/* Adjust width/height/depth for MSAA */
mt->msaa_layout = compute_msaa_layout(brw, format, mt->target);
if (mt->msaa_layout == INTEL_MSAA_LAYOUT_IMS) {
- /* In the Sandy Bridge PRM, volume 4, part 1, page 31, it says:
+ /* From the Ivybridge PRM, Volume 1, Part 1, page 108:
+ * "If the surface is multisampled and it is a depth or stencil
+ * surface or Multisampled Surface StorageFormat in SURFACE_STATE is
+ * MSFMT_DEPTH_STENCIL, WL and HL must be adjusted as follows before
+ * proceeding:
+ *
+ * +----------------------------------------------------------------+
+ * | Num Multisamples | W_l = | H_l = |
+ * +----------------------------------------------------------------+
+ * | 2 | ceiling(W_l / 2) * 4 | H_l (no adjustment) |
+ * | 4 | ceiling(W_l / 2) * 4 | ceiling(H_l / 2) * 4 |
+ * | 8 | ceiling(W_l / 2) * 8 | ceiling(H_l / 2) * 4 |
+ * | 16 | ceiling(W_l / 2) * 8 | ceiling(H_l / 2) * 8 |
+ * +----------------------------------------------------------------+
+ * "
+ *
+ * Note that MSFMT_DEPTH_STENCIL just means the IMS (interleaved)
+ * format rather than UMS/CMS (array slices). The Sandybridge PRM,
+ * Volume 1, Part 1, Page 111 has the same formula for 4x MSAA.
+ *
+ * Another more complicated explanation for these adjustments comes
+ * from the Sandybridge PRM, volume 4, part 1, page 31:
*
* "Any of the other messages (sample*, LOD, load4) used with a
* (4x) multisampled surface will in-effect sample a surface with
base_format == GL_DEPTH_STENCIL_EXT)
return I915_TILING_Y;
+ /* 1D textures (and 1D array textures) don't get any benefit from tiling,
+ * in fact it leads to a less efficient use of memory space and bandwidth
+ * due to tile alignment.
+ */
+ if (mt->logical_height0 == 1)
+ return I915_TILING_NONE;
+
int minimum_pitch = mt->total_width * mt->cpp;
/* If the width is much smaller than a tile, don't bother tiling. */
/* From the Sandybridge PRM, Volume 1, Part 2, page 32:
* "NOTE: 128BPE Format Color Buffer ( render target ) MUST be either TileX
* or Linear."
- * 128 bits per pixel translates to 16 bytes per pixel. This is necessary
- * all the way back to 965, but is explicitly permitted on Gen7.
+ * 128 bits per pixel translates to 16 bytes per pixel. This is necessary
+ * all the way back to 965, but is permitted on Gen7+.
*/
- if (brw->gen != 7 && mt->cpp >= 16)
+ if (brw->gen < 7 && mt->cpp >= 16)
return I915_TILING_X;
/* From the Ivy Bridge PRM, Vol4 Part1 2.12.2.1 (SURFACE_STATE for most
mt->pitch = pitch;
/* If the BO is too large to fit in the aperture, we need to use the
- * BLT engine to support it. The BLT paths can't currently handle Y-tiling,
- * so we need to fall back to X.
+ * BLT engine to support it. Prior to Sandybridge, the BLT paths can't
+ * handle Y-tiling, so we need to fall back to X.
