r300_state.c revision 560ad7e59940423a696e93e62ec3c41b1e553421
1/*
2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2009 Marek Olšák <maraeo@gmail.com>
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
23
24#include "draw/draw_context.h"
25
26#include "util/u_framebuffer.h"
27#include "util/u_math.h"
28#include "util/u_mm.h"
29#include "util/u_memory.h"
30#include "util/u_pack_color.h"
31
32#include "tgsi/tgsi_parse.h"
33
34#include "pipe/p_config.h"
35
36#include "r300_cb.h"
37#include "r300_context.h"
38#include "r300_emit.h"
39#include "r300_reg.h"
40#include "r300_screen.h"
41#include "r300_screen_buffer.h"
42#include "r300_state_inlines.h"
43#include "r300_fs.h"
44#include "r300_texture.h"
45#include "r300_vs.h"
46#include "r300_winsys.h"
47#include "r300_hyperz.h"
48
49/* r300_state: Functions used to intialize state context by translating
50 * Gallium state objects into semi-native r300 state objects. */
51
52#define UPDATE_STATE(cso, atom) \
53    if (cso != atom.state) { \
54        atom.state = cso;    \
55        atom.dirty = TRUE;   \
56    }
57
58static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA,
59                                            unsigned dstRGB, unsigned dstA)
60{
61    /* If the blend equation is ADD or REVERSE_SUBTRACT,
62     * SRC_ALPHA == 0, and the following state is set, the colorbuffer
63     * will not be changed.
64     * Notice that the dst factors are the src factors inverted. */
65    return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
66            srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
67            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
68           (srcA == PIPE_BLENDFACTOR_SRC_COLOR ||
69            srcA == PIPE_BLENDFACTOR_SRC_ALPHA ||
70            srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
71            srcA == PIPE_BLENDFACTOR_ZERO) &&
72           (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
73            dstRGB == PIPE_BLENDFACTOR_ONE) &&
74           (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
75            dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
76            dstA == PIPE_BLENDFACTOR_ONE);
77}
78
79static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA,
80                                            unsigned dstRGB, unsigned dstA)
81{
82    /* If the blend equation is ADD or REVERSE_SUBTRACT,
83     * SRC_ALPHA == 1, and the following state is set, the colorbuffer
84     * will not be changed.
85     * Notice that the dst factors are the src factors inverted. */
86    return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
87            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
88           (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
89            srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
90            srcA == PIPE_BLENDFACTOR_ZERO) &&
91           (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
92            dstRGB == PIPE_BLENDFACTOR_ONE) &&
93           (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
94            dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
95            dstA == PIPE_BLENDFACTOR_ONE);
96}
97
98static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA,
99                                            unsigned dstRGB, unsigned dstA)
100{
101    /* If the blend equation is ADD or REVERSE_SUBTRACT,
102     * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer
103     * will not be changed.
104     * Notice that the dst factors are the src factors inverted. */
105    return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
106            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
107           (srcA == PIPE_BLENDFACTOR_ZERO) &&
108           (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
109            dstRGB == PIPE_BLENDFACTOR_ONE) &&
110           (dstA == PIPE_BLENDFACTOR_ONE);
111}
112
113static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA,
114                                            unsigned dstRGB, unsigned dstA)
115{
116    /* If the blend equation is ADD or REVERSE_SUBTRACT,
117     * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer
118     * will not be changed.
119     * Notice that the dst factors are the src factors inverted. */
120    return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
121            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
122           (srcA == PIPE_BLENDFACTOR_ZERO) &&
123           (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
124            dstRGB == PIPE_BLENDFACTOR_ONE) &&
125           (dstA == PIPE_BLENDFACTOR_ONE);
126}
127
128static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA,
129                                                  unsigned dstRGB, unsigned dstA)
130{
131    /* If the blend equation is ADD or REVERSE_SUBTRACT,
132     * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set,
133     * the colorbuffer will not be changed.
134     * Notice that the dst factors are the src factors inverted. */
135    return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
136            srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
137            srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
138            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
139           (srcA == PIPE_BLENDFACTOR_SRC_COLOR ||
140            srcA == PIPE_BLENDFACTOR_SRC_ALPHA ||
141            srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
142            srcA == PIPE_BLENDFACTOR_ZERO) &&
143           (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
144            dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
145            dstRGB == PIPE_BLENDFACTOR_ONE) &&
146           (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
147            dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
148            dstA == PIPE_BLENDFACTOR_ONE);
149}
150
151static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA,
152                                                  unsigned dstRGB, unsigned dstA)
153{
154    /* If the blend equation is ADD or REVERSE_SUBTRACT,
155     * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set,
156     * the colorbuffer will not be changed.
157     * Notice that the dst factors are the src factors inverted. */
158    return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
159            srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
160            srcRGB == PIPE_BLENDFACTOR_ZERO) &&
161           (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
162            srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
163            srcA == PIPE_BLENDFACTOR_ZERO) &&
164           (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
165            dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
166            dstRGB == PIPE_BLENDFACTOR_ONE) &&
167           (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
168            dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
169            dstA == PIPE_BLENDFACTOR_ONE);
170}
171
172static unsigned bgra_cmask(unsigned mask)
173{
174    /* Gallium uses RGBA color ordering while R300 expects BGRA. */
175
176    return ((mask & PIPE_MASK_R) << 2) |
177           ((mask & PIPE_MASK_B) >> 2) |
178           (mask & (PIPE_MASK_G | PIPE_MASK_A));
179}
180
181/* Create a new blend state based on the CSO blend state.
182 *
183 * This encompasses alpha blending, logic/raster ops, and blend dithering. */
184static void* r300_create_blend_state(struct pipe_context* pipe,
185                                     const struct pipe_blend_state* state)
186{
187    struct r300_screen* r300screen = r300_screen(pipe->screen);
188    struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state);
189    uint32_t blend_control = 0;       /* R300_RB3D_CBLEND: 0x4e04 */
190    uint32_t alpha_blend_control = 0; /* R300_RB3D_ABLEND: 0x4e08 */
191    uint32_t color_channel_mask = 0;  /* R300_RB3D_COLOR_CHANNEL_MASK: 0x4e0c */
192    uint32_t rop = 0;                 /* R300_RB3D_ROPCNTL: 0x4e18 */
193    uint32_t dither = 0;              /* R300_RB3D_DITHER_CTL: 0x4e50 */
194    CB_LOCALS;
195
196    if (state->rt[0].blend_enable)
197    {
198        unsigned eqRGB = state->rt[0].rgb_func;
199        unsigned srcRGB = state->rt[0].rgb_src_factor;
200        unsigned dstRGB = state->rt[0].rgb_dst_factor;
201
202        unsigned eqA = state->rt[0].alpha_func;
203        unsigned srcA = state->rt[0].alpha_src_factor;
204        unsigned dstA = state->rt[0].alpha_dst_factor;
205
206        /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha,
207         * this is just the crappy D3D naming */
208        blend_control = R300_ALPHA_BLEND_ENABLE |
209            r300_translate_blend_function(eqRGB) |
210            ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) |
211            ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT);
212
213        /* Optimization: some operations do not require the destination color.
214         *
215         * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled,
216         * otherwise blending gives incorrect results. It seems to be
217         * a hardware bug. */
218        if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN ||
219            eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX ||
220            dstRGB != PIPE_BLENDFACTOR_ZERO ||
221            dstA != PIPE_BLENDFACTOR_ZERO ||
222            srcRGB == PIPE_BLENDFACTOR_DST_COLOR ||
223            srcRGB == PIPE_BLENDFACTOR_DST_ALPHA ||
224            srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR ||
225            srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
226            srcA == PIPE_BLENDFACTOR_DST_COLOR ||
227            srcA == PIPE_BLENDFACTOR_DST_ALPHA ||
228            srcA == PIPE_BLENDFACTOR_INV_DST_COLOR ||
229            srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
230            srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) {
231            /* Enable reading from the colorbuffer. */
232            blend_control |= R300_READ_ENABLE;
233
234            if (r300screen->caps.is_r500) {
235                /* Optimization: Depending on incoming pixels, we can
236                 * conditionally disable the reading in hardware... */
237                if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN &&
238                    eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) {
239                    /* Disable reading if SRC_ALPHA == 0. */
240                    if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
241                         dstRGB == PIPE_BLENDFACTOR_ZERO) &&
242                        (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
243                         dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
244                         dstA == PIPE_BLENDFACTOR_ZERO)) {
245                         blend_control |= R500_SRC_ALPHA_0_NO_READ;
246                    }
247
248                    /* Disable reading if SRC_ALPHA == 1. */
249                    if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
250                         dstRGB == PIPE_BLENDFACTOR_ZERO) &&
251                        (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
252                         dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
253                         dstA == PIPE_BLENDFACTOR_ZERO)) {
254                         blend_control |= R500_SRC_ALPHA_1_NO_READ;
255                    }
256                }
257            }
258        }
259
260        /* Optimization: discard pixels which don't change the colorbuffer.
