r300g: fix some bugs with zbuffer compression (v4)

This drops the memblock manager for ZMASK. Instead, only one zbuffer can be
compressed at a time. Note that this does not necessarily have to be slower.
When there is a large number of zbuffers, compression might be used more often
than it was before. It's also easier to debug.

How it works:
1) 'clear' turns the compression on.
2) If some other zbuffer is set or the currently-bound zbuffer is used
   for texturing, the driver decompresses it and then turns the compression off.

Notes:
- The ZMASK clear has been refactored, so that only one packet3 is used to clear
  ZMASK.
- The 8x8 compression mode is disabled. I couldn't make it work without issues.
- Also removed driver-specific stuff from u_blitter.

Driver status:
- RV530 and R580 appear to just work (finally).
- RV570 should work, but there may be an issue that we don't correctly
  calculate the number of dwords to clear, resulting in a partially
  uninitialized zbuffer.
- RS690 misrenders as if no ZMASK clear happened. No idea what's going on.
- RV350 may even hardlock. This issue was already present and this patch doesn't
  fix it.

I think we are still missing some hardware info we need to make the zbuffer
compression work fully.

Note that there is also an issue with HiZ, resulting in a sort of blocky
zigzagged corruption around some objects.

3 files changed, 65 insertions, 40 deletions

diff --git a/src/gallium/auxiliary/util/u_blitter.c b/src/gallium/auxiliary/util/u_blitter.c
index e72f267f53d..1f9d50932e3 100644
--- a/src/gallium/auxiliary/util/u_blitter.c
+++ b/src/gallium/auxiliary/util/u_blitter.c
@@ -86,7 +86,6 @@ struct blitter_context_priv
    void *dsa_write_depth_keep_stencil;
    void *dsa_keep_depth_stencil;
    void *dsa_keep_depth_write_stencil;
-   void *dsa_flush_depth_stencil;
 
    void *velem_state;
 
@@ -158,10 +157,6 @@ struct blitter_context *util_blitter_create(struct pipe_context *pipe)
    ctx->dsa_keep_depth_stencil =
       pipe->create_depth_stencil_alpha_state(pipe, &dsa);
 
-   dsa.depth.writemask = 1;
-   ctx->dsa_flush_depth_stencil =
-      pipe->create_depth_stencil_alpha_state(pipe, &dsa);
-
    dsa.depth.enabled = 1;
    dsa.depth.writemask = 1;
    dsa.depth.func = PIPE_FUNC_ALWAYS;
@@ -247,7 +242,6 @@ void util_blitter_destroy(struct blitter_context *blitter)
                                           ctx->dsa_write_depth_keep_stencil);
    pipe->delete_depth_stencil_alpha_state(pipe, ctx->dsa_write_depth_stencil);
    pipe->delete_depth_stencil_alpha_state(pipe, ctx->dsa_keep_depth_write_stencil);
-   pipe->delete_depth_stencil_alpha_state(pipe, ctx->dsa_flush_depth_stencil);
 
    pipe->delete_rasterizer_state(pipe, ctx->rs_state);
    pipe->delete_vs_state(pipe, ctx->vs);
@@ -665,12 +659,13 @@ static void blitter_draw_rectangle(struct blitter_context *blitter,
    blitter_draw_quad(ctx);
 }
 
-void util_blitter_clear(struct blitter_context *blitter,
-                        unsigned width, unsigned height,
-                        unsigned num_cbufs,
-                        unsigned clear_buffers,
-                        const float *rgba,
-                        double depth, unsigned stencil)
+static void util_blitter_clear_custom(struct blitter_context *blitter,
+                                      unsigned width, unsigned height,
+                                      unsigned num_cbufs,
+                                      unsigned clear_buffers,
+                                      const float *rgba,
+                                      double depth, unsigned stencil,
+                                      void *custom_blend, void *custom_dsa)
 {
    struct blitter_context_priv *ctx = (struct blitter_context_priv*)blitter;
    struct pipe_context *pipe = ctx->base.pipe;
@@ -681,26 +676,28 @@ void util_blitter_clear(struct blitter_context *blitter,
    blitter_check_saved_CSOs(ctx);
 
