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
, unsigned shader
,
170 struct svga_compile_key
*key
)
174 assert(shader
< Elements(svga
->curr
.num_sampler_views
));
176 for (i
= 0; i
< svga
->curr
.num_sampler_views
[shader
]; i
++) {
177 struct pipe_sampler_view
*view
= svga
->curr
.sampler_views
[shader
][i
];
179 assert(svga
->curr
.sampler
[shader
][i
]);
180 assert(view
->texture
);
181 assert(view
->texture
->target
< (1 << 4)); /* texture_target:4 */
183 /* 1D/2D array textures with one slice are treated as non-arrays
184 * by the SVGA3D device. Convert the texture type here so that
185 * we emit the right TEX/SAMPLE instruction in the shader.
187 if (view
->texture
->target
== PIPE_TEXTURE_1D_ARRAY
||
188 view
->texture
->target
== PIPE_TEXTURE_2D_ARRAY
) {
189 if (view
->texture
->array_size
== 1) {
190 key
->tex
[i
].is_array
= 0;
193 assert(view
->texture
->array_size
> 1);
194 key
->tex
[i
].is_array
= 1;
198 if (!svga
->curr
.sampler
[shader
][i
]->normalized_coords
) {
199 assert(idx
< (1 << 5)); /* width_height_idx:5 bitfield */
200 key
->tex
[i
].width_height_idx
= idx
++;
201 key
->tex
[i
].unnormalized
= TRUE
;
202 ++key
->num_unnormalized_coords
;
205 key
->tex
[i
].swizzle_r
= view
->swizzle_r
;
206 key
->tex
[i
].swizzle_g
= view
->swizzle_g
;
207 key
->tex
[i
].swizzle_b
= view
->swizzle_b
;
208 key
->tex
[i
].swizzle_a
= view
->swizzle_a
;
211 key
->num_textures
= svga
->curr
.num_sampler_views
[shader
];
215 /** Search for a compiled shader variant with the same compile key */
216 struct svga_shader_variant
*
217 svga_search_shader_key(const struct svga_shader
*shader
,
218 const struct svga_compile_key
*key
)
220 struct svga_shader_variant
*variant
= shader
->variants
;
224 for ( ; variant
; variant
= variant
->next
) {
225 if (svga_compile_keys_equal(key
, &variant
->key
))
231 /** Search for a shader with the same token key */
233 svga_search_shader_token_key(struct svga_shader
*pshader
,
234 const struct svga_token_key
*key
)
236 struct svga_shader
*shader
= pshader
;
240 for ( ; shader
; shader
= shader
->next
) {
241 if (memcmp(key
, &shader
->token_key
, sizeof(struct svga_token_key
)) == 0)
248 * Helper function to define a gb shader for non-vgpu10 device
250 static enum pipe_error
251 define_gb_shader_vgpu9(struct svga_context
*svga
,
252 SVGA3dShaderType type
,
253 struct svga_shader_variant
*variant
,
256 struct svga_winsys_screen
*sws
= svga_screen(svga
->pipe
.screen
)->sws
;
260 * Create gb memory for the shader and upload the shader code.
261 * Kernel module will allocate an id for the shader and issue
262 * the DefineGBShader command.
264 variant
->gb_shader
= sws
->shader_create(sws
, type
,
265 variant
->tokens
, codeLen
);
267 if (!variant
->gb_shader
)
268 return PIPE_ERROR_OUT_OF_MEMORY
;
270 ret
= SVGA3D_BindGBShader(svga
->swc
, variant
->gb_shader
);
276 * Helper function to define a gb shader for vgpu10 device
278 static enum pipe_error
279 define_gb_shader_vgpu10(struct svga_context
*svga
,
280 SVGA3dShaderType type
,
281 struct svga_shader_variant
*variant
,
284 struct svga_winsys_context
*swc
= svga
->swc
;
288 * Shaders in VGPU10 enabled device reside in the device COTable.
