mode |=
1222          r300_translate_polygon_mode_back(state->fill_back);
1223    }
1224
1225    if (state->cull_face & PIPE_FACE_FRONT) {
1226        cull_mode |= R300_CULL_FRONT;
1227    }
1228    if (state->cull_face & PIPE_FACE_BACK) {
1229        cull_mode |= R300_CULL_BACK;
1230    }
1231
1232    if (state->line_stipple_enable) {
1233        line_stipple_config =
1234            R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE |
1235            (fui((float)state->line_stipple_factor) &
1236                R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK);
1237        /* XXX this might need to be scaled up */
1238        line_stipple_value = state->line_stipple_pattern;
1239    } else {
1240        line_stipple_config = 0;
1241        line_stipple_value = 0;
1242    }
1243
1244    if (state->flatshade) {
1245        rs->color_control = R300_SHADE_MODEL_FLAT;
1246    } else {
1247        rs->color_control = R300_SHADE_MODEL_SMOOTH;
1248    }
1249
1250    clip_rule = state->scissor ? 0xAAAA : 0xFFFF;
1251
1252    /* Point sprites coord mode */
1253    if (rs->rs.sprite_coord_enable) {
1254        switch (state->sprite_coord_mode) {
1255            case PIPE_SPRITE_COORD_UPPER_LEFT:
1256                point_texcoord_top = 0.0f;
1257                point_texcoord_bottom = 1.0f;
1258                break;
1259            case PIPE_SPRITE_COORD_LOWER_LEFT:
1260                point_texcoord_top = 1.0f;
1261                point_texcoord_bottom = 0.0f;
1262                break;
1263        }
1264    }
1265
1266    if (r300_screen(pipe->screen)->caps.has_tcl) {
1267       vap_clip_cntl = (state->clip_plane_enable & 63) |
1268                       R300_PS_UCP_MODE_CLIP_AS_TRIFAN;
1269    } else {
1270       vap_clip_cntl = R300_CLIP_DISABLE;
1271    }
1272
1273    /* Vertex color clamping. FP20 means no clamping. */
1274    round_mode =
1275      R300_GA_ROUND_MODE_GEOMETRY_ROUND_NEAREST |
1276      (!vclamp ? (R300_GA_ROUND_MODE_RGB_CLAMP_FP20 |
1277                  R300_GA_ROUND_MODE_ALPHA_CLAMP_FP20) : 0);
1278
1279    /* Build the main command buffer. */
1280    BEGIN_CB(rs->cb_main, RS_STATE_MAIN_SIZE);
1281    OUT_CB_REG(R300_VAP_CNTL_STATUS, vap_control_status);
1282    OUT_CB_REG(R300_VAP_CLIP_CNTL, vap_clip_cntl);
1283    OUT_CB_REG(R300_GA_POINT_SIZE, point_size);
1284    OUT_CB_REG_SEQ(R300_GA_POINT_MINMAX, 2);
1285    OUT_CB(point_minmax);
1286    OUT_CB(line_control);
1287    OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_ENABLE, 2);
1288    OUT_CB(polygon_offset_enable);
1289    rs->cull_mode_index = 11;
1290    OUT_CB(cull_mode);
1291    OUT_CB_REG(R300_GA_LINE_STIPPLE_CONFIG, line_stipple_config);
1292    OUT_CB_REG(R300_GA_LINE_STIPPLE_VALUE, line_stipple_value);
1293    OUT_CB_REG(R300_GA_POLY_MODE, polygon_mode);
1294    OUT_CB_REG(R300_GA_ROUND_MODE, round_mode);
1295    OUT_CB_REG(R300_SC_CLIP_RULE, clip_rule);
1296    OUT_CB_REG_SEQ(R300_GA_POINT_S0, 4);
1297    OUT_CB_32F(point_texcoord_left);
1298    OUT_CB_32F(point_texcoord_bottom);
1299    OUT_CB_32F(point_texcoord_right);
1300    OUT_CB_32F(point_texcoord_top);
1301    END_CB;
1302
1303    /* Build the two command buffers for polygon offset setup. */
1304    if (polygon_offset_enable) {
1305        float scale = state->offset_scale * 12;
1306        float offset = state->offset_units * 4;
1307
1308        BEGIN_CB(rs->cb_poly_offset_zb16, 5);
1309        OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4);
1310        OUT_CB_32F(scale);
1311        OUT_CB_32F(offset);
1312        OUT_CB_32F(scale);
1313        OUT_CB_32F(offset);
1314        END_CB;
1315
1316        offset = state->offset_units * 2;
1317
1318        BEGIN_CB(rs->cb_poly_offset_zb24, 5);
1319        OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4);
1320        OUT_CB_32F(scale);
1321        OUT_CB_32F(offset);
1322        OUT_CB_32F(scale);
