src/gallium/drivers/freedreno/a6xx/fd6_resource.c

   1 /*
   2  * Copyright (C) 2018 Rob Clark <robclark@freedesktop.org>
   3  * Copyright © 2018 Google, Inc.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   9  * and/or sell copies of the Software, and to permit persons to whom the
  10  * Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  22  * SOFTWARE.
  23  *
  24  * Authors:
  25  *    Rob Clark <robclark@freedesktop.org>
  26  */
  27
  28 #include "fd6_resource.h"
  29 #include "fd6_format.h"
  30
  31 #include "a6xx.xml.h"
  32
  33 /* A subset of the valid tiled formats can be compressed.  We do
  34  * already require tiled in order to be compressed, but just because
  35  * it can be tiled doesn't mean it can be compressed.
  36  */
  37 static bool
  38 ok_ubwc_format(enum pipe_format pfmt)
  39 {
  40         /* NOTE: both x24s8 and z24s8 map to RB6_X8Z24_UNORM, but UBWC
  41          * does not seem to work properly when sampling x24s8.. possibly
  42          * because we sample it as TFMT6_8_8_8_8_UINT.
  43          *
  44          * This could possibly be a hw limitation, or maybe something
  45          * else wrong somewhere (although z24s8 blits and sampling with
  46          * UBWC seem fine).  Recheck on a later revision of a6xx
  47          */
  48         if (pfmt == PIPE_FORMAT_X24S8_UINT)
  49                 return false;
  50
  51         switch (fd6_pipe2color(pfmt)) {
  52         case RB6_R10G10B10A2_UINT:
  53         case RB6_R10G10B10A2_UNORM:
  54         case RB6_R11G11B10_FLOAT:
  55         case RB6_R16_FLOAT:
  56         case RB6_R16G16B16A16_FLOAT:
  57         case RB6_R16G16B16A16_SINT:
  58         case RB6_R16G16B16A16_UINT:
  59         case RB6_R16G16_FLOAT:
  60         case RB6_R16G16_SINT:
  61         case RB6_R16G16_UINT:
  62         case RB6_R16_SINT:
  63         case RB6_R16_UINT:
  64         case RB6_R32G32B32A32_SINT:
  65         case RB6_R32G32B32A32_UINT:
  66         case RB6_R32G32_SINT:
  67         case RB6_R32G32_UINT:
  68         case RB6_R5G6B5_UNORM:
  69         case RB6_R8G8B8A8_SINT:
  70         case RB6_R8G8B8A8_UINT:
  71         case RB6_R8G8B8A8_UNORM:
  72         case RB6_R8G8B8_UNORM:
  73         case RB6_R8G8_SINT:
  74         case RB6_R8G8_UINT:
  75         case RB6_R8G8_UNORM:
  76         case RB6_Z24_UNORM_S8_UINT:
  77         case RB6_Z24_UNORM_S8_UINT_AS_R8G8B8A8:
  78                 return true;
  79         default:
  80                 return false;
  81         }
  82 }
  83
  84 uint32_t
  85 fd6_fill_ubwc_buffer_sizes(struct fd_resource *rsc)
  86 {
  87 #define RBG_TILE_WIDTH_ALIGNMENT 64
  88 #define RGB_TILE_HEIGHT_ALIGNMENT 16
  89 #define UBWC_PLANE_SIZE_ALIGNMENT 4096
  90
  91         struct pipe_resource *prsc = &rsc->base;
  92         uint32_t width = prsc->width0;
  93         uint32_t height = prsc->height0;
  94
  95         if (!ok_ubwc_format(prsc->format))
  96                 return 0;
  97
  98         /* limit things to simple single level 2d for now: */
  99         if ((prsc->depth0 != 1) || (prsc->array_size != 1) || (prsc->last_level != 0))
 100                 return 0;
 101
 102         uint32_t block_width, block_height;
 103         switch (rsc->layout.cpp) {
 104         case 2:
 105         case 4:
 106                 block_width = 16;
 107                 block_height = 4;
 108                 break;
 109         case 8:
 110                 block_width = 8;
 111                 block_height = 4;
 112                 break;
 113         case 16:
 114                 block_width = 4;
 115                 block_height = 4;
 116                 break;
 117         default:
 118                 return 0;
 119         }
 120
 121         uint32_t meta_stride =
 122                 ALIGN_POT(DIV_ROUND_UP(width, block_width), RBG_TILE_WIDTH_ALIGNMENT);
 123         uint32_t meta_height =
 124                 ALIGN_POT(DIV_ROUND_UP(height, block_height), RGB_TILE_HEIGHT_ALIGNMENT);
 125         uint32_t meta_size =
 126                 ALIGN_POT(meta_stride * meta_height, UBWC_PLANE_SIZE_ALIGNMENT);
 127
 128         /* UBWC goes first, then color data.. this constraint is mainly only
 129          * because it is what the kernel expects for scanout.  For non-2D we
 130          * could just use a separate UBWC buffer..
