1 /****************************************************************************
2 * Copyright (C) 2015 Intel Corporation. All Rights Reserved.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
22 ***************************************************************************/
24 #include "swr_context.h"
25 #include "swr_public.h"
26 #include "swr_screen.h"
27 #include "swr_resource.h"
28 #include "swr_fence.h"
29 #include "gen_knobs.h"
31 #include "pipe/p_screen.h"
32 #include "pipe/p_defines.h"
33 #include "util/u_memory.h"
34 #include "util/u_format.h"
35 #include "util/u_inlines.h"
36 #include "util/u_cpu_detect.h"
37 #include "util/u_format_s3tc.h"
38 #include "util/u_string.h"
40 #include "state_tracker/sw_winsys.h"
44 #include "memory/TilingFunctions.h"
51 * XXX Check max texture size values against core and sampler.
53 #define SWR_MAX_TEXTURE_SIZE (4 * 1024 * 1024 * 1024ULL) /* 4GB */
54 #define SWR_MAX_TEXTURE_2D_LEVELS 14 /* 8K x 8K for now */
55 #define SWR_MAX_TEXTURE_3D_LEVELS 12 /* 2K x 2K x 2K for now */
56 #define SWR_MAX_TEXTURE_CUBE_LEVELS 14 /* 8K x 8K for now */
57 #define SWR_MAX_TEXTURE_ARRAY_LAYERS 512 /* 8K x 512 / 8K x 8K x 512 */
59 /* Default max client_copy_limit */
60 #define SWR_CLIENT_COPY_LIMIT 32768
62 /* Flag indicates creation of alternate surface, to prevent recursive loop
63 * in resource creation when msaa_force_enable is set. */
64 #define SWR_RESOURCE_FLAG_ALT_SURFACE (PIPE_RESOURCE_FLAG_DRV_PRIV << 0)
68 swr_get_name(struct pipe_screen
*screen
)
71 util_snprintf(buf
, sizeof(buf
), "SWR (LLVM %u.%u, %u bits)",
72 HAVE_LLVM
>> 8, HAVE_LLVM
& 0xff,
73 lp_native_vector_width
);
78 swr_get_vendor(struct pipe_screen
*screen
)
80 return "Intel Corporation";
84 swr_is_format_supported(struct pipe_screen
*_screen
,
85 enum pipe_format format
,
86 enum pipe_texture_target target
,
87 unsigned sample_count
,
90 struct swr_screen
*screen
= swr_screen(_screen
);
91 struct sw_winsys
*winsys
= screen
->winsys
;
92 const struct util_format_description
*format_desc
;
94 assert(target
== PIPE_BUFFER
|| target
== PIPE_TEXTURE_1D
95 || target
== PIPE_TEXTURE_1D_ARRAY
96 || target
== PIPE_TEXTURE_2D
97 || target
== PIPE_TEXTURE_2D_ARRAY
98 || target
== PIPE_TEXTURE_RECT
99 || target
== PIPE_TEXTURE_3D
100 || target
== PIPE_TEXTURE_CUBE
101 || target
== PIPE_TEXTURE_CUBE_ARRAY
);
103 format_desc
= util_format_description(format
);
107 if ((sample_count
> screen
->msaa_max_count
)
108 || !util_is_power_of_two(sample_count
))
111 if (bind
& PIPE_BIND_DISPLAY_TARGET
) {
112 if (!winsys
->is_displaytarget_format_supported(winsys
, bind
, format
))
116 if (bind
& PIPE_BIND_RENDER_TARGET
) {
117 if (format_desc
->colorspace
== UTIL_FORMAT_COLORSPACE_ZS
)
120 if (mesa_to_swr_format(format
) == (SWR_FORMAT
)-1)
124 * Although possible, it is unnatural to render into compressed or YUV
125 * surfaces. So disable these here to avoid going into weird paths
126 * inside the state trackers.
128 if (format_desc
->block
.width
!= 1 || format_desc
->block
.height
!= 1)
132 if (bind
& PIPE_BIND_DEPTH_STENCIL
) {
133 if (format_desc
->colorspace
!= UTIL_FORMAT_COLORSPACE_ZS
)
136 if (mesa_to_swr_format(format
) == (SWR_FORMAT
)-1)
140 if (format_desc
->layout
== UTIL_FORMAT_LAYOUT_BPTC
||
141 format_desc
->layout
== UTIL_FORMAT_LAYOUT_ASTC
) {
145 if (format_desc
->layout
== UTIL_FORMAT_LAYOUT_ETC
&&
146 format
!= PIPE_FORMAT_ETC1_RGB8
) {
150 if (format_desc
->layout
== UTIL_FORMAT_LAYOUT_S3TC
) {
151 return util_format_s3tc_enabled
;
158 swr_get_param(struct pipe_screen
*screen
, enum pipe_cap param
)
162 case PIPE_CAP_MAX_RENDER_TARGETS
:
163 return PIPE_MAX_COLOR_BUFS
;
164 case PIPE_CAP_MAX_TEXTURE_2D_LEVELS
:
165 return SWR_MAX_TEXTURE_2D_LEVELS
;
166 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS
:
167 return SWR_MAX_TEXTURE_3D_LEVELS
;
168 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS
:
169 return SWR_MAX_TEXTURE_CUBE_LEVELS
;
170 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS
:
171 return MAX_SO_STREAMS
;
172 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS
:
173 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS
:
174 return MAX_ATTRIBUTES
* 4;
175 case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES
:
176 case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS
:
178 case PIPE_CAP_MAX_VERTEX_STREAMS
:
180 case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE
:
182 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS
:
183 return SWR_MAX_TEXTURE_ARRAY_LAYERS
;
184 case PIPE_CAP_MIN_TEXEL_OFFSET
:
186 case PIPE_CAP_MAX_TEXEL_OFFSET
:
188 case PIPE_CAP_GLSL_FEATURE_LEVEL
:
190 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT
:
192 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT
:
194 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE
:
196 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT
:
198 case PIPE_CAP_MAX_VIEWPORTS
:
200 case PIPE_CAP_ENDIANNESS
:
201 return PIPE_ENDIAN_NATIVE
;
202 case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET
:
203 case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET
:
206 /* supported features */
207 case PIPE_CAP_NPOT_TEXTURES
:
208 case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES
:
209 case PIPE_CAP_MIXED_COLOR_DEPTH_BITS
:
210 case PIPE_CAP_TWO_SIDED_STENCIL
:
212 case PIPE_CAP_POINT_SPRITE
:
213 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS
:
214 case PIPE_CAP_OCCLUSION_QUERY
:
215 case PIPE_CAP_QUERY_TIME_ELAPSED
:
216 case PIPE_CAP_QUERY_PIPELINE_STATISTICS
:
217 case PIPE_CAP_TEXTURE_MIRROR_CLAMP
:
218 case PIPE_CAP_TEXTURE_SHADOW_MAP
:
219 case PIPE_CAP_TEXTURE_SWIZZLE
:
220 case PIPE_CAP_BLEND_EQUATION_SEPARATE
:
221 case PIPE_CAP_INDEP_BLEND_ENABLE
:
222 case PIPE_CAP_INDEP_BLEND_FUNC
:
223 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT
:
224 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER
:
225 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER
:
226 case PIPE_CAP_DEPTH_CLIP_DISABLE
:
227 case PIPE_CAP_PRIMITIVE_RESTART
:
228 case PIPE_CAP_TGSI_INSTANCEID
:
229 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR
:
230 case PIPE_CAP_START_INSTANCE
:
231 case PIPE_CAP_SEAMLESS_CUBE_MAP
:
232 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE
:
233 case PIPE_CAP_CONDITIONAL_RENDER
:
234 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED
:
235 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS
:
236 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION
:
237 case PIPE_CAP_USER_VERTEX_BUFFERS
:
238 case PIPE_CAP_USER_CONSTANT_BUFFERS
:
239 case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS
:
240 case PIPE_CAP_QUERY_TIMESTAMP
:
241 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS
:
242 case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT
:
243 case PIPE_CAP_DRAW_INDIRECT
:
245 case PIPE_CAP_CONDITIONAL_RENDER_INVERTED
:
246 case PIPE_CAP_CLIP_HALFZ
:
247 case PIPE_CAP_POLYGON_OFFSET_CLAMP
:
248 case PIPE_CAP_DEPTH_BOUNDS_TEST
:
249 case PIPE_CAP_CLEAR_TEXTURE
:
250 case PIPE_CAP_TEXTURE_FLOAT_LINEAR
:
251 case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR
:
252 case PIPE_CAP_CULL_DISTANCE
:
253 case PIPE_CAP_CUBE_MAP_ARRAY
:
254 case PIPE_CAP_DOUBLES
:
258 * If user has explicitly set max_sample_count = 1 (via SWR_MSAA_MAX_COUNT)
259 * then disable all MSAA support and go back to old (FAKE_SW_MSAA) caps. */
260 case PIPE_CAP_TEXTURE_MULTISAMPLE
:
261 case PIPE_CAP_MULTISAMPLE_Z_RESOLVE
:
262 return (swr_screen(screen
)->msaa_max_count
> 1) ? 1 : 0;
263 case PIPE_CAP_FAKE_SW_MSAA
:
264 return (swr_screen(screen
)->msaa_max_count
> 1) ? 0 : 1;
266 /* fetch jit change for 2-4GB buffers requires alignment */
267 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY
:
268 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY
:
269 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY
:
272 /* unsupported features */
273 case PIPE_CAP_ANISOTROPIC_FILTER
:
274 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK
:
275 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT
:
276 case PIPE_CAP_SHADER_STENCIL_EXPORT
:
277 case PIPE_CAP_TEXTURE_BARRIER
:
278 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED
:
279 case PIPE_CAP_VERTEX_COLOR_CLAMPED
:
280 case PIPE_CAP_COMPUTE
:
281 case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT
:
282 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS
:
283 case PIPE_CAP_TGSI_TEXCOORD
:
284 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER
:
285 case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS
:
286 case PIPE_CAP_TEXTURE_GATHER_SM5
:
287 case PIPE_CAP_TEXTURE_QUERY_LOD
:
288 case PIPE_CAP_SAMPLE_SHADING
:
289 case PIPE_CAP_TEXTURE_GATHER_OFFSETS
:
290 case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION
:
291 case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE
:
292 case PIPE_CAP_SAMPLER_VIEW_TARGET
:
293 case PIPE_CAP_VERTEXID_NOBASE
:
294 case PIPE_CAP_RESOURCE_FROM_USER_MEMORY
:
295 case PIPE_CAP_DEVICE_RESET_STATUS_QUERY
:
296 case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS
:
297 case PIPE_CAP_TGSI_TXQS
:
298 case PIPE_CAP_FORCE_PERSAMPLE_INTERP
:
299 case PIPE_CAP_SHAREABLE_SHADERS
:
300 case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS
:
301 case PIPE_CAP_DRAW_PARAMETERS
:
302 case PIPE_CAP_TGSI_PACK_HALF_FLOAT
:
303 case PIPE_CAP_MULTI_DRAW_INDIRECT
:
304 case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS
:
305 case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL
:
306 case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL
:
307 case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT
:
308 case PIPE_CAP_INVALIDATE_BUFFER
:
309 case PIPE_CAP_GENERATE_MIPMAP
:
310 case PIPE_CAP_STRING_MARKER
:
311 case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY
:
312 case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS
:
313 case PIPE_CAP_QUERY_BUFFER_OBJECT
:
314 case PIPE_CAP_QUERY_MEMORY_INFO
:
315 case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR
:
316 case PIPE_CAP_PCI_GROUP
:
317 case PIPE_CAP_PCI_BUS
:
318 case PIPE_CAP_PCI_DEVICE
:
319 case PIPE_CAP_PCI_FUNCTION
:
320 case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT
:
321 case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES
:
322 case PIPE_CAP_TGSI_VOTE
:
323 case PIPE_CAP_MAX_WINDOW_RECTANGLES
:
324 case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED
:
325 case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS
:
326 case PIPE_CAP_TGSI_ARRAY_COMPONENTS
:
327 case PIPE_CAP_TGSI_CAN_READ_OUTPUTS
:
328 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME
:
329 case PIPE_CAP_NATIVE_FENCE_FD
:
330 case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY
:
331 case PIPE_CAP_TGSI_FS_FBFETCH
:
332 case PIPE_CAP_TGSI_MUL_ZERO_WINS
:
334 case PIPE_CAP_INT64_DIVMOD
:
335 case PIPE_CAP_TGSI_TEX_TXF_LZ
:
336 case PIPE_CAP_TGSI_CLOCK
:
337 case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE
:
338 case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE
:
339 case PIPE_CAP_TGSI_BALLOT
:
340 case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT
:
341 case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX
:
342 case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION
:
343 case PIPE_CAP_POST_DEPTH_COVERAGE
:
344 case PIPE_CAP_BINDLESS_TEXTURE
:
345 case PIPE_CAP_NIR_SAMPLERS_AS_DEREF
:
346 case PIPE_CAP_QUERY_SO_OVERFLOW
:
347 case PIPE_CAP_MEMOBJ
:
348 case PIPE_CAP_LOAD_CONSTBUF
:
351 case PIPE_CAP_VENDOR_ID
:
353 case PIPE_CAP_DEVICE_ID
:
355 case PIPE_CAP_ACCELERATED
:
357 case PIPE_CAP_VIDEO_MEMORY
: {
358 /* XXX: Do we want to return the full amount of system memory ? */
359 uint64_t system_memory
;
361 if (!os_get_total_physical_memory(&system_memory
))
364 return (int)(system_memory
>> 20);
368 /* should only get here on unhandled cases */
369 debug_printf("Unexpected PIPE_CAP %d query\n", param
);
374 swr_get_shader_param(struct pipe_screen
*screen
,
375 enum pipe_shader_type shader
,
376 enum pipe_shader_cap param
)
378 if (shader
== PIPE_SHADER_VERTEX
||
379 shader
== PIPE_SHADER_FRAGMENT
||
380 shader
== PIPE_SHADER_GEOMETRY
)
381 return gallivm_get_shader_param(param
);
383 // Todo: tesselation, compute
389 swr_get_paramf(struct pipe_screen
*screen
, enum pipe_capf param
)
392 case PIPE_CAPF_MAX_LINE_WIDTH
:
393 case PIPE_CAPF_MAX_LINE_WIDTH_AA
:
394 case PIPE_CAPF_MAX_POINT_WIDTH
:
395 return 255.0; /* arbitrary */
396 case PIPE_CAPF_MAX_POINT_WIDTH_AA
:
398 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY
:
400 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS
:
401 return 16.0; /* arbitrary */
402 case PIPE_CAPF_GUARD_BAND_LEFT
:
403 case PIPE_CAPF_GUARD_BAND_TOP
:
404 case PIPE_CAPF_GUARD_BAND_RIGHT
:
405 case PIPE_CAPF_GUARD_BAND_BOTTOM
:
408 /* should only get here on unhandled cases */
409 debug_printf("Unexpected PIPE_CAPF %d query\n", param
);
414 mesa_to_swr_format(enum pipe_format format
)
416 static const std::map
<pipe_format
,SWR_FORMAT
> mesa2swr
= {
417 /* depth / stencil */
418 {PIPE_FORMAT_Z16_UNORM
, R16_UNORM
}, // z
419 {PIPE_FORMAT_Z32_FLOAT
, R32_FLOAT
}, // z
420 {PIPE_FORMAT_Z24_UNORM_S8_UINT
, R24_UNORM_X8_TYPELESS
}, // z
421 {PIPE_FORMAT_Z24X8_UNORM
, R24_UNORM_X8_TYPELESS
}, // z
422 {PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
, R32_FLOAT_X8X24_TYPELESS
}, // z
425 {PIPE_FORMAT_A8_UNORM
, A8_UNORM
},
426 {PIPE_FORMAT_A16_UNORM
, A16_UNORM
},
427 {PIPE_FORMAT_A16_FLOAT
, A16_FLOAT
},
428 {PIPE_FORMAT_A32_FLOAT
, A32_FLOAT
},
431 {PIPE_FORMAT_B5G6R5_UNORM
, B5G6R5_UNORM
},
432 {PIPE_FORMAT_B5G6R5_SRGB
, B5G6R5_UNORM_SRGB
},
433 {PIPE_FORMAT_B5G5R5A1_UNORM
, B5G5R5A1_UNORM
},
434 {PIPE_FORMAT_B5G5R5X1_UNORM
, B5G5R5X1_UNORM
},
435 {PIPE_FORMAT_B4G4R4A4_UNORM
, B4G4R4A4_UNORM
},
436 {PIPE_FORMAT_B8G8R8A8_UNORM
, B8G8R8A8_UNORM
},
437 {PIPE_FORMAT_B8G8R8A8_SRGB
, B8G8R8A8_UNORM_SRGB
},
438 {PIPE_FORMAT_B8G8R8X8_UNORM
, B8G8R8X8_UNORM
},
439 {PIPE_FORMAT_B8G8R8X8_SRGB
, B8G8R8X8_UNORM_SRGB
},
442 {PIPE_FORMAT_R10G10B10A2_UNORM
, R10G10B10A2_UNORM
},
443 {PIPE_FORMAT_R10G10B10A2_SNORM
, R10G10B10A2_SNORM
},
444 {PIPE_FORMAT_R10G10B10A2_USCALED
, R10G10B10A2_USCALED
},
445 {PIPE_FORMAT_R10G10B10A2_SSCALED
