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 static void
 170 panfrost_emit_texture_payload(
 171         mali_ptr *payload,
 172         const struct util_format_description *desc,
 173         enum mali_format mali_format,
 174         enum mali_texture_type type,
 175         enum mali_texture_layout layout,
 176         unsigned first_level, unsigned last_level,
 177         unsigned first_layer, unsigned last_layer,
 178         unsigned cube_stride,
 179         bool manual_stride,
 180         mali_ptr base,
 181         struct panfrost_slice *slices)
 182 {
 183         base |= panfrost_compression_tag(desc, mali_format, layout);
 184
 185         /* Inject the addresses in, interleaving array indices, mip levels,
 186          * cube faces, and strides in that order */
 187
 188         unsigned first_face  = 0, last_face = 0, face_mult = 1;
 189
 190         if (type == MALI_TEX_CUBE) {
 191                 face_mult = 6;
 192                 panfrost_adjust_cube_dimensions(&first_face, &last_face, &first_layer, &last_layer);
 193         }
 194
 195         unsigned idx = 0;
 196
 197         for (unsigned w = first_layer; w <= last_layer; ++w) {
 198                 for (unsigned l = first_level; l <= last_level; ++l) {
 199                         for (unsigned f = first_face; f <= last_face; ++f) {
 200                                 payload[idx++] = base + panfrost_texture_offset(
 201                                                 slices, type == MALI_TEX_3D,
 202                                                 cube_stride, l, w * face_mult + f);
 203
 204                                 if (manual_stride)
 205                                         payload[idx++] = slices[l].stride;
 206                         }
 207                 }
 208         }
 209 }
 210
 211 void
 212 panfrost_new_texture(
 213         void *out,
 214         uint16_t width, uint16_t height,
 215         uint16_t depth, uint16_t array_size,
 216         enum pipe_format format,
 217         enum mali_texture_type type,
 218         enum mali_texture_layout layout,
 219         unsigned first_level, unsigned last_level,
 220         unsigned first_layer, unsigned last_layer,
 221         unsigned cube_stride,
 222         unsigned swizzle,
 223         mali_ptr base,
 224         struct panfrost_slice *slices)
 225 {
 226         const struct util_format_description *desc =
 227                 util_format_description(format);
 228
 229         unsigned bytes_per_pixel = util_format_get_blocksize(format);
 230
 231         enum mali_format mali_format = panfrost_find_format(desc);
 232
 233         bool manual_stride = (layout == MALI_TEXTURE_LINEAR)
 234                 && panfrost_needs_explicit_stride(slices, width,
 235                                 first_level, last_level, bytes_per_pixel);
 236
 237         struct mali_texture_descriptor descriptor = {
 238                 .width = MALI_POSITIVE(u_minify(width, first_level)),
 239                 .height = MALI_POSITIVE(u_minify(height, first_level)),
 240                 .depth = MALI_POSITIVE(u_minify(depth, first_level)),
 241                 .array_size = MALI_POSITIVE(array_size),
 242                 .format = {
 243                         .swizzle = panfrost_translate_swizzle_4(desc->swizzle),
 244                         .format = mali_format,
 245                         .srgb = (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB),
 246                         .type = type,
 247                         .layout = layout,
 248                         .manual_stride = manual_stride,
 249                         .unknown2 = 1,
 250                 },
 251                 .levels = last_level - first_level,
 252                 .swizzle = swizzle
 253         };
 254
 255         memcpy(out, &descriptor, sizeof(descriptor));
 256
 257         mali_ptr *payload = (mali_ptr *) (out + sizeof(struct mali_texture_descriptor));
 258         panfrost_emit_texture_payload(
 259                 payload,
 260                 desc,
 261                 mali_format,
 262                 type,
 263                 layout,
 264                 first_level, last_level,
