1 /*
2  * Copyright © 2011 Intel Corporation
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21  * IN THE SOFTWARE.
22  */
23 
24 #include "intel_batchbuffer.h"
25 #include "intel_mipmap_tree.h"
26 #include "intel_regions.h"
27 #include "intel_fbo.h"
28 #include "brw_context.h"
29 #include "brw_state.h"
30 #include "brw_defines.h"
31 
emit_depthbuffer(struct brw_context * brw)32 static void emit_depthbuffer(struct brw_context *brw)
33 {
34    struct intel_context *intel = &brw->intel;
35    struct gl_context *ctx = &intel->ctx;
36    struct gl_framebuffer *fb = ctx->DrawBuffer;
37 
38    /* _NEW_BUFFERS */
39    struct intel_renderbuffer *drb = intel_get_renderbuffer(fb, BUFFER_DEPTH);
40    struct intel_renderbuffer *srb = intel_get_renderbuffer(fb, BUFFER_STENCIL);
41    struct intel_mipmap_tree *depth_mt = NULL,
42 			    *stencil_mt = NULL,
43 			    *hiz_mt = NULL;
44 
45    /* Amount by which drawing should be offset in order to draw to the
46     * appropriate miplevel/zoffset/cubeface.  We will extract these values
47     * from depth_irb or stencil_irb once we determine which is present.
48     */
49    uint32_t draw_x = 0, draw_y = 0;
50 
51    /* Masks used to determine how much of the draw_x and draw_y offsets should
52     * be performed using the fine adjustment of "depth coordinate offset X/Y"
53     * (dw5 of 3DSTATE_DEPTH_BUFFER).  Any remaining coarse adjustment will be
54     * performed by changing the base addresses of the buffers.
55     *
56     * Since the HiZ, depth, and stencil buffers all use the same "depth
57     * coordinate offset X/Y" values, we need to make sure that the coarse
58     * adjustment will be possible to apply to all three buffers.  Since coarse
59     * adjustment can only be applied in multiples of the tile size, we will OR
60     * together the tile masks of all the buffers to determine which offsets to
61     * perform as fine adjustments.
62     */
63    uint32_t tile_mask_x = 0, tile_mask_y = 0;
64 
65    if (drb)
66       depth_mt = drb->mt;
67 
68    if (depth_mt) {
69       hiz_mt = depth_mt->hiz_mt;
70 
71       intel_region_get_tile_masks(depth_mt->region,
72                                   &tile_mask_x, &tile_mask_y, false);
73 
74       if (hiz_mt) {
75          uint32_t hiz_tile_mask_x, hiz_tile_mask_y;
76          intel_region_get_tile_masks(hiz_mt->region,
77                                      &hiz_tile_mask_x, &hiz_tile_mask_y,
78                                      false);
79 
80          /* Each HiZ row represents 2 rows of pixels */
81          hiz_tile_mask_y = hiz_tile_mask_y << 1 | 1;
82 
83          tile_mask_x |= hiz_tile_mask_x;
84          tile_mask_y |= hiz_tile_mask_y;
85       }
86    }
87 
88    if (srb) {
89       stencil_mt = srb->mt;
90       if (stencil_mt->stencil_mt)
91 	 stencil_mt = stencil_mt->stencil_mt;
92 
93       assert(stencil_mt->format == MESA_FORMAT_S8);
94 
95       /* Stencil buffer uses 64x64 tiles. */
96       tile_mask_x |= 63;
97       tile_mask_y |= 63;
98    }
99 
100    /* Gen7 doesn't support packed depth/stencil */
101    assert(stencil_mt == NULL || depth_mt != stencil_mt);
102    assert(!depth_mt || !_mesa_is_format_packed_depth_stencil(depth_mt->format));
103 
104    intel_emit_depth_stall_flushes(intel);
105 
106    if (depth_mt == NULL) {
107       uint32_t dw1 = BRW_DEPTHFORMAT_D32_FLOAT << 18;
108       uint32_t dw3 = 0;
109       uint32_t tile_x = 0, tile_y = 0;
110 
111       if (stencil_mt == NULL) {
112 	 dw1 |= (BRW_SURFACE_NULL << 29);
113       } else {
114 	 /* _NEW_STENCIL: enable stencil buffer writes */
115 	 dw1 |= ((ctx->Stencil.WriteMask != 0) << 27);
116 
117          draw_x = srb->draw_x;
118          draw_y = srb->draw_y;
119          tile_x = draw_x & tile_mask_x;
120          tile_y = draw_y & tile_mask_y;
121 
122          /* According to the Sandy Bridge PRM, volume 2 part 1, pp326-327
123           * (3DSTATE_DEPTH_BUFFER dw5), in the documentation for "Depth
124           * Coordinate Offset X/Y":
125           *
126           *   "The 3 LSBs of both offsets must be zero to ensure correct
127           *   alignment"
128           *
129           * We have no guarantee that tile_x and tile_y are correctly aligned,
130           * since they are determined by the mipmap layout, which is only
131           * aligned to multiples of 4.