*/
- if (y_or_x && mt->bo->size >= brw->max_gtt_map_object_size) {
+ if (brw->gen < 6 && y_or_x && mt->bo->size >= brw->max_gtt_map_object_size) {
perf_debug("%dx%d miptree larger than aperture; falling back to X-tiled\n",
mt->total_width, mt->total_height);
uint32_t offset,
uint32_t width,
uint32_t height,
+ uint32_t depth,
int pitch)
{
struct intel_mipmap_tree *mt;
uint32_t tiling, swizzle;
+ GLenum target;
drm_intel_bo_get_tiling(bo, &tiling, &swizzle);
*/
assert(pitch >= 0);
- mt = intel_miptree_create_layout(brw, GL_TEXTURE_2D, format,
+ target = depth > 1 ? GL_TEXTURE_2D_ARRAY : GL_TEXTURE_2D;
+
+ mt = intel_miptree_create_layout(brw, target, format,
0, 0,
- width, height, 1,
+ width, height, depth,
true, 0, false);
if (!mt) {
free(mt);
0,
width,
height,
+ 1,
pitch);
if (!singlesample_mt)
goto fail;
drm_intel_bo_unreference((*mt)->bo);
intel_miptree_release(&(*mt)->stencil_mt);
- intel_miptree_release(&(*mt)->hiz_mt);
+ if ((*mt)->hiz_buf) {
+ if ((*mt)->hiz_buf->mt)
+ intel_miptree_release(&(*mt)->hiz_buf->mt);
+ else
+ drm_intel_bo_unreference((*mt)->hiz_buf->bo);
+ free((*mt)->hiz_buf);
+ }
intel_miptree_release(&(*mt)->mcs_mt);
intel_resolve_map_clear(&(*mt)->hiz_map);
intel_texture_object(intelImage->base.Base.TexObject);
int level = intelImage->base.Base.Level;
int face = intelImage->base.Base.Face;
- GLuint depth = intelImage->base.Base.Depth;
+
+ GLuint depth;
+ if (intel_obj->base.Target == GL_TEXTURE_1D_ARRAY)
+ depth = intelImage->base.Base.Height;
+ else
+ depth = intelImage->base.Base.Depth;
if (!invalidate) {
for (int slice = 0; slice < depth; slice++) {
struct intel_mipmap_tree *mt,
uint32_t level)
{
- assert(mt->hiz_mt);
+ assert(mt->hiz_buf);
if (brw->gen >= 8 || brw->is_haswell) {
uint32_t width = minify(mt->physical_width0, level);
}
+/**
+ * Helper for intel_miptree_alloc_hiz() that determines the required hiz
+ * buffer dimensions and allocates a bo for the hiz buffer.
+ */
+static struct intel_miptree_aux_buffer *
+intel_gen7_hiz_buf_create(struct brw_context *brw,
+ struct intel_mipmap_tree *mt)
+{
+ unsigned z_width = mt->logical_width0;
+ unsigned z_height = mt->logical_height0;
+ const unsigned z_depth = MAX2(mt->logical_depth0, 1);
+ unsigned hz_width, hz_height;
+ struct intel_miptree_aux_buffer *buf = calloc(sizeof(*buf), 1);
+
+ if (!buf)
+ return NULL;
+
+ /* Gen7 PRM Volume 2, Part 1, 11.5.3 "Hierarchical Depth Buffer" documents
+ * adjustments required for Z_Height and Z_Width based on multisampling.
