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_tgsi.h"
38 #include "svga_hw_reg.h"
42 static INLINE
int compare_fs_keys( const struct svga_fs_compile_key
*a
,
43 const struct svga_fs_compile_key
*b
)
45 unsigned keysize_a
= svga_fs_key_size( a
);
46 unsigned keysize_b
= svga_fs_key_size( b
);
48 if (keysize_a
!= keysize_b
) {
49 return (int)(keysize_a
- keysize_b
);
51 return memcmp( a
, b
, keysize_a
);
55 static struct svga_shader_result
*search_fs_key( struct svga_fragment_shader
*fs
,
56 const struct svga_fs_compile_key
*key
)
58 struct svga_shader_result
*result
= fs
->base
.results
;
62 for ( ; result
; result
= result
->next
) {
63 if (compare_fs_keys( key
, &result
->key
.fkey
) == 0)
72 * If we fail to compile a fragment shader (because it uses too many
73 * registers, for example) we'll use a dummy/fallback shader that
74 * simply emits a constant color.
76 static const struct tgsi_token
*
77 get_dummy_fragment_shader(void)
79 static const float red
[4] = { 1.0, 0.0, 0.0, 0.0 };
80 struct ureg_program
*ureg
;
81 const struct tgsi_token
*tokens
;
86 ureg
= ureg_create(TGSI_PROCESSOR_FRAGMENT
);
90 dst
= ureg_DECL_output(ureg
, TGSI_SEMANTIC_COLOR
, 0);
91 src
= ureg_DECL_immediate(ureg
, red
, 4);
92 ureg_MOV(ureg
, dst
, src
);
95 tokens
= ureg_get_tokens(ureg
, &num_tokens
);
103 static enum pipe_error
compile_fs( struct svga_context
*svga
,
104 struct svga_fragment_shader
*fs
,
105 const struct svga_fs_compile_key
*key
,
106 struct svga_shader_result
**out_result
)
108 struct svga_shader_result
*result
;
109 enum pipe_error ret
= PIPE_ERROR
;
111 result
= svga_translate_fragment_program( fs
, key
);
112 if (result
== NULL
) {
113 /* some problem during translation, try the dummy shader */
114 const struct tgsi_token
*dummy
= get_dummy_fragment_shader();
116 ret
= PIPE_ERROR_OUT_OF_MEMORY
;
119 debug_printf("Failed to compile fragment shader, using dummy shader instead.\n");
120 FREE((void *) fs
->base
.tokens
);
121 fs
->base
.tokens
= dummy
;
122 result
= svga_translate_fragment_program(fs
, key
);
123 if (result
== NULL
) {
129 result
->id
= util_bitmask_add(svga
->fs_bm
);
130 if(result
->id
== UTIL_BITMASK_INVALID_INDEX
) {
131 ret
= PIPE_ERROR_OUT_OF_MEMORY
;
135 ret
= SVGA3D_DefineShader(svga
->swc
,
137 SVGA3D_SHADERTYPE_PS
,
139 result
->nr_tokens
* sizeof result
->tokens
[0]);
143 *out_result
= result
;
144 result
->next
= fs
->base
.results
;
145 fs
->base
.results
= result
;
150 if (result
->id
!= UTIL_BITMASK_INVALID_INDEX
)
151 util_bitmask_clear( svga
->fs_bm
, result
->id
);
152 svga_destroy_shader_result( result
);
158 /* SVGA_NEW_TEXTURE_BINDING
160 * SVGA_NEW_NEED_SWTNL
163 static enum pipe_error
164 make_fs_key(const struct svga_context
*svga
,
165 struct svga_fragment_shader
*fs
,
166 struct svga_fs_compile_key
*key
)
171 memset(key
, 0, sizeof *key
);
173 /* Only need fragment shader fixup for twoside lighting if doing
174 * hwtnl. Otherwise the draw module does the whole job for us.
