Merge remote-tracking branch 'public/master' into vulkan
[mesa.git] / src / mesa / state_tracker / st_cb_texture.c
index 9d90d031f8818e7d67564a7cf00f7b4434a1b411..460c1790663a00b61779b168e83fd15ef8ddf391 100644 (file)
@@ -706,15 +706,13 @@ st_init_pbo_upload(struct st_context *st)
    st->pbo_upload.rgba_only =
       screen->get_param(screen, PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY);
 
-   /* Create the vertex shader */
-   {
-      unsigned semantic_names[] = { TGSI_SEMANTIC_POSITION };
-      unsigned semantic_indexes[] = { 0 };
-
-      st->pbo_upload.vs = util_make_vertex_passthrough_shader(pipe, 1,
-                                                              semantic_names,
-                                                              semantic_indexes,
-                                                              FALSE);
+   if (screen->get_param(screen, PIPE_CAP_TGSI_INSTANCEID)) {
+      if (screen->get_param(screen, PIPE_CAP_TGSI_VS_LAYER_VIEWPORT)) {
+         st->pbo_upload.upload_layers = true;
+      } else if (screen->get_param(screen, PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES) >= 3) {
+         st->pbo_upload.upload_layers = true;
+         st->pbo_upload.use_gs = true;
+      }
    }
 
    /* Blend state */
@@ -734,6 +732,11 @@ st_destroy_pbo_upload(struct st_context *st)
       st->pbo_upload.fs = NULL;
    }
 
+   if (st->pbo_upload.gs) {
+      cso_delete_geometry_shader(st->cso_context, st->pbo_upload.gs);
+      st->pbo_upload.gs = NULL;
+   }
+
    if (st->pbo_upload.vs) {
       cso_delete_vertex_shader(st->cso_context, st->pbo_upload.vs);
       st->pbo_upload.vs = NULL;
@@ -1120,7 +1123,94 @@ reinterpret_formats(enum pipe_format *src_format, enum pipe_format *dst_format)
 }
 
 static void *
-create_pbo_upload_shader(struct st_context *st)
+create_pbo_upload_vs(struct st_context *st)
+{
+   struct ureg_program *ureg;
+   struct ureg_src in_pos;
+   struct ureg_src in_instanceid;
+   struct ureg_dst out_pos;
+   struct ureg_dst out_layer;
+
+   ureg = ureg_create(TGSI_PROCESSOR_VERTEX);
+   if (!ureg)
+      return NULL;
+
+   in_pos = ureg_DECL_vs_input(ureg, TGSI_SEMANTIC_POSITION);
+
+   out_pos = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
+
+   if (st->pbo_upload.upload_layers) {
+      in_instanceid = ureg_DECL_system_value(ureg, TGSI_SEMANTIC_INSTANCEID, 0);
+
+      if (!st->pbo_upload.use_gs)
+         out_layer = ureg_DECL_output(ureg, TGSI_SEMANTIC_LAYER, 0);
+   }
+
+   /* out_pos = in_pos */
+   ureg_MOV(ureg, out_pos, in_pos);
+
+   if (st->pbo_upload.upload_layers) {
+      if (st->pbo_upload.use_gs) {
+         /* out_pos.z = i2f(gl_InstanceID) */
+         ureg_I2F(ureg, ureg_writemask(out_pos, TGSI_WRITEMASK_Z),
+                        ureg_scalar(in_instanceid, TGSI_SWIZZLE_X));
+      } else {
+         /* out_layer = gl_InstanceID */
+         ureg_MOV(ureg, out_layer, in_instanceid);
+      }
+   }
+
+   ureg_END(ureg);
+
+   return ureg_create_shader_and_destroy(ureg, st->pipe);
+}
+
+static void *
+create_pbo_upload_gs(struct st_context *st)
+{
+   static const int zero = 0;
+   struct ureg_program *ureg;
+   struct ureg_dst out_pos;
+   struct ureg_dst out_layer;
+   struct ureg_src in_pos;
+   struct ureg_src imm;
+   unsigned i;
+
+   ureg = ureg_create(TGSI_PROCESSOR_GEOMETRY);
+   if (!ureg)
+      return NULL;
+
+   ureg_property(ureg, TGSI_PROPERTY_GS_INPUT_PRIM, PIPE_PRIM_TRIANGLES);
+   ureg_property(ureg, TGSI_PROPERTY_GS_OUTPUT_PRIM, PIPE_PRIM_TRIANGLE_STRIP);
+   ureg_property(ureg, TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES, 3);
+
+   out_pos = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
+   out_layer = ureg_DECL_output(ureg, TGSI_SEMANTIC_LAYER, 0);
+
+   in_pos = ureg_DECL_input(ureg, TGSI_SEMANTIC_POSITION, 0, 0, 1);
+
+   imm = ureg_DECL_immediate_int(ureg, &zero, 1);
+
+   for (i = 0; i < 3; ++i) {
+      struct ureg_src in_pos_vertex = ureg_src_dimension(in_pos, i);
+
+      /* out_pos = in_pos[i] */
+      ureg_MOV(ureg, out_pos, in_pos_vertex);
+
+      /* out_layer.x = f2i(in_pos[i].z) */
+      ureg_F2I(ureg, ureg_writemask(out_layer, TGSI_WRITEMASK_X),
+                     ureg_scalar(in_pos_vertex, TGSI_SWIZZLE_Z));
+
+      ureg_EMIT(ureg, ureg_scalar(imm, TGSI_SWIZZLE_X));
+   }
+
+   ureg_END(ureg);
+
+   return ureg_create_shader_and_destroy(ureg, st->pipe);
+}
+
+static void *
+create_pbo_upload_fs(struct st_context *st)
 {
    struct pipe_context *pipe = st->pipe;
    struct pipe_screen *screen = pipe->screen;
@@ -1128,10 +1218,14 @@ create_pbo_upload_shader(struct st_context *st)
    struct ureg_dst out;
    struct ureg_src sampler;
    struct ureg_src pos;
+   struct ureg_src layer;
    struct ureg_src const0;
    struct ureg_dst temp0;
 