    /* bind CSOs */
-   if (clear_buffers & PIPE_CLEAR_COLOR)
+   if (custom_blend) {
+      pipe->bind_blend_state(pipe, custom_blend);
+   } else if (clear_buffers & PIPE_CLEAR_COLOR) {
       pipe->bind_blend_state(pipe, ctx->blend_write_color);
-   else
+   } else {
       pipe->bind_blend_state(pipe, ctx->blend_keep_color);
+   }
 
-   if ((clear_buffers & PIPE_CLEAR_DEPTHSTENCIL) == PIPE_CLEAR_DEPTHSTENCIL) {
-      sr.ref_value[0] = stencil & 0xff;
+   if (custom_dsa) {
+      pipe->bind_depth_stencil_alpha_state(pipe, custom_dsa);
+   } else if ((clear_buffers & PIPE_CLEAR_DEPTHSTENCIL) == PIPE_CLEAR_DEPTHSTENCIL) {
       pipe->bind_depth_stencil_alpha_state(pipe, ctx->dsa_write_depth_stencil);
-      pipe->set_stencil_ref(pipe, &sr);
-   }
-   else if (clear_buffers & PIPE_CLEAR_DEPTH) {
+   } else if (clear_buffers & PIPE_CLEAR_DEPTH) {
       pipe->bind_depth_stencil_alpha_state(pipe, ctx->dsa_write_depth_keep_stencil);
-   }
-   else if (clear_buffers & PIPE_CLEAR_STENCIL) {
-      sr.ref_value[0] = stencil & 0xff;
+   } else if (clear_buffers & PIPE_CLEAR_STENCIL) {
       pipe->bind_depth_stencil_alpha_state(pipe, ctx->dsa_keep_depth_write_stencil);
-      pipe->set_stencil_ref(pipe, &sr);
-   }
-   else
+   } else {
       pipe->bind_depth_stencil_alpha_state(pipe, ctx->dsa_keep_depth_stencil);
+   }
+
+   sr.ref_value[0] = stencil & 0xff;
+   pipe->set_stencil_ref(pipe, &sr);
 
    pipe->bind_rasterizer_state(pipe, ctx->rs_state);
    pipe->bind_vertex_elements_state(pipe, ctx->velem_state);
@@ -713,6 +710,27 @@ void util_blitter_clear(struct blitter_context *blitter,
    blitter_restore_CSOs(ctx);
 }
 