, R10G10B10A2_SSCALED
},
446 {PIPE_FORMAT_R10G10B10A2_UINT
, R10G10B10A2_UINT
},
449 {PIPE_FORMAT_R10G10B10X2_USCALED
, R10G10B10X2_USCALED
},
452 {PIPE_FORMAT_B10G10R10A2_UNORM
, B10G10R10A2_UNORM
},
453 {PIPE_FORMAT_B10G10R10A2_SNORM
, B10G10R10A2_SNORM
},
454 {PIPE_FORMAT_B10G10R10A2_USCALED
, B10G10R10A2_USCALED
},
455 {PIPE_FORMAT_B10G10R10A2_SSCALED
, B10G10R10A2_SSCALED
},
456 {PIPE_FORMAT_B10G10R10A2_UINT
, B10G10R10A2_UINT
},
459 {PIPE_FORMAT_B10G10R10X2_UNORM
, B10G10R10X2_UNORM
},
462 {PIPE_FORMAT_R11G11B10_FLOAT
, R11G11B10_FLOAT
},
464 /* 32 bits per component */
465 {PIPE_FORMAT_R32_FLOAT
, R32_FLOAT
},
466 {PIPE_FORMAT_R32G32_FLOAT
, R32G32_FLOAT
},
467 {PIPE_FORMAT_R32G32B32_FLOAT
, R32G32B32_FLOAT
},
468 {PIPE_FORMAT_R32G32B32A32_FLOAT
, R32G32B32A32_FLOAT
},
469 {PIPE_FORMAT_R32G32B32X32_FLOAT
, R32G32B32X32_FLOAT
},
471 {PIPE_FORMAT_R32_USCALED
, R32_USCALED
},
472 {PIPE_FORMAT_R32G32_USCALED
, R32G32_USCALED
},
473 {PIPE_FORMAT_R32G32B32_USCALED
, R32G32B32_USCALED
},
474 {PIPE_FORMAT_R32G32B32A32_USCALED
, R32G32B32A32_USCALED
},
476 {PIPE_FORMAT_R32_SSCALED
, R32_SSCALED
},
477 {PIPE_FORMAT_R32G32_SSCALED
, R32G32_SSCALED
},
478 {PIPE_FORMAT_R32G32B32_SSCALED
, R32G32B32_SSCALED
},
479 {PIPE_FORMAT_R32G32B32A32_SSCALED
, R32G32B32A32_SSCALED
},
481 {PIPE_FORMAT_R32_UINT
, R32_UINT
},
482 {PIPE_FORMAT_R32G32_UINT
, R32G32_UINT
},
483 {PIPE_FORMAT_R32G32B32_UINT
, R32G32B32_UINT
},
484 {PIPE_FORMAT_R32G32B32A32_UINT
, R32G32B32A32_UINT
},
486 {PIPE_FORMAT_R32_SINT
, R32_SINT
},
487 {PIPE_FORMAT_R32G32_SINT
, R32G32_SINT
},
488 {PIPE_FORMAT_R32G32B32_SINT
, R32G32B32_SINT
},
489 {PIPE_FORMAT_R32G32B32A32_SINT
, R32G32B32A32_SINT
},
491 /* 16 bits per component */
492 {PIPE_FORMAT_R16_UNORM
, R16_UNORM
},
493 {PIPE_FORMAT_R16G16_UNORM
, R16G16_UNORM
},
494 {PIPE_FORMAT_R16G16B16_UNORM
, R16G16B16_UNORM
},
495 {PIPE_FORMAT_R16G16B16A16_UNORM
, R16G16B16A16_UNORM
},
496 {PIPE_FORMAT_R16G16B16X16_UNORM
, R16G16B16X16_UNORM
},
498 {PIPE_FORMAT_R16_USCALED
, R16_USCALED
},
499 {PIPE_FORMAT_R16G16_USCALED
, R16G16_USCALED
},
500 {PIPE_FORMAT_R16G16B16_USCALED
, R16G16B16_USCALED
},
501 {PIPE_FORMAT_R16G16B16A16_USCALED
, R16G16B16A16_USCALED
},
503 {PIPE_FORMAT_R16_SNORM
, R16_SNORM
},
504 {PIPE_FORMAT_R16G16_SNORM
, R16G16_SNORM
},
505 {PIPE_FORMAT_R16G16B16_SNORM
, R16G16B16_SNORM
},
506 {PIPE_FORMAT_R16G16B16A16_SNORM
, R16G16B16A16_SNORM
},
508 {PIPE_FORMAT_R16_SSCALED
, R16_SSCALED
},
509 {PIPE_FORMAT_R16G16_SSCALED
, R16G16_SSCALED
},
510 {PIPE_FORMAT_R16G16B16_SSCALED
, R16G16B16_SSCALED
},
511 {PIPE_FORMAT_R16G16B16A16_SSCALED
, R16G16B16A16_SSCALED
},
513 {PIPE_FORMAT_R16_UINT
, R16_UINT
},
514 {PIPE_FORMAT_R16G16_UINT
, R16G16_UINT
},
515 {PIPE_FORMAT_R16G16B16_UINT
, R16G16B16_UINT
},
516 {PIPE_FORMAT_R16G16B16A16_UINT
, R16G16B16A16_UINT
},
518 {PIPE_FORMAT_R16_SINT
, R16_SINT
},
519 {PIPE_FORMAT_R16G16_SINT
, R16G16_SINT
},
520 {PIPE_FORMAT_R16G16B16_SINT
, R16G16B16_SINT
},
521 {PIPE_FORMAT_R16G16B16A16_SINT
, R16G16B16A16_SINT
},
523 {PIPE_FORMAT_R16_FLOAT
, R16_FLOAT
},
524 {PIPE_FORMAT_R16G16_FLOAT
, R16G16_FLOAT
},
525 {PIPE_FORMAT_R16G16B16_FLOAT
, R16G16B16_FLOAT
},
526 {PIPE_FORMAT_R16G16B16A16_FLOAT
, R16G16B16A16_FLOAT
},
527 {PIPE_FORMAT_R16G16B16X16_FLOAT
, R16G16B16X16_FLOAT
},
529 /* 8 bits per component */
530 {PIPE_FORMAT_R8_UNORM
, R8_UNORM
},
531 {PIPE_FORMAT_R8G8_UNORM
, R8G8_UNORM
},
532 {PIPE_FORMAT_R8G8B8_UNORM
, R8G8B8_UNORM
},
533 {PIPE_FORMAT_R8G8B8_SRGB
, R8G8B8_UNORM_SRGB
},
534 {PIPE_FORMAT_R8G8B8A8_UNORM
, R8G8B8A8_UNORM
},
535 {PIPE_FORMAT_R8G8B8A8_SRGB
, R8G8B8A8_UNORM_SRGB
},
536 {PIPE_FORMAT_R8G8B8X8_UNORM
, R8G8B8X8_UNORM
},
537 {PIPE_FORMAT_R8G8B8X8_SRGB
, R8G8B8X8_UNORM_SRGB
},
539 {PIPE_FORMAT_R8_USCALED
, R8_USCALED
},
540 {PIPE_FORMAT_R8G8_USCALED
, R8G8_USCALED
},
541 {PIPE_FORMAT_R8G8B8_USCALED
, R8G8B8_USCALED
},
542 {PIPE_FORMAT_R8G8B8A8_USCALED
, R8G8B8A8_USCALED
},
544 {PIPE_FORMAT_R8_SNORM
, R8_SNORM
},
545 {PIPE_FORMAT_R8G8_SNORM
, R8G8_SNORM
},
546 {PIPE_FORMAT_R8G8B8_SNORM
, R8G8B8_SNORM
},
547 {PIPE_FORMAT_R8G8B8A8_SNORM
, R8G8B8A8_SNORM
},
549 {PIPE_FORMAT_R8_SSCALED
, R8_SSCALED
},
550 {PIPE_FORMAT_R8G8_SSCALED
, R8G8_SSCALED
},
551 {PIPE_FORMAT_R8G8B8_SSCALED
, R8G8B8_SSCALED
},
552 {PIPE_FORMAT_R8G8B8A8_SSCALED
, R8G8B8A8_SSCALED
},
554 {PIPE_FORMAT_R8_UINT
, R8_UINT
},
555 {PIPE_FORMAT_R8G8_UINT
, R8G8_UINT
},
556 {PIPE_FORMAT_R8G8B8_UINT
, R8G8B8_UINT
},
557 {PIPE_FORMAT_R8G8B8A8_UINT
, R8G8B8A8_UINT
},
559 {PIPE_FORMAT_R8_SINT
, R8_SINT
},
560 {PIPE_FORMAT_R8G8_SINT
, R8G8_SINT
},
561 {PIPE_FORMAT_R8G8B8_SINT
, R8G8B8_SINT
},
562 {PIPE_FORMAT_R8G8B8A8_SINT
, R8G8B8A8_SINT
},
564 /* These formats are valid for vertex data, but should not be used
565 * for render targets.
