2 * Copyright (c) 2012-2015 Etnaviv Project
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, sub license,
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
12 * next paragraph) shall be included in all copies or substantial portions
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 NON-INFRINGEMENT. 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
21 * DEALINGS IN THE SOFTWARE.
24 * Wladimir J. van der Laan <laanwj@gmail.com>
25 * Christian Gmeiner <christian.gmeiner@gmail.com>
28 #include "etnaviv_screen.h"
30 #include "hw/common.xml.h"
32 #include "etnaviv_compiler.h"
33 #include "etnaviv_context.h"
34 #include "etnaviv_debug.h"
35 #include "etnaviv_fence.h"
36 #include "etnaviv_format.h"
37 #include "etnaviv_query.h"
38 #include "etnaviv_resource.h"
39 #include "etnaviv_translate.h"
41 #include "util/hash_table.h"
42 #include "util/os_time.h"
43 #include "util/u_math.h"
44 #include "util/u_memory.h"
45 #include "util/u_screen.h"
46 #include "util/u_string.h"
48 #include "state_tracker/drm_driver.h"
50 #include "drm-uapi/drm_fourcc.h"
52 #define ETNA_DRM_VERSION(major, minor) ((major) << 16 | (minor))
53 #define ETNA_DRM_VERSION_FENCE_FD ETNA_DRM_VERSION(1, 1)
54 #define ETNA_DRM_VERSION_PERFMON ETNA_DRM_VERSION(1, 2)
56 static const struct debug_named_value debug_options
[] = {
57 {"dbg_msgs", ETNA_DBG_MSGS
, "Print debug messages"},
58 {"frame_msgs", ETNA_DBG_FRAME_MSGS
, "Print frame messages"},
59 {"resource_msgs", ETNA_DBG_RESOURCE_MSGS
, "Print resource messages"},
60 {"compiler_msgs", ETNA_DBG_COMPILER_MSGS
, "Print compiler messages"},
61 {"linker_msgs", ETNA_DBG_LINKER_MSGS
, "Print linker messages"},
62 {"dump_shaders", ETNA_DBG_DUMP_SHADERS
, "Dump shaders"},
63 {"no_ts", ETNA_DBG_NO_TS
, "Disable TS"},
64 {"no_autodisable", ETNA_DBG_NO_AUTODISABLE
, "Disable autodisable"},
65 {"no_supertile", ETNA_DBG_NO_SUPERTILE
, "Disable supertiles"},
66 {"no_early_z", ETNA_DBG_NO_EARLY_Z
, "Disable early z"},
67 {"cflush_all", ETNA_DBG_CFLUSH_ALL
, "Flush every cache before state update"},
68 {"msaa2x", ETNA_DBG_MSAA_2X
, "Force 2x msaa"},
69 {"msaa4x", ETNA_DBG_MSAA_4X
, "Force 4x msaa"},
70 {"flush_all", ETNA_DBG_FLUSH_ALL
, "Flush after every rendered primitive"},
71 {"zero", ETNA_DBG_ZERO
, "Zero all resources after allocation"},
72 {"draw_stall", ETNA_DBG_DRAW_STALL
, "Stall FE/PE after each rendered primitive"},
73 {"shaderdb", ETNA_DBG_SHADERDB
, "Enable shaderdb output"},
74 {"no_singlebuffer",ETNA_DBG_NO_SINGLEBUF
, "Disable single buffer feature"},
75 {"nir", ETNA_DBG_NIR
, "use new NIR compiler"},
76 {"deqp", ETNA_DBG_DEQP
, "Hacks to run dEQP GLES3 tests"}, /* needs MESA_GLES_VERSION_OVERRIDE=3.0 */
80 DEBUG_GET_ONCE_FLAGS_OPTION(etna_mesa_debug
, "ETNA_MESA_DEBUG", debug_options
, 0)
81 int etna_mesa_debug
= 0;
84 etna_screen_destroy(struct pipe_screen
*pscreen
)
86 struct etna_screen
*screen
= etna_screen(pscreen
);
89 etna_perfmon_del(screen
->perfmon
);
92 etna_pipe_del(screen
->pipe
);
95 etna_gpu_del(screen
->gpu
);
101 etna_device_del(screen
->dev
);
107 etna_screen_get_name(struct pipe_screen
*pscreen
)
109 struct etna_screen
*priv
= etna_screen(pscreen
);
110 static char buffer
[128];
112 snprintf(buffer
, sizeof(buffer
), "Vivante GC%x rev %04x", priv
->model
,
119 etna_screen_get_vendor(struct pipe_screen
*pscreen
)
125 etna_screen_get_device_vendor(struct pipe_screen
*pscreen
)
131 etna_screen_get_param(struct pipe_screen
*pscreen
, enum pipe_cap param
)
133 struct etna_screen
*screen
= etna_screen(pscreen
);
136 /* Supported features (boolean caps). */
137 case PIPE_CAP_ANISOTROPIC_FILTER
:
138 case PIPE_CAP_POINT_SPRITE
:
139 case PIPE_CAP_BLEND_EQUATION_SEPARATE
:
140 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT
:
141 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER
:
142 case PIPE_CAP_FRAGMENT_SHADER_TEXTURE_LOD
:
143 case PIPE_CAP_FRAGMENT_SHADER_DERIVATIVES
:
144 case PIPE_CAP_VERTEX_SHADER_SATURATE
:
145 case PIPE_CAP_TEXTURE_BARRIER
:
146 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION
:
147 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY
:
148 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY
:
149 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY
:
150 case PIPE_CAP_TGSI_TEXCOORD
:
151 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED
:
152 case PIPE_CAP_MIXED_COLOR_DEPTH_BITS
:
153 case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES
:
154 case PIPE_CAP_STRING_MARKER
:
155 case PIPE_CAP_SHAREABLE_SHADERS
:
157 case PIPE_CAP_NATIVE_FENCE_FD
:
158 return screen
->drm_version
>= ETNA_DRM_VERSION_FENCE_FD
;
159 case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL
:
160 case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL
: /* note: not integer */
161 return DBG_ENABLED(ETNA_DBG_NIR
);
162 case PIPE_CAP_TGSI_FS_POINT_IS_SYSVAL
:
166 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT
:
168 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT
:
169 return 4; /* XXX could easily be supported */
171 case PIPE_CAP_NPOT_TEXTURES
:
172 return true; /* VIV_FEATURE(priv->dev, chipMinorFeatures1,
173 NON_POWER_OF_TWO); */
175 case PIPE_CAP_TEXTURE_SWIZZLE
:
176 case PIPE_CAP_PRIMITIVE_RESTART
:
177 return VIV_FEATURE(screen
, chipMinorFeatures1
, HALTI0
);
179 /* Unsupported features. */
180 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT
:
181 case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY
:
182 case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION
:
186 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS
:
187 return DBG_ENABLED(ETNA_DBG_DEQP
) ? 4 : 0;
188 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS
:
189 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS
:
192 case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE
:
194 case PIPE_CAP_MAX_VERTEX_ELEMENT_SRC_OFFSET
:
196 case PIPE_CAP_MAX_VERTEX_BUFFERS
:
197 return screen
->specs
.stream_count
;
198 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR
:
199 return VIV_FEATURE(screen
, chipMinorFeatures4
, HALTI2
);
203 case PIPE_CAP_TEXTURE_SHADOW_MAP
:
204 return DBG_ENABLED(ETNA_DBG_NIR
) && screen
->specs
.halti
>= 2;
205 case PIPE_CAP_MAX_TEXTURE_2D_SIZE
:
206 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS
: /* TODO: verify */
207 return screen
->specs
.max_texture_size
;
208 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS
:
209 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS
:
211 int log2_max_tex_size
= util_last_bit(screen
->specs
.max_texture_size
);
212 assert(log2_max_tex_size
> 0);
213 return log2_max_tex_size
;
216 case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET
:
217 case PIPE_CAP_MIN_TEXEL_OFFSET
:
219 case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET
:
220 case PIPE_CAP_MAX_TEXEL_OFFSET
:
222 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE
:
223 return VIV_FEATURE(screen
, chipMinorFeatures2
, SEAMLESS_CUBE_MAP
);
226 case PIPE_CAP_OCCLUSION_QUERY
:
227 return VIV_FEATURE(screen
, chipMinorFeatures1
, HALTI0
);
228 case PIPE_CAP_QUERY_TIMESTAMP
:
232 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER
:
234 case PIPE_CAP_MAX_TEXTURE_UPLOAD_MEMORY_BUDGET
: {
235 /* etnaviv is being run on systems as small as 256MB total RAM so
236 * we need to provide a sane value for such a device. Limit the
237 * memory budget to min(~3% of pyhiscal memory, 64MB).
239 * a simple divison by 32 provides the numbers we want.