+ */
+ switch (mt->num_samples) {
+ case 0:
+ case 1:
+ break;
+ case 2:
+ case 4:
+ z_width *= 2;
+ z_height *= 2;
+ break;
+ case 8:
+ z_width *= 4;
+ z_height *= 2;
+ break;
+ default:
+ unreachable("unsupported sample count");
+ }
+
+ const unsigned vertical_align = 8; /* 'j' in the docs */
+ const unsigned H0 = z_height;
+ const unsigned h0 = ALIGN(H0, vertical_align);
+ const unsigned h1 = ALIGN(minify(H0, 1), vertical_align);
+ const unsigned Z0 = z_depth;
+
+ /* HZ_Width (bytes) = ceiling(Z_Width / 16) * 16 */
+ hz_width = ALIGN(z_width, 16);
+
+ if (mt->target == GL_TEXTURE_3D) {
+ unsigned H_i = H0;
+ unsigned Z_i = Z0;
+ hz_height = 0;
+ for (int level = mt->first_level; level <= mt->last_level; ++level) {
+ unsigned h_i = ALIGN(H_i, vertical_align);
+ /* sum(i=0 to m; h_i * max(1, floor(Z_Depth/2**i))) */
+ hz_height += h_i * Z_i;
+ H_i = minify(H_i, 1);
+ Z_i = minify(Z_i, 1);
+ }
+ /* HZ_Height =
+ * (1/2) * sum(i=0 to m; h_i * max(1, floor(Z_Depth/2**i)))
+ */
+ hz_height = DIV_ROUND_UP(hz_height, 2);
+ } else {
+ const unsigned hz_qpitch = h0 + h1 + (12 * vertical_align);
+ if (mt->target == GL_TEXTURE_CUBE_MAP_ARRAY ||
+ mt->target == GL_TEXTURE_CUBE_MAP) {
+ /* HZ_Height (rows) = Ceiling ( ( Q_pitch * Z_depth * 6/2) /8 ) * 8 */
+ hz_height = DIV_ROUND_UP(hz_qpitch * Z0 * 6, 2 * 8) * 8;
+ } else {
+ /* HZ_Height (rows) = Ceiling ( ( Q_pitch * Z_depth/2) /8 ) * 8 */
+ hz_height = DIV_ROUND_UP(hz_qpitch * Z0, 2 * 8) * 8;
+ }
+ }
+
+ unsigned long pitch;
+ uint32_t tiling = I915_TILING_Y;
+ buf->bo = drm_intel_bo_alloc_tiled(brw->bufmgr, "hiz",
+ hz_width, hz_height, 1,
+ &tiling, &pitch,
+ BO_ALLOC_FOR_RENDER);
+ if (!buf->bo) {
+ free(buf);
+ return NULL;
+ } else if (tiling != I915_TILING_Y) {
+ drm_intel_bo_unreference(buf->bo);
+ free(buf);
+ return NULL;
+ }
+
+ buf->pitch = pitch;
+
+ return buf;
+}
+
+
+/**
+ * Helper for intel_miptree_alloc_hiz() that determines the required hiz
+ * buffer dimensions and allocates a bo for the hiz buffer.
+ */
+static struct intel_miptree_aux_buffer *
+intel_gen8_hiz_buf_create(struct brw_context *brw,
+ struct intel_mipmap_tree *mt)
+{
+ unsigned z_width = mt->logical_width0;
+ unsigned z_height = mt->logical_height0;
+ const unsigned z_depth = MAX2(mt->logical_depth0, 1);
+ unsigned hz_width, hz_height;
+ struct intel_miptree_aux_buffer *buf = calloc(sizeof(*buf), 1);
+
+ if (!buf)
+ return NULL;
+
+ /* Gen7 PRM Volume 2, Part 1, 11.5.3 "Hierarchical Depth Buffer" documents
+ * adjustments required for Z_Height and Z_Width based on multisampling.
+ */
+ switch (mt->num_samples) {
+ case 0:
+ case 1:
+ break;
+ case 2:
+ case 4:
+ z_width *= 2;
+ z_height *= 2;
+ break;
+ case 8:
+ z_width *= 4;
+ z_height *= 2;
+ break;
+ default:
+ unreachable("unsupported sample count");
+ }
+
+ const unsigned vertical_align = 8; /* 'j' in the docs */
+ const unsigned H0 = z_height;
+ const unsigned h0 = ALIGN(H0, vertical_align);
+ const unsigned h1 = ALIGN(minify(H0, 1), vertical_align);
+ const unsigned Z0 = z_depth;
+
+ /* HZ_Width (bytes) = ceiling(Z_Width / 16) * 16 */
+ hz_width = ALIGN(z_width, 16);
+
+ unsigned H_i = H0;
+ unsigned Z_i = Z0;
+ unsigned sum_h_i = 0;
+ unsigned hz_height_3d_sum = 0;
+ for (int level = mt->first_level; level <= mt->last_level; ++level) {
+ unsigned i = level - mt->first_level;