1323        OUT_CB_32F(offset);
1324        END_CB;
1325    }
1326
1327    return (void*)rs;
1328}
1329
1330/* Bind rasterizer state. */
1331static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
1332{
1333    struct r300_context* r300 = r300_context(pipe);
1334    struct r300_rs_state* rs = (struct r300_rs_state*)state;
1335    int last_sprite_coord_enable = r300->sprite_coord_enable;
1336    boolean last_two_sided_color = r300->two_sided_color;
1337    boolean last_msaa_enable = r300->msaa_enable;
1338    boolean last_flatshade = r300->flatshade;
1339    boolean last_clip_halfz = r300->clip_halfz;
1340
1341    if (r300->draw && rs) {
1342        draw_set_rasterizer_state(r300->draw, &rs->rs_draw, state);
1343    }
1344
1345    if (rs) {
1346        r300->polygon_offset_enabled = rs->polygon_offset_enable;
1347        r300->sprite_coord_enable = rs->rs.sprite_coord_enable;
1348        r300->two_sided_color = rs->rs.light_twoside;
1349        r300->msaa_enable = rs->rs.multisample;
1350        r300->flatshade = rs->rs.flatshade;
1351        r300->clip_halfz = rs->rs.clip_halfz;
1352    } else {
1353        r300->polygon_offset_enabled = FALSE;
1354        r300->sprite_coord_enable = 0;
1355        r300->two_sided_color = FALSE;
1356        r300->msaa_enable = FALSE;
1357        r300->flatshade = FALSE;
1358        r300->clip_halfz = FALSE;
1359    }
1360
1361    UPDATE_STATE(state, r300->rs_state);
1362    r300->rs_state.size = RS_STATE_MAIN_SIZE + (r300->polygon_offset_enabled ? 5 : 0);
1363
1364    if (last_sprite_coord_enable != r300->sprite_coord_enable ||
1365        last_two_sided_color != r300->two_sided_color ||
1366        last_flatshade != r300->flatshade) {
1367        r300_mark_atom_dirty(r300, &r300->rs_block_state);
1368    }
1369
1370    if (last_msaa_enable != r300->msaa_enable) {
1371        if (r300->alpha_to_coverage) {
1372            r300_mark_atom_dirty(r300, &r300->dsa_state);
1373        }
1374
1375        if (r300->alpha_to_one &&
1376            r300->fs_status == FRAGMENT_SHADER_VALID) {
1377            r300->fs_status = FRAGMENT_SHADER_MAYBE_DIRTY;
1378        }
1379    }
1380
1381    if (r300->screen->caps.has_tcl && last_clip_halfz != r300->clip_halfz) {
1382        r300_mark_atom_dirty(r300, &r300->vs_state);
1383    }
1384}
1385
1386/* Free rasterizer state. */
1387static void r300_delete_rs_state(struct pipe_context* pipe, void* state)
1388{
1389    FREE(state);
1390}
1391
1392static void*
1393        r300_create_sampler_state(struct pipe_context* pipe,
1394                                  const struct pipe_sampler_state* state)
1395{
1396    struct r300_context* r300 = r300_context(pipe);
1397    struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state);
1398    boolean is_r500 = r300->screen->caps.is_r500;
1399    int lod_bias;
1400
1401    sampler->state = *state;
1402
1403    /* r300 doesn't handle CLAMP and MIRROR_CLAMP correctly when either MAG
1404     * or MIN filter is NEAREST. Since texwrap produces same results
1405     * for CLAMP and CLAMP_TO_EDGE, we use them instead. */
1406    if (sampler->state.min_img_filter == PIPE_TEX_FILTER_NEAREST ||
1407        sampler->state.mag_img_filter == PIPE_TEX_FILTER_NEAREST) {
1408        /* Wrap S. */
1409        if (sampler->state.wrap_s == PIPE_TEX_WRAP_CLAMP)
1410            sampler->state.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1411        else if (sampler->state.wrap_s == PIPE_TEX_WRAP_MIRROR_CLAMP)
1412            sampler->state.wrap_s = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE;
1413
1414        /* Wrap T. */
1415        if (sampler->state.wrap_t == PIPE_TEX_WRAP_CLAMP)
1416            sampler->state.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1417        else if (sampler->state.wrap_t == PIPE_TEX_WRAP_MIRROR_CLAMP)
1418            sampler->state.wrap_t = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE;