243 uint64_t system_memory
;
245 if (!os_get_total_physical_memory(&system_memory
))
246 system_memory
= 4096 << 20;
248 return MIN2(system_memory
/ 32, 64 * 1024 * 1024);
251 case PIPE_CAP_MAX_VARYINGS
:
252 return screen
->specs
.max_varyings
;
254 case PIPE_CAP_PCI_GROUP
:
255 case PIPE_CAP_PCI_BUS
:
256 case PIPE_CAP_PCI_DEVICE
:
257 case PIPE_CAP_PCI_FUNCTION
:
259 case PIPE_CAP_ACCELERATED
:
261 case PIPE_CAP_VIDEO_MEMORY
:
266 return u_pipe_screen_get_param_defaults(pscreen
, param
);
271 etna_screen_get_paramf(struct pipe_screen
*pscreen
, enum pipe_capf param
)
273 struct etna_screen
*screen
= etna_screen(pscreen
);
276 case PIPE_CAPF_MAX_LINE_WIDTH
:
277 case PIPE_CAPF_MAX_LINE_WIDTH_AA
:
278 case PIPE_CAPF_MAX_POINT_WIDTH
:
279 case PIPE_CAPF_MAX_POINT_WIDTH_AA
:
281 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY
:
283 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS
:
284 return util_last_bit(screen
->specs
.max_texture_size
);
285 case PIPE_CAPF_MIN_CONSERVATIVE_RASTER_DILATE
:
286 case PIPE_CAPF_MAX_CONSERVATIVE_RASTER_DILATE
:
287 case PIPE_CAPF_CONSERVATIVE_RASTER_DILATE_GRANULARITY
:
291 debug_printf("unknown paramf %d", param
);
296 etna_screen_get_shader_param(struct pipe_screen
*pscreen
,
297 enum pipe_shader_type shader
,
298 enum pipe_shader_cap param
)
300 struct etna_screen
*screen
= etna_screen(pscreen
);
301 bool ubo_enable
= screen
->specs
.halti
>= 2 && DBG_ENABLED(ETNA_DBG_NIR
);
303 if (DBG_ENABLED(ETNA_DBG_DEQP
))
307 case PIPE_SHADER_FRAGMENT
:
308 case PIPE_SHADER_VERTEX
:
310 case PIPE_SHADER_COMPUTE
:
311 case PIPE_SHADER_GEOMETRY
:
312 case PIPE_SHADER_TESS_CTRL
:
313 case PIPE_SHADER_TESS_EVAL
:
316 DBG("unknown shader type %d", shader
);
321 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS
:
322 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS
:
323 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS
:
324 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS
:
325 return ETNA_MAX_TOKENS
;
326 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH
:
327 return ETNA_MAX_DEPTH
; /* XXX */
328 case PIPE_SHADER_CAP_MAX_INPUTS
:
329 /* Maximum number of inputs for the vertex shader is the number
330 * of vertex elements - each element defines one vertex shader
331 * input register. For the fragment shader, this is the number
333 return shader
== PIPE_SHADER_FRAGMENT
? screen
->specs
.max_varyings
334 : screen
->specs
.vertex_max_elements
;
335 case PIPE_SHADER_CAP_MAX_OUTPUTS
:
336 return 16; /* see VIVS_VS_OUTPUT */
337 case PIPE_SHADER_CAP_MAX_TEMPS
:
338 return 64; /* Max native temporaries. */
339 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS
:
340 return ubo_enable
? ETNA_MAX_CONST_BUF
: 1;
341 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED
:
343 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR
:
344 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR
:
345 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR
:
346 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR
:
348 case PIPE_SHADER_CAP_SUBROUTINES
:
350 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED
:
351 return VIV_FEATURE(screen
, chipMinorFeatures0
, HAS_SQRT_TRIG
);
352 case PIPE_SHADER_CAP_INT64_ATOMICS
:
353 case PIPE_SHADER_CAP_FP16
:
355 case PIPE_SHADER_CAP_INTEGERS
:
356 return DBG_ENABLED(ETNA_DBG_NIR
) && screen
->specs
.halti
>= 2;
357 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS
:
358 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS
:
359 return shader
== PIPE_SHADER_FRAGMENT
360 ? screen
->specs
.fragment_sampler_count
361 : screen
->specs
.vertex_sampler_count
;
362 case PIPE_SHADER_CAP_PREFERRED_IR
:
363 return DBG_ENABLED(ETNA_DBG_NIR
) ? PIPE_SHADER_IR_NIR
: PIPE_SHADER_IR_TGSI
;
364 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE
:
366 return 16384; /* 16384 so state tracker enables UBOs */
367 return shader
== PIPE_SHADER_FRAGMENT
368 ? screen
->specs
.max_ps_uniforms
* sizeof(float[4])
369 : screen
->specs
.max_vs_uniforms
* sizeof(float[4]);
370 case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED
:
371 case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED
:
372 case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED
:
373 case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED
:
374 case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE
:
376 case PIPE_SHADER_CAP_SUPPORTED_IRS
:
378 case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT
:
