src/panfrost/encoder/pan_texture.c

   1 /*
   2  * Copyright (C) 2008 VMware, Inc.
   3  * Copyright (C) 2014 Broadcom
   4  * Copyright (C) 2018-2019 Alyssa Rosenzweig
   5  * Copyright (C) 2019-2020 Collabora, Ltd.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a
   8  * copy of this software and associated documentation files (the "Software"),
   9  * to deal in the Software without restriction, including without limitation
  10  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  11  * and/or sell copies of the Software, and to permit persons to whom the
  12  * Software is furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the next
  15  * paragraph) shall be included in all copies or substantial portions of the
  16  * Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  24  * SOFTWARE.
  25  *
  26  */
  27
  28 #include "util/macros.h"
  29 #include "util/u_math.h"
  30 #include "pan_texture.h"
  31
  32 /* Generates a texture descriptor. Ideally, descriptors are immutable after the
  33  * texture is created, so we can keep these hanging around in GPU memory in a
  34  * dedicated BO and not have to worry. In practice there are some minor gotchas
  35  * with this (the driver sometimes will change the format of a texture on the
  36  * fly for compression) but it's fast enough to just regenerate the descriptor
  37  * in those cases, rather than monkeypatching at drawtime.
  38  *
  39  * A texture descriptor consists of a 32-byte mali_texture_descriptor structure
  40  * followed by a variable number of pointers. Due to this variance and
  41  * potentially large size, we actually upload directly rather than returning
  42  * the descriptor. Whether the user does a copy themselves or not is irrelevant
  43  * to us here.
  44  */
  45
  46 /* Check if we need to set a custom stride by computing the "expected"
  47  * stride and comparing it to what the user actually wants. Only applies
  48  * to linear textures, since tiled/compressed textures have strict
  49  * alignment requirements for their strides as it is */
  50
  51 static bool
  52 panfrost_needs_explicit_stride(
  53                 struct panfrost_slice *slices,
  54                 uint16_t width,
  55                 unsigned first_level, unsigned last_level,
  56                 unsigned bytes_per_pixel)
  57 {
  58         for (unsigned l = first_level; l <= last_level; ++l) {
  59                 unsigned actual = slices[l].stride;
  60                 unsigned expected = u_minify(width, l) * bytes_per_pixel;
  61
  62                 if (actual != expected)
  63                         return true;
  64         }
  65
  66         return false;
  67 }
  68
  69 /* A Scalable Texture Compression (ASTC) corresponds to just a few texture type
  70  * in the hardware, but in fact can be parametrized to have various widths and
  71  * heights for the so-called "stretch factor". It turns out these parameters
  72  * are stuffed in the bottom bits of the payload pointers. This functions
  73  * computes these magic stuffing constants based on the ASTC format in use. The
  74  * constant in a given dimension is 3-bits, and two are stored side-by-side for
  75  * each active dimension.
  76  */
  77
  78 static unsigned
  79 panfrost_astc_stretch(unsigned dim)
  80 {
  81         assert(dim >= 4 && dim <= 12);
  82         return MIN2(dim, 11) - 4;
  83 }
  84
  85 /* Texture addresses are tagged with information about AFBC (colour AFBC?) xor
  86  * ASTC (stretch factor) if in use. */
  87
  88 static unsigned
  89 panfrost_compression_tag(
  90                 const struct util_format_description *desc,
  91                 enum mali_format format, enum mali_texture_layout layout)
  92 {
  93         if (layout == MALI_TEXTURE_AFBC)
  94                 return util_format_has_depth(desc) ? 0x0 : 0x1;
  95         else if (format == MALI_ASTC_HDR_SUPP || format == MALI_ASTC_SRGB_SUPP)
  96                 return (panfrost_astc_stretch(desc->block.height) << 3) |
  97                         panfrost_astc_stretch(desc->block.width);
  98         else
  99                 return 0;
 100 }
 101
 102
 103 /* Cubemaps have 6 faces as "layers" in between each actual layer. We
 104  * need to fix this up. TODO: logic wrong in the asserted out cases ...