1419
1420        /* Wrap R. */
1421        if (sampler->state.wrap_r == PIPE_TEX_WRAP_CLAMP)
1422            sampler->state.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1423        else if (sampler->state.wrap_r == PIPE_TEX_WRAP_MIRROR_CLAMP)
1424            sampler->state.wrap_r = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE;
1425    }
1426
1427    sampler->filter0 |=
1428        (r300_translate_wrap(sampler->state.wrap_s) << R300_TX_WRAP_S_SHIFT) |
1429        (r300_translate_wrap(sampler->state.wrap_t) << R300_TX_WRAP_T_SHIFT) |
1430        (r300_translate_wrap(sampler->state.wrap_r) << R300_TX_WRAP_R_SHIFT);
1431
1432    sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter,
1433                                                   state->mag_img_filter,
1434                                                   state->min_mip_filter,
1435                                                   state->max_anisotropy > 1);
1436
1437    sampler->filter0 |= r300_anisotropy(state->max_anisotropy);
1438
1439    /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
1440    /* We must pass these to the merge function to clamp them properly. */
1441    sampler->min_lod = (unsigned)MAX2(state->min_lod, 0);
1442    sampler->max_lod = (unsigned)MAX2(ceilf(state->max_lod), 0);
1443
1444    lod_bias = CLAMP((int)(state->lod_bias * 32 + 1), -(1 << 9), (1 << 9) - 1);
1445
1446    sampler->filter1 |= (lod_bias << R300_LOD_BIAS_SHIFT) & R300_LOD_BIAS_MASK;
1447
1448    /* This is very high quality anisotropic filtering for R5xx.
1449     * It's good for benchmarking the performance of texturing but
1450     * in practice we don't want to slow down the driver because it's
1451     * a pretty good performance killer. Feel free to play with it. */
1452    if (DBG_ON(r300, DBG_ANISOHQ) && is_r500) {
1453        sampler->filter1 |= r500_anisotropy(state->max_anisotropy);
1454    }
1455
1456    /* R500-specific fixups and optimizations */
1457    if (r300->screen->caps.is_r500) {
1458        sampler->filter1 |= R500_BORDER_FIX;
1459    }
1460
1461    return (void*)sampler;
1462}
1463
1464static void r300_bind_sampler_states(struct pipe_context* pipe,
1465                                     enum pipe_shader_type shader,
1466                                     unsigned start, unsigned count,
1467                                     void** states)
1468{
1469    struct r300_context* r300 = r300_context(pipe);
1470    struct r300_textures_state* state =
1471        (struct r300_textures_state*)r300->textures_state.state;
1472    unsigned tex_units = r300->screen->caps.num_tex_units;
1473
1474    assert(start == 0);
1475
1476    if (shader != PIPE_SHADER_FRAGMENT)
1477       return;
1478
1479    if (count > tex_units)
1480       return;
1481
1482    memcpy(state->sampler_states, states, sizeof(void*) * count);
1483    state->sampler_state_count = count;
1484
1485    r300_mark_atom_dirty(r300, &r300->textures_state);
1486}
1487
1488static void r300_delete_sampler_state(struct pipe_context* pipe, void* state)
1489{
1490    FREE(state);
1491}
1492
1493static uint32_t r300_assign_texture_cache_region(unsigned index, unsigned num)
1494{
1495    /* This looks like a hack, but I believe it's suppose to work like
1496     * that. To illustrate how this works, let's assume you have 5 textures.
1497     * From docs, 5 and the successive numbers are:
1498     *
1499     * FOURTH_1     = 5
1500     * FOURTH_2     = 6
1501     * FOURTH_3     = 7
1502     * EIGHTH_0     = 8
1503     * EIGHTH_1     = 9
1504     *
1505     * First 3 textures will get 3/4 of size of the cache, divided evenly
1506     * between them. The last 1/4 of the cache must be divided between
1507     * the last 2 textures, each will therefore get 1/8 of the cache.
1508     * Why not just to use "5 + texture_index" ?
1509     *
1510     * This simple trick works for all "num" <= 16.
