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         /* Set flags for alternative formats */
  83
  84         if (surf->format == PIPE_FORMAT_B5G6R5_UNORM) {
  85                 fmt.unk1 = 0x14000000;
  86                 fmt.nr_channels = MALI_POSITIVE(2);
  87                 fmt.unk3 |= 0x1;
  88         } else if (surf->format == PIPE_FORMAT_R11G11B10_FLOAT) {
  89                 fmt.unk1 = 0x88000000;
  90                 fmt.unk3 = 0x0;
  91                 fmt.nr_channels = MALI_POSITIVE(4);
  92         }
  93
  94         return fmt;
  95 }
  96
  97
  98 static void
  99 panfrost_mfbd_clear(
 100                 struct panfrost_job *job,
 101                 struct bifrost_framebuffer *fb,
 102                 struct bifrost_fb_extra *fbx,
 103                 struct bifrost_render_target *rt)
 104 {
 105         if (job->clear & PIPE_CLEAR_COLOR) {
 106                 rt->clear_color_1 = job->clear_color;
 107                 rt->clear_color_2 = job->clear_color;
 108                 rt->clear_color_3 = job->clear_color;
 109                 rt->clear_color_4 = job->clear_color;
 110         }
 111
 112         if (job->clear & PIPE_CLEAR_DEPTH) {
 113                 fb->clear_depth = job->clear_depth;
 114         }
 115
 116         if (job->clear & PIPE_CLEAR_STENCIL) {
 117                 fb->clear_stencil = job->clear_stencil;
 118         }
 119 }
 120
 121 static void
 122 panfrost_mfbd_set_cbuf(
 123                 struct bifrost_render_target *rt,
 124                 struct pipe_surface *surf)
 125 {
 126         struct panfrost_resource *rsrc = pan_resource(surf->texture);
 127
 128         unsigned level = surf->u.tex.level;
 129         unsigned first_layer = surf->u.tex.first_layer;
 130         assert(surf->u.tex.last_layer == first_layer);
 131         int stride = rsrc->slices[level].stride;
 132
 133         mali_ptr base = panfrost_get_texture_address(rsrc, level, first_layer);
 134
 135         rt->format = panfrost_mfbd_format(surf);
 136
 137         /* Now, we set the layout specific pieces */
 138
 139         if (rsrc->layout == PAN_LINEAR) {
 140                 rt->format.block = MALI_MFBD_BLOCK_LINEAR;
 141                 rt->framebuffer = base;
 142                 rt->framebuffer_stride = stride / 16;
 143         } else if (rsrc->layout == PAN_TILED) {
 144                 rt->format.block = MALI_MFBD_BLOCK_TILED;
 145                 rt->framebuffer = base;
 146                 rt->framebuffer_stride = stride;
 147         } else if (rsrc->layout == PAN_AFBC) {
 148                 rt->format.block = MALI_MFBD_BLOCK_AFBC;
 149
 150                 unsigned header_size = rsrc->slices[level].header_size;
 151
 152                 rt->framebuffer = base + header_size;
 153                 rt->afbc.metadata = base;
 154                 rt->afbc.stride = 0;
 155                 rt->afbc.unk = 0x30009;
 156
 157                 /* TODO: Investigate shift */
 158                 rt->framebuffer_stride = stride << 1;
 159         } else {
 160                 fprintf(stderr, "Invalid render layout (cbuf)");
 161                 assert(0);
 162         }
 163 }
 164
 165 static void
 166 panfrost_mfbd_set_zsbuf(
 167                 struct bifrost_framebuffer *fb,
 168                 struct bifrost_fb_extra *fbx,
 169                 struct pipe_surface *surf)
 170 {
 171         struct panfrost_resource *rsrc = pan_resource(surf->texture);
 172
 173         unsigned level = surf->u.tex.level;
 174         assert(surf->u.tex.first_layer == 0);
 175
 176         unsigned offset = rsrc->slices[level].offset;
 177
 178         if (rsrc->layout == PAN_AFBC) {
 179                 mali_ptr base = rsrc->bo->gpu + offset;
 180                 unsigned header_size = rsrc->slices[level].header_size;
 181
 182                 fb->mfbd_flags |= MALI_MFBD_EXTRA;
 183
 184                 fbx->flags =
 185                         MALI_EXTRA_PRESENT |
 186                         MALI_EXTRA_AFBC |
 187                         MALI_EXTRA_AFBC_ZS |
 188                         MALI_EXTRA_ZS |
 189                         0x1; /* unknown */
 190
 191                 fbx->ds_afbc.depth_stencil = base + header_size;
 192                 fbx->ds_afbc.depth_stencil_afbc_metadata = base;
 193                 fbx->ds_afbc.depth_stencil_afbc_stride = 0;
 194
 195                 fbx->ds_afbc.zero1 = 0x10009;
 196                 fbx->ds_afbc.padding = 0x1000;
 197         } else if (rsrc->layout == PAN_LINEAR) {
 198                 int stride = rsrc->slices[level].stride;
 199                 fb->mfbd_flags |= MALI_MFBD_EXTRA;
 200
 201                 fbx->flags |= MALI_EXTRA_PRESENT | MALI_EXTRA_ZS | 0x1;
 202
 203                 fbx->ds_linear.depth = rsrc->bo->gpu + offset;
 204                 fbx->ds_linear.depth_stride = stride;
 205         } else {
 206                 assert(0);
 207         }
 208 }
 209
 210 /* Helper for sequential uploads used for MFBD */
 211
 212 #define UPLOAD(dest, offset, src, max) { \
 213         size_t sz = sizeof(*src); \
 214         memcpy(dest.cpu + offset, src, sz); \