132           *
133           * So, to avoid hanging the GPU, just smash the low order 3 bits of
134           * tile_x and tile_y to 0.  This is a temporary workaround until we
135           * come up with a better solution.
136           */
137          tile_x &= ~7;
138          tile_y &= ~7;
139 
140 	 /* 3DSTATE_STENCIL_BUFFER inherits surface type and dimensions. */
141 	 dw1 |= (BRW_SURFACE_2D << 29);
142 	 dw3 = ((srb->Base.Base.Width + tile_x - 1) << 4) |
143 	       ((srb->Base.Base.Height + tile_y - 1) << 18);
144       }
145 
146       BEGIN_BATCH(7);
147       OUT_BATCH(GEN7_3DSTATE_DEPTH_BUFFER << 16 | (7 - 2));
148       OUT_BATCH(dw1);
149       OUT_BATCH(0);
150       OUT_BATCH(dw3);
151       OUT_BATCH(0);
152       OUT_BATCH(tile_x | (tile_y << 16));
153       OUT_BATCH(0);
154       ADVANCE_BATCH();
155    } else {
156       struct intel_region *region = depth_mt->region;
157       uint32_t tile_x, tile_y, offset;
158 
159       draw_x = drb->draw_x;
160       draw_y = drb->draw_y;
161       tile_x = draw_x & tile_mask_x;
162       tile_y = draw_y & tile_mask_y;
163 
164       /* According to the Sandy Bridge PRM, volume 2 part 1, pp326-327
165        * (3DSTATE_DEPTH_BUFFER dw5), in the documentation for "Depth
166        * Coordinate Offset X/Y":
167        *
168        *   "The 3 LSBs of both offsets must be zero to ensure correct
169        *   alignment"
170        *
171        * We have no guarantee that tile_x and tile_y are correctly aligned,
172        * since they are determined by the mipmap layout, which is only aligned
173        * to multiples of 4.
174        *
175        * So, to avoid hanging the GPU, just smash the low order 3 bits of
176        * tile_x and tile_y to 0.  This is a temporary workaround until we come
177        * up with a better solution.
