1 /**********************************************************
2 * Copyright 2008-2009 VMware, Inc. All rights reserved.
4 * Permission is hereby granted, free of charge, to any person
5 * obtaining a copy of this software and associated documentation
6 * files (the "Software"), to deal in the Software without
7 * restriction, including without limitation the rights to use, copy,
8 * modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
24 **********************************************************/
26 #include "util/u_inlines.h"
27 #include "pipe/p_defines.h"
28 #include "util/u_math.h"
29 #include "util/u_memory.h"
30 #include "util/u_bitmask.h"
31 #include "tgsi/tgsi_ureg.h"
33 #include "svga_context.h"
34 #include "svga_state.h"
36 #include "svga_resource_texture.h"
37 #include "svga_tgsi.h"
39 #include "svga_hw_reg.h"
43 static INLINE
int compare_fs_keys( const struct svga_fs_compile_key
*a
,
44 const struct svga_fs_compile_key
*b
)
46 unsigned keysize_a
= svga_fs_key_size( a
);
47 unsigned keysize_b
= svga_fs_key_size( b
);
49 if (keysize_a
!= keysize_b
) {
50 return (int)(keysize_a
- keysize_b
);
52 return memcmp( a
, b
, keysize_a
);
56 static struct svga_shader_result
*search_fs_key( struct svga_fragment_shader
*fs
,
57 const struct svga_fs_compile_key
*key
)
59 struct svga_shader_result
*result
= fs
->base
.results
;
63 for ( ; result
; result
= result
->next
) {
64 if (compare_fs_keys( key
, &result
->key
.fkey
) == 0)
73 * If we fail to compile a fragment shader (because it uses too many
74 * registers, for example) we'll use a dummy/fallback shader that
75 * simply emits a constant color.
77 static const struct tgsi_token
*
78 get_dummy_fragment_shader(void)
80 static const float red
[4] = { 1.0, 0.0, 0.0, 0.0 };
81 struct ureg_program
*ureg
;
82 const struct tgsi_token
*tokens
;
87 ureg
= ureg_create(TGSI_PROCESSOR_FRAGMENT
);
91 dst
= ureg_DECL_output(ureg
, TGSI_SEMANTIC_COLOR
, 0);
92 src
= ureg_DECL_immediate(ureg
, red
, 4);
93 ureg_MOV(ureg
, dst
, src
);
96 tokens
= ureg_get_tokens(ureg
, &num_tokens
);
104 static enum pipe_error
compile_fs( struct svga_context
*svga
,
105 struct svga_fragment_shader
*fs
,
106 const struct svga_fs_compile_key
*key
,
107 struct svga_shader_result
**out_result
)
109 struct svga_shader_result
*result
;
110 enum pipe_error ret
= PIPE_ERROR
;
112 result
= svga_translate_fragment_program( fs
, key
);
113 if (result
== NULL
) {
114 /* some problem during translation, try the dummy shader */
115 const struct tgsi_token
*dummy
= get_dummy_fragment_shader();
117 ret
= PIPE_ERROR_OUT_OF_MEMORY
;
120 debug_printf("Failed to compile fragment shader, using dummy shader instead.\n");
121 FREE((void *) fs
->base
.tokens
);
122 fs
->base
.tokens
= dummy
;
123 result
= svga_translate_fragment_program(fs
, key
);
124 if (result
== NULL
) {
130 result
->id
= util_bitmask_add(svga
->fs_bm
);
131 if(result
->id
== UTIL_BITMASK_INVALID_INDEX
) {
132 ret
= PIPE_ERROR_OUT_OF_MEMORY
;
136 ret
= SVGA3D_DefineShader(svga
->swc
,
138 SVGA3D_SHADERTYPE_PS
,
140 result
->nr_tokens
* sizeof result
->tokens
[0]);
144 *out_result
= result
;
145 result
->next
= fs
->base
.results
;
146 fs
->base
.results
= result
;
151 if (result
->id
!= UTIL_BITMASK_INVALID_INDEX
)
152 util_bitmask_clear( svga
->fs_bm
, result
->id
);
153 svga_destroy_shader_result( result
);
159 /* SVGA_NEW_TEXTURE_BINDING
161 * SVGA_NEW_NEED_SWTNL
164 static enum pipe_error
165 make_fs_key(const struct svga_context
*svga
,
166 struct svga_fragment_shader
*fs
,
167 struct svga_fs_compile_key
*key
)
172 memset(key
, 0, sizeof *key
);
174 /* Only need fragment shader fixup for twoside lighting if doing
175 * hwtnl. Otherwise the draw module does the whole job for us.
179 if (!svga
->state
.sw
.need_swtnl
) {
182 key
->light_twoside
= svga
->curr
.rast
->templ
.light_twoside
;
183 key
->front_ccw
= svga
->curr
.rast
->templ
.front_ccw
;
186 /* The blend workaround for simulating logicop xor behaviour
187 * requires that the incoming fragment color be white. This change
188 * achieves that by creating a variant of the current fragment
189 * shader that overrides all output colors with 1,1,1,1
191 * This will work for most shaders, including those containing
192 * TEXKIL and/or depth-write. However, it will break on the
193 * combination of xor-logicop plus alphatest.
195 * Ultimately, we could implement alphatest in the shader using
196 * texkil prior to overriding the outgoing fragment color.
