OSDN Git Service

i965: Fix out-of-order sampler unit usage in ARB fragment programs.
[android-x86/external-mesa.git] / src / mesa / drivers / dri / i965 / brw_wm_surface_state.c
1 /*
2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
3  Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4  develop this 3D driver.
5  
6  Permission is hereby granted, free of charge, to any person obtaining
7  a copy of this software and associated documentation files (the
8  "Software"), to deal in the Software without restriction, including
9  without limitation the rights to use, copy, modify, merge, publish,
10  distribute, sublicense, and/or sell copies of the Software, and to
11  permit persons to whom the Software is furnished to do so, subject to
12  the following conditions:
13  
14  The above copyright notice and this permission notice (including the
15  next paragraph) shall be included in all copies or substantial
16  portions of the Software.
17  
18  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21  IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25  
26  **********************************************************************/
27  /*
28   * Authors:
29   *   Keith Whitwell <keith@tungstengraphics.com>
30   */
31                    
32
33 #include "main/mtypes.h"
34 #include "main/samplerobj.h"
35 #include "program/prog_parameter.h"
36
37 #include "intel_mipmap_tree.h"
38 #include "intel_batchbuffer.h"
39 #include "intel_tex.h"
40 #include "intel_fbo.h"
41 #include "intel_buffer_objects.h"
42
43 #include "brw_context.h"
44 #include "brw_state.h"
45 #include "brw_defines.h"
46 #include "brw_wm.h"
47
48 GLuint
49 translate_tex_target(GLenum target)
50 {
51    switch (target) {
52    case GL_TEXTURE_1D: 
53    case GL_TEXTURE_1D_ARRAY_EXT:
54       return BRW_SURFACE_1D;
55
56    case GL_TEXTURE_RECTANGLE_NV: 
57       return BRW_SURFACE_2D;
58
59    case GL_TEXTURE_2D: 
60    case GL_TEXTURE_2D_ARRAY_EXT:
61    case GL_TEXTURE_EXTERNAL_OES:
62       return BRW_SURFACE_2D;
63
64    case GL_TEXTURE_3D: 
65       return BRW_SURFACE_3D;
66
67    case GL_TEXTURE_CUBE_MAP: 
68       return BRW_SURFACE_CUBE;
69
70    default: 
71       assert(0); 
72       return 0;
73    }
74 }
75
76 struct surface_format_info {
77    bool exists;
78    int sampling;
79    int filtering;
80    int shadow_compare;
81    int chroma_key;
82    int render_target;
83    int alpha_blend;
84    int input_vb;
85    int streamed_output_vb;
86    int color_processing;
87 };
88
89 /* This macro allows us to write the table almost as it appears in the PRM,
90  * while restructuring it to turn it into the C code we want.
91  */
92 #define SF(sampl, filt, shad, ck, rt, ab, vb, so, color, sf) \
93    [sf] = { true, sampl, filt, shad, ck, rt, ab, vb, so, color },
94
95 #define Y 0
96 #define x 999
97 /**
98  * This is the table of support for surface (texture, renderbuffer, and vertex
99  * buffer, but not depthbuffer) formats across the various hardware generations.
100  *
101  * The table is formatted to match the documentation, except that the docs have
102  * this ridiculous mapping of Y[*+~^#&] for "supported on DevWhatever".  To put
103  * it in our table, here's the mapping:
104  *
105  * Y*: 45
106  * Y+: 45 (g45/gm45)
107  * Y~: 50 (gen5)
108  * Y^: 60 (gen6)
109  * Y#: 70 (gen7)
110  *
111  * See page 88 of the Sandybridge PRM VOL4_Part1 PDF.
112  */
113 const struct surface_format_info surface_formats[] = {
114 /* smpl filt shad CK  RT  AB  VB  SO  color */
115    SF( Y, 50,  x,  x,  Y,  Y,  Y,  Y,  x, BRW_SURFACEFORMAT_R32G32B32A32_FLOAT)
116    SF( Y,  x,  x,  x,  Y,  x,  Y,  Y,  x, BRW_SURFACEFORMAT_R32G32B32A32_SINT)
117    SF( Y,  x,  x,  x,  Y,  x,  Y,  Y,  x, BRW_SURFACEFORMAT_R32G32B32A32_UINT)
118    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32G32B32A32_UNORM)
119    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32G32B32A32_SNORM)
120    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R64G64_FLOAT)
121    SF( Y, 50,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_R32G32B32X32_FLOAT)
122    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32G32B32A32_SSCALED)
123    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32G32B32A32_USCALED)
124    SF( Y, 50,  x,  x,  x,  x,  Y,  Y,  x, BRW_SURFACEFORMAT_R32G32B32_FLOAT)
125    SF( Y,  x,  x,  x,  x,  x,  Y,  Y,  x, BRW_SURFACEFORMAT_R32G32B32_SINT)
126    SF( Y,  x,  x,  x,  x,  x,  Y,  Y,  x, BRW_SURFACEFORMAT_R32G32B32_UINT)
127    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32G32B32_UNORM)
128    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32G32B32_SNORM)
129    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32G32B32_SSCALED)
130    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32G32B32_USCALED)
131    SF( Y,  Y,  x,  x,  Y, 45,  Y,  x, 60, BRW_SURFACEFORMAT_R16G16B16A16_UNORM)
132    SF( Y,  Y,  x,  x,  Y, 60,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16B16A16_SNORM)
133    SF( Y,  x,  x,  x,  Y,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16B16A16_SINT)
134    SF( Y,  x,  x,  x,  Y,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16B16A16_UINT)
135    SF( Y,  Y,  x,  x,  Y,  Y,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16B16A16_FLOAT)
136    SF( Y, 50,  x,  x,  Y,  Y,  Y,  Y,  x, BRW_SURFACEFORMAT_R32G32_FLOAT)
137    SF( Y,  x,  x,  x,  Y,  x,  Y,  Y,  x, BRW_SURFACEFORMAT_R32G32_SINT)
138    SF( Y,  x,  x,  x,  Y,  x,  Y,  Y,  x, BRW_SURFACEFORMAT_R32G32_UINT)
139    SF( Y, 50,  Y,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_R32_FLOAT_X8X24_TYPELESS)
140    SF( Y,  x,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_X32_TYPELESS_G8X24_UINT)
141    SF( Y, 50,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_L32A32_FLOAT)
142    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32G32_UNORM)
143    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32G32_SNORM)
144    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R64_FLOAT)
145    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_R16G16B16X16_UNORM)
146    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_R16G16B16X16_FLOAT)
147    SF( Y, 50,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_A32X32_FLOAT)
148    SF( Y, 50,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_L32X32_FLOAT)
149    SF( Y, 50,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_I32X32_FLOAT)
150    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16B16A16_SSCALED)
151    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16B16A16_USCALED)
152    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32G32_SSCALED)
153    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32G32_USCALED)
154    SF( Y,  Y,  x,  Y,  Y,  Y,  Y,  x, 60, BRW_SURFACEFORMAT_B8G8R8A8_UNORM)
155    SF( Y,  Y,  x,  x,  Y,  Y,  x,  x,  x, BRW_SURFACEFORMAT_B8G8R8A8_UNORM_SRGB)
156 /* smpl filt shad CK  RT  AB  VB  SO  color */
157    SF( Y,  Y,  x,  x,  Y,  Y,  Y,  x, 60, BRW_SURFACEFORMAT_R10G10B10A2_UNORM)
158    SF( Y,  Y,  x,  x,  x,  x,  x,  x, 60, BRW_SURFACEFORMAT_R10G10B10A2_UNORM_SRGB)
159    SF( Y,  x,  x,  x,  Y,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R10G10B10A2_UINT)
160    SF( Y,  Y,  x,  x,  x,  Y,  Y,  x,  x, BRW_SURFACEFORMAT_R10G10B10_SNORM_A2_UNORM)
161    SF( Y,  Y,  x,  x,  Y,  Y,  Y,  x, 60, BRW_SURFACEFORMAT_R8G8B8A8_UNORM)
162    SF( Y,  Y,  x,  x,  Y,  Y,  x,  x, 60, BRW_SURFACEFORMAT_R8G8B8A8_UNORM_SRGB)
163    SF( Y,  Y,  x,  x,  Y, 60,  Y,  x,  x, BRW_SURFACEFORMAT_R8G8B8A8_SNORM)
164    SF( Y,  x,  x,  x,  Y,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8G8B8A8_SINT)
165    SF( Y,  x,  x,  x,  Y,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8G8B8A8_UINT)
166    SF( Y,  Y,  x,  x,  Y, 45,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16_UNORM)
167    SF( Y,  Y,  x,  x,  Y, 60,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16_SNORM)
168    SF( Y,  x,  x,  x,  Y,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16_SINT)
169    SF( Y,  x,  x,  x,  Y,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16_UINT)
170    SF( Y,  Y,  x,  x,  Y,  Y,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16_FLOAT)
171    SF( Y,  Y,  x,  x,  Y,  Y,  x,  x, 60, BRW_SURFACEFORMAT_B10G10R10A2_UNORM)
172    SF( Y,  Y,  x,  x,  Y,  Y,  x,  x, 60, BRW_SURFACEFORMAT_B10G10R10A2_UNORM_SRGB)
173    SF( Y,  Y,  x,  x,  Y,  Y,  Y,  x,  x, BRW_SURFACEFORMAT_R11G11B10_FLOAT)
