1/*
2 * Copyright © 2006,2008 Intel Corporation
3 * Copyright © 2007 Red Hat, Inc.
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 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 *
24 * Authors:
25 *    Wang Zhenyu <zhenyu.z.wang@intel.com>
26 *    Eric Anholt <eric@anholt.net>
27 *    Carl Worth <cworth@redhat.com>
28 *    Keith Packard <keithp@keithp.com>
29 *
30 */
31
32#ifdef HAVE_CONFIG_H
33#include "config.h"
34#endif
35
36#include <assert.h>
37#include "xf86.h"
38#include "i830.h"
39#include "i915_reg.h"
40
41/* bring in brw structs */
42#include "brw_defines.h"
43#include "brw_structs.h"
44
45/* 24 = 4 vertices/composite * 3 texcoords/vertex * 2 floats/texcoord
46 *
47 * This is an upper-bound based on the case of a non-affine
48 * transformation and with a mask, but useful for sizing all cases for
49 * simplicity.
50 */
51#define VERTEX_FLOATS_PER_COMPOSITE	24
52#define VERTEX_BUFFER_SIZE		(256 * VERTEX_FLOATS_PER_COMPOSITE)
53
54struct blendinfo {
55    Bool dst_alpha;
56    Bool src_alpha;
57    uint32_t src_blend;
58    uint32_t dst_blend;
59};
60
61struct formatinfo {
62    int fmt;
63    uint32_t card_fmt;
64};
65
66// refer vol2, 3d rasterization 3.8.1
67
68/* defined in brw_defines.h */
69static struct blendinfo i965_blend_op[] = {
70    /* Clear */
71    {0, 0, BRW_BLENDFACTOR_ZERO,          BRW_BLENDFACTOR_ZERO},
72    /* Src */
73    {0, 0, BRW_BLENDFACTOR_ONE,           BRW_BLENDFACTOR_ZERO},
74    /* Dst */
75    {0, 0, BRW_BLENDFACTOR_ZERO,          BRW_BLENDFACTOR_ONE},
76    /* Over */
77    {0, 1, BRW_BLENDFACTOR_ONE,           BRW_BLENDFACTOR_INV_SRC_ALPHA},
78    /* OverReverse */
79    {1, 0, BRW_BLENDFACTOR_INV_DST_ALPHA, BRW_BLENDFACTOR_ONE},
80    /* In */
81    {1, 0, BRW_BLENDFACTOR_DST_ALPHA,     BRW_BLENDFACTOR_ZERO},
82    /* InReverse */
83    {0, 1, BRW_BLENDFACTOR_ZERO,          BRW_BLENDFACTOR_SRC_ALPHA},
84    /* Out */
85    {1, 0, BRW_BLENDFACTOR_INV_DST_ALPHA, BRW_BLENDFACTOR_ZERO},
86    /* OutReverse */
87    {0, 1, BRW_BLENDFACTOR_ZERO,          BRW_BLENDFACTOR_INV_SRC_ALPHA},
88    /* Atop */
89    {1, 1, BRW_BLENDFACTOR_DST_ALPHA,     BRW_BLENDFACTOR_INV_SRC_ALPHA},
90    /* AtopReverse */
91    {1, 1, BRW_BLENDFACTOR_INV_DST_ALPHA, BRW_BLENDFACTOR_SRC_ALPHA},
92    /* Xor */
93    {1, 1, BRW_BLENDFACTOR_INV_DST_ALPHA, BRW_BLENDFACTOR_INV_SRC_ALPHA},
94    /* Add */
95    {0, 0, BRW_BLENDFACTOR_ONE,           BRW_BLENDFACTOR_ONE},
96};
97/**
98 * Highest-valued BLENDFACTOR used in i965_blend_op.
99 *
100 * This leaves out BRW_BLENDFACTOR_INV_DST_COLOR,
101 * BRW_BLENDFACTOR_INV_CONST_{COLOR,ALPHA},
102 * BRW_BLENDFACTOR_INV_SRC1_{COLOR,ALPHA}
103 */
104#define BRW_BLENDFACTOR_COUNT (BRW_BLENDFACTOR_INV_DST_ALPHA + 1)
105
106/* FIXME: surface format defined in brw_defines.h, shared Sampling engine
107 * 1.7.2
108 */
109static struct formatinfo i965_tex_formats[] = {
110    {PICT_a8r8g8b8, BRW_SURFACEFORMAT_B8G8R8A8_UNORM },
111    {PICT_x8r8g8b8, BRW_SURFACEFORMAT_B8G8R8X8_UNORM },
112    {PICT_a8b8g8r8, BRW_SURFACEFORMAT_R8G8B8A8_UNORM },
113    {PICT_x8b8g8r8, BRW_SURFACEFORMAT_R8G8B8X8_UNORM },
114    {PICT_r5g6b5,   BRW_SURFACEFORMAT_B5G6R5_UNORM   },
115    {PICT_a1r5g5b5, BRW_SURFACEFORMAT_B5G5R5A1_UNORM },
116    {PICT_a8,       BRW_SURFACEFORMAT_A8_UNORM	 },
117};
118
119static void i965_get_blend_cntl(int op, PicturePtr pMask, uint32_t dst_format,
120				uint32_t *sblend, uint32_t *dblend)
121{
122
123    *sblend = i965_blend_op[op].src_blend;
124    *dblend = i965_blend_op[op].dst_blend;
125
126    /* If there's no dst alpha channel, adjust the blend op so that we'll treat
127     * it as always 1.
128     */
129    if (PICT_FORMAT_A(dst_format) == 0 && i965_blend_op[op].dst_alpha) {
130        if (*sblend == BRW_BLENDFACTOR_DST_ALPHA)
131            *sblend = BRW_BLENDFACTOR_ONE;
132        else if (*sblend == BRW_BLENDFACTOR_INV_DST_ALPHA)
133            *sblend = BRW_BLENDFACTOR_ZERO;
134    }
135
136    /* If the source alpha is being used, then we should only be in a case where
137     * the source blend factor is 0, and the source blend value is the mask
138     * channels multiplied by the source picture's alpha.
139     */
140    if (pMask && pMask->componentAlpha && PICT_FORMAT_RGB(pMask->format)
141            && i965_blend_op[op].src_alpha) {
142        if (*dblend == BRW_BLENDFACTOR_SRC_ALPHA) {
143	    *dblend = BRW_BLENDFACTOR_SRC_COLOR;
144        } else if (*dblend == BRW_BLENDFACTOR_INV_SRC_ALPHA) {
145	    *dblend = BRW_BLENDFACTOR_INV_SRC_COLOR;
146        }
147    }
148
149}
150
151static Bool i965_get_dest_format(PicturePtr pDstPicture, uint32_t *dst_format)
152{
153    ScrnInfoPtr pScrn = xf86Screens[pDstPicture->pDrawable->pScreen->myNum];
154
155    switch (pDstPicture->format) {
156    case PICT_a8r8g8b8:
157    case PICT_x8r8g8b8:
158        *dst_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
159        break;
160    case PICT_r5g6b5:
161        *dst_format = BRW_SURFACEFORMAT_B5G6R5_UNORM;
162        break;
163    case PICT_a1r5g5b5:
164    	*dst_format = BRW_SURFACEFORMAT_B5G5R5A1_UNORM;
165	break;
166    case PICT_x1r5g5b5:
167        *dst_format = BRW_SURFACEFORMAT_B5G5R5X1_UNORM;
168        break;
169    case PICT_a8:
170        *dst_format = BRW_SURFACEFORMAT_A8_UNORM;
171        break;
172    case PICT_a4r4g4b4:
173    case PICT_x4r4g4b4:
174	*dst_format = BRW_SURFACEFORMAT_B4G4R4A4_UNORM;
175	break;
176    default:
177        I830FALLBACK("Unsupported dest format 0x%x\n",
178		     (int)pDstPicture->format);
179    }
180
181    return TRUE;
182}
183
184static Bool i965_check_composite_texture(ScrnInfoPtr pScrn, PicturePtr pPict, int unit)
185{
186    if (pPict->repeatType > RepeatReflect)
187	I830FALLBACK("extended repeat (%d) not supported\n",
188		     pPict->repeatType);
189
190    if (pPict->filter != PictFilterNearest &&
191        pPict->filter != PictFilterBilinear)
192    {
193        I830FALLBACK("Unsupported filter 0x%x\n", pPict->filter);
194    }
195
196    if (pPict->pDrawable)
197    {
198	int w, h, i;
199
200	w = pPict->pDrawable->width;
201	h = pPict->pDrawable->height;
202	if ((w > 8192) || (h > 8192))
203	    I830FALLBACK("Picture w/h too large (%dx%d)\n", w, h);
204
205	for (i = 0; i < sizeof(i965_tex_formats) / sizeof(i965_tex_formats[0]);
206	     i++)
207	{
208	    if (i965_tex_formats[i].fmt == pPict->format)
209		break;
210	}
211	if (i == sizeof(i965_tex_formats) / sizeof(i965_tex_formats[0]))
212	    I830FALLBACK("Unsupported picture format 0x%x\n",
213			 (int)pPict->format);
214    }
215
216    return TRUE;
217}
218
219Bool
220i965_check_composite(int op, PicturePtr pSrcPicture, PicturePtr pMaskPicture,
221		     PicturePtr pDstPicture)
222{
223    ScrnInfoPtr pScrn = xf86Screens[pDstPicture->pDrawable->pScreen->myNum];
224    uint32_t tmp1;
225
226    /* Check for unsupported compositing operations. */
227    if (op >= sizeof(i965_blend_op) / sizeof(i965_blend_op[0]))
228        I830FALLBACK("Unsupported Composite op 0x%x\n", op);
229
230    if (pMaskPicture && pMaskPicture->componentAlpha &&
231            PICT_FORMAT_RGB(pMaskPicture->format)) {
232        /* Check if it's component alpha that relies on a source alpha and on
233         * the source value.  We can only get one of those into the single
234         * source value that we get to blend with.
235         */
236        if (i965_blend_op[op].src_alpha &&
237            (i965_blend_op[op].src_blend != BRW_BLENDFACTOR_ZERO))
238	{
239	    I830FALLBACK("Component alpha not supported with source "
240			 "alpha and source value blending.\n");
241	}
242    }
243
244    if (!i965_check_composite_texture(pScrn, pSrcPicture, 0))
245        I830FALLBACK("Check Src picture texture\n");
246    if (pMaskPicture != NULL && !i965_check_composite_texture(pScrn, pMaskPicture, 1))
247        I830FALLBACK("Check Mask picture texture\n");
248
249    if (!i965_get_dest_format(pDstPicture, &tmp1))
250	I830FALLBACK("Get Color buffer format\n");
251
252    return TRUE;
253
254}
255
256#define BRW_GRF_BLOCKS(nreg)    ((nreg + 15) / 16 - 1)
257
258/* Set up a default static partitioning of the URB, which is supposed to
259 * allow anything we would want to do, at potentially lower performance.