 131          */
 132         for (int level = 0; level <= prsc->last_level; level++) {
 133                 struct fdl_slice *slice = fd_resource_slice(rsc, level);
 134                 slice->offset += meta_size;
 135         }
 136
 137         rsc->layout.ubwc_slices[0].offset = 0;
 138         rsc->layout.ubwc_slices[0].pitch = meta_stride;
 139         rsc->layout.ubwc_size = meta_size >> 2;   /* in dwords??? */
 140         rsc->layout.tile_mode = TILE6_3;
 141
 142         return meta_size;
 143 }
 144
 145 /**
 146  * Ensure the rsc is in an ok state to be used with the specified format.
 147  * This handles the case of UBWC buffers used with non-UBWC compatible
 148  * formats, by triggering an uncompress.
 149  */
 150 void
 151 fd6_validate_format(struct fd_context *ctx, struct fd_resource *rsc,
 152                 enum pipe_format format)
 153 {
 154         if (!rsc->layout.ubwc_size)
 155                 return;
 156
 157         if (ok_ubwc_format(format))
 158                 return;
 159
 160         fd_resource_uncompress(ctx, rsc);
 161 }
 162
 163 static void
 164 setup_lrz(struct fd_resource *rsc)
 165 {
 166         struct fd_screen *screen = fd_screen(rsc->base.screen);
 167         const uint32_t flags = DRM_FREEDRENO_GEM_CACHE_WCOMBINE |
 168                         DRM_FREEDRENO_GEM_TYPE_KMEM; /* TODO */
 169         unsigned width0 = rsc->base.width0;
 170         unsigned height0 = rsc->base.height0;
 171
 172         /* LRZ buffer is super-sampled: */
 173         switch (rsc->base.nr_samples) {
 174         case 4:
 175                 width0 *= 2;
 176                 /* fallthru */
 177         case 2:
 178                 height0 *= 2;
 179         }
 180
 181         unsigned lrz_pitch  = align(DIV_ROUND_UP(width0, 8), 32);
 182         unsigned lrz_height = align(DIV_ROUND_UP(height0, 8), 16);
 183
 184         unsigned size = lrz_pitch * lrz_height * 2;
 185
 186         rsc->lrz_height = lrz_height;
 187         rsc->lrz_width = lrz_pitch;
 188         rsc->lrz_pitch = lrz_pitch;
 189         rsc->lrz = fd_bo_new(screen->dev, size, flags, "lrz");
 190 }
 191
 192 uint32_t
 193 fd6_setup_slices(struct fd_resource *rsc)
 194 {
 195         struct pipe_resource *prsc = &rsc->base;
 196
 197         if (!(fd_mesa_debug & FD_DBG_NOLRZ) && has_depth(rsc->base.format))
 198                 setup_lrz(rsc);
 199
 200         fdl6_layout(&rsc->layout, prsc->format, fd_resource_nr_samples(prsc),
 201                         prsc->width0, prsc->height0, prsc->depth0,
 202                         prsc->last_level + 1, prsc->array_size,
 203                         prsc->target == PIPE_TEXTURE_3D, false);
 204
 205         /* The caller does this bit of layout setup again. */
 206         if (rsc->layout.layer_first)
 207                 return rsc->layout.size / prsc->array_size;
 208         else
 209                 return rsc->layout.size;
 210 }