174    SF( Y,  x,  x,  x,  Y,  x,  Y,  Y,  x, BRW_SURFACEFORMAT_R32_SINT)
175    SF( Y,  x,  x,  x,  Y,  x,  Y,  Y,  x, BRW_SURFACEFORMAT_R32_UINT)
176    SF( Y, 50,  Y,  x,  Y,  Y,  Y,  Y,  x, BRW_SURFACEFORMAT_R32_FLOAT)
177    SF( Y, 50,  Y,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_R24_UNORM_X8_TYPELESS)
178    SF( Y,  x,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_X24_TYPELESS_G8_UINT)
179    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_L16A16_UNORM)
180    SF( Y, 50,  Y,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_I24X8_UNORM)
181    SF( Y, 50,  Y,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_L24X8_UNORM)
182    SF( Y, 50,  Y,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_A24X8_UNORM)
183    SF( Y, 50,  Y,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_I32_FLOAT)
184    SF( Y, 50,  Y,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_L32_FLOAT)
185    SF( Y, 50,  Y,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_A32_FLOAT)
186    SF( Y,  Y,  x,  Y,  x,  x,  x,  x, 60, BRW_SURFACEFORMAT_B8G8R8X8_UNORM)
187    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_B8G8R8X8_UNORM_SRGB)
188    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_R8G8B8X8_UNORM)
189    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_R8G8B8X8_UNORM_SRGB)
190    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_R9G9B9E5_SHAREDEXP)
191    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_B10G10R10X2_UNORM)
192    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_L16A16_FLOAT)
193    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32_UNORM)
194    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32_SNORM)
195 /* smpl filt shad CK  RT  AB  VB  SO  color */
196    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R10G10B10X2_USCALED)
197    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8G8B8A8_SSCALED)
198    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8G8B8A8_USCALED)
199    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16_SSCALED)
200    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16_USCALED)
201    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32_SSCALED)
202    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R32_USCALED)
203    SF( Y,  Y,  x,  Y,  Y,  Y,  x,  x,  x, BRW_SURFACEFORMAT_B5G6R5_UNORM)
204    SF( Y,  Y,  x,  x,  Y,  Y,  x,  x,  x, BRW_SURFACEFORMAT_B5G6R5_UNORM_SRGB)
205    SF( Y,  Y,  x,  Y,  Y,  Y,  x,  x,  x, BRW_SURFACEFORMAT_B5G5R5A1_UNORM)
206    SF( Y,  Y,  x,  x,  Y,  Y,  x,  x,  x, BRW_SURFACEFORMAT_B5G5R5A1_UNORM_SRGB)
207    SF( Y,  Y,  x,  Y,  Y,  Y,  x,  x,  x, BRW_SURFACEFORMAT_B4G4R4A4_UNORM)
208    SF( Y,  Y,  x,  x,  Y,  Y,  x,  x,  x, BRW_SURFACEFORMAT_B4G4R4A4_UNORM_SRGB)
209    SF( Y,  Y,  x,  x,  Y,  Y,  Y,  x,  x, BRW_SURFACEFORMAT_R8G8_UNORM)
210    SF( Y,  Y,  x,  Y,  Y, 60,  Y,  x,  x, BRW_SURFACEFORMAT_R8G8_SNORM)
211    SF( Y,  x,  x,  x,  Y,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8G8_SINT)
212    SF( Y,  x,  x,  x,  Y,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8G8_UINT)
213    SF( Y,  Y,  Y,  x,  Y, 45,  Y,  x, 70, BRW_SURFACEFORMAT_R16_UNORM)
214    SF( Y,  Y,  x,  x,  Y, 60,  Y,  x,  x, BRW_SURFACEFORMAT_R16_SNORM)
215    SF( Y,  x,  x,  x,  Y,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16_SINT)
216    SF( Y,  x,  x,  x,  Y,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16_UINT)
217    SF( Y,  Y,  x,  x,  Y,  Y,  Y,  x,  x, BRW_SURFACEFORMAT_R16_FLOAT)
218    SF( Y,  Y,  Y,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_I16_UNORM)
219    SF( Y,  Y,  Y,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_L16_UNORM)
220    SF( Y,  Y,  Y,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_A16_UNORM)
221    SF( Y,  Y,  x,  Y,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_L8A8_UNORM)
222    SF( Y,  Y,  Y,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_I16_FLOAT)
223    SF( Y,  Y,  Y,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_L16_FLOAT)
224    SF( Y,  Y,  Y,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_A16_FLOAT)
225    SF(45, 45,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_L8A8_UNORM_SRGB)
226    SF( Y,  Y,  x,  Y,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_R5G5_SNORM_B6_UNORM)
227    SF( x,  x,  x,  x,  Y,  Y,  x,  x,  x, BRW_SURFACEFORMAT_B5G5R5X1_UNORM)
228    SF( x,  x,  x,  x,  Y,  Y,  x,  x,  x, BRW_SURFACEFORMAT_B5G5R5X1_UNORM_SRGB)
229    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8G8_SSCALED)
230    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8G8_USCALED)
231 /* smpl filt shad CK  RT  AB  VB  SO  color */
232    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16_SSCALED)
233    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16_USCALED)
234    SF( Y,  Y,  x, 45,  Y,  Y,  Y,  x,  x, BRW_SURFACEFORMAT_R8_UNORM)
235    SF( Y,  Y,  x,  x,  Y, 60,  Y,  x,  x, BRW_SURFACEFORMAT_R8_SNORM)
236    SF( Y,  x,  x,  x,  Y,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8_SINT)
237    SF( Y,  x,  x,  x,  Y,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8_UINT)
238    SF( Y,  Y,  x,  Y,  Y,  Y,  x,  x,  x, BRW_SURFACEFORMAT_A8_UNORM)
239    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_I8_UNORM)
240    SF( Y,  Y,  x,  Y,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_L8_UNORM)
241    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_P4A4_UNORM)
242    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_A4P4_UNORM)
243    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8_SSCALED)
244    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8_USCALED)
245    SF(45, 45,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_L8_UNORM_SRGB)
246    SF(45, 45,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_DXT1_RGB_SRGB)
247    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_R1_UINT)
248    SF( Y,  Y,  x,  Y,  Y,  x,  x,  x, 60, BRW_SURFACEFORMAT_YCRCB_NORMAL)
249    SF( Y,  Y,  x,  Y,  Y,  x,  x,  x, 60, BRW_SURFACEFORMAT_YCRCB_SWAPUVY)
250    SF( Y,  Y,  x,  Y,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_BC1_UNORM)
251    SF( Y,  Y,  x,  Y,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_BC2_UNORM)
252    SF( Y,  Y,  x,  Y,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_BC3_UNORM)
253    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_BC4_UNORM)
254    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_BC5_UNORM)
255    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_BC1_UNORM_SRGB)
256    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_BC2_UNORM_SRGB)
257    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_BC3_UNORM_SRGB)
258    SF( Y,  x,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_MONO8)
259    SF( Y,  Y,  x,  x,  Y,  x,  x,  x, 60, BRW_SURFACEFORMAT_YCRCB_SWAPUV)
260    SF( Y,  Y,  x,  x,  Y,  x,  x,  x, 60, BRW_SURFACEFORMAT_YCRCB_SWAPY)
261    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_DXT1_RGB)
262 /* smpl filt shad CK  RT  AB  VB  SO  color */
263    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_FXT1)
264    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8G8B8_UNORM)
265    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8G8B8_SNORM)
266    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8G8B8_SSCALED)
267    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R8G8B8_USCALED)
268    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R64G64B64A64_FLOAT)
269    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R64G64B64_FLOAT)
270    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_BC4_SNORM)
271    SF( Y,  Y,  x,  x,  x,  x,  x,  x,  x, BRW_SURFACEFORMAT_BC5_SNORM)
272    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16B16_UNORM)
273    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16B16_SNORM)
274    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16B16_SSCALED)
275    SF( x,  x,  x,  x,  x,  x,  Y,  x,  x, BRW_SURFACEFORMAT_R16G16B16_USCALED)
276 };
277 #undef x
278 #undef Y
279
280 uint32_t
281 brw_format_for_mesa_format(gl_format mesa_format)
282 {
283    /* This table is ordered according to the enum ordering in formats.h.  We do
284     * expect that enum to be extended without our explicit initialization
285     * staying in sync, so we initialize to 0 even though
286     * BRW_SURFACEFORMAT_R32G32B32A32_FLOAT happens to also be 0.
