2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
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:
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.
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.
26 **********************************************************************/
29 * Keith Whitwell <keith@tungstengraphics.com>
34 #include "intel_fbo.h"
35 #include "brw_context.h"
36 #include "brw_state.h"
37 #include "brw_defines.h"
40 /***********************************************************************
41 * WM unit - fragment programs and rasterization
45 brw_color_buffer_write_enabled(struct brw_context
*brw
)
47 struct gl_context
*ctx
= &brw
->intel
.ctx
;
48 const struct gl_fragment_program
*fp
= brw
->fragment_program
;
52 for (i
= 0; i
< ctx
->DrawBuffer
->_NumColorDrawBuffers
; i
++) {
53 struct gl_renderbuffer
*rb
= ctx
->DrawBuffer
->_ColorDrawBuffers
[i
];
57 (fp
->Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_COLOR
) ||
58 fp
->Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DATA0
+ i
)) &&
59 (ctx
->Color
.ColorMask
[i
][0] ||
60 ctx
->Color
.ColorMask
[i
][1] ||
61 ctx
->Color
.ColorMask
[i
][2] ||
62 ctx
->Color
.ColorMask
[i
][3])) {
71 * Setup wm hardware state. See page 225 of Volume 2
74 brw_prepare_wm_unit(struct brw_context
*brw
)
76 struct intel_context
*intel
= &brw
->intel
;
77 struct gl_context
*ctx
= &intel
->ctx
;
78 const struct gl_fragment_program
*fp
= brw
->fragment_program
;
79 struct brw_wm_unit_state
*wm
;
81 wm
= brw_state_batch(brw
, sizeof(*wm
), 32, &brw
->wm
.state_offset
);
82 memset(wm
, 0, sizeof(*wm
));
84 if (brw
->wm
.prog_data
->prog_offset_16
) {
85 /* These two fields should be the same pre-gen6, which is why we
86 * only have one hardware field to program for both dispatch
89 assert(brw
->wm
.prog_data
->first_curbe_grf
==
90 brw
->wm
.prog_data
->first_curbe_grf_16
);
93 /* CACHE_NEW_WM_PROG */
94 wm
->thread0
.grf_reg_count
= brw
->wm
.prog_data
->reg_blocks
;
95 wm
->wm9
.grf_reg_count_2
= brw
->wm
.prog_data
->reg_blocks_16
;
96 wm
->thread0
.kernel_start_pointer
= brw
->wm
.prog_bo
->offset
>> 6; /* reloc */
98 wm
->wm9
.kernel_start_pointer_2
= (brw
->wm
.prog_bo
->offset
+
99 brw
->wm
.prog_data
->prog_offset_16
) >> 6;
100 wm
->thread1
.depth_coef_urb_read_offset
= 1;
101 wm
->thread1
.floating_point_mode
= BRW_FLOATING_POINT_NON_IEEE_754
;
104 wm
->thread1
.binding_table_entry_count
= 0; /* hardware requirement */
106 /* BRW_NEW_NR_SURFACES */
107 wm
->thread1
.binding_table_entry_count
= brw
->wm
.nr_surfaces
;
110 if (brw
->wm
.prog_data
->total_scratch
!= 0) {
111 wm
->thread2
.scratch_space_base_pointer
=
112 brw
->wm
.scratch_bo
->offset
>> 10; /* reloc */
113 wm
->thread2
.per_thread_scratch_space
=
114 ffs(brw
->wm
.prog_data
->total_scratch
) - 11;
116 wm
->thread2
.scratch_space_base_pointer
= 0;
117 wm
->thread2
.per_thread_scratch_space
= 0;
120 wm
->thread3
.dispatch_grf_start_reg
= brw
->wm
.prog_data
->first_curbe_grf
;
121 wm
->thread3
.urb_entry_read_length
= brw
->wm
.prog_data
->urb_read_length
;
122 wm
->thread3
.urb_entry_read_offset
= 0;
123 wm
->thread3
.const_urb_entry_read_length
=
124 brw
->wm
.prog_data
->curb_read_length
;
125 /* BRW_NEW_CURBE_OFFSETS */
126 wm
->thread3
.const_urb_entry_read_offset
= brw
->curbe
.wm_start
* 2;
129 wm
->wm4
.sampler_count
= 0; /* hardware requirement */
131 /* CACHE_NEW_SAMPLER */
132 wm
->wm4
.sampler_count
= (brw
->wm
.sampler_count
+ 1) / 4;
135 if (brw
->wm
.sampler_count
) {
137 wm
->wm4
.sampler_state_pointer
= (intel
->batch
.bo
->offset
+
138 brw
->wm
.sampler_offset
) >> 5;
140 wm
->wm4
.sampler_state_pointer
= 0;
143 /* BRW_NEW_FRAGMENT_PROGRAM */
144 wm
->wm5
.program_uses_depth
= (fp
->Base
.InputsRead
&
145 (1 << FRAG_ATTRIB_WPOS
)) != 0;
146 wm
->wm5
.program_computes_depth
= (fp
->Base
.OutputsWritten
&
147 BITFIELD64_BIT(FRAG_RESULT_DEPTH
)) != 0;
149 * Override for NULL depthbuffer case, required by the Pixel Shader Computed
152 if (!intel_get_renderbuffer(ctx
->DrawBuffer
, BUFFER_DEPTH
))
153 wm
->wm5
.program_computes_depth
= 0;
156 wm
->wm5
.program_uses_killpixel
= fp
->UsesKill
|| ctx
->Color
.AlphaEnabled
;
159 /* BRW_NEW_FRAGMENT_PROGRAM
161 * If using the fragment shader backend, the program is always
162 * 8-wide. If not, it's always 16.
