src/panfrost/lib/pan_afbc.c

   1 /*
   2  * Copyright (C) 2019 Collabora, Ltd.
   3  *
   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:
  10  *
  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
  13  * Software.
  14  *
  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 FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  *
  23  * Authors:
  24  *   Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
  25  */
  26
  27 #include "pan_texture.h"
  28
  29 /* Arm FrameBuffer Compression (AFBC) is a lossless compression scheme natively
  30  * implemented in Mali GPUs (as well as many display controllers paired with
  31  * Mali GPUs, etc). Where possible, Panfrost prefers to use AFBC for both
  32  * rendering and texturing. In most cases, this is a performance-win due to a
  33  * dramatic reduction in memory bandwidth and cache locality compared to a
  34  * linear resources.
  35  *
  36  * AFBC divides the framebuffer into 16x16 tiles (other sizes possible, TODO:
  37  * do we need to support this?). So, the width and height each must be aligned
  38  * up to 16 pixels. This is inherently good for performance; note that for a 4
  39  * byte-per-pixel format like RGBA8888, that means that rows are 16*4=64 byte
  40  * aligned, which is the cache-line size.
  41  *
  42  * For each AFBC-compressed resource, there is a single contiguous
  43  * (CPU/GPU-shared) buffer. This buffer itself is divided into two parts:
  44  * header and body, placed immediately after each other.
  45  *
  46  * The AFBC header contains 16 bytes of metadata per tile.
  47  *
  48  * The AFBC body is the same size as the original linear resource (padded to
  49  * the nearest tile). Although the body comes immediately after the header, it
  50  * must also be cache-line aligned, so there can sometimes be a bit of padding
  51  * between the header and body.
  52  *
  53  * As an example, a 64x64 RGBA framebuffer contains 64/16 = 4 tiles horizontally and
  54  * 4 tiles vertically. There are 4*4=16 tiles in total, each containing 16
  55  * bytes of metadata, so there is a 16*16=256 byte header. 64x64 is already
  56  * tile aligned, so the body is 64*64 * 4 bytes per pixel = 16384 bytes of
  57  * body.
  58  *
  59  * From userspace, Panfrost needs to be able to calculate these sizes. It
  60  * explicitly does not and can not know the format of the data contained within
  61  * this header and body. The GPU has native support for AFBC encode/decode. For
  62  * an internal FBO or a framebuffer used for scanout with an AFBC-compatible
  63  * winsys/display-controller, the buffer is maintained AFBC throughout flight,
  64  * and the driver never needs to know the internal data. For edge cases where
  65  * the driver really does need to read/write from the AFBC resource, we
  66  * generate a linear staging buffer and use the GPU to blit AFBC<--->linear.
  67  * TODO: Implement me. */
  68
  69 #define AFBC_TILE_WIDTH 16
  70 #define AFBC_TILE_HEIGHT 16
  71 #define AFBC_HEADER_BYTES_PER_TILE 16
  72 #define AFBC_CACHE_ALIGN 64
  73
  74 /* Is it possible to AFBC compress a particular format? Common formats (and
  75  * YUV) are compressible. Some obscure formats are not and fallback on linear,
  76  * at a performance hit. Also, if you need to disable AFBC entirely in the
  77  * driver for debug/profiling, just always return false here. */
  78
  79 bool
  80 panfrost_format_supports_afbc(enum pipe_format format)
  81 {
  82         const struct util_format_description *desc =
  83                 util_format_description(format);
  84
  85         /* sRGB cannot be AFBC, but it can be tiled. TODO: Verify. The blob
  86          * does not do AFBC for SRGB8_ALPHA8, but it's not clear why it
  87          * shouldn't be able to. */
  88
  89         if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
  90                 return false;
  91
  92         if (util_format_is_rgba8_variant(desc))
  93                 return true;
  94
  95         /* Only Z24S8 variants are compressible as Z/S */
  96
  97         if (panfrost_is_z24s8_variant(format))
  98                 return true;
  99
 100         /* Lookup special formats */
 101         switch (format) {
 102         case PIPE_FORMAT_R8G8B8_UNORM:
 103         case PIPE_FORMAT_B8G8R8_UNORM:
 104         case PIPE_FORMAT_R5G6B5_UNORM:
 105         case PIPE_FORMAT_B5G6R5_UNORM:
 106                 return true;
 107         default:
 108                 return false;
 109         }
 110 }
 111
 112 unsigned
 113 panfrost_afbc_header_size(unsigned width, unsigned height)
 114 {
 115         /* Align to tile */
 116         unsigned aligned_width  = ALIGN_POT(width,  AFBC_TILE_WIDTH);
 117         unsigned aligned_height = ALIGN_POT(height, AFBC_TILE_HEIGHT);
 118
 119         /* Compute size in tiles, rather than pixels */
 120         unsigned tile_count_x = aligned_width  / AFBC_TILE_WIDTH;
 121         unsigned tile_count_y = aligned_height / AFBC_TILE_HEIGHT;
 122         unsigned tile_count = tile_count_x * tile_count_y;
 123
 124         /* Multiply to find the header size */
 125         unsigned header_bytes = tile_count * AFBC_HEADER_BYTES_PER_TILE;
 126
 127         /* Align and go */
 128         return ALIGN_POT(header_bytes, AFBC_CACHE_ALIGN);
 129
 130 }