<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Wed, May 3, 2017 at 2:22 AM, Topi Pohjolainen <span dir="ltr"><<a href="mailto:topi.pohjolainen@gmail.com" target="_blank">topi.pohjolainen@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">leaving y-tiled (r8stencil) copies still as they were.<br>
<br>
Signed-off-by: Topi Pohjolainen <<a href="mailto:topi.pohjolainen@intel.com">topi.pohjolainen@intel.com</a>><br>
---<br>
src/mesa/drivers/dri/i965/brw_<wbr>blorp.c | 24 +++++++----<br>
src/mesa/drivers/dri/i965/brw_<wbr>tex_layout.c | 57 ---------------------------<br>
src/mesa/drivers/dri/i965/<wbr>gen6_depth_state.c | 31 +++++++--------<br>
src/mesa/drivers/dri/i965/<wbr>intel_mipmap_tree.c | 45 ++++++++++++++-------<br>
4 files changed, 63 insertions(+), 94 deletions(-)<br>
<br>
diff --git a/src/mesa/drivers/dri/i965/<wbr>brw_blorp.c b/src/mesa/drivers/dri/i965/<wbr>brw_blorp.c<br>
index 158cf66..7b6c13d 100644<br>
--- a/src/mesa/drivers/dri/i965/<wbr>brw_blorp.c<br>
+++ b/src/mesa/drivers/dri/i965/<wbr>brw_blorp.c<br>
@@ -146,8 +146,7 @@ blorp_surf_for_miptree(struct brw_context *brw,<br>
.write_domain = is_render_target ? I915_GEM_DOMAIN_RENDER : 0,<br>
};<br>
<br>
- if (brw->gen == 6 && mt->format == MESA_FORMAT_S_UINT8 &&<br>
- mt->array_layout == ALL_SLICES_AT_EACH_LOD) {<br>
+ if (brw->gen == 6 && mt->format == MESA_FORMAT_S_UINT8) {<br>
/* Sandy bridge stencil and HiZ use this ALL_SLICES_AT_EACH_LOD hack in<br>
* order to allow for layered rendering. The hack makes each LOD of the<br>
* stencil or HiZ buffer a single tightly packed array surface at some<br>
@@ -166,12 +165,23 @@ blorp_surf_for_miptree(struct brw_context *brw,<br>
* consulted. Otherwise surf is ignored and there is no need to adjust<br>
* it any further. See blorp_emit_depth_stencil_<wbr>config().<br>
*/<br>
- surf->addr.offset += brw_stencil_all_slices_at_<wbr>each_lod_offset(<br>
- surf->surf, *level);<br>
+ uint32_t x_offset_sa, y_offset_sa;<br>
+ get_image_offset_sa_gen6_back_<wbr>to_back(<br>
+ &mt->surf, *level, 0, 0, &x_offset_sa, &y_offset_sa);<br>
+ assert(x_offset_sa == 0);<br>
+<br>
+ /* The stencil buffer has quirky pitch requirements. From Vol 2a,<br>
+ * 11.5.6.2.1 3DSTATE_STENCIL_BUFFER, field "Surface Pitch":<br>
+ * The pitch must be set to 2x the value computed based on width, as<br>
+ * the stencil buffer is stored with two rows interleaved.<br>
+ *<br>
+ * As ISL uses twice the width, byte offset calculation needs to use<br>
+ * half of that.<br>
+ */<br>
+ const uint32_t offset = y_offset_sa * mt->surf.row_pitch / 2;<br></blockquote><div><br></div>If you used isl_tiling_get_intratile_offset_sa, it would take care of this for you because it does a trip through logical tile sizes.<br></div><div class="gmail_quote"><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
+ assert(offset % 4096 == 0);<br>
<br>