380 case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS
:
381 case PIPE_SHADER_CAP_MAX_SHADER_IMAGES
:
382 case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD
:
383 case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS
:
384 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS
:
385 case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS
:
389 debug_printf("unknown shader param %d", param
);
394 etna_screen_get_timestamp(struct pipe_screen
*pscreen
)
396 return os_time_get_nano();
400 gpu_supports_texture_target(struct etna_screen
*screen
,
401 enum pipe_texture_target target
)
403 if (target
== PIPE_TEXTURE_CUBE_ARRAY
)
406 /* pre-halti has no array/3D */
407 if (screen
->specs
.halti
< 0 &&
408 (target
== PIPE_TEXTURE_1D_ARRAY
||
409 target
== PIPE_TEXTURE_2D_ARRAY
||
410 target
== PIPE_TEXTURE_3D
))
417 gpu_supports_texture_format(struct etna_screen
*screen
, uint32_t fmt
,
418 enum pipe_format format
)
420 bool supported
= true;
422 if (fmt
== TEXTURE_FORMAT_ETC1
)
423 supported
= VIV_FEATURE(screen
, chipFeatures
, ETC1_TEXTURE_COMPRESSION
);
425 if (fmt
>= TEXTURE_FORMAT_DXT1
&& fmt
<= TEXTURE_FORMAT_DXT4_DXT5
)
426 supported
= VIV_FEATURE(screen
, chipFeatures
, DXT_TEXTURE_COMPRESSION
);
428 if (util_format_is_srgb(format
))
429 supported
= VIV_FEATURE(screen
, chipMinorFeatures1
, HALTI0
);
431 if (fmt
& EXT_FORMAT
)
432 supported
= VIV_FEATURE(screen
, chipMinorFeatures1
, HALTI0
);
434 if (fmt
& ASTC_FORMAT
) {
435 supported
= screen
->specs
.tex_astc
;
438 if (util_format_is_snorm(format
))
439 supported
= VIV_FEATURE(screen
, chipMinorFeatures2
, HALTI1
);
441 if (format
!= PIPE_FORMAT_S8_UINT_Z24_UNORM
&&
442 (util_format_is_pure_integer(format
) || util_format_is_float(format
)))
443 supported
= VIV_FEATURE(screen
, chipMinorFeatures4
, HALTI2
);
449 if (texture_format_needs_swiz(format
))
450 return VIV_FEATURE(screen
, chipMinorFeatures1
, HALTI0
);
456 gpu_supports_render_format(struct etna_screen
*screen
, enum pipe_format format
,
457 unsigned sample_count
)
459 const uint32_t fmt
= translate_pe_format(format
);
461 if (fmt
== ETNA_NO_MATCH
)
464 /* Validate MSAA; number of samples must be allowed, and render target
465 * must have MSAA'able format. */
466 if (sample_count
> 1) {
467 if (!VIV_FEATURE(screen
, chipFeatures
, MSAA
))
469 if (!translate_samples_to_xyscale(sample_count
, NULL
, NULL
))
471 if (translate_ts_format(format
) == ETNA_NO_MATCH
)
475 if (format
== PIPE_FORMAT_R8_UNORM
)
476 return VIV_FEATURE(screen
, chipMinorFeatures5
, HALTI5
);
478 /* figure out 8bpp RS clear to enable these formats */
479 if (format
== PIPE_FORMAT_R8_SINT
|| format
== PIPE_FORMAT_R8_UINT
)
480 return VIV_FEATURE(screen
, chipMinorFeatures5
, HALTI5
);
482 if (util_format_is_srgb(format
))
483 return VIV_FEATURE(screen
, chipMinorFeatures5
, HALTI3
);
485 if (util_format_is_pure_integer(format
) || util_format_is_float(format
))
486 return VIV_FEATURE(screen
, chipMinorFeatures4
, HALTI2
);
488 if (format
== PIPE_FORMAT_R8G8_UNORM
)
489 return VIV_FEATURE(screen
, chipMinorFeatures4
, HALTI2
);
491 /* any other extended format is HALTI0 (only R10G10B10A2?) */
492 if (fmt
>= PE_FORMAT_R16F
)
493 return VIV_FEATURE(screen
, chipMinorFeatures1
, HALTI0
);
499 gpu_supports_vertex_format(struct etna_screen
*screen
, enum pipe_format format
)
501 if (translate_vertex_format_type(format
) == ETNA_NO_MATCH
)
504 if (util_format_is_pure_integer(format
))
505 return VIV_FEATURE(screen
, chipMinorFeatures4
, HALTI2
);
511 etna_screen_is_format_supported(struct pipe_screen
*pscreen
,
512 enum pipe_format format
,
513 enum pipe_texture_target target
,
514 unsigned sample_count
,
515 unsigned storage_sample_count
,
518 struct etna_screen
*screen
= etna_screen(pscreen
);
519 unsigned allowed
= 0;
521 if (!gpu_supports_texture_target(screen
, target
))
524 if (MAX2(1, sample_count
) != MAX2(1, storage_sample_count
))
527 if (usage
& PIPE_BIND_RENDER_TARGET
) {
528 if (gpu_supports_render_format(screen
, format
, sample_count
))
529 allowed
|= PIPE_BIND_RENDER_TARGET
;
532 if (usage
& PIPE_BIND_DEPTH_STENCIL
) {
533 if (translate_depth_format(format
) != ETNA_NO_MATCH
)
534 allowed
|= PIPE_BIND_DEPTH_STENCIL
;
537 if (usage
& PIPE_BIND_SAMPLER_VIEW
) {
538 uint32_t fmt
= translate_texture_format(format
);
540 if (!gpu_supports_texture_format(screen
, fmt
, format
))
543 if (sample_count
< 2 && fmt
!= ETNA_NO_MATCH
)
544 allowed
|= PIPE_BIND_SAMPLER_VIEW
;
547 if (usage