+ unsigned h_i = ALIGN(H_i, vertical_align);
+ /* sum(i=2 to m; h_i) */
+ if (i >= 2) {
+ sum_h_i += h_i;
+ }
+ /* sum(i=0 to m; h_i * max(1, floor(Z_Depth/2**i))) */
+ hz_height_3d_sum += h_i * Z_i;
+ H_i = minify(H_i, 1);
+ Z_i = minify(Z_i, 1);
+ }
+ /* HZ_QPitch = h0 + max(h1, sum(i=2 to m; h_i)) */
+ buf->qpitch = h0 + MAX2(h1, sum_h_i);
+
+ if (mt->target == GL_TEXTURE_3D) {
+ /* (1/2) * sum(i=0 to m; h_i * max(1, floor(Z_Depth/2**i))) */
+ hz_height = DIV_ROUND_UP(hz_height_3d_sum, 2);
+ } else {
+ /* HZ_Height (rows) = ceiling( (HZ_QPitch/2)/8) *8 * Z_Depth */
+ hz_height = DIV_ROUND_UP(buf->qpitch, 2 * 8) * 8 * Z0;
+ if (mt->target == GL_TEXTURE_CUBE_MAP_ARRAY ||
+ mt->target == GL_TEXTURE_CUBE_MAP) {
+ /* HZ_Height (rows) = ceiling( (HZ_QPitch/2)/8) *8 * 6 * Z_Depth
+ *
+ * We can can just take our hz_height calculation from above, and
+ * multiply by 6 for the cube map and cube map array types.
+ */
+ hz_height *= 6;
+ }
+ }
+
+ unsigned long pitch;
+ uint32_t tiling = I915_TILING_Y;
+ buf->bo = drm_intel_bo_alloc_tiled(brw->bufmgr, "hiz",
+ hz_width, hz_height, 1,
+ &tiling, &pitch,
+ BO_ALLOC_FOR_RENDER);
+ if (!buf->bo) {
+ free(buf);
+ return NULL;
+ } else if (tiling != I915_TILING_Y) {
+ drm_intel_bo_unreference(buf->bo);
+ free(buf);
+ return NULL;
+ }
+
+ buf->pitch = pitch;
+
+ return buf;
+}
+
+
+static struct intel_miptree_aux_buffer *
+intel_hiz_miptree_buf_create(struct brw_context *brw,
+ struct intel_mipmap_tree *mt)
+{
+ struct intel_miptree_aux_buffer *buf = calloc(sizeof(*buf), 1);
+ const bool force_all_slices_at_each_lod = brw->gen == 6;
+
+ if (!buf)
+ return NULL;
+
+ buf->mt = intel_miptree_create(brw,
+ mt->target,
+ mt->format,
+ mt->first_level,
+ mt->last_level,
+ mt->logical_width0,
+ mt->logical_height0,
+ mt->logical_depth0,
+ true,
+ mt->num_samples,
+ INTEL_MIPTREE_TILING_ANY,
+ force_all_slices_at_each_lod);
+ if (!buf->mt) {
+ free(buf);
+ return NULL;
+ }
+
+ buf->bo = buf->mt->bo;
+ buf->pitch = buf->mt->pitch;
+ buf->qpitch = buf->mt->qpitch;
+
+ return buf;
+}
+
bool
intel_miptree_alloc_hiz(struct brw_context *brw,
struct intel_mipmap_tree *mt)
{
- assert(mt->hiz_mt == NULL);
- const bool force_all_slices_at_each_lod = brw->gen == 6;
- mt->hiz_mt = intel_miptree_create(brw,
- mt->target,
- mt->format,
- mt->first_level,
- mt->last_level,
- mt->logical_width0,
- mt->logical_height0,
- mt->logical_depth0,
- true,
- mt->num_samples,
- INTEL_MIPTREE_TILING_ANY,
- force_all_slices_at_each_lod);
+ assert(mt->hiz_buf == NULL);
+
+ if (brw->gen == 7) {
+ mt->hiz_buf = intel_gen7_hiz_buf_create(brw, mt);
+ } else if (brw->gen >= 8) {
+ mt->hiz_buf = intel_gen8_hiz_buf_create(brw, mt);
+ } else {
+ mt->hiz_buf = intel_hiz_miptree_buf_create(brw, mt);
+ }
- if (!mt->hiz_mt)
+ if (!mt->hiz_buf)
return false;
/* Mark that all slices need a HiZ resolve. */
}
map->stride = map->mt->pitch;
- if (!intel_miptree_blit(brw,
- mt, level, slice,
- map->x, map->y, false,
- map->mt, 0, 0,
- 0, 0, false,
- map->w, map->h, GL_COPY)) {
- fprintf(stderr, "Failed to blit\n");
- goto fail;
+ /* One of either READ_BIT or WRITE_BIT or both is set. READ_BIT implies no
+ * INVALIDATE_RANGE_BIT. WRITE_BIT needs the original values read in unless
+ * invalidate is set, since we'll be writing the whole rectangle from our
+ * temporary buffer back out.