568 {PIPE_FORMAT_R32_FIXED
, R32_SFIXED
},
569 {PIPE_FORMAT_R32G32_FIXED
, R32G32_SFIXED
},
570 {PIPE_FORMAT_R32G32B32_FIXED
, R32G32B32_SFIXED
},
571 {PIPE_FORMAT_R32G32B32A32_FIXED
, R32G32B32A32_SFIXED
},
573 {PIPE_FORMAT_R64_FLOAT
, R64_FLOAT
},
574 {PIPE_FORMAT_R64G64_FLOAT
, R64G64_FLOAT
},
575 {PIPE_FORMAT_R64G64B64_FLOAT
, R64G64B64_FLOAT
},
576 {PIPE_FORMAT_R64G64B64A64_FLOAT
, R64G64B64A64_FLOAT
},
578 /* These formats have entries in SWR but don't have Load/StoreTile
579 * implementations. That means these aren't renderable, and thus having
580 * a mapping entry here is detrimental.
584 {PIPE_FORMAT_L8_UNORM, L8_UNORM},
585 {PIPE_FORMAT_I8_UNORM, I8_UNORM},
586 {PIPE_FORMAT_L8A8_UNORM, L8A8_UNORM},
587 {PIPE_FORMAT_L16_UNORM, L16_UNORM},
588 {PIPE_FORMAT_UYVY, YCRCB_SWAPUVY},
590 {PIPE_FORMAT_L8_SRGB, L8_UNORM_SRGB},
591 {PIPE_FORMAT_L8A8_SRGB, L8A8_UNORM_SRGB},
593 {PIPE_FORMAT_DXT1_RGBA, BC1_UNORM},
594 {PIPE_FORMAT_DXT3_RGBA, BC2_UNORM},
595 {PIPE_FORMAT_DXT5_RGBA, BC3_UNORM},
597 {PIPE_FORMAT_DXT1_SRGBA, BC1_UNORM_SRGB},
598 {PIPE_FORMAT_DXT3_SRGBA, BC2_UNORM_SRGB},
599 {PIPE_FORMAT_DXT5_SRGBA, BC3_UNORM_SRGB},
601 {PIPE_FORMAT_RGTC1_UNORM, BC4_UNORM},
602 {PIPE_FORMAT_RGTC1_SNORM, BC4_SNORM},
603 {PIPE_FORMAT_RGTC2_UNORM, BC5_UNORM},
604 {PIPE_FORMAT_RGTC2_SNORM, BC5_SNORM},
606 {PIPE_FORMAT_L16A16_UNORM, L16A16_UNORM},
607 {PIPE_FORMAT_I16_UNORM, I16_UNORM},
608 {PIPE_FORMAT_L16_FLOAT, L16_FLOAT},
609 {PIPE_FORMAT_L16A16_FLOAT, L16A16_FLOAT},
610 {PIPE_FORMAT_I16_FLOAT, I16_FLOAT},
611 {PIPE_FORMAT_L32_FLOAT, L32_FLOAT},
612 {PIPE_FORMAT_L32A32_FLOAT, L32A32_FLOAT},
613 {PIPE_FORMAT_I32_FLOAT, I32_FLOAT},
615 {PIPE_FORMAT_I8_UINT, I8_UINT},
616 {PIPE_FORMAT_L8_UINT, L8_UINT},
617 {PIPE_FORMAT_L8A8_UINT, L8A8_UINT},
619 {PIPE_FORMAT_I8_SINT, I8_SINT},
620 {PIPE_FORMAT_L8_SINT, L8_SINT},
621 {PIPE_FORMAT_L8A8_SINT, L8A8_SINT},
626 auto it
= mesa2swr
.find(format
);
627 if (it
== mesa2swr
.end())
628 return (SWR_FORMAT
)-1;
634 swr_displaytarget_layout(struct swr_screen
*screen
, struct swr_resource
*res
)
636 struct sw_winsys
*winsys
= screen
->winsys
;
637 struct sw_displaytarget
*dt
;
639 const unsigned width
= align(res
->swr
.width
, res
->swr
.halign
);
640 const unsigned height
= align(res
->swr
.height
, res
->swr
.valign
);
643 dt
= winsys
->displaytarget_create(winsys
,
653 void *map
= winsys
->displaytarget_map(winsys
, dt
, 0);
655 res
->display_target
= dt
;
656 res
->swr
.xpBaseAddress
= (gfxptr_t
)map
;
658 /* Clear the display target surface */
660 memset(map
, 0, height
* stride
);
662 winsys
->displaytarget_unmap(winsys
, dt
);
668 swr_texture_layout(struct swr_screen
*screen
,
669 struct swr_resource
*res
,
672 struct pipe_resource
*pt
= &res
->base
;
674 pipe_format fmt
= pt
->format
;
675 const struct util_format_description
*desc
= util_format_description(fmt
);
677 res
->has_depth
= util_format_has_depth(desc
);
678 res
->has_stencil
= util_format_has_stencil(desc
);
680 if (res
->has_stencil
&& !res
->has_depth
)
681 fmt
= PIPE_FORMAT_R8_UINT
;
683 /* We always use the SWR layout. For 2D and 3D textures this looks like:
685 * |<------- pitch ------->|
686 * +=======================+-------
692 * +-----------+-----------+ |
694 * | Level 1 | L3L3 | |
696 * +===========+===========+-------
702 * +-----------+-----------+
706 * +===========+===========+
708 * The overall width in bytes is known as the pitch, while the overall
709 * height in rows is the qpitch. Array slices are laid out logically below
710 * one another, qpitch rows apart. For 3D surfaces, the "level" values are
711 * just invalid for the higher array numbers (since depth is also
712 * minified). 1D and 1D array surfaces are stored effectively the same way,
713 * except that pitch never plays into it. All the levels are logically
714 * adjacent to each other on the X axis. The qpitch becomes the number of
715 * elements between array slices, while the pitch is unused.