& PIPE_BIND_VERTEX_BUFFER
) {
548 if (gpu_supports_vertex_format(screen
, format
))
549 allowed
|= PIPE_BIND_VERTEX_BUFFER
;
552 if (usage
& PIPE_BIND_INDEX_BUFFER
) {
553 /* must be supported index format */
554 if (format
== PIPE_FORMAT_I8_UINT
|| format
== PIPE_FORMAT_I16_UINT
||
555 (format
== PIPE_FORMAT_I32_UINT
&&
556 VIV_FEATURE(screen
, chipFeatures
, 32_BIT_INDICES
))) {
557 allowed
|= PIPE_BIND_INDEX_BUFFER
;
563 usage
& (PIPE_BIND_DISPLAY_TARGET
| PIPE_BIND_SCANOUT
| PIPE_BIND_SHARED
);
565 if (usage
!= allowed
) {
566 DBG("not supported: format=%s, target=%d, sample_count=%d, "
567 "usage=%x, allowed=%x",
568 util_format_name(format
), target
, sample_count
, usage
, allowed
);
571 return usage
== allowed
;
574 const uint64_t supported_modifiers
[] = {
575 DRM_FORMAT_MOD_LINEAR
,
576 DRM_FORMAT_MOD_VIVANTE_TILED
,
577 DRM_FORMAT_MOD_VIVANTE_SUPER_TILED
,
578 DRM_FORMAT_MOD_VIVANTE_SPLIT_TILED
,
579 DRM_FORMAT_MOD_VIVANTE_SPLIT_SUPER_TILED
,
583 etna_screen_query_dmabuf_modifiers(struct pipe_screen
*pscreen
,
584 enum pipe_format format
, int max
,
586 unsigned int *external_only
, int *count
)
588 struct etna_screen
*screen
= etna_screen(pscreen
);
589 int i
, num_modifiers
= 0;
591 if (max
> ARRAY_SIZE(supported_modifiers
))
592 max
= ARRAY_SIZE(supported_modifiers
);
596 max
= ARRAY_SIZE(supported_modifiers
);
599 for (i
= 0; num_modifiers
< max
; i
++) {
600 /* don't advertise split tiled formats on single pipe/buffer GPUs */
601 if ((screen
->specs
.pixel_pipes
== 1 || screen
->specs
.single_buffer
) &&
606 modifiers
[num_modifiers
] = supported_modifiers
[i
];
608 external_only
[num_modifiers
] = util_format_is_yuv(format
) ? 1 : 0;
612 *count
= num_modifiers
;
616 etna_determine_uniform_limits(struct etna_screen
*screen
)
618 /* values for the non unified case are taken from
619 * gcmCONFIGUREUNIFORMS in the Vivante kernel driver file
620 * drivers/mxc/gpu-viv/hal/kernel/inc/gc_hal_base.h.
622 if (screen
->model
== chipModel_GC2000
&&
623 (screen
->revision
== 0x5118 || screen
->revision
== 0x5140)) {
624 screen
->specs
.max_vs_uniforms
= 256;
625 screen
->specs
.max_ps_uniforms
= 64;
626 } else if (screen
->specs
.num_constants
== 320) {
627 screen
->specs
.max_vs_uniforms
= 256;
628 screen
->specs
.max_ps_uniforms
= 64;
629 } else if (screen
->specs
.num_constants
> 256 &&
630 screen
->model
== chipModel_GC1000
) {
631 /* All GC1000 series chips can only support 64 uniforms for ps on non-unified const mode. */
632 screen
->specs
.max_vs_uniforms
= 256;
633 screen
->specs
.max_ps_uniforms
= 64;
634 } else if (screen
->specs
.num_constants
> 256) {
635 screen
->specs
.max_vs_uniforms
= 256;
636 screen
->specs
.max_ps_uniforms
= 256;
637 } else if (screen
->specs
.num_constants
== 256) {
638 screen
->specs
.max_vs_uniforms
= 256;
639 screen
->specs
.max_ps_uniforms
= 256;
641 screen
->specs
.max_vs_uniforms
= 168;
642 screen
->specs
.max_ps_uniforms
= 64;
647 etna_get_specs(struct etna_screen
*screen
)
650 uint32_t instruction_count
;
652 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_INSTRUCTION_COUNT
, &val
)) {
653 DBG("could not get ETNA_GPU_INSTRUCTION_COUNT");
656 instruction_count
= val
;
658 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_VERTEX_OUTPUT_BUFFER_SIZE
,
660 DBG("could not get ETNA_GPU_VERTEX_OUTPUT_BUFFER_SIZE");
663 screen
->specs
.vertex_output_buffer_size
= val
;
665 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_VERTEX_CACHE_SIZE
, &val
)) {
666 DBG("could not get ETNA_GPU_VERTEX_CACHE_SIZE");
669 screen
->specs
.vertex_cache_size
= val
;
671 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_SHADER_CORE_COUNT
, &val
)) {
672 DBG("could not get ETNA_GPU_SHADER_CORE_COUNT");
675 screen
->specs
.shader_core_count
= val
;
677 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_STREAM_COUNT
, &val
)) {
678 DBG("could not get ETNA_GPU_STREAM_COUNT");
681 screen
->specs
.stream_count
= val
;
683 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_REGISTER_MAX
, &val
)) {
684 DBG("could not get ETNA_GPU_REGISTER_MAX");
687 screen
->specs
.max_registers
= val
;
689 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_PIXEL_PIPES
, &val
)) {
690 DBG("could not get ETNA_GPU_PIXEL_PIPES");
693 screen
->specs
.pixel_pipes
= val
;
695 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_NUM_CONSTANTS
, &val
)) {
696 DBG("could not get %s", "ETNA_GPU_NUM_CONSTANTS");
700 fprintf(stderr
, "Warning: zero num constants (update kernel?)\n");