289 * SVGA driver will allocate an integer ID for the shader and
290 * issue DXDefineShader and DXBindShader commands.
292 variant
->id
= util_bitmask_add(svga
->shader_id_bm
);
293 if (variant
->id
== UTIL_BITMASK_INVALID_INDEX
) {
294 return PIPE_ERROR_OUT_OF_MEMORY
;
297 /* Create gb memory for the shader and upload the shader code */
298 variant
->gb_shader
= swc
->shader_create(swc
,
300 variant
->tokens
, codeLen
);
302 if (!variant
->gb_shader
) {
303 /* Free the shader ID */
304 assert(variant
->id
!= UTIL_BITMASK_INVALID_INDEX
);
305 goto fail_no_allocation
;
309 * Since we don't want to do any flush within state emission to avoid
310 * partial state in a command buffer, it's important to make sure that
311 * there is enough room to send both the DXDefineShader & DXBindShader
312 * commands in the same command buffer. So let's send both
313 * commands in one command reservation. If it fails, we'll undo
314 * the shader creation and return an error.
316 ret
= SVGA3D_vgpu10_DefineAndBindShader(swc
, variant
->gb_shader
,
317 variant
->id
, type
, codeLen
);
325 swc
->shader_destroy(swc
, variant
->gb_shader
);
326 variant
->gb_shader
= NULL
;
329 util_bitmask_clear(svga
->shader_id_bm
, variant
->id
);
330 variant
->id
= UTIL_BITMASK_INVALID_INDEX
;
332 return PIPE_ERROR_OUT_OF_MEMORY
;
336 * Issue the SVGA3D commands to define a new shader.
337 * \param variant contains the shader tokens, etc. The result->id field will
341 svga_define_shader(struct svga_context
*svga
,
342 SVGA3dShaderType type
,
343 struct svga_shader_variant
*variant
)
345 unsigned codeLen
= variant
->nr_tokens
* sizeof(variant
->tokens
[0]);
348 variant
->id
= UTIL_BITMASK_INVALID_INDEX
;
350 if (svga_have_gb_objects(svga
)) {
351 if (svga_have_vgpu10(svga
))
352 return define_gb_shader_vgpu10(svga
, type
, variant
, codeLen
);
354 return define_gb_shader_vgpu9(svga
, type
, variant
, codeLen
);
357 /* Allocate an integer ID for the shader */
358 variant
->id
= util_bitmask_add(svga
->shader_id_bm
);
359 if (variant
->id
== UTIL_BITMASK_INVALID_INDEX
) {
360 return PIPE_ERROR_OUT_OF_MEMORY
;
363 /* Issue SVGA3D device command to define the shader */
364 ret
= SVGA3D_DefineShader(svga
->swc
,
369 if (ret
!= PIPE_OK
) {
371 assert(variant
->id
!= UTIL_BITMASK_INVALID_INDEX
);
372 util_bitmask_clear(svga
->shader_id_bm
, variant
->id
);
373 variant
->id
= UTIL_BITMASK_INVALID_INDEX
;
382 * Issue the SVGA3D commands to set/bind a shader.
383 * \param result the shader to bind.