- assert(brw_stencil_all_slices_<wbr>at_each_lod_offset(surf->surf, *level) ==<br>
- mt->level[*level].level_y * mt->pitch +<br>
- mt->level[*level].level_x * 64);<br>
+ surf->addr.offset += offset;<br>
<br>
*level = 0;<br>
}<br>
diff --git a/src/mesa/drivers/dri/i965/<wbr>brw_tex_layout.c b/src/mesa/drivers/dri/i965/<wbr>brw_tex_layout.c<br>
index deac692..dec04d7 100644<br>
--- a/src/mesa/drivers/dri/i965/<wbr>brw_tex_layout.c<br>
+++ b/src/mesa/drivers/dri/i965/<wbr>brw_tex_layout.c<br>
@@ -163,63 +163,6 @@ gen9_miptree_layout_1d(struct intel_mipmap_tree *mt)<br>
}<br>
}<br>
<br>
-static unsigned<br>
-all_slices_at_each_lod_x_<wbr>offset(unsigned w0, unsigned align, unsigned level)<br>
-{<br>
- const unsigned w = level >= 2 ? minify(w0, 1) : 0;<br>
- return ALIGN(w, align);<br>
-}<br>
-<br>
-static unsigned<br>
-all_slices_at_each_lod_y_<wbr>offset(const struct isl_extent4d *phys_level0_sa,<br>
- enum isl_surf_dim dim, unsigned align,<br>
- unsigned level)<br>
-{<br>
- unsigned y = 0;<br>
-<br>
- /* Add vertical space taken by lower levels one by one. Levels one and two<br>
- * are side-by-side just below level zero. Levels three and greater are<br>
- * stacked one after another below level two.<br>
- */<br>
- for (unsigned i = 1; i <= level; ++i) {<br>
- const unsigned d = dim == ISL_SURF_DIM_3D ?<br>
- minify(phys_level0_sa->depth, i - 1) :<br>
- phys_level0_sa->array_len;<br>
-<br>
- /* Levels two and greater are stacked just below level zero. */<br>
- if (i != 2) {<br>
- const unsigned h = minify(phys_level0_sa->height, i - 1);<br>
- y += d * ALIGN(h, align);<br>
- }<br>
- }<br>
-<br>
- return y;<br>
-}<br>
-<br>
-uint32_t<br>
-brw_stencil_all_slices_at_<wbr>each_lod_offset(const struct isl_surf *surf,<br>
- unsigned level)<br>
-{<br>
- const unsigned halign = 64;<br>
- const unsigned valign = 64;<br>
- const unsigned level_x = all_slices_at_each_lod_x_<wbr>offset(<br>
- surf->phys_level0_sa.width, halign, level);<br>
- const unsigned level_y = all_slices_at_each_lod_y_<wbr>offset(<br>
- &surf->phys_level0_sa, surf->dim, valign, level);<br>
-<br>
- /* From Vol 2a, 11.5.6.2.1 3DSTATE_STENCIL_BUFFER, field "Surface Pitch":<br>
- * The pitch must be set to 2x the value computed based on width, as<br>
- * the stencil buffer is stored with two rows interleaved.<br>
- *<br>
- * While ISL surface stores the pitch expected by hardware, the offset<br>
- * into individual slices needs to be calculated as if rows are<br>
- * interleaved.<br>
- */<br>
- const unsigned two_rows_interleaved_pitch = surf->row_pitch / 2;<br>
-<br>
- return level_y * two_rows_interleaved_pitch + level_x * 64;<br>
-}<br>
-<br>
uint32_t<br>
brw_get_mipmap_total_width(<wbr>unsigned w0, unsigned num_levels, unsigned halign)<br>
{<br>
diff --git a/src/mesa/drivers/dri/i965/<wbr>gen6_depth_state.c b/src/mesa/drivers/dri/i965/<wbr>gen6_depth_state.c<br>
index 16fb209..1850934 100644<br>