 265                 first_layer, last_layer,
 266                 cube_stride,
 267                 manual_stride,
 268                 base,
 269                 slices);
 270 }
 271
 272 void
 273 panfrost_new_texture_bifrost(
 274         struct bifrost_texture_descriptor *descriptor,
 275         uint16_t width, uint16_t height,
 276         uint16_t depth, uint16_t array_size,
 277         enum pipe_format format,
 278         enum mali_texture_type type,
 279         enum mali_texture_layout layout,
 280         unsigned first_level, unsigned last_level,
 281         unsigned first_layer, unsigned last_layer,
 282         unsigned cube_stride,
 283         unsigned swizzle,
 284         mali_ptr base,
 285         struct panfrost_slice *slices,
 286         struct panfrost_bo *payload)
 287 {
 288         const struct util_format_description *desc =
 289                 util_format_description(format);
 290
 291         enum mali_format mali_format = panfrost_find_format(desc);
 292
 293         /* Apparently it's always needed in Bifrost? */
 294         bool manual_stride = true;
 295
 296         panfrost_emit_texture_payload(
 297                 (mali_ptr *) payload->cpu,
 298                 desc,
 299                 mali_format,
 300                 type,
 301                 layout,
 302                 first_level, last_level,
 303                 first_layer, last_layer,
 304                 cube_stride,
 305                 manual_stride,
 306                 base,
 307                 slices);
 308
 309         descriptor->format_unk = 0x2;
 310         descriptor->type = type;
 311         descriptor->format_unk2 = 0x100;
 312         descriptor->format = mali_format;
 313         descriptor->srgb = (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB);
 314         descriptor->format_unk3 = 0x0;
 315         descriptor->width = MALI_POSITIVE(u_minify(width, first_level));
 316         descriptor->height = MALI_POSITIVE(u_minify(height, first_level));
 317         descriptor->swizzle = swizzle;
 318         descriptor->layout = layout;
 319         descriptor->levels = last_level - first_level;
 320         descriptor->unk1 = 0x0;
 321         descriptor->levels_unk = 0;
 322         descriptor->level_2 = 0;
 323         descriptor->payload = payload->gpu;
 324         descriptor->array_size = MALI_POSITIVE(array_size);
 325         descriptor->unk4 = 0x0;
 326         descriptor->depth = MALI_POSITIVE(u_minify(depth, first_level));
 327         descriptor->unk5 = 0x0;
 328 }
 329
 330 /* Computes sizes for checksumming, which is 8 bytes per 16x16 tile.
 331  * Checksumming is believed to be a CRC variant (CRC64 based on the size?).
 332  * This feature is also known as "transaction elimination". */
 333
 334 #define CHECKSUM_TILE_WIDTH 16
 335 #define CHECKSUM_TILE_HEIGHT 16
 336 #define CHECKSUM_BYTES_PER_TILE 8
 337
 338 unsigned
 339 panfrost_compute_checksum_size(
 340         struct panfrost_slice *slice,
 341         unsigned width,
 342         unsigned height)
 343 {
 344         unsigned aligned_width = ALIGN_POT(width, CHECKSUM_TILE_WIDTH);
 345         unsigned aligned_height = ALIGN_POT(height, CHECKSUM_TILE_HEIGHT);
 346
 347         unsigned tile_count_x = aligned_width / CHECKSUM_TILE_WIDTH;
 348         unsigned tile_count_y = aligned_height / CHECKSUM_TILE_HEIGHT;
 349
 350         slice->checksum_stride = tile_count_x * CHECKSUM_BYTES_PER_TILE;
 351
 352         return slice->checksum_stride * tile_count_y;
 353 }
 354
 355 unsigned
 356 panfrost_get_layer_stride(struct panfrost_slice *slices, bool is_3d, unsigned cube_stride, unsigned level)
 357 {
 358         return is_3d ? slices[level].size0 : cube_stride;
 359 }
 360
 361 /* Computes the offset into a texture at a particular level/face. Add to
 362  * the base address of a texture to get the address to that level/face */
 363
 364 unsigned
 365 panfrost_texture_offset(struct panfrost_slice *slices, bool is_3d, unsigned cube_stride, unsigned level, unsigned face)
 366 {
 367         unsigned layer_stride = panfrost_get_layer_stride(slices, is_3d, cube_stride, level);
 368         return slices[level].offset + (face * layer_stride);
 369 }