 215         assert((offset + sz) <= max); \
 216         offset += sz; \
 217 }
 218
 219 static mali_ptr
 220 panfrost_mfbd_upload(
 221                 struct panfrost_context *ctx,
 222                 struct bifrost_framebuffer *fb,
 223                 struct bifrost_fb_extra *fbx,
 224                 struct bifrost_render_target *rts,
 225                 unsigned cbufs)
 226 {
 227         off_t offset = 0;
 228
 229         /* There may be extra data stuck in the middle */
 230         bool has_extra = fb->mfbd_flags & MALI_MFBD_EXTRA;
 231
 232         /* Compute total size for transfer */
 233
 234         size_t total_sz =
 235                 sizeof(struct bifrost_framebuffer) +
 236                 (has_extra ? sizeof(struct bifrost_fb_extra) : 0) +
 237                 sizeof(struct bifrost_render_target) * cbufs;
 238
 239         struct panfrost_transfer m_f_trans =
 240                 panfrost_allocate_transient(ctx, total_sz);
 241
 242         /* Do the transfer */
 243
 244         UPLOAD(m_f_trans, offset, fb, total_sz);
 245
 246         if (has_extra)
 247                 UPLOAD(m_f_trans, offset, fbx, total_sz);
 248
 249         for (unsigned c = 0; c < cbufs; ++c) {
 250                 UPLOAD(m_f_trans, offset, &rts[c], total_sz);
 251         }
 252
 253         /* Return pointer suitable for the fragment section */
 254         return m_f_trans.gpu | MALI_MFBD | (has_extra ? 2 : 0);
 255 }
 256
 257 #undef UPLOAD
 258
 259 /* Creates an MFBD for the FRAGMENT section of the bound framebuffer */
 260
 261 mali_ptr
 262 panfrost_mfbd_fragment(struct panfrost_context *ctx, bool has_draws)
 263 {
 264         struct panfrost_job *job = panfrost_get_job_for_fbo(ctx);
 265
 266         struct bifrost_framebuffer fb = panfrost_emit_mfbd(ctx, has_draws);
 267         struct bifrost_fb_extra fbx = {};
 268         struct bifrost_render_target rts[4] = {};
 269
 270         /* XXX: MRT case */
 271         fb.rt_count_2 = 1;
 272         fb.mfbd_flags = 0x100;
 273
 274         /* TODO: MRT clear */
 275         panfrost_mfbd_clear(job, &fb, &fbx, &rts[0]);
 276
 277         for (int cb = 0; cb < ctx->pipe_framebuffer.nr_cbufs; ++cb) {
 278                 struct pipe_surface *surf = ctx->pipe_framebuffer.cbufs[cb];
 279                 panfrost_mfbd_set_cbuf(&rts[cb], surf);
 280         }
 281
 282         if (ctx->pipe_framebuffer.zsbuf) {
 283                 panfrost_mfbd_set_zsbuf(&fb, &fbx, ctx->pipe_framebuffer.zsbuf);
 284         }
 285
 286         /* For the special case of a depth-only FBO, we need to attach a dummy render target */
 287
 288         if (ctx->pipe_framebuffer.nr_cbufs == 0) {
 289                 struct mali_rt_format null_rt = {
 290                         .unk1 = 0x4000000,
 291                         .unk4 = 0x8
 292                 };
 293
 294                 rts[0].format = null_rt;
 295                 rts[0].framebuffer = 0;
 296                 rts[0].framebuffer_stride = 0;
 297         }
 298
 299         /* When scanning out, the depth buffer is immediately invalidated, so
 300          * we don't need to waste bandwidth writing it out. This can improve
 301          * performance substantially (Z32_UNORM 1080p @ 60fps is 475 MB/s of
 302          * memory bandwidth!).
 303          *
 304          * The exception is ReadPixels, but this is not supported on GLES so we
 305          * can safely ignore it. */
 306
 307         if (panfrost_is_scanout(ctx)) {
 308                 job->requirements &= ~PAN_REQ_DEPTH_WRITE;
 309         }
 310
 311         /* Actualize the requirements */
 312
 313         if (job->requirements & PAN_REQ_MSAA) {
 314                 rts[0].format.flags |= MALI_MFBD_FORMAT_MSAA;
 315
 316                 /* XXX */
 317                 fb.unk1 |= (1 << 4) | (1 << 1);
 318                 fb.rt_count_2 = 4;
 319         }
 320
 321         if (job->requirements & PAN_REQ_DEPTH_WRITE)
 322                 fb.mfbd_flags |= MALI_MFBD_DEPTH_WRITE;
 323
 324         /* Checksumming only works with a single render target */
 325
 326         if (ctx->pipe_framebuffer.nr_cbufs == 1) {
 327                 struct pipe_surface *surf = ctx->pipe_framebuffer.cbufs[0];
 328                 struct panfrost_resource *rsrc = pan_resource(surf->texture);
 329                 struct panfrost_bo *bo = rsrc->bo;
 330
 331                 if (rsrc->checksummed) {
 332                         unsigned level = surf->u.tex.level;
 333                         struct panfrost_slice *slice = &rsrc->slices[level];
 334
 335                         fb.mfbd_flags |= MALI_MFBD_EXTRA;
 336                         fbx.flags |= MALI_EXTRA_PRESENT;
 337                         fbx.checksum_stride = slice->checksum_stride;
 338                         fbx.checksum = bo->gpu + slice->checksum_offset;
 339                 }
 340         }
 341
 342         /* We always upload at least one (dummy) cbuf */
 343         unsigned cbufs = MAX2(ctx->pipe_framebuffer.nr_cbufs, 1);
 344
 345         return panfrost_mfbd_upload(ctx, &fb, &fbx, rts, cbufs);
 346 }