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 */
  40
  41         struct mali_rt_format fmt = {
  42                 .unk1 = 0x4000000,
  43                 .unk2 = 0x1,
  44                 .nr_channels = MALI_POSITIVE(desc->nr_channels),
  45                 .flags = 0x444,
  46                 .swizzle = panfrost_translate_swizzle_4(desc->swizzle),
  47                 .unk4 = 0x8
  48         };
  49
  50         return fmt;
  51 }
  52
  53
  54 static void
  55 panfrost_mfbd_clear(
  56                 struct panfrost_job *job,
  57                 struct bifrost_framebuffer *fb,
  58                 struct bifrost_fb_extra *fbx,
  59                 struct bifrost_render_target *rt)
  60 {
  61         if (job->clear & PIPE_CLEAR_COLOR) {
  62                 rt->clear_color_1 = job->clear_color;
  63                 rt->clear_color_2 = job->clear_color;
  64                 rt->clear_color_3 = job->clear_color;
  65                 rt->clear_color_4 = job->clear_color;
  66         }
  67
  68         if (job->clear & PIPE_CLEAR_DEPTH) {
  69                 fb->clear_depth = job->clear_depth;
  70         }
  71
  72         if (job->clear & PIPE_CLEAR_STENCIL) {
  73                 fb->clear_stencil = job->clear_stencil;
  74         }
  75 }
  76
  77 static void
  78 panfrost_mfbd_set_cbuf(
  79                 struct bifrost_render_target *rt,
  80                 struct pipe_surface *surf,
  81                 bool flip_y)
  82 {
  83         struct panfrost_resource *rsrc = pan_resource(surf->texture);
  84
  85         signed stride =
  86                 util_format_get_stride(surf->format, surf->texture->width0);
  87
  88         rt->format = panfrost_mfbd_format(surf);
  89
  90         /* Now, we set the layout specific pieces */
  91
  92         if (rsrc->bo->layout == PAN_LINEAR) {
  93                 mali_ptr framebuffer = rsrc->bo->gpu[0];
  94
  95                 if (flip_y) {
  96                         framebuffer += stride * (surf->texture->height0 - 1);
  97                         stride = -stride;
  98                 }
  99
 100                 /* MFBD specifies stride in tiles */
 101                 rt->framebuffer = framebuffer;
 102                 rt->framebuffer_stride = stride / 16;
 103         } else if (rsrc->bo->layout == PAN_AFBC) {
 104                 rt->afbc.metadata = rsrc->bo->afbc_slab.gpu;
 105                 rt->afbc.stride = 0;
 106                 rt->afbc.unk = 0x30009;
 107
 108                 rt->format.flags |= MALI_MFBD_FORMAT_AFBC;
 109
 110                 mali_ptr afbc_main = rsrc->bo->afbc_slab.gpu + rsrc->bo->afbc_metadata_size;
 111                 rt->framebuffer = afbc_main;
 112
 113                 /* TODO: Investigate shift */
 114                 rt->framebuffer_stride = stride << 1;
 115         } else {
 116                 fprintf(stderr, "Invalid render layout (cbuf)");
 117                 assert(0);
 118         }
 119 }
 120
 121 static void
 122 panfrost_mfbd_set_zsbuf(
 123                 struct bifrost_framebuffer *fb,
 124                 struct bifrost_fb_extra *fbx,
 125                 struct pipe_surface *surf)
 126 {
 127         struct panfrost_resource *rsrc = pan_resource(surf->texture);
 128
 129         if (rsrc->bo->layout == PAN_AFBC) {
 130                 fb->unk3 |= MALI_MFBD_EXTRA;
 131
 132                 fbx->flags =
 133                         MALI_EXTRA_PRESENT |
 134                         MALI_EXTRA_AFBC |
 135                         MALI_EXTRA_AFBC_ZS |
 136                         MALI_EXTRA_ZS |
 137                         0x1; /* unknown */
 138
 139                 fbx->ds_afbc.depth_stencil_afbc_metadata = rsrc->bo->afbc_slab.gpu;
 140                 fbx->ds_afbc.depth_stencil_afbc_stride = 0;
 141
 142                 fbx->ds_afbc.depth_stencil = rsrc->bo->afbc_slab.gpu + rsrc->bo->afbc_metadata_size;
 143
 144                 fbx->ds_afbc.zero1 = 0x10009;
 145                 fbx->ds_afbc.padding = 0x1000;
 146         } else if (rsrc->bo->layout == PAN_LINEAR) {
 147                 fb->unk3 |= MALI_MFBD_EXTRA;
 148                 fbx->flags |= MALI_EXTRA_PRESENT | MALI_EXTRA_ZS | 0x1;
 149
 150                 fbx->ds_linear.depth = rsrc->bo->gpu[0];
 151                 fbx->ds_linear.depth_stride =
 152                         util_format_get_stride(surf->format, surf->texture->width0);
 153         } else {
 154                 assert(0);
 155         }
 156 }
 157
 158 /* Helper for sequential uploads used for MFBD */
 159
 160 #define UPLOAD(dest, offset, src, max) { \
 161         size_t sz = sizeof(*src); \
 162         memcpy(dest.cpu + offset, src, sz); \
 163         assert((offset + sz) <= max); \
 164         offset += sz; \
 165 }
 166
