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 RGBA8888 (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         /* Set flags for alternative formats */
  52
  53         if (surf->texture->format == PIPE_FORMAT_B5G6R5_UNORM) {
  54                 fmt.unk1 = 0x14000000;
  55                 fmt.nr_channels = MALI_POSITIVE(2);
  56                 fmt.unk3 |= 0x1;
  57         }
  58
  59         return fmt;
  60 }
  61
  62
  63 static void
  64 panfrost_mfbd_clear(
  65                 struct panfrost_job *job,
  66                 struct bifrost_framebuffer *fb,
  67                 struct bifrost_fb_extra *fbx,
  68                 struct bifrost_render_target *rt)
  69 {
  70         if (job->clear & PIPE_CLEAR_COLOR) {
  71                 rt->clear_color_1 = job->clear_color;
  72                 rt->clear_color_2 = job->clear_color;
  73                 rt->clear_color_3 = job->clear_color;
  74                 rt->clear_color_4 = job->clear_color;
  75         }
  76
  77         if (job->clear & PIPE_CLEAR_DEPTH) {
  78                 fb->clear_depth = job->clear_depth;
  79         }
  80
  81         if (job->clear & PIPE_CLEAR_STENCIL) {
  82                 fb->clear_stencil = job->clear_stencil;
  83         }
  84 }
  85
  86 static void
  87 panfrost_mfbd_set_cbuf(
  88                 struct bifrost_render_target *rt,
  89                 struct pipe_surface *surf)
  90 {
  91         struct panfrost_resource *rsrc = pan_resource(surf->texture);
  92
  93         unsigned level = surf->u.tex.level;
  94         assert(surf->u.tex.first_layer == 0);
  95
  96         int stride = rsrc->bo->slices[level].stride;
  97         unsigned offset = rsrc->bo->slices[level].offset;
  98
  99         rt->format = panfrost_mfbd_format(surf);
 100
 101         /* Now, we set the layout specific pieces */
 102
 103         if (rsrc->bo->layout == PAN_LINEAR) {
 104                 rt->format.block = MALI_MFBD_BLOCK_LINEAR;
 105                 rt->framebuffer = rsrc->bo->gpu + offset;
 106                 rt->framebuffer_stride = stride / 16;
 107         } else if (rsrc->bo->layout == PAN_TILED) {
 108                 rt->format.block = MALI_MFBD_BLOCK_TILED;
 109                 rt->framebuffer = rsrc->bo->gpu + offset;
 110                 rt->framebuffer_stride = stride;
 111         } else if (rsrc->bo->layout == PAN_AFBC) {
 112                 assert(level == 0);
 113                 rt->afbc.metadata = rsrc->bo->afbc_slab.gpu;
 114                 rt->afbc.stride = 0;
 115                 rt->afbc.unk = 0x30009;
 116
 117                 rt->format.block = MALI_MFBD_BLOCK_AFBC;
 118
 119                 mali_ptr afbc_main = rsrc->bo->afbc_slab.gpu + rsrc->bo->afbc_metadata_size;
 120                 rt->framebuffer = afbc_main;
 121
 122                 /* TODO: Investigate shift */
 123                 rt->framebuffer_stride = stride << 1;
 124         } else {
 125                 fprintf(stderr, "Invalid render layout (cbuf)");
 126                 assert(0);
 127         }
 128 }
 129
 130 static void
 131 panfrost_mfbd_set_zsbuf(
 132                 struct bifrost_framebuffer *fb,
 133                 struct bifrost_fb_extra *fbx,
 134                 struct pipe_surface *surf)
 135 {
 136         struct panfrost_resource *rsrc = pan_resource(surf->texture);
 137
 138         unsigned level = surf->u.tex.level;
 139         assert(surf->u.tex.first_layer == 0);
 140
 141         unsigned offset = rsrc->bo->slices[level].offset;
 142
 143         if (rsrc->bo->layout == PAN_AFBC) {
 144                 assert(level == 0);
 145                 fb->mfbd_flags |= MALI_MFBD_EXTRA;
 146
 147                 fbx->flags =
 148                         MALI_EXTRA_PRESENT |
 149                         MALI_EXTRA_AFBC |
 150                         MALI_EXTRA_AFBC_ZS |
 151                         MALI_EXTRA_ZS |
 152                         0x1; /* unknown */
 153
 154                 fbx->ds_afbc.depth_stencil_afbc_metadata = rsrc->bo->afbc_slab.gpu;
 155                 fbx->ds_afbc.depth_stencil_afbc_stride = 0;
 156
 157                 fbx->ds_afbc.depth_stencil = rsrc->bo->afbc_slab.gpu + rsrc->bo->afbc_metadata_size;
 158
 159                 fbx->ds_afbc.zero1 = 0x10009;
 160                 fbx->ds_afbc.padding = 0x1000;
 161         } else if (rsrc->bo->layout == PAN_LINEAR) {
 162                 int stride = rsrc->bo->slices[level].stride;
 163                 fb->mfbd_flags |= MALI_MFBD_EXTRA;
 164
 165                 fbx->flags |= MALI_EXTRA_PRESENT | MALI_EXTRA_ZS | 0x1;
 166
 167                 fbx->ds_linear.depth = rsrc->bo->gpu + offset;
 168                 fbx->ds_linear.depth_stride = stride;
 169         } else {
 170                 assert(0);
 171         }
 172 }
 173
 174 /* Helper for sequential uploads used for MFBD */
 175
 176 #define UPLOAD(dest, offset, src, max) { \