260 */
261#define URB_CS_ENTRY_SIZE     0
262#define URB_CS_ENTRIES	      0
263
264#define URB_VS_ENTRY_SIZE     1	  // each 512-bit row
265#define URB_VS_ENTRIES	      8	  // we needs at least 8 entries
266
267#define URB_GS_ENTRY_SIZE     0
268#define URB_GS_ENTRIES	      0
269
270#define URB_CLIP_ENTRY_SIZE   0
271#define URB_CLIP_ENTRIES      0
272
273#define URB_SF_ENTRY_SIZE     2
274#define URB_SF_ENTRIES	      1
275
276static const uint32_t sip_kernel_static[][4] = {
277/*    wait (1) a0<1>UW a145<0,1,0>UW { align1 +  } */
278    { 0x00000030, 0x20000108, 0x00001220, 0x00000000 },
279/*    nop (4) g0<1>UD { align1 +  } */
280    { 0x0040007e, 0x20000c21, 0x00690000, 0x00000000 },
281/*    nop (4) g0<1>UD { align1 +  } */
282    { 0x0040007e, 0x20000c21, 0x00690000, 0x00000000 },
283/*    nop (4) g0<1>UD { align1 +  } */
284    { 0x0040007e, 0x20000c21, 0x00690000, 0x00000000 },
285/*    nop (4) g0<1>UD { align1 +  } */
286    { 0x0040007e, 0x20000c21, 0x00690000, 0x00000000 },
287/*    nop (4) g0<1>UD { align1 +  } */
288    { 0x0040007e, 0x20000c21, 0x00690000, 0x00000000 },
289/*    nop (4) g0<1>UD { align1 +  } */
290    { 0x0040007e, 0x20000c21, 0x00690000, 0x00000000 },
291/*    nop (4) g0<1>UD { align1 +  } */
292    { 0x0040007e, 0x20000c21, 0x00690000, 0x00000000 },
293/*    nop (4) g0<1>UD { align1 +  } */
294    { 0x0040007e, 0x20000c21, 0x00690000, 0x00000000 },
295/*    nop (4) g0<1>UD { align1 +  } */
296    { 0x0040007e, 0x20000c21, 0x00690000, 0x00000000 },
297};
298
299/*
300 * this program computes dA/dx and dA/dy for the texture coordinates along
301 * with the base texture coordinate. It was extracted from the Mesa driver
302 */
303
304#define SF_KERNEL_NUM_GRF  16
305#define SF_MAX_THREADS	   2
306
307static const uint32_t sf_kernel_static[][4] = {
308#include "exa_sf.g4b"
309};
310
311static const uint32_t sf_kernel_mask_static[][4] = {
312#include "exa_sf_mask.g4b"
313};
314
315/* ps kernels */
316#define PS_KERNEL_NUM_GRF   32
317#define PS_MAX_THREADS	    48
318
319static const uint32_t ps_kernel_nomask_affine_static [][4] = {
320#include "exa_wm_xy.g4b"
321#include "exa_wm_src_affine.g4b"
322#include "exa_wm_src_sample_argb.g4b"
323#include "exa_wm_write.g4b"
324};
325
326static const uint32_t ps_kernel_nomask_projective_static [][4] = {
327#include "exa_wm_xy.g4b"
328#include "exa_wm_src_projective.g4b"
329#include "exa_wm_src_sample_argb.g4b"
330#include "exa_wm_write.g4b"
331};
332
333static const uint32_t ps_kernel_maskca_affine_static [][4] = {
334#include "exa_wm_xy.g4b"
335#include "exa_wm_src_affine.g4b"
336#include "exa_wm_src_sample_argb.g4b"
337#include "exa_wm_mask_affine.g4b"
338#include "exa_wm_mask_sample_argb.g4b"
339#include "exa_wm_ca.g4b"
340#include "exa_wm_write.g4b"
341};
342
343static const uint32_t ps_kernel_maskca_projective_static [][4] = {
344#include "exa_wm_xy.g4b"
345#include "exa_wm_src_projective.g4b"
346#include "exa_wm_src_sample_argb.g4b"
347#include "exa_wm_mask_projective.g4b"
348#include "exa_wm_mask_sample_argb.g4b"
349#include "exa_wm_ca.g4b"
350#include "exa_wm_write.g4b"
351};
352
353static const uint32_t ps_kernel_maskca_srcalpha_affine_static [][4] = {
354#include "exa_wm_xy.g4b"
355#include "exa_wm_src_affine.g4b"
356#include "exa_wm_src_sample_a.g4b"
357#include "exa_wm_mask_affine.g4b"
358#include "exa_wm_mask_sample_argb.g4b"
359#include "exa_wm_ca_srcalpha.g4b"
360#include "exa_wm_write.g4b"
361};
362
363static const uint32_t ps_kernel_maskca_srcalpha_projective_static [][4] = {
364#include "exa_wm_xy.g4b"
365#include "exa_wm_src_projective.g4b"
366#include "exa_wm_src_sample_a.g4b"
367#include "exa_wm_mask_projective.g4b"
368#include "exa_wm_mask_sample_argb.g4b"
369#include "exa_wm_ca_srcalpha.g4b"
370#include "exa_wm_write.g4b"
371};
372
373static const uint32_t ps_kernel_masknoca_affine_static [][4] = {
374#include "exa_wm_xy.g4b"
375#include "exa_wm_src_affine.g4b"
376#include "exa_wm_src_sample_argb.g4b"
377#include "exa_wm_mask_affine.g4b"
378#include "exa_wm_mask_sample_a.g4b"
379#include "exa_wm_noca.g4b"
380#include "exa_wm_write.g4b"
381};
382
383static const uint32_t ps_kernel_masknoca_projective_static [][4] = {
384#include "exa_wm_xy.g4b"
385#include "exa_wm_src_projective.g4b"
386#include "exa_wm_src_sample_argb.g4b"
387#include "exa_wm_mask_projective.g4b"
388#include "exa_wm_mask_sample_a.g4b"
389#include "exa_wm_noca.g4b"
390#include "exa_wm_write.g4b"
391};
392
393/* new programs for IGDNG */
394static const uint32_t sf_kernel_static_gen5[][4] = {
395#include "exa_sf.g4b.gen5"
396};
397
398static const uint32_t sf_kernel_mask_static_gen5[][4] = {
399#include "exa_sf_mask.g4b.gen5"
400};
401
402static const uint32_t ps_kernel_nomask_affine_static_gen5 [][4] = {
403#include "exa_wm_xy.g4b.gen5"
404#include "exa_wm_src_affine.g4b.gen5"
405#include "exa_wm_src_sample_argb.g4b.gen5"
406#include "exa_wm_write.g4b.gen5"
407};
408
409static const uint32_t ps_kernel_nomask_projective_static_gen5 [][4] = {
410#include "exa_wm_xy.g4b.gen5"
411#include "exa_wm_src_projective.g4b.gen5"
412#include "exa_wm_src_sample_argb.g4b.gen5"
413#include "exa_wm_write.g4b.gen5"
414};
415
416static const uint32_t ps_kernel_maskca_affine_static_gen5 [][4] = {
417#include "exa_wm_xy.g4b.gen5"
418#include "exa_wm_src_affine.g4b.gen5"
419#include "exa_wm_src_sample_argb.g4b.gen5"
420#include "exa_wm_mask_affine.g4b.gen5"
421#include "exa_wm_mask_sample_argb.g4b.gen5"
422#include "exa_wm_ca.g4b.gen5"
423#include "exa_wm_write.g4b.gen5"
424};
425
426static const uint32_t ps_kernel_maskca_projective_static_gen5 [][4] = {
427#include "exa_wm_xy.g4b.gen5"
428#include "exa_wm_src_projective.g4b.gen5"
429#include "exa_wm_src_sample_argb.g4b.gen5"
430#include "exa_wm_mask_projective.g4b.gen5"
431#include "exa_wm_mask_sample_argb.g4b.gen5"
432#include "exa_wm_ca.g4b.gen5"
433#include "exa_wm_write.g4b.gen5"
434};
435
436static const uint32_t ps_kernel_maskca_srcalpha_affine_static_gen5 [][4] = {
437#include "exa_wm_xy.g4b.gen5"
438#include "exa_wm_src_affine.g4b.gen5"
439#include "exa_wm_src_sample_a.g4b.gen5"
440#include "exa_wm_mask_affine.g4b.gen5"
441#include "exa_wm_mask_sample_argb.g4b.gen5"
442#include "exa_wm_ca_srcalpha.g4b.gen5"
443#include "exa_wm_write.g4b.gen5"
444};
445
446static const uint32_t ps_kernel_maskca_srcalpha_projective_static_gen5 [][4] = {
447#include "exa_wm_xy.g4b.gen5"
448#include "exa_wm_src_projective.g4b.gen5"
449#include "exa_wm_src_sample_a.g4b.gen5"
450#include "exa_wm_mask_projective.g4b.gen5"
451#include "exa_wm_mask_sample_argb.g4b.gen5"
452#include "exa_wm_ca_srcalpha.g4b.gen5"
453#include "exa_wm_write.g4b.gen5"
454};
455
456static const uint32_t ps_kernel_masknoca_affine_static_gen5 [][4] = {
457#include "exa_wm_xy.g4b.gen5"
458#include "exa_wm_src_affine.g4b.gen5"
459#include "exa_wm_src_sample_argb.g4b.gen5"
460#include "exa_wm_mask_affine.g4b.gen5"
461#include "exa_wm_mask_sample_a.g4b.gen5"
462#include "exa_wm_noca.g4b.gen5"
463#include "exa_wm_write.g4b.gen5"
464};
465
466static const uint32_t ps_kernel_masknoca_projective_static_gen5 [][4] = {
467#include "exa_wm_xy.g4b.gen5"
468#include "exa_wm_src_projective.g4b.gen5"
469#include "exa_wm_src_sample_argb.g4b.gen5"
470#include "exa_wm_mask_projective.g4b.gen5"
471#include "exa_wm_mask_sample_a.g4b.gen5"
472#include "exa_wm_noca.g4b.gen5"
473#include "exa_wm_write.g4b.gen5"
474};
475
476#define WM_STATE_DECL(kernel) \
477    struct brw_wm_unit_state wm_state_ ## kernel[SAMPLER_STATE_FILTER_COUNT] \
478						[SAMPLER_STATE_EXTEND_COUNT] \
479						[SAMPLER_STATE_FILTER_COUNT] \
480						[SAMPLER_STATE_EXTEND_COUNT]
481
482/* Many of the fields in the state structure must be aligned to a
483 * 64-byte boundary, (or a 32-byte boundary, but 64 is good enough for
484 * those too).
