src/gallium/drivers/panfrost/pan_mfbd.c

   1 /*
   2  * Copyright 2018-2019 Alyssa Rosenzweig
   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  */
  24
  25 #include "pan_context.h"
  26 #include "pan_util.h"
  27 #include "pan_format.h"
  28
  29 #include "util/u_format.h"
  30
  31 static struct mali_rt_format
  32 panfrost_mfbd_format(struct pipe_surface *surf)
  33 {
  34         /* Explode details on the format */
  35
  36         const struct util_format_description *desc =
  37                 util_format_description(surf->texture->format);
  38
  39         /* Fill in accordingly, defaulting to 8-bit UNORM */
  40
  41         struct mali_rt_format fmt = {
  42                 .unk1 = 0x4000000,
  43                 .unk2 = 0x1,
  44                 .nr_channels = MALI_POSITIVE(desc->nr_channels),
  45                 .unk3 = 0x4,
  46                 .flags = 0x8,
  47                 .swizzle = panfrost_translate_swizzle_4(desc->swizzle),
  48                 .unk4 = 0x8
  49         };
  50
  51         if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
  52                 fmt.flags |= MALI_MFBD_FORMAT_SRGB;
  53
  54         /* Set flags for alternative formats */
  55
  56         if (surf->texture->format == PIPE_FORMAT_B5G6R5_UNORM) {
  57                 fmt.unk1 = 0x14000000;
  58                 fmt.nr_channels = MALI_POSITIVE(2);
  59                 fmt.unk3 |= 0x1;
  60         }
  61
  62         return fmt;
  63 }
  64
  65
  66 static void
  67 panfrost_mfbd_clear(
  68                 struct panfrost_job *job,
  69                 struct bifrost_framebuffer *fb,
  70                 struct bifrost_fb_extra *fbx,
  71                 struct bifrost_render_target *rt)
  72 {
  73         if (job->clear & PIPE_CLEAR_COLOR) {
  74                 rt->clear_color_1 = job->clear_color;
  75                 rt->clear_color_2 = job->clear_color;
  76                 rt->clear_color_3 = job->clear_color;
  77                 rt->clear_color_4 = job->clear_color;
  78         }
  79
  80         if (job->clear & PIPE_CLEAR_DEPTH) {
  81                 fb->clear_depth = job->clear_depth;
  82         }
  83
  84         if (job->clear & PIPE_CLEAR_STENCIL) {
  85                 fb->clear_stencil = job->clear_stencil;
  86         }
  87 }
  88
  89 static void
  90 panfrost_mfbd_set_cbuf(
  91                 struct bifrost_render_target *rt,
  92                 struct pipe_surface *surf)
  93 {
  94         struct panfrost_resource *rsrc = pan_resource(surf->texture);
  95
  96         unsigned level = surf->u.tex.level;
  97         assert(surf->u.tex.first_layer == 0);
  98
  99         int stride = rsrc->bo->slices[level].stride;
 100         unsigned offset = rsrc->bo->slices[level].offset;
 101
 102         rt->format = panfrost_mfbd_format(surf);
 103
 104         /* Now, we set the layout specific pieces */
 105
 106         if (rsrc->bo->layout == PAN_LINEAR) {
 107                 rt->format.block = MALI_MFBD_BLOCK_LINEAR;
 108                 rt->framebuffer = rsrc->bo->gpu + offset;
 109                 rt->framebuffer_stride = stride / 16;
 110         } else if (rsrc->bo->layout == PAN_TILED) {
 111                 rt->format.block = MALI_MFBD_BLOCK_TILED;
 112                 rt->framebuffer = rsrc->bo->gpu + offset;
 113                 rt->framebuffer_stride = stride;
 114         } else if (rsrc->bo->layout == PAN_AFBC) {
 115                 rt->format.block = MALI_MFBD_BLOCK_AFBC;
 116
 117                 mali_ptr base = rsrc->bo->gpu + offset;
 118                 unsigned header_size = rsrc->bo->slices[level].header_size;