 105  * can they happen, perhaps from cubemap arrays? */
 106
 107 static void
 108 panfrost_adjust_cube_dimensions(
 109                 unsigned *first_face, unsigned *last_face,
 110                 unsigned *first_layer, unsigned *last_layer)
 111 {
 112         *first_face = *first_layer % 6;
 113         *last_face = *last_layer % 6;
 114         *first_layer /= 6;
 115         *last_layer /= 6;
 116
 117         assert((*first_layer == *last_layer) || (*first_face == 0 && *last_face == 5));
 118 }
 119
 120 /* Following the texture descriptor is a number of pointers. How many? */
 121
 122 static unsigned
 123 panfrost_texture_num_elements(
 124                 unsigned first_level, unsigned last_level,
 125                 unsigned first_layer, unsigned last_layer,
 126                 bool is_cube, bool manual_stride)
 127 {
 128         unsigned first_face  = 0, last_face = 0;
 129
 130         if (is_cube) {
 131                 panfrost_adjust_cube_dimensions(&first_face, &last_face,
 132                                 &first_layer, &last_layer);
 133         }
 134
 135         unsigned levels = 1 + last_level - first_level;
 136         unsigned layers = 1 + last_layer - first_layer;
 137         unsigned faces  = 1 + last_face  - first_face;
 138         unsigned num_elements = levels * layers * faces;
 139
 140         if (manual_stride)
 141                 num_elements *= 2;
 142
 143         return num_elements;
 144 }
 145
 146 /* Conservative estimate of the size of the texture payload a priori.
 147  * Average case, size equal to the actual size. Worst case, off by 2x (if
 148  * a manual stride is not needed on a linear texture). Returned value
 149  * must be greater than or equal to the actual size, so it's safe to use
 150  * as an allocation amount */
 151
 152 unsigned
 153 panfrost_estimate_texture_payload_size(
 154                 unsigned first_level, unsigned last_level,
 155                 unsigned first_layer, unsigned last_layer,
 156                 enum mali_texture_type type, enum mali_texture_layout layout)
 157 {
 158         /* Assume worst case */
 159         unsigned manual_stride = (layout == MALI_TEXTURE_LINEAR);
 160
 161         unsigned elements = panfrost_texture_num_elements(
 162                         first_level, last_level,
 163                         first_layer, last_layer,
 164                         type == MALI_TEX_CUBE, manual_stride);
 165
 166         return sizeof(mali_ptr) * elements;
 167 }
 168
 169 /* Bifrost requires a tile stride for tiled textures. This stride is computed
 170  * as (16 * bpp * width) assuming there is at least one tile (width >= 16).
 171  * Otherwise if width < 16, the blob puts zero. Interactions with AFBC are
 172  * currently unknown.
 173  */
 174
 175 static unsigned
 176 panfrost_nonlinear_stride(enum mali_texture_layout layout,
 177                 unsigned bytes_per_pixel,
 178                 unsigned width)
 179 {
 180         if (layout == MALI_TEXTURE_TILED) {
 181                 return (width < 16) ? 0 : (16 * bytes_per_pixel * width);
 182         } else {
 183                 unreachable("TODO: AFBC on Bifrost");
 184         }
 185 }
 186
 187 static void
 188 panfrost_emit_texture_payload(
 189         mali_ptr *payload,
 190         const struct util_format_description *desc,
 191         enum mali_format mali_format,
 192         enum mali_texture_type type,
 193         enum mali_texture_layout layout,
 194         unsigned width,
 195         unsigned first_level, unsigned last_level,
 196         unsigned first_layer, unsigned last_layer,
 197         unsigned cube_stride,
 198         bool manual_stride,
 199         mali_ptr base,
 200         struct panfrost_slice *slices)
 201 {
 202         base |= panfrost_compression_tag(desc, mali_format, layout);
 203
 204         /* Inject the addresses in, interleaving array indices, mip levels,
 205          * cube faces, and strides in that order */
 206
 207         unsigned first_face  = 0, last_face = 0, face_mult = 1;
 208
 209         if (type == MALI_TEX_CUBE) {
 210                 face_mult = 6;
 211                 panfrost_adjust_cube_dimensions(&first_face, &last_face, &first_layer, &last_layer);
 212         }
 213
 214         unsigned idx = 0;
 215
 216         for (unsigned w = first_layer; w <= last_layer; ++w) {
 217                 for (unsigned l = first_level; l <= last_level; ++l) {
 218                         for (unsigned f = first_face; f <= last_face; ++f) {
 219                                 payload[idx++] = base + panfrost_texture_offset(
 220                                                 slices, type == MALI_TEX_3D,
 221                                                 cube_stride, l, w * face_mult + f);
 222
 223                                 if (manual_stride) {
 224                                         payload[idx++] = (layout == MALI_TEXTURE_LINEAR) ?