703 screen
->specs
.num_constants
= val
;
705 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_NUM_VARYINGS
, &val
)) {
706 DBG("could not get ETNA_GPU_NUM_VARYINGS");
709 screen
->specs
.max_varyings
= MAX2(val
, ETNA_NUM_VARYINGS
);
711 /* Figure out gross GPU architecture. See rnndb/common.xml for a specific
712 * description of the differences. */
713 if (VIV_FEATURE(screen
, chipMinorFeatures5
, HALTI5
))
714 screen
->specs
.halti
= 5; /* New GC7000/GC8x00 */
715 else if (VIV_FEATURE(screen
, chipMinorFeatures5
, HALTI4
))
716 screen
->specs
.halti
= 4; /* Old GC7000/GC7400 */
717 else if (VIV_FEATURE(screen
, chipMinorFeatures5
, HALTI3
))
718 screen
->specs
.halti
= 3; /* None? */
719 else if (VIV_FEATURE(screen
, chipMinorFeatures4
, HALTI2
))
720 screen
->specs
.halti
= 2; /* GC2500/GC3000/GC5000/GC6400 */
721 else if (VIV_FEATURE(screen
, chipMinorFeatures2
, HALTI1
))
722 screen
->specs
.halti
= 1; /* GC900/GC4000/GC7000UL */
723 else if (VIV_FEATURE(screen
, chipMinorFeatures1
, HALTI0
))
724 screen
->specs
.halti
= 0; /* GC880/GC2000/GC7000TM */
726 screen
->specs
.halti
= -1; /* GC7000nanolite / pre-GC2000 except GC880 */
727 if (screen
->specs
.halti
>= 0)
728 DBG("etnaviv: GPU arch: HALTI%d", screen
->specs
.halti
);
730 DBG("etnaviv: GPU arch: pre-HALTI");
732 screen
->specs
.can_supertile
=
733 VIV_FEATURE(screen
, chipMinorFeatures0
, SUPER_TILED
);
734 screen
->specs
.bits_per_tile
=
735 VIV_FEATURE(screen
, chipMinorFeatures0
, 2BITPERTILE
) ? 2 : 4;
736 screen
->specs
.ts_clear_value
=
737 VIV_FEATURE(screen
, chipMinorFeatures5
, BLT_ENGINE
) ? 0xffffffff :
738 VIV_FEATURE(screen
, chipMinorFeatures0
, 2BITPERTILE
) ? 0x55555555 :
742 /* vertex and fragment samplers live in one address space */
743 screen
->specs
.vertex_sampler_offset
= 8;
744 screen
->specs
.fragment_sampler_count
= 8;
745 screen
->specs
.vertex_sampler_count
= 4;
747 if (screen
->model
== 0x400)
748 screen
->specs
.vertex_sampler_count
= 0;
750 screen
->specs
.vs_need_z_div
=
751 screen
->model
< 0x1000 && screen
->model
!= 0x880;
752 screen
->specs
.has_sin_cos_sqrt
=
753 VIV_FEATURE(screen
, chipMinorFeatures0
, HAS_SQRT_TRIG
);
754 screen
->specs
.has_sign_floor_ceil
=
755 VIV_FEATURE(screen
, chipMinorFeatures0
, HAS_SIGN_FLOOR_CEIL
);
756 screen
->specs
.has_shader_range_registers
=
757 screen
->model
>= 0x1000 || screen
->model
== 0x880;
758 screen
->specs
.npot_tex_any_wrap
=
759 VIV_FEATURE(screen
, chipMinorFeatures1
, NON_POWER_OF_TWO
);
760 screen
->specs
.has_new_transcendentals
=
761 VIV_FEATURE(screen
, chipMinorFeatures3
, HAS_FAST_TRANSCENDENTALS
);
762 screen
->specs
.has_halti2_instructions
=
763 VIV_FEATURE(screen
, chipMinorFeatures4
, HALTI2
);
764 screen
->specs
.v4_compression
=
765 VIV_FEATURE(screen
, chipMinorFeatures6
, V4_COMPRESSION
);
767 if (screen
->specs
.halti
>= 5) {
768 /* GC7000 - this core must load shaders from memory. */
769 screen
->specs
.vs_offset
= 0;
770 screen
->specs
.ps_offset
= 0;
771 screen
->specs
.max_instructions
= 0; /* Do not program shaders manually */
772 screen
->specs
.has_icache
= true;
773 } else if (VIV_FEATURE(screen
, chipMinorFeatures3
, INSTRUCTION_CACHE
)) {
774 /* GC3000 - this core is capable of loading shaders from
775 * memory. It can also run shaders from registers, as a fallback, but
776 * "max_instructions" does not have the correct value. It has place for
777 * 2*256 instructions just like GC2000, but the offsets are slightly
780 screen
->specs
.vs_offset
= 0xC000;
781 /* State 08000-0C000 mirrors 0C000-0E000, and the Vivante driver uses
782 * this mirror for writing PS instructions, probably safest to do the
785 screen
->specs
.ps_offset
= 0x8000 + 0x1000;
786 screen
->specs
.max_instructions
= 256; /* maximum number instructions for non-icache use */
787 screen
->specs
.has_icache
= true;
789 if (instruction_count
> 256) { /* unified instruction memory? */
790 screen
->specs
.vs_offset
= 0xC000;
791 screen
->specs
.ps_offset
= 0xD000; /* like vivante driver */
792 screen
->specs
.max_instructions
= 256;
794 screen
->specs
.vs_offset
= 0x4000;
795 screen
->specs
.ps_offset
= 0x6000;
796 screen
->specs
.max_instructions
= instruction_count
/ 2;
798 screen
->specs
.has_icache
= false;
801 if (VIV_FEATURE(screen
, chipMinorFeatures1
, HALTI0
)) {
802 screen
->specs
.vertex_max_elements
= 16;
804 /* Etna_viv documentation seems confused over the correct value
805 * here so choose the lower to be safe: HALTI0 says 16 i.s.o.