1511     */
1512    if (num <= 1)
1513        return R300_TX_CACHE(R300_TX_CACHE_WHOLE);
1514    else
1515        return R300_TX_CACHE(num + index);
1516}
1517
1518static void r300_set_sampler_views(struct pipe_context* pipe,
1519                                   enum pipe_shader_type shader,
1520                                   unsigned start, unsigned count,
1521                                   unsigned unbind_num_trailing_slots,
1522                                   bool take_ownership,
1523                                   struct pipe_sampler_view** views)
1524{
1525    struct r300_context* r300 = r300_context(pipe);
1526    struct r300_textures_state* state =
1527        (struct r300_textures_state*)r300->textures_state.state;
1528    struct r300_resource *texture;
1529    unsigned i, real_num_views = 0, view_index = 0;
1530    unsigned tex_units = r300->screen->caps.num_tex_units;
1531    boolean dirty_tex = FALSE;
1532
1533    assert(start == 0);  /* non-zero not handled yet */
1534
1535    if (shader != PIPE_SHADER_FRAGMENT || count > tex_units) {
1536       if (take_ownership) {
1537          for (unsigned i = 0; i < count; i++) {
1538             struct pipe_sampler_view *view = views[i];
1539             pipe_sampler_view_reference(&view, NULL);
1540          }
1541       }
1542       return;
1543    }
1544
1545    /* Calculate the real number of views. */
1546    for (i = 0; i < count; i++) {
1547        if (views[i])
1548            real_num_views++;
1549    }
1550
1551    for (i = 0; i < count; i++) {
1552        if (take_ownership) {
1553            pipe_sampler_view_reference(
1554                    (struct pipe_sampler_view**)&state->sampler_views[i], NULL);
1555            state->sampler_views[i] = (struct r300_sampler_view*)views[i];
1556        } else {
1557            pipe_sampler_view_reference(
1558                    (struct pipe_sampler_view**)&state->sampler_views[i],
1559                    views[i]);
1560        }
1561
1562        if (!views[i]) {
1563            continue;
1564        }
1565
1566        /* A new sampler view (= texture)... */
1567        dirty_tex = TRUE;
1568
1569        /* Set the texrect factor in the fragment shader.
1570             * Needed for RECT and NPOT fallback. */
1571        texture = r300_resource(views[i]->texture);
1572        if (texture->tex.is_npot) {
1573            r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state);
1574        }
1575
1576        state->sampler_views[i]->texcache_region =
1577                r300_assign_texture_cache_region(view_index, real_num_views);
1578        view_index++;
1579    }
1580
1581    for (i = count; i < tex_units; i++) {
1582        if (state->sampler_views[i]) {
1583            pipe_sampler_view_reference(
1584                    (struct pipe_sampler_view**)&state->sampler_views[i],
1585                    NULL);
1586        }
1587    }
1588
1589    state->sampler_view_count = count;
1590
1591    r300_mark_atom_dirty(r300, &r300->textures_state);
1592
1593    if (dirty_tex) {
1594        r300_mark_atom_dirty(r300, &r300->texture_cache_inval);
1595    }
1596}
1597
1598struct pipe_sampler_view *
1599r300_create_sampler_view_custom(struct pipe_context *pipe,
1600                         struct pipe_resource *texture,
1601                         const struct pipe_sampler_view *templ,
1602                         unsigned width0_override,
1603                         unsigned height0_override)
1604{
1605    struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view);
1606    struct r300_resource *tex = r300_resource(texture);
1607    boolean is_r500 = r300_screen(pipe->screen)->caps.is_r500;
1608    boolean dxtc_swizzle = r300_screen(pipe->screen)->caps.dxtc_swizzle;
1609
1610    if (view) {
1611        unsigned hwformat;
1612
1613        view->base = *templ;
1614        view->base.reference.count = 1;
1615        view->base.context = pipe;
1616        view->base.texture = NULL;
1617        pipe_resource_reference(&view->base.texture, texture);
1618
1619	view->width0_override = width0_override;
1620	view->height0_override = height0_override;
1621        view->swizzle[0] = templ->swizzle_r;
1622        view->swizzle[1] = templ->swizzle_g;
1623        view->swizzle[2] = templ->swizzle_b;
1624        view->swizzle[3] = templ->swizzle_a;
1625
1626        hwformat = r300_translate_texformat(templ->format,
1627                                            view->swizzle,
1628                                            is_r500,
1629                                            dxtc_swizzle);
1630
1631        if (hwformat == ~0) {
1632            fprintf(stderr, "r300: Oops. Got unsupported format %s in %s.\n",