-   ureg    = ureg_create(TGSI_PROCESSOR_FRAGMENT);
+   ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT);
+   if (!ureg)
+      return NULL;
+
    out     = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
    sampler = ureg_DECL_sampler(ureg, 0);
    if (screen->get_param(screen, PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL)) {
@@ -1140,10 +1234,14 @@ create_pbo_upload_shader(struct st_context *st)
       pos = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_POSITION, 0,
                                TGSI_INTERPOLATE_LINEAR);
    }
+   if (st->pbo_upload.upload_layers) {
+      layer = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_LAYER, 0,
+                                       TGSI_INTERPOLATE_CONSTANT);
+   }
    const0  = ureg_DECL_constant(ureg, 0);
    temp0   = ureg_DECL_temporary(ureg);
 
-   /* Note: const0 = [ -xoffset + skip_pixels, -yoffset, stride, 0 ] */
+   /* Note: const0 = [ -xoffset + skip_pixels, -yoffset, stride, image_height ] */
 
    /* temp0.xy = f2i(temp0.xy) */
    ureg_F2I(ureg, ureg_writemask(temp0, TGSI_WRITEMASK_XY),
@@ -1166,10 +1264,19 @@ create_pbo_upload_shader(struct st_context *st)
                    ureg_scalar(ureg_src(temp0), TGSI_SWIZZLE_Y),
                    ureg_scalar(ureg_src(temp0), TGSI_SWIZZLE_X));
 
+   if (st->pbo_upload.upload_layers) {
+      /* temp0.x = const0.w * layer + temp0.x */
+      ureg_UMAD(ureg, ureg_writemask(temp0, TGSI_WRITEMASK_X),
+                      ureg_scalar(const0, TGSI_SWIZZLE_W),
+                      ureg_scalar(layer, TGSI_SWIZZLE_X),
+                      ureg_scalar(ureg_src(temp0), TGSI_SWIZZLE_X));
+   }
+
+   /* temp0.w = 0 */
+   ureg_MOV(ureg, ureg_writemask(temp0, TGSI_WRITEMASK_W), ureg_imm1u(ureg, 0));
+
    /* out = txf(sampler, temp0.x) */
-   ureg_TXF(ureg, out, TGSI_TEXTURE_BUFFER,
-                  ureg_scalar(ureg_src(temp0), TGSI_SWIZZLE_X),
-                  sampler);
+   ureg_TXF(ureg, out, TGSI_TEXTURE_BUFFER, ureg_src(temp0), sampler);
 
    ureg_release_temporary(ureg, temp0);
 