261         *
262         * The code below is non-trivial and some math is involved.
263         *
264         * Discarding pixels must be disabled when FP16 AA is enabled.
265         * This is a hardware bug. Also, this implementation wouldn't work
266         * with FP blending enabled and equation clamping disabled.
267         *
268         * Equations other than ADD are rarely used and therefore won't be
269         * optimized. */
270        if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) &&
271            (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) {
272            /* ADD: X+Y
273             * REVERSE_SUBTRACT: Y-X
274             *
275             * The idea is:
276             * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1,
277             * then CB will not be changed.
278             *
279             * Given the srcFactor and dstFactor variables, we can derive
280             * what src and dst should be equal to and discard appropriate
281             * pixels.
282             */
283            if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) {
284                blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0;
285            } else if (blend_discard_if_src_alpha_1(srcRGB, srcA,
286                                                    dstRGB, dstA)) {
287                blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1;
288            } else if (blend_discard_if_src_color_0(srcRGB, srcA,
289                                                    dstRGB, dstA)) {
290                blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0;
291            } else if (blend_discard_if_src_color_1(srcRGB, srcA,
292                                                    dstRGB, dstA)) {
293                blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1;
294            } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA,
295                                                          dstRGB, dstA)) {
296                blend_control |=
297                    R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0;
298            } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA,
299                                                          dstRGB, dstA)) {
300                blend_control |=
301                    R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1;
302            }
303        }
304
305        /* separate alpha */
306        if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) {
307            blend_control |= R300_SEPARATE_ALPHA_ENABLE;
308            alpha_blend_control =
309                r300_translate_blend_function(eqA) |
310                (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) |
311                (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT);
312        }
313    }
314
315    /* PIPE_LOGICOP_* don't need to be translated, fortunately. */
316    if (state->logicop_enable) {
317        rop = R300_RB3D_ROPCNTL_ROP_ENABLE |
318                (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT;
319    }
320
321    /* Color channel masks for all MRTs. */
322    color_channel_mask = bgra_cmask(state->rt[0].colormask);
323    if (r300screen->caps.is_r500 && state->independent_blend_enable) {
324        if (state->rt[1].blend_enable) {
325            color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4;
326        }
327        if (state->rt[2].blend_enable) {
328            color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8;
329        }
330        if (state->rt[3].blend_enable) {
331            color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12;
332        }
333    }
334
335    /* Neither fglrx nor classic r300 ever set this, regardless of dithering
336     * state. Since it's an optional implementation detail, we can leave it
337     * out and never dither.
338     *
339     * This could be revisited if we ever get quality or conformance hints.
340     *
341    if (state->dither) {
342        dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT |
343                        R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT;
344    }
345    */
346
347    /* Build a command buffer. */
348    BEGIN_CB(blend->cb, 8);
349    OUT_CB_REG(R300_RB3D_ROPCNTL, rop);
350    OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3);
351    OUT_CB(blend_control);
352    OUT_CB(alpha_blend_control);
353    OUT_CB(color_channel_mask);
354    OUT_CB_REG(R300_RB3D_DITHER_CTL, dither);
355    END_CB;
356
357    /* The same as above, but with no colorbuffer reads and writes. */
358    BEGIN_CB(blend->cb_no_readwrite, 8);
359    OUT_CB_REG(R300_RB3D_ROPCNTL, rop);
360    OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3);
361    OUT_CB(0);
362    OUT_CB(0);
363    OUT_CB(0);
364    OUT_CB_REG(R300_RB3D_DITHER_CTL, dither);
365    END_CB;
366
367    return (void*)blend;
368}
369
370/* Bind blend state. */
371static void r300_bind_blend_state(struct pipe_context* pipe,
372                                  void* state)
373{
374    struct r300_context* r300 = r300_context(pipe);
375
376    UPDATE_STATE(state, r300->blend_state);
377}
378
379/* Free blend state. */
380static void r300_delete_blend_state(struct pipe_context* pipe,
381                                    void* state)
382{
383    FREE(state);
384}
385
386/* Convert float to 10bit integer */
387static unsigned float_to_fixed10(float f)
388{
389    return CLAMP((unsigned)(f * 1023.9f), 0, 1023);
390}
391
392/* Set blend color.
393 * Setup both R300 and R500 registers, figure out later which one to write. */
394static void r300_set_blend_color(struct pipe_context* pipe,
395                                 const struct pipe_blend_color* color)
396{
397    struct r300_context* r300 = r300_context(pipe);
398    struct r300_blend_color_state* state =
399        (struct r300_blend_color_state*)r300->blend_color_state.state;
400    CB_LOCALS;
401
402    if (r300->screen->caps.is_r500) {
403        /* XXX if FP16 blending is enabled, we should use the FP16 format */
404        BEGIN_CB(state->cb, 3);
405        OUT_CB_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2);
406        OUT_CB(float_to_fixed10(color->color[0]) |
407               (float_to_fixed10(color->color[3]) << 16));
408        OUT_CB(float_to_fixed10(color->color[2]) |
409               (float_to_fixed10(color->color[1]) << 16));
410        END_CB;
411    } else {
412        union util_color uc;
413        util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
414
415        BEGIN_CB(state->cb, 2);
416        OUT_CB_REG(R300_RB3D_BLEND_COLOR, uc.ui);
417        END_CB;
418    }
419
420    r300->blend_color_state.dirty = TRUE;
421}
422
423static void r300_set_clip_state(struct pipe_context* pipe,
424                                const struct pipe_clip_state* state)
425{
426    struct r300_context* r300 = r300_context(pipe);
427    struct r300_clip_state *clip =
428            (struct r300_clip_state*)r300->clip_state.state;
429    CB_LOCALS;
430
431    clip->clip = *state;
432
433    if (r300->screen->caps.has_tcl) {
434        r300->clip_state.size = 2 + !!state->nr * 3 + state->nr * 4;
435
436        BEGIN_CB(clip->cb, r300->clip_state.size);
437        if (state->nr) {
438           OUT_CB_REG(R300_VAP_PVS_VECTOR_INDX_REG,
439                   (r300->screen->caps.is_r500 ?
440                    R500_PVS_UCP_START : R300_PVS_UCP_START));
441           OUT_CB_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, state->nr * 4);
442           OUT_CB_TABLE(state->ucp, state->nr * 4);
443        }
444        OUT_CB_REG(R300_VAP_CLIP_CNTL, ((1 << state->nr) - 1) |
445                R300_PS_UCP_MODE_CLIP_AS_TRIFAN |
446                (state->depth_clamp ? R300_CLIP_DISABLE : 0));
447        END_CB;
448
449        r300->clip_state.dirty = TRUE;
450    } else {
451        draw_set_clip_state(r300->draw, state);
452    }
453}
454
455static void
456r300_set_sample_mask(struct pipe_context *pipe,
457                     unsigned sample_mask)
458{
459}
460
461
462/* Create a new depth, stencil, and alpha state based on the CSO dsa state.
463 *
464 * This contains the depth buffer, stencil buffer, alpha test, and such.
