1 /**********************************************************
2 * Copyright 2008-2012 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_bitmask.h"
27 #include "util/u_memory.h"
28 #include "svga_context.h"
30 #include "svga_format.h"
31 #include "svga_shader.h"
35 * This bit isn't really used anywhere. It only serves to help
36 * generate a unique "signature" for the vertex shader output bitmask.
37 * Shader input/output signatures are used to resolve shader linking
40 #define FOG_GENERIC_BIT (((uint64_t) 1) << 63)
44 * Use the shader info to generate a bitmask indicating which generic
45 * inputs are used by the shader. A set bit indicates that GENERIC[i]
49 svga_get_generic_inputs_mask(const struct tgsi_shader_info
*info
)
54 for (i
= 0; i
< info
->num_inputs
; i
++) {
55 if (info
->input_semantic_name
[i
] == TGSI_SEMANTIC_GENERIC
) {
56 unsigned j
= info
->input_semantic_index
[i
];
57 assert(j
< sizeof(mask
) * 8);
58 mask
|= ((uint64_t) 1) << j
;
67 * Scan shader info to return a bitmask of written outputs.
70 svga_get_generic_outputs_mask(const struct tgsi_shader_info
*info
)
75 for (i
= 0; i
< info
->num_outputs
; i
++) {
76 switch (info
->output_semantic_name
[i
]) {
77 case TGSI_SEMANTIC_GENERIC
:
79 unsigned j
= info
->output_semantic_index
[i
];
80 assert(j
< sizeof(mask
) * 8);
81 mask
|= ((uint64_t) 1) << j
;
84 case TGSI_SEMANTIC_FOG
:
85 mask
|= FOG_GENERIC_BIT
;
96 * Given a mask of used generic variables (as returned by the above functions)
97 * fill in a table which maps those indexes to small integers.
98 * This table is used by the remap_generic_index() function in
99 * svga_tgsi_decl_sm30.c
100 * Example: if generics_mask = binary(1010) it means that GENERIC[1] and
101 * GENERIC[3] are used. The remap_table will contain:
104 * The remaining table entries will be filled in with the next unused
105 * generic index (in this example, 2).
108 svga_remap_generics(uint64_t generics_mask
,
109 int8_t remap_table
[MAX_GENERIC_VARYING
])
111 /* Note texcoord[0] is reserved so start at 1 */
112 unsigned count
= 1, i
;
114 for (i
= 0; i
< MAX_GENERIC_VARYING
; i
++) {
118 /* for each bit set in generic_mask */
119 while (generics_mask
) {
120 unsigned index
= ffsll(generics_mask
) - 1;
121 remap_table
[index
] = count
++;
122 generics_mask
&= ~((uint64_t) 1 << index
);
128 * Use the generic remap table to map a TGSI generic varying variable
129 * index to a small integer. If the remapping table doesn't have a
130 * valid value for the given index (the table entry is -1) it means
131 * the fragment shader doesn't use that VS output. Just allocate
132 * the next free value in that case. Alternately, we could cull
133 * VS instructions that write to register, or replace the register
134 * with a dummy temp register.
135 * XXX TODO: we should do one of the later as it would save precious
136 * texcoord registers.
139 svga_remap_generic_index(int8_t remap_table
[MAX_GENERIC_VARYING
],
142 assert(generic_index
< MAX_GENERIC_VARYING
);
144 if (generic_index
>= MAX_GENERIC_VARYING
) {
145 /* just don't return a random/garbage value */
146 generic_index
= MAX_GENERIC_VARYING
- 1;
149 if (remap_table
[generic_index
] == -1) {
150 /* This is a VS output that has no matching PS input. Find a
154 for (i
= 0; i
< MAX_GENERIC_VARYING
; i
++) {
155 max
= MAX2(max
, remap_table
[i
]);
157 remap_table
[generic_index
] = max
+ 1;
160 return remap_table
[generic_index
];
165 * Initialize the shader-neutral fields of svga_compile_key from context
166 * state. This is basically the texture-related state.