+ */
+ if (!(map->mode & GL_MAP_INVALIDATE_RANGE_BIT)) {
+ if (!intel_miptree_blit(brw,
+ mt, level, slice,
+ map->x, map->y, false,
+ map->mt, 0, 0,
+ 0, 0, false,
+ map->w, map->h, GL_COPY)) {
+ fprintf(stderr, "Failed to blit\n");
+ goto fail;
+ }
}
map->ptr = intel_miptree_map_raw(brw, map->mt);
if (image_x >= 32768 || image_y >= 32768)
return false;
+ /* See intel_miptree_blit() for details on the 32k pitch limit. */
if (mt->pitch >= 32768)
return false;
return true;
}
+static bool
+use_intel_mipree_map_blit(struct brw_context *brw,
+ struct intel_mipmap_tree *mt,
+ GLbitfield mode,
+ unsigned int level,
+ unsigned int slice)
+{
+ if (brw->has_llc &&
+ /* It's probably not worth swapping to the blit ring because of
+ * all the overhead involved.
+ */
+ !(mode & GL_MAP_WRITE_BIT) &&
+ !mt->compressed &&
+ (mt->tiling == I915_TILING_X ||
+ /* Prior to Sandybridge, the blitter can't handle Y tiling */
+ (brw->gen >= 6 && mt->tiling == I915_TILING_Y)) &&
+ can_blit_slice(mt, level, slice))
+ return true;
+
+ if (mt->tiling != I915_TILING_NONE &&
+ mt->bo->size >= brw->max_gtt_map_object_size) {
+ assert(can_blit_slice(mt, level, slice));
+ return true;
+ }
+
+ return false;
+}
+
/**
* Parameter \a out_stride has type ptrdiff_t not because the buffer stride may
* exceed 32 bits but to diminish the likelihood subtle bugs in pointer
intel_miptree_map_etc(brw, mt, map, level, slice);
} else if (mt->stencil_mt && !(mode & BRW_MAP_DIRECT_BIT)) {
intel_miptree_map_depthstencil(brw, mt, map, level, slice);
- }
- /* See intel_miptree_blit() for details on the 32k pitch limit. */
- else if (brw->has_llc &&
- !(mode & GL_MAP_WRITE_BIT) &&
- !mt->compressed &&
- (mt->tiling == I915_TILING_X ||
- (brw->gen >= 6 && mt->tiling == I915_TILING_Y)) &&
- can_blit_slice(mt, level, slice)) {
- intel_miptree_map_blit(brw, mt, map, level, slice);
- } else if (mt->tiling != I915_TILING_NONE &&
- mt->bo->size >= brw->max_gtt_map_object_size) {
- assert(can_blit_slice(mt, level, slice));
+ } else if (use_intel_mipree_map_blit(brw, mt, mode, level, slice)) {
intel_miptree_map_blit(brw, mt, map, level, slice);
#if defined(USE_SSE41)
} else if (!(mode & GL_MAP_WRITE_BIT) && !mt->compressed && cpu_has_sse4_1) {