--- a/src/mesa/drivers/dri/i965/<wbr>gen6_depth_state.c<br>
+++ b/src/mesa/drivers/dri/i965/<wbr>gen6_depth_state.c<br>
@@ -189,29 +189,28 @@ gen6_emit_depth_stencil_hiz(<wbr>struct brw_context *brw,<br>
<br>
/* Emit stencil buffer. */<br>
if (separate_stencil) {<br>
- uint32_t offset = 0;<br>
+ assert(stencil_mt->format == MESA_FORMAT_S_UINT8);<br>
<br>
- if (stencil_mt->array_layout == ALL_SLICES_AT_EACH_LOD) {<br>
- assert(stencil_mt->format == MESA_FORMAT_S_UINT8);<br>
-<br>
- struct isl_surf temp_surf;<br>
- intel_miptree_get_isl_surf(<wbr>brw, stencil_mt, &temp_surf);<br>
-<br>
- offset = brw_stencil_all_slices_at_<wbr>each_lod_offset(<br>
- &temp_surf, lod);<br>
- assert(offset ==<br>
- stencil_mt->level[lod].level_y * stencil_mt->pitch +<br>
- stencil_mt->level[lod].level_x * 64);<br>
- }<br>
+ uint32_t x_offset_sa, y_offset_sa;<br>
+ get_image_offset_sa_gen6_back_<wbr>to_back(&stencil_mt->surf,<br>
+ lod, 0, 0,<br>
+ &x_offset_sa, &y_offset_sa);<br></blockquote><div><br></div><div>Why are you calling this directly and not just using isl_surf_get_image_offset_sa?<br></div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
+ assert(x_offset_sa == 0);<br>
<br>
- BEGIN_BATCH(3);<br>
- OUT_BATCH((_3DSTATE_STENCIL_<wbr>BUFFER << 16) | (3 - 2));<br>
/* The stencil buffer has quirky pitch requirements. From Vol 2a,<br>
* 11.5.6.2.1 3DSTATE_STENCIL_BUFFER, field "Surface Pitch":<br>
* The pitch must be set to 2x the value computed based on width, as<br>
* the stencil buffer is stored with two rows interleaved.<br>
+ *<br>
+ * As ISL uses twice the width, byte offset calculation needs to use<br>
+ * half of that.<br>
*/<br>
- OUT_BATCH(2 * stencil_mt->pitch - 1);<br>
+ const uint32_t offset = y_offset_sa * stencil_mt->surf.row_pitch / 2;<br></blockquote><div><br></div><div>Again, use isl_tiling_get_intratile_offset_sa<br></div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
+ assert(offset % 4096 == 0);<br>
+<br>
+ BEGIN_BATCH(3);<br>
+ OUT_BATCH((_3DSTATE_STENCIL_<wbr>BUFFER << 16) | (3 - 2));<br>
+ OUT_BATCH(stencil_mt->surf.<wbr>row_pitch - 1);<br>
OUT_RELOC(stencil_mt->bo,<br>
I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,<br>
offset);<br>
diff --git a/src/mesa/drivers/dri/i965/<wbr>intel_mipmap_tree.c b/src/mesa/drivers/dri/i965/<wbr>intel_mipmap_tree.c<br>
index ced1e0e..febe880 100644<br>
--- a/src/mesa/drivers/dri/i965/<wbr>intel_mipmap_tree.c<br>
+++ b/src/mesa/drivers/dri/i965/<wbr>intel_mipmap_tree.c<br>
@@ -678,6 +678,14 @@ miptree_create(struct brw_context *brw,<br>
GLuint num_samples,<br>
uint32_t layout_flags)<br>
{<br>
+ if (format == MESA_FORMAT_S_UINT8)<br>
+ return make_surface(brw, target, format, first_level, last_level,<br>
+ width0, height0, depth0, num_samples,<br>
+ ISL_TILING_W,<br>
+ ISL_SURF_USAGE_STENCIL_BIT |<br>
+ ISL_SURF_USAGE_TEXTURE_BIT,<br>
+ BO_ALLOC_FOR_RENDER);<br>
+<br>
