[Mesa-dev] [PATCH 6/8] i965: Implement HiZ resolves on Broadwell.

Chris Forbes chrisf at ijw.co.nz
Mon Feb 10 21:56:59 PST 2014


Since we don't have to do nonzero miplevels for multisampling, can't
you just use mt->logical_width0 and mt->logical_height0 if
multisampling?

On Tue, Feb 11, 2014 at 6:48 PM, Kenneth Graunke <kenneth at whitecape.org> wrote:
> On 02/07/2014 10:43 PM, Kenneth Graunke wrote:
>> Broadwell's 3DSTATE_WM_HZ_OP packet makes this much easier.
>>
>> Instead of programming the whole pipeline, we simply have to emit the
>> depth/stencil packets, a state override, and a pipe control.  Then
>> arrange for the state to be put back.  This is easily done from a single
>> function.
>>
>> Signed-off-by: Kenneth Graunke <kenneth at whitecape.org>
>> ---
>>  src/mesa/drivers/dri/i965/brw_blorp.cpp      |   8 +-
>>  src/mesa/drivers/dri/i965/brw_context.h      |   4 +
>>  src/mesa/drivers/dri/i965/gen8_depth_state.c | 105 +++++++++++++++++++++++++++
>>  3 files changed, 115 insertions(+), 2 deletions(-)
>>
>> diff --git a/src/mesa/drivers/dri/i965/brw_blorp.cpp b/src/mesa/drivers/dri/i965/brw_blorp.cpp
>> index 76537c8..4a5001d 100644
>> --- a/src/mesa/drivers/dri/i965/brw_blorp.cpp
>> +++ b/src/mesa/drivers/dri/i965/brw_blorp.cpp
>> @@ -195,8 +195,12 @@ intel_hiz_exec(struct brw_context *brw, struct intel_mipmap_tree *mt,
>>     DBG("%s %s to mt %p level %d layer %d\n",
>>         __FUNCTION__, opname, mt, level, layer);
>>
>> -   brw_hiz_op_params params(mt, level, layer, op);
>> -   brw_blorp_exec(brw, &params);
>> +   if (brw->gen >= 8) {
>> +      gen8_hiz_exec(brw, mt, level, layer, op);
>> +   } else {
>> +      brw_hiz_op_params params(mt, level, layer, op);
>> +      brw_blorp_exec(brw, &params);
>> +   }
>>  }
>>
>>  } /* extern "C" */
>> diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h
>> index cf0fe98..c107683 100644
>> --- a/src/mesa/drivers/dri/i965/brw_context.h
>> +++ b/src/mesa/drivers/dri/i965/brw_context.h
>> @@ -61,6 +61,7 @@ extern "C" {
>>  #include "intel_debug.h"
>>  #include "intel_screen.h"
>>  #include "intel_tex_obj.h"
>> +#include "intel_resolve_map.h"
>>
>>  /* Glossary:
>>   *
>> @@ -1891,6 +1892,9 @@ gen8_emit_depth_stencil_hiz(struct brw_context *brw,
>>                              uint32_t width, uint32_t height,
>>                              uint32_t tile_x, uint32_t tile_y);
>>
>> +void gen8_hiz_exec(struct brw_context *brw, struct intel_mipmap_tree *mt,
>> +                   unsigned int level, unsigned int layer, enum gen6_hiz_op op);
>> +
>>  extern const GLuint prim_to_hw_prim[GL_TRIANGLE_STRIP_ADJACENCY+1];
>>
>>  void
>> diff --git a/src/mesa/drivers/dri/i965/gen8_depth_state.c b/src/mesa/drivers/dri/i965/gen8_depth_state.c
>> index f30ff28..3fa20c8 100644
>> --- a/src/mesa/drivers/dri/i965/gen8_depth_state.c
>> +++ b/src/mesa/drivers/dri/i965/gen8_depth_state.c
>> @@ -203,3 +203,108 @@ gen8_emit_depth_stencil_hiz(struct brw_context *brw,
>>                        brw->depthstencil.stencil_offset,
>>                        hiz, width, height, depth, lod, min_array_element);
>>  }
>> +
>> +/**
>> + * Emit packets to perform a depth/HiZ resolve or fast depth/stencil clear.
>> + *
>> + * See the "Optimized Depth Buffer Clear and/or Stencil Buffer Clear" section
>> + * of the hardware documentation for details.
>> + */
>> +void
>> +gen8_hiz_exec(struct brw_context *brw, struct intel_mipmap_tree *mt,
>> +              unsigned int level, unsigned int layer, enum gen6_hiz_op op)
>> +{
>> +   if (op == GEN6_HIZ_OP_NONE)
>> +      return;
>> +
>> +   assert(mt->first_level == 0);
>> +
>> +   struct intel_mipmap_level *miplevel = &mt->level[level];
>> +
>> +   /* The basic algorithm is:
>> +    * - If needed, emit 3DSTATE_{DEPTH,HIER_DEPTH,STENCIL}_BUFFER and