287     */
288    static const uint32_t table[MESA_FORMAT_COUNT] =
289    {
290       [MESA_FORMAT_RGBA8888] = 0,
291       [MESA_FORMAT_RGBA8888_REV] = BRW_SURFACEFORMAT_R8G8B8A8_UNORM,
292       [MESA_FORMAT_ARGB8888] = BRW_SURFACEFORMAT_B8G8R8A8_UNORM,
293       [MESA_FORMAT_ARGB8888_REV] = 0,
294       [MESA_FORMAT_RGBX8888] = 0,
295       [MESA_FORMAT_RGBX8888_REV] = BRW_SURFACEFORMAT_R8G8B8X8_UNORM,
296       [MESA_FORMAT_XRGB8888] = BRW_SURFACEFORMAT_B8G8R8X8_UNORM,
297       [MESA_FORMAT_XRGB8888_REV] = 0,
298       [MESA_FORMAT_RGB888] = 0,
299       [MESA_FORMAT_BGR888] = 0,
300       [MESA_FORMAT_RGB565] = BRW_SURFACEFORMAT_B5G6R5_UNORM,
301       [MESA_FORMAT_RGB565_REV] = 0,
302       [MESA_FORMAT_ARGB4444] = BRW_SURFACEFORMAT_B4G4R4A4_UNORM,
303       [MESA_FORMAT_ARGB4444_REV] = 0,
304       [MESA_FORMAT_RGBA5551] = 0,
305       [MESA_FORMAT_ARGB1555] = BRW_SURFACEFORMAT_B5G5R5A1_UNORM,
306       [MESA_FORMAT_ARGB1555_REV] = 0,
307       [MESA_FORMAT_AL44] = 0,
308       [MESA_FORMAT_AL88] = BRW_SURFACEFORMAT_L8A8_UNORM,
309       [MESA_FORMAT_AL88_REV] = 0,
310       [MESA_FORMAT_AL1616] = BRW_SURFACEFORMAT_L16A16_UNORM,
311       [MESA_FORMAT_AL1616_REV] = 0,
312       [MESA_FORMAT_RGB332] = 0,
313       [MESA_FORMAT_A8] = BRW_SURFACEFORMAT_A8_UNORM,
314       [MESA_FORMAT_A16] = BRW_SURFACEFORMAT_A16_UNORM,
315       [MESA_FORMAT_L8] = BRW_SURFACEFORMAT_L8_UNORM,
316       [MESA_FORMAT_L16] = BRW_SURFACEFORMAT_L16_UNORM,
317       [MESA_FORMAT_I8] = BRW_SURFACEFORMAT_I8_UNORM,
318       [MESA_FORMAT_I16] = BRW_SURFACEFORMAT_I16_UNORM,
319       [MESA_FORMAT_YCBCR_REV] = BRW_SURFACEFORMAT_YCRCB_NORMAL,
320       [MESA_FORMAT_YCBCR] = BRW_SURFACEFORMAT_YCRCB_SWAPUVY,
321       [MESA_FORMAT_R8] = BRW_SURFACEFORMAT_R8_UNORM,
322       [MESA_FORMAT_GR88] = BRW_SURFACEFORMAT_R8G8_UNORM,
323       [MESA_FORMAT_RG88] = 0,
324       [MESA_FORMAT_R16] = BRW_SURFACEFORMAT_R16_UNORM,
325       [MESA_FORMAT_RG1616] = BRW_SURFACEFORMAT_R16G16_UNORM,
326       [MESA_FORMAT_RG1616_REV] = 0,
327       [MESA_FORMAT_ARGB2101010] = BRW_SURFACEFORMAT_B10G10R10A2_UNORM,
328       [MESA_FORMAT_ABGR2101010_UINT] = BRW_SURFACEFORMAT_R10G10B10A2_UINT,
329       [MESA_FORMAT_Z24_S8] = 0,
330       [MESA_FORMAT_S8_Z24] = 0,
331       [MESA_FORMAT_Z16] = 0,
332       [MESA_FORMAT_X8_Z24] = 0,
333       [MESA_FORMAT_Z24_X8] = 0,
334       [MESA_FORMAT_Z32] = 0,
335       [MESA_FORMAT_S8] = 0,
336
337       [MESA_FORMAT_SRGB8] = 0,
338       [MESA_FORMAT_SRGBA8] = 0,
339       [MESA_FORMAT_SARGB8] = BRW_SURFACEFORMAT_B8G8R8A8_UNORM_SRGB,
340       [MESA_FORMAT_SL8] = BRW_SURFACEFORMAT_L8_UNORM_SRGB,
341       [MESA_FORMAT_SLA8] = BRW_SURFACEFORMAT_L8A8_UNORM_SRGB,
342       [MESA_FORMAT_SRGB_DXT1] = BRW_SURFACEFORMAT_DXT1_RGB_SRGB,
343       [MESA_FORMAT_SRGBA_DXT1] = BRW_SURFACEFORMAT_BC1_UNORM_SRGB,
344       [MESA_FORMAT_SRGBA_DXT3] = BRW_SURFACEFORMAT_BC2_UNORM_SRGB,
345       [MESA_FORMAT_SRGBA_DXT5] = BRW_SURFACEFORMAT_BC3_UNORM_SRGB,
346
347       [MESA_FORMAT_RGB_FXT1] = BRW_SURFACEFORMAT_FXT1,
348       [MESA_FORMAT_RGBA_FXT1] = BRW_SURFACEFORMAT_FXT1,
349       [MESA_FORMAT_RGB_DXT1] = BRW_SURFACEFORMAT_DXT1_RGB,
350       [MESA_FORMAT_RGBA_DXT1] = BRW_SURFACEFORMAT_BC1_UNORM,
351       [MESA_FORMAT_RGBA_DXT3] = BRW_SURFACEFORMAT_BC2_UNORM,
352       [MESA_FORMAT_RGBA_DXT5] = BRW_SURFACEFORMAT_BC3_UNORM,
353
354       [MESA_FORMAT_RGBA_FLOAT32] = BRW_SURFACEFORMAT_R32G32B32A32_FLOAT,
355       [MESA_FORMAT_RGBA_FLOAT16] = BRW_SURFACEFORMAT_R16G16B16A16_FLOAT,
356       [MESA_FORMAT_RGB_FLOAT32] = 0,
357       [MESA_FORMAT_RGB_FLOAT16] = 0,
358       [MESA_FORMAT_ALPHA_FLOAT32] = BRW_SURFACEFORMAT_A32_FLOAT,
359       [MESA_FORMAT_ALPHA_FLOAT16] = BRW_SURFACEFORMAT_A16_FLOAT,
360       [MESA_FORMAT_LUMINANCE_FLOAT32] = BRW_SURFACEFORMAT_L32_FLOAT,
361       [MESA_FORMAT_LUMINANCE_FLOAT16] = BRW_SURFACEFORMAT_L16_FLOAT,
362       [MESA_FORMAT_LUMINANCE_ALPHA_FLOAT32] = BRW_SURFACEFORMAT_L32A32_FLOAT,
363       [MESA_FORMAT_LUMINANCE_ALPHA_FLOAT16] = BRW_SURFACEFORMAT_L16A16_FLOAT,
364       [MESA_FORMAT_INTENSITY_FLOAT32] = BRW_SURFACEFORMAT_I32_FLOAT,
365       [MESA_FORMAT_INTENSITY_FLOAT16] = BRW_SURFACEFORMAT_I16_FLOAT,
366       [MESA_FORMAT_R_FLOAT32] = BRW_SURFACEFORMAT_R32_FLOAT,
367       [MESA_FORMAT_R_FLOAT16] = BRW_SURFACEFORMAT_R16_FLOAT,
368       [MESA_FORMAT_RG_FLOAT32] = BRW_SURFACEFORMAT_R32G32_FLOAT,
369       [MESA_FORMAT_RG_FLOAT16] = BRW_SURFACEFORMAT_R16G16_FLOAT,
370
371       [MESA_FORMAT_ALPHA_UINT8] = 0,
372       [MESA_FORMAT_ALPHA_UINT16] = 0,
373       [MESA_FORMAT_ALPHA_UINT32] = 0,
374       [MESA_FORMAT_ALPHA_INT8] = 0,
375       [MESA_FORMAT_ALPHA_INT16] = 0,
376       [MESA_FORMAT_ALPHA_INT32] = 0,
377
378       [MESA_FORMAT_INTENSITY_UINT8] = 0,
379       [MESA_FORMAT_INTENSITY_UINT16] = 0,
380       [MESA_FORMAT_INTENSITY_UINT32] = 0,
381       [MESA_FORMAT_INTENSITY_INT8] = 0,
382       [MESA_FORMAT_INTENSITY_INT16] = 0,
383       [MESA_FORMAT_INTENSITY_INT32] = 0,
384
385       [MESA_FORMAT_LUMINANCE_UINT8] = 0,
386       [MESA_FORMAT_LUMINANCE_UINT16] = 0,
387       [MESA_FORMAT_LUMINANCE_UINT32] = 0,
388       [MESA_FORMAT_LUMINANCE_INT8] = 0,
389       [MESA_FORMAT_LUMINANCE_INT16] = 0,
390       [MESA_FORMAT_LUMINANCE_INT32] = 0,
391
392       [MESA_FORMAT_LUMINANCE_ALPHA_UINT8] = 0,
393       [MESA_FORMAT_LUMINANCE_ALPHA_UINT16] = 0,
394       [MESA_FORMAT_LUMINANCE_ALPHA_UINT32] = 0,
395       [MESA_FORMAT_LUMINANCE_ALPHA_INT8] = 0,
396       [MESA_FORMAT_LUMINANCE_ALPHA_INT16] = 0,
397       [MESA_FORMAT_LUMINANCE_ALPHA_INT32] = 0,
398
399       [MESA_FORMAT_R_INT8] = BRW_SURFACEFORMAT_R8_SINT,
400       [MESA_FORMAT_RG_INT8] = BRW_SURFACEFORMAT_R8G8_SINT,
401       [MESA_FORMAT_RGB_INT8] = 0,
402       [MESA_FORMAT_RGBA_INT8] = BRW_SURFACEFORMAT_R8G8B8A8_SINT,
403       [MESA_FORMAT_R_INT16] = BRW_SURFACEFORMAT_R16_SINT,
404       [MESA_FORMAT_RG_INT16] = BRW_SURFACEFORMAT_R16G16_SINT,
405       [MESA_FORMAT_RGB_INT16] = 0,
406       [MESA_FORMAT_RGBA_INT16] = BRW_SURFACEFORMAT_R16G16B16A16_SINT,
407       [MESA_FORMAT_R_INT32] = BRW_SURFACEFORMAT_R32_SINT,
408       [MESA_FORMAT_RG_INT32] = BRW_SURFACEFORMAT_R32G32_SINT,
409       [MESA_FORMAT_RGB_INT32] = BRW_SURFACEFORMAT_R32G32B32_SINT,
410       [MESA_FORMAT_RGBA_INT32] = BRW_SURFACEFORMAT_R32G32B32A32_SINT,
411
412       [MESA_FORMAT_R_UINT8] = BRW_SURFACEFORMAT_R8_UINT,
413       [MESA_FORMAT_RG_UINT8] = BRW_SURFACEFORMAT_R8G8_UINT,
414       [MESA_FORMAT_RGB_UINT8] = 0,