 225                                                 slices[l].stride :
 226                                                 panfrost_nonlinear_stride(layout,
 227                                                                 MAX2(desc->block.bits / 8, 1),
 228                                                                 u_minify(width, l));
 229                                 }
 230                         }
 231                 }
 232         }
 233 }
 234
 235 void
 236 panfrost_new_texture(
 237         void *out,
 238         uint16_t width, uint16_t height,
 239         uint16_t depth, uint16_t array_size,
 240         enum pipe_format format,
 241         enum mali_texture_type type,
 242         enum mali_texture_layout layout,
 243         unsigned first_level, unsigned last_level,
 244         unsigned first_layer, unsigned last_layer,
 245         unsigned cube_stride,
 246         unsigned swizzle,
 247         mali_ptr base,
 248         struct panfrost_slice *slices)
 249 {
 250         const struct util_format_description *desc =
 251                 util_format_description(format);
 252
 253         unsigned bytes_per_pixel = util_format_get_blocksize(format);
 254
 255         enum mali_format mali_format = panfrost_pipe_format_table[desc->format].hw;
 256         assert(mali_format);
 257
 258         bool manual_stride = (layout == MALI_TEXTURE_LINEAR)
 259                 && panfrost_needs_explicit_stride(slices, width,
 260                                 first_level, last_level, bytes_per_pixel);
 261
 262         struct mali_texture_descriptor descriptor = {
 263                 .width = MALI_POSITIVE(u_minify(width, first_level)),
 264                 .height = MALI_POSITIVE(u_minify(height, first_level)),
 265                 .depth = MALI_POSITIVE(u_minify(depth, first_level)),
 266                 .array_size = MALI_POSITIVE(array_size),
 267                 .format = {
 268                         .swizzle = panfrost_translate_swizzle_4(desc->swizzle),
 269                         .format = mali_format,
 270                         .srgb = (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB),
 271                         .type = type,
 272                         .layout = layout,
 273                         .manual_stride = manual_stride,
 274                         .unknown2 = 1,
 275                 },
 276                 .levels = last_level - first_level,
 277                 .swizzle = swizzle
 278         };
 279
 280         memcpy(out, &descriptor, sizeof(descriptor));
 281
 282         mali_ptr *payload = (mali_ptr *) (out + sizeof(struct mali_texture_descriptor));
 283         panfrost_emit_texture_payload(
 284                 payload,
 285                 desc,
 286                 mali_format,
 287                 type,
 288                 layout,
 289                 width,
 290                 first_level, last_level,
 291                 first_layer, last_layer,
 292                 cube_stride,
 293                 manual_stride,
 294                 base,
 295                 slices);
 296 }
 297
 298 void
 299 panfrost_new_texture_bifrost(
 300         struct bifrost_texture_descriptor *descriptor,
 301         uint16_t width, uint16_t height,
 302         uint16_t depth, uint16_t array_size,
 303         enum pipe_format format,
 304         enum mali_texture_type type,
 305         enum mali_texture_layout layout,
 306         unsigned first_level, unsigned last_level,
 307         unsigned first_layer, unsigned last_layer,