485 */
486#define PAD64_MULTI(previous, idx, factor) char previous ## _pad ## idx [(64 - (sizeof(struct previous) * (factor)) % 64) % 64]
487#define PAD64(previous, idx) PAD64_MULTI(previous, idx, 1)
488
489typedef enum {
490    SAMPLER_STATE_FILTER_NEAREST,
491    SAMPLER_STATE_FILTER_BILINEAR,
492    SAMPLER_STATE_FILTER_COUNT
493} sampler_state_filter_t;
494
495typedef enum {
496    SAMPLER_STATE_EXTEND_NONE,
497    SAMPLER_STATE_EXTEND_REPEAT,
498    SAMPLER_STATE_EXTEND_PAD,
499    SAMPLER_STATE_EXTEND_REFLECT,
500    SAMPLER_STATE_EXTEND_COUNT
501} sampler_state_extend_t;
502
503typedef enum {
504    WM_KERNEL_NOMASK_AFFINE,
505    WM_KERNEL_NOMASK_PROJECTIVE,
506    WM_KERNEL_MASKCA_AFFINE,
507    WM_KERNEL_MASKCA_PROJECTIVE,
508    WM_KERNEL_MASKCA_SRCALPHA_AFFINE,
509    WM_KERNEL_MASKCA_SRCALPHA_PROJECTIVE,
510    WM_KERNEL_MASKNOCA_AFFINE,
511    WM_KERNEL_MASKNOCA_PROJECTIVE,
512    WM_KERNEL_COUNT
513} wm_kernel_t;
514
515#define KERNEL(kernel_enum, kernel, masked) \
516    [kernel_enum] = {__UNCONST(&kernel), sizeof(kernel), masked}
517struct wm_kernel_info {
518    void *data;
519    unsigned int size;
520    Bool has_mask;
521};
522
523static struct wm_kernel_info wm_kernels[] = {
524    KERNEL(WM_KERNEL_NOMASK_AFFINE,
525	   ps_kernel_nomask_affine_static, FALSE),
526    KERNEL(WM_KERNEL_NOMASK_PROJECTIVE,
527	   ps_kernel_nomask_projective_static, FALSE),
528    KERNEL(WM_KERNEL_MASKCA_AFFINE,
529	   ps_kernel_maskca_affine_static, TRUE),
530    KERNEL(WM_KERNEL_MASKCA_PROJECTIVE,
531	   ps_kernel_maskca_projective_static, TRUE),
532    KERNEL(WM_KERNEL_MASKCA_SRCALPHA_AFFINE,
533	   ps_kernel_maskca_srcalpha_affine_static, TRUE),
534    KERNEL(WM_KERNEL_MASKCA_SRCALPHA_PROJECTIVE,
535	   ps_kernel_maskca_srcalpha_projective_static, TRUE),
536    KERNEL(WM_KERNEL_MASKNOCA_AFFINE,
537	   ps_kernel_masknoca_affine_static, TRUE),
538    KERNEL(WM_KERNEL_MASKNOCA_PROJECTIVE,
539	   ps_kernel_masknoca_projective_static, TRUE),
540};
541
542static struct wm_kernel_info wm_kernels_gen5[] = {
543    KERNEL(WM_KERNEL_NOMASK_AFFINE,
544	   ps_kernel_nomask_affine_static_gen5, FALSE),
545    KERNEL(WM_KERNEL_NOMASK_PROJECTIVE,
546	   ps_kernel_nomask_projective_static_gen5, FALSE),
547    KERNEL(WM_KERNEL_MASKCA_AFFINE,
548	   ps_kernel_maskca_affine_static_gen5, TRUE),
549    KERNEL(WM_KERNEL_MASKCA_PROJECTIVE,
550	   ps_kernel_maskca_projective_static_gen5, TRUE),
551    KERNEL(WM_KERNEL_MASKCA_SRCALPHA_AFFINE,
552	   ps_kernel_maskca_srcalpha_affine_static_gen5, TRUE),
553    KERNEL(WM_KERNEL_MASKCA_SRCALPHA_PROJECTIVE,
554	   ps_kernel_maskca_srcalpha_projective_static_gen5, TRUE),
555    KERNEL(WM_KERNEL_MASKNOCA_AFFINE,
556	   ps_kernel_masknoca_affine_static_gen5, TRUE),
557    KERNEL(WM_KERNEL_MASKNOCA_PROJECTIVE,
558	   ps_kernel_masknoca_projective_static_gen5, TRUE),
559};
560#undef KERNEL
561
562typedef struct _brw_cc_unit_state_padded {
563    struct brw_cc_unit_state state;
564    char pad[64 - sizeof (struct brw_cc_unit_state)];
565} brw_cc_unit_state_padded;
566
567typedef struct brw_surface_state_padded {
568    struct brw_surface_state state;
569    char pad[32 - sizeof (struct brw_surface_state)];
570} brw_surface_state_padded;
571
572struct gen4_cc_unit_state {
573    /* Index by [src_blend][dst_blend] */
574    brw_cc_unit_state_padded cc_state[BRW_BLENDFACTOR_COUNT]
575				     [BRW_BLENDFACTOR_COUNT];
576};
577
578typedef float gen4_vertex_buffer[VERTEX_BUFFER_SIZE];
579
580typedef struct gen4_composite_op {
581    int		op;
582    PicturePtr	source_picture;
583    PicturePtr	mask_picture;
584    PicturePtr	dest_picture;
585    PixmapPtr	source;
586    PixmapPtr	mask;
587    PixmapPtr	dest;
588    drm_intel_bo *binding_table_bo;
589    sampler_state_filter_t src_filter;
590    sampler_state_filter_t mask_filter;
591    sampler_state_extend_t src_extend;
592    sampler_state_extend_t mask_extend;
593    Bool is_affine;
594    wm_kernel_t wm_kernel;
595} gen4_composite_op;
596
597/** Private data for gen4 render accel implementation. */
598struct gen4_render_state {
599    drm_intel_bo *vs_state_bo;
600    drm_intel_bo *sf_state_bo;
601    drm_intel_bo *sf_mask_state_bo;
602    drm_intel_bo *cc_state_bo;
603    drm_intel_bo *wm_state_bo[WM_KERNEL_COUNT]
604			     [SAMPLER_STATE_FILTER_COUNT]
605			     [SAMPLER_STATE_EXTEND_COUNT]
606			     [SAMPLER_STATE_FILTER_COUNT]
607			     [SAMPLER_STATE_EXTEND_COUNT];
608    drm_intel_bo *wm_kernel_bo[WM_KERNEL_COUNT];
609
610    drm_intel_bo *sip_kernel_bo;
611    dri_bo* vertex_buffer_bo;
612
613    gen4_composite_op composite_op;
614
615    int vb_offset;
616    int vertex_size;
617
618    Bool needs_state_emit;
619};
620
621/**
622 * Sets up the SF state pointing at an SF kernel.
623 *
624 * The SF kernel does coord interp: for each attribute,
625 * calculate dA/dx and dA/dy.  Hand these interpolation coefficients
626 * back to SF which then hands pixels off to WM.
627 */
628static drm_intel_bo *
629gen4_create_sf_state(ScrnInfoPtr scrn, drm_intel_bo *kernel_bo)
630{
631    I830Ptr pI830 = I830PTR(scrn);
632    struct brw_sf_unit_state *sf_state;
633    drm_intel_bo *sf_state_bo;
634
635    sf_state_bo = drm_intel_bo_alloc(pI830->bufmgr, "gen4 SF state",
636				     sizeof(*sf_state), 4096);
637    drm_intel_bo_map(sf_state_bo, TRUE);
638    sf_state = sf_state_bo->virtual;
639
640    memset(sf_state, 0, sizeof(*sf_state));
641    sf_state->thread0.grf_reg_count = BRW_GRF_BLOCKS(SF_KERNEL_NUM_GRF);
642    sf_state->thread0.kernel_start_pointer =
643	intel_emit_reloc(sf_state_bo,
644			 offsetof(struct brw_sf_unit_state, thread0),
645			 kernel_bo, sf_state->thread0.grf_reg_count << 1,
646			 I915_GEM_DOMAIN_INSTRUCTION, 0) >> 6;
647    sf_state->sf1.single_program_flow = 1;
648    sf_state->sf1.binding_table_entry_count = 0;
649    sf_state->sf1.thread_priority = 0;
650    sf_state->sf1.floating_point_mode = 0; /* Mesa does this */
651    sf_state->sf1.illegal_op_exception_enable = 1;
652    sf_state->sf1.mask_stack_exception_enable = 1;
653    sf_state->sf1.sw_exception_enable = 1;
654    sf_state->thread2.per_thread_scratch_space = 0;
655    /* scratch space is not used in our kernel */
656    sf_state->thread2.scratch_space_base_pointer = 0;
657    sf_state->thread3.const_urb_entry_read_length = 0; /* no const URBs */
658    sf_state->thread3.const_urb_entry_read_offset = 0; /* no const URBs */
659    sf_state->thread3.urb_entry_read_length = 1; /* 1 URB per vertex */
660    /* don't smash vertex header, read start from dw8 */
661    sf_state->thread3.urb_entry_read_offset = 1;
662    sf_state->thread3.dispatch_grf_start_reg = 3;
663    sf_state->thread4.max_threads = SF_MAX_THREADS - 1;
664    sf_state->thread4.urb_entry_allocation_size = URB_SF_ENTRY_SIZE - 1;
665    sf_state->thread4.nr_urb_entries = URB_SF_ENTRIES;
666    sf_state->thread4.stats_enable = 1;
667    sf_state->sf5.viewport_transform = FALSE; /* skip viewport */
668    sf_state->sf6.cull_mode = BRW_CULLMODE_NONE;
669    sf_state->sf6.scissor = 0;
670    sf_state->sf7.trifan_pv = 2;
671    sf_state->sf6.dest_org_vbias = 0x8;
672    sf_state->sf6.dest_org_hbias = 0x8;
673
674    drm_intel_bo_unmap(sf_state_bo);
675
676    return sf_state_bo;
677}
678
679static drm_intel_bo *
680sampler_border_color_create(ScrnInfoPtr scrn)
681{
682    struct brw_sampler_legacy_border_color sampler_border_color;
683
684    /* Set up the sampler border color (always transparent black) */
685    memset(&sampler_border_color, 0, sizeof(sampler_border_color));
686    sampler_border_color.color[0] = 0; /* R */
687    sampler_border_color.color[1] = 0; /* G */
688    sampler_border_color.color[2] = 0; /* B */
689    sampler_border_color.color[3] = 0; /* A */
690
691    return intel_bo_alloc_for_data(scrn,
692				   &sampler_border_color,
693				   sizeof(sampler_border_color),
694				   "gen4 render sampler border color");
695}
696
697static void
698sampler_state_init (drm_intel_bo *sampler_state_bo,
699		    struct brw_sampler_state *sampler_state,
700		    sampler_state_filter_t filter,
701		    sampler_state_extend_t extend,
702		    drm_intel_bo *border_color_bo)
703{
704    uint32_t sampler_state_offset;
705
706    sampler_state_offset = (char *)sampler_state -
707	(char *)sampler_state_bo->virtual;
708
709    /* PS kernel use this sampler */
710    memset(sampler_state, 0, sizeof(*sampler_state));
711
712    sampler_state->ss0.lod_preclamp = 1; /* GL mode */
713
714    /* We use the legacy mode to get the semantics specified by
715     * the Render extension. */
716    sampler_state->ss0.border_color_mode = BRW_BORDER_COLOR_MODE_LEGACY;
717
718    switch(filter) {
719    default:
720    case SAMPLER_STATE_FILTER_NEAREST:
721	sampler_state->ss0.min_filter = BRW_MAPFILTER_NEAREST;
722	sampler_state->ss0.mag_filter = BRW_MAPFILTER_NEAREST;
723	break;
724    case SAMPLER_STATE_FILTER_BILINEAR:
725	sampler_state->ss0.min_filter = BRW_MAPFILTER_LINEAR;
726	sampler_state->ss0.mag_filter = BRW_MAPFILTER_LINEAR;
727	break;
728    }
729
730    switch (extend) {
731    default:
732    case SAMPLER_STATE_EXTEND_NONE:
733	sampler_state->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CLAMP_BORDER;
734	sampler_state->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CLAMP_BORDER;
735	sampler_state->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CLAMP_BORDER;
736	break;
737    case SAMPLER_STATE_EXTEND_REPEAT:
738	sampler_state->ss1.r_wrap_mode = BRW_TEXCOORDMODE_WRAP;
739	sampler_state->ss1.s_wrap_mode = BRW_TEXCOORDMODE_WRAP;
740	sampler_state->ss1.t_wrap_mode = BRW_TEXCOORDMODE_WRAP;
741	break;
742    case SAMPLER_STATE_EXTEND_PAD:
743	sampler_state->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
744	sampler_state->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
745	sampler_state->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