415       [MESA_FORMAT_RGBA_UINT8] = BRW_SURFACEFORMAT_R8G8B8A8_UINT,
416       [MESA_FORMAT_R_UINT16] = BRW_SURFACEFORMAT_R16_UINT,
417       [MESA_FORMAT_RG_UINT16] = BRW_SURFACEFORMAT_R16G16_UINT,
418       [MESA_FORMAT_RGB_UINT16] = 0,
419       [MESA_FORMAT_RGBA_UINT16] = BRW_SURFACEFORMAT_R16G16B16A16_UINT,
420       [MESA_FORMAT_R_UINT32] = BRW_SURFACEFORMAT_R32_UINT,
421       [MESA_FORMAT_RG_UINT32] = BRW_SURFACEFORMAT_R32G32_UINT,
422       [MESA_FORMAT_RGB_UINT32] = BRW_SURFACEFORMAT_R32G32B32_UINT,
423       [MESA_FORMAT_RGBA_UINT32] = BRW_SURFACEFORMAT_R32G32B32A32_UINT,
424
425       [MESA_FORMAT_DUDV8] = BRW_SURFACEFORMAT_R8G8_SNORM,
426       [MESA_FORMAT_SIGNED_R8] = BRW_SURFACEFORMAT_R8_SNORM,
427       [MESA_FORMAT_SIGNED_RG88_REV] = BRW_SURFACEFORMAT_R8G8_SNORM,
428       [MESA_FORMAT_SIGNED_RGBX8888] = 0,
429       [MESA_FORMAT_SIGNED_RGBA8888] = 0,
430       [MESA_FORMAT_SIGNED_RGBA8888_REV] = BRW_SURFACEFORMAT_R8G8B8A8_SNORM,
431       [MESA_FORMAT_SIGNED_R16] = BRW_SURFACEFORMAT_R16_SNORM,
432       [MESA_FORMAT_SIGNED_GR1616] = BRW_SURFACEFORMAT_R16G16_SNORM,
433       [MESA_FORMAT_SIGNED_RGB_16] = 0,
434       [MESA_FORMAT_SIGNED_RGBA_16] = BRW_SURFACEFORMAT_R16G16B16A16_SNORM,
435       [MESA_FORMAT_RGBA_16] = BRW_SURFACEFORMAT_R16G16B16A16_UNORM,
436
437       [MESA_FORMAT_RED_RGTC1] = BRW_SURFACEFORMAT_BC4_UNORM,
438       [MESA_FORMAT_SIGNED_RED_RGTC1] = BRW_SURFACEFORMAT_BC4_SNORM,
439       [MESA_FORMAT_RG_RGTC2] = BRW_SURFACEFORMAT_BC5_UNORM,
440       [MESA_FORMAT_SIGNED_RG_RGTC2] = BRW_SURFACEFORMAT_BC5_SNORM,
441
442       [MESA_FORMAT_L_LATC1] = 0,
443       [MESA_FORMAT_SIGNED_L_LATC1] = 0,
444       [MESA_FORMAT_LA_LATC2] = 0,
445       [MESA_FORMAT_SIGNED_LA_LATC2] = 0,
446
447       [MESA_FORMAT_SIGNED_A8] = 0,
448       [MESA_FORMAT_SIGNED_L8] = 0,
449       [MESA_FORMAT_SIGNED_AL88] = 0,
450       [MESA_FORMAT_SIGNED_I8] = 0,
451       [MESA_FORMAT_SIGNED_A16] = 0,
452       [MESA_FORMAT_SIGNED_L16] = 0,
453       [MESA_FORMAT_SIGNED_AL1616] = 0,
454       [MESA_FORMAT_SIGNED_I16] = 0,
455
456       [MESA_FORMAT_RGB9_E5_FLOAT] = BRW_SURFACEFORMAT_R9G9B9E5_SHAREDEXP,
457       [MESA_FORMAT_R11_G11_B10_FLOAT] = BRW_SURFACEFORMAT_R11G11B10_FLOAT,
458
459       [MESA_FORMAT_Z32_FLOAT] = 0,
460       [MESA_FORMAT_Z32_FLOAT_X24S8] = 0,
461    };
462    assert(mesa_format < MESA_FORMAT_COUNT);
463    return table[mesa_format];
464 }
465
466 void
467 brw_init_surface_formats(struct brw_context *brw)
468 {
469    struct intel_context *intel = &brw->intel;
470    struct gl_context *ctx = &intel->ctx;
471    int gen;
472    gl_format format;
473
474    gen = intel->gen * 10;
475    if (intel->is_g4x)
476       gen += 5;
477
478    for (format = MESA_FORMAT_NONE + 1; format < MESA_FORMAT_COUNT; format++) {
479       uint32_t texture, render;
480       const struct surface_format_info *rinfo, *tinfo;
481       bool is_integer = _mesa_is_format_integer_color(format);
482
483       render = texture = brw_format_for_mesa_format(format);
484       tinfo = &surface_formats[texture];
485
486       /* The value of BRW_SURFACEFORMAT_R32G32B32A32_FLOAT is 0, so don't skip
487        * it.
488        */
489       if (texture == 0 && format != MESA_FORMAT_RGBA_FLOAT32)
490          continue;
491
492       if (gen >= tinfo->sampling && (gen >= tinfo->filtering || is_integer))
493          ctx->TextureFormatSupported[format] = true;
494
495       /* Re-map some render target formats to make them supported when they
496        * wouldn't be using their format for texturing.
497        */
498       switch (render) {
499          /* For these formats, we just need to read/write the first
500           * channel into R, which is to say that we just treat them as
501           * GL_RED.
502           */
503       case BRW_SURFACEFORMAT_I32_FLOAT:
504       case BRW_SURFACEFORMAT_L32_FLOAT:
505          render = BRW_SURFACEFORMAT_R32_FLOAT;
506          break;
507       case BRW_SURFACEFORMAT_I16_FLOAT:
508       case BRW_SURFACEFORMAT_L16_FLOAT:
509          render = BRW_SURFACEFORMAT_R16_FLOAT;
510          break;
511       case BRW_SURFACEFORMAT_B8G8R8X8_UNORM:
512          /* XRGB is handled as ARGB because the chips in this family
513           * cannot render to XRGB targets.  This means that we have to
514           * mask writes to alpha (ala glColorMask) and reconfigure the
515           * alpha blending hardware to use GL_ONE (or GL_ZERO) for
516           * cases where GL_DST_ALPHA (or GL_ONE_MINUS_DST_ALPHA) is
517           * used.
518           */
519          render = BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
520          break;
521       }
522
523       rinfo = &surface_formats[render];
524
525       /* Note that GL_EXT_texture_integer says that blending doesn't occur for
526        * integer, so we don't need hardware support for blending on it.  Other
527        * than that, GL in general requires alpha blending for render targets,
528        * even though we don't support it for some formats.
529        */
530       if (gen >= rinfo->render_target &&
531           (gen >= rinfo->alpha_blend || is_integer)) {
532          brw->render_target_format[format] = render;
533          brw->format_supported_as_render_target[format] = true;
534       }
535    }
536
537    /* We will check this table for FBO completeness, but the surface format
538     * table above only covered color rendering.
539     */
540    brw->format_supported_as_render_target[MESA_FORMAT_S8_Z24] = true;
541    brw->format_supported_as_render_target[MESA_FORMAT_X8_Z24] = true;
542    brw->format_supported_as_render_target[MESA_FORMAT_S8] = true;
543    brw->format_supported_as_render_target[MESA_FORMAT_Z16] = true;
544    brw->format_supported_as_render_target[MESA_FORMAT_Z32_FLOAT] = true;
545    brw->format_supported_as_render_target[MESA_FORMAT_Z32_FLOAT_X24S8] = true;
546
547    /* We remap depth formats to a supported texturing format in
548     * translate_tex_format().
549     */
550    ctx->TextureFormatSupported[MESA_FORMAT_S8_Z24] = true;
551    ctx->TextureFormatSupported[MESA_FORMAT_X8_Z24] = true;
552    ctx->TextureFormatSupported[MESA_FORMAT_Z32_FLOAT] = true;
553    ctx->TextureFormatSupported[MESA_FORMAT_Z32_FLOAT_X24S8] = true;
554    ctx->TextureFormatSupported[MESA_FORMAT_Z16] = true;
555
556    /* On hardware that lacks support for ETC1, we map ETC1 to RGBX
557     * during glCompressedTexImage2D(). See intel_mipmap_tree::wraps_etc1.