806 * 10, but VERTEX_ELEMENT_CONFIG register says 16 i.s.o. 12. */
807 screen
->specs
.vertex_max_elements
= 10;
810 etna_determine_uniform_limits(screen
);
812 if (screen
->specs
.halti
>= 5) {
813 screen
->specs
.has_unified_uniforms
= true;
814 screen
->specs
.vs_uniforms_offset
= VIVS_SH_HALTI5_UNIFORMS_MIRROR(0);
815 screen
->specs
.ps_uniforms_offset
= VIVS_SH_HALTI5_UNIFORMS(screen
->specs
.max_vs_uniforms
*4);
816 } else if (screen
->specs
.halti
>= 1) {
817 /* unified uniform memory on GC3000 - HALTI1 feature bit is just a guess
819 screen
->specs
.has_unified_uniforms
= true;
820 screen
->specs
.vs_uniforms_offset
= VIVS_SH_UNIFORMS(0);
821 /* hardcode PS uniforms to start after end of VS uniforms -
822 * for more flexibility this offset could be variable based on the
825 screen
->specs
.ps_uniforms_offset
= VIVS_SH_UNIFORMS(screen
->specs
.max_vs_uniforms
*4);
827 screen
->specs
.has_unified_uniforms
= false;
828 screen
->specs
.vs_uniforms_offset
= VIVS_VS_UNIFORMS(0);
829 screen
->specs
.ps_uniforms_offset
= VIVS_PS_UNIFORMS(0);
832 screen
->specs
.max_texture_size
=
833 VIV_FEATURE(screen
, chipMinorFeatures0
, TEXTURE_8K
) ? 8192 : 2048;
834 screen
->specs
.max_rendertarget_size
=
835 VIV_FEATURE(screen
, chipMinorFeatures0
, RENDERTARGET_8K
) ? 8192 : 2048;
837 screen
->specs
.single_buffer
= VIV_FEATURE(screen
, chipMinorFeatures4
, SINGLE_BUFFER
);
838 if (screen
->specs
.single_buffer
)
839 DBG("etnaviv: Single buffer mode enabled with %d pixel pipes", screen
->specs
.pixel_pipes
);
841 screen
->specs
.tex_astc
= VIV_FEATURE(screen
, chipMinorFeatures4
, TEXTURE_ASTC
) &&
842 !VIV_FEATURE(screen
, chipMinorFeatures6
, NO_ASTC
);
844 screen
->specs
.use_blt
= VIV_FEATURE(screen
, chipMinorFeatures5
, BLT_ENGINE
);
853 etna_screen_bo_from_handle(struct pipe_screen
*pscreen
,
854 struct winsys_handle
*whandle
, unsigned *out_stride
)
856 struct etna_screen
*screen
= etna_screen(pscreen
);
859 if (whandle
->type
== WINSYS_HANDLE_TYPE_SHARED
) {
860 bo
= etna_bo_from_name(screen
->dev
, whandle
->handle
);
861 } else if (whandle
->type
== WINSYS_HANDLE_TYPE_FD
) {
862 bo
= etna_bo_from_dmabuf(screen
->dev
, whandle
->handle
);
864 DBG("Attempt to import unsupported handle type %d", whandle
->type
);
869 DBG("ref name 0x%08x failed", whandle
->handle
);
873 *out_stride
= whandle
->stride
;
879 etna_get_compiler_options(struct pipe_screen
*pscreen
,
880 enum pipe_shader_ir ir
, unsigned shader
)
882 return &etna_screen(pscreen
)->options
;
886 etna_screen_create(struct etna_device
*dev
, struct etna_gpu
*gpu
,
887 struct renderonly
*ro
)
889 struct etna_screen
*screen
= CALLOC_STRUCT(etna_screen
);
890 struct pipe_screen
*pscreen
;
891 drmVersionPtr version
;
897 pscreen
= &screen
->base
;
900 screen
->ro
= renderonly_dup(ro
);
904 DBG("could not create renderonly object");
908 version
= drmGetVersion(screen
->ro
->gpu_fd
);
909 screen
->drm_version
= ETNA_DRM_VERSION(version
->version_major
,
910 version
->version_minor
);
911 drmFreeVersion(version
);
913 etna_mesa_debug
= debug_get_option_etna_mesa_debug();
915 /* Disable autodisable for correct rendering with TS */
916 etna_mesa_debug
|= ETNA_DBG_NO_AUTODISABLE
;
918 screen
->pipe
= etna_pipe_new(gpu
, ETNA_PIPE_3D
);
920 DBG("could not create 3d pipe");
924 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_MODEL
, &val
)) {
925 DBG("could not get ETNA_GPU_MODEL");
930 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_REVISION
, &val
)) {
931 DBG("could not get ETNA_GPU_REVISION");
934 screen
->revision
= val
;
936 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_FEATURES_0
, &val
)) {