746	break;
747    case SAMPLER_STATE_EXTEND_REFLECT:
748	sampler_state->ss1.r_wrap_mode = BRW_TEXCOORDMODE_MIRROR;
749	sampler_state->ss1.s_wrap_mode = BRW_TEXCOORDMODE_MIRROR;
750	sampler_state->ss1.t_wrap_mode = BRW_TEXCOORDMODE_MIRROR;
751	break;
752    }
753
754    sampler_state->ss2.border_color_pointer =
755	intel_emit_reloc(sampler_state_bo, sampler_state_offset +
756			 offsetof(struct brw_sampler_state, ss2),
757			 border_color_bo, 0,
758			 I915_GEM_DOMAIN_SAMPLER, 0) >> 5;
759
760    sampler_state->ss3.chroma_key_enable = 0; /* disable chromakey */
761}
762
763static drm_intel_bo *
764gen4_create_sampler_state(ScrnInfoPtr scrn,
765			  sampler_state_filter_t src_filter,
766			  sampler_state_extend_t src_extend,
767			  sampler_state_filter_t mask_filter,
768			  sampler_state_extend_t mask_extend,
769			  drm_intel_bo *border_color_bo)
770{
771    I830Ptr pI830 = I830PTR(scrn);
772    drm_intel_bo *sampler_state_bo;
773    struct brw_sampler_state *sampler_state;
774
775    sampler_state_bo = drm_intel_bo_alloc(pI830->bufmgr, "gen4 sampler state",
776					  sizeof(struct brw_sampler_state) * 2,
777					  4096);
778    drm_intel_bo_map(sampler_state_bo, TRUE);
779    sampler_state = sampler_state_bo->virtual;
780
781    sampler_state_init(sampler_state_bo,
782		       &sampler_state[0],
783		       src_filter, src_extend,
784		       border_color_bo);
785    sampler_state_init(sampler_state_bo,
786		       &sampler_state[1],
787		       mask_filter, mask_extend,
788		       border_color_bo);
789
790    drm_intel_bo_unmap(sampler_state_bo);
791
792    return sampler_state_bo;
793}
794
795static void
796cc_state_init (drm_intel_bo *cc_state_bo,
797	       uint32_t cc_state_offset,
798	       int src_blend,
799	       int dst_blend,
800	       drm_intel_bo *cc_vp_bo)
801{
802    struct brw_cc_unit_state *cc_state;
803
804    cc_state = (struct brw_cc_unit_state *)((char *)cc_state_bo->virtual +
805					    cc_state_offset);
806
807    memset(cc_state, 0, sizeof(*cc_state));
808    cc_state->cc0.stencil_enable = 0;   /* disable stencil */
809    cc_state->cc2.depth_test = 0;       /* disable depth test */
810    cc_state->cc2.logicop_enable = 0;   /* disable logic op */
811    cc_state->cc3.ia_blend_enable = 0;  /* blend alpha same as colors */
812    cc_state->cc3.blend_enable = 1;     /* enable color blend */
813    cc_state->cc3.alpha_test = 0;       /* disable alpha test */
814
815    cc_state->cc4.cc_viewport_state_offset =
816	intel_emit_reloc(cc_state_bo, cc_state_offset +
817			 offsetof(struct brw_cc_unit_state, cc4),
818			 cc_vp_bo, 0,
819			 I915_GEM_DOMAIN_INSTRUCTION, 0) >> 5;
820
821    cc_state->cc5.dither_enable = 0;    /* disable dither */
822    cc_state->cc5.logicop_func = 0xc;   /* COPY */
823    cc_state->cc5.statistics_enable = 1;
824    cc_state->cc5.ia_blend_function = BRW_BLENDFUNCTION_ADD;
825
826    /* Fill in alpha blend factors same as color, for the future. */
827    cc_state->cc5.ia_src_blend_factor = src_blend;
828    cc_state->cc5.ia_dest_blend_factor = dst_blend;
829
830    cc_state->cc6.blend_function = BRW_BLENDFUNCTION_ADD;
831    cc_state->cc6.clamp_post_alpha_blend = 1;
832    cc_state->cc6.clamp_pre_alpha_blend = 1;
833    cc_state->cc6.clamp_range = 0;  /* clamp range [0,1] */
834
835    cc_state->cc6.src_blend_factor = src_blend;
836    cc_state->cc6.dest_blend_factor = dst_blend;
837}
838
839static drm_intel_bo *
840gen4_create_wm_state(ScrnInfoPtr scrn,
841		     Bool has_mask, drm_intel_bo *kernel_bo,
842		     drm_intel_bo *sampler_bo)
843{
844    I830Ptr pI830 = I830PTR(scrn);
845    struct brw_wm_unit_state *wm_state;
846    drm_intel_bo *wm_state_bo;
847
848    wm_state_bo = drm_intel_bo_alloc(pI830->bufmgr, "gen4 WM state",
849				     sizeof(*wm_state), 4096);
850    drm_intel_bo_map(wm_state_bo, TRUE);
851    wm_state = wm_state_bo->virtual;
852
853    memset(wm_state, 0, sizeof (*wm_state));
854    wm_state->thread0.grf_reg_count = BRW_GRF_BLOCKS(PS_KERNEL_NUM_GRF);
855    wm_state->thread0.kernel_start_pointer =
856	intel_emit_reloc(wm_state_bo,
857			 offsetof(struct brw_wm_unit_state, thread0),
858                         kernel_bo, wm_state->thread0.grf_reg_count << 1,
859                         I915_GEM_DOMAIN_INSTRUCTION, 0) >> 6;
860
861    wm_state->thread1.single_program_flow = 0;
862
863    /* scratch space is not used in our kernel */
864    wm_state->thread2.scratch_space_base_pointer = 0;
865    wm_state->thread2.per_thread_scratch_space = 0;
866
867    wm_state->thread3.const_urb_entry_read_length = 0;
868    wm_state->thread3.const_urb_entry_read_offset = 0;
869
870    wm_state->thread3.urb_entry_read_offset = 0;
871    /* wm kernel use urb from 3, see wm_program in compiler module */
872    wm_state->thread3.dispatch_grf_start_reg = 3; /* must match kernel */
873
874    wm_state->wm4.stats_enable = 1;  /* statistic */
875
876    if (IS_IGDNG(pI830))
877        wm_state->wm4.sampler_count = 0; /* hardware requirement */
878    else
879        wm_state->wm4.sampler_count = 1; /* 1-4 samplers used */
880
881    wm_state->wm4.sampler_state_pointer =
882	intel_emit_reloc(wm_state_bo, offsetof(struct brw_wm_unit_state, wm4),
883			 sampler_bo,
884			 wm_state->wm4.stats_enable +
885			 (wm_state->wm4.sampler_count << 2),
886			 I915_GEM_DOMAIN_INSTRUCTION, 0) >> 5;
887    wm_state->wm5.max_threads = PS_MAX_THREADS - 1;
888    wm_state->wm5.transposed_urb_read = 0;
889    wm_state->wm5.thread_dispatch_enable = 1;
890    /* just use 16-pixel dispatch (4 subspans), don't need to change kernel
891     * start point
892     */
893    wm_state->wm5.enable_16_pix = 1;
894    wm_state->wm5.enable_8_pix = 0;
895    wm_state->wm5.early_depth_test = 1;
896
897    /* Each pair of attributes (src/mask coords) is two URB entries */
898    if (has_mask) {
899	wm_state->thread1.binding_table_entry_count = 3; /* 2 tex and fb */
900	wm_state->thread3.urb_entry_read_length = 4;
901    } else {
902	wm_state->thread1.binding_table_entry_count = 2; /* 1 tex and fb */
903	wm_state->thread3.urb_entry_read_length = 2;
904    }
905
906    /* binding table entry count is only used for prefetching, and it has to
907     * be set 0 for IGDNG
908     */
909    if (IS_IGDNG(pI830))
910        wm_state->thread1.binding_table_entry_count = 0;
911
912    drm_intel_bo_unmap(wm_state_bo);
913
914    return wm_state_bo;
915}
916
917static drm_intel_bo *
918gen4_create_cc_viewport(ScrnInfoPtr scrn)
919{
920    I830Ptr pI830 = I830PTR(scrn);
921    drm_intel_bo *bo;
922    struct brw_cc_viewport cc_viewport;
923
924    cc_viewport.min_depth = -1.e35;
925    cc_viewport.max_depth = 1.e35;
926
927    bo = drm_intel_bo_alloc(pI830->bufmgr, "gen4 render unit state",
928			    sizeof(cc_viewport), 4096);
929    drm_intel_bo_subdata(bo, 0, sizeof(cc_viewport), &cc_viewport);
930
931    return bo;
932}
933
934static drm_intel_bo *
935gen4_create_vs_unit_state(ScrnInfoPtr scrn)
936{
937    I830Ptr pI830 = I830PTR(scrn);
938    struct brw_vs_unit_state vs_state;
939    memset(&vs_state, 0, sizeof(vs_state));
940
941    /* Set up the vertex shader to be disabled (passthrough) */
942    if (IS_IGDNG(pI830))
943        vs_state.thread4.nr_urb_entries = URB_VS_ENTRIES >> 2; /* hardware requirement */
944    else
945        vs_state.thread4.nr_urb_entries = URB_VS_ENTRIES;
946    vs_state.thread4.urb_entry_allocation_size = URB_VS_ENTRY_SIZE - 1;
947    vs_state.vs6.vs_enable = 0;
948    vs_state.vs6.vert_cache_disable = 1;
949
950    return intel_bo_alloc_for_data(scrn, &vs_state, sizeof(vs_state),
951				   "gen4 render VS state");
952}
953
954/**
955 * Set up all combinations of cc state: each blendfactor for source and
956 * dest.
957 */
958static drm_intel_bo *
959gen4_create_cc_unit_state(ScrnInfoPtr scrn)
960{
961    I830Ptr pI830 = I830PTR(scrn);
962    struct gen4_cc_unit_state *cc_state;
963    drm_intel_bo *cc_state_bo, *cc_vp_bo;
964    int i, j;
965
966    cc_vp_bo = gen4_create_cc_viewport(scrn);
967
968    cc_state_bo = drm_intel_bo_alloc(pI830->bufmgr, "gen4 CC state",
969				     sizeof(*cc_state), 4096);
970    drm_intel_bo_map(cc_state_bo, TRUE);
971    cc_state = cc_state_bo->virtual;
972    for (i = 0; i < BRW_BLENDFACTOR_COUNT; i++) {
973	for (j = 0; j < BRW_BLENDFACTOR_COUNT; j++) {
974	    cc_state_init(cc_state_bo,
975			  offsetof(struct gen4_cc_unit_state,
976				   cc_state[i][j].state),
977			  i, j, cc_vp_bo);
978	}
979    }
980    drm_intel_bo_unmap(cc_state_bo);
981
982    drm_intel_bo_unreference(cc_vp_bo);
983
984    return cc_state_bo;
985}
986
987static uint32_t
988i965_get_card_format(PicturePtr pPict)
989{
990    int i;
991
992    for (i = 0; i < sizeof(i965_tex_formats) / sizeof(i965_tex_formats[0]);
993	 i++)
994    {
995	if (i965_tex_formats[i].fmt == pPict->format)
996	    break;
997    }
998    assert(i != sizeof(i965_tex_formats) / sizeof(i965_tex_formats[0]));
999
1000    return i965_tex_formats[i].card_fmt;
1001}
1002
1003static sampler_state_filter_t
1004sampler_state_filter_from_picture (int filter)
1005{
1006    switch (filter) {
1007    case PictFilterNearest:
1008	return SAMPLER_STATE_FILTER_NEAREST;
1009    case PictFilterBilinear:
1010	return SAMPLER_STATE_FILTER_BILINEAR;
1011    default:
1012	return -1;
1013    }
1014}
1015
1016static sampler_state_extend_t
1017sampler_state_extend_from_picture (int repeat_type)
1018{
1019    switch (repeat_type) {
1020    case RepeatNone:
1021	return SAMPLER_STATE_EXTEND_NONE;
1022    case RepeatNormal:
1023	return SAMPLER_STATE_EXTEND_REPEAT;
1024    case RepeatPad:
1025	return SAMPLER_STATE_EXTEND_PAD;
1026    case RepeatReflect:
1027	return SAMPLER_STATE_EXTEND_REFLECT;
1028    default:
1029	return -1;
1030    }
1031}
1032
1033/**
1034 * Sets up the common fields for a surface state buffer for the given
1035 * picture in the given surface state buffer.