465 * On the Radeon, depth and stencil buffer setup are intertwined, which is
466 * the reason for some of the strange-looking assignments across registers. */
467static void*
468        r300_create_dsa_state(struct pipe_context* pipe,
469                              const struct pipe_depth_stencil_alpha_state* state)
470{
471    struct r300_capabilities *caps = &r300_screen(pipe->screen)->caps;
472    struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state);
473    CB_LOCALS;
474
475    dsa->dsa = *state;
476
477    /* Depth test setup. - separate write mask depth for decomp flush */
478    if (state->depth.writemask) {
479        dsa->z_buffer_control |= R300_Z_WRITE_ENABLE;
480    }
481
482    if (state->depth.enabled) {
483        dsa->z_buffer_control |= R300_Z_ENABLE;
484
485        dsa->z_stencil_control |=
486            (r300_translate_depth_stencil_function(state->depth.func) <<
487                R300_Z_FUNC_SHIFT);
488    }
489
490    /* Stencil buffer setup. */
491    if (state->stencil[0].enabled) {
492        dsa->z_buffer_control |= R300_STENCIL_ENABLE;
493        dsa->z_stencil_control |=
494            (r300_translate_depth_stencil_function(state->stencil[0].func) <<
495                R300_S_FRONT_FUNC_SHIFT) |
496            (r300_translate_stencil_op(state->stencil[0].fail_op) <<
497                R300_S_FRONT_SFAIL_OP_SHIFT) |
498            (r300_translate_stencil_op(state->stencil[0].zpass_op) <<
499                R300_S_FRONT_ZPASS_OP_SHIFT) |
500            (r300_translate_stencil_op(state->stencil[0].zfail_op) <<
501                R300_S_FRONT_ZFAIL_OP_SHIFT);
502
503        dsa->stencil_ref_mask =
504                (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) |
505                (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT);
506
507        if (state->stencil[1].enabled) {
508            dsa->two_sided = TRUE;
509
510            dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK;
511            dsa->z_stencil_control |=
512            (r300_translate_depth_stencil_function(state->stencil[1].func) <<
513                R300_S_BACK_FUNC_SHIFT) |
514            (r300_translate_stencil_op(state->stencil[1].fail_op) <<
515                R300_S_BACK_SFAIL_OP_SHIFT) |
516            (r300_translate_stencil_op(state->stencil[1].zpass_op) <<
517                R300_S_BACK_ZPASS_OP_SHIFT) |
518            (r300_translate_stencil_op(state->stencil[1].zfail_op) <<
519                R300_S_BACK_ZFAIL_OP_SHIFT);
520
521            dsa->stencil_ref_bf =
522                (state->stencil[1].valuemask << R300_STENCILMASK_SHIFT) |
523                (state->stencil[1].writemask << R300_STENCILWRITEMASK_SHIFT);
524
525            if (caps->is_r500) {
526                dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK;
527            } else {
528                dsa->two_sided_stencil_ref =
529                  (state->stencil[0].valuemask != state->stencil[1].valuemask ||
530                   state->stencil[0].writemask != state->stencil[1].writemask);
531            }
532        }
533    }
534
535    /* Alpha test setup. */
536    if (state->alpha.enabled) {
537        dsa->alpha_function =
538            r300_translate_alpha_function(state->alpha.func) |
539            R300_FG_ALPHA_FUNC_ENABLE;
540
541        /* We could use 10bit alpha ref but who needs that? */
542        dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value);
543
544        if (caps->is_r500)
545            dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT;
546    }
547
548    BEGIN_CB(&dsa->cb_begin, 8);
549    OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function);
550    OUT_CB_REG_SEQ(R300_ZB_CNTL, 3);
551    OUT_CB(dsa->z_buffer_control);
552    OUT_CB(dsa->z_stencil_control);
553    OUT_CB(dsa->stencil_ref_mask);
554    OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf);
555    END_CB;
556
557    BEGIN_CB(dsa->cb_no_readwrite, 8);
558    OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function);
559    OUT_CB_REG_SEQ(R300_ZB_CNTL, 3);
560    OUT_CB(0);
561    OUT_CB(0);
562    OUT_CB(0);
563    OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, 0);
564    END_CB;
565
566    return (void*)dsa;
567}
568
569static void r300_dsa_inject_stencilref(struct r300_context *r300)
570{
571    struct r300_dsa_state *dsa =
572            (struct r300_dsa_state*)r300->dsa_state.state;
573
574    if (!dsa)
575        return;
576
577    dsa->stencil_ref_mask =
578        (dsa->stencil_ref_mask & ~R300_STENCILREF_MASK) |
579        r300->stencil_ref.ref_value[0];
580    dsa->stencil_ref_bf =
581        (dsa->stencil_ref_bf & ~R300_STENCILREF_MASK) |
582        r300->stencil_ref.ref_value[1];
583}
584
585/* Bind DSA state. */
586static void r300_bind_dsa_state(struct pipe_context* pipe,
587                                void* state)
588{
589    struct r300_context* r300 = r300_context(pipe);
590
591    if (!state) {
592        return;
593    }
594
595    UPDATE_STATE(state, r300->dsa_state);
596
597    r300->hyperz_state.dirty = TRUE; /* Will be updated before the emission. */
598    r300_dsa_inject_stencilref(r300);
599}
600
601/* Free DSA state. */
602static void r300_delete_dsa_state(struct pipe_context* pipe,
603                                  void* state)
604{
605    FREE(state);
606}
607
608static void r300_set_stencil_ref(struct pipe_context* pipe,
609                                 const struct pipe_stencil_ref* sr)
610{
611    struct r300_context* r300 = r300_context(pipe);
612
613    r300->stencil_ref = *sr;
614
615    r300_dsa_inject_stencilref(r300);
616    r300->dsa_state.dirty = TRUE;
617}
618
619static void r300_tex_set_tiling_flags(struct r300_context *r300,
620                                      struct r300_texture *tex, unsigned level)
621{
622    /* Check if the macrotile flag needs to be changed.
623     * Skip changing the flags otherwise. */
624    if (tex->desc.macrotile[tex->surface_level] !=
625        tex->desc.macrotile[level]) {
626        /* Tiling determines how DRM treats the buffer data.
627         * We must flush CS when changing it if the buffer is referenced. */
628        if (r300->rws->cs_is_buffer_referenced(r300->cs,
629                                               tex->buffer, R300_REF_CS))
630            r300->context.flush(&r300->context, 0, NULL);
631
632        r300->rws->buffer_set_tiling(r300->rws, tex->buffer,
633                tex->desc.microtile, tex->desc.macrotile[level],
634                tex->desc.stride_in_bytes[0]);
635
636        tex->surface_level = level;
637    }
638}
639
640/* This switcheroo is needed just because of goddamned MACRO_SWITCH. */
641static void r300_fb_set_tiling_flags(struct r300_context *r300,
642                               const struct pipe_framebuffer_state *state)
643{
644    unsigned i;
645
646    /* Set tiling flags for new surfaces. */
647    for (i = 0; i < state->nr_cbufs; i++) {
648        r300_tex_set_tiling_flags(r300,
649                                  r300_texture(state->cbufs[i]->texture),
650                                  state->cbufs[i]->level);
651    }
652    if (state->zsbuf) {
653        r300_tex_set_tiling_flags(r300,
654                                  r300_texture(state->zsbuf->texture),
655                                  state->zsbuf->level);
656    }
657}
658
659static void r300_print_fb_surf_info(struct pipe_surface *surf, unsigned index,
660                                    const char *binding)
661{
662    struct pipe_resource *tex = surf->texture;
663    struct r300_texture *rtex = r300_texture(tex);
664
665    fprintf(stderr,
666            "r300:   %s[%i] Dim: %ix%i, Offset: %i, ZSlice: %i, "
667            "Face: %i, Level: %i, Format: %s\n"
668
669            "r300:     TEX: Macro: %s, Micro: %s, Pitch: %i, "
670            "Dim: %ix%ix%i, LastLevel: %i, Format: %s\n",
671
672            binding, index, surf->width, surf->height, surf->offset,
673            surf->zslice, surf->face, surf->level,
674            util_format_short_name(surf->format),
675
676            rtex->desc.macrotile[0] ? "YES" : " NO",
677            rtex->desc.microtile ? "YES" : " NO",
678            rtex->desc.stride_in_pixels[0],
679            tex->width0, tex->height0, tex->depth0,
680            tex->last_level, util_format_short_name(tex->format));
681}
682
683void r300_mark_fb_state_dirty(struct r300_context *r300,
684                              enum r300_fb_state_change change)
685{
686    struct pipe_framebuffer_state *state = r300->fb_state.state;
687    boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
688
689    /* What is marked as dirty depends on the enum r300_fb_state_change. */
690    r300->gpu_flush.dirty = TRUE;
691    r300->fb_state.dirty = TRUE;
692    r300->hyperz_state.dirty = TRUE;
693
694    if (change == R300_CHANGED_FB_STATE) {
695        r300->aa_state.dirty = TRUE;
696        r300->fb_state_pipelined.dirty = TRUE;
697    }
698
699    /* Now compute the fb_state atom size. */
700    r300->fb_state.size = 2 + (8 * state->nr_cbufs);
701
702    if (r300->cbzb_clear)
703        r300->fb_state.size += 10;
704    else if (state->zsbuf) {
705        r300->fb_state.size += 10;
706        if (has_hyperz)
707            r300->fb_state.size += r300->screen->caps.hiz_ram ? 8 : 4;
708    }
709
710    /* The size of the rest of atoms stays the same. */
711}
712
713static void
714    r300_set_framebuffer_state(struct pipe_context* pipe,
715                               const struct pipe_framebuffer_state* state)
716{
717    struct r300_context* r300 = r300_context(pipe);
718    struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
719    struct pipe_framebuffer_state *old_state = r300->fb_state.state;