937 DBG("could not get ETNA_GPU_FEATURES_0");
940 screen
->features
[0] = val
;
942 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_FEATURES_1
, &val
)) {
943 DBG("could not get ETNA_GPU_FEATURES_1");
946 screen
->features
[1] = val
;
948 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_FEATURES_2
, &val
)) {
949 DBG("could not get ETNA_GPU_FEATURES_2");
952 screen
->features
[2] = val
;
954 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_FEATURES_3
, &val
)) {
955 DBG("could not get ETNA_GPU_FEATURES_3");
958 screen
->features
[3] = val
;
960 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_FEATURES_4
, &val
)) {
961 DBG("could not get ETNA_GPU_FEATURES_4");
964 screen
->features
[4] = val
;
966 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_FEATURES_5
, &val
)) {
967 DBG("could not get ETNA_GPU_FEATURES_5");
970 screen
->features
[5] = val
;
972 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_FEATURES_6
, &val
)) {
973 DBG("could not get ETNA_GPU_FEATURES_6");
976 screen
->features
[6] = val
;
978 if (etna_gpu_get_param(screen
->gpu
, ETNA_GPU_FEATURES_7
, &val
)) {
979 DBG("could not get ETNA_GPU_FEATURES_7");
982 screen
->features
[7] = val
;
984 if (!etna_get_specs(screen
))
987 if (screen
->specs
.halti
>= 5 && !etnaviv_device_softpin_capable(dev
)) {
988 DBG("halti5 requires softpin");
992 screen
->options
= (nir_shader_compiler_options
) {
995 .lower_ftrunc
= true,
997 .lower_bitops
= true,
998 .lower_all_io_to_temps
= true,
999 .vertex_id_zero_based
= true,
1000 .lower_flrp32
= true,
1002 .lower_vector_cmp
= true,
1004 .lower_fdiv
= true, /* !screen->specs.has_new_transcendentals */
1005 .lower_fsign
= !screen
->specs
.has_sign_floor_ceil
,
1006 .lower_ffloor
= !screen
->specs
.has_sign_floor_ceil
,
1007 .lower_fceil
= !screen
->specs
.has_sign_floor_ceil
,
1008 .lower_fsqrt
= !screen
->specs
.has_sin_cos_sqrt
,
1009 .lower_sincos
= !screen
->specs
.has_sin_cos_sqrt
,
1012 /* apply debug options that disable individual features */
1013 if (DBG_ENABLED(ETNA_DBG_NO_EARLY_Z
))
1014 screen
->features
[viv_chipFeatures
] |= chipFeatures_NO_EARLY_Z
;
1015 if (DBG_ENABLED(ETNA_DBG_NO_TS
))
1016 screen
->features
[viv_chipFeatures
] &= ~chipFeatures_FAST_CLEAR
;
1017 if (DBG_ENABLED(ETNA_DBG_NO_AUTODISABLE
))
1018 screen
->features
[viv_chipMinorFeatures1
] &= ~chipMinorFeatures1_AUTO_DISABLE
;
1019 if (DBG_ENABLED(ETNA_DBG_NO_SUPERTILE
))
1020 screen
->specs
.can_supertile
= 0;
1021 if (DBG_ENABLED(ETNA_DBG_NO_SINGLEBUF
))
1022 screen
->specs
.single_buffer
= 0;
1024 pscreen
->destroy
= etna_screen_destroy
;
1025 pscreen
->get_param
= etna_screen_get_param
;
1026 pscreen
->get_paramf
= etna_screen_get_paramf
;
1027 pscreen
->get_shader_param
= etna_screen_get_shader_param
;
1028 pscreen
->get_compiler_options
= etna_get_compiler_options
;
1030 pscreen
->get_name
= etna_screen_get_name
;
1031 pscreen
->get_vendor
= etna_screen_get_vendor
;
1032 pscreen
->get_device_vendor
= etna_screen_get_device_vendor
;
1034 pscreen
->get_timestamp
= etna_screen_get_timestamp
;
1035 pscreen
->context_create
= etna_context_create
;
1036 pscreen
->is_format_supported
= etna_screen_is_format_supported
;
1037 pscreen
->query_dmabuf_modifiers
= etna_screen_query_dmabuf_modifiers
;
1039 etna_fence_screen_init(pscreen
);
1040 etna_query_screen_init(pscreen
);
1041 etna_resource_screen_init(pscreen
);
1043 util_dynarray_init(&screen
->supported_pm_queries
, NULL
);
1044 slab_create_parent(&screen
->transfer_pool
, sizeof(struct etna_transfer
), 16);
1046 if (screen
->drm_version
>= ETNA_DRM_VERSION_PERFMON
)
1047 etna_pm_query_setup(screen
);
1052 etna_screen_destroy(pscreen
);