 119
 120                 rt->framebuffer = base + header_size;
 121                 rt->afbc.metadata = base;
 122                 rt->afbc.stride = 0;
 123                 rt->afbc.unk = 0x30009;
 124
 125                 /* TODO: Investigate shift */
 126                 rt->framebuffer_stride = stride << 1;
 127         } else {
 128                 fprintf(stderr, "Invalid render layout (cbuf)");
 129                 assert(0);
 130         }
 131 }
 132
 133 static void
 134 panfrost_mfbd_set_zsbuf(
 135                 struct bifrost_framebuffer *fb,
 136                 struct bifrost_fb_extra *fbx,
 137                 struct pipe_surface *surf)
 138 {
 139         struct panfrost_resource *rsrc = pan_resource(surf->texture);
 140
 141         unsigned level = surf->u.tex.level;
 142         assert(surf->u.tex.first_layer == 0);
 143
 144         unsigned offset = rsrc->bo->slices[level].offset;
 145
 146         if (rsrc->bo->layout == PAN_AFBC) {
 147                 mali_ptr base = rsrc->bo->gpu + offset;
 148                 unsigned header_size = rsrc->bo->slices[level].header_size;
 149
 150                 fb->mfbd_flags |= MALI_MFBD_EXTRA;
 151
 152                 fbx->flags =
 153                         MALI_EXTRA_PRESENT |
 154                         MALI_EXTRA_AFBC |
 155                         MALI_EXTRA_AFBC_ZS |
 156                         MALI_EXTRA_ZS |
 157                         0x1; /* unknown */
 158
 159                 fbx->ds_afbc.depth_stencil = base + header_size;
 160                 fbx->ds_afbc.depth_stencil_afbc_metadata = base;
 161                 fbx->ds_afbc.depth_stencil_afbc_stride = 0;
 162
 163                 fbx->ds_afbc.zero1 = 0x10009;
 164                 fbx->ds_afbc.padding = 0x1000;
 165         } else if (rsrc->bo->layout == PAN_LINEAR) {
 166                 int stride = rsrc->bo->slices[level].stride;
 167                 fb->mfbd_flags |= MALI_MFBD_EXTRA;
 168
 169                 fbx->flags |= MALI_EXTRA_PRESENT | MALI_EXTRA_ZS | 0x1;
 170
 171                 fbx->ds_linear.depth = rsrc->bo->gpu + offset;
 172                 fbx->ds_linear.depth_stride = stride;
 173         } else {
 174                 assert(0);
 175         }
 176 }
 177
 178 /* Helper for sequential uploads used for MFBD */
 179
 180 #define UPLOAD(dest, offset, src, max) { \
 181         size_t sz = sizeof(*src); \
 182         memcpy(dest.cpu + offset, src, sz); \
 183         assert((offset + sz) <= max); \
 184         offset += sz; \
 185 }
 186
 187 static mali_ptr
 188 panfrost_mfbd_upload(
 189                 struct panfrost_context *ctx,
 190                 struct bifrost_framebuffer *fb,
 191                 struct bifrost_fb_extra *fbx,
 192                 struct bifrost_render_target *rts,
 193                 unsigned cbufs)
 194 {
 195         off_t offset = 0;
 196
 197         /* There may be extra data stuck in the middle */
 198         bool has_extra = fb->mfbd_flags & MALI_MFBD_EXTRA;
 199
 200         /* Compute total size for transfer */
 201
 202         size_t total_sz =
 203                 sizeof(struct bifrost_framebuffer) +
 204                 (has_extra ? sizeof(struct bifrost_fb_extra) : 0) +
 205                 sizeof(struct bifrost_render_target) * cbufs;
 206
 207         struct panfrost_transfer m_f_trans =
 208                 panfrost_allocate_transient(ctx, total_sz);
 209
 210         /* Do the transfer */
 211
 212         UPLOAD(m_f_trans, offset, fb, total_sz);
 213
 214         if (has_extra)
 215                 UPLOAD(m_f_trans, offset, fbx, total_sz);
 216