717 * Each level's sizes are subject to the valign and halign settings of the
718 * surface. For compressed formats that swr is unaware of, we will use an
719 * appropriately-sized uncompressed format, and scale the widths/heights.
721 * This surface is stored inside res->swr. For depth/stencil textures,
722 * res->secondary will have an identically-laid-out but R8_UINT-formatted
723 * stencil tree. In the Z32F_S8 case, the primary surface still has 64-bpp
724 * texels, to simplify map/unmap logic which copies the stencil values
728 res
->swr
.width
= pt
->width0
;
729 res
->swr
.height
= pt
->height0
;
730 res
->swr
.type
= swr_convert_target_type(pt
->target
);
731 res
->swr
.tileMode
= SWR_TILE_NONE
;
732 res
->swr
.format
= mesa_to_swr_format(fmt
);
733 res
->swr
.numSamples
= std::max(1u, pt
->nr_samples
);
735 if (pt
->bind
& (PIPE_BIND_RENDER_TARGET
| PIPE_BIND_DEPTH_STENCIL
)) {
736 res
->swr
.halign
= KNOB_MACROTILE_X_DIM
;
737 res
->swr
.valign
= KNOB_MACROTILE_Y_DIM
;
739 /* If SWR_MSAA_FORCE_ENABLE is set, turn on MSAA and override requested
740 * surface sample count. */
741 if (screen
->msaa_force_enable
) {
742 res
->swr
.numSamples
= screen
->msaa_max_count
;
743 fprintf(stderr
,"swr_texture_layout: forcing sample count: %d\n",
744 res
->swr
.numSamples
);
751 unsigned halign
= res
->swr
.halign
* util_format_get_blockwidth(fmt
);
752 unsigned width
= align(pt
->width0
, halign
);
753 if (pt
->target
== PIPE_TEXTURE_1D
|| pt
->target
== PIPE_TEXTURE_1D_ARRAY
) {
754 for (int level
= 1; level
<= pt
->last_level
; level
++)
755 width
+= align(u_minify(pt
->width0
, level
), halign
);
756 res
->swr
.pitch
= util_format_get_blocksize(fmt
);
757 res
->swr
.qpitch
= util_format_get_nblocksx(fmt
, width
);
759 // The pitch is the overall width of the texture in bytes. Most of the
760 // time this is the pitch of level 0 since all the other levels fit
761 // underneath it. However in some degenerate situations, the width of
762 // level1 + level2 may be larger. In that case, we use those
763 // widths. This can happen if, e.g. halign is 32, and the width of level
764 // 0 is 32 or less. In that case, the aligned levels 1 and 2 will also
765 // be 32 each, adding up to 64.
766 unsigned valign
= res
->swr
.valign
* util_format_get_blockheight(fmt
);
767 if (pt
->last_level
> 1) {
768 width
= std::max
<uint32_t>(
770 align(u_minify(pt
->width0
, 1), halign
) +
771 align(u_minify(pt
->width0
, 2), halign
));
773 res
->swr
.pitch
= util_format_get_stride(fmt
, width
);
775 // The qpitch is controlled by either the height of the second LOD, or
776 // the combination of all the later LODs.
777 unsigned height
= align(pt
->height0
, valign
);
778 if (pt
->last_level
== 1) {
779 height
+= align(u_minify(pt
->height0
, 1), valign
);
780 } else if (pt
->last_level
> 1) {
781 unsigned level1
= align(u_minify(pt
->height0
, 1), valign
);
783 for (int level
= 2; level
<= pt
->last_level
; level
++) {
784 level2
+= align(u_minify(pt
->height0
, level
), valign
);
786 height
+= std::max(level1
, level2
);
788 res
->swr
.qpitch
= util_format_get_nblocksy(fmt
, height
);
791 if (pt
->target
== PIPE_TEXTURE_3D
)
792 res
->swr
.depth
= pt
->depth0
;
794 res
->swr
.depth
= pt
->array_size
;
796 // Fix up swr format if necessary so that LOD offset computation works
797 if (res
->swr
.format
== (SWR_FORMAT
)-1) {
798 switch (util_format_get_blocksize(fmt
)) {
800 unreachable("Unexpected format block size");
801 case 1: res
->swr
.format
= R8_UINT
; break;
802 case 2: res
->swr
.format
= R16_UINT
; break;
803 case 4: res
->swr
.format
= R32_UINT
; break;
805 if (util_format_is_compressed(fmt
))
806 res
->swr
.format
= BC4_UNORM
;
808 res
->swr
.format
= R32G32_UINT
;
811 if (util_format_is_compressed(fmt
))
812 res
->swr
.format
= BC5_UNORM
;
814 res
->swr
.format
= R32G32B32A32_UINT
;
819 for (int level
= 0; level
<= pt
->last_level
; level
++) {
820 res
->mip_offsets
[level
] =
821 ComputeSurfaceOffset
<false>(0, 0, 0, 0, 0, level
, &res
->swr
);
824 size_t total_size
= res
->swr
.depth
* res
->swr
.qpitch
* res
->swr
.pitch
*
826 if (total_size
> SWR_MAX_TEXTURE_SIZE
)
830 res
->swr
.xpBaseAddress
= (gfxptr_t
)AlignedMalloc(total_size
, 64);
832 if (res
->has_depth
&& res
->has_stencil
) {
833 res
->secondary
= res
->swr
;
834 res
->secondary
.format
= R8_UINT
;
835 res
->secondary
.pitch
= res
->swr
.pitch
/ util_format_get_blocksize(fmt
);
837 for (int level
= 0; level
<= pt
->last_level
; level
++) {
838 res
->secondary_mip_offsets
[level
] =
839 ComputeSurfaceOffset
<false>(0, 0, 0, 0, 0, level
, &res
->secondary
);
842 total_size
= res
->secondary
.depth
* res
->secondary
.qpitch
*
843 res
->secondary
.pitch
* res
->secondary
.numSamples
;
845 res
->secondary
.xpBaseAddress
= (gfxptr_t
) AlignedMalloc(total_size
, 64);