1633                    util_format_short_name(templ->format), __func__);
1634        }
1635        assert(hwformat != ~0);
1636
1637	r300_texture_setup_format_state(r300_screen(pipe->screen), tex,
1638					templ->format, 0,
1639	                                width0_override, height0_override,
1640					&view->format);
1641        view->format.format1 |= hwformat;
1642        if (is_r500) {
1643            view->format.format2 |= r500_tx_format_msb_bit(templ->format);
1644        }
1645    }
1646
1647    return (struct pipe_sampler_view*)view;
1648}
1649
1650static struct pipe_sampler_view *
1651r300_create_sampler_view(struct pipe_context *pipe,
1652                         struct pipe_resource *texture,
1653                         const struct pipe_sampler_view *templ)
1654{
1655    return r300_create_sampler_view_custom(pipe, texture, templ,
1656                                           r300_resource(texture)->tex.width0,
1657                                           r300_resource(texture)->tex.height0);
1658}
1659
1660
1661static void
1662r300_sampler_view_destroy(struct pipe_context *pipe,
1663                          struct pipe_sampler_view *view)
1664{
1665   pipe_resource_reference(&view->texture, NULL);
1666   FREE(view);
1667}
1668
1669static void r300_set_sample_mask(struct pipe_context *pipe,
1670                                 unsigned mask)
1671{
1672    struct r300_context* r300 = r300_context(pipe);
1673
1674    *((unsigned*)r300->sample_mask.state) = mask;
1675
1676    r300_mark_atom_dirty(r300, &r300->sample_mask);
1677}
1678
1679static void r300_set_scissor_states(struct pipe_context* pipe,
1680                                    unsigned start_slot,
1681                                    unsigned num_scissors,
1682                                    const struct pipe_scissor_state* state)
1683{
1684    struct r300_context* r300 = r300_context(pipe);
1685
1686    memcpy(r300->scissor_state.state, state,
1687        sizeof(struct pipe_scissor_state));
1688
1689    r300_mark_atom_dirty(r300, &r300->scissor_state);
1690}
1691
1692static void r300_set_viewport_states(struct pipe_context* pipe,
1693                                     unsigned start_slot,
1694                                     unsigned num_viewports,
1695                                     const struct pipe_viewport_state* state)
1696{
1697    struct r300_context* r300 = r300_context(pipe);
1698    struct r300_viewport_state* viewport =
1699        (struct r300_viewport_state*)r300->viewport_state.state;
1700
1701    r300->viewport = *state;
1702
1703    if (r300->draw) {
1704        draw_set_viewport_states(r300->draw, start_slot, num_viewports, state);
1705        viewport->vte_control = R300_VTX_XY_FMT | R300_VTX_Z_FMT;
1706        return;
1707    }
1708
1709    /* Do the transform in HW. */
1710    viewport->vte_control = R300_VTX_W0_FMT;
1711
1712    if (state->scale[0] != 1.0f) {
1713        viewport->xscale = state->scale[0];
1714        viewport->vte_control |= R300_VPORT_X_SCALE_ENA;
1715    }
1716    if (state->scale[1] != 1.0f) {
1717        viewport->yscale = state->scale[1];
1718        viewport->vte_control |= R300_VPORT_Y_SCALE_ENA;
1719    }
1720    if (state->scale[2] != 1.0f) {
1721        viewport->zscale = state->scale[2];
1722        viewport->vte_control |= R300_VPORT_Z_SCALE_ENA;
1723    }
1724    if (state->translate[0] != 0.0f) {
1725        viewport->xoffset = state->translate[0];
1726        viewport->vte_control |= R300_VPORT_X_OFFSET_ENA;
1727    }
1728    if (state->translate[1] != 0.0f) {
1729        viewport->yoffset = state->translate[1];
1730        viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA;
1731    }
1732    if (state->translate[2] != 0.0f) {
1733        viewport->zoffset = state->translate[2];
1734        viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA;
1735    }
1736
1737    r300_mark_atom_dirty(r300, &r300->viewport_state);
1738    if (r300->fs.state && r300_fs(r300)->shader &&
1739        r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) {
1740        r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state);
1741    }
1742}
1743
1744static void r300_set_vertex_buffers_hwtcl(struct pipe_context* pipe,
1745                                    unsigned start_slot, unsigned count,
1746                                    unsigned unbind_num_trailing_slots,
1747                                    bool take_ownership,
1748                                    const struct pipe_vertex_buffer* buffers)
1749{
1750    struct r300_context* r300 = r300_context(pipe);
1751
1752    util_set_vertex_buffers_count(r300->vertex_buffer,
1753                                  &r300->nr_vertex_buffers,