 308         unsigned cube_stride,
 309         unsigned swizzle,
 310         mali_ptr base,
 311         struct panfrost_slice *slices,
 312         struct panfrost_bo *payload)
 313 {
 314         const struct util_format_description *desc =
 315                 util_format_description(format);
 316
 317         enum mali_format mali_format = panfrost_pipe_format_table[desc->format].hw;
 318         assert(mali_format);
 319
 320         panfrost_emit_texture_payload(
 321                 (mali_ptr *) payload->cpu,
 322                 desc,
 323                 mali_format,
 324                 type,
 325                 layout,
 326                 width,
 327                 first_level, last_level,
 328                 first_layer, last_layer,
 329                 cube_stride,
 330                 true, /* Stride explicit on Bifrost */
 331                 base,
 332                 slices);
 333
 334         descriptor->format_unk = 0x2;
 335         descriptor->type = type;
 336         descriptor->format = mali_format;
 337         descriptor->srgb = (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB);
 338         descriptor->format_unk3 = 0x0;
 339         descriptor->width = MALI_POSITIVE(u_minify(width, first_level));
 340         descriptor->height = MALI_POSITIVE(u_minify(height, first_level));
 341         descriptor->swizzle = swizzle;
 342         descriptor->layout = layout;
 343         descriptor->levels = last_level - first_level;
 344         descriptor->unk1 = 0x0;
 345         descriptor->levels_unk = 0;
 346         descriptor->level_2 = 0;
 347         descriptor->payload = payload->gpu;
 348         descriptor->array_size = MALI_POSITIVE(array_size);
 349         descriptor->unk4 = 0x0;
 350         descriptor->depth = MALI_POSITIVE(u_minify(depth, first_level));
 351         descriptor->unk5 = 0x0;
 352 }
 353
 354 /* Computes sizes for checksumming, which is 8 bytes per 16x16 tile.
 355  * Checksumming is believed to be a CRC variant (CRC64 based on the size?).
 356  * This feature is also known as "transaction elimination". */
 357
 358 #define CHECKSUM_TILE_WIDTH 16
 359 #define CHECKSUM_TILE_HEIGHT 16
 360 #define CHECKSUM_BYTES_PER_TILE 8
 361
 362 unsigned
 363 panfrost_compute_checksum_size(
 364         struct panfrost_slice *slice,
 365         unsigned width,
 366         unsigned height)
 367 {
 368         unsigned aligned_width = ALIGN_POT(width, CHECKSUM_TILE_WIDTH);
 369         unsigned aligned_height = ALIGN_POT(height, CHECKSUM_TILE_HEIGHT);
 370
 371         unsigned tile_count_x = aligned_width / CHECKSUM_TILE_WIDTH;
 372         unsigned tile_count_y = aligned_height / CHECKSUM_TILE_HEIGHT;
 373
 374         slice->checksum_stride = tile_count_x * CHECKSUM_BYTES_PER_TILE;
 375
 376         return slice->checksum_stride * tile_count_y;
 377 }
 378
 379 unsigned
 380 panfrost_get_layer_stride(struct panfrost_slice *slices, bool is_3d, unsigned cube_stride, unsigned level)
 381 {
 382         return is_3d ? slices[level].size0 : cube_stride;
 383 }
 384
 385 /* Computes the offset into a texture at a particular level/face. Add to
 386  * the base address of a texture to get the address to that level/face */
 387
 388 unsigned
 389 panfrost_texture_offset(struct panfrost_slice *slices, bool is_3d, unsigned cube_stride, unsigned level, unsigned face)
 390 {
 391         unsigned layer_stride = panfrost_get_layer_stride(slices, is_3d, cube_stride, level);
 392         return slices[level].offset + (face * layer_stride);
 393 }