853 swr_can_create_resource(struct pipe_screen
*screen
,
854 const struct pipe_resource
*templat
)
856 struct swr_resource res
;
857 memset(&res
, 0, sizeof(res
));
859 return swr_texture_layout(swr_screen(screen
), &res
, false);
862 /* Helper function that conditionally creates a single-sample resolve resource
863 * and attaches it to main multisample resource. */
865 swr_create_resolve_resource(struct pipe_screen
*_screen
,
866 struct swr_resource
*msaa_res
)
868 struct swr_screen
*screen
= swr_screen(_screen
);
870 /* If resource is multisample, create a single-sample resolve resource */
871 if (msaa_res
->base
.nr_samples
> 1 || (screen
->msaa_force_enable
&&
872 !(msaa_res
->base
.flags
& SWR_RESOURCE_FLAG_ALT_SURFACE
))) {
874 /* Create a single-sample copy of the resource. Copy the original
875 * resource parameters and set flag to prevent recursion when re-calling
877 struct pipe_resource alt_template
= msaa_res
->base
;
878 alt_template
.nr_samples
= 0;
879 alt_template
.flags
|= SWR_RESOURCE_FLAG_ALT_SURFACE
;
881 /* Note: Display_target is a special single-sample resource, only the
882 * display_target has been created already. */
883 if (msaa_res
->base
.bind
& (PIPE_BIND_DISPLAY_TARGET
| PIPE_BIND_SCANOUT
884 | PIPE_BIND_SHARED
)) {
885 /* Allocate the multisample buffers. */
886 if (!swr_texture_layout(screen
, msaa_res
, true))
889 /* Alt resource will only be bound as PIPE_BIND_RENDER_TARGET
890 * remove the DISPLAY_TARGET, SCANOUT, and SHARED bindings */
891 alt_template
.bind
= PIPE_BIND_RENDER_TARGET
;
894 /* Allocate single-sample resolve surface */
895 struct pipe_resource
*alt
;
896 alt
= _screen
->resource_create(_screen
, &alt_template
);
900 /* Attach it to the multisample resource */
901 msaa_res
->resolve_target
= alt
;
903 /* Hang resolve surface state off the multisample surface state to so
904 * StoreTiles knows where to resolve the surface. */
905 msaa_res
->swr
.xpAuxBaseAddress
= (gfxptr_t
)&swr_resource(alt
)->swr
;
908 return true; /* success */
911 static struct pipe_resource
*
912 swr_resource_create(struct pipe_screen
*_screen
,
913 const struct pipe_resource
*templat
)
915 struct swr_screen
*screen
= swr_screen(_screen
);
916 struct swr_resource
*res
= CALLOC_STRUCT(swr_resource
);
920 res
->base
= *templat
;
921 pipe_reference_init(&res
->base
.reference
, 1);
922 res
->base
.screen
= &screen
->base
;
924 if (swr_resource_is_texture(&res
->base
)) {
925 if (res
->base
.bind
& (PIPE_BIND_DISPLAY_TARGET
| PIPE_BIND_SCANOUT
926 | PIPE_BIND_SHARED
)) {
927 /* displayable surface
928 * first call swr_texture_layout without allocating to finish
929 * filling out the SWR_SURFACE_STATE in res */
930 swr_texture_layout(screen
, res
, false);
931 if (!swr_displaytarget_layout(screen
, res
))
935 if (!swr_texture_layout(screen
, res
, true))
939 /* If resource was multisample, create resolve resource and attach
940 * it to multisample resource. */
941 if (!swr_create_resolve_resource(_screen
, res
))
945 /* other data (vertex buffer, const buffer, etc) */
946 assert(util_format_get_blocksize(templat
->format
) == 1);
947 assert(templat
->height0
== 1);
948 assert(templat
->depth0
== 1);
949 assert(templat
->last_level
== 0);
951 /* Easiest to just call swr_texture_layout, as it sets up
952 * SWR_SURFACE_STATE in res */
953 if (!swr_texture_layout(screen
, res
, true))
965 swr_resource_destroy(struct pipe_screen
*p_screen
, struct pipe_resource
*pt
)
967 struct swr_screen
*screen
= swr_screen(p_screen
);
968 struct swr_resource
*spr
= swr_resource(pt
);
970 if (spr
->display_target
) {
971 /* If resource is display target, winsys manages the buffer and will
972 * free it on displaytarget_destroy. */
973 swr_fence_finish(p_screen
, NULL
, screen
->flush_fence
, 0);
975 struct sw_winsys
*winsys
= screen
->winsys
;
976 winsys
->displaytarget_destroy(winsys
, spr
->display_target
);
978 if (spr
->swr
.numSamples
> 1) {
979 /* Free an attached resolve resource */
980 struct swr_resource
*alt
= swr_resource(spr
->resolve_target
);
981 swr_fence_work_free(screen
->flush_fence
, (void*)(alt
->swr
.xpBaseAddress
), true);
983 /* Free multisample buffer */
984 swr_fence_work_free(screen
->flush_fence
, (void*)(spr
->swr
.xpBaseAddress
), true);
987 /* For regular resources, defer deletion */
988 swr_resource_unused(pt
);
990 if (spr
->swr
.numSamples
> 1) {
991 /* Free an attached resolve resource */
992 struct swr_resource
*alt
= swr_resource(spr
->resolve_target
);
993 swr_fence_work_free(screen
->flush_fence
, (void*)(alt
->swr
.xpBaseAddress
), true);
996 swr_fence_work_free(screen
->flush_fence
, (void*)(spr
->swr
.xpBaseAddress
), true);
997 swr_fence_work_free(screen
->flush_fence
,
998 (void*)(spr
->secondary
.xpBaseAddress
), true);
1000 /* If work queue grows too large, submit a fence to force queue to