@@ -1191,10 +1298,11 @@ try_pbo_upload_common(struct gl_context *ctx,
                       unsigned image_height)
 {
    struct st_context *st = st_context(ctx);
+   struct cso_context *cso = st->cso_context;
    struct pipe_context *pipe = st->pipe;
-   struct pipe_sampler_view *sampler_view = NULL;
    unsigned depth = surface->u.tex.last_layer - surface->u.tex.first_layer + 1;
    unsigned skip_pixels = 0;
+   bool success = false;
 
    /* Check alignment. */
    {
@@ -1209,26 +1317,56 @@ try_pbo_upload_common(struct gl_context *ctx,
    }
 
    /* Create the shaders */
+   if (!st->pbo_upload.vs) {
+      st->pbo_upload.vs = create_pbo_upload_vs(st);
+      if (!st->pbo_upload.vs)
+         return false;
+   }
+
+   if (depth != 1 && st->pbo_upload.use_gs && !st->pbo_upload.gs) {
+      st->pbo_upload.gs = create_pbo_upload_gs(st);
+      if (!st->pbo_upload.gs)
+         return false;
+   }
+
    if (!st->pbo_upload.fs) {
-      st->pbo_upload.fs = create_pbo_upload_shader(st);
+      st->pbo_upload.fs = create_pbo_upload_fs(st);
       if (!st->pbo_upload.fs)
          return false;
    }
 
+   cso_save_state(cso, (CSO_BIT_FRAGMENT_SAMPLER_VIEWS |
+                        CSO_BIT_FRAGMENT_SAMPLERS |
+                        CSO_BIT_VERTEX_ELEMENTS |
+                        CSO_BIT_AUX_VERTEX_BUFFER_SLOT |
+                        CSO_BIT_FRAMEBUFFER |
+                        CSO_BIT_VIEWPORT |
+                        CSO_BIT_BLEND |
+                        CSO_BIT_DEPTH_STENCIL_ALPHA |
+                        CSO_BIT_RASTERIZER |
+                        CSO_BIT_STREAM_OUTPUTS |
+                        CSO_BITS_ALL_SHADERS));
+   cso_save_constant_buffer_slot0(cso, PIPE_SHADER_FRAGMENT);
+
+
    /* Set up the sampler_view */
    {
       unsigned first_element = buf_offset;
       unsigned last_element = buf_offset + skip_pixels + upload_width - 1
          + (upload_height - 1 + (depth - 1) * image_height) * stride;
       struct pipe_sampler_view templ;
+      struct pipe_sampler_view *sampler_view;
+      struct pipe_sampler_state sampler = {0};
+      const struct pipe_sampler_state *samplers[1] = {&sampler};
 
       /* This should be ensured by Mesa before calling our callbacks */
       assert((last_element + 1) * bytes_per_pixel <= buffer->width0);
 
       if (last_element - first_element > ctx->Const.MaxTextureBufferSize - 1)
-         return false;
+         goto fail;
 
       memset(&templ, 0, sizeof(templ));
+      templ.target = PIPE_BUFFER;
       templ.format = src_format;
       templ.u.buf.first_element = first_element;
       templ.u.buf.last_element = last_element;
@@ -1239,51 +1377,15 @@ try_pbo_upload_common(struct gl_context *ctx,
 
       sampler_view = pipe->create_sampler_view(pipe, buffer, &templ);
       if (sampler_view == NULL)
-         return false;
-   }
-
-   /* Begin setting state. This is the point of no return. */
-   cso_save_fragment_sampler_views(st->cso_context);
-   cso_set_sampler_views(st->cso_context, PIPE_SHADER_FRAGMENT, 1,
-                         &sampler_view);
-
-   /* Framebuffer_state */
-   {
-      struct pipe_framebuffer_state fb;
-      memset(&fb, 0, sizeof(fb));
-      fb.width = surface->width;
-      fb.height = surface->height;
-      fb.nr_cbufs = 1;
-      pipe_surface_reference(&fb.cbufs[0], surface);
-
-      cso_save_framebuffer(st->cso_context);
-      cso_set_framebuffer(st->cso_context, &fb);
+         goto fail;
 