169 svga_init_shader_key_common(const struct svga_context
*svga
,
170 enum pipe_shader_type shader
,
171 struct svga_compile_key
*key
)
175 assert(shader
< ARRAY_SIZE(svga
->curr
.num_sampler_views
));
177 /* In case the number of samplers and sampler_views doesn't match,
178 * loop over the lower of the two counts.
180 key
->num_textures
= MIN2(svga
->curr
.num_sampler_views
[shader
],
181 svga
->curr
.num_samplers
[shader
]);
183 for (i
= 0; i
< key
->num_textures
; i
++) {
184 struct pipe_sampler_view
*view
= svga
->curr
.sampler_views
[shader
][i
];
185 const struct svga_sampler_state
*sampler
= svga
->curr
.sampler
[shader
][i
];
186 if (view
&& sampler
) {
187 assert(view
->texture
);
188 assert(view
->texture
->target
< (1 << 4)); /* texture_target:4 */
190 /* 1D/2D array textures with one slice are treated as non-arrays
191 * by the SVGA3D device. Convert the texture type here so that
192 * we emit the right TEX/SAMPLE instruction in the shader.
194 if (view
->texture
->target
== PIPE_TEXTURE_1D_ARRAY
||
195 view
->texture
->target
== PIPE_TEXTURE_2D_ARRAY
) {
196 if (view
->texture
->array_size
== 1) {
197 key
->tex
[i
].is_array
= 0;
200 assert(view
->texture
->array_size
> 1);
201 key
->tex
[i
].is_array
= 1;
205 if (!sampler
->normalized_coords
) {
206 assert(idx
< (1 << 5)); /* width_height_idx:5 bitfield */
207 key
->tex
[i
].width_height_idx
= idx
++;
208 key
->tex
[i
].unnormalized
= TRUE
;
209 ++key
->num_unnormalized_coords
;
212 key
->tex
[i
].swizzle_r
= view
->swizzle_r
;
213 key
->tex
[i
].swizzle_g
= view
->swizzle_g
;
214 key
->tex
[i
].swizzle_b
= view
->swizzle_b
;
215 key
->tex
[i
].swizzle_a
= view
->swizzle_a
;
221 /** Search for a compiled shader variant with the same compile key */
222 struct svga_shader_variant
*
223 svga_search_shader_key(const struct svga_shader
*shader
,
224 const struct svga_compile_key
*key
)
226 struct svga_shader_variant
*variant
= shader
->variants
;
230 for ( ; variant
; variant
= variant
->next
) {
231 if (svga_compile_keys_equal(key
, &variant
->key
))
237 /** Search for a shader with the same token key */
239 svga_search_shader_token_key(struct svga_shader
*pshader
,
240 const struct svga_token_key
*key
)
242 struct svga_shader
*shader
= pshader
;
246 for ( ; shader
; shader
= shader
->next
) {
247 if (memcmp(key
, &shader
->token_key
, sizeof(struct svga_token_key
)) == 0)
254 * Helper function to define a gb shader for non-vgpu10 device
256 static enum pipe_error
257 define_gb_shader_vgpu9(struct svga_context
*svga
,
258 SVGA3dShaderType type
,
259 struct svga_shader_variant
*variant
,
262 struct svga_winsys_screen
*sws
= svga_screen(svga
->pipe
.screen
)->sws
;
266 * Create gb memory for the shader and upload the shader code.
267 * Kernel module will allocate an id for the shader and issue
268 * the DefineGBShader command.
270 variant
->gb_shader
= sws
->shader_create(sws
, type
,
271 variant
->tokens
, codeLen
);
273 if (!variant
->gb_shader
)
274 return PIPE_ERROR_OUT_OF_MEMORY
;
276 ret
= SVGA3D_BindGBShader(svga
->swc
, variant
->gb_shader
);
282 * Helper function to define a gb shader for vgpu10 device
284 static enum pipe_error
285 define_gb_shader_vgpu10(struct svga_context
*svga
,
286 SVGA3dShaderType type
,
287 struct svga_shader_variant
*variant
,
290 struct svga_winsys_context
*swc
= svga
->swc
;
294 * Shaders in VGPU10 enabled device reside in the device COTable.