200 if (svga
->curr
.blend
->need_white_fragments
) {
201 key
->white_fragments
= 1;
204 /* XXX: want to limit this to the textures that the shader actually
207 * SVGA_NEW_TEXTURE_BINDING | SVGA_NEW_SAMPLER
209 for (i
= 0; i
< svga
->curr
.num_sampler_views
; i
++) {
210 if (svga
->curr
.sampler_views
[i
]) {
211 assert(svga
->curr
.sampler
[i
]);
212 assert(svga
->curr
.sampler_views
[i
]->texture
);
213 key
->tex
[i
].texture_target
= svga
->curr
.sampler_views
[i
]->texture
->target
;
214 if (!svga
->curr
.sampler
[i
]->normalized_coords
) {
215 key
->tex
[i
].width_height_idx
= idx
++;
216 key
->tex
[i
].unnormalized
= TRUE
;
217 ++key
->num_unnormalized_coords
;
220 key
->tex
[i
].swizzle_r
= svga
->curr
.sampler_views
[i
]->swizzle_r
;
221 key
->tex
[i
].swizzle_g
= svga
->curr
.sampler_views
[i
]->swizzle_g
;
222 key
->tex
[i
].swizzle_b
= svga
->curr
.sampler_views
[i
]->swizzle_b
;
223 key
->tex
[i
].swizzle_a
= svga
->curr
.sampler_views
[i
]->swizzle_a
;
226 key
->num_textures
= svga
->curr
.num_sampler_views
;
229 for (i
= 0; i
< svga
->curr
.num_samplers
; ++i
) {
230 if (svga
->curr
.sampler_views
[i
]) {
231 struct pipe_resource
*tex
= svga
->curr
.sampler_views
[i
]->texture
;
232 struct svga_texture
*stex
= svga_texture(tex
);
233 SVGA3dSurfaceFormat format
= stex
->key
.format
;
235 if (format
== SVGA3D_Z_D16
||
236 format
== SVGA3D_Z_D24X8
||
237 format
== SVGA3D_Z_D24S8
) {
238 /* If we're sampling from a SVGA3D_Z_D16, SVGA3D_Z_D24X8,
239 * or SVGA3D_Z_D24S8 surface, we'll automatically get
240 * shadow comparison. But we only get LEQUAL mode.
241 * Set TEX_COMPARE_NONE here so we don't emit the extra FS
242 * code for shadow comparison.
244 key
->tex
[i
].compare_mode
= PIPE_TEX_COMPARE_NONE
;
245 key
->tex
[i
].compare_func
= PIPE_FUNC_NEVER
;
246 /* These depth formats _only_ support comparison mode and
247 * not ordinary sampling so warn if the later is expected.
249 if (svga
->curr
.sampler
[i
]->compare_mode
!=
250 PIPE_TEX_COMPARE_R_TO_TEXTURE
) {
251 debug_warn_once("Unsupported shadow compare mode");
253 /* The only supported comparison mode is LEQUAL */
254 if (svga
->curr
.sampler
[i
]->compare_func
!= PIPE_FUNC_LEQUAL
) {
255 debug_warn_once("Unsupported shadow compare function");
259 /* For other texture formats, just use the compare func/mode
260 * as-is. Should be no-ops for color textures. For depth
261 * textures, we do not get automatic depth compare. We have
262 * to do it ourselves in the shader. And we don't get PCF.
264 key
->tex
[i
].compare_mode
= svga
->curr
.sampler
[i
]->compare_mode
;
265 key
->tex
[i
].compare_func
= svga
->curr
.sampler
[i
]->compare_func
;
270 /* sprite coord gen state */
271 for (i
= 0; i
< svga
->curr
.num_samplers
; ++i
) {
272 key
->tex
[i
].sprite_texgen
=
273 svga
->curr
.rast
->templ
.sprite_coord_enable
& (1 << i
);
276 key
->sprite_origin_lower_left
= (svga
->curr
.rast
->templ
.sprite_coord_mode
277 == PIPE_SPRITE_COORD_LOWER_LEFT
);
279 /* SVGA_NEW_FRAME_BUFFER */
280 if (fs
->base
.info
.color0_writes_all_cbufs
) {
281 /* Replicate color0 output to N colorbuffers */
282 key
->write_color0_to_n_cbufs
= svga
->curr
.framebuffer
.nr_cbufs
;
290 static enum pipe_error
291 emit_hw_fs(struct svga_context
*svga
, unsigned dirty
)
293 struct svga_shader_result
*result
= NULL
;
294 unsigned id
= SVGA3D_INVALID_ID
;
295 enum pipe_error ret
= PIPE_OK
;
297 struct svga_fragment_shader
*fs
= svga
->curr
.fs
;
298 struct svga_fs_compile_key key
;
301 * SVGA_NEW_TEXTURE_BINDING
303 * SVGA_NEW_NEED_SWTNL
305 * SVGA_NEW_FRAME_BUFFER
307 ret
= make_fs_key( svga
, fs
, &key
);
311 result
= search_fs_key( fs
, &key
);
313 ret
= compile_fs( svga
, fs
, &key
, &result
);
321 assert(id
!= SVGA3D_INVALID_ID
);
323 if (result
!= svga
->state
.hw_draw
.fs
) {
324 ret
= SVGA3D_SetShader(svga
->swc
,
325 SVGA3D_SHADERTYPE_PS
,
330 svga
->dirty
|= SVGA_NEW_FS_RESULT
;
331 svga
->state
.hw_draw
.fs
= result
;
337 struct svga_tracked_state svga_hw_fs
=
339 "fragment shader (hwtnl)",
341 SVGA_NEW_TEXTURE_BINDING
|
342 SVGA_NEW_NEED_SWTNL
|
345 SVGA_NEW_FRAME_BUFFER
|