558     */
559    ctx->TextureFormatSupported[MESA_FORMAT_ETC1_RGB8] = true;
560 }
561
562 bool
563 brw_render_target_supported(struct intel_context *intel,
564                             struct gl_renderbuffer *rb)
565 {
566    struct brw_context *brw = brw_context(&intel->ctx);
567    gl_format format = rb->Format;
568
569    /* Many integer formats are promoted to RGBA (like XRGB8888 is), which means
570     * we would consider them renderable even though we don't have surface
571     * support for their alpha behavior and don't have the blending unit
572     * available to fake it like we do for XRGB8888.  Force them to being
573     * unsupported.
574     */
575    if ((rb->_BaseFormat != GL_RGBA &&
576         rb->_BaseFormat != GL_RG &&
577         rb->_BaseFormat != GL_RED) && _mesa_is_format_integer_color(format))
578       return false;
579
580    /* Under some conditions, MSAA is not supported for formats whose width is
581     * more than 64 bits.
582     */
583    if (rb->NumSamples > 0 && _mesa_get_format_bytes(format) > 8) {
584       /* Gen6: MSAA on >64 bit formats is unsupported. */
585       if (intel->gen <= 6)
586          return false;
587
588       /* Gen7: 8x MSAA on >64 bit formats is unsupported. */
589       if (rb->NumSamples >= 8)
590          return false;
591    }
592
593    return brw->format_supported_as_render_target[format];
594 }
595
596 GLuint
597 translate_tex_format(gl_format mesa_format,
598                      GLenum internal_format,
599                      GLenum depth_mode,
600                      GLenum srgb_decode)
601 {
602    if (srgb_decode == GL_SKIP_DECODE_EXT)
603       mesa_format = _mesa_get_srgb_format_linear(mesa_format);
604
605    switch( mesa_format ) {
606
607    case MESA_FORMAT_Z16:
608       return BRW_SURFACEFORMAT_I16_UNORM;
609
610    case MESA_FORMAT_S8_Z24:
611    case MESA_FORMAT_X8_Z24:
612       return BRW_SURFACEFORMAT_I24X8_UNORM;
613
614    case MESA_FORMAT_Z32_FLOAT:
615       return BRW_SURFACEFORMAT_I32_FLOAT;
616
617    case MESA_FORMAT_Z32_FLOAT_X24S8:
618       return BRW_SURFACEFORMAT_R32G32_FLOAT;
619
620    case MESA_FORMAT_RGBA_FLOAT32:
621       /* The value of this BRW_SURFACEFORMAT is 0, which tricks the
622        * assertion below.
623        */
624       return BRW_SURFACEFORMAT_R32G32B32A32_FLOAT;
625
626    default:
627       assert(brw_format_for_mesa_format(mesa_format) != 0);
628       return brw_format_for_mesa_format(mesa_format);
629    }
630 }
631
632 uint32_t
633 brw_get_surface_tiling_bits(uint32_t tiling)
634 {
635    switch (tiling) {
636    case I915_TILING_X:
637       return BRW_SURFACE_TILED;
638    case I915_TILING_Y:
639       return BRW_SURFACE_TILED | BRW_SURFACE_TILED_Y;
640    default:
641       return 0;
642    }
643 }
644
645
646 uint32_t
647 brw_get_surface_num_multisamples(unsigned num_samples)
648 {
649    if (num_samples > 1)
650       return BRW_SURFACE_MULTISAMPLECOUNT_4;
651    else
652       return BRW_SURFACE_MULTISAMPLECOUNT_1;
653 }
654
655
656 static void
657 brw_update_buffer_texture_surface(struct gl_context *ctx,
658                                   unsigned unit,
659                                   uint32_t *binding_table,
660                                   unsigned surf_index)
661 {
662    struct brw_context *brw = brw_context(ctx);
663    struct intel_context *intel = &brw->intel;
664    struct gl_texture_object *tObj = ctx->Texture.Unit[unit]._Current;
665    uint32_t *surf;
666    struct intel_buffer_object *intel_obj =
667       intel_buffer_object(tObj->BufferObject);
668    drm_intel_bo *bo = intel_obj ? intel_obj->buffer : NULL;
669    gl_format format = tObj->_BufferObjectFormat;
670    uint32_t brw_format = brw_format_for_mesa_format(format);
671    int texel_size = _mesa_get_format_bytes(format);
672
673    if (brw_format == 0 && format != MESA_FORMAT_RGBA_FLOAT32) {
674       _mesa_problem(NULL, "bad format %s for texture buffer\n",
675                     _mesa_get_format_name(format));
676    }
677
678    surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
679                           6 * 4, 32, &binding_table[surf_index]);
680
681    surf[0] = (BRW_SURFACE_BUFFER << BRW_SURFACE_TYPE_SHIFT |
682               (brw_format_for_mesa_format(format) << BRW_SURFACE_FORMAT_SHIFT));
683
684    if (intel->gen >= 6)
685       surf[0] |= BRW_SURFACE_RC_READ_WRITE;
686
687    if (bo) {
688       surf[1] = bo->offset; /* reloc */
689
690       /* Emit relocation to surface contents. */
691       drm_intel_bo_emit_reloc(brw->intel.batch.bo,
692                               binding_table[surf_index] + 4,
693                               bo, 0, I915_GEM_DOMAIN_SAMPLER, 0);
694
695       int w = intel_obj->Base.Size / texel_size;
696       surf[2] = ((w & 0x7f) << BRW_SURFACE_WIDTH_SHIFT |
697                  ((w >> 7) & 0x1fff) << BRW_SURFACE_HEIGHT_SHIFT);
698       surf[3] = (((w >> 20) & 0x7f) << BRW_SURFACE_DEPTH_SHIFT |
699                  (texel_size - 1) << BRW_SURFACE_PITCH_SHIFT);
700    } else {
701       surf[1] = 0;
702       surf[2] = 0;
703       surf[3] = 0;
704    }
705
706    surf[4] = 0;
707    surf[5] = 0;
708 }
709
710 static void
711 brw_update_texture_surface(struct gl_context *ctx,
712                            unsigned unit,
713                            uint32_t *binding_table,
714                            unsigned surf_index)
715 {
716    struct brw_context *brw = brw_context(ctx);
717    struct gl_texture_object *tObj = ctx->Texture.Unit[unit]._Current;
718    struct intel_texture_object *intelObj = intel_texture_object(tObj);
719    struct intel_mipmap_tree *mt = intelObj->mt;
720    struct gl_texture_image *firstImage = tObj->Image[0][tObj->BaseLevel];
721    struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit);
722    uint32_t *surf;
723    int width, height, depth;
724
725    if (tObj->Target == GL_TEXTURE_BUFFER) {
726       brw_update_buffer_texture_surface(ctx, unit, binding_table, surf_index);
727       return;
728    }
729
730    intel_miptree_get_dimensions_for_image(firstImage, &width, &height, &depth);
731
732    surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
733                           6 * 4, 32, &binding_table[surf_index]);
734
735    surf[0] = (translate_tex_target(tObj->Target) << BRW_SURFACE_TYPE_SHIFT |
736               BRW_SURFACE_MIPMAPLAYOUT_BELOW << BRW_SURFACE_MIPLAYOUT_SHIFT |
737               BRW_SURFACE_CUBEFACE_ENABLES |
738               (translate_tex_format(mt->format,
739                                     firstImage->InternalFormat,
740                                     tObj->DepthMode,
741                                     sampler->sRGBDecode) <<
742                BRW_SURFACE_FORMAT_SHIFT));
743
744    surf[1] = intelObj->mt->region->bo->offset + intelObj->mt->offset; /* reloc */
745
746    surf[2] = ((intelObj->_MaxLevel - tObj->BaseLevel) << BRW_SURFACE_LOD_SHIFT |
747               (width - 1) << BRW_SURFACE_WIDTH_SHIFT |
748               (height - 1) << BRW_SURFACE_HEIGHT_SHIFT);
749
750    surf[3] = (brw_get_surface_tiling_bits(intelObj->mt->region->tiling) |
751               (depth - 1) << BRW_SURFACE_DEPTH_SHIFT |
752               ((intelObj->mt->region->pitch * intelObj->mt->cpp) - 1) <<
753               BRW_SURFACE_PITCH_SHIFT);
754
755    surf[4] = 0;
756
757    surf[5] = (mt->align_h == 4) ? BRW_SURFACE_VERTICAL_ALIGN_ENABLE : 0;
758
759    /* Emit relocation to surface contents */
760    drm_intel_bo_emit_reloc(brw->intel.batch.bo,
761                            binding_table[surf_index] + 4,
762                            intelObj->mt->region->bo,
763                            intelObj->mt->offset,
764                            I915_GEM_DOMAIN_SAMPLER, 0);
765 }
766
767 /**
768  * Create the constant buffer surface.  Vertex/fragment shader constants will be
769  * read from this buffer with Data Port Read instructions/messages.