 217         for (unsigned c = 0; c < cbufs; ++c) {
 218                 UPLOAD(m_f_trans, offset, &rts[c], total_sz);
 219         }
 220
 221         /* Return pointer suitable for the fragment section */
 222         return m_f_trans.gpu | MALI_MFBD | (has_extra ? 2 : 0);
 223 }
 224
 225 #undef UPLOAD
 226
 227 /* Creates an MFBD for the FRAGMENT section of the bound framebuffer */
 228
 229 mali_ptr
 230 panfrost_mfbd_fragment(struct panfrost_context *ctx, bool has_draws)
 231 {
 232         struct panfrost_job *job = panfrost_get_job_for_fbo(ctx);
 233
 234         struct bifrost_framebuffer fb = panfrost_emit_mfbd(ctx, has_draws);
 235         struct bifrost_fb_extra fbx = {};
 236         struct bifrost_render_target rts[4] = {};
 237
 238         /* XXX: MRT case */
 239         fb.rt_count_2 = 1;
 240         fb.mfbd_flags = 0x100;
 241
 242         /* TODO: MRT clear */
 243         panfrost_mfbd_clear(job, &fb, &fbx, &rts[0]);
 244
 245         for (int cb = 0; cb < ctx->pipe_framebuffer.nr_cbufs; ++cb) {
 246                 struct pipe_surface *surf = ctx->pipe_framebuffer.cbufs[cb];
 247                 panfrost_mfbd_set_cbuf(&rts[cb], surf);
 248         }
 249
 250         if (ctx->pipe_framebuffer.zsbuf) {
 251                 panfrost_mfbd_set_zsbuf(&fb, &fbx, ctx->pipe_framebuffer.zsbuf);
 252         }
 253
 254         /* For the special case of a depth-only FBO, we need to attach a dummy render target */
 255
 256         if (ctx->pipe_framebuffer.nr_cbufs == 0) {
 257                 struct mali_rt_format null_rt = {
 258                         .unk1 = 0x4000000,
 259                         .unk4 = 0x8
 260                 };
 261
 262                 rts[0].format = null_rt;
 263                 rts[0].framebuffer = 0;
 264                 rts[0].framebuffer_stride = 0;
 265         }
 266
 267         /* When scanning out, the depth buffer is immediately invalidated, so
 268          * we don't need to waste bandwidth writing it out. This can improve
 269          * performance substantially (Z32_UNORM 1080p @ 60fps is 475 MB/s of
 270          * memory bandwidth!).
 271          *
 272          * The exception is ReadPixels, but this is not supported on GLES so we
 273          * can safely ignore it. */
 274
 275         if (panfrost_is_scanout(ctx)) {
 276                 job->requirements &= ~PAN_REQ_DEPTH_WRITE;
 277         }
 278
 279         /* Actualize the requirements */
 280
 281         if (job->requirements & PAN_REQ_MSAA) {
 282                 rts[0].format.flags |= MALI_MFBD_FORMAT_MSAA;
 283
 284                 /* XXX */
 285                 fb.unk1 |= (1 << 4) | (1 << 1);
 286                 fb.rt_count_2 = 4;
 287         }
 288
 289         if (job->requirements & PAN_REQ_DEPTH_WRITE)
 290                 fb.mfbd_flags |= MALI_MFBD_DEPTH_WRITE;
 291
 292         if (ctx->pipe_framebuffer.nr_cbufs == 1) {
 293                 struct panfrost_resource *rsrc = (struct panfrost_resource *) ctx->pipe_framebuffer.cbufs[0]->texture;
 294
 295                 if (rsrc->bo->has_checksum) {
 296                         fb.mfbd_flags |= MALI_MFBD_EXTRA;
 297                         fbx.flags |= MALI_EXTRA_PRESENT;
 298                         fbx.checksum_stride = rsrc->bo->checksum_stride;
 299                         fbx.checksum = rsrc->bo->gpu + rsrc->bo->slices[0].stride * rsrc->base.height0;
 300                 }
 301         }
 302
 303         /* We always upload at least one (dummy) cbuf */
 304         unsigned cbufs = MAX2(ctx->pipe_framebuffer.nr_cbufs, 1);
 305
 306         return panfrost_mfbd_upload(ctx, &fb, &fbx, rts, cbufs);
 307 }