1001 * drain. This is mainly to decrease the amount of memory used by the
1002 * piglit streaming-texture-leak test */
1003 if (screen
->pipe
&& swr_fence(screen
->flush_fence
)->work
.count
> 64)
1004 swr_fence_submit(swr_context(screen
->pipe
), screen
->flush_fence
);
1012 swr_flush_frontbuffer(struct pipe_screen
*p_screen
,
1013 struct pipe_resource
*resource
,
1016 void *context_private
,
1017 struct pipe_box
*sub_box
)
1019 struct swr_screen
*screen
= swr_screen(p_screen
);
1020 struct sw_winsys
*winsys
= screen
->winsys
;
1021 struct swr_resource
*spr
= swr_resource(resource
);
1022 struct pipe_context
*pipe
= screen
->pipe
;
1023 struct swr_context
*ctx
= swr_context(pipe
);
1026 swr_fence_finish(p_screen
, NULL
, screen
->flush_fence
, 0);
1027 swr_resource_unused(resource
);
1028 ctx
->api
.pfnSwrEndFrame(ctx
->swrContext
);
1031 /* Multisample resolved into resolve_target at flush with store_resource */
1032 if (pipe
&& spr
->swr
.numSamples
> 1) {
1033 struct pipe_resource
*resolve_target
= spr
->resolve_target
;
1035 /* Once resolved, copy into display target */
1036 SWR_SURFACE_STATE
*resolve
= &swr_resource(resolve_target
)->swr
;
1038 void *map
= winsys
->displaytarget_map(winsys
, spr
->display_target
,
1039 PIPE_TRANSFER_WRITE
);
1040 memcpy(map
, (void*)(resolve
->xpBaseAddress
), resolve
->pitch
* resolve
->height
);
1041 winsys
->displaytarget_unmap(winsys
, spr
->display_target
);
1044 debug_assert(spr
->display_target
);
1045 if (spr
->display_target
)
1046 winsys
->displaytarget_display(
1047 winsys
, spr
->display_target
, context_private
, sub_box
);
1052 swr_destroy_screen(struct pipe_screen
*p_screen
)
1054 struct swr_screen
*screen
= swr_screen(p_screen
);
1055 struct sw_winsys
*winsys
= screen
->winsys
;
1057 fprintf(stderr
, "SWR destroy screen!\n");
1059 swr_fence_finish(p_screen
, NULL
, screen
->flush_fence
, 0);
1060 swr_fence_reference(p_screen
, &screen
->flush_fence
, NULL
);
1062 JitDestroyContext(screen
->hJitMgr
);
1064 if (winsys
->destroy
)
1065 winsys
->destroy(winsys
);
1072 swr_validate_env_options(struct swr_screen
*screen
)
1074 /* The client_copy_limit sets a maximum on the amount of user-buffer memory
1075 * copied to scratch space on a draw. Past this, the draw will access
1076 * user-buffer directly and then block. This is faster than queuing many
1077 * large client draws. */
1078 screen
->client_copy_limit
= SWR_CLIENT_COPY_LIMIT
;
1079 int client_copy_limit
=
1080 debug_get_num_option("SWR_CLIENT_COPY_LIMIT", SWR_CLIENT_COPY_LIMIT
);
1081 if (client_copy_limit
> 0)
1082 screen
->client_copy_limit
= client_copy_limit
;
1084 /* XXX msaa under development, disable by default for now */
1085 screen
->msaa_max_count
= 1; /* was SWR_MAX_NUM_MULTISAMPLES; */
1087 /* validate env override values, within range and power of 2 */
1088 int msaa_max_count
= debug_get_num_option("SWR_MSAA_MAX_COUNT", 1);
1089 if (msaa_max_count
!= 1) {
1090 if ((msaa_max_count
< 1) || (msaa_max_count
> SWR_MAX_NUM_MULTISAMPLES
)
1091 || !util_is_power_of_two(msaa_max_count
)) {
1092 fprintf(stderr
, "SWR_MSAA_MAX_COUNT invalid: %d\n", msaa_max_count
);
1093 fprintf(stderr
, "must be power of 2 between 1 and %d" \
1094 " (or 1 to disable msaa)\n",
1095 SWR_MAX_NUM_MULTISAMPLES
);
1099 fprintf(stderr
, "SWR_MSAA_MAX_COUNT: %d\n", msaa_max_count
);
1100 if (msaa_max_count
== 1)
1101 fprintf(stderr
, "(msaa disabled)\n");
1103 screen
->msaa_max_count
= msaa_max_count
;
1106 screen
->msaa_force_enable
= debug_get_bool_option(
1107 "SWR_MSAA_FORCE_ENABLE", false);
1108 if (screen
->msaa_force_enable
)
1109 fprintf(stderr
, "SWR_MSAA_FORCE_ENABLE: true\n");
1114 struct pipe_screen
*
1115 swr_create_screen_internal(struct sw_winsys
*winsys
)
1117 struct swr_screen
*screen
= CALLOC_STRUCT(swr_screen
);
1122 if (!lp_build_init()) {
1127 screen
->winsys
= winsys
;
1128 screen
->base
.get_name
= swr_get_name
;
1129 screen
->base
.get_vendor
= swr_get_vendor
;
1130 screen
->base
.is_format_supported
= swr_is_format_supported
;
1131 screen
->base
.context_create
= swr_create_context
;
1132 screen
->base
.can_create_resource
= swr_can_create_resource
;
1134 screen
->base
.destroy
= swr_destroy_screen
;
1135 screen
->base
.get_param
= swr_get_param
;
1136 screen
->base
.get_shader_param
= swr_get_shader_param
;
1137 screen
->base
.get_paramf
= swr_get_paramf
;
1139 screen
->base
.resource_create
= swr_resource_create
;
1140 screen
->base
.resource_destroy
= swr_resource_destroy
;
1142 screen
->base
.flush_frontbuffer
= swr_flush_frontbuffer
;
1144 // Pass in "" for architecture for run-time determination
1145 screen
->hJitMgr
= JitCreateContext(KNOB_SIMD_WIDTH
, "", "swr");
1147 swr_fence_init(&screen
->base
);
1149 util_format_s3tc_init();
1151 swr_validate_env_options(screen
);
1153 return &screen
->base
;