720    boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
721    unsigned max_width, max_height, i;
722    uint32_t zbuffer_bpp = 0;
723    int blocksize;
724
725    if (r300->screen->caps.is_r500) {
726        max_width = max_height = 4096;
727    } else if (r300->screen->caps.is_r400) {
728        max_width = max_height = 4021;
729    } else {
730        max_width = max_height = 2560;
731    }
732
733    if (state->width > max_width || state->height > max_height) {
734        fprintf(stderr, "r300: Implementation error: Render targets are too "
735        "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__);
736        return;
737    }
738
739    /* If nr_cbufs is changed from zero to non-zero or vice versa... */
740    if (!!old_state->nr_cbufs != !!state->nr_cbufs) {
741        r300->blend_state.dirty = TRUE;
742    }
743    /* If zsbuf is set from NULL to non-NULL or vice versa.. */
744    if (!!old_state->zsbuf != !!state->zsbuf) {
745        r300->dsa_state.dirty = TRUE;
746    }
747
748    /* The tiling flags are dependent on the surface miplevel, unfortunately. */
749    r300_fb_set_tiling_flags(r300, state);
750
751    util_copy_framebuffer_state(r300->fb_state.state, state);
752
753    r300_mark_fb_state_dirty(r300, R300_CHANGED_FB_STATE);
754
755    r300->z_compression = false;
756
757    if (state->zsbuf) {
758        blocksize = util_format_get_blocksize(state->zsbuf->texture->format);
759        switch (blocksize) {
760        case 2:
761            zbuffer_bpp = 16;
762            break;
763        case 4:
764            zbuffer_bpp = 24;
765            break;
766        }
767        if (has_hyperz) {
768            struct r300_surface *zs_surf = r300_surface(state->zsbuf);
769            struct r300_texture *tex;
770            int compress = r300->screen->caps.is_rv350 ? RV350_Z_COMPRESS_88 : R300_Z_COMPRESS_44;
771            int level = zs_surf->base.level;
772
773            tex = r300_texture(zs_surf->base.texture);
774
775            /* work out whether we can support hiz on this buffer */
776            r300_hiz_alloc_block(r300, zs_surf);
777
778            /* work out whether we can support zmask features on this buffer */
779            r300_zmask_alloc_block(r300, zs_surf, compress);
780
781            if (tex->zmask_mem[level]) {
782                /* compression causes hangs on 16-bit */
783                if (zbuffer_bpp == 24)
784                    r300->z_compression = compress;
785            }
786            DBG(r300, DBG_HYPERZ,
787                "hyper-z features: hiz: %d @ %08x z-compression: %d z-fastfill: %d @ %08x\n", tex->hiz_mem[level] ? 1 : 0,
788                tex->hiz_mem[level] ? tex->hiz_mem[level]->ofs : 0xdeadbeef,
789                r300->z_compression, tex->zmask_mem[level] ? 1 : 0,
790                tex->zmask_mem[level] ? tex->zmask_mem[level]->ofs : 0xdeadbeef);
791        }
792
793        /* Polygon offset depends on the zbuffer bit depth. */
794        if (r300->zbuffer_bpp != zbuffer_bpp) {
795            r300->zbuffer_bpp = zbuffer_bpp;
796
797            if (r300->polygon_offset_enabled)
798                r300->rs_state.dirty = TRUE;
799        }
800    }
801
802    /* Set up AA config. */
803    if (r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0)) {
804        if (state->nr_cbufs && state->cbufs[0]->texture->nr_samples > 1) {
805            aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE;
806
807            switch (state->cbufs[0]->texture->nr_samples) {
808                case 2:
809                    aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2;
810                    break;
811                case 3:
812                    aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_3;
813                    break;
814                case 4:
815                    aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4;
816                    break;
817                case 6:
818                    aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6;
819                    break;
820            }
821        } else {
822            aa->aa_config = 0;
823        }
824    }
825
826    if (DBG_ON(r300, DBG_FB)) {
827        fprintf(stderr, "r300: set_framebuffer_state:\n");
828        for (i = 0; i < state->nr_cbufs; i++) {
829            r300_print_fb_surf_info(state->cbufs[i], i, "CB");
830        }
831        if (state->zsbuf) {
832            r300_print_fb_surf_info(state->zsbuf, 0, "ZB");
833        }
834    }
835}
836
837/* Create fragment shader state. */
838static void* r300_create_fs_state(struct pipe_context* pipe,
839                                  const struct pipe_shader_state* shader)
840{
841    struct r300_fragment_shader* fs = NULL;
842
843    fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader);
844
845    /* Copy state directly into shader. */
846    fs->state = *shader;
847    fs->state.tokens = tgsi_dup_tokens(shader->tokens);
848
849    return (void*)fs;
850}
851
852void r300_mark_fs_code_dirty(struct r300_context *r300)
853{
854    struct r300_fragment_shader* fs = r300_fs(r300);
855
856    r300->fs.dirty = TRUE;
857    r300->fs_rc_constant_state.dirty = TRUE;
858    r300->fs_constants.dirty = TRUE;
859    r300->fs.size = fs->shader->cb_code_size;
860
861    if (r300->screen->caps.is_r500) {
862        r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 7;
863        r300->fs_constants.size = fs->shader->externals_count * 4 + 3;
864    } else {
865        r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 5;
866        r300->fs_constants.size = fs->shader->externals_count * 4 + 1;
867    }
868
869    ((struct r300_constant_buffer*)r300->fs_constants.state)->remap_table =
870            fs->shader->code.constants_remap_table;
871}
872
873/* Bind fragment shader state. */
874static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
875{
876    struct r300_context* r300 = r300_context(pipe);
877    struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
878
879    if (fs == NULL) {
880        r300->fs.state = NULL;
881        return;
882    }
883
884    r300->fs.state = fs;
885    r300_pick_fragment_shader(r300);
886    r300_mark_fs_code_dirty(r300);
887
888    r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
889}
890
891/* Delete fragment shader state. */
892static void r300_delete_fs_state(struct pipe_context* pipe, void* shader)
893{
894    struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
895    struct r300_fragment_shader_code *tmp, *ptr = fs->first;
896
897    while (ptr) {
898        tmp = ptr;
899        ptr = ptr->next;
900        rc_constants_destroy(&tmp->code.constants);
901        FREE(tmp->cb_code);
902        FREE(tmp);
903    }
904    FREE((void*)fs->state.tokens);
905    FREE(shader);
906}
907
908static void r300_set_polygon_stipple(struct pipe_context* pipe,
909                                     const struct pipe_poly_stipple* state)
910{
911    /* XXX no idea how to set this up, but not terribly important */
912}
913
914/* Create a new rasterizer state based on the CSO rasterizer state.
915 *
916 * This is a very large chunk of state, and covers most of the graphics
917 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
918 *
919 * In a not entirely unironic sidenote, this state has nearly nothing to do
920 * with the actual block on the Radeon called the rasterizer (RS). */
921static void* r300_create_rs_state(struct pipe_context* pipe,
922                                  const struct pipe_rasterizer_state* state)
923{
924    struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state);
925    float psiz;
926    uint32_t vap_control_status;    /* R300_VAP_CNTL_STATUS: 0x2140 */
927    uint32_t point_size;            /* R300_GA_POINT_SIZE: 0x421c */
928    uint32_t point_minmax;          /* R300_GA_POINT_MINMAX: 0x4230 */
929    uint32_t line_control;          /* R300_GA_LINE_CNTL: 0x4234 */
930    uint32_t polygon_offset_enable; /* R300_SU_POLY_OFFSET_ENABLE: 0x42b4 */
931    uint32_t cull_mode;             /* R300_SU_CULL_MODE: 0x42b8 */
932    uint32_t line_stipple_config;   /* R300_GA_LINE_STIPPLE_CONFIG: 0x4328 */
933    uint32_t line_stipple_value;    /* R300_GA_LINE_STIPPLE_VALUE: 0x4260 */
934    uint32_t polygon_mode;          /* R300_GA_POLY_MODE: 0x4288 */
935    uint32_t clip_rule;             /* R300_SC_CLIP_RULE: 0x43D0 */
936
937    /* Point sprites texture coordinates, 0: lower left, 1: upper right */
938    float point_texcoord_left = 0;  /* R300_GA_POINT_S0: 0x4200 */
939    float point_texcoord_bottom = 0;/* R300_GA_POINT_T0: 0x4204 */
940    float point_texcoord_right = 1; /* R300_GA_POINT_S1: 0x4208 */
941    float point_texcoord_top = 0;   /* R300_GA_POINT_T1: 0x420c */
942    CB_LOCALS;
943
944    /* Copy rasterizer state. */
945    rs->rs = *state;
946    rs->rs_draw = *state;
947
948    rs->rs.sprite_coord_enable = state->point_quad_rasterization *
949                                 state->sprite_coord_enable;
950
951    /* Override some states for Draw. */
952    rs->rs_draw.sprite_coord_enable = 0; /* We can do this in HW. */
953
954#ifdef PIPE_ARCH_LITTLE_ENDIAN
955    vap_control_status = R300_VC_NO_SWAP;
956#else
957    vap_control_status = R300_VC_32BIT_SWAP;
958#endif
959
960    /* If no TCL engine is present, turn off the HW TCL. */
961    if (!r300_screen(pipe->screen)->caps.has_tcl) {
962        vap_control_status |= R300_VAP_TCL_BYPASS;
963    }
964
965    /* Point size width and height. */
966    point_size =
967        pack_float_16_6x(state->point_size) |
968        (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT);
969
970    /* Point size clamping. */
971    if (state->point_size_per_vertex) {
972        /* Per-vertex point size.