1036 */
1037static void
1038i965_set_picture_surface_state(dri_bo *ss_bo, int ss_index,
1039			       PicturePtr pPicture, PixmapPtr pPixmap,
1040			       Bool is_dst)
1041{
1042    struct brw_surface_state_padded *ss;
1043    struct brw_surface_state local_ss;
1044    dri_bo *pixmap_bo = i830_get_pixmap_bo(pPixmap);
1045
1046    ss = (struct brw_surface_state_padded *)ss_bo->virtual + ss_index;
1047
1048    /* Since ss is a pointer to WC memory, do all of our bit operations
1049     * into a local temporary first.
1050     */
1051    memset(&local_ss, 0, sizeof(local_ss));
1052    local_ss.ss0.surface_type = BRW_SURFACE_2D;
1053    if (is_dst) {
1054	uint32_t dst_format = 0;
1055	Bool ret = TRUE;
1056
1057	ret = i965_get_dest_format(pPicture, &dst_format);
1058	assert(ret == TRUE);
1059	local_ss.ss0.surface_format = dst_format;
1060    } else {
1061	local_ss.ss0.surface_format = i965_get_card_format(pPicture);
1062    }
1063
1064    local_ss.ss0.data_return_format = BRW_SURFACERETURNFORMAT_FLOAT32;
1065    local_ss.ss0.writedisable_alpha = 0;
1066    local_ss.ss0.writedisable_red = 0;
1067    local_ss.ss0.writedisable_green = 0;
1068    local_ss.ss0.writedisable_blue = 0;
1069    local_ss.ss0.color_blend = 1;
1070    local_ss.ss0.vert_line_stride = 0;
1071    local_ss.ss0.vert_line_stride_ofs = 0;
1072    local_ss.ss0.mipmap_layout_mode = 0;
1073    local_ss.ss0.render_cache_read_mode = 0;
1074    if (pixmap_bo != NULL)
1075	local_ss.ss1.base_addr = pixmap_bo->offset;
1076    else
1077	local_ss.ss1.base_addr = intel_get_pixmap_offset(pPixmap);
1078
1079    local_ss.ss2.mip_count = 0;
1080    local_ss.ss2.render_target_rotation = 0;
1081    local_ss.ss2.height = pPixmap->drawable.height - 1;
1082    local_ss.ss2.width = pPixmap->drawable.width - 1;
1083    local_ss.ss3.pitch = intel_get_pixmap_pitch(pPixmap) - 1;
1084    local_ss.ss3.tile_walk = 0; /* Tiled X */
1085    local_ss.ss3.tiled_surface = i830_pixmap_tiled(pPixmap) ? 1 : 0;
1086
1087    memcpy(ss, &local_ss, sizeof(local_ss));
1088
1089    if (pixmap_bo != NULL) {
1090	uint32_t write_domain, read_domains;
1091
1092	if (is_dst) {
1093	    write_domain = I915_GEM_DOMAIN_RENDER;
1094	    read_domains = I915_GEM_DOMAIN_RENDER;
1095	} else {
1096	    write_domain = 0;
1097	    read_domains = I915_GEM_DOMAIN_SAMPLER;
1098	}
1099	dri_bo_emit_reloc(ss_bo, read_domains, write_domain,
1100			  0,
1101			  ss_index * sizeof(*ss) +
1102			  offsetof(struct brw_surface_state, ss1),
1103			  pixmap_bo);
1104    }
1105}
1106
1107static void
1108i965_emit_composite_state(ScrnInfoPtr pScrn)
1109{
1110    I830Ptr pI830 = I830PTR(pScrn);
1111    struct gen4_render_state *render_state= pI830->gen4_render_state;
1112    gen4_composite_op *composite_op = &render_state->composite_op;
1113    int op = composite_op->op;
1114    PicturePtr pMaskPicture = composite_op->mask_picture;
1115    PicturePtr pDstPicture = composite_op->dest_picture;
1116    PixmapPtr pMask = composite_op->mask;
1117    PixmapPtr pDst = composite_op->dest;
1118    sampler_state_filter_t src_filter = composite_op->src_filter;
1119    sampler_state_filter_t mask_filter = composite_op->mask_filter;
1120    sampler_state_extend_t src_extend = composite_op->src_extend;
1121    sampler_state_extend_t mask_extend = composite_op->mask_extend;
1122    Bool is_affine = composite_op->is_affine;
1123    int urb_vs_start, urb_vs_size;
1124    int urb_gs_start, urb_gs_size;
1125    int urb_clip_start, urb_clip_size;
1126    int urb_sf_start, urb_sf_size;
1127    int urb_cs_start, urb_cs_size;
1128    uint32_t src_blend, dst_blend;
1129    dri_bo *binding_table_bo = composite_op->binding_table_bo;
1130
1131    render_state->needs_state_emit = FALSE;
1132
1133    IntelEmitInvarientState(pScrn);
1134    pI830->last_3d = LAST_3D_RENDER;
1135
1136    urb_vs_start = 0;
1137    urb_vs_size = URB_VS_ENTRIES * URB_VS_ENTRY_SIZE;
1138    urb_gs_start = urb_vs_start + urb_vs_size;
1139    urb_gs_size = URB_GS_ENTRIES * URB_GS_ENTRY_SIZE;
1140    urb_clip_start = urb_gs_start + urb_gs_size;
1141    urb_clip_size = URB_CLIP_ENTRIES * URB_CLIP_ENTRY_SIZE;
1142    urb_sf_start = urb_clip_start + urb_clip_size;
1143    urb_sf_size = URB_SF_ENTRIES * URB_SF_ENTRY_SIZE;
1144    urb_cs_start = urb_sf_start + urb_sf_size;
1145    urb_cs_size = URB_CS_ENTRIES * URB_CS_ENTRY_SIZE;
1146
1147    i965_get_blend_cntl(op, pMaskPicture, pDstPicture->format,
1148			&src_blend, &dst_blend);
1149
1150    /* Begin the long sequence of commands needed to set up the 3D
1151     * rendering pipe
1152     */
1153    {
1154	BEGIN_BATCH(2);
1155	OUT_BATCH(MI_FLUSH |
1156		  MI_STATE_INSTRUCTION_CACHE_FLUSH |
1157		  BRW_MI_GLOBAL_SNAPSHOT_RESET);
1158	OUT_BATCH(MI_NOOP);
1159	ADVANCE_BATCH();
1160    }
1161    {
1162        if (IS_IGDNG(pI830))
1163            BEGIN_BATCH(14);
1164        else
1165            BEGIN_BATCH(12);
1166
1167        /* Match Mesa driver setup */
1168	if (IS_G4X(pI830) || IS_IGDNG(pI830))
1169	    OUT_BATCH(NEW_PIPELINE_SELECT | PIPELINE_SELECT_3D);
1170	else
1171	    OUT_BATCH(BRW_PIPELINE_SELECT | PIPELINE_SELECT_3D);
1172
1173	OUT_BATCH(BRW_CS_URB_STATE | 0);
1174	OUT_BATCH((0 << 4) |  /* URB Entry Allocation Size */
1175		  (0 << 0));  /* Number of URB Entries */
1176
1177	/* Zero out the two base address registers so all offsets are
1178	 * absolute.
1179	 */
1180        if (IS_IGDNG(pI830)) {
1181            OUT_BATCH(BRW_STATE_BASE_ADDRESS | 6);
1182            OUT_BATCH(0 | BASE_ADDRESS_MODIFY);  /* Generate state base address */
1183            OUT_BATCH(0 | BASE_ADDRESS_MODIFY);  /* Surface state base address */
1184            OUT_BATCH(0 | BASE_ADDRESS_MODIFY);  /* media base addr, don't care */
1185            OUT_BATCH(0 | BASE_ADDRESS_MODIFY);  /* Instruction base address */
1186            /* general state max addr, disabled */
1187            OUT_BATCH(0x10000000 | BASE_ADDRESS_MODIFY);
1188            /* media object state max addr, disabled */
1189            OUT_BATCH(0x10000000 | BASE_ADDRESS_MODIFY);
1190            /* Instruction max addr, disabled */
1191            OUT_BATCH(0x10000000 | BASE_ADDRESS_MODIFY);
1192        } else {
1193            OUT_BATCH(BRW_STATE_BASE_ADDRESS | 4);
1194            OUT_BATCH(0 | BASE_ADDRESS_MODIFY);  /* Generate state base address */
1195            OUT_BATCH(0 | BASE_ADDRESS_MODIFY);  /* Surface state base address */
1196            OUT_BATCH(0 | BASE_ADDRESS_MODIFY);  /* media base addr, don't care */
1197            /* general state max addr, disabled */
1198            OUT_BATCH(0x10000000 | BASE_ADDRESS_MODIFY);
1199            /* media object state max addr, disabled */
1200            OUT_BATCH(0x10000000 | BASE_ADDRESS_MODIFY);
1201        }
1202	/* Set system instruction pointer */
1203	OUT_BATCH(BRW_STATE_SIP | 0);
1204	OUT_RELOC(render_state->sip_kernel_bo,
1205		  I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
1206	OUT_BATCH(MI_NOOP);
1207	ADVANCE_BATCH();
1208    }
1209    {
1210	int pipe_ctrl;
1211	BEGIN_BATCH(26);
1212	/* Pipe control */
1213
1214	if (IS_IGDNG(pI830))
1215            pipe_ctrl = BRW_PIPE_CONTROL_NOWRITE;
1216	else
1217            pipe_ctrl = BRW_PIPE_CONTROL_NOWRITE | BRW_PIPE_CONTROL_IS_FLUSH;
1218
1219	OUT_BATCH(BRW_PIPE_CONTROL | pipe_ctrl | 2);
1220	OUT_BATCH(0);			       /* Destination address */
1221	OUT_BATCH(0);			       /* Immediate data low DW */
1222	OUT_BATCH(0);			       /* Immediate data high DW */
1223
1224	/* Binding table pointers */
1225	OUT_BATCH(BRW_3DSTATE_BINDING_TABLE_POINTERS | 4);
1226	OUT_BATCH(0); /* vs */
1227	OUT_BATCH(0); /* gs */
1228	OUT_BATCH(0); /* clip */
1229	OUT_BATCH(0); /* sf */
1230	/* Only the PS uses the binding table */
1231	OUT_RELOC(binding_table_bo, I915_GEM_DOMAIN_SAMPLER, 0, 0);
1232
1233	/* The drawing rectangle clipping is always on.  Set it to values that
1234	 * shouldn't do any clipping.
