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 void
  32 panfrost_invert_swizzle(const unsigned char *in, unsigned char *out)
  33 {
  34         /* First, default to all zeroes to prevent uninitialized junk */
  35
  36         for (unsigned c = 0; c < 4; ++c)
  37                 out[c] = PIPE_SWIZZLE_0;
  38
  39         /* Now "do" what the swizzle says */
  40
  41         for (unsigned c = 0; c < 4; ++c) {
  42                 unsigned char i = in[c];
  43
  44                 /* Who cares? */
  45                 if (i < PIPE_SWIZZLE_X || i > PIPE_SWIZZLE_W)
  46                         continue;
  47
  48                 /* Invert */
  49                 unsigned idx = i - PIPE_SWIZZLE_X;
  50                 out[idx] = PIPE_SWIZZLE_X + c;
  51         }
  52 }
  53
  54 static struct mali_rt_format
  55 panfrost_mfbd_format(struct pipe_surface *surf)
  56 {
  57         /* Explode details on the format */
  58
  59         const struct util_format_description *desc =
  60                 util_format_description(surf->format);
  61
  62         /* The swizzle for rendering is inverted from texturing */
  63
  64         unsigned char swizzle[4];
  65         panfrost_invert_swizzle(desc->swizzle, swizzle);
  66
  67         /* Fill in accordingly, defaulting to 8-bit UNORM */
  68
  69         struct mali_rt_format fmt = {
  70                 .unk1 = 0x4000000,
  71                 .unk2 = 0x1,
  72                 .nr_channels = MALI_POSITIVE(desc->nr_channels),
  73                 .unk3 = 0x4,
  74                 .flags = 0x8,
  75                 .swizzle = panfrost_translate_swizzle_4(swizzle),
  76                 .unk4 = 0x8
  77         };
  78
  79         if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
  80                 fmt.flags |= MALI_MFBD_FORMAT_SRGB;
  81
  82         /* sRGB handled as a dedicated flag */
  83         enum pipe_format linearized = util_format_linear(surf->format);
  84
  85         /* If RGB, we're good to go */
  86         if (util_format_is_unorm8(desc))
  87                 return fmt;
  88
  89         /* Set flags for alternative formats */
  90
  91         switch (linearized) {
  92                 case PIPE_FORMAT_B5G6R5_UNORM:
  93                         fmt.unk1 = 0x14000000;
  94                         fmt.nr_channels = MALI_POSITIVE(2);
  95                         fmt.unk3 |= 0x1;
  96                         break;
  97
  98                 case PIPE_FORMAT_A4B4G4R4_UNORM:
  99                 case PIPE_FORMAT_B4G4R4A4_UNORM:
 100                         fmt.unk1 = 0x10000000;
 101                         fmt.unk3 = 0x5;
 102                         fmt.nr_channels = MALI_POSITIVE(1);
 103                         break;
 104
 105                 case PIPE_FORMAT_R10G10B10A2_UNORM:
 106                 case PIPE_FORMAT_B10G10R10A2_UNORM:
 107                 case PIPE_FORMAT_R10G10B10X2_UNORM:
 108                 case PIPE_FORMAT_B10G10R10X2_UNORM:
 109                         fmt.unk1 = 0x08000000;
 110                         fmt.unk3 = 0x6;
 111                         fmt.nr_channels = MALI_POSITIVE(1);
 112                         break;
 113
 114                         /* Generic 8-bit */
 115                 case PIPE_FORMAT_R8_UINT:
 116                 case PIPE_FORMAT_R8_SINT:
 117                         fmt.unk1 = 0x80000000;
 118                         fmt.unk3 = 0x0;
 119                         fmt.nr_channels = MALI_POSITIVE(1);
 120                         break;
 121
 122                         /* Generic 32-bit */
 123                 case PIPE_FORMAT_R11G11B10_FLOAT:
 124                 case PIPE_FORMAT_R8G8B8A8_UINT:
 125                 case PIPE_FORMAT_R8G8B8A8_SINT:
 126                 case PIPE_FORMAT_R16G16_FLOAT:
 127                 case PIPE_FORMAT_R16G16_UINT:
 128                 case PIPE_FORMAT_R16G16_SINT:
 129                 case PIPE_FORMAT_R32_FLOAT:
 130                 case PIPE_FORMAT_R32_UINT:
 131                 case PIPE_FORMAT_R32_SINT:
 132                 case PIPE_FORMAT_R10G10B10A2_UINT:
 133                         fmt.unk1 = 0x88000000;
 134                         fmt.unk3 = 0x0;
 135                         fmt.nr_channels = MALI_POSITIVE(4);
 136                         break;
 137
 138                         /* Generic 16-bit */
 139                 case PIPE_FORMAT_R8G8_UINT:
 140                 case PIPE_FORMAT_R8G8_SINT:
 141                 case PIPE_FORMAT_R16_FLOAT:
 142                 case PIPE_FORMAT_R16_UINT:
 143                 case PIPE_FORMAT_R16_SINT:
 144                 case PIPE_FORMAT_B5G5R5A1_UNORM:
 145                         fmt.unk1 = 0x84000000;
 146                         fmt.unk3 = 0x0;
 147                         fmt.nr_channels = MALI_POSITIVE(2);
 148                         break;
 149
 150                         /* Generic 64-bit */
 151                 case PIPE_FORMAT_R32G32_FLOAT:
 152                 case PIPE_FORMAT_R32G32_SINT:
 153                 case PIPE_FORMAT_R32G32_UINT:
 154                 case PIPE_FORMAT_R16G16B16A16_FLOAT:
 155                 case PIPE_FORMAT_R16G16B16A16_SINT:
 156                 case PIPE_FORMAT_R16G16B16A16_UINT:
 157                         fmt.unk1 = 0x8c000000;
 158                         fmt.unk3 = 0x1;
 159                         fmt.nr_channels = MALI_POSITIVE(2);
 160                         break;
 161
 162                         /* Generic 128-bit */
 163                 case PIPE_FORMAT_R32G32B32A32_FLOAT:
 164                 case PIPE_FORMAT_R32G32B32A32_SINT:
 165                 case PIPE_FORMAT_R32G32B32A32_UINT:
 166                         fmt.unk1 = 0x90000000;
 167                         fmt.unk3 = 0x1;
 168                         fmt.nr_channels = MALI_POSITIVE(4);
 169                         break;
 170
 171                 default:
 172                         unreachable("Invalid format rendering");
 173         }
 174
 175         return fmt;
 176 }
 177
 178
 179 static void
 180 panfrost_mfbd_clear(
 181                 struct panfrost_job *job,
 182                 struct bifrost_framebuffer *fb,
 183                 struct bifrost_fb_extra *fbx,
 184                 struct bifrost_render_target *rts,
 185                 unsigned rt_count)
 186 {
 187         for (unsigned i = 0; i < rt_count; ++i) {
 188                 if (!(job->clear & (PIPE_CLEAR_COLOR0 << i)))
 189                         continue;
 190
 191                 rts[i].clear_color_1 = job->clear_color[i][0];
 192                 rts[i].clear_color_2 = job->clear_color[i][1];
 193                 rts[i].clear_color_3 = job->clear_color[i][2];
 194                 rts[i].clear_color_4 = job->clear_color[i][3];
 195         }
 196
 197         if (job->clear & PIPE_CLEAR_DEPTH) {
 198                 fb->clear_depth = job->clear_depth;
 199         }
 200
 201         if (job->clear & PIPE_CLEAR_STENCIL) {
 202                 fb->clear_stencil = job->clear_stencil;
 203         }
 204 }
 205
 206 static void
 207 panfrost_mfbd_set_cbuf(
 208                 struct bifrost_render_target *rt,
 209                 struct pipe_surface *surf)
 210 {
 211         struct panfrost_resource *rsrc = pan_resource(surf->texture);
 212
 213         unsigned level = surf->u.tex.level;
 214         unsigned first_layer = surf->u.tex.first_layer;
 215         assert(surf->u.tex.last_layer == first_layer);
 216         int stride = rsrc->slices[level].stride;
 217
 218         mali_ptr base = panfrost_get_texture_address(rsrc, level, first_layer);
 219
 220         rt->format = panfrost_mfbd_format(surf);
 221
 222         /* Now, we set the layout specific pieces */
 223
 224         if (rsrc->layout == PAN_LINEAR) {
 225                 rt->format.block = MALI_MFBD_BLOCK_LINEAR;
 226                 rt->framebuffer = base;
 227                 rt->framebuffer_stride = stride / 16;
 228         } else if (rsrc->layout == PAN_TILED) {
 229                 rt->format.block = MALI_MFBD_BLOCK_TILED;
 230                 rt->framebuffer = base;
 231                 rt->framebuffer_stride = stride;
 232         } else if (rsrc->layout == PAN_AFBC) {
 233                 rt->format.block = MALI_MFBD_BLOCK_AFBC;
 234
 235                 unsigned header_size = rsrc->slices[level].header_size;
 236
 237                 rt->framebuffer = base + header_size;
 238                 rt->afbc.metadata = base;
 239                 rt->afbc.stride = 0;
 240                 rt->afbc.unk = 0x30009;
 241
 242                 /* TODO: Investigate shift */
 243                 rt->framebuffer_stride = stride << 1;
 244         } else {
 245                 fprintf(stderr, "Invalid render layout (cbuf)");
 246                 assert(0);
 247         }
 248 }
 249
 250 static void
 251 panfrost_mfbd_set_zsbuf(