-      pipe_surface_reference(&fb.cbufs[0], NULL);
-   }
+      cso_set_sampler_views(cso, PIPE_SHADER_FRAGMENT, 1, &sampler_view);
 
-   /* Viewport state */
-   {
-      struct pipe_viewport_state vp;
-      vp.scale[0] = 0.5f * surface->width;
-      vp.scale[1] = 0.5f * surface->height;
-      vp.scale[2] = 1.0f;
-      vp.translate[0] = 0.5f * surface->width;
-      vp.translate[1] = 0.5f * surface->height;
-      vp.translate[2] = 0.0f;
+      pipe_sampler_view_reference(&sampler_view, NULL);
 
-      cso_save_viewport(st->cso_context);
-      cso_set_viewport(st->cso_context, &vp);
+      cso_set_samplers(cso, PIPE_SHADER_FRAGMENT, 1, samplers);
    }
 
-   /* Blend state */
-   cso_save_blend(st->cso_context);
-   cso_set_blend(st->cso_context, &st->pbo_upload.blend);
-
-   /* Rasterizer state */
-   cso_save_rasterizer(st->cso_context);
-   cso_set_rasterizer(st->cso_context, &st->pbo_upload.raster);
-
    /* Upload vertices */
    {
       struct pipe_vertex_buffer vbo;
@@ -1302,6 +1404,8 @@ try_pbo_upload_common(struct gl_context *ctx,
 
       u_upload_alloc(st->uploader, 0, 8 * sizeof(float), 4,
                      &vbo.buffer_offset, &vbo.buffer, (void **) &verts);
+      if (!verts)
+         goto fail;
 
       verts[0] = x0;
       verts[1] = y0;
@@ -1316,39 +1420,42 @@ try_pbo_upload_common(struct gl_context *ctx,
 
       velem.src_offset = 0;
       velem.instance_divisor = 0;
-      velem.vertex_buffer_index = cso_get_aux_vertex_buffer_slot(st->cso_context);
+      velem.vertex_buffer_index = cso_get_aux_vertex_buffer_slot(cso);
       velem.src_format = PIPE_FORMAT_R32G32_FLOAT;
 
-      cso_save_vertex_elements(st->cso_context);
-      cso_set_vertex_elements(st->cso_context, 1, &velem);
+      cso_set_vertex_elements(cso, 1, &velem);
+
+      cso_set_vertex_buffers(cso, velem.vertex_buffer_index, 1, &vbo);
 
-      cso_save_aux_vertex_buffer_slot(st->cso_context);
-      cso_set_vertex_buffers(st->cso_context, velem.vertex_buffer_index,
-                             1, &vbo);
+      pipe_resource_reference(&vbo.buffer, NULL);
    }
 
    /* Upload constants */
+   /* Note: the user buffer must be valid until draw time */
+   struct {
+      int32_t xoffset;
+      int32_t yoffset;
+      int32_t stride;
+      int32_t image_size;
+   } constants;
+
    {
       struct pipe_constant_buffer cb;
 
-      struct {
-         int32_t xoffset;
-         int32_t yoffset;
-         int32_t stride;
-         int32_t pad;
-      } constants;
-
       constants.xoffset = -xoffset + skip_pixels;
       constants.yoffset = -yoffset;
       constants.stride = stride;
-      constants.pad = 0;
+      constants.image_size = stride * image_height;
 
       if (st->constbuf_uploader) {
          cb.buffer = NULL;
          cb.user_buffer = NULL;
          u_upload_data(st->constbuf_uploader, 0, sizeof(constants),
-                       st->ctx->Const.UniformBufferOffsetAlignment,
+                       ctx->Const.UniformBufferOffsetAlignment,
                        &constants, &cb.buffer_offset, &cb.buffer);
+         if (!cb.buffer)
+            goto fail;
+
          u_upload_unmap(st->constbuf_uploader);
       } else {
          cb.buffer = NULL;
@@ -1357,50 +1464,68 @@ try_pbo_upload_common(struct gl_context *ctx,
       }
       cb.buffer_size = sizeof(constants);
 