1235	 */
1236	OUT_BATCH(BRW_3DSTATE_DRAWING_RECTANGLE | 2);
1237	OUT_BATCH(0x00000000);	/* ymin, xmin */
1238	OUT_BATCH(DRAW_YMAX(pDst->drawable.height - 1) |
1239		  DRAW_XMAX(pDst->drawable.width - 1)); /* ymax, xmax */
1240	OUT_BATCH(0x00000000);	/* yorigin, xorigin */
1241
1242	/* skip the depth buffer */
1243	/* skip the polygon stipple */
1244	/* skip the polygon stipple offset */
1245	/* skip the line stipple */
1246
1247	/* Set the pointers to the 3d pipeline state */
1248	OUT_BATCH(BRW_3DSTATE_PIPELINED_POINTERS | 5);
1249	OUT_RELOC(render_state->vs_state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
1250	OUT_BATCH(BRW_GS_DISABLE);   /* disable GS, resulting in passthrough */
1251	OUT_BATCH(BRW_CLIP_DISABLE); /* disable CLIP, resulting in passthrough */
1252	if (pMask) {
1253	    OUT_RELOC(render_state->sf_mask_state_bo,
1254		      I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
1255	} else {
1256	    OUT_RELOC(render_state->sf_state_bo,
1257		      I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
1258	}
1259
1260	OUT_RELOC(render_state->wm_state_bo[composite_op->wm_kernel]
1261		  [src_filter][src_extend]
1262		  [mask_filter][mask_extend],
1263		  I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
1264
1265	OUT_RELOC(render_state->cc_state_bo,
1266		  I915_GEM_DOMAIN_INSTRUCTION, 0,
1267		  offsetof(struct gen4_cc_unit_state,
1268			   cc_state[src_blend][dst_blend]));
1269
1270	/* URB fence */
1271	OUT_BATCH(BRW_URB_FENCE |
1272		  UF0_CS_REALLOC |
1273		  UF0_SF_REALLOC |
1274		  UF0_CLIP_REALLOC |
1275		  UF0_GS_REALLOC |
1276		  UF0_VS_REALLOC |
1277		  1);
1278	OUT_BATCH(((urb_clip_start + urb_clip_size) << UF1_CLIP_FENCE_SHIFT) |
1279		  ((urb_gs_start + urb_gs_size) << UF1_GS_FENCE_SHIFT) |
1280		  ((urb_vs_start + urb_vs_size) << UF1_VS_FENCE_SHIFT));
1281	OUT_BATCH(((urb_cs_start + urb_cs_size) << UF2_CS_FENCE_SHIFT) |
1282		  ((urb_sf_start + urb_sf_size) << UF2_SF_FENCE_SHIFT));
1283
1284	/* Constant buffer state */
1285	OUT_BATCH(BRW_CS_URB_STATE | 0);
1286	OUT_BATCH(((URB_CS_ENTRY_SIZE - 1) << 4) |
1287		  (URB_CS_ENTRIES << 0));
1288	ADVANCE_BATCH();
1289    }
1290    {
1291	/*
1292	 * number of extra parameters per vertex
1293	 */
1294        int nelem = pMask ? 2: 1;
1295	/*
1296	 * size of extra parameters:
1297	 *  3 for homogenous (xyzw)
1298	 *  2 for cartesian (xy)
1299	 */
1300	int selem = is_affine ? 2 : 3;
1301	uint32_t    w_component;
1302	uint32_t    src_format;
1303
1304	render_state->vertex_size = 4 * (2 + nelem * selem);
1305
1306	if (is_affine)
1307	{
1308	    src_format = BRW_SURFACEFORMAT_R32G32_FLOAT;
1309	    w_component = BRW_VFCOMPONENT_STORE_1_FLT;
1310	}
1311	else
1312	{
1313	    src_format = BRW_SURFACEFORMAT_R32G32B32_FLOAT;
1314	    w_component = BRW_VFCOMPONENT_STORE_SRC;
1315	}
1316
1317        if (IS_IGDNG(pI830)) {
1318            BEGIN_BATCH(pMask?9:7);
1319	    /*
1320	     * The reason to add this extra vertex element in the header is that
1321	     * IGDNG has different vertex header definition and origin method to
1322	     * set destination element offset doesn't exist anymore, which means
1323	     * hardware requires a predefined vertex element layout.
1324	     *
1325	     * haihao proposed this approach to fill the first vertex element, so
1326	     * origin layout for Gen4 doesn't need to change, and origin shader
1327	     * programs behavior is also kept.
1328	     *
1329	     * I think this is not bad. - zhenyu
1330	     */
1331
1332	    OUT_BATCH(BRW_3DSTATE_VERTEX_ELEMENTS | ((2 * (2 + nelem)) - 1));
1333	    OUT_BATCH((0 << VE0_VERTEX_BUFFER_INDEX_SHIFT) |
1334                      VE0_VALID |
1335                      (BRW_SURFACEFORMAT_R32G32_FLOAT << VE0_FORMAT_SHIFT) |
1336                      (0 << VE0_OFFSET_SHIFT));
1337
1338	    OUT_BATCH((BRW_VFCOMPONENT_STORE_0 << VE1_VFCOMPONENT_0_SHIFT) |
1339                      (BRW_VFCOMPONENT_STORE_0 << VE1_VFCOMPONENT_1_SHIFT) |
1340                      (BRW_VFCOMPONENT_STORE_0 << VE1_VFCOMPONENT_2_SHIFT) |
1341                      (BRW_VFCOMPONENT_STORE_0 << VE1_VFCOMPONENT_3_SHIFT));
1342        } else {
1343            BEGIN_BATCH(pMask?7:5);
1344            /* Set up our vertex elements, sourced from the single vertex buffer.
1345             * that will be set up later.
1346             */
1347            OUT_BATCH(BRW_3DSTATE_VERTEX_ELEMENTS | ((2 * (1 + nelem)) - 1));
1348        }
1349
1350	/* x,y */
1351	OUT_BATCH((0 << VE0_VERTEX_BUFFER_INDEX_SHIFT) |
1352		  VE0_VALID |
1353		  (BRW_SURFACEFORMAT_R32G32_FLOAT << VE0_FORMAT_SHIFT) |
1354		  (0                            << VE0_OFFSET_SHIFT));
1355
1356        if (IS_IGDNG(pI830))
1357            OUT_BATCH((BRW_VFCOMPONENT_STORE_SRC	<< VE1_VFCOMPONENT_0_SHIFT) |
1358                      (BRW_VFCOMPONENT_STORE_SRC	<< VE1_VFCOMPONENT_1_SHIFT) |
1359                      (BRW_VFCOMPONENT_STORE_1_FLT	<< VE1_VFCOMPONENT_2_SHIFT) |
1360                      (BRW_VFCOMPONENT_STORE_1_FLT	<< VE1_VFCOMPONENT_3_SHIFT));
1361        else
1362            OUT_BATCH((BRW_VFCOMPONENT_STORE_SRC	<< VE1_VFCOMPONENT_0_SHIFT) |
1363                      (BRW_VFCOMPONENT_STORE_SRC	<< VE1_VFCOMPONENT_1_SHIFT) |
1364                      (BRW_VFCOMPONENT_STORE_1_FLT	<< VE1_VFCOMPONENT_2_SHIFT) |
1365                      (BRW_VFCOMPONENT_STORE_1_FLT	<< VE1_VFCOMPONENT_3_SHIFT) |
1366                      (4				<< VE1_DESTINATION_ELEMENT_OFFSET_SHIFT));
1367	/* u0, v0, w0 */
1368	OUT_BATCH((0				<< VE0_VERTEX_BUFFER_INDEX_SHIFT) |
1369		  VE0_VALID					     |
1370		  (src_format			<< VE0_FORMAT_SHIFT) |
1371		  ((2 * 4)                      << VE0_OFFSET_SHIFT)); /* offset vb in bytes */
1372
1373        if (IS_IGDNG(pI830))
1374            OUT_BATCH((BRW_VFCOMPONENT_STORE_SRC	<< VE1_VFCOMPONENT_0_SHIFT) |
1375                      (BRW_VFCOMPONENT_STORE_SRC	<< VE1_VFCOMPONENT_1_SHIFT) |
1376                      (w_component			<< VE1_VFCOMPONENT_2_SHIFT) |
1377                      (BRW_VFCOMPONENT_STORE_1_FLT	<< VE1_VFCOMPONENT_3_SHIFT));
1378        else
1379            OUT_BATCH((BRW_VFCOMPONENT_STORE_SRC	<< VE1_VFCOMPONENT_0_SHIFT) |
1380                      (BRW_VFCOMPONENT_STORE_SRC	<< VE1_VFCOMPONENT_1_SHIFT) |
1381                      (w_component			<< VE1_VFCOMPONENT_2_SHIFT) |
1382                      (BRW_VFCOMPONENT_STORE_1_FLT	<< VE1_VFCOMPONENT_3_SHIFT) |
1383                      ((4 + 4)			<< VE1_DESTINATION_ELEMENT_OFFSET_SHIFT)); /* VUE offset in dwords */
1384	/* u1, v1, w1 */
1385   	if (pMask) {
1386	    OUT_BATCH((0			    << VE0_VERTEX_BUFFER_INDEX_SHIFT) |
1387		      VE0_VALID							    |
1388		      (src_format		    << VE0_FORMAT_SHIFT) |
1389		      (((2 + selem) * 4)            << VE0_OFFSET_SHIFT));  /* vb offset in bytes */
1390
1391            if (IS_IGDNG(pI830))
1392                OUT_BATCH((BRW_VFCOMPONENT_STORE_SRC    << VE1_VFCOMPONENT_0_SHIFT) |
1393                          (BRW_VFCOMPONENT_STORE_SRC    << VE1_VFCOMPONENT_1_SHIFT) |
1394                          (w_component		    << VE1_VFCOMPONENT_2_SHIFT) |
1395                          (BRW_VFCOMPONENT_STORE_1_FLT  << VE1_VFCOMPONENT_3_SHIFT));
1396            else
1397                OUT_BATCH((BRW_VFCOMPONENT_STORE_SRC    << VE1_VFCOMPONENT_0_SHIFT) |
1398                          (BRW_VFCOMPONENT_STORE_SRC    << VE1_VFCOMPONENT_1_SHIFT) |
1399                          (w_component		    << VE1_VFCOMPONENT_2_SHIFT) |
1400                          (BRW_VFCOMPONENT_STORE_1_FLT  << VE1_VFCOMPONENT_3_SHIFT) |
1401                          ((4 + 4 + 4)		    << VE1_DESTINATION_ELEMENT_OFFSET_SHIFT)); /* VUE offset in dwords */
1402   	}
1403
1404	ADVANCE_BATCH();
1405    }
1406}
1407
1408/**
1409 * Returns whether the current set of composite state plus vertex buffer is
1410 * expected to fit in the aperture.