770  */
771 void
772 brw_create_constant_surface(struct brw_context *brw,
773                             drm_intel_bo *bo,
774                             uint32_t offset,
775                             int width,
776                             uint32_t *out_offset)
777 {
778    struct intel_context *intel = &brw->intel;
779    const GLint w = width - 1;
780    uint32_t *surf;
781
782    surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
783                           6 * 4, 32, out_offset);
784
785    surf[0] = (BRW_SURFACE_BUFFER << BRW_SURFACE_TYPE_SHIFT |
786               BRW_SURFACE_MIPMAPLAYOUT_BELOW << BRW_SURFACE_MIPLAYOUT_SHIFT |
787               BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_SURFACE_FORMAT_SHIFT);
788
789    if (intel->gen >= 6)
790       surf[0] |= BRW_SURFACE_RC_READ_WRITE;
791
792    surf[1] = bo->offset + offset; /* reloc */
793
794    surf[2] = ((w & 0x7f) << BRW_SURFACE_WIDTH_SHIFT |
795               ((w >> 7) & 0x1fff) << BRW_SURFACE_HEIGHT_SHIFT);
796
797    surf[3] = (((w >> 20) & 0x7f) << BRW_SURFACE_DEPTH_SHIFT |
798               (16 - 1) << BRW_SURFACE_PITCH_SHIFT); /* ignored */
799
800    surf[4] = 0;
801    surf[5] = 0;
802
803    /* Emit relocation to surface contents.  Section 5.1.1 of the gen4
804     * bspec ("Data Cache") says that the data cache does not exist as
805     * a separate cache and is just the sampler cache.
806     */
807    drm_intel_bo_emit_reloc(brw->intel.batch.bo,
808                            *out_offset + 4,
809                            bo, offset,
810                            I915_GEM_DOMAIN_SAMPLER, 0);
811 }
812
813 /**
814  * Set up a binding table entry for use by stream output logic (transform
815  * feedback).
816  *
817  * buffer_size_minus_1 must me less than BRW_MAX_NUM_BUFFER_ENTRIES.
818  */
819 void
820 brw_update_sol_surface(struct brw_context *brw,
821                        struct gl_buffer_object *buffer_obj,
822                        uint32_t *out_offset, unsigned num_vector_components,
823                        unsigned stride_dwords, unsigned offset_dwords)
824 {
825    struct intel_context *intel = &brw->intel;
826    struct intel_buffer_object *intel_bo = intel_buffer_object(buffer_obj);
827    drm_intel_bo *bo =
828       intel_bufferobj_buffer(intel, intel_bo, INTEL_WRITE_PART);
829    uint32_t *surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE, 6 * 4, 32,
830                                     out_offset);
831    uint32_t pitch_minus_1 = 4*stride_dwords - 1;
832    uint32_t offset_bytes = 4 * offset_dwords;
833    size_t size_dwords = buffer_obj->Size / 4;
834    uint32_t buffer_size_minus_1, width, height, depth, surface_format;
835
836    /* FIXME: can we rely on core Mesa to ensure that the buffer isn't
837     * too big to map using a single binding table entry?
838     */
839    assert((size_dwords - offset_dwords) / stride_dwords
840           <= BRW_MAX_NUM_BUFFER_ENTRIES);
841
842    if (size_dwords > offset_dwords + num_vector_components) {
843       /* There is room for at least 1 transform feedback output in the buffer.
844        * Compute the number of additional transform feedback outputs the
845        * buffer has room for.
846        */
847       buffer_size_minus_1 =
848          (size_dwords - offset_dwords - num_vector_components) / stride_dwords;
849    } else {
850       /* There isn't even room for a single transform feedback output in the
851        * buffer.  We can't configure the binding table entry to prevent output
852        * entirely; we'll have to rely on the geometry shader to detect
853        * overflow.  But to minimize the damage in case of a bug, set up the
854        * binding table entry to just allow a single output.
855        */
856       buffer_size_minus_1 = 0;
857    }
858    width = buffer_size_minus_1 & 0x7f;
859    height = (buffer_size_minus_1 & 0xfff80) >> 7;
860    depth = (buffer_size_minus_1 & 0x7f00000) >> 20;
861
862    switch (num_vector_components) {
863    case 1:
864       surface_format = BRW_SURFACEFORMAT_R32_FLOAT;
865       break;
866    case 2:
867       surface_format = BRW_SURFACEFORMAT_R32G32_FLOAT;
868       break;
869    case 3:
870       surface_format = BRW_SURFACEFORMAT_R32G32B32_FLOAT;
871       break;
872    case 4:
873       surface_format = BRW_SURFACEFORMAT_R32G32B32A32_FLOAT;
874       break;
875    default:
876       assert(!"Invalid vector size for transform feedback output");
877       surface_format = BRW_SURFACEFORMAT_R32_FLOAT;
878       break;
879    }
880
881    surf[0] = BRW_SURFACE_BUFFER << BRW_SURFACE_TYPE_SHIFT |
882       BRW_SURFACE_MIPMAPLAYOUT_BELOW << BRW_SURFACE_MIPLAYOUT_SHIFT |
883       surface_format << BRW_SURFACE_FORMAT_SHIFT |
884       BRW_SURFACE_RC_READ_WRITE;
885    surf[1] = bo->offset + offset_bytes; /* reloc */
886    surf[2] = (width << BRW_SURFACE_WIDTH_SHIFT |
887               height << BRW_SURFACE_HEIGHT_SHIFT);
888    surf[3] = (depth << BRW_SURFACE_DEPTH_SHIFT |
889               pitch_minus_1 << BRW_SURFACE_PITCH_SHIFT);
890    surf[4] = 0;
891    surf[5] = 0;
892
893    /* Emit relocation to surface contents. */
894    drm_intel_bo_emit_reloc(brw->intel.batch.bo,
895                            *out_offset + 4,
896                            bo, offset_bytes,
897                            I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER);
898 }
899
900 /* Creates a new WM constant buffer reflecting the current fragment program's
901  * constants, if needed by the fragment program.
902  *
903  * Otherwise, constants go through the CURBEs using the brw_constant_buffer
904  * state atom.
905  */
906 static void
907 brw_upload_wm_pull_constants(struct brw_context *brw)
908 {
909    struct gl_context *ctx = &brw->intel.ctx;
910    struct intel_context *intel = &brw->intel;
911    /* BRW_NEW_FRAGMENT_PROGRAM */
912    struct brw_fragment_program *fp =
913       (struct brw_fragment_program *) brw->fragment_program;
914    struct gl_program_parameter_list *params = fp->program.Base.Parameters;
915    const int size = brw->wm.prog_data->nr_pull_params * sizeof(float);
916    const int surf_index = SURF_INDEX_FRAG_CONST_BUFFER;
917    float *constants;
918    unsigned int i;
919
920    _mesa_load_state_parameters(ctx, params);
921
922    /* CACHE_NEW_WM_PROG */
923    if (brw->wm.prog_data->nr_pull_params == 0) {
924       if (brw->wm.const_bo) {
925          drm_intel_bo_unreference(brw->wm.const_bo);
926          brw->wm.const_bo = NULL;
927          brw->wm.surf_offset[surf_index] = 0;
928          brw->state.dirty.brw |= BRW_NEW_SURFACES;
929       }
930       return;
931    }
932
933    drm_intel_bo_unreference(brw->wm.const_bo);
934    brw->wm.const_bo = drm_intel_bo_alloc(intel->bufmgr, "WM const bo",
935                                          size, 64);
936
937    /* _NEW_PROGRAM_CONSTANTS */
938    drm_intel_gem_bo_map_gtt(brw->wm.const_bo);
939    constants = brw->wm.const_bo->virtual;
940    for (i = 0; i < brw->wm.prog_data->nr_pull_params; i++) {
941       constants[i] = *brw->wm.prog_data->pull_param[i];
942    }
943    drm_intel_gem_bo_unmap_gtt(brw->wm.const_bo);
944
945    intel->vtbl.create_constant_surface(brw, brw->wm.const_bo, 0,
946                                        params->NumParameters,
947                                        &brw->wm.surf_offset[surf_index]);
948
949    brw->state.dirty.brw |= BRW_NEW_SURFACES;
950 }
951
952 const struct brw_tracked_state brw_wm_pull_constants = {
953    .dirty = {
954       .mesa = (_NEW_PROGRAM_CONSTANTS),
955       .brw = (BRW_NEW_BATCH | BRW_NEW_FRAGMENT_PROGRAM),
956       .cache = CACHE_NEW_WM_PROG,
957    },
958    .emit = brw_upload_wm_pull_constants,
959 };
960
961 static void
962 brw_update_null_renderbuffer_surface(struct brw_context *brw, unsigned int unit)
963 {
964    /* From the Sandy bridge PRM, Vol4 Part1 p71 (Surface Type: Programming
965     * Notes):
966     *
967     *     A null surface will be used in instances where an actual surface is
968     *     not bound. When a write message is generated to a null surface, no
969     *     actual surface is written to. When a read message (including any
970     *     sampling engine message) is generated to a null surface, the result
971     *     is all zeros. Note that a null surface type is allowed to be used
972     *     with all messages, even if it is not specificially indicated as
973     *     supported. All of the remaining fields in surface state are ignored
974     *     for null surfaces, with the following exceptions:
975     *
976     *     - [DevSNB+]: Width, Height, Depth, and LOD fields must match the
977     *       depth buffer’s corresponding state for all render target surfaces,
978     *       including null.
979     *
980     *     - Surface Format must be R8G8B8A8_UNORM.