386 svga_set_shader(struct svga_context
*svga
,
387 SVGA3dShaderType type
,
388 struct svga_shader_variant
*variant
)
391 unsigned id
= variant
? variant
->id
: SVGA3D_INVALID_ID
;
393 assert(type
== SVGA3D_SHADERTYPE_VS
||
394 type
== SVGA3D_SHADERTYPE_GS
||
395 type
== SVGA3D_SHADERTYPE_PS
);
397 if (svga_have_gb_objects(svga
)) {
398 struct svga_winsys_gb_shader
*gbshader
=
399 variant
? variant
->gb_shader
: NULL
;
401 if (svga_have_vgpu10(svga
))
402 ret
= SVGA3D_vgpu10_SetShader(svga
->swc
, type
, gbshader
, id
);
404 ret
= SVGA3D_SetGBShader(svga
->swc
, type
, gbshader
);
407 ret
= SVGA3D_SetShader(svga
->swc
, type
, id
);
414 struct svga_shader_variant
*
415 svga_new_shader_variant(struct svga_context
*svga
)
417 svga
->hud
.num_shaders
++;
418 return CALLOC_STRUCT(svga_shader_variant
);
423 svga_destroy_shader_variant(struct svga_context
*svga
,
424 SVGA3dShaderType type
,
425 struct svga_shader_variant
*variant
)
427 enum pipe_error ret
= PIPE_OK
;
429 if (svga_have_gb_objects(svga
) && variant
->gb_shader
) {
430 if (svga_have_vgpu10(svga
)) {
431 struct svga_winsys_context
*swc
= svga
->swc
;
432 swc
->shader_destroy(swc
, variant
->gb_shader
);
433 ret
= SVGA3D_vgpu10_DestroyShader(svga
->swc
, variant
->id
);
434 if (ret
!= PIPE_OK
) {
435 /* flush and try again */
436 svga_context_flush(svga
, NULL
);
437 ret
= SVGA3D_vgpu10_DestroyShader(svga
->swc
, variant
->id
);
439 util_bitmask_clear(svga
->shader_id_bm
, variant
->id
);
442 struct svga_winsys_screen
*sws
= svga_screen(svga
->pipe
.screen
)->sws
;
443 sws
->shader_destroy(sws
, variant
->gb_shader
);
445 variant
->gb_shader
= NULL
;
448 if (variant
->id
!= UTIL_BITMASK_INVALID_INDEX
) {
449 ret
= SVGA3D_DestroyShader(svga
->swc
, variant
->id
, type
);
450 if (ret
!= PIPE_OK
) {
451 /* flush and try again */
452 svga_context_flush(svga
, NULL
);
453 ret
= SVGA3D_DestroyShader(svga
->swc
, variant
->id
, type
);
454 assert(ret
== PIPE_OK
);
456 util_bitmask_clear(svga
->shader_id_bm
, variant
->id
);
460 FREE((unsigned *)variant
->tokens
);
463 svga
->hud
.num_shaders
--;
470 * Called at the beginning of every new command buffer to ensure that
471 * shaders are properly paged-in. Instead of sending the SetShader
472 * command, this function sends a private allocation command to
473 * page in a shader. This avoids emitting redundant state to the device
474 * just to page in a resource.
477 svga_rebind_shaders(struct svga_context
*svga
)
479 struct svga_winsys_context
*swc
= svga
->swc
;
480 struct svga_hw_draw_state
*hw
= &svga
->state
.hw_draw
;
483 assert(svga_have_vgpu10(svga
));
486 * If the underlying winsys layer does not need resource rebinding,
487 * just clear the rebind flags and return.
489 if (swc
->resource_rebind
== NULL
) {
490 svga
->rebind
.flags
.vs
= 0;
491 svga
->rebind
.flags
.gs
= 0;
492 svga
->rebind
.flags
.fs
= 0;
497 if (svga
->rebind
.flags
.vs
&& hw
->vs
&& hw
->vs
->gb_shader
) {
498 ret
= swc
->resource_rebind(swc
, NULL
, hw
->vs
->gb_shader
, SVGA_RELOC_READ
);
502 svga
->rebind
.flags
.vs
= 0;
504 if (svga
->rebind
.flags
.gs
&& hw
->gs
&& hw
->gs
->gb_shader
) {
505 ret
= swc
->resource_rebind(swc
, NULL
, hw
->gs
->gb_shader
, SVGA_RELOC_READ
);
509 svga
->rebind
.flags
.gs
= 0;
511 if (svga
->rebind
.flags
.fs
&& hw
->fs
&& hw
->fs
->gb_shader
) {
512 ret
= swc
->resource_rebind(swc
, NULL
, hw
->fs
->gb_shader
, SVGA_RELOC_READ
);
516 svga
->rebind
.flags
.fs
= 0;