-      cso_save_constant_buffer_slot0(st->cso_context, PIPE_SHADER_FRAGMENT);
-      cso_set_constant_buffer(st->cso_context, PIPE_SHADER_FRAGMENT, 0, &cb);
+      cso_set_constant_buffer(cso, PIPE_SHADER_FRAGMENT, 0, &cb);
+
+      pipe_resource_reference(&cb.buffer, NULL);
+   }
+
+   /* Framebuffer_state */
+   {
+      struct pipe_framebuffer_state fb;
+      memset(&fb, 0, sizeof(fb));
+      fb.width = surface->width;
+      fb.height = surface->height;
+      fb.nr_cbufs = 1;
+      pipe_surface_reference(&fb.cbufs[0], surface);
+
+      cso_set_framebuffer(cso, &fb);
+
+      pipe_surface_reference(&fb.cbufs[0], NULL);
    }
 
+   cso_set_viewport_dims(cso, surface->width, surface->height, FALSE);
+
+   /* Blend state */
+   cso_set_blend(cso, &st->pbo_upload.blend);
+
+   /* Depth/stencil/alpha state */
+   {
+      struct pipe_depth_stencil_alpha_state dsa;
+      memset(&dsa, 0, sizeof(dsa));
+      cso_set_depth_stencil_alpha(cso, &dsa);
+   }
+
+   /* Rasterizer state */
+   cso_set_rasterizer(cso, &st->pbo_upload.raster);
+
    /* Set up the shaders */
-   cso_save_vertex_shader(st->cso_context);
-   cso_set_vertex_shader_handle(st->cso_context, st->pbo_upload.vs);
+   cso_set_vertex_shader_handle(cso, st->pbo_upload.vs);
 
-   cso_save_geometry_shader(st->cso_context);
-   cso_set_geometry_shader_handle(st->cso_context, NULL);
+   cso_set_geometry_shader_handle(cso, depth != 1 ? st->pbo_upload.gs : NULL);
 
-   cso_save_tessctrl_shader(st->cso_context);
-   cso_set_tessctrl_shader_handle(st->cso_context, NULL);
+   cso_set_tessctrl_shader_handle(cso, NULL);
 
-   cso_save_tesseval_shader(st->cso_context);
-   cso_set_tesseval_shader_handle(st->cso_context, NULL);
+   cso_set_tesseval_shader_handle(cso, NULL);
 
-   cso_save_fragment_shader(st->cso_context);
-   cso_set_fragment_shader_handle(st->cso_context, st->pbo_upload.fs);
+   cso_set_fragment_shader_handle(cso, st->pbo_upload.fs);
 
    /* Disable stream output */
-   cso_save_stream_outputs(st->cso_context);
-   cso_set_stream_outputs(st->cso_context, 0, NULL, 0);
-
-   cso_draw_arrays(st->cso_context, PIPE_PRIM_TRIANGLE_STRIP, 0, 4);
-
-   cso_restore_fragment_sampler_views(st->cso_context);
-   cso_restore_framebuffer(st->cso_context);
-   cso_restore_viewport(st->cso_context);
-   cso_restore_blend(st->cso_context);
-   cso_restore_rasterizer(st->cso_context);
-   cso_restore_vertex_elements(st->cso_context);
-   cso_restore_aux_vertex_buffer_slot(st->cso_context);
-   cso_restore_constant_buffer_slot0(st->cso_context, PIPE_SHADER_FRAGMENT);
-   cso_restore_vertex_shader(st->cso_context);
-   cso_restore_geometry_shader(st->cso_context);
-   cso_restore_tessctrl_shader(st->cso_context);
-   cso_restore_tesseval_shader(st->cso_context);
-   cso_restore_fragment_shader(st->cso_context);
-   cso_restore_stream_outputs(st->cso_context);
-
-   pipe_sampler_view_reference(&sampler_view, NULL);
+   cso_set_stream_outputs(cso, 0, NULL, 0);
 
-   return true;
+   if (depth == 1) {
+      cso_draw_arrays(cso, PIPE_PRIM_TRIANGLE_STRIP, 0, 4);
+   } else {
+      cso_draw_arrays_instanced(cso, PIPE_PRIM_TRIANGLE_STRIP,
+                                0, 4, 0, depth);
+   }
+
+   success = true;
+
+fail:
+   cso_restore_state(cso);
+   cso_restore_constant_buffer_slot0(cso, PIPE_SHADER_FRAGMENT);
+
+   return success;
 }
 
 static bool
@@ -1442,8 +1567,7 @@ try_pbo_upload(struct gl_context *ctx, GLuint dims,
       image_height = unpack->ImageHeight > 0 ? unpack->ImageHeight : height;
    }
 