973         * Clamp to [0, max FB size] */
974        psiz = pipe->screen->get_paramf(pipe->screen,
975                                        PIPE_CAP_MAX_POINT_WIDTH);
976        point_minmax =
977            pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT;
978    } else {
979        /* We cannot disable the point-size vertex output,
980         * so clamp it. */
981        psiz = state->point_size;
982        point_minmax =
983            (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MIN_SHIFT) |
984            (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT);
985    }
986
987    /* Line control. */
988    line_control = pack_float_16_6x(state->line_width) |
989        R300_GA_LINE_CNTL_END_TYPE_COMP;
990
991    /* Enable polygon mode */
992    polygon_mode = 0;
993    if (state->fill_front != PIPE_POLYGON_MODE_FILL ||
994        state->fill_back != PIPE_POLYGON_MODE_FILL) {
995        polygon_mode = R300_GA_POLY_MODE_DUAL;
996    }
997
998    /* Front face */
999    if (state->front_ccw)
1000        cull_mode = R300_FRONT_FACE_CCW;
1001    else
1002        cull_mode = R300_FRONT_FACE_CW;
1003
1004    /* Polygon offset */
1005    polygon_offset_enable = 0;
1006    if (util_get_offset(state, state->fill_front)) {
1007       polygon_offset_enable |= R300_FRONT_ENABLE;
1008    }
1009    if (util_get_offset(state, state->fill_back)) {
1010       polygon_offset_enable |= R300_BACK_ENABLE;
1011    }
1012
1013    rs->polygon_offset_enable = polygon_offset_enable != 0;
1014
1015    /* Polygon mode */
1016    if (polygon_mode) {
1017       polygon_mode |=
1018          r300_translate_polygon_mode_front(state->fill_front);
1019       polygon_mode |=
1020          r300_translate_polygon_mode_back(state->fill_back);
1021    }
1022
1023    if (state->cull_face & PIPE_FACE_FRONT) {
1024        cull_mode |= R300_CULL_FRONT;
1025    }
1026    if (state->cull_face & PIPE_FACE_BACK) {
1027        cull_mode |= R300_CULL_BACK;
1028    }
1029
1030    if (state->line_stipple_enable) {
1031        line_stipple_config =
1032            R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE |
1033            (fui((float)state->line_stipple_factor) &
1034                R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK);
1035        /* XXX this might need to be scaled up */
1036        line_stipple_value = state->line_stipple_pattern;
1037    } else {
1038        line_stipple_config = 0;
1039        line_stipple_value = 0;
1040    }
1041
1042    if (state->flatshade) {
1043        rs->color_control = R300_SHADE_MODEL_FLAT;
1044    } else {
1045        rs->color_control = R300_SHADE_MODEL_SMOOTH;
1046    }
1047
1048    clip_rule = state->scissor ? 0xAAAA : 0xFFFF;
1049
1050    /* Point sprites coord mode */
1051    if (rs->rs.sprite_coord_enable) {
1052        switch (state->sprite_coord_mode) {
1053            case PIPE_SPRITE_COORD_UPPER_LEFT:
1054                point_texcoord_top = 0.0f;
1055                point_texcoord_bottom = 1.0f;
1056                break;
1057            case PIPE_SPRITE_COORD_LOWER_LEFT:
1058                point_texcoord_top = 1.0f;
1059                point_texcoord_bottom = 0.0f;
1060                break;
1061        }
1062    }
1063
1064    /* Build the main command buffer. */
1065    BEGIN_CB(rs->cb_main, RS_STATE_MAIN_SIZE);
1066    OUT_CB_REG(R300_VAP_CNTL_STATUS, vap_control_status);
1067    OUT_CB_REG(R300_GA_POINT_SIZE, point_size);
1068    OUT_CB_REG_SEQ(R300_GA_POINT_MINMAX, 2);
1069    OUT_CB(point_minmax);
1070    OUT_CB(line_control);
1071    OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_ENABLE, 2);
1072    OUT_CB(polygon_offset_enable);
1073    rs->cull_mode_index = 9;
1074    OUT_CB(cull_mode);
1075    OUT_CB_REG(R300_GA_LINE_STIPPLE_CONFIG, line_stipple_config);
1076    OUT_CB_REG(R300_GA_LINE_STIPPLE_VALUE, line_stipple_value);
1077    OUT_CB_REG(R300_GA_POLY_MODE, polygon_mode);
1078    OUT_CB_REG(R300_SC_CLIP_RULE, clip_rule);
1079    OUT_CB_REG_SEQ(R300_GA_POINT_S0, 4);
1080    OUT_CB_32F(point_texcoord_left);
1081    OUT_CB_32F(point_texcoord_bottom);
1082    OUT_CB_32F(point_texcoord_right);
1083    OUT_CB_32F(point_texcoord_top);
1084    END_CB;
1085
1086    /* Build the two command buffers for polygon offset setup. */
1087    if (polygon_offset_enable) {
1088        float scale = state->offset_scale * 12;
1089        float offset = state->offset_units * 4;
1090
1091        BEGIN_CB(rs->cb_poly_offset_zb16, 5);
1092        OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4);
1093        OUT_CB_32F(scale);
1094        OUT_CB_32F(offset);
1095        OUT_CB_32F(scale);
1096        OUT_CB_32F(offset);
1097        END_CB;
1098
1099        offset = state->offset_units * 2;
1100
1101        BEGIN_CB(rs->cb_poly_offset_zb24, 5);
1102        OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4);
1103        OUT_CB_32F(scale);
1104        OUT_CB_32F(offset);
1105        OUT_CB_32F(scale);
1106        OUT_CB_32F(offset);
1107        END_CB;
1108    }
1109
1110    return (void*)rs;
1111}
1112
1113/* Bind rasterizer state. */
1114static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
1115{
1116    struct r300_context* r300 = r300_context(pipe);
1117    struct r300_rs_state* rs = (struct r300_rs_state*)state;
1118    int last_sprite_coord_enable = r300->sprite_coord_enable;
1119    boolean last_two_sided_color = r300->two_sided_color;
1120
1121    if (r300->draw && rs) {
1122        draw_set_rasterizer_state(r300->draw, &rs->rs_draw, state);
1123    }
1124
1125    if (rs) {
1126        r300->polygon_offset_enabled = rs->polygon_offset_enable;
1127        r300->sprite_coord_enable = rs->rs.sprite_coord_enable;
1128        r300->two_sided_color = rs->rs.light_twoside;
1129    } else {
1130        r300->polygon_offset_enabled = FALSE;
1131        r300->sprite_coord_enable = 0;
1132        r300->two_sided_color = FALSE;
1133    }
1134
1135    UPDATE_STATE(state, r300->rs_state);
1136    r300->rs_state.size = RS_STATE_MAIN_SIZE + (r300->polygon_offset_enabled ? 5 : 0);
1137
1138    if (last_sprite_coord_enable != r300->sprite_coord_enable ||
1139        last_two_sided_color != r300->two_sided_color) {
1140        r300->rs_block_state.dirty = TRUE;
1141    }
1142}
1143
1144/* Free rasterizer state. */
1145static void r300_delete_rs_state(struct pipe_context* pipe, void* state)
1146{
1147    FREE(state);
1148}
1149
1150static void*
1151        r300_create_sampler_state(struct pipe_context* pipe,
1152                                  const struct pipe_sampler_state* state)
1153{
1154    struct r300_context* r300 = r300_context(pipe);
1155    struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state);
1156    boolean is_r500 = r300->screen->caps.is_r500;
1157    int lod_bias;
1158
1159    sampler->state = *state;
1160
1161    /* r300 doesn't handle CLAMP and MIRROR_CLAMP correctly when either MAG
1162     * or MIN filter is NEAREST. Since texwrap produces same results
1163     * for CLAMP and CLAMP_TO_EDGE, we use them instead. */
1164    if (sampler->state.min_img_filter == PIPE_TEX_FILTER_NEAREST ||
1165        sampler->state.mag_img_filter == PIPE_TEX_FILTER_NEAREST) {
1166        /* Wrap S. */
1167        if (sampler->state.wrap_s == PIPE_TEX_WRAP_CLAMP)
1168            sampler->state.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1169        else if (sampler->state.wrap_s == PIPE_TEX_WRAP_MIRROR_CLAMP)
1170            sampler->state.wrap_s = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE;
1171
1172        /* Wrap T. */
1173        if (sampler->state.wrap_t == PIPE_TEX_WRAP_CLAMP)
1174            sampler->state.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1175        else if (sampler->state.wrap_t == PIPE_TEX_WRAP_MIRROR_CLAMP)
1176            sampler->state.wrap_t = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE;
1177
1178        /* Wrap R. */
1179        if (sampler->state.wrap_r == PIPE_TEX_WRAP_CLAMP)
1180            sampler->state.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
1181        else if (sampler->state.wrap_r == PIPE_TEX_WRAP_MIRROR_CLAMP)
1182            sampler->state.wrap_r = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE;
1183    }
1184
1185    sampler->filter0 |=
1186        (r300_translate_wrap(sampler->state.wrap_s) << R300_TX_WRAP_S_SHIFT) |
1187        (r300_translate_wrap(sampler->state.wrap_t) << R300_TX_WRAP_T_SHIFT) |
1188        (r300_translate_wrap(sampler->state.wrap_r) << R300_TX_WRAP_R_SHIFT);
1189
1190    sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter,
1191                                                   state->mag_img_filter,
1192                                                   state->min_mip_filter,
1193                                                   state->max_anisotropy > 0);
1194
1195    sampler->filter0 |= r300_anisotropy(state->max_anisotropy);
1196
1197    /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
1198    /* We must pass these to the merge function to clamp them properly. */
1199    sampler->min_lod = (unsigned)MAX2(state->min_lod, 0);
1200    sampler->max_lod = (unsigned)MAX2(ceilf(state->max_lod), 0);
1201
1202    lod_bias = CLAMP((int)(state->lod_bias * 32 + 1), -(1 << 9), (1 << 9) - 1);
1203
1204    sampler->filter1 |= (lod_bias << R300_LOD_BIAS_SHIFT) & R300_LOD_BIAS_MASK;
1205
1206    /* This is very high quality anisotropic filtering for R5xx.