178 if (!svga
->state
.sw
.need_swtnl
) {
181 key
->light_twoside
= svga
->curr
.rast
->templ
.light_twoside
;
182 key
->front_ccw
= svga
->curr
.rast
->templ
.front_ccw
;
185 /* The blend workaround for simulating logicop xor behaviour
186 * requires that the incoming fragment color be white. This change
187 * achieves that by creating a variant of the current fragment
188 * shader that overrides all output colors with 1,1,1,1
190 * This will work for most shaders, including those containing
191 * TEXKIL and/or depth-write. However, it will break on the
192 * combination of xor-logicop plus alphatest.
194 * Ultimately, we could implement alphatest in the shader using
195 * texkil prior to overriding the outgoing fragment color.
199 if (svga
->curr
.blend
->need_white_fragments
) {
200 key
->white_fragments
= 1;
203 /* XXX: want to limit this to the textures that the shader actually
206 * SVGA_NEW_TEXTURE_BINDING | SVGA_NEW_SAMPLER
208 for (i
= 0; i
< svga
->curr
.num_sampler_views
; i
++) {
209 if (svga
->curr
.sampler_views
[i
]) {
210 assert(svga
->curr
.sampler
[i
]);
211 assert(svga
->curr
.sampler_views
[i
]->texture
);
212 key
->tex
[i
].texture_target
= svga
->curr
.sampler_views
[i
]->texture
->target
;
213 if (!svga
->curr
.sampler
[i
]->normalized_coords
) {
214 key
->tex
[i
].width_height_idx
= idx
++;
215 key
->tex
[i
].unnormalized
= TRUE
;
216 ++key
->num_unnormalized_coords
;
219 key
->tex
[i
].swizzle_r
= svga
->curr
.sampler_views
[i
]->swizzle_r
;
220 key
->tex
[i
].swizzle_g
= svga
->curr
.sampler_views
[i
]->swizzle_g
;
221 key
->tex
[i
].swizzle_b
= svga
->curr
.sampler_views
[i
]->swizzle_b
;
222 key
->tex
[i
].swizzle_a
= svga
->curr
.sampler_views
[i
]->swizzle_a
;
225 key
->num_textures
= svga
->curr
.num_sampler_views
;
228 for (i
= 0; i
< svga
->curr
.num_samplers
; ++i
) {
229 if (svga
->curr
.sampler
[i
]) {
230 key
->tex
[i
].compare_mode
= svga
->curr
.sampler
[i
]->compare_mode
;
231 key
->tex
[i
].compare_func
= svga
->curr
.sampler
[i
]->compare_func
;
235 /* sprite coord gen state */
236 for (i
= 0; i
< svga
->curr
.num_samplers
; ++i
) {
237 key
->tex
[i
].sprite_texgen
=
238 svga
->curr
.rast
->templ
.sprite_coord_enable
& (1 << i
);
241 key
->sprite_origin_lower_left
= (svga
->curr
.rast
->templ
.sprite_coord_mode
242 == PIPE_SPRITE_COORD_LOWER_LEFT
);
249 static enum pipe_error
250 emit_hw_fs(struct svga_context
*svga
, unsigned dirty
)
252 struct svga_shader_result
*result
= NULL
;
253 unsigned id
= SVGA3D_INVALID_ID
;
254 enum pipe_error ret
= PIPE_OK
;
256 struct svga_fragment_shader
*fs
= svga
->curr
.fs
;
257 struct svga_fs_compile_key key
;
260 * SVGA_NEW_TEXTURE_BINDING
262 * SVGA_NEW_NEED_SWTNL
265 ret
= make_fs_key( svga
, fs
, &key
);
269 result
= search_fs_key( fs
, &key
);
271 ret
= compile_fs( svga
, fs
, &key
, &result
);
279 assert(id
!= SVGA3D_INVALID_ID
);
281 if (result
!= svga
->state
.hw_draw
.fs
) {
282 ret
= SVGA3D_SetShader(svga
->swc
,
283 SVGA3D_SHADERTYPE_PS
,
288 svga
->dirty
|= SVGA_NEW_FS_RESULT
;
289 svga
->state
.hw_draw
.fs
= result
;
295 struct svga_tracked_state svga_hw_fs
=
297 "fragment shader (hwtnl)",
299 SVGA_NEW_TEXTURE_BINDING
|
300 SVGA_NEW_NEED_SWTNL
|