 177         size_t sz = sizeof(*src); \
 178         memcpy(dest.cpu + offset, src, sz); \
 179         assert((offset + sz) <= max); \
 180         offset += sz; \
 181 }
 182
 183 static mali_ptr
 184 panfrost_mfbd_upload(
 185                 struct panfrost_context *ctx,
 186                 struct bifrost_framebuffer *fb,
 187                 struct bifrost_fb_extra *fbx,
 188                 struct bifrost_render_target *rts,
 189                 unsigned cbufs)
 190 {
 191         off_t offset = 0;
 192
 193         /* There may be extra data stuck in the middle */
 194         bool has_extra = fb->mfbd_flags & MALI_MFBD_EXTRA;
 195
 196         /* Compute total size for transfer */
 197
 198         size_t total_sz =
 199                 sizeof(struct bifrost_framebuffer) +
 200                 (has_extra ? sizeof(struct bifrost_fb_extra) : 0) +
 201                 sizeof(struct bifrost_render_target) * cbufs;
 202
 203         struct panfrost_transfer m_f_trans =
 204                 panfrost_allocate_transient(ctx, total_sz);
 205
 206         /* Do the transfer */
 207
 208         UPLOAD(m_f_trans, offset, fb, total_sz);
 209
 210         if (has_extra)
 211                 UPLOAD(m_f_trans, offset, fbx, total_sz);
 212
 213         for (unsigned c = 0; c < cbufs; ++c) {
 214                 UPLOAD(m_f_trans, offset, &rts[c], total_sz);
 215         }
 216
 217         /* Return pointer suitable for the fragment section */
 218         return m_f_trans.gpu | MALI_MFBD | (has_extra ? 2 : 0);
 219 }
 220
 221 #undef UPLOAD
 222
 223 /* Creates an MFBD for the FRAGMENT section of the bound framebuffer */
 224
 225 mali_ptr
 226 panfrost_mfbd_fragment(struct panfrost_context *ctx, bool has_draws)
 227 {
 228         struct panfrost_job *job = panfrost_get_job_for_fbo(ctx);
 229
 230         struct bifrost_framebuffer fb = panfrost_emit_mfbd(ctx, has_draws);
 231         struct bifrost_fb_extra fbx = {};
 232         struct bifrost_render_target rts[4] = {};
 233
 234         /* XXX: MRT case */
 235         fb.rt_count_2 = 1;
 236         fb.mfbd_flags = 0x100;
 237
 238         /* TODO: MRT clear */
 239         panfrost_mfbd_clear(job, &fb, &fbx, &rts[0]);
 240
 241         for (int cb = 0; cb < ctx->pipe_framebuffer.nr_cbufs; ++cb) {
 242                 struct pipe_surface *surf = ctx->pipe_framebuffer.cbufs[cb];
 243                 panfrost_mfbd_set_cbuf(&rts[cb], surf);
 244         }
 245
 246         if (ctx->pipe_framebuffer.zsbuf) {
 247                 panfrost_mfbd_set_zsbuf(&fb, &fbx, ctx->pipe_framebuffer.zsbuf);
 248         }
 249
 250         /* For the special case of a depth-only FBO, we need to attach a dummy render target */
 251
 252         if (ctx->pipe_framebuffer.nr_cbufs == 0) {
 253                 struct mali_rt_format null_rt = {
 254                         .unk1 = 0x4000000,
 255                         .unk4 = 0x8
 256                 };
 257
 258                 rts[0].format = null_rt;
 259                 rts[0].framebuffer = 0;
 260                 rts[0].framebuffer_stride = 0;
 261         }
 262
 263         /* When scanning out, the depth buffer is immediately invalidated, so
 264          * we don't need to waste bandwidth writing it out. This can improve
 265          * performance substantially (Z32_UNORM 1080p @ 60fps is 475 MB/s of
 266          * memory bandwidth!).
 267          *
 268          * The exception is ReadPixels, but this is not supported on GLES so we
 269          * can safely ignore it. */
 270
 271         if (panfrost_is_scanout(ctx)) {
 272                 job->requirements &= ~PAN_REQ_DEPTH_WRITE;
 273         }
 274
 275         /* Actualize the requirements */
 276
 277         if (job->requirements & PAN_REQ_MSAA) {
 278                 rts[0].format.flags |= MALI_MFBD_FORMAT_MSAA;
 279
 280                 /* XXX */
 281                 fb.unk1 |= (1 << 4) | (1 << 1);
 282                 fb.rt_count_2 = 4;
 283         }
 284
 285         if (job->requirements & PAN_REQ_DEPTH_WRITE)
 286                 fb.mfbd_flags |= MALI_MFBD_DEPTH_WRITE;
 287
 288         if (ctx->pipe_framebuffer.nr_cbufs == 1) {
 289                 struct panfrost_resource *rsrc = (struct panfrost_resource *) ctx->pipe_framebuffer.cbufs[0]->texture;
 290
 291                 if (rsrc->bo->has_checksum) {
 292                         fb.mfbd_flags |= MALI_MFBD_EXTRA;
 293                         fbx.flags |= MALI_EXTRA_PRESENT;
 294                         fbx.checksum_stride = rsrc->bo->checksum_stride;
 295                         fbx.checksum = rsrc->bo->gpu + rsrc->bo->slices[0].stride * rsrc->base.height0;
 296                 }
 297         }
 298
 299         /* We always upload at least one (dummy) cbuf */
 300         unsigned cbufs = MAX2(ctx->pipe_framebuffer.nr_cbufs, 1);
 301
 302         return panfrost_mfbd_upload(ctx, &fb, &fbx, rts, cbufs);
 303 }