1411 */
1412static Bool
1413i965_composite_check_aperture(ScrnInfoPtr pScrn)
1414{
1415    I830Ptr pI830 = I830PTR(pScrn);
1416    struct gen4_render_state *render_state= pI830->gen4_render_state;
1417    gen4_composite_op *composite_op = &render_state->composite_op;
1418    drm_intel_bo *bo_table[] = {
1419	pI830->batch_bo,
1420	composite_op->binding_table_bo,
1421	render_state->vertex_buffer_bo,
1422	render_state->vs_state_bo,
1423	render_state->sf_state_bo,
1424	render_state->sf_mask_state_bo,
1425	render_state->wm_state_bo[composite_op->wm_kernel]
1426				 [composite_op->src_filter]
1427				 [composite_op->src_extend]
1428				 [composite_op->mask_filter]
1429				 [composite_op->mask_extend],
1430	render_state->cc_state_bo,
1431	render_state->sip_kernel_bo,
1432    };
1433
1434    return drm_intel_bufmgr_check_aperture_space(bo_table,
1435						 ARRAY_SIZE(bo_table)) == 0;
1436}
1437
1438Bool
1439i965_prepare_composite(int op, PicturePtr pSrcPicture,
1440		       PicturePtr pMaskPicture, PicturePtr pDstPicture,
1441		       PixmapPtr pSrc, PixmapPtr pMask, PixmapPtr pDst)
1442{
1443    ScrnInfoPtr pScrn = xf86Screens[pDstPicture->pDrawable->pScreen->myNum];
1444    I830Ptr pI830 = I830PTR(pScrn);
1445    struct gen4_render_state *render_state= pI830->gen4_render_state;
1446    gen4_composite_op *composite_op = &render_state->composite_op;
1447    uint32_t *binding_table;
1448    drm_intel_bo *binding_table_bo, *surface_state_bo;
1449
1450    if (composite_op->src_filter < 0)
1451	I830FALLBACK("Bad src filter 0x%x\n", pSrcPicture->filter);
1452    composite_op->src_extend =
1453	sampler_state_extend_from_picture(pSrcPicture->repeatType);
1454    if (composite_op->src_extend < 0)
1455	I830FALLBACK("Bad src repeat 0x%x\n", pSrcPicture->repeatType);
1456
1457    if (pMaskPicture) {
1458	composite_op->mask_filter =
1459	    sampler_state_filter_from_picture(pMaskPicture->filter);
1460	if (composite_op->mask_filter < 0)
1461	    I830FALLBACK("Bad mask filter 0x%x\n", pMaskPicture->filter);
1462	composite_op->mask_extend =
1463	    sampler_state_extend_from_picture(pMaskPicture->repeatType);
1464	if (composite_op->mask_extend < 0)
1465	    I830FALLBACK("Bad mask repeat 0x%x\n", pMaskPicture->repeatType);
1466    } else {
1467	composite_op->mask_filter = SAMPLER_STATE_FILTER_NEAREST;
1468	composite_op->mask_extend = SAMPLER_STATE_EXTEND_NONE;
1469    }
1470
1471    /* Set up the surface states. */
1472    surface_state_bo = dri_bo_alloc(pI830->bufmgr, "surface_state",
1473				    3 * sizeof (brw_surface_state_padded),
1474				    4096);
1475    if (dri_bo_map(surface_state_bo, 1) != 0) {
1476	dri_bo_unreference(surface_state_bo);
1477	return FALSE;
1478    }
1479    /* Set up the state buffer for the destination surface */
1480    i965_set_picture_surface_state(surface_state_bo, 0,
1481				   pDstPicture, pDst, TRUE);
1482    /* Set up the source surface state buffer */
1483    i965_set_picture_surface_state(surface_state_bo, 1,
1484				   pSrcPicture, pSrc, FALSE);
1485    if (pMask) {
1486	/* Set up the mask surface state buffer */
1487	i965_set_picture_surface_state(surface_state_bo, 2,
1488				       pMaskPicture, pMask,
1489				       FALSE);
1490    }
1491    dri_bo_unmap(surface_state_bo);
1492
1493    /* Set up the binding table of surface indices to surface state. */
1494    binding_table_bo = dri_bo_alloc(pI830->bufmgr, "binding_table",
1495				    3 * sizeof(uint32_t), 4096);
1496    if (dri_bo_map (binding_table_bo, 1) != 0) {
1497	dri_bo_unreference(binding_table_bo);
1498	dri_bo_unreference(surface_state_bo);
1499	return FALSE;
1500    }
1501
1502    binding_table = binding_table_bo->virtual;
1503    binding_table[0] = intel_emit_reloc(binding_table_bo,
1504					0 * sizeof(uint32_t),
1505					surface_state_bo,
1506					0 * sizeof(brw_surface_state_padded),
1507					I915_GEM_DOMAIN_INSTRUCTION, 0);
1508
1509    binding_table[1] = intel_emit_reloc(binding_table_bo,
1510					1 * sizeof(uint32_t),
1511					surface_state_bo,
1512					1 * sizeof(brw_surface_state_padded),
1513					I915_GEM_DOMAIN_INSTRUCTION, 0);
1514
1515    if (pMask) {
1516	binding_table[2] = intel_emit_reloc(binding_table_bo,
1517					    2 * sizeof(uint32_t),
1518					    surface_state_bo,
1519					    2 * sizeof(brw_surface_state_padded),
1520					    I915_GEM_DOMAIN_INSTRUCTION, 0);
1521    } else {
1522	binding_table[2] = 0;
1523    }
1524    dri_bo_unmap(binding_table_bo);
1525    /* All refs to surface_state are now contained in binding_table_bo. */
1526    drm_intel_bo_unreference(surface_state_bo);
1527
1528    composite_op->op = op;
1529    composite_op->source_picture = pSrcPicture;
1530    composite_op->mask_picture = pMaskPicture;
1531    composite_op->dest_picture = pDstPicture;
1532    composite_op->source = pSrc;
1533    composite_op->mask = pMask;
1534    composite_op->dest = pDst;
1535    drm_intel_bo_unreference(composite_op->binding_table_bo);
1536    composite_op->binding_table_bo = binding_table_bo;
1537    composite_op->src_filter =
1538	sampler_state_filter_from_picture(pSrcPicture->filter);
1539
1540    pI830->scale_units[0][0] = pSrc->drawable.width;
1541    pI830->scale_units[0][1] = pSrc->drawable.height;
1542
1543    pI830->transform[0] = pSrcPicture->transform;
1544    composite_op->is_affine =
1545	i830_transform_is_affine(pI830->transform[0]);
1546
1547    if (!pMask) {
1548	pI830->transform[1] = NULL;
1549	pI830->scale_units[1][0] = -1;
1550	pI830->scale_units[1][1] = -1;
1551    } else {
1552	pI830->transform[1] = pMaskPicture->transform;
1553	pI830->scale_units[1][0] = pMask->drawable.width;
1554	pI830->scale_units[1][1] = pMask->drawable.height;
1555	composite_op->is_affine |=
1556	    i830_transform_is_affine(pI830->transform[1]);
1557    }
1558
1559
1560    if (pMask) {
1561	if (pMaskPicture->componentAlpha &&
1562	    PICT_FORMAT_RGB(pMaskPicture->format))
1563	{
1564	    if (i965_blend_op[op].src_alpha) {
1565		if (composite_op->is_affine)
1566		    composite_op->wm_kernel = WM_KERNEL_MASKCA_SRCALPHA_AFFINE;
1567		else
1568		    composite_op->wm_kernel = WM_KERNEL_MASKCA_SRCALPHA_PROJECTIVE;
1569	    } else {
1570		if (composite_op->is_affine)
1571		    composite_op->wm_kernel = WM_KERNEL_MASKCA_AFFINE;
1572		else
1573		    composite_op->wm_kernel = WM_KERNEL_MASKCA_PROJECTIVE;
1574	    }
1575	} else {
1576	    if (composite_op->is_affine)
1577		composite_op->wm_kernel = WM_KERNEL_MASKNOCA_AFFINE;
1578	    else
1579		composite_op->wm_kernel = WM_KERNEL_MASKNOCA_PROJECTIVE;
1580	}
1581    } else {
1582	if (composite_op->is_affine)
1583	    composite_op->wm_kernel = WM_KERNEL_NOMASK_AFFINE;
1584	else
1585	    composite_op->wm_kernel = WM_KERNEL_NOMASK_PROJECTIVE;
1586    }
1587
1588    if (!i965_composite_check_aperture(pScrn)) {
1589	intel_batch_flush(pScrn, FALSE);
1590	if (!i965_composite_check_aperture(pScrn))
1591	    I830FALLBACK("Couldn't fit render operation in aperture\n");
1592    }
1593
1594    render_state->needs_state_emit = TRUE;
1595
1596    return TRUE;
1597}
1598
1599static drm_intel_bo *
1600i965_get_vb_space(ScrnInfoPtr pScrn)
1601{
1602    I830Ptr pI830 = I830PTR(pScrn);
1603    struct gen4_render_state *render_state = pI830->gen4_render_state;
1604
1605    /* If the vertex buffer is too full, then we free the old and a new one
1606     * gets made.
1607     */
1608    if (render_state->vb_offset + VERTEX_FLOATS_PER_COMPOSITE >
1609	VERTEX_BUFFER_SIZE) {
1610	drm_intel_bo_unreference(render_state->vertex_buffer_bo);
1611	render_state->vertex_buffer_bo = NULL;
1612    }
1613
1614    /* Alloc a new vertex buffer if necessary. */
1615    if (render_state->vertex_buffer_bo == NULL) {
1616	render_state->vertex_buffer_bo = drm_intel_bo_alloc(pI830->bufmgr, "vb",
1617							    sizeof(gen4_vertex_buffer),
1618							    4096);
1619	render_state->vb_offset = 0;
1620    }
1621
1622    drm_intel_bo_reference(render_state->vertex_buffer_bo);
1623    return render_state->vertex_buffer_bo;
1624}
1625
1626void
1627i965_composite(PixmapPtr pDst, int srcX, int srcY, int maskX, int maskY,
1628	       int dstX, int dstY, int w, int h)
1629{
1630    ScrnInfoPtr pScrn = xf86Screens[pDst->drawable.pScreen->myNum];
1631    I830Ptr pI830 = I830PTR(pScrn);
1632    struct gen4_render_state *render_state = pI830->gen4_render_state;
1633    Bool has_mask;
1634    float src_x[3], src_y[3], src_w[3], mask_x[3], mask_y[3], mask_w[3];
1635    int i;
1636    drm_intel_bo *vb_bo;
1637    float vb[18];
1638    Bool is_affine = render_state->composite_op.is_affine;
1639
1640    if (is_affine)
1641    {
1642	if (!i830_get_transformed_coordinates(srcX, srcY,
1643					      pI830->transform[0],
1644					      &src_x[0], &src_y[0]))
1645	    return;
1646	if (!i830_get_transformed_coordinates(srcX, srcY + h,
1647					      pI830->transform[0],
1648					      &src_x[1], &src_y[1]))
1649	    return;
1650	if (!i830_get_transformed_coordinates(srcX + w, srcY + h,
1651					      pI830->transform[0],
1652					      &src_x[2], &src_y[2]))
1653	    return;
1654    }
1655    else
1656    {
1657	if (!i830_get_transformed_coordinates_3d(srcX, srcY,
1658						 pI830->transform[0],
1659						 &src_x[0], &src_y[0],
1660						 &src_w[0]))
1661	    return;
1662	if (!i830_get_transformed_coordinates_3d(srcX, srcY + h,
1663						 pI830->transform[0],
1664						 &src_x[1], &src_y[1],
1665						 &src_w[1]))
1666	    return;
1667	if (!i830_get_transformed_coordinates_3d(srcX + w, srcY + h,
1668						 pI830->transform[0],
1669						 &src_x[2], &src_y[2],
1670						 &src_w[2]))
1671	    return;
1672    }
1673
1674    if (pI830->scale_units[1][0] == -1 || pI830->scale_units[1][1] == -1) {
1675	has_mask = FALSE;
1676    } else {
1677	has_mask = TRUE;
1678	if (is_affine) {
1679	    if (!i830_get_transformed_coordinates(maskX, maskY,