178        */
179       tile_x &= ~7;
180       tile_y &= ~7;
181 
182       offset = intel_region_get_aligned_offset(region,
183                                                draw_x & ~tile_mask_x,
184                                                draw_y & ~tile_mask_y,
185                                                false);
186 
187       assert(region->tiling == I915_TILING_Y);
188 
189       /* _NEW_DEPTH, _NEW_STENCIL */
190       BEGIN_BATCH(7);
191       OUT_BATCH(GEN7_3DSTATE_DEPTH_BUFFER << 16 | (7 - 2));
192       OUT_BATCH(((region->pitch * region->cpp) - 1) |
193 		(brw_depthbuffer_format(brw) << 18) |
194 		((hiz_mt ? 1 : 0) << 22) | /* hiz enable */
195 		((stencil_mt != NULL && ctx->Stencil.WriteMask != 0) << 27) |
196 		((ctx->Depth.Mask != 0) << 28) |
197 		(BRW_SURFACE_2D << 29));
198       OUT_RELOC(region->bo,
199 	        I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
200 		offset);
201       OUT_BATCH((((drb->Base.Base.Width + tile_x) - 1) << 4) |
202                 (((drb->Base.Base.Height + tile_y) - 1) << 18));
203       OUT_BATCH(0);
204       OUT_BATCH(tile_x | (tile_y << 16));
205       OUT_BATCH(0);
206       ADVANCE_BATCH();
207    }
208 
209    if (hiz_mt == NULL) {
210       BEGIN_BATCH(3);
211       OUT_BATCH(GEN7_3DSTATE_HIER_DEPTH_BUFFER << 16 | (3 - 2));
212       OUT_BATCH(0);
213       OUT_BATCH(0);
214       ADVANCE_BATCH();
215    } else {
216       uint32_t hiz_offset =
217          intel_region_get_aligned_offset(hiz_mt->region,
218                                          draw_x & ~tile_mask_x,
219                                          (draw_y & ~tile_mask_y) / 2,
220                                          false);
221       BEGIN_BATCH(3);
222       OUT_BATCH(GEN7_3DSTATE_HIER_DEPTH_BUFFER << 16 | (3 - 2));
223       OUT_BATCH(hiz_mt->region->pitch * hiz_mt->region->cpp - 1);
224       OUT_RELOC(hiz_mt->region->bo,
225                 I915_GEM_DOMAIN_RENDER,
226                 I915_GEM_DOMAIN_RENDER,
227                 hiz_offset);
228       ADVANCE_BATCH();
229    }
230 
231    if (stencil_mt == NULL) {
232       BEGIN_BATCH(3);
233       OUT_BATCH(GEN7_3DSTATE_STENCIL_BUFFER << 16 | (3 - 2));
234       OUT_BATCH(0);
235       OUT_BATCH(0);
236       ADVANCE_BATCH();
237    } else {
238       const int enabled = intel->is_haswell ? HSW_STENCIL_ENABLED : 0;
239 
240       /* Note: We can't compute the stencil offset using
241        * intel_region_get_aligned_offset(), because the stencil region claims
242        * that the region is untiled; in fact it's W tiled.
243        */
244       uint32_t stencil_offset =
245          (draw_y & ~tile_mask_y) * stencil_mt->region->pitch +
246          (draw_x & ~tile_mask_x) * 64;
247 
248       BEGIN_BATCH(3);
249       OUT_BATCH(GEN7_3DSTATE_STENCIL_BUFFER << 16 | (3 - 2));
250       /* The stencil buffer has quirky pitch requirements.  From the Graphics
251        * BSpec: vol2a.11 3D Pipeline Windower > Early Depth/Stencil Processing
252        * > Depth/Stencil Buffer State > 3DSTATE_STENCIL_BUFFER [DevIVB+],
253        * field "Surface Pitch":
254        *
255        *    The pitch must be set to 2x the value computed based on width, as
256        *    the stencil buffer is stored with two rows interleaved.
257        *
258        * (Note that it is not 100% clear whether this intended to apply to
259        * Gen7; the BSpec flags this comment as "DevILK,DevSNB" (which would
260        * imply that it doesn't), however the comment appears on a "DevIVB+"
261        * page (which would imply that it does).  Experiments with the hardware
262        * indicate that it does.
263        */
264       OUT_BATCH(enabled |
265 	        (2 * stencil_mt->region->pitch * stencil_mt->region->cpp - 1));
266       OUT_RELOC(stencil_mt->region->bo,
267 	        I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
268 		stencil_offset);
269       ADVANCE_BATCH();
270    }
271 
272    BEGIN_BATCH(3);
273    OUT_BATCH(GEN7_3DSTATE_CLEAR_PARAMS << 16 | (3 - 2));
274    OUT_BATCH(depth_mt ? depth_mt->depth_clear_value : 0);
275    OUT_BATCH(1);
276    ADVANCE_BATCH();
277 }
278 
279 /**
280  * \see brw_context.state.depth_region
281  */
282 const struct brw_tracked_state gen7_depthbuffer = {
283    .dirty = {
284       .mesa = (_NEW_BUFFERS | _NEW_DEPTH | _NEW_STENCIL),
285       .brw = BRW_NEW_BATCH,
286       .cache = 0,
287    },
288    .emit = emit_depthbuffer,
289 };
290