 167 static mali_ptr
 168 panfrost_mfbd_upload(
 169                 struct panfrost_context *ctx,
 170                 struct bifrost_framebuffer *fb,
 171                 struct bifrost_fb_extra *fbx,
 172                 struct bifrost_render_target *rts,
 173                 unsigned cbufs)
 174 {
 175         off_t offset = 0;
 176
 177         /* There may be extra data stuck in the middle */
 178         bool has_extra = fb->unk3 & MALI_MFBD_EXTRA;
 179
 180         /* Compute total size for transfer */
 181
 182         size_t total_sz =
 183                 sizeof(struct bifrost_framebuffer) +
 184                 (has_extra ? sizeof(struct bifrost_fb_extra) : 0) +
 185                 sizeof(struct bifrost_render_target) * cbufs;
 186
 187         struct panfrost_transfer m_f_trans =
 188                 panfrost_allocate_transient(ctx, total_sz);
 189
 190         /* Do the transfer */
 191
 192         UPLOAD(m_f_trans, offset, fb, total_sz);
 193
 194         if (has_extra)
 195                 UPLOAD(m_f_trans, offset, fbx, total_sz);
 196
 197         for (unsigned c = 0; c < cbufs; ++c) {
 198                 UPLOAD(m_f_trans, offset, &rts[c], total_sz);
 199         }
 200
 201         /* Return pointer suitable for the fragment seciton */
 202         return m_f_trans.gpu | MALI_MFBD | (has_extra ? 2 : 0);
 203 }
 204
 205 #undef UPLOAD
 206
 207 /* Creates an MFBD for the FRAGMENT section of the bound framebuffer */
 208
 209 mali_ptr
 210 panfrost_mfbd_fragment(struct panfrost_context *ctx, bool flip_y)
 211 {
 212         struct panfrost_job *job = panfrost_get_job_for_fbo(ctx);
 213
 214         struct bifrost_framebuffer fb = panfrost_emit_mfbd(ctx);
 215         struct bifrost_fb_extra fbx = {};
 216         struct bifrost_render_target rts[4] = {};
 217
 218         /* XXX: MRT case */
 219         fb.rt_count_2 = 1;
 220         fb.unk3 = 0x100;
 221
 222         /* TODO: MRT clear */
 223         panfrost_mfbd_clear(job, &fb, &fbx, &rts[0]);
 224
 225         for (int cb = 0; cb < ctx->pipe_framebuffer.nr_cbufs; ++cb) {
 226                 struct pipe_surface *surf = ctx->pipe_framebuffer.cbufs[cb];
 227                 panfrost_mfbd_set_cbuf(&rts[cb], surf, flip_y);
 228         }
 229
 230         if (ctx->pipe_framebuffer.zsbuf) {
 231                 panfrost_mfbd_set_zsbuf(&fb, &fbx, ctx->pipe_framebuffer.zsbuf);
 232         }
 233
 234         /* For the special case of a depth-only FBO, we need to attach a dummy render target */
 235
 236         if (ctx->pipe_framebuffer.nr_cbufs == 0) {
 237                 struct mali_rt_format null_rt = {
 238                         .unk1 = 0x4000000,
 239                         .unk4 = 0x8
 240                 };
 241
 242                 rts[0].format = null_rt;
 243                 rts[0].framebuffer = 0;
 244                 rts[0].framebuffer_stride = 0;
 245         }
 246
 247         /* When scanning out, the depth buffer is immediately invalidated, so
 248          * we don't need to waste bandwidth writing it out. This can improve
 249          * performance substantially (Z32_UNORM 1080p @ 60fps is 475 MB/s of
 250          * memory bandwidth!).
 251          *
 252          * The exception is ReadPixels, but this is not supported on GLES so we
 253          * can safely ignore it. */
 254
 255         if (panfrost_is_scanout(ctx)) {
 256                 job->requirements &= ~PAN_REQ_DEPTH_WRITE;
 257         }
 258
 259         /* Actualize the requirements */
 260
 261         if (job->requirements & PAN_REQ_MSAA) {
 262                 rts[0].format.flags |= MALI_MFBD_FORMAT_MSAA;
 263
 264                 /* XXX */
 265                 fb.unk1 |= (1 << 4) | (1 << 1);
 266                 fb.rt_count_2 = 4;
 267         }
 268
 269         if (job->requirements & PAN_REQ_DEPTH_WRITE)
 270                 fb.unk3 |= MALI_MFBD_DEPTH_WRITE;
 271
 272         if (ctx->pipe_framebuffer.nr_cbufs == 1) {
 273                 struct panfrost_resource *rsrc = (struct panfrost_resource *) ctx->pipe_framebuffer.cbufs[0]->texture;
 274
 275                 if (rsrc->bo->has_checksum) {
 276                         int stride = util_format_get_stride(rsrc->base.format, rsrc->base.width0);
 277
 278                         fb.unk3 |= MALI_MFBD_EXTRA;
 279                         fbx.flags |= MALI_EXTRA_PRESENT;
 280                         fbx.checksum_stride = rsrc->bo->checksum_stride;
 281                         fbx.checksum = rsrc->bo->gpu[0] + stride * rsrc->base.height0;
 282                 }
 283         }
 284
 285         /* We always upload at least one (dummy) cbuf */
 286         unsigned cbufs = MAX2(ctx->pipe_framebuffer.nr_cbufs, 1);
 287
 288         return panfrost_mfbd_upload(ctx, &fb, &fbx, rts, cbufs);
 289 }