+void util_blitter_clear(struct blitter_context *blitter,
+                        unsigned width, unsigned height,
+                        unsigned num_cbufs,
+                        unsigned clear_buffers,
+                        const float *rgba,
+                        double depth, unsigned stencil)
+{
+   util_blitter_clear_custom(blitter, width, height, num_cbufs,
+                             clear_buffers, rgba, depth, stencil,
+                             NULL, NULL);
+}
+
+void util_blitter_clear_depth_custom(struct blitter_context *blitter,
+                                     unsigned width, unsigned height,
+                                     double depth, void *custom_dsa)
+{
+    const float rgba[4] = {0, 0, 0, 0};
+    util_blitter_clear_custom(blitter, width, height, 0,
+                              0, rgba, depth, 0, NULL, custom_dsa);
+}
+
 static
 boolean is_overlap(unsigned sx1, unsigned sx2, unsigned sy1, unsigned sy2,
                    unsigned dx1, unsigned dx2, unsigned dy1, unsigned dy2)
@@ -1024,12 +1042,3 @@ void util_blitter_custom_depth_stencil(struct blitter_context *blitter,
                            UTIL_BLITTER_ATTRIB_NONE, NULL);
    blitter_restore_CSOs(ctx);
 }
-
-/* flush a region of a depth stencil surface for r300g */
-void util_blitter_flush_depth_stencil(struct blitter_context *blitter,
-                                      struct pipe_surface *dstsurf)
-{
-	struct blitter_context_priv *ctx = (struct blitter_context_priv*)blitter;
-	util_blitter_custom_depth_stencil(blitter, dstsurf, NULL,
-					  ctx->dsa_flush_depth_stencil, 0.0f);
-}
diff --git a/src/gallium/auxiliary/util/u_blitter.h b/src/gallium/auxiliary/util/u_blitter.h
index 922a8580ac1..47494c94110 100644
--- a/src/gallium/auxiliary/util/u_blitter.h
+++ b/src/gallium/auxiliary/util/u_blitter.h
@@ -55,13 +55,13 @@ struct blitter_context
     * \param y1      A Y coordinate of the top-left corner.
     * \param x2      An X coordinate of the bottom-right corner.
     * \param y2      A Y coordinate of the bottom-right corner.
-    * \param depth  A depth which the rectangle is rendered at.
+    * \param depth   A depth which the rectangle is rendered at.
     *
     * \param type   Semantics of the attributes "attrib".
     *               If type is UTIL_BLITTER_ATTRIB_NONE, ignore them.
     *               If type is UTIL_BLITTER_ATTRIB_COLOR, the attributes
-    *               make up a constant RGBA color, and should go to the COLOR0
-    *               varying slot of a fragment shader.
+    *               make up a constant RGBA color, and should go
+    *               to the GENERIC0 varying slot of a fragment shader.
     *               If type is UTIL_BLITTER_ATTRIB_TEXCOORD, {a1, a2} and
     *               {a3, a4} specify top-left and bottom-right texture
     *               coordinates of the rectangle, respectively, and should go
@@ -141,6 +141,10 @@ void util_blitter_clear(struct blitter_context *blitter,
                         const float *rgba,
                         double depth, unsigned stencil);
 
+void util_blitter_clear_depth_custom(struct blitter_context *blitter,
+                                     unsigned width, unsigned height,
+                                     double depth, void *custom_dsa);
+
 /**
  * Copy a block of pixels from one surface to another.
  *
@@ -200,9 +204,6 @@ void util_blitter_clear_depth_stencil(struct blitter_context *blitter,
                                       unsigned dstx, unsigned dsty,
                                       unsigned width, unsigned height);
 
-void util_blitter_flush_depth_stencil(struct blitter_context *blitter,
-                                      struct pipe_surface *dstsurf);
-
 void util_blitter_custom_depth_stencil(struct blitter_context *blitter,
 				       struct pipe_surface *zsurf,
 				       struct pipe_surface *cbsurf,
diff --git a/src/gallium/auxiliary/util/u_inlines.h b/src/gallium/auxiliary/util/u_inlines.h
index b4870bce981..fc4eb8fdd2e 100644
--- a/src/gallium/auxiliary/util/u_inlines.h
+++ b/src/gallium/auxiliary/util/u_inlines.h
@@ -160,6 +160,21 @@ pipe_surface_init(struct pipe_context *ctx, struct pipe_surface* ps,
    pipe_surface_reset(ctx, ps, pt, level, layer, flags);
 }
 
+/* Return true if the surfaces are equal. */
+static INLINE boolean
+pipe_surface_equal(struct pipe_surface *s1, struct pipe_surface *s2)
+{
+   return s1->texture == s2->texture &&
+          s1->format == s2->format &&
+          (s1->texture->target != PIPE_BUFFER ||
+           (s1->u.buf.first_element == s2->u.buf.first_element &&
+            s1->u.buf.last_element == s2->u.buf.last_element)) &&
+          (s1->texture->target == PIPE_BUFFER ||
+           (s1->u.tex.level == s2->u.tex.level &&
+            s1->u.tex.first_layer == s2->u.tex.first_layer &&
+            s1->u.tex.last_layer == s2->u.tex.last_layer));
+}
+
 /*
  * Convenience wrappers for screen buffer functions.
  */