295 * SVGA driver will allocate an integer ID for the shader and
296 * issue DXDefineShader and DXBindShader commands.
298 variant
->id
= util_bitmask_add(svga
->shader_id_bm
);
299 if (variant
->id
== UTIL_BITMASK_INVALID_INDEX
) {
300 return PIPE_ERROR_OUT_OF_MEMORY
;
303 /* Create gb memory for the shader and upload the shader code */
304 variant
->gb_shader
= swc
->shader_create(swc
,
306 variant
->tokens
, codeLen
);
308 if (!variant
->gb_shader
) {
309 /* Free the shader ID */
310 assert(variant
->id
!= UTIL_BITMASK_INVALID_INDEX
);
311 goto fail_no_allocation
;
315 * Since we don't want to do any flush within state emission to avoid
316 * partial state in a command buffer, it's important to make sure that
317 * there is enough room to send both the DXDefineShader & DXBindShader
318 * commands in the same command buffer. So let's send both
319 * commands in one command reservation. If it fails, we'll undo
320 * the shader creation and return an error.
322 ret
= SVGA3D_vgpu10_DefineAndBindShader(swc
, variant
->gb_shader
,
323 variant
->id
, type
, codeLen
);
331 swc
->shader_destroy(swc
, variant
->gb_shader
);
332 variant
->gb_shader
= NULL
;
335 util_bitmask_clear(svga
->shader_id_bm
, variant
->id
);
336 variant
->id
= UTIL_BITMASK_INVALID_INDEX
;
338 return PIPE_ERROR_OUT_OF_MEMORY
;
342 * Issue the SVGA3D commands to define a new shader.
343 * \param variant contains the shader tokens, etc. The result->id field will
347 svga_define_shader(struct svga_context
*svga
,
348 SVGA3dShaderType type
,
349 struct svga_shader_variant
*variant
)
351 unsigned codeLen
= variant
->nr_tokens
* sizeof(variant
->tokens
[0]);
354 SVGA_STATS_TIME_PUSH(svga_sws(svga
), SVGA_STATS_TIME_DEFINESHADER
);
356 variant
->id
= UTIL_BITMASK_INVALID_INDEX
;
358 if (svga_have_gb_objects(svga
)) {
359 if (svga_have_vgpu10(svga
))
360 ret
= define_gb_shader_vgpu10(svga
, type
, variant
, codeLen
);
362 ret
= define_gb_shader_vgpu9(svga
, type
, variant
, codeLen
);
365 /* Allocate an integer ID for the shader */
366 variant
->id
= util_bitmask_add(svga
->shader_id_bm
);
367 if (variant
->id
== UTIL_BITMASK_INVALID_INDEX
) {
368 ret
= PIPE_ERROR_OUT_OF_MEMORY
;
372 /* Issue SVGA3D device command to define the shader */
373 ret
= SVGA3D_DefineShader(svga
->swc
,
378 if (ret
!= PIPE_OK
) {
380 assert(variant
->id
!= UTIL_BITMASK_INVALID_INDEX
);
381 util_bitmask_clear(svga
->shader_id_bm
, variant
->id
);
382 variant
->id
= UTIL_BITMASK_INVALID_INDEX
;
387 SVGA_STATS_TIME_POP(svga_sws(svga
));
393 * Issue the SVGA3D commands to set/bind a shader.
394 * \param result the shader to bind.
