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