981     */
982    struct intel_context *intel = &brw->intel;
983    struct gl_context *ctx = &intel->ctx;
984    uint32_t *surf;
985    unsigned surface_type = BRW_SURFACE_NULL;
986    drm_intel_bo *bo = NULL;
987    unsigned pitch_minus_1 = 0;
988    uint32_t multisampling_state = 0;
989
990    /* _NEW_BUFFERS */
991    const struct gl_framebuffer *fb = ctx->DrawBuffer;
992
993    surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
994                           6 * 4, 32, &brw->wm.surf_offset[unit]);
995
996    if (fb->Visual.samples > 1) {
997       /* On Gen6, null render targets seem to cause GPU hangs when
998        * multisampling.  So work around this problem by rendering into dummy
999        * color buffer.
1000        *
1001        * To decrease the amount of memory needed by the workaround buffer, we
1002        * set its pitch to 128 bytes (the width of a Y tile).  This means that
1003        * the amount of memory needed for the workaround buffer is
1004        * (width_in_tiles + height_in_tiles - 1) tiles.
1005        *
1006        * Note that since the workaround buffer will be interpreted by the
1007        * hardware as an interleaved multisampled buffer, we need to compute
1008        * width_in_tiles and height_in_tiles by dividing the width and height
1009        * by 16 rather than the normal Y-tile size of 32.
1010        */
1011       unsigned width_in_tiles = ALIGN(fb->Width, 16) / 16;
1012       unsigned height_in_tiles = ALIGN(fb->Height, 16) / 16;
1013       unsigned size_needed = (width_in_tiles + height_in_tiles - 1) * 4096;
1014       brw_get_scratch_bo(intel, &brw->wm.multisampled_null_render_target_bo,
1015                          size_needed);
1016       bo = brw->wm.multisampled_null_render_target_bo;
1017       surface_type = BRW_SURFACE_2D;
1018       pitch_minus_1 = 127;
1019       multisampling_state =
1020          brw_get_surface_num_multisamples(fb->Visual.samples);
1021    }
1022
1023    surf[0] = (surface_type << BRW_SURFACE_TYPE_SHIFT |
1024               BRW_SURFACEFORMAT_B8G8R8A8_UNORM << BRW_SURFACE_FORMAT_SHIFT);
1025    if (intel->gen < 6) {
1026       surf[0] |= (1 << BRW_SURFACE_WRITEDISABLE_R_SHIFT |
1027                   1 << BRW_SURFACE_WRITEDISABLE_G_SHIFT |
1028                   1 << BRW_SURFACE_WRITEDISABLE_B_SHIFT |
1029                   1 << BRW_SURFACE_WRITEDISABLE_A_SHIFT);
1030    }
1031    surf[1] = bo ? bo->offset : 0;
1032    surf[2] = ((fb->Width - 1) << BRW_SURFACE_WIDTH_SHIFT |
1033               (fb->Height - 1) << BRW_SURFACE_HEIGHT_SHIFT);
1034
1035    /* From Sandy bridge PRM, Vol4 Part1 p82 (Tiled Surface: Programming
1036     * Notes):
1037     *
1038     *     If Surface Type is SURFTYPE_NULL, this field must be TRUE
1039     */
1040    surf[3] = (BRW_SURFACE_TILED | BRW_SURFACE_TILED_Y |
1041               pitch_minus_1 << BRW_SURFACE_PITCH_SHIFT);
1042    surf[4] = multisampling_state;
1043    surf[5] = 0;
1044
1045    if (bo) {
1046       drm_intel_bo_emit_reloc(brw->intel.batch.bo,
1047                               brw->wm.surf_offset[unit] + 4,
1048                               bo, 0,
1049                               I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER);
1050    }
1051 }
1052
1053 /**
1054  * Sets up a surface state structure to point at the given region.
1055  * While it is only used for the front/back buffer currently, it should be
1056  * usable for further buffers when doing ARB_draw_buffer support.
1057  */
1058 static void
1059 brw_update_renderbuffer_surface(struct brw_context *brw,
1060                                 struct gl_renderbuffer *rb,
1061                                 unsigned int unit)
1062 {
1063    struct intel_context *intel = &brw->intel;
1064    struct gl_context *ctx = &intel->ctx;
1065    struct intel_renderbuffer *irb = intel_renderbuffer(rb);
1066    struct intel_mipmap_tree *mt = irb->mt;
1067    struct intel_region *region;
1068    uint32_t *surf;
1069    uint32_t tile_x, tile_y;
1070    uint32_t format = 0;
1071    gl_format rb_format = intel_rb_format(irb);
1072
1073    if (irb->tex_image && !brw->has_surface_tile_offset) {
1074       intel_renderbuffer_tile_offsets(irb, &tile_x, &tile_y);
1075
1076       if (tile_x != 0 || tile_y != 0) {
1077          /* Original gen4 hardware couldn't draw to a non-tile-aligned
1078           * destination in a miptree unless you actually setup your renderbuffer
1079           * as a miptree and used the fragile lod/array_index/etc. controls to
1080           * select the image.  So, instead, we just make a new single-level
1081           * miptree and render into that.
1082           */
1083          struct intel_context *intel = intel_context(ctx);
1084          struct intel_texture_image *intel_image =
1085             intel_texture_image(irb->tex_image);
1086          struct intel_mipmap_tree *new_mt;
1087          int width, height, depth;
1088
1089          intel_miptree_get_dimensions_for_image(irb->tex_image, &width, &height, &depth);
1090
1091          new_mt = intel_miptree_create(intel, irb->tex_image->TexObject->Target,
1092                                        intel_image->base.Base.TexFormat,
1093                                        intel_image->base.Base.Level,
1094                                        intel_image->base.Base.Level,
1095                                        width, height, depth,
1096                                        true,
1097                                        0 /* num_samples */,
1098                                        INTEL_MSAA_LAYOUT_NONE);
1099
1100          intel_miptree_copy_teximage(intel, intel_image, new_mt);
1101          intel_miptree_reference(&irb->mt, intel_image->mt);
1102          intel_renderbuffer_set_draw_offset(irb);
1103          intel_miptree_release(&new_mt);
1104
1105          mt = irb->mt;
1106       }
1107    }
1108
1109    region = irb->mt->region;
1110
1111    surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
1112                           6 * 4, 32, &brw->wm.surf_offset[unit]);
1113
1114    switch (rb_format) {
1115    case MESA_FORMAT_SARGB8:
1116       /* _NEW_BUFFERS
1117        *
1118        * Without GL_EXT_framebuffer_sRGB we shouldn't bind sRGB surfaces to the
1119        * blend/update as sRGB.
1120        */
1121       if (ctx->Color.sRGBEnabled)
1122          format = brw_format_for_mesa_format(rb_format);
1123       else
1124          format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
1125       break;
1126    default:
1127       format = brw->render_target_format[rb_format];
1128       if (unlikely(!brw->format_supported_as_render_target[rb_format])) {
1129          _mesa_problem(ctx, "%s: renderbuffer format %s unsupported\n",
1130                        __FUNCTION__, _mesa_get_format_name(rb_format));
1131       }
1132       break;
1133    }
1134
1135    surf[0] = (BRW_SURFACE_2D << BRW_SURFACE_TYPE_SHIFT |
1136               format << BRW_SURFACE_FORMAT_SHIFT);
1137
1138    /* reloc */
1139    surf[1] = (intel_renderbuffer_tile_offsets(irb, &tile_x, &tile_y) +
1140               region->bo->offset);
1141
1142    surf[2] = ((rb->Width - 1) << BRW_SURFACE_WIDTH_SHIFT |
1143               (rb->Height - 1) << BRW_SURFACE_HEIGHT_SHIFT);
1144
1145    surf[3] = (brw_get_surface_tiling_bits(region->tiling) |
1146               ((region->pitch * region->cpp) - 1) << BRW_SURFACE_PITCH_SHIFT);
1147
1148    surf[4] = brw_get_surface_num_multisamples(mt->num_samples);
1149
1150    assert(brw->has_surface_tile_offset || (tile_x == 0 && tile_y == 0));
1151    /* Note that the low bits of these fields are missing, so
1152     * there's the possibility of getting in trouble.
1153     */
1154    assert(tile_x % 4 == 0);
1155    assert(tile_y % 2 == 0);
1156    surf[5] = ((tile_x / 4) << BRW_SURFACE_X_OFFSET_SHIFT |
1157               (tile_y / 2) << BRW_SURFACE_Y_OFFSET_SHIFT |
1158               (mt->align_h == 4 ? BRW_SURFACE_VERTICAL_ALIGN_ENABLE : 0));
1159
1160    if (intel->gen < 6) {
1161       /* _NEW_COLOR */
1162       if (!ctx->Color.ColorLogicOpEnabled &&
1163           (ctx->Color.BlendEnabled & (1 << unit)))
1164          surf[0] |= BRW_SURFACE_BLEND_ENABLED;
1165
1166       if (!ctx->Color.ColorMask[unit][0])
1167          surf[0] |= 1 << BRW_SURFACE_WRITEDISABLE_R_SHIFT;
1168       if (!ctx->Color.ColorMask[unit][1])
1169          surf[0] |= 1 << BRW_SURFACE_WRITEDISABLE_G_SHIFT;
1170       if (!ctx->Color.ColorMask[unit][2])
1171          surf[0] |= 1 << BRW_SURFACE_WRITEDISABLE_B_SHIFT;
1172
1173       /* As mentioned above, disable writes to the alpha component when the
1174        * renderbuffer is XRGB.