1207     * It's good for benchmarking the performance of texturing but
1208     * in practice we don't want to slow down the driver because it's
1209     * a pretty good performance killer. Feel free to play with it. */
1210    if (DBG_ON(r300, DBG_ANISOHQ) && is_r500) {
1211        sampler->filter1 |= r500_anisotropy(state->max_anisotropy);
1212    }
1213
1214    /* R500-specific fixups and optimizations */
1215    if (r300->screen->caps.is_r500) {
1216        sampler->filter1 |= R500_BORDER_FIX;
1217    }
1218
1219    return (void*)sampler;
1220}
1221
1222static void r300_bind_sampler_states(struct pipe_context* pipe,
1223                                     unsigned count,
1224                                     void** states)
1225{
1226    struct r300_context* r300 = r300_context(pipe);
1227    struct r300_textures_state* state =
1228        (struct r300_textures_state*)r300->textures_state.state;
1229    unsigned tex_units = r300->screen->caps.num_tex_units;
1230
1231    if (count > tex_units) {
1232        return;
1233    }
1234
1235    memcpy(state->sampler_states, states, sizeof(void*) * count);
1236    state->sampler_state_count = count;
1237
1238    r300->textures_state.dirty = TRUE;
1239}
1240
1241static void r300_lacks_vertex_textures(struct pipe_context* pipe,
1242                                       unsigned count,
1243                                       void** states)
1244{
1245}
1246
1247static void r300_delete_sampler_state(struct pipe_context* pipe, void* state)
1248{
1249    FREE(state);
1250}
1251
1252static uint32_t r300_assign_texture_cache_region(unsigned index, unsigned num)
1253{
1254    /* This looks like a hack, but I believe it's suppose to work like
1255     * that. To illustrate how this works, let's assume you have 5 textures.
1256     * From docs, 5 and the successive numbers are:
1257     *
1258     * FOURTH_1     = 5
1259     * FOURTH_2     = 6
1260     * FOURTH_3     = 7
1261     * EIGHTH_0     = 8
1262     * EIGHTH_1     = 9
1263     *
1264     * First 3 textures will get 3/4 of size of the cache, divived evenly
1265     * between them. The last 1/4 of the cache must be divided between
1266     * the last 2 textures, each will therefore get 1/8 of the cache.
1267     * Why not just to use "5 + texture_index" ?
1268     *
1269     * This simple trick works for all "num" <= 16.
1270     */
1271    if (num <= 1)
1272        return R300_TX_CACHE(R300_TX_CACHE_WHOLE);
1273    else
1274        return R300_TX_CACHE(num + index);
1275}
1276
1277static void r300_set_fragment_sampler_views(struct pipe_context* pipe,
1278                                            unsigned count,
1279                                            struct pipe_sampler_view** views)
1280{
1281    struct r300_context* r300 = r300_context(pipe);
1282    struct r300_textures_state* state =
1283        (struct r300_textures_state*)r300->textures_state.state;
1284    struct r300_texture *texture;
1285    unsigned i, real_num_views = 0, view_index = 0;
1286    unsigned tex_units = r300->screen->caps.num_tex_units;
1287    boolean dirty_tex = FALSE;
1288
1289    if (count > tex_units) {
1290        return;
1291    }
1292
1293    /* Calculate the real number of views. */
1294    for (i = 0; i < count; i++) {
1295        if (views[i])
1296            real_num_views++;
1297    }
1298
1299    for (i = 0; i < count; i++) {
1300        if (&state->sampler_views[i]->base != views[i]) {
1301            pipe_sampler_view_reference(
1302                    (struct pipe_sampler_view**)&state->sampler_views[i],
1303                    views[i]);
1304
1305            if (!views[i]) {
1306                continue;
1307            }
1308
1309            /* A new sampler view (= texture)... */
1310            dirty_tex = TRUE;
1311
1312            /* Set the texrect factor in the fragment shader.
1313             * Needed for RECT and NPOT fallback. */
1314            texture = r300_texture(views[i]->texture);
1315            if (texture->desc.is_npot) {
1316                r300->fs_rc_constant_state.dirty = TRUE;
1317            }
1318
1319            state->sampler_views[i]->texcache_region =
1320                r300_assign_texture_cache_region(view_index, real_num_views);
1321            view_index++;
1322        }
1323    }
1324
1325    for (i = count; i < tex_units; i++) {
1326        if (state->sampler_views[i]) {
1327            pipe_sampler_view_reference(
1328                    (struct pipe_sampler_view**)&state->sampler_views[i],
1329                    NULL);
1330        }
1331    }
1332
1333    state->sampler_view_count = count;
1334
1335    r300->textures_state.dirty = TRUE;
1336
1337    if (dirty_tex) {
1338        r300->texture_cache_inval.dirty = TRUE;
1339    }
1340}
1341
1342static struct pipe_sampler_view *
1343r300_create_sampler_view(struct pipe_context *pipe,
1344                         struct pipe_resource *texture,
1345                         const struct pipe_sampler_view *templ)
1346{
1347    struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view);
1348    struct r300_texture *tex = r300_texture(texture);
1349    boolean is_r500 = r300_screen(pipe->screen)->caps.is_r500;
1350
1351    if (view) {
1352        view->base = *templ;
1353        view->base.reference.count = 1;
1354        view->base.context = pipe;
1355        view->base.texture = NULL;
1356        pipe_resource_reference(&view->base.texture, texture);
1357
1358        view->swizzle[0] = templ->swizzle_r;
1359        view->swizzle[1] = templ->swizzle_g;
1360        view->swizzle[2] = templ->swizzle_b;
1361        view->swizzle[3] = templ->swizzle_a;
1362
1363        view->format = tex->tx_format;
1364        view->format.format1 |= r300_translate_texformat(templ->format,
1365                                                         view->swizzle,
1366                                                         is_r500);
1367        if (is_r500) {
1368            view->format.format2 |= r500_tx_format_msb_bit(templ->format);
1369        }
1370    }
1371
1372    return (struct pipe_sampler_view*)view;
1373}
1374
1375static void
1376r300_sampler_view_destroy(struct pipe_context *pipe,
1377                          struct pipe_sampler_view *view)
1378{
1379   pipe_resource_reference(&view->texture, NULL);
1380   FREE(view);
1381}
1382
1383static void r300_set_scissor_state(struct pipe_context* pipe,
1384                                   const struct pipe_scissor_state* state)
1385{
1386    struct r300_context* r300 = r300_context(pipe);
1387
1388    memcpy(r300->scissor_state.state, state,
1389        sizeof(struct pipe_scissor_state));
1390
1391    r300->scissor_state.dirty = TRUE;
1392}
1393
1394static void r300_set_viewport_state(struct pipe_context* pipe,
1395                                    const struct pipe_viewport_state* state)
1396{
1397    struct r300_context* r300 = r300_context(pipe);
1398    struct r300_viewport_state* viewport =
1399        (struct r300_viewport_state*)r300->viewport_state.state;
1400
1401    r300->viewport = *state;
1402
1403    if (r300->draw) {
1404        draw_set_viewport_state(r300->draw, state);
1405        viewport->vte_control = R300_VTX_XY_FMT | R300_VTX_Z_FMT;
1406        return;
1407    }
1408
1409    /* Do the transform in HW. */
1410    viewport->vte_control = R300_VTX_W0_FMT;
1411
1412    if (state->scale[0] != 1.0f) {
1413        viewport->xscale = state->scale[0];
1414        viewport->vte_control |= R300_VPORT_X_SCALE_ENA;
1415    }
1416    if (state->scale[1] != 1.0f) {
1417        viewport->yscale = state->scale[1];
1418        viewport->vte_control |= R300_VPORT_Y_SCALE_ENA;
1419    }
1420    if (state->scale[2] != 1.0f) {
1421        viewport->zscale = state->scale[2];
1422        viewport->vte_control |= R300_VPORT_Z_SCALE_ENA;
1423    }
1424    if (state->translate[0] != 0.0f) {
1425        viewport->xoffset = state->translate[0];
1426        viewport->vte_control |= R300_VPORT_X_OFFSET_ENA;
1427    }
1428    if (state->translate[1] != 0.0f) {
1429        viewport->yoffset = state->translate[1];
1430        viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA;
1431    }
1432    if (state->translate[2] != 0.0f) {
1433        viewport->zoffset = state->translate[2];
1434        viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA;
1435    }
1436
1437    r300->viewport_state.dirty = TRUE;
1438    if (r300->fs.state && r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) {
1439        r300->fs_rc_constant_state.dirty = TRUE;
1440    }
1441}
1442
1443static void r300_set_vertex_buffers(struct pipe_context* pipe,
1444                                    unsigned count,
1445                                    const struct pipe_vertex_buffer* buffers)
1446{
1447    struct r300_context* r300 = r300_context(pipe);