1680						  pI830->transform[1],
1681						  &mask_x[0], &mask_y[0]))
1682		return;
1683	    if (!i830_get_transformed_coordinates(maskX, maskY + h,
1684						  pI830->transform[1],
1685						  &mask_x[1], &mask_y[1]))
1686		return;
1687	    if (!i830_get_transformed_coordinates(maskX + w, maskY + h,
1688						  pI830->transform[1],
1689						  &mask_x[2], &mask_y[2]))
1690		return;
1691	} else {
1692	    if (!i830_get_transformed_coordinates_3d(maskX, maskY,
1693						     pI830->transform[1],
1694						     &mask_x[0], &mask_y[0],
1695						     &mask_w[0]))
1696		return;
1697	    if (!i830_get_transformed_coordinates_3d(maskX, maskY + h,
1698						     pI830->transform[1],
1699						     &mask_x[1], &mask_y[1],
1700						     &mask_w[1]))
1701		return;
1702	    if (!i830_get_transformed_coordinates_3d(maskX + w, maskY + h,
1703						     pI830->transform[1],
1704						     &mask_x[2], &mask_y[2],
1705						     &mask_w[2]))
1706		return;
1707	}
1708    }
1709
1710    vb_bo = i965_get_vb_space(pScrn);
1711    if (vb_bo == NULL)
1712	return;
1713    i = 0;
1714    /* rect (x2,y2) */
1715    vb[i++] = (float)(dstX + w);
1716    vb[i++] = (float)(dstY + h);
1717    vb[i++] = src_x[2] / pI830->scale_units[0][0];
1718    vb[i++] = src_y[2] / pI830->scale_units[0][1];
1719    if (!is_affine)
1720	vb[i++] = src_w[2];
1721    if (has_mask) {
1722        vb[i++] = mask_x[2] / pI830->scale_units[1][0];
1723        vb[i++] = mask_y[2] / pI830->scale_units[1][1];
1724	if (!is_affine)
1725	    vb[i++] = mask_w[2];
1726    }
1727
1728    /* rect (x1,y2) */
1729    vb[i++] = (float)dstX;
1730    vb[i++] = (float)(dstY + h);
1731    vb[i++] = src_x[1] / pI830->scale_units[0][0];
1732    vb[i++] = src_y[1] / pI830->scale_units[0][1];
1733    if (!is_affine)
1734	vb[i++] = src_w[1];
1735    if (has_mask) {
1736        vb[i++] = mask_x[1] / pI830->scale_units[1][0];
1737        vb[i++] = mask_y[1] / pI830->scale_units[1][1];
1738	if (!is_affine)
1739	    vb[i++] = mask_w[1];
1740    }
1741
1742    /* rect (x1,y1) */
1743    vb[i++] = (float)dstX;
1744    vb[i++] = (float)dstY;
1745    vb[i++] = src_x[0] / pI830->scale_units[0][0];
1746    vb[i++] = src_y[0] / pI830->scale_units[0][1];
1747    if (!is_affine)
1748	vb[i++] = src_w[0];
1749    if (has_mask) {
1750        vb[i++] = mask_x[0] / pI830->scale_units[1][0];
1751        vb[i++] = mask_y[0] / pI830->scale_units[1][1];
1752	if (!is_affine)
1753	    vb[i++] = mask_w[0];
1754    }
1755    assert (i <= VERTEX_BUFFER_SIZE);
1756    drm_intel_bo_subdata(vb_bo, render_state->vb_offset * 4, i * 4, vb);
1757
1758    if (!i965_composite_check_aperture(pScrn))
1759	intel_batch_flush(pScrn, FALSE);
1760
1761    intel_batch_start_atomic(pScrn, 200);
1762    if (render_state->needs_state_emit)
1763	i965_emit_composite_state(pScrn);
1764
1765    BEGIN_BATCH(12);
1766    OUT_BATCH(MI_FLUSH);
1767    /* Set up the pointer to our (single) vertex buffer */
1768    OUT_BATCH(BRW_3DSTATE_VERTEX_BUFFERS | 3);
1769    OUT_BATCH((0 << VB0_BUFFER_INDEX_SHIFT) |
1770	      VB0_VERTEXDATA |
1771	      (render_state->vertex_size << VB0_BUFFER_PITCH_SHIFT));
1772    OUT_RELOC(vb_bo, I915_GEM_DOMAIN_VERTEX, 0, render_state->vb_offset * 4);
1773
1774    if (IS_IGDNG(pI830))
1775        OUT_RELOC(vb_bo, I915_GEM_DOMAIN_VERTEX, 0, render_state->vb_offset * 4 + i * 4);
1776    else
1777        OUT_BATCH(3);
1778
1779    OUT_BATCH(0); // ignore for VERTEXDATA, but still there
1780
1781    OUT_BATCH(BRW_3DPRIMITIVE |
1782	      BRW_3DPRIMITIVE_VERTEX_SEQUENTIAL |
1783	      (_3DPRIM_RECTLIST << BRW_3DPRIMITIVE_TOPOLOGY_SHIFT) |
1784	      (0 << 9) |  /* CTG - indirect vertex count */
1785	      4);
1786    OUT_BATCH(3);  /* vertex count per instance */
1787    OUT_BATCH(0); /* start vertex offset */
1788    OUT_BATCH(1); /* single instance */
1789    OUT_BATCH(0); /* start instance location */
1790    OUT_BATCH(0); /* index buffer offset, ignored */
1791    ADVANCE_BATCH();
1792
1793    render_state->vb_offset += i;
1794    drm_intel_bo_unreference(vb_bo);
1795
1796    intel_batch_end_atomic(pScrn);
1797
1798    i830_debug_sync(pScrn);
1799}
1800
1801void
1802i965_batch_flush_notify(ScrnInfoPtr pScrn)
1803{
1804    I830Ptr pI830 = I830PTR(pScrn);
1805    struct gen4_render_state *render_state = pI830->gen4_render_state;
1806
1807    /* Once a batch is emitted, we never want to map again any buffer
1808     * object being referenced by that batch, (which would be very
1809     * expensive). */
1810    if (render_state->vertex_buffer_bo) {
1811	dri_bo_unreference (render_state->vertex_buffer_bo);
1812	render_state->vertex_buffer_bo = NULL;
1813    }
1814
1815    render_state->needs_state_emit = TRUE;
1816}
1817
1818/**
1819 * Called at EnterVT so we can set up our offsets into the state buffer.
1820 */
1821void
1822gen4_render_state_init(ScrnInfoPtr pScrn)
1823{
1824    I830Ptr pI830 = I830PTR(pScrn);
1825    struct gen4_render_state *render_state;
1826    int i, j, k, l, m;
1827    drm_intel_bo *sf_kernel_bo, *sf_kernel_mask_bo;
1828    drm_intel_bo *border_color_bo;
1829
1830    if (pI830->gen4_render_state == NULL)
1831	pI830->gen4_render_state = calloc(sizeof(*render_state), 1);
1832
1833    render_state = pI830->gen4_render_state;
1834    render_state->vb_offset = 0;
1835
1836    render_state->vs_state_bo = gen4_create_vs_unit_state(pScrn);
1837
1838    /* Set up the two SF states (one for blending with a mask, one without) */
1839    if (IS_IGDNG(pI830)) {
1840	sf_kernel_bo = intel_bo_alloc_for_data(pScrn,
1841					       sf_kernel_static_gen5,
1842					       sizeof(sf_kernel_static_gen5),
1843					       "sf kernel gen5");
1844	sf_kernel_mask_bo = intel_bo_alloc_for_data(pScrn,
1845						    sf_kernel_mask_static_gen5,
1846						    sizeof(sf_kernel_mask_static_gen5),
1847						    "sf mask kernel");
1848    } else {
1849	sf_kernel_bo = intel_bo_alloc_for_data(pScrn,
1850					       sf_kernel_static,
1851					       sizeof(sf_kernel_static),
1852					       "sf kernel");
1853	sf_kernel_mask_bo = intel_bo_alloc_for_data(pScrn,
1854						    sf_kernel_mask_static,
1855						    sizeof(sf_kernel_mask_static),
1856						    "sf mask kernel");
1857    }
1858    render_state->sf_state_bo = gen4_create_sf_state(pScrn, sf_kernel_bo);
1859    render_state->sf_mask_state_bo = gen4_create_sf_state(pScrn,
1860							  sf_kernel_mask_bo);
1861    drm_intel_bo_unreference(sf_kernel_bo);
1862    drm_intel_bo_unreference(sf_kernel_mask_bo);
1863
1864    for (m = 0; m < WM_KERNEL_COUNT; m++) {
1865	if (IS_IGDNG(pI830))
1866	    render_state->wm_kernel_bo[m] =
1867		intel_bo_alloc_for_data(pScrn,
1868				        wm_kernels_gen5[m].data, wm_kernels_gen5[m].size,
1869				        "WM kernel gen5");
1870	else
1871	    render_state->wm_kernel_bo[m] =
1872		intel_bo_alloc_for_data(pScrn,
1873				        wm_kernels[m].data, wm_kernels[m].size,
1874				        "WM kernel");
1875    }
1876
1877    /* Set up the WM states: each filter/extend type for source and mask, per
1878     * kernel.
1879     */
1880    border_color_bo = sampler_border_color_create(pScrn);
1881    for (i = 0; i < SAMPLER_STATE_FILTER_COUNT; i++) {
1882	for (j = 0; j < SAMPLER_STATE_EXTEND_COUNT; j++) {
1883	    for (k = 0; k < SAMPLER_STATE_FILTER_COUNT; k++) {
1884		for (l = 0; l < SAMPLER_STATE_EXTEND_COUNT; l++) {
1885		    drm_intel_bo *sampler_state_bo;
1886
1887		    sampler_state_bo =
1888			gen4_create_sampler_state(pScrn,
1889						  i, j,
1890						  k, l,
1891						  border_color_bo);
1892
1893		    for (m = 0; m < WM_KERNEL_COUNT; m++) {
1894			if (IS_IGDNG(pI830))
1895			    render_state->wm_state_bo[m][i][j][k][l] =
1896				gen4_create_wm_state(pScrn,
1897						     wm_kernels_gen5[m].has_mask,
1898						     render_state->wm_kernel_bo[m],
1899						     sampler_state_bo);
1900			else
1901			    render_state->wm_state_bo[m][i][j][k][l] =
1902				gen4_create_wm_state(pScrn,
1903						     wm_kernels[m].has_mask,
1904						     render_state->wm_kernel_bo[m],
1905						     sampler_state_bo);
1906		    }
1907		    drm_intel_bo_unreference(sampler_state_bo);
1908		}
1909	    }
1910	}
1911    }
1912    drm_intel_bo_unreference(border_color_bo);
1913
1914    render_state->cc_state_bo = gen4_create_cc_unit_state(pScrn);
1915    render_state->sip_kernel_bo = intel_bo_alloc_for_data(pScrn,
1916							  sip_kernel_static,
1917							  sizeof(sip_kernel_static),
1918							  "sip kernel");
1919}
1920
1921/**
1922 * Called at LeaveVT.
1923 */
1924void
1925gen4_render_state_cleanup(ScrnInfoPtr pScrn)
1926{
1927    I830Ptr pI830 = I830PTR(pScrn);
1928    struct gen4_render_state *render_state= pI830->gen4_render_state;
1929    int i, j, k, l, m;
1930    gen4_composite_op *composite_op = &render_state->composite_op;
1931
1932    drm_intel_bo_unreference(composite_op->binding_table_bo);
1933    drm_intel_bo_unreference(render_state->vertex_buffer_bo);
1934
1935    drm_intel_bo_unreference(render_state->vs_state_bo);
1936    drm_intel_bo_unreference(render_state->sf_state_bo);
1937    drm_intel_bo_unreference(render_state->sf_mask_state_bo);
1938
1939    for (i = 0; i < WM_KERNEL_COUNT; i++)
1940	drm_intel_bo_unreference(render_state->wm_kernel_bo[i]);
1941
1942    for (i = 0; i < SAMPLER_STATE_FILTER_COUNT; i++)
1943	for (j = 0; j < SAMPLER_STATE_EXTEND_COUNT; j++)
1944	    for (k = 0; k < SAMPLER_STATE_FILTER_COUNT; k++)
1945		for (l = 0; l < SAMPLER_STATE_EXTEND_COUNT; l++)
1946		    for (m = 0; m < WM_KERNEL_COUNT; m++)
1947			drm_intel_bo_unreference(render_state->wm_state_bo[m][i][j][k][l]);
1948
1949    drm_intel_bo_unreference(render_state->cc_state_bo);
1950    drm_intel_bo_unreference(render_state->sip_kernel_bo);
1951
1952    free(pI830->gen4_render_state);
1953    pI830->gen4_render_state = NULL;
1954}
1955