1175        */
1176       if (ctx->DrawBuffer->Visual.alphaBits == 0 ||
1177           !ctx->Color.ColorMask[unit][3]) {
1178          surf[0] |= 1 << BRW_SURFACE_WRITEDISABLE_A_SHIFT;
1179       }
1180    }
1181
1182    drm_intel_bo_emit_reloc(brw->intel.batch.bo,
1183                            brw->wm.surf_offset[unit] + 4,
1184                            region->bo,
1185                            surf[1] - region->bo->offset,
1186                            I915_GEM_DOMAIN_RENDER,
1187                            I915_GEM_DOMAIN_RENDER);
1188 }
1189
1190 /**
1191  * Construct SURFACE_STATE objects for renderbuffers/draw buffers.
1192  */
1193 static void
1194 brw_update_renderbuffer_surfaces(struct brw_context *brw)
1195 {
1196    struct intel_context *intel = &brw->intel;
1197    struct gl_context *ctx = &brw->intel.ctx;
1198    GLuint i;
1199
1200    /* _NEW_BUFFERS | _NEW_COLOR */
1201    /* Update surfaces for drawing buffers */
1202    if (ctx->DrawBuffer->_NumColorDrawBuffers >= 1) {
1203       for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) {
1204          if (intel_renderbuffer(ctx->DrawBuffer->_ColorDrawBuffers[i])) {
1205             intel->vtbl.update_renderbuffer_surface(brw, ctx->DrawBuffer->_ColorDrawBuffers[i], i);
1206          } else {
1207             intel->vtbl.update_null_renderbuffer_surface(brw, i);
1208          }
1209       }
1210    } else {
1211       intel->vtbl.update_null_renderbuffer_surface(brw, 0);
1212    }
1213    brw->state.dirty.brw |= BRW_NEW_SURFACES;
1214 }
1215
1216 const struct brw_tracked_state brw_renderbuffer_surfaces = {
1217    .dirty = {
1218       .mesa = (_NEW_COLOR |
1219                _NEW_BUFFERS),
1220       .brw = BRW_NEW_BATCH,
1221       .cache = 0
1222    },
1223    .emit = brw_update_renderbuffer_surfaces,
1224 };
1225
1226 const struct brw_tracked_state gen6_renderbuffer_surfaces = {
1227    .dirty = {
1228       .mesa = _NEW_BUFFERS,
1229       .brw = BRW_NEW_BATCH,
1230       .cache = 0
1231    },
1232    .emit = brw_update_renderbuffer_surfaces,
1233 };
1234
1235 /**
1236  * Construct SURFACE_STATE objects for enabled textures.
1237  */
1238 static void
1239 brw_update_texture_surfaces(struct brw_context *brw)
1240 {
1241    struct intel_context *intel = &brw->intel;
1242    struct gl_context *ctx = &intel->ctx;
1243
1244    /* BRW_NEW_VERTEX_PROGRAM and BRW_NEW_FRAGMENT_PROGRAM:
1245     * Unfortunately, we're stuck using the gl_program structs until the
1246     * ARB_fragment_program front-end gets converted to GLSL IR.  These
1247     * have the downside that SamplerUnits is split and only contains the
1248     * mappings for samplers active in that stage.
1249     */
1250    struct gl_program *vs = (struct gl_program *) brw->vertex_program;
1251    struct gl_program *fs = (struct gl_program *) brw->fragment_program;
1252
1253    unsigned num_samplers = _mesa_fls(vs->SamplersUsed | fs->SamplersUsed);
1254
1255    for (unsigned s = 0; s < num_samplers; s++) {
1256       brw->vs.surf_offset[SURF_INDEX_VS_TEXTURE(s)] = 0;
1257       brw->wm.surf_offset[SURF_INDEX_TEXTURE(s)] = 0;
1258
1259       if (vs->SamplersUsed & (1 << s)) {
1260          const unsigned unit = vs->SamplerUnits[s];
1261
1262          /* _NEW_TEXTURE */
1263          if (ctx->Texture.Unit[unit]._ReallyEnabled) {
1264             intel->vtbl.update_texture_surface(ctx, unit,
1265                                                brw->vs.surf_offset,
1266                                                SURF_INDEX_VS_TEXTURE(s));
1267          }
1268       }
1269
1270       if (fs->SamplersUsed & (1 << s)) {
1271          const unsigned unit = fs->SamplerUnits[s];
1272
1273          /* _NEW_TEXTURE */
1274          if (ctx->Texture.Unit[unit]._ReallyEnabled) {
1275             intel->vtbl.update_texture_surface(ctx, unit,
1276                                                brw->wm.surf_offset,
1277                                                SURF_INDEX_TEXTURE(s));
1278          }
1279       }
1280    }
1281
1282    brw->state.dirty.brw |= BRW_NEW_SURFACES;
1283 }
1284
1285 const struct brw_tracked_state brw_texture_surfaces = {
1286    .dirty = {
1287       .mesa = _NEW_TEXTURE,
1288       .brw = BRW_NEW_BATCH |
1289              BRW_NEW_VERTEX_PROGRAM |
1290              BRW_NEW_FRAGMENT_PROGRAM,
1291       .cache = 0
1292    },
1293    .emit = brw_update_texture_surfaces,
1294 };
1295
1296 void
1297 brw_upload_ubo_surfaces(struct brw_context *brw,
1298                         struct gl_shader *shader,
1299                         uint32_t *surf_offsets)
1300 {
1301    struct gl_context *ctx = &brw->intel.ctx;
1302    struct intel_context *intel = &brw->intel;
1303
1304    if (!shader)
1305       return;
1306
1307    for (int i = 0; i < shader->NumUniformBlocks; i++) {
1308       struct gl_uniform_buffer_binding *binding;
1309       struct intel_buffer_object *intel_bo;
1310
1311       binding = &ctx->UniformBufferBindings[shader->UniformBlocks[i].Binding];
1312       intel_bo = intel_buffer_object(binding->BufferObject);
1313       drm_intel_bo *bo = intel_bufferobj_buffer(intel, intel_bo, INTEL_READ);
1314
1315       /* Because behavior for referencing outside of the binding's size in the
1316        * glBindBufferRange case is undefined, we can just bind the whole buffer
1317        * glBindBufferBase wants and be a correct implementation.
1318        */
1319       int size = bo->size - binding->Offset;
1320       size = ALIGN(size, 16) / 16; /* The interface takes a number of vec4s */
1321
1322       intel->vtbl.create_constant_surface(brw, bo, binding->Offset,
1323                                           size,
1324                                           &surf_offsets[i]);
1325    }
1326
1327    if (shader->NumUniformBlocks)
1328       brw->state.dirty.brw |= BRW_NEW_SURFACES;
1329 }
1330
1331 static void
1332 brw_upload_wm_ubo_surfaces(struct brw_context *brw)
1333 {
1334    struct gl_context *ctx = &brw->intel.ctx;
1335    /* _NEW_PROGRAM */
1336    struct gl_shader_program *prog = ctx->Shader._CurrentFragmentProgram;
1337
1338    if (!prog)
1339       return;
1340
1341    brw_upload_ubo_surfaces(brw, prog->_LinkedShaders[MESA_SHADER_FRAGMENT],
1342                            &brw->wm.surf_offset[SURF_INDEX_WM_UBO(0)]);
1343 }
1344
1345 const struct brw_tracked_state brw_wm_ubo_surfaces = {
1346    .dirty = {
1347       .mesa = (_NEW_PROGRAM |
1348                _NEW_BUFFER_OBJECT),
1349       .brw = BRW_NEW_BATCH,
1350       .cache = 0,
1351    },
1352    .emit = brw_upload_wm_ubo_surfaces,
1353 };
1354
1355 /**
1356  * Constructs the binding table for the WM surface state, which maps unit
1357  * numbers to surface state objects.
1358  */
1359 static void
1360 brw_upload_wm_binding_table(struct brw_context *brw)
1361 {
1362    uint32_t *bind;
1363    int i;
1364
1365    /* Might want to calculate nr_surfaces first, to avoid taking up so much
1366     * space for the binding table.
1367     */
1368    bind = brw_state_batch(brw, AUB_TRACE_BINDING_TABLE,
1369                           sizeof(uint32_t) * BRW_MAX_WM_SURFACES,
1370                           32, &brw->wm.bind_bo_offset);
1371
1372    /* BRW_NEW_SURFACES */
1373    for (i = 0; i < BRW_MAX_WM_SURFACES; i++) {
1374       bind[i] = brw->wm.surf_offset[i];
1375    }
1376
1377    brw->state.dirty.brw |= BRW_NEW_PS_BINDING_TABLE;
1378 }
1379
1380 const struct brw_tracked_state brw_wm_binding_table = {
1381    .dirty = {
1382       .mesa = 0,
1383       .brw = (BRW_NEW_BATCH |
1384               BRW_NEW_SURFACES),
1385       .cache = 0
1386    },
1387    .emit = brw_upload_wm_binding_table,
1388 };
1389
1390 void
1391 gen4_init_vtable_surface_functions(struct brw_context *brw)
1392 {
1393    struct intel_context *intel = &brw->intel;
1394
1395    intel->vtbl.update_texture_surface = brw_update_texture_surface;
1396    intel->vtbl.update_renderbuffer_surface = brw_update_renderbuffer_surface;
1397    intel->vtbl.update_null_renderbuffer_surface =
1398       brw_update_null_renderbuffer_surface;
1399    intel->vtbl.create_constant_surface = brw_create_constant_surface;
1400 }