1754                                  buffers, start_slot, count,
1755                                  unbind_num_trailing_slots, take_ownership);
1756
1757    /* There must be at least one vertex buffer set, otherwise it locks up. */
1758    if (!r300->nr_vertex_buffers) {
1759        util_set_vertex_buffers_count(r300->vertex_buffer,
1760                                      &r300->nr_vertex_buffers,
1761                                      &r300->dummy_vb, 0, 1, 0, false);
1762    }
1763
1764    r300->vertex_arrays_dirty = TRUE;
1765}
1766
1767static void r300_set_vertex_buffers_swtcl(struct pipe_context* pipe,
1768                                    unsigned start_slot, unsigned count,
1769                                    unsigned unbind_num_trailing_slots,
1770                                    bool take_ownership,
1771                                    const struct pipe_vertex_buffer* buffers)
1772{
1773    struct r300_context* r300 = r300_context(pipe);
1774    unsigned i;
1775
1776    util_set_vertex_buffers_count(r300->vertex_buffer,
1777                                  &r300->nr_vertex_buffers,
1778                                  buffers, start_slot, count,
1779                                  unbind_num_trailing_slots, take_ownership);
1780    draw_set_vertex_buffers(r300->draw, start_slot, count,
1781                            unbind_num_trailing_slots, buffers);
1782
1783    if (!buffers)
1784        return;
1785
1786    for (i = 0; i < count; i++) {
1787        if (buffers[i].is_user_buffer) {
1788            draw_set_mapped_vertex_buffer(r300->draw, start_slot + i,
1789                                          buffers[i].buffer.user, ~0);
1790        } else if (buffers[i].buffer.resource) {
1791            draw_set_mapped_vertex_buffer(r300->draw, start_slot + i,
1792                                          r300_resource(buffers[i].buffer.resource)->malloced_buffer, ~0);
1793        }
1794    }
1795}
1796
1797/* Initialize the PSC tables. */
1798static void r300_vertex_psc(struct r300_vertex_element_state *velems)
1799{
1800    struct r300_vertex_stream_state *vstream = &velems->vertex_stream;
1801    uint16_t type, swizzle;
1802    enum pipe_format format;
1803    unsigned i;
1804
1805    /* Vertex shaders have no semantics on their inputs,
1806     * so PSC should just route stuff based on the vertex elements,
1807     * and not on attrib information. */
1808    for (i = 0; i < velems->count; i++) {
1809        format = velems->velem[i].src_format;
1810
1811        type = r300_translate_vertex_data_type(format);
1812        if (type == R300_INVALID_FORMAT) {
1813            fprintf(stderr, "r300: Bad vertex format %s.\n",
1814                    util_format_short_name(format));
1815            assert(0);
1816            abort();
1817        }
1818
1819        type |= i << R300_DST_VEC_LOC_SHIFT;
1820        swizzle = r300_translate_vertex_data_swizzle(format);
1821
1822        if (i & 1) {
1823            vstream->vap_prog_stream_cntl[i >> 1] |= type << 16;
1824            vstream->vap_prog_stream_cntl_ext[i >> 1] |= (uint32_t)swizzle << 16;
1825        } else {
1826            vstream->vap_prog_stream_cntl[i >> 1] |= type;
1827            vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
1828        }
1829    }
1830
1831    /* Set the last vector in the PSC. */
1832    if (i) {
1833        i -= 1;
1834    }
1835    vstream->vap_prog_stream_cntl[i >> 1] |=
1836        (R300_LAST_VEC << (i & 1 ? 16 : 0));
1837
1838    vstream->count = (i >> 1) + 1;
1839}
1840
1841static void* r300_create_vertex_elements_state(struct pipe_context* pipe,
1842                                               unsigned count,
1843                                               const struct pipe_vertex_element* attribs)
1844{
1845    struct r300_vertex_element_state *velems;
1846    unsigned i;
1847    struct pipe_vertex_element dummy_attrib = {0};
1848
1849    /* R300 Programmable Stream Control (PSC) doesn't support 0 vertex elements. */
1850    if (!count) {
1851        dummy_attrib.src_format = PIPE_FORMAT_R8G8B8A8_UNORM;
1852        attribs = &dummy_attrib;
1853        count = 1;
1854    } else if (count > 16) {
1855        fprintf(stderr, "r300: More than 16 vertex elements are not supported,"
1856                " requested %i, using 16.\n", count);
1857        count = 16;
1858    }
1859
1860    velems = CALLOC_STRUCT(r300_vertex_element_state);
1861    if (!velems)
1862        return NULL;
1863
1864    velems->count = count;
1865    memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count);
1866
1867    if (r300_screen(pipe->screen)->caps.has_tcl) {
1868        /* Setup PSC.