>> +    *   3DSTATE_CLEAR_PARAMS packets to set up the relevant buffers.
>> +    * - If needed, emit 3DSTATE_DRAWING_RECTANGLE.
>> +    * - Emit 3DSTATE_WM_HZ_OP with a bit set for the particular operation.
>> +    * - Do a special PIPE_CONTROL to trigger an implicit rectangle primitive.
>> +    * - Emit 3DSTATE_WM_HZ_OP with no bits set to return to normal rendering.
>> +    */
>> +   emit_depth_packets(brw, mt,
>> +                      brw_depth_format(brw, mt->format),
>> +                      BRW_SURFACE_2D,
>> +                      true, /* depth writes */
>> +                      NULL, false, 0, /* no stencil for now */
>> +                      true, /* hiz */
>> +                      mt->logical_width0,
>> +                      mt->logical_height0,
>> +                      MAX2(mt->logical_depth0, 1),
>> +                      level,
>> +                      layer); /* min_array_element */
>> +
>> +   BEGIN_BATCH(4);
>> +   OUT_BATCH(_3DSTATE_DRAWING_RECTANGLE << 16 | (4 - 2));
>> +   OUT_BATCH(0);
>> +   OUT_BATCH(((mt->logical_width0 - 1) & 0xffff) |
>> +             ((mt->logical_height0 - 1) << 16));
>> +   OUT_BATCH(0);
>> +   ADVANCE_BATCH();
>> +
>> +   /* Emit 3DSTATE_WM_HZ_OP to override pipeline state for the particular
>> +    * resolve or clear operation we want to perform.
>> +    */
>> +   uint32_t dw1 = 0;
>> +
>> +   switch (op) {
>> +   case GEN6_HIZ_OP_DEPTH_RESOLVE:
>> +      dw1 |= GEN8_WM_HZ_DEPTH_RESOLVE;
>> +      break;
>> +   case GEN6_HIZ_OP_HIZ_RESOLVE:
>> +      dw1 |= GEN8_WM_HZ_HIZ_RESOLVE;
>> +      break;
>> +   case GEN6_HIZ_OP_DEPTH_CLEAR:
>> +      dw1 |= GEN8_WM_HZ_DEPTH_CLEAR;
>> +      break;
>> +   case GEN6_HIZ_OP_NONE:
>> +      assert(!"Should not get here.");
>> +   }
>> +
>> +   uint32_t sample_mask = 0xFFFF;
>> +   if (mt->num_samples > 0) {
>> +      dw1 |= SET_FIELD(ffs(mt->num_samples) - 1, GEN8_WM_HZ_NUM_SAMPLES);
>> +      sample_mask = gen6_determine_sample_mask(brw);
>> +   }
>> +
>> +   BEGIN_BATCH(5);
>> +   OUT_BATCH(_3DSTATE_WM_HZ_OP << 16 | (5 - 2));
>> +   OUT_BATCH(dw1);
>> +   OUT_BATCH(0);
>> +   OUT_BATCH(SET_FIELD(miplevel->width, GEN8_WM_HZ_CLEAR_RECTANGLE_X_MAX) |
>> +             SET_FIELD(miplevel->height, GEN8_WM_HZ_CLEAR_RECTANGLE_Y_MAX));
>
> I'm concerned about a possible interaction with multisampling:
>
> miplevel->width/height are based on the physical dimensions, rather than
> the logical ones.  For example, a (logical) 256x256 texture may have a
> level 0 miplevel->width of 512 (4x MSAA) or 1024 (8x MSAA).
>
> I'm at a bit of a loss for what to do instead, though.
>
>> +   OUT_BATCH(SET_FIELD(sample_mask, GEN8_WM_HZ_SAMPLE_MASK));
>> +   ADVANCE_BATCH();
>> +
>> +   /* Emit a PIPE_CONTROL with "Post-Sync Operation" set to "Write Immediate
>> +    * Data", and no other bits set.  This causes 3DSTATE_WM_HZ_OP's state to
>> +    * take effect, and spawns a rectangle primitive.
>> +    */
>> +   brw_emit_pipe_control_write(brw,
>> +                               PIPE_CONTROL_WRITE_IMMEDIATE,
>> +                               brw->batch.workaround_bo, 0, 0, 0);
>> +
>> +   /* Emit 3DSTATE_WM_HZ_OP again to disable the state overrides. */
>> +   BEGIN_BATCH(5);
>> +   OUT_BATCH(_3DSTATE_WM_HZ_OP << 16 | (5 - 2));
>> +   OUT_BATCH(0);
>> +   OUT_BATCH(0);
>> +   OUT_BATCH(0);
>> +   OUT_BATCH(0);
>> +   ADVANCE_BATCH();
>> +
>> +   /* We've clobbered all of the depth packets, and the drawing rectangle,
>> +    * so we need to ensure those packets are re-emitted before the next
>> +    * primitive.
>> +    *
>> +    * Setting _NEW_DEPTH and _NEW_BUFFERS covers it, but is rather overkill.
>> +    */
>> +   brw->state.dirty.mesa |= _NEW_DEPTH | _NEW_BUFFERS;
>> +}
>>
>
>
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
>


More information about the mesa-dev mailing list