 252                 struct bifrost_framebuffer *fb,
 253                 struct bifrost_fb_extra *fbx,
 254                 struct pipe_surface *surf)
 255 {
 256         struct panfrost_resource *rsrc = pan_resource(surf->texture);
 257
 258         unsigned level = surf->u.tex.level;
 259         assert(surf->u.tex.first_layer == 0);
 260
 261         unsigned offset = rsrc->slices[level].offset;
 262
 263         if (rsrc->layout == PAN_AFBC) {
 264                 mali_ptr base = rsrc->bo->gpu + offset;
 265                 unsigned header_size = rsrc->slices[level].header_size;
 266
 267                 fb->mfbd_flags |= MALI_MFBD_EXTRA;
 268
 269                 fbx->flags =
 270                         MALI_EXTRA_PRESENT |
 271                         MALI_EXTRA_AFBC |
 272                         MALI_EXTRA_AFBC_ZS |
 273                         MALI_EXTRA_ZS |
 274                         0x1; /* unknown */
 275
 276                 fbx->ds_afbc.depth_stencil = base + header_size;
 277                 fbx->ds_afbc.depth_stencil_afbc_metadata = base;
 278                 fbx->ds_afbc.depth_stencil_afbc_stride = 0;
 279
 280                 fbx->ds_afbc.zero1 = 0x10009;
 281                 fbx->ds_afbc.padding = 0x1000;
 282         } else if (rsrc->layout == PAN_LINEAR) {
 283                 int stride = rsrc->slices[level].stride;
 284                 fb->mfbd_flags |= MALI_MFBD_EXTRA;
 285
 286                 fbx->flags |= MALI_EXTRA_PRESENT | MALI_EXTRA_ZS | 0x1;
 287
 288                 fbx->ds_linear.depth = rsrc->bo->gpu + offset;
 289                 fbx->ds_linear.depth_stride = stride;
 290         } else {
 291                 assert(0);
 292         }
 293 }
 294
 295 /* Helper for sequential uploads used for MFBD */
 296
 297 #define UPLOAD(dest, offset, src, max) { \
 298         size_t sz = sizeof(*src); \
 299         memcpy(dest.cpu + offset, src, sz); \
 300         assert((offset + sz) <= max); \
 301         offset += sz; \
 302 }
 303
 304 static mali_ptr
 305 panfrost_mfbd_upload(
 306                 struct panfrost_context *ctx,
 307                 struct bifrost_framebuffer *fb,
 308                 struct bifrost_fb_extra *fbx,
 309                 struct bifrost_render_target *rts,
 310                 unsigned cbufs)
 311 {
 312         off_t offset = 0;
 313
 314         /* There may be extra data stuck in the middle */
 315         bool has_extra = fb->mfbd_flags & MALI_MFBD_EXTRA;
 316
 317         /* Compute total size for transfer */
 318
 319         size_t total_sz =
 320                 sizeof(struct bifrost_framebuffer) +
 321                 (has_extra ? sizeof(struct bifrost_fb_extra) : 0) +
 322                 sizeof(struct bifrost_render_target) * cbufs;
 323
 324         struct panfrost_transfer m_f_trans =
 325                 panfrost_allocate_transient(ctx, total_sz);
 326
 327         /* Do the transfer */
 328
 329         UPLOAD(m_f_trans, offset, fb, total_sz);
 330
 331         if (has_extra)
 332                 UPLOAD(m_f_trans, offset, fbx, total_sz);
 333
 334         for (unsigned c = 0; c < cbufs; ++c) {
 335                 UPLOAD(m_f_trans, offset, &rts[c], total_sz);
 336         }
 337
 338         /* Return pointer suitable for the fragment section */
 339         return m_f_trans.gpu | MALI_MFBD | (has_extra ? 2 : 0);
 340 }
 341
 342 #undef UPLOAD
 343
 344 /* Creates an MFBD for the FRAGMENT section of the bound framebuffer */
 345
 346 mali_ptr
 347 panfrost_mfbd_fragment(struct panfrost_context *ctx, bool has_draws)
 348 {
 349         struct panfrost_job *job = panfrost_get_job_for_fbo(ctx);
 350
 351         struct bifrost_framebuffer fb = panfrost_emit_mfbd(ctx, has_draws);
 352         struct bifrost_fb_extra fbx = {};
 353         struct bifrost_render_target rts[4] = {};
 354
 355         /* XXX: MRT case */
 356         fb.rt_count_2 = 1;
 357         fb.mfbd_flags = 0x100;
 358
 359         /* TODO: MRT clear */
 360         panfrost_mfbd_clear(job, &fb, &fbx, rts, fb.rt_count_2);
 361
 362         for (int cb = 0; cb < ctx->pipe_framebuffer.nr_cbufs; ++cb) {
 363                 struct pipe_surface *surf = ctx->pipe_framebuffer.cbufs[cb];
 364                 unsigned bpp = util_format_get_blocksize(surf->format);
 365
 366                 panfrost_mfbd_set_cbuf(&rts[cb], surf);
 367
 368                 /* What is this? Looks like some extension of the bpp field.
 369                  * Maybe it establishes how much internal tilebuffer space is
 370                  * reserved? */
 371                 fb.rt_count_2 = MAX2(fb.rt_count_2, ALIGN_POT(bpp, 4) / 4);
 372         }
 373
 374         if (ctx->pipe_framebuffer.zsbuf) {
 375                 panfrost_mfbd_set_zsbuf(&fb, &fbx, ctx->pipe_framebuffer.zsbuf);
 376         }
 377
 378         /* For the special case of a depth-only FBO, we need to attach a dummy render target */
 379
 380         if (ctx->pipe_framebuffer.nr_cbufs == 0) {
 381                 struct mali_rt_format null_rt = {
 382                         .unk1 = 0x4000000,
 383                         .unk4 = 0x8
 384                 };
 385
 386                 rts[0].format = null_rt;
 387                 rts[0].framebuffer = 0;
 388                 rts[0].framebuffer_stride = 0;
 389         }
 390
 391         /* When scanning out, the depth buffer is immediately invalidated, so
 392          * we don't need to waste bandwidth writing it out. This can improve
 393          * performance substantially (Z32_UNORM 1080p @ 60fps is 475 MB/s of
 394          * memory bandwidth!).
 395          *
 396          * The exception is ReadPixels, but this is not supported on GLES so we
 397          * can safely ignore it. */
 398
 399         if (panfrost_is_scanout(ctx)) {
 400                 job->requirements &= ~PAN_REQ_DEPTH_WRITE;
 401         }
 402
 403         /* Actualize the requirements */
 404
 405         if (job->requirements & PAN_REQ_MSAA) {
 406                 rts[0].format.flags |= MALI_MFBD_FORMAT_MSAA;
 407
 408                 /* XXX */
 409                 fb.unk1 |= (1 << 4) | (1 << 1);
 410                 fb.rt_count_2 = 4;
 411         }
 412
 413         if (job->requirements & PAN_REQ_DEPTH_WRITE)
 414                 fb.mfbd_flags |= MALI_MFBD_DEPTH_WRITE;
 415
 416         /* Checksumming only works with a single render target */
 417
 418         if (ctx->pipe_framebuffer.nr_cbufs == 1) {
 419                 struct pipe_surface *surf = ctx->pipe_framebuffer.cbufs[0];
 420                 struct panfrost_resource *rsrc = pan_resource(surf->texture);
 421                 struct panfrost_bo *bo = rsrc->bo;
 422
 423                 if (rsrc->checksummed) {
 424                         unsigned level = surf->u.tex.level;
 425                         struct panfrost_slice *slice = &rsrc->slices[level];
 426
 427                         fb.mfbd_flags |= MALI_MFBD_EXTRA;
 428                         fbx.flags |= MALI_EXTRA_PRESENT;
 429                         fbx.checksum_stride = slice->checksum_stride;
 430                         fbx.checksum = bo->gpu + slice->checksum_offset;
 431                 }
 432         }
 433
 434         /* We always upload at least one (dummy) cbuf */
 435         unsigned cbufs = MAX2(ctx->pipe_framebuffer.nr_cbufs, 1);
 436
 437         return panfrost_mfbd_upload(ctx, &fb, &fbx, rts, cbufs);
 438 }