397 svga_set_shader(struct svga_context
*svga
,
398 SVGA3dShaderType type
,
399 struct svga_shader_variant
*variant
)
402 unsigned id
= variant
? variant
->id
: SVGA3D_INVALID_ID
;
404 assert(type
== SVGA3D_SHADERTYPE_VS
||
405 type
== SVGA3D_SHADERTYPE_GS
||
406 type
== SVGA3D_SHADERTYPE_PS
);
408 if (svga_have_gb_objects(svga
)) {
409 struct svga_winsys_gb_shader
*gbshader
=
410 variant
? variant
->gb_shader
: NULL
;
412 if (svga_have_vgpu10(svga
))
413 ret
= SVGA3D_vgpu10_SetShader(svga
->swc
, type
, gbshader
, id
);
415 ret
= SVGA3D_SetGBShader(svga
->swc
, type
, gbshader
);
418 ret
= SVGA3D_SetShader(svga
->swc
, type
, id
);
425 struct svga_shader_variant
*
426 svga_new_shader_variant(struct svga_context
*svga
)
428 svga
->hud
.num_shaders
++;
429 return CALLOC_STRUCT(svga_shader_variant
);
434 svga_destroy_shader_variant(struct svga_context
*svga
,
435 SVGA3dShaderType type
,
436 struct svga_shader_variant
*variant
)
438 enum pipe_error ret
= PIPE_OK
;
440 if (svga_have_gb_objects(svga
) && variant
->gb_shader
) {
441 if (svga_have_vgpu10(svga
)) {
442 struct svga_winsys_context
*swc
= svga
->swc
;
443 swc
->shader_destroy(swc
, variant
->gb_shader
);
444 ret
= SVGA3D_vgpu10_DestroyShader(svga
->swc
, variant
->id
);
445 if (ret
!= PIPE_OK
) {
446 /* flush and try again */
447 svga_context_flush(svga
, NULL
);
448 ret
= SVGA3D_vgpu10_DestroyShader(svga
->swc
, variant
->id
);
450 util_bitmask_clear(svga
->shader_id_bm
, variant
->id
);
453 struct svga_winsys_screen
*sws
= svga_screen(svga
->pipe
.screen
)->sws
;
454 sws
->shader_destroy(sws
, variant
->gb_shader
);
456 variant
->gb_shader
= NULL
;
459 if (variant
->id
!= UTIL_BITMASK_INVALID_INDEX
) {
460 ret
= SVGA3D_DestroyShader(svga
->swc
, variant
->id
, type
);
461 if (ret
!= PIPE_OK
) {
462 /* flush and try again */
463 svga_context_flush(svga
, NULL
);
464 ret
= SVGA3D_DestroyShader(svga
->swc
, variant
->id
, type
);
465 assert(ret
== PIPE_OK
);
467 util_bitmask_clear(svga
->shader_id_bm
, variant
->id
);
471 FREE((unsigned *)variant
->tokens
);
474 svga
->hud
.num_shaders
--;
481 * Called at the beginning of every new command buffer to ensure that
482 * shaders are properly paged-in. Instead of sending the SetShader
483 * command, this function sends a private allocation command to
484 * page in a shader. This avoids emitting redundant state to the device
485 * just to page in a resource.
488 svga_rebind_shaders(struct svga_context
*svga
)
490 struct svga_winsys_context
*swc
= svga
->swc
;
491 struct svga_hw_draw_state
*hw
= &svga
->state
.hw_draw
;
494 assert(svga_have_vgpu10(svga
));
497 * If the underlying winsys layer does not need resource rebinding,
498 * just clear the rebind flags and return.
500 if (swc
->resource_rebind
== NULL
) {
501 svga
->rebind
.flags
.vs
= 0;
502 svga
->rebind
.flags
.gs
= 0;
503 svga
->rebind
.flags
.fs
= 0;
508 if (svga
->rebind
.flags
.vs
&& hw
->vs
&& hw
->vs
->gb_shader
) {
509 ret
= swc
->resource_rebind(swc
, NULL
, hw
->vs
->gb_shader
, SVGA_RELOC_READ
);
513 svga
->rebind
.flags
.vs
= 0;
515 if (svga
->rebind
.flags
.gs
&& hw
->gs
&& hw
->gs
->gb_shader
) {
516 ret
= swc
->resource_rebind(swc
, NULL
, hw
->gs
->gb_shader
, SVGA_RELOC_READ
);
520 svga
->rebind
.flags
.gs
= 0;
522 if (svga
->rebind
.flags
.fs
&& hw
->fs
&& hw
->fs
->gb_shader
) {
523 ret
= swc
->resource_rebind(swc
, NULL
, hw
->fs
->gb_shader
, SVGA_RELOC_READ
);
527 svga
->rebind
.flags
.fs
= 0;