-   /* XXX We only support updating a single layer */
-   if (depth != 1)
+   if (depth != 1 && !st->pbo_upload.upload_layers)
       return false;
 
    /* Choose the source format. Initially, we do so without checking driver
@@ -1799,13 +1923,166 @@ st_TexImage(struct gl_context * ctx, GLuint dims,
 }
 
 
+static void
+st_CompressedTexSubImage(struct gl_context *ctx, GLuint dims,
+                         struct gl_texture_image *texImage,
+                         GLint x, GLint y, GLint z,
+                         GLsizei w, GLsizei h, GLsizei d,
+                         GLenum format, GLsizei imageSize, const GLvoid *data)
+{
+   struct st_context *st = st_context(ctx);
+   struct st_texture_image *stImage = st_texture_image(texImage);
+   struct st_texture_object *stObj = st_texture_object(texImage->TexObject);
+   struct pipe_resource *texture = stImage->pt;
+   struct pipe_context *pipe = st->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_resource *dst = stImage->pt;
+   struct pipe_surface *surface = NULL;
+   struct compressed_pixelstore store;
+   enum pipe_format copy_format;
+   unsigned bytes_per_block;
+   unsigned bw, bh;
+   intptr_t buf_offset;
+   bool success = false;
+
+   /* Check basic pre-conditions for PBO upload */
+   if (!st->prefer_blit_based_texture_transfer) {
+      goto fallback;
+   }
+
+   if (!_mesa_is_bufferobj(ctx->Unpack.BufferObj))
+      goto fallback;
+
+   if ((_mesa_is_format_etc2(texImage->TexFormat) && !st->has_etc2) ||
+       (texImage->TexFormat == MESA_FORMAT_ETC1_RGB8 && !st->has_etc1)) {
+      /* ETC isn't supported and is represented by uncompressed formats. */
+      goto fallback;
+   }
+
+   if (!dst) {
+      goto fallback;
+   }
+
+   if (!st->pbo_upload.enabled ||
+       !screen->get_param(screen, PIPE_CAP_SURFACE_REINTERPRET_BLOCKS)) {
+      goto fallback;
+   }
+
+   /* Choose the pipe format for the upload. */
+   bytes_per_block = util_format_get_blocksize(dst->format);
+   bw = util_format_get_blockwidth(dst->format);
+   bh = util_format_get_blockheight(dst->format);
+
+   switch (bytes_per_block) {
+   case 8:
+      copy_format = PIPE_FORMAT_R16G16B16A16_UINT;
+      break;
+   case 16:
+      copy_format = PIPE_FORMAT_R32G32B32A32_UINT;
+      break;
+   default:
+      goto fallback;
+   }
+
+   if (!screen->is_format_supported(screen, copy_format, PIPE_BUFFER, 0,
+                                    PIPE_BIND_SAMPLER_VIEW)) {
+      goto fallback;
+   }
+
+   if (!screen->is_format_supported(screen, copy_format, dst->target,
+                                    dst->nr_samples, PIPE_BIND_RENDER_TARGET)) {
+      goto fallback;
+   }
+
+   /* Interpret the pixelstore settings. */
+   _mesa_compute_compressed_pixelstore(dims, texImage->TexFormat, w, h, d,
+                                       &ctx->Unpack, &store);
+   assert(store.CopyBytesPerRow % bytes_per_block == 0);
+   assert(store.SkipBytes % bytes_per_block == 0);
+
+   /* Compute the offset into the buffer */
+   buf_offset = (intptr_t)data + store.SkipBytes;
+
+   if (buf_offset % bytes_per_block) {
+      goto fallback;
+   }
+
+   buf_offset = buf_offset / bytes_per_block;
+
+   /* Set up the surface. */
+   {
+      unsigned level = stObj->pt != stImage->pt ? 0 : texImage->TexObject->MinLevel + texImage->Level;
+      unsigned max_layer = util_max_layer(texture, level);
+
+      z += texImage->Face + texImage->TexObject->MinLayer;
+
+      struct pipe_surface templ;
+      memset(&templ, 0, sizeof(templ));
+      templ.format = copy_format;
+      templ.u.tex.level = level;
+      templ.u.tex.first_layer = MIN2(z, max_layer);
+      templ.u.tex.last_layer = MIN2(z + d - 1, max_layer);
+
+      surface = pipe->create_surface(pipe, texture, &templ);
+      if (!surface)
+         goto fallback;
+   }
+
+   success = try_pbo_upload_common(ctx, surface,
+                                   x / bw, y / bh,
+                                   store.CopyBytesPerRow / bytes_per_block,
+                                   store.CopyRowsPerSlice,
+                                   st_buffer_object(ctx->Unpack.BufferObj)->buffer,
+                                   copy_format,
+                                   buf_offset,
+                                   bytes_per_block,
+                                   store.TotalBytesPerRow / bytes_per_block,
+                                   store.TotalRowsPerSlice);
+
+   pipe_surface_reference(&surface, NULL);
+
+   if (success)
+      return;
+
+fallback:
+   _mesa_store_compressed_texsubimage(ctx, dims, texImage,
+                                      x, y, z, w, h, d,
+                                      format, imageSize, data);
+}
+
 static void
 st_CompressedTexImage(struct gl_context *ctx, GLuint dims,
                       struct gl_texture_image *texImage,
                       GLsizei imageSize, const GLvoid *data)
 {
    prep_teximage(ctx, texImage, GL_NONE, GL_NONE);
-   _mesa_store_compressed_teximage(ctx, dims, texImage, imageSize, data);
+
+   /* only 2D and 3D compressed images are supported at this time */
+   if (dims == 1) {
+      _mesa_problem(ctx, "Unexpected glCompressedTexImage1D call");
+      return;
+   }
+
+   /* This is pretty simple, because unlike the general texstore path we don't
+    * have to worry about the usual image unpacking or image transfer
+    * operations.
+    */
+   assert(texImage);
+   assert(texImage->Width > 0);
+   assert(texImage->Height > 0);
+   assert(texImage->Depth > 0);
+
+   /* allocate storage for texture data */
+   if (!st_AllocTextureImageBuffer(ctx, texImage)) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glCompressedTexImage%uD", dims);
+      return;
+   }
+
+   st_CompressedTexSubImage(ctx, dims, texImage,
+                            0, 0, 0,
+                            texImage->Width, texImage->Height, texImage->Depth,
+                            texImage->TexFormat,
+                            imageSize, data);
 }
 