1448    struct pipe_vertex_buffer *vbo;
1449    unsigned i, max_index = (1 << 24) - 1;
1450    boolean any_user_buffer = FALSE;
1451
1452    if (count == r300->vertex_buffer_count &&
1453        memcmp(r300->vertex_buffer, buffers,
1454            sizeof(struct pipe_vertex_buffer) * count) == 0) {
1455        return;
1456    }
1457
1458    if (r300->screen->caps.has_tcl) {
1459        /* HW TCL. */
1460        r300->incompatible_vb_layout = FALSE;
1461
1462        /* Check if the strides and offsets are aligned to the size of DWORD. */
1463        for (i = 0; i < count; i++) {
1464            if (buffers[i].buffer) {
1465                if (buffers[i].stride % 4 != 0 ||
1466                    buffers[i].buffer_offset % 4 != 0) {
1467                    r300->incompatible_vb_layout = TRUE;
1468                    break;
1469                }
1470            }
1471        }
1472
1473        for (i = 0; i < count; i++) {
1474            /* Why, yes, I AM casting away constness. How did you know? */
1475            vbo = (struct pipe_vertex_buffer*)&buffers[i];
1476
1477            /* Skip NULL buffers */
1478            if (!buffers[i].buffer) {
1479                continue;
1480            }
1481
1482            if (r300_buffer_is_user_buffer(vbo->buffer)) {
1483                any_user_buffer = TRUE;
1484            }
1485
1486            if (vbo->max_index == ~0) {
1487                /* if no VBO stride then only one vertex value so max index is 1 */
1488                /* should think about converting to VS constants like svga does */
1489                if (!vbo->stride)
1490                    vbo->max_index = 1;
1491                else
1492                    vbo->max_index =
1493                             (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride;
1494            }
1495
1496            max_index = MIN2(vbo->max_index, max_index);
1497        }
1498
1499        r300->any_user_vbs = any_user_buffer;
1500        r300->vertex_buffer_max_index = max_index;
1501
1502    } else {
1503        /* SW TCL. */
1504        draw_set_vertex_buffers(r300->draw, count, buffers);
1505    }
1506
1507    /* Common code. */
1508    for (i = 0; i < count; i++) {
1509        /* Reference our buffer. */
1510        pipe_resource_reference(&r300->vertex_buffer[i].buffer, buffers[i].buffer);
1511    }
1512    for (; i < r300->vertex_buffer_count; i++) {
1513        /* Dereference any old buffers. */
1514        pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL);
1515    }
1516
1517    memcpy(r300->vertex_buffer, buffers,
1518        sizeof(struct pipe_vertex_buffer) * count);
1519    r300->vertex_buffer_count = count;
1520}
1521
1522static void r300_set_index_buffer(struct pipe_context* pipe,
1523                                  const struct pipe_index_buffer *ib)
1524{
1525    struct r300_context* r300 = r300_context(pipe);
1526
1527    if (ib) {
1528        pipe_resource_reference(&r300->index_buffer.buffer, ib->buffer);
1529        memcpy(&r300->index_buffer, ib, sizeof(r300->index_buffer));
1530    }
1531    else {
1532        pipe_resource_reference(&r300->index_buffer.buffer, NULL);
1533        memset(&r300->index_buffer, 0, sizeof(r300->index_buffer));
1534    }
1535
1536    if (r300->screen->caps.has_tcl) {
1537       /* TODO make this more like a state */
1538    }
1539    else {
1540       draw_set_index_buffer(r300->draw, ib);
1541    }
1542}
1543
1544/* Initialize the PSC tables. */
1545static void r300_vertex_psc(struct r300_vertex_element_state *velems)
1546{
1547    struct r300_vertex_stream_state *vstream = &velems->vertex_stream;
1548    uint16_t type, swizzle;
1549    enum pipe_format format;
1550    unsigned i;
1551
1552    if (velems->count > 16) {
1553        fprintf(stderr, "r300: More than 16 vertex elements are not supported,"
1554                " requested %i, using 16.\n", velems->count);
1555        velems->count = 16;
1556    }
1557
1558    /* Vertex shaders have no semantics on their inputs,
1559     * so PSC should just route stuff based on the vertex elements,
1560     * and not on attrib information. */
1561    for (i = 0; i < velems->count; i++) {
1562        format = velems->hw_format[i];
1563
1564        type = r300_translate_vertex_data_type(format);
1565        if (type == R300_INVALID_FORMAT) {
1566            fprintf(stderr, "r300: Bad vertex format %s.\n",
1567                    util_format_short_name(format));
1568            assert(0);
1569            abort();
1570        }
1571
1572        type |= i << R300_DST_VEC_LOC_SHIFT;
1573        swizzle = r300_translate_vertex_data_swizzle(format);
1574
1575        if (i & 1) {
1576            vstream->vap_prog_stream_cntl[i >> 1] |= type << 16;
1577            vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16;
1578        } else {
1579            vstream->vap_prog_stream_cntl[i >> 1] |= type;
1580            vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
1581        }
1582    }
1583
1584    /* Set the last vector in the PSC. */
1585    if (i) {
1586        i -= 1;
1587    }
1588    vstream->vap_prog_stream_cntl[i >> 1] |=
1589        (R300_LAST_VEC << (i & 1 ? 16 : 0));
1590
1591    vstream->count = (i >> 1) + 1;
1592}
1593
1594#define FORMAT_REPLACE(what, withwhat) \
1595    case PIPE_FORMAT_##what: *format = PIPE_FORMAT_##withwhat; break
1596
1597static void* r300_create_vertex_elements_state(struct pipe_context* pipe,
1598                                               unsigned count,
1599                                               const struct pipe_vertex_element* attribs)
1600{
1601    struct r300_vertex_element_state *velems;
1602    unsigned i;
1603    enum pipe_format *format;
1604
1605    assert(count <= PIPE_MAX_ATTRIBS);
1606    velems = CALLOC_STRUCT(r300_vertex_element_state);
1607    if (velems != NULL) {
1608        velems->count = count;
1609        memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count);
1610
1611        if (r300_screen(pipe->screen)->caps.has_tcl) {
1612            /* Set the best hw format in case the original format is not
1613             * supported by hw. */
1614            for (i = 0; i < count; i++) {
1615                velems->hw_format[i] = velems->velem[i].src_format;
1616                format = &velems->hw_format[i];
1617
1618                /* This is basically the list of unsupported formats.
1619                 * For now we don't care about the alignment, that's going to
1620                 * be sorted out after the PSC setup. */
1621                switch (*format) {
1622                    FORMAT_REPLACE(R64_FLOAT,           R32_FLOAT);
1623                    FORMAT_REPLACE(R64G64_FLOAT,        R32G32_FLOAT);
1624                    FORMAT_REPLACE(R64G64B64_FLOAT,     R32G32B32_FLOAT);
1625                    FORMAT_REPLACE(R64G64B64A64_FLOAT,  R32G32B32A32_FLOAT);
1626
1627                    FORMAT_REPLACE(R32_UNORM,           R32_FLOAT);
1628                    FORMAT_REPLACE(R32G32_UNORM,        R32G32_FLOAT);
1629                    FORMAT_REPLACE(R32G32B32_UNORM,     R32G32B32_FLOAT);
1630                    FORMAT_REPLACE(R32G32B32A32_UNORM,  R32G32B32A32_FLOAT);
1631
1632                    FORMAT_REPLACE(R32_USCALED,         R32_FLOAT);
1633                    FORMAT_REPLACE(R32G32_USCALED,      R32G32_FLOAT);
1634                    FORMAT_REPLACE(R32G32B32_USCALED,   R32G32B32_FLOAT);
1635                    FORMAT_REPLACE(R32G32B32A32_USCALED,R32G32B32A32_FLOAT);
1636
1637                    FORMAT_REPLACE(R32_SNORM,           R32_FLOAT);
1638                    FORMAT_REPLACE(R32G32_SNORM,        R32G32_FLOAT);
1639                    FORMAT_REPLACE(R32G32B32_SNORM,     R32G32B32_FLOAT);
1640                    FORMAT_REPLACE(R32G32B32A32_SNORM,  R32G32B32A32_FLOAT);
1641
1642                    FORMAT_REPLACE(R32_SSCALED,         R32_FLOAT);
1643                    FORMAT_REPLACE(R32G32_SSCALED,      R32G32_FLOAT);
1644                    FORMAT_REPLACE(R32G32B32_SSCALED,   R32G32B32_FLOAT);
1645                    FORMAT_REPLACE(R32G32B32A32_SSCALED,R32G32B32A32_FLOAT);
1646
1647                    FORMAT_REPLACE(R32_FIXED,           R32_FLOAT);
1648                    FORMAT_REPLACE(R32G32_FIXED,        R32G32_FLOAT);
1649                    FORMAT_REPLACE(R32G32B32_FIXED,     R32G32B32_FLOAT);
1650                    FORMAT_REPLACE(R32G32B32A32_FIXED,  R32G32B32A32_FLOAT);
1651
1652                    default:;
1653                }
1654
1655                velems->incompatible_layout =
1656                        velems->incompatible_layout ||
1657                        velems->velem[i].src_format != velems->hw_format[i] ||
1658                        velems->velem[i].src_offset % 4 != 0;
1659            }
1660
1661            /* Now setup PSC.
