On Tue, Feb 27, 2018 at 4:33 AM, Pohjolainen, Topi < topi.pohjolai...@gmail.com> wrote:
> On Thu, Feb 22, 2018 at 11:06:45PM -0800, Jason Ekstrand wrote: > > With Yf and Ys tiling, everything is actually four dimensional because > > we can have multiple depth or multisampled array slices in the same > > tile. This commit just enhances the calculations so they can handle it. > > --- > > src/intel/isl/isl.c | 71 ++++++++++++++++++++++++++++++ > +++++++++-------------- > > 1 file changed, 53 insertions(+), 18 deletions(-) > > > > diff --git a/src/intel/isl/isl.c b/src/intel/isl/isl.c > > index 5bdc726..46aff46 100644 > > --- a/src/intel/isl/isl.c > > +++ b/src/intel/isl/isl.c > > @@ -988,7 +988,7 @@ isl_calc_phys_total_extent_el_gen4_2d( > > const struct isl_extent4d *phys_level0_sa, > > enum isl_array_pitch_span array_pitch_span, > > uint32_t *array_pitch_el_rows, > > - struct isl_extent2d *total_extent_el) > > + struct isl_extent4d *phys_total_el) > > { > > const struct isl_format_layout *fmtl = isl_format_get_layout(info-> > format); > > > > @@ -1001,10 +1001,12 @@ isl_calc_phys_total_extent_el_gen4_2d( > > image_align_sa, > phys_level0_sa, > > array_pitch_span, > > &phys_slice0_sa); > > - *total_extent_el = (struct isl_extent2d) { > > + *phys_total_el = (struct isl_extent4d) { > > .w = isl_assert_div(phys_slice0_sa.w, fmtl->bw), > > .h = *array_pitch_el_rows * (phys_level0_sa->array_len - 1) + > > isl_assert_div(phys_slice0_sa.h, fmtl->bh), > > + .d = 1, > > + .a = 1, > > }; > > } > > > > @@ -1019,7 +1021,7 @@ isl_calc_phys_total_extent_el_gen4_3d( > > const struct isl_extent3d *image_align_sa, > > const struct isl_extent4d *phys_level0_sa, > > uint32_t *array_pitch_el_rows, > > - struct isl_extent2d *phys_total_el) > > + struct isl_extent4d *phys_total_el) > > { > > const struct isl_format_layout *fmtl = isl_format_get_layout(info-> > format); > > > > @@ -1066,9 +1068,11 @@ isl_calc_phys_total_extent_el_gen4_3d( > > */ > > *array_pitch_el_rows = > > isl_align_npot(phys_level0_sa->h, image_align_sa->h) / fmtl->bw; > > - *phys_total_el = (struct isl_extent2d) { > > + *phys_total_el = (struct isl_extent4d) { > > .w = isl_assert_div(total_w, fmtl->bw), > > .h = isl_assert_div(total_h, fmtl->bh), > > + .d = 1, > > + .a = 1, > > }; > > } > > > > @@ -1084,7 +1088,7 @@ isl_calc_phys_total_extent_el_gen6_stencil_hiz( > > const struct isl_extent3d *image_align_sa, > > const struct isl_extent4d *phys_level0_sa, > > uint32_t *array_pitch_el_rows, > > - struct isl_extent2d *phys_total_el) > > + struct isl_extent4d *phys_total_el) > > { > > const struct isl_format_layout *fmtl = isl_format_get_layout(info-> > format); > > > > @@ -1127,9 +1131,11 @@ isl_calc_phys_total_extent_el_gen6_stencil_hiz( > > > > *array_pitch_el_rows = > > isl_assert_div(isl_align(H0, image_align_sa->h), fmtl->bh); > > - *phys_total_el = (struct isl_extent2d) { > > + *phys_total_el = (struct isl_extent4d) { > > .w = isl_assert_div(MAX(total_top_w, total_bottom_w), fmtl->bw), > > .h = isl_assert_div(total_h, fmtl->bh), > > + .d = 1, > > + .a = 1, > > }; > > } > > > > @@ -1144,7 +1150,7 @@ isl_calc_phys_total_extent_el_gen9_1d( > > const struct isl_extent3d *image_align_sa, > > const struct isl_extent4d *phys_level0_sa, > > uint32_t *array_pitch_el_rows, > > - struct isl_extent2d *phys_total_el) > > + struct isl_extent4d *phys_total_el) > > { > > MAYBE_UNUSED const struct isl_format_layout *fmtl = > isl_format_get_layout(info->format); > > > > @@ -1164,9 +1170,11 @@ isl_calc_phys_total_extent_el_gen9_1d( > > } > > > > *array_pitch_el_rows = 1; > > - *phys_total_el = (struct isl_extent2d) { > > + *phys_total_el = (struct isl_extent4d) { > > .w = isl_assert_div(slice_w, fmtl->bw), > > .h = phys_level0_sa->array_len, > > + .d = 1, > > + .a = 1, > > }; > > } > > > > @@ -1184,7 +1192,7 @@ isl_calc_phys_total_extent_el(const struct > isl_device *dev, > > const struct isl_extent4d *phys_level0_sa, > > enum isl_array_pitch_span > array_pitch_span, > > uint32_t *array_pitch_el_rows, > > - struct isl_extent2d *total_extent_el) > > + struct isl_extent4d *phys_total_el) > > { > > switch (dim_layout) { > > case ISL_DIM_LAYOUT_GEN9_1D: > > @@ -1192,14 +1200,14 @@ isl_calc_phys_total_extent_el(const struct > isl_device *dev, > > isl_calc_phys_total_extent_el_gen9_1d(dev, info, > > image_align_sa, > phys_level0_sa, > > array_pitch_el_rows, > > - total_extent_el); > > + phys_total_el); > > return; > > case ISL_DIM_LAYOUT_GEN4_2D: > > isl_calc_phys_total_extent_el_gen4_2d(dev, info, tile_info, > msaa_layout, > > image_align_sa, > phys_level0_sa, > > array_pitch_span, > > array_pitch_el_rows, > > - total_extent_el); > > + phys_total_el); > > return; > > case ISL_DIM_LAYOUT_GEN6_STENCIL_HIZ: > > assert(array_pitch_span == ISL_ARRAY_PITCH_SPAN_COMPACT); > > @@ -1207,14 +1215,14 @@ isl_calc_phys_total_extent_el(const struct > isl_device *dev, > > image_align_sa, > > phys_level0_sa, > > > array_pitch_el_rows, > > - total_extent_el); > > + phys_total_el); > > return; > > case ISL_DIM_LAYOUT_GEN4_3D: > > assert(array_pitch_span == ISL_ARRAY_PITCH_SPAN_COMPACT); > > isl_calc_phys_total_extent_el_gen4_3d(dev, info, > > image_align_sa, > phys_level0_sa, > > array_pitch_el_rows, > > - total_extent_el); > > + phys_total_el); > > return; > > } > > } > > @@ -1257,7 +1265,7 @@ isl_calc_row_pitch_alignment(const struct > isl_surf_init_info *surf_info, > > static uint32_t > > isl_calc_linear_min_row_pitch(const struct isl_device *dev, > > const struct isl_surf_init_info *info, > > - const struct isl_extent2d *phys_total_el, > > + const struct isl_extent4d *phys_total_el, > > uint32_t alignment) > > { > > const struct isl_format_layout *fmtl = isl_format_get_layout(info-> > format); > > @@ -1270,7 +1278,7 @@ static uint32_t > > isl_calc_tiled_min_row_pitch(const struct isl_device *dev, > > const struct isl_surf_init_info *surf_info, > > const struct isl_tile_info *tile_info, > > - const struct isl_extent2d *phys_total_el, > > + const struct isl_extent4d *phys_total_el, > > uint32_t alignment) > > { > > const struct isl_format_layout *fmtl = isl_format_get_layout(surf_ > info->format); > > @@ -1290,7 +1298,7 @@ static uint32_t > > isl_calc_min_row_pitch(const struct isl_device *dev, > > const struct isl_surf_init_info *surf_info, > > const struct isl_tile_info *tile_info, > > - const struct isl_extent2d *phys_total_el, > > + const struct isl_extent4d *phys_total_el, > > uint32_t alignment) > > { > > if (tile_info->tiling == ISL_TILING_LINEAR) { > > @@ -1323,7 +1331,7 @@ isl_calc_row_pitch(const struct isl_device *dev, > > const struct isl_surf_init_info *surf_info, > > const struct isl_tile_info *tile_info, > > enum isl_dim_layout dim_layout, > > - const struct isl_extent2d *phys_total_el, > > + const struct isl_extent4d *phys_total_el, > > uint32_t *out_row_pitch) > > { > > uint32_t alignment = > > @@ -1446,7 +1454,7 @@ isl_surf_init_s(const struct isl_device *dev, > > isl_choose_array_pitch_span(dev, info, dim_layout, > &phys_level0_sa); > > > > uint32_t array_pitch_el_rows; > > - struct isl_extent2d phys_total_el; > > + struct isl_extent4d phys_total_el; > > isl_calc_phys_total_extent_el(dev, info, &tile_info, > > dim_layout, msaa_layout, > > &image_align_sa, &phys_level0_sa, > > @@ -1461,6 +1469,9 @@ isl_surf_init_s(const struct isl_device *dev, > > uint32_t base_alignment; > > uint64_t size; > > if (tiling == ISL_TILING_LINEAR) { > > + /* LINEAR tiling has no concept of intra-tile arrays */ > > + assert(phys_total_el.d == 1 && phys_total_el.a == 1); > > + > > size = (uint64_t) row_pitch * phys_total_el.h; > > > > /* From the Broadwell PRM Vol 2d, RENDER_SURFACE_STATE:: > SurfaceBaseAddress: > > @@ -1482,7 +1493,31 @@ isl_surf_init_s(const struct isl_device *dev, > > } > > base_alignment = isl_round_up_to_power_of_two(base_alignment); > > } else { > > + /* Pitches must make sense with the tiling */ > > + assert(row_pitch % tile_info.phys_extent_B.width == 0); > > + assert(array_pitch_el_rows % tile_info.logical_extent_el.d == 0); > > + assert(array_pitch_el_rows % tile_info.logical_extent_el.a == 0); > > + > > + uint32_t array_slices; > > + if (phys_total_el.d > 1) { > > + assert(phys_total_el.a == 1); > > + array_slices = isl_align_div(phys_total_el.d, > > + tile_info.logical_extent_el.d); > > + } else if (phys_total_el.a > 1) { > > + assert(phys_total_el.d == 1); > > + array_slices = isl_align_div(phys_total_el.a, > > + tile_info.logical_extent_el.a); > > + } else { > > + assert(phys_total_el.d == 1 && phys_total_el.a == 1); > > + array_slices = 1; > > + } > > + > > + uint32_t array_pitch_tl_rows = > > + array_pitch_el_rows / MAX2(tile_info.logical_extent_el.d, > > + tile_info.logical_extent_el.a); > > + > > const uint32_t total_h_tl = > > + (array_slices - 1) * array_pitch_tl_rows + > > I probably need to read further to understand the subtraction of one, and > why > we use "array_slices == 1" for the case where "phys_total_el.d == 1 && > phys_total_el.a == 1". It's because the array stride is used to get us to the last slice and the bit below with phys_total_el.h is used to get us to the toal height. This lets us avoid aligning the total height to the array stride in the case where it's overly large. > Patches 1-4 are: > > Reviewed-by: Topi Pohjolainen <topi.pohjolai...@intel.com> > Thanks! > > isl_align_div(phys_total_el.h, tile_info.logical_extent_el. > height); > > > > size = (uint64_t) total_h_tl * tile_info.phys_extent_B.height * > row_pitch; > > -- > > 2.5.0.400.gff86faf > > > > _______________________________________________ > > mesa-dev mailing list > > mesa-dev@lists.freedesktop.org > > https://lists.freedesktop.org/mailman/listinfo/mesa-dev >
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev