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         struct panfrost_context *ctx = job->ctx;
  62
  63         if (job->clear & PIPE_CLEAR_COLOR) {
  64                 rt->clear_color_1 = job->clear_color;
  65                 rt->clear_color_2 = job->clear_color;
  66                 rt->clear_color_3 = job->clear_color;
  67                 rt->clear_color_4 = job->clear_color;
  68         }
  69
  70         if (job->clear & PIPE_CLEAR_DEPTH) {
  71                 fb->clear_depth = job->clear_depth;
  72         }
  73
  74         if (job->clear & PIPE_CLEAR_STENCIL) {
  75                 fb->clear_stencil = job->clear_stencil;
  76         }
  77
  78         if (job->clear & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) {
  79                 /* Setup combined 24/8 depth/stencil */
  80                 fb->unk3 |= MALI_MFBD_EXTRA;
  81                 fbx->flags = 0x405;
  82                 fbx->ds_linear.depth = ctx->depth_stencil_buffer.gpu;
  83                 fbx->ds_linear.depth_stride = ctx->pipe_framebuffer.width * 4;
  84         }
  85 }
  86
  87 static void
  88 panfrost_mfbd_set_cbuf(
  89                 struct bifrost_render_target *rt,
  90                 struct pipe_surface *surf,
  91                 bool flip_y)
  92 {
  93         struct panfrost_resource *rsrc = pan_resource(surf->texture);
  94
  95         signed stride =
  96                 util_format_get_stride(surf->format, surf->texture->width0);
  97
  98         rt->format = panfrost_mfbd_format(surf);
  99
 100         /* Now, we set the layout specific pieces */
 101
 102         if (rsrc->bo->layout == PAN_LINEAR) {
 103                 mali_ptr framebuffer = rsrc->bo->gpu[0];
 104
 105                 if (flip_y) {
 106                         framebuffer += stride * (surf->texture->height0 - 1);
 107                         stride = -stride;
 108                 }
 109
 110                 /* MFBD specifies stride in tiles */
 111                 rt->framebuffer = framebuffer;
 112                 rt->framebuffer_stride = stride / 16;
 113         } else if (rsrc->bo->layout == PAN_AFBC) {
 114                 rt->afbc.metadata = rsrc->bo->afbc_slab.gpu;
 115                 rt->afbc.stride = 0;
 116                 rt->afbc.unk = 0x30009;
 117
 118                 rt->format.flags |= MALI_MFBD_FORMAT_AFBC;
 119
 120                 mali_ptr afbc_main = rsrc->bo->afbc_slab.gpu + rsrc->bo->afbc_metadata_size;
 121                 rt->framebuffer = afbc_main;
 122
 123                 /* TODO: Investigate shift */
 124                 rt->framebuffer_stride = stride << 1;
 125         } else {
 126                 fprintf(stderr, "Invalid render layout (cbuf)");
 127                 assert(0);
 128         }
 129 }
 130
 131 static void
 132 panfrost_mfbd_set_zsbuf(
 133                 struct bifrost_framebuffer *fb,
 134                 struct bifrost_fb_extra *fbx,
 135                 struct pipe_surface *surf)
 136 {
 137         struct panfrost_resource *rsrc = pan_resource(surf->texture);
 138
 139         if (rsrc->bo->layout == PAN_AFBC) {
 140                 fb->unk3 |= MALI_MFBD_EXTRA;
 141
 142                 fbx->flags =
 143                         MALI_EXTRA_PRESENT |
 144                         MALI_EXTRA_AFBC |
 145                         MALI_EXTRA_AFBC_ZS |
 146                         MALI_EXTRA_ZS |
 147                         0x1; /* unknown */
 148
 149                 fbx->ds_afbc.depth_stencil_afbc_metadata = rsrc->bo->afbc_slab.gpu;
 150                 fbx->ds_afbc.depth_stencil_afbc_stride = 0;
 151
 152                 fbx->ds_afbc.depth_stencil = rsrc->bo->afbc_slab.gpu + rsrc->bo->afbc_metadata_size;
 153
 154                 fbx->ds_afbc.zero1 = 0x10009;
 155                 fbx->ds_afbc.padding = 0x1000;
 156
 157                 fb->unk3 |= MALI_MFBD_DEPTH_WRITE;
 158         }
 159 }
 160
 161 /* Helper for sequential uploads used for MFBD */
 162
 163 #define UPLOAD(dest, offset, src, max) { \
 164         size_t sz = sizeof(*src); \
 165         memcpy(dest.cpu + offset, src, sz); \