1662             * The unused components will be replaced by (..., 0, 1). */
1663            r300_vertex_psc(velems);
1664
1665            /* Align the formats to the size of DWORD.
1666             * We only care about the blocksizes of the formats since
1667             * swizzles are already set up.
1668             * Also compute the vertex size. */
1669            for (i = 0; i < count; i++) {
1670                /* This is OK because we check for aligned strides too. */
1671                velems->hw_format_size[i] =
1672                    align(util_format_get_blocksize(velems->hw_format[i]), 4);
1673                velems->vertex_size_dwords += velems->hw_format_size[i] / 4;
1674            }
1675        }
1676    }
1677    return velems;
1678}
1679
1680static void r300_bind_vertex_elements_state(struct pipe_context *pipe,
1681                                            void *state)
1682{
1683    struct r300_context *r300 = r300_context(pipe);
1684    struct r300_vertex_element_state *velems = state;
1685
1686    if (velems == NULL) {
1687        return;
1688    }
1689
1690    r300->velems = velems;
1691
1692    if (r300->draw) {
1693        draw_set_vertex_elements(r300->draw, velems->count, velems->velem);
1694        return;
1695    }
1696
1697    UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state);
1698    r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2;
1699}
1700
1701static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state)
1702{
1703   FREE(state);
1704}
1705
1706static void* r300_create_vs_state(struct pipe_context* pipe,
1707                                  const struct pipe_shader_state* shader)
1708{
1709    struct r300_context* r300 = r300_context(pipe);
1710    struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader);
1711
1712    /* Copy state directly into shader. */
1713    vs->state = *shader;
1714    vs->state.tokens = tgsi_dup_tokens(shader->tokens);
1715
1716    if (r300->screen->caps.has_tcl) {
1717        r300_init_vs_outputs(vs);
1718        r300_translate_vertex_shader(r300, vs);
1719    } else {
1720        r300_draw_init_vertex_shader(r300->draw, vs);
1721    }
1722
1723    return vs;
1724}
1725
1726static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
1727{
1728    struct r300_context* r300 = r300_context(pipe);
1729    struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
1730
1731    if (vs == NULL) {
1732        r300->vs_state.state = NULL;
1733        return;
1734    }
1735    if (vs == r300->vs_state.state) {
1736        return;
1737    }
1738    r300->vs_state.state = vs;
1739
1740    /* The majority of the RS block bits is dependent on the vertex shader. */
1741    r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
1742
1743    if (r300->screen->caps.has_tcl) {
1744        unsigned fc_op_dwords = r300->screen->caps.is_r500 ? 3 : 2;
1745        r300->vs_state.dirty = TRUE;
1746        r300->vs_state.size =
1747                vs->code.length + 9 +
1748                (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0) +
1749        (vs->code.num_fc_ops ? vs->code.num_fc_ops * fc_op_dwords + 4 : 0);
1750
1751        if (vs->externals_count) {
1752            r300->vs_constants.dirty = TRUE;
1753            r300->vs_constants.size = vs->externals_count * 4 + 3;
1754        } else {
1755            r300->vs_constants.size = 0;
1756        }
1757
1758        ((struct r300_constant_buffer*)r300->vs_constants.state)->remap_table =
1759                vs->code.constants_remap_table;
1760
1761        r300->pvs_flush.dirty = TRUE;
1762    } else {
1763        draw_bind_vertex_shader(r300->draw,
1764                (struct draw_vertex_shader*)vs->draw_vs);
1765    }
1766}
1767
1768static void r300_delete_vs_state(struct pipe_context* pipe, void* shader)
1769{
1770    struct r300_context* r300 = r300_context(pipe);
1771    struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
1772
1773    if (r300->screen->caps.has_tcl) {
1774        rc_constants_destroy(&vs->code.constants);
1775        if (vs->code.constants_remap_table)
1776            FREE(vs->code.constants_remap_table);
1777    } else {
1778        draw_delete_vertex_shader(r300->draw,
1779                (struct draw_vertex_shader*)vs->draw_vs);
1780    }
1781
1782    FREE((void*)vs->state.tokens);
1783    FREE(shader);
1784}
1785
1786static void r300_set_constant_buffer(struct pipe_context *pipe,
1787                                     uint shader, uint index,
1788                                     struct pipe_resource *buf)
1789{
1790    struct r300_context* r300 = r300_context(pipe);
1791    struct r300_constant_buffer *cbuf;
1792    uint32_t *mapped;
1793
1794    switch (shader) {
1795        case PIPE_SHADER_VERTEX:
1796            cbuf = (struct r300_constant_buffer*)r300->vs_constants.state;
1797            break;
1798        case PIPE_SHADER_FRAGMENT:
1799            cbuf = (struct r300_constant_buffer*)r300->fs_constants.state;
1800            break;
1801        default:
1802            assert(0);
1803            return;
1804    }
1805
1806    if (buf == NULL || buf->width0 == 0 ||
1807        (mapped = r300_buffer(buf)->constant_buffer) == NULL) {
1808        return;
1809    }
1810
1811    if (shader == PIPE_SHADER_FRAGMENT ||
1812        (shader == PIPE_SHADER_VERTEX && r300->screen->caps.has_tcl)) {
1813        assert((buf->width0 % (4 * sizeof(float))) == 0);
1814        cbuf->ptr = mapped + index*4;
1815    }
1816
1817    if (shader == PIPE_SHADER_VERTEX) {
1818        if (r300->screen->caps.has_tcl) {
1819            if (r300->vs_constants.size) {
1820                r300->vs_constants.dirty = TRUE;
1821            }
1822            r300->pvs_flush.dirty = TRUE;
1823        } else if (r300->draw) {
1824            draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX,
1825                0, mapped, buf->width0);
1826        }
1827    } else if (shader == PIPE_SHADER_FRAGMENT) {
1828        r300->fs_constants.dirty = TRUE;
1829    }
1830}
1831
1832void r300_init_state_functions(struct r300_context* r300)
1833{
1834    r300->context.create_blend_state = r300_create_blend_state;
1835    r300->context.bind_blend_state = r300_bind_blend_state;
1836    r300->context.delete_blend_state = r300_delete_blend_state;
1837
1838    r300->context.set_blend_color = r300_set_blend_color;
1839
1840    r300->context.set_clip_state = r300_set_clip_state;
1841    r300->context.set_sample_mask = r300_set_sample_mask;
1842
1843    r300->context.set_constant_buffer = r300_set_constant_buffer;
1844
1845    r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state;
1846    r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state;
1847    r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state;
1848
1849    r300->context.set_stencil_ref = r300_set_stencil_ref;
1850
1851    r300->context.set_framebuffer_state = r300_set_framebuffer_state;
1852
1853    r300->context.create_fs_state = r300_create_fs_state;
1854    r300->context.bind_fs_state = r300_bind_fs_state;
1855    r300->context.delete_fs_state = r300_delete_fs_state;
1856
1857    r300->context.set_polygon_stipple = r300_set_polygon_stipple;
1858
1859    r300->context.create_rasterizer_state = r300_create_rs_state;
1860    r300->context.bind_rasterizer_state = r300_bind_rs_state;
1861    r300->context.delete_rasterizer_state = r300_delete_rs_state;
1862
1863    r300->context.create_sampler_state = r300_create_sampler_state;
1864    r300->context.bind_fragment_sampler_states = r300_bind_sampler_states;
1865    r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures;
1866    r300->context.delete_sampler_state = r300_delete_sampler_state;
1867
1868    r300->context.set_fragment_sampler_views = r300_set_fragment_sampler_views;
1869    r300->context.create_sampler_view = r300_create_sampler_view;
1870    r300->context.sampler_view_destroy = r300_sampler_view_destroy;
1871
1872    r300->context.set_scissor_state = r300_set_scissor_state;
1873
1874    r300->context.set_viewport_state = r300_set_viewport_state;
1875
1876    r300->context.set_vertex_buffers = r300_set_vertex_buffers;
1877    r300->context.set_index_buffer = r300_set_index_buffer;
1878
1879    r300->context.create_vertex_elements_state = r300_create_vertex_elements_state;
1880    r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state;
1881    r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state;
1882
1883    r300->context.create_vs_state = r300_create_vs_state;
1884    r300->context.bind_vs_state = r300_bind_vs_state;
1885    r300->context.delete_vs_state = r300_delete_vs_state;
1886}
1887