164 if (ctx
->Shader
.CurrentFragmentProgram
) {
165 struct brw_shader
*shader
= (struct brw_shader
*)
166 ctx
->Shader
.CurrentFragmentProgram
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
168 if (shader
!= NULL
&& shader
->ir
!= NULL
) {
169 wm
->wm5
.enable_8_pix
= 1;
170 if (brw
->wm
.prog_data
->prog_offset_16
)
171 wm
->wm5
.enable_16_pix
= 1;
174 if (!wm
->wm5
.enable_8_pix
)
175 wm
->wm5
.enable_16_pix
= 1;
177 wm
->wm5
.max_threads
= brw
->wm_max_threads
- 1;
179 /* _NEW_BUFFERS | _NEW_COLOR */
180 if (brw_color_buffer_write_enabled(brw
) ||
181 wm
->wm5
.program_uses_killpixel
||
182 wm
->wm5
.program_computes_depth
) {
183 wm
->wm5
.thread_dispatch_enable
= 1;
186 wm
->wm5
.legacy_line_rast
= 0;
187 wm
->wm5
.legacy_global_depth_bias
= 0;
188 wm
->wm5
.early_depth_test
= 1; /* never need to disable */
189 wm
->wm5
.line_aa_region_width
= 0;
190 wm
->wm5
.line_endcap_aa_region_width
= 1;
192 /* _NEW_POLYGONSTIPPLE */
193 wm
->wm5
.polygon_stipple
= ctx
->Polygon
.StippleFlag
;
196 if (ctx
->Polygon
.OffsetFill
) {
197 wm
->wm5
.depth_offset
= 1;
198 /* Something wierd going on with legacy_global_depth_bias,
199 * offset_constant, scaling and MRD. This value passes glean
200 * but gives some odd results elsewere (eg. the
201 * quad-offset-units test).
203 wm
->global_depth_offset_constant
= ctx
->Polygon
.OffsetUnits
* 2;
205 /* This is the only value that passes glean:
207 wm
->global_depth_offset_scale
= ctx
->Polygon
.OffsetFactor
;
211 wm
->wm5
.line_stipple
= ctx
->Line
.StippleFlag
;
214 if (unlikely(INTEL_DEBUG
& DEBUG_STATS
) || intel
->stats_wm
)
215 wm
->wm4
.stats_enable
= 1;
217 /* Emit WM program relocation */
218 drm_intel_bo_emit_reloc(intel
->batch
.bo
,
219 brw
->wm
.state_offset
+
220 offsetof(struct brw_wm_unit_state
, thread0
),
221 brw
->wm
.prog_bo
, wm
->thread0
.grf_reg_count
<< 1,
222 I915_GEM_DOMAIN_INSTRUCTION
, 0);
224 if (brw
->wm
.prog_data
->prog_offset_16
) {
225 drm_intel_bo_emit_reloc(intel
->batch
.bo
,
226 brw
->wm
.state_offset
+
227 offsetof(struct brw_wm_unit_state
, wm9
),
229 ((wm
->wm9
.grf_reg_count_2
<< 1) +
230 brw
->wm
.prog_data
->prog_offset_16
),
231 I915_GEM_DOMAIN_INSTRUCTION
, 0);
234 /* Emit scratch space relocation */
235 if (brw
->wm
.prog_data
->total_scratch
!= 0) {
236 drm_intel_bo_emit_reloc(intel
->batch
.bo
,
237 brw
->wm
.state_offset
+
238 offsetof(struct brw_wm_unit_state
, thread2
),
240 wm
->thread2
.per_thread_scratch_space
,
241 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
);
244 /* Emit sampler state relocation */
245 if (brw
->wm
.sampler_count
!= 0) {
246 drm_intel_bo_emit_reloc(intel
->batch
.bo
,
247 brw
->wm
.state_offset
+
248 offsetof(struct brw_wm_unit_state
, wm4
),
249 intel
->batch
.bo
, (brw
->wm
.sampler_offset
|
250 wm
->wm4
.stats_enable
|
251 (wm
->wm4
.sampler_count
<< 2)),
252 I915_GEM_DOMAIN_INSTRUCTION
, 0);
255 brw
->state
.dirty
.cache
|= CACHE_NEW_WM_UNIT
;
258 const struct brw_tracked_state brw_wm_unit
= {
260 .mesa
= (_NEW_POLYGON
|
261 _NEW_POLYGONSTIPPLE
|
267 .brw
= (BRW_NEW_BATCH
|
268 BRW_NEW_FRAGMENT_PROGRAM
|
269 BRW_NEW_CURBE_OFFSETS
|
270 BRW_NEW_NR_WM_SURFACES
),
272 .cache
= (CACHE_NEW_WM_PROG
|
275 .prepare
= brw_prepare_wm_unit
,