 166         assert((offset + sz) <= max); \
 167         offset += sz; \
 168 }
 169
 170 static mali_ptr
 171 panfrost_mfbd_upload(
 172                 struct panfrost_context *ctx,
 173                 struct bifrost_framebuffer *fb,
 174                 struct bifrost_fb_extra *fbx,
 175                 struct bifrost_render_target *rts,
 176                 unsigned cbufs)
 177 {
 178         off_t offset = 0;
 179
 180         /* There may be extra data stuck in the middle */
 181         bool has_extra = fb->unk3 & MALI_MFBD_EXTRA;
 182
 183         /* Compute total size for transfer */
 184
 185         size_t total_sz =
 186                 sizeof(struct bifrost_framebuffer) +
 187                 (has_extra ? sizeof(struct bifrost_fb_extra) : 0) +
 188                 sizeof(struct bifrost_render_target) * cbufs;
 189
 190         struct panfrost_transfer m_f_trans =
 191                 panfrost_allocate_transient(ctx, total_sz);
 192
 193         /* Do the transfer */
 194
 195         UPLOAD(m_f_trans, offset, fb, total_sz);
 196
 197         if (has_extra)
 198                 UPLOAD(m_f_trans, offset, fbx, total_sz);
 199
 200         for (unsigned c = 0; c < cbufs; ++c) {
 201                 UPLOAD(m_f_trans, offset, &rts[c], total_sz);
 202         }
 203
 204         /* Return pointer suitable for the fragment seciton */
 205         return m_f_trans.gpu | MALI_MFBD | (has_extra ? 2 : 0);
 206 }
 207
 208 #undef UPLOAD
 209
 210 /* Creates an MFBD for the FRAGMENT section of the bound framebuffer */
 211
 212 mali_ptr
 213 panfrost_mfbd_fragment(struct panfrost_context *ctx, bool flip_y)
 214 {
 215         struct panfrost_job *job = panfrost_get_job_for_fbo(ctx);
 216
 217         struct bifrost_framebuffer fb = panfrost_emit_mfbd(ctx);
 218         struct bifrost_fb_extra fbx = {};
 219         struct bifrost_render_target rts[4] = {};
 220
 221         /* XXX: MRT case */
 222         fb.rt_count_2 = 1;
 223         fb.unk3 = 0x100;
 224
 225         /* TODO: MRT clear */
 226         panfrost_mfbd_clear(job, &fb, &fbx, &rts[0]);
 227
 228         for (int cb = 0; cb < ctx->pipe_framebuffer.nr_cbufs; ++cb) {
 229                 struct pipe_surface *surf = ctx->pipe_framebuffer.cbufs[cb];
 230                 panfrost_mfbd_set_cbuf(&rts[cb], surf, flip_y);
 231         }
 232
 233         if (ctx->pipe_framebuffer.zsbuf) {
 234                 panfrost_mfbd_set_zsbuf(&fb, &fbx, ctx->pipe_framebuffer.zsbuf);
 235         }
 236
 237         /* For the special case of a depth-only FBO, we need to attach a dummy render target */
 238
 239         if (ctx->pipe_framebuffer.nr_cbufs == 0) {
 240                 struct mali_rt_format null_rt = {
 241                         .unk1 = 0x4000000,
 242                         .unk4 = 0x8
 243                 };
 244
 245                 rts[0].format = null_rt;
 246                 rts[0].framebuffer = 0;
 247                 rts[0].framebuffer_stride = 0;
 248         }
 249
 250         if (job->msaa) {
 251                 rts[0].format.flags |= MALI_MFBD_FORMAT_MSAA;
 252
 253                 /* XXX */
 254                 fb.unk1 |= (1 << 4) | (1 << 1);
 255                 fb.rt_count_2 = 4;
 256         }
 257
 258         if (ctx->pipe_framebuffer.nr_cbufs == 1) {
 259                 struct panfrost_resource *rsrc = (struct panfrost_resource *) ctx->pipe_framebuffer.cbufs[0]->texture;
 260
 261                 if (rsrc->bo->has_checksum) {
 262                         int stride = util_format_get_stride(rsrc->base.format, rsrc->base.width0);
 263
 264                         fb.unk3 |= MALI_MFBD_EXTRA;
 265                         fbx.flags |= MALI_EXTRA_PRESENT;
 266                         fbx.checksum_stride = rsrc->bo->checksum_stride;
 267                         fbx.checksum = rsrc->bo->gpu[0] + stride * rsrc->base.height0;
 268                 }
 269         }
 270
 271         /* We always upload at least one (dummy) cbuf */
 272         unsigned cbufs = MAX2(ctx->pipe_framebuffer.nr_cbufs, 1);
 273
 274         return panfrost_mfbd_upload(ctx, &fb, &fbx, rts, cbufs);
 275 }