1869         * The unused components will be replaced by (..., 0, 1). */
1870        r300_vertex_psc(velems);
1871
1872        for (i = 0; i < count; i++) {
1873            velems->format_size[i] =
1874                align(util_format_get_blocksize(velems->velem[i].src_format), 4);
1875            velems->vertex_size_dwords += velems->format_size[i] / 4;
1876        }
1877    }
1878
1879    return velems;
1880}
1881
1882static void r300_bind_vertex_elements_state(struct pipe_context *pipe,
1883                                            void *state)
1884{
1885    struct r300_context *r300 = r300_context(pipe);
1886    struct r300_vertex_element_state *velems = state;
1887
1888    if (!velems) {
1889        return;
1890    }
1891
1892    r300->velems = velems;
1893
1894    if (r300->draw) {
1895        draw_set_vertex_elements(r300->draw, velems->count, velems->velem);
1896        return;
1897    }
1898
1899    UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state);
1900    r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2;
1901    r300->vertex_arrays_dirty = TRUE;
1902}
1903
1904static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state)
1905{
1906    FREE(state);
1907}
1908
1909static void* r300_create_vs_state(struct pipe_context* pipe,
1910                                  const struct pipe_shader_state* shader)
1911{
1912    struct r300_context* r300 = r300_context(pipe);
1913    struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader);
1914
1915    /* Copy state directly into shader. */
1916    vs->state = *shader;
1917    vs->state.tokens = tgsi_dup_tokens(shader->tokens);
1918
1919    if (r300->screen->caps.has_tcl) {
1920        r300_init_vs_outputs(r300, vs);
1921        r300_translate_vertex_shader(r300, vs);
1922    } else {
1923        r300_draw_init_vertex_shader(r300, vs);
1924    }
1925
1926    return vs;
1927}
1928
1929static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
1930{
1931    struct r300_context* r300 = r300_context(pipe);
1932    struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
1933
1934    if (!vs) {
1935        r300->vs_state.state = NULL;
1936        return;
1937    }
1938    if (vs == r300->vs_state.state) {
1939        return;
1940    }
1941    r300->vs_state.state = vs;
1942
1943    /* The majority of the RS block bits is dependent on the vertex shader. */
1944    r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */
1945
1946    if (r300->screen->caps.has_tcl) {
1947        unsigned fc_op_dwords = r300->screen->caps.is_r500 ? 3 : 2;
1948        r300_mark_atom_dirty(r300, &r300->vs_state);
1949        r300->vs_state.size = vs->code.length + 9 +
1950			(R300_VS_MAX_FC_OPS * fc_op_dwords + 4);
1951
1952        r300_mark_atom_dirty(r300, &r300->vs_constants);
1953        r300->vs_constants.size =
1954                2 +
1955                (vs->externals_count ? vs->externals_count * 4 + 3 : 0) +
1956                (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0);
1957
1958        ((struct r300_constant_buffer*)r300->vs_constants.state)->remap_table =
1959                vs->code.constants_remap_table;
1960
1961        r300_mark_atom_dirty(r300, &r300->pvs_flush);
1962    } else {
1963        draw_bind_vertex_shader(r300->draw,
1964                (struct draw_vertex_shader*)vs->draw_vs);
1965    }
1966}
1967
1968static void r300_delete_vs_state(struct pipe_context* pipe, void* shader)
1969{
1970    struct r300_context* r300 = r300_context(pipe);
1971    struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
1972
1973    if (r300->screen->caps.has_tcl) {
1974        rc_constants_destroy(&vs->code.constants);
1975        FREE(vs->code.constants_remap_table);
1976    } else {
1977        draw_delete_vertex_shader(r300->draw,
1978                (struct draw_vertex_shader*)vs->draw_vs);
1979    }
1980
1981    FREE((void*)vs->state.tokens);
1982    FREE(shader);
1983}
1984
1985static void r300_set_constant_buffer(struct pipe_context *pipe,
1986                                     enum pipe_shader_type shader, uint index,
1987                                     bool take_ownership,
1988                                     const struct pipe_constant_buffer *cb)
1989{
1990    struct r300_context* r300 = r300_context(pipe);
1991    struct r300_constant_buffer *cbuf;
1992    uint32_t *mapped;
1993
1994    if (!cb || (!cb->buffer && !cb->user_buffer))
1995        return;
1996
1997    switch (shader) {
1998        case PIPE_SHADER_VERTEX:
1999            cbuf = (struct r300_constant_buffer*)r300->vs_constants.state;
2000            break;
2001        case PIPE_SHADER_FRAGMENT:
2002            cbuf = (struct r300_constant_buffer*)r300->fs_constants.state;
2003            break;
2004        default:
2005            return;
2006    }
2007
2008
2009    if (cb->user_buffer)
2010        mapped = (uint32_t*)cb->user_buffer;
2011    else {
2012        struct r300_resource *rbuf = r300_resource(cb->buffer);
2013
2014        if (rbuf && rbuf->malloced_buffer)
2015            mapped = (uint32_t*)rbuf->malloced_buffer;
2016        else
2017            return;
2018    }
2019
2020    if (shader == PIPE_SHADER_FRAGMENT ||
2021        (shader == PIPE_SHADER_VERTEX && r300->screen->caps.has_tcl)) {
2022        cbuf->ptr = mapped;
2023    }
2024
2025    if (shader == PIPE_SHADER_VERTEX) {
2026        if (r300->screen->caps.has_tcl) {
2027            struct r300_vertex_shader *vs =
2028                    (struct r300_vertex_shader*)r300->vs_state.state;
2029
2030            if (!vs) {
2031                cbuf->buffer_base = 0;
2032                return;
2033            }
2034
2035            cbuf->buffer_base = r300->vs_const_base;
2036            r300->vs_const_base += vs->code.constants.Count;
2037            if (r300->vs_const_base > R500_MAX_PVS_CONST_VECS) {
2038                r300->vs_const_base = vs->code.constants.Count;
2039                cbuf->buffer_base = 0;
2040                r300_mark_atom_dirty(r300, &r300->pvs_flush);
2041            }
2042            r300_mark_atom_dirty(r300, &r300->vs_constants);
2043        } else if (r300->draw) {
2044            draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX,
2045                0, mapped, cb->buffer_size);
2046        }
2047    } else if (shader == PIPE_SHADER_FRAGMENT) {
2048        r300_mark_atom_dirty(r300, &r300->fs_constants);
2049    }
2050}
2051
2052static void r300_texture_barrier(struct pipe_context *pipe, unsigned flags)
2053{
2054    struct r300_context *r300 = r300_context(pipe);
2055
2056    r300_mark_atom_dirty(r300, &r300->gpu_flush);
2057    r300_mark_atom_dirty(r300, &r300->texture_cache_inval);
2058}
2059
2060static void r300_memory_barrier(struct pipe_context *pipe, unsigned flags)
2061{
2062}
2063
2064void r300_init_state_functions(struct r300_context* r300)
2065{
2066    r300->context.create_blend_state = r300_create_blend_state;
2067    r300->context.bind_blend_state = r300_bind_blend_state;
2068    r300->context.delete_blend_state = r300_delete_blend_state;
2069
2070    r300->context.set_blend_color = r300_set_blend_color;
2071
2072    r300->context.set_clip_state = r300_set_clip_state;
2073    r300->context.set_sample_mask = r300_set_sample_mask;
2074
2075    r300->context.set_constant_buffer = r300_set_constant_buffer;
2076
2077    r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state;
2078    r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state;
2079    r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state;
2080
2081    r300->context.set_stencil_ref = r300_set_stencil_ref;
2082
2083    r300->context.set_framebuffer_state = r300_set_framebuffer_state;
2084
2085    r300->context.create_fs_state = r300_create_fs_state;
2086    r300->context.bind_fs_state = r300_bind_fs_state;
2087    r300->context.delete_fs_state = r300_delete_fs_state;
2088
2089    r300->context.set_polygon_stipple = r300_set_polygon_stipple;
2090
2091    r300->context.create_rasterizer_state = r300_create_rs_state;
2092    r300->context.bind_rasterizer_state = r300_bind_rs_state;
2093    r300->context.delete_rasterizer_state = r300_delete_rs_state;
2094
2095    r300->context.create_sampler_state = r300_create_sampler_state;
2096    r300->context.bind_sampler_states = r300_bind_sampler_states;
2097    r300->context.delete_sampler_state = r300_delete_sampler_state;
2098
2099    r300->context.set_sampler_views = r300_set_sampler_views;
2100    r300->context.create_sampler_view = r300_create_sampler_view;
2101    r300->context.sampler_view_destroy = r300_sampler_view_destroy;
2102
2103    r300->context.set_scissor_states = r300_set_scissor_states;
2104
2105    r300->context.set_viewport_states = r300_set_viewport_states;
2106
2107    if (r300->screen->caps.has_tcl) {
2108        r300->context.set_vertex_buffers = r300_set_vertex_buffers_hwtcl;
2109    } else {
2110        r300->context.set_vertex_buffers = r300_set_vertex_buffers_swtcl;
2111    }
2112
2113    r300->context.create_vertex_elements_state = r300_create_vertex_elements_state;
2114    r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state;
2115    r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state;
2116
2117    r300->context.create_vs_state = r300_create_vs_state;
2118    r300->context.bind_vs_state = r300_bind_vs_state;
2119    r300->context.delete_vs_state = r300_delete_vs_state;
2120
2121    r300->context.texture_barrier = r300_texture_barrier;
2122    r300->context.memory_barrier = r300_memory_barrier;
2123}
2124