struct intel_mipmap_tree *mt;<br>
mesa_format tex_format = format;<br>
mesa_format etc_format = MESA_FORMAT_NONE;<br>
@@ -2275,28 +2283,35 @@ intel_update_r8stencil(struct brw_context *brw,<br>
!src->r8stencil_needs_update)<br>
return;<br>
<br>
+ assert(src->surf.size > 0);<br>
+<br>
if (!mt->r8stencil_mt) {<br>
const uint32_t r8stencil_flags =<br>
MIPTREE_LAYOUT_ACCELERATED_<wbr>UPLOAD | MIPTREE_LAYOUT_TILING_Y |<br>
MIPTREE_LAYOUT_DISABLE_AUX;<br>
assert(brw->gen > 6); /* Handle MIPTREE_LAYOUT_FORCE_ALL_<wbr>SLICE_AT_LOD */<br>
- mt->r8stencil_mt = intel_miptree_create(brw,<br>
- src->target,<br>
- MESA_FORMAT_R_UINT8,<br>
- src->first_level,<br>
- src->last_level,<br>
- src->logical_width0,<br>
- src->logical_height0,<br>
- src->logical_depth0,<br>
- src->num_samples,<br>
- r8stencil_flags);<br>
+ mt->r8stencil_mt = intel_miptree_create(<br>
+ brw,<br>
+ src->target,<br>
+ MESA_FORMAT_R_UINT8,<br>
+ 0, src->surf.levels - 1,<br>
+ src->surf.logical_level0_px.<wbr>width,<br>
+ src->surf.logical_level0_px.<wbr>height,<br>
+ src->surf.dim_layout == ISL_DIM_LAYOUT_GEN4_3D ?<br>
+ src->surf.logical_level0_px.<wbr>depth :<br>
+ src->surf.logical_level0_px.<wbr>array_len,<br>
+ src->surf.samples,<br>
+ r8stencil_flags);<br>
assert(mt->r8stencil_mt);<br>
}<br>
<br>
struct intel_mipmap_tree *dst = mt->r8stencil_mt;<br>
<br>
- for (int level = src->first_level; level <= src->last_level; level++) {<br>
- const unsigned depth = src->level[level].depth;<br>
+ for (int level = 0; level < src->surf.levels; level++) {<br>
+ const unsigned depth = src->surf.dim_layout == ISL_DIM_LAYOUT_GEN4_3D ?<br></blockquote><div><br></div><div>ISL_SURF_DIM_3D<br></div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
+ minify(src->surf.logical_<wbr>level0_px.depth, level) :<br>
+ src->surf.logical_level0_px.<wbr>array_len;<br>
+<br>
const int layers_per_blit =<br>
(dst->msaa_layout == INTEL_MSAA_LAYOUT_UMS ||<br>
dst->msaa_layout == INTEL_MSAA_LAYOUT_CMS) ?<br>
@@ -2309,8 +2324,10 @@ intel_update_r8stencil(struct brw_context *brw,<br>
dst, level, layers_per_blit * layer,<br>
MESA_FORMAT_R_UNORM8,<br>
0, 0,<br>
- minify(src->logical_width0, level),<br>
- minify(src->logical_height0, level),<br>
+ minify(src->surf.logical_<wbr>level0_px.width,<br>
+ level),<br>
+ minify(src->surf.logical_<wbr>level0_px.height,<br>
+ level),<br>
0, 0,<br>
minify(dst->logical_width0, level),<br>
minify(dst->logical_height0, level),<br>
<span class="HOEnZb"><font color="#888888">--<br>
2.9.3<br>
<br>
______________________________<wbr>_________________<br>
mesa-dev mailing list<br>
<a href="mailto:mesa-dev@lists.freedesktop.org">mesa-dev@lists.freedesktop.org</a><br>
<a href="https://lists.freedesktop.org/mailman/listinfo/mesa-dev" rel="noreferrer" target="_blank">https://lists.freedesktop.org/<wbr>mailman/listinfo/mesa-dev</a><br>
</font></span></blockquote></div><br></div></div>