 
@@ -2463,7 +2740,7 @@ st_finalize_texture(struct gl_context *ctx,
 {
    struct st_context *st = st_context(ctx);
    struct st_texture_object *stObj = st_texture_object(tObj);
-   const GLuint nr_faces = (stObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1;
+   const GLuint nr_faces = _mesa_num_tex_faces(stObj->base.Target);
    GLuint face;
    const struct st_texture_image *firstImage;
    enum pipe_format firstImageFormat;
@@ -2609,10 +2886,13 @@ st_finalize_texture(struct gl_context *ctx,
          /* Need to import images in main memory or held in other textures.
           */
          if (stImage && stObj->pt != stImage->pt) {
+            GLuint depth = stObj->depth0;
+            if (stObj->base.Target == GL_TEXTURE_3D)
+               depth = u_minify(depth, level);
             if (level == 0 ||
                 (stImage->base.Width == u_minify(stObj->width0, level) &&
                  stImage->base.Height == u_minify(stObj->height0, level) &&
-                 stImage->base.Depth == u_minify(stObj->depth0, level))) {
+                 stImage->base.Depth == depth)) {
                /* src image fits expected dest mipmap level size */
                copy_image_data_to_texture(st, stObj, level, stImage);
             }
@@ -2827,10 +3107,10 @@ void
 st_init_texture_functions(struct dd_function_table *functions)
 {
    functions->ChooseTextureFormat = st_ChooseTextureFormat;
-   functions->QuerySamplesForFormat = st_QuerySamplesForFormat;
+   functions->QueryInternalFormat = st_QueryInternalFormat;
    functions->TexImage = st_TexImage;
    functions->TexSubImage = st_TexSubImage;
-   functions->CompressedTexSubImage = _mesa_store_compressed_texsubimage;
+   functions->CompressedTexSubImage = st_CompressedTexSubImage;
    functions->CopyTexSubImage = st_CopyTexSubImage;
    functions->GenerateMipmap = st_generate_mipmap;