[g3dvl] split idct code into state and buffers

6 files changed, 343 insertions, 277 deletions

diff --git a/src/gallium/auxiliary/vl/vl_idct.c b/src/gallium/auxiliary/vl/vl_idct.c
index f19cfc38d7b..168e6e88b2b 100644
--- a/src/gallium/auxiliary/vl/vl_idct.c
+++ b/src/gallium/auxiliary/vl/vl_idct.c
@@ -197,9 +197,6 @@ matrix_mul(struct ureg_program *shader, struct ureg_dst dst, struct ureg_dst l[2
 static void *
 create_transpose_frag_shader(struct vl_idct *idct)
 {
-   struct pipe_resource *transpose = idct->textures.individual.transpose;
-   struct pipe_resource *intermediate = idct->textures.individual.intermediate;
-
    struct ureg_program *shader;
 
    struct ureg_src block, tex, sampler[2];
@@ -221,8 +218,8 @@ create_transpose_frag_shader(struct vl_idct *idct)
    start[0] = ureg_imm1f(shader, 0.0f);
    start[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 
-   fetch_four(shader, l, block, sampler[0], start[0], block, false, false, transpose->width0);
-   fetch_four(shader, r, tex, sampler[1], start[1], block, true, false, intermediate->height0);
+   fetch_four(shader, l, block, sampler[0], start[0], block, false, false, BLOCK_WIDTH / 4);
+   fetch_four(shader, r, tex, sampler[1], start[1], block, true, false, idct->buffer_height / 4);
 
    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 
@@ -244,9 +241,6 @@ create_transpose_frag_shader(struct vl_idct *idct)
 static void *
 create_matrix_frag_shader(struct vl_idct *idct)
 {
-   struct pipe_resource *matrix = idct->textures.individual.matrix;
-   struct pipe_resource *source = idct->textures.individual.source;
-
    struct ureg_program *shader;
 
    struct ureg_src tex, block, sampler[2];
@@ -278,21 +272,21 @@ create_matrix_frag_shader(struct vl_idct *idct)
 
    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), tex);
    for (i = 0; i < 4; ++i) {
-      fetch_four(shader, l[i], ureg_src(t_tc), sampler[0], start[0], block, false, false, source->width0);
+      fetch_four(shader, l[i], ureg_src(t_tc), sampler[0], start[0], block, false, false, idct->buffer_width / 4);
       ureg_MUL(shader, l[i][0], ureg_src(l[i][0]), ureg_imm1f(shader, STAGE1_SCALE));
       ureg_MUL(shader, l[i][1], ureg_src(l[i][1]), ureg_imm1f(shader, STAGE1_SCALE));
       if(i != 3)
          ureg_ADD(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), 
-            ureg_src(t_tc), ureg_imm1f(shader, 1.0f / source->height0));
+            ureg_src(t_tc), ureg_imm1f(shader, 1.0f / idct->buffer_height));
    }
    
    for (i = 0; i < NR_RENDER_TARGETS; ++i) {
 
 #if NR_RENDER_TARGETS == 8
       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_imm1f(shader, 1.0f / BLOCK_WIDTH * i));
-      fetch_four(shader, r, ureg_src(t_tc), sampler[1], start[1], block, true, true, matrix->width0);
+      fetch_four(shader, r, ureg_src(t_tc), sampler[1], start[1], block, true, true, BLOCK_WIDTH / 4);
 #elif NR_RENDER_TARGETS == 1
-      fetch_four(shader, r, block, sampler[1], start[1], block, true, true, matrix->width0);
+      fetch_four(shader, r, block, sampler[1], start[1], block, true, true, BLOCK_WIDTH / 4);
 #else
 #error invalid number of render targets
 #endif
@@ -339,13 +333,81 @@ cleanup_shaders(struct vl_idct *idct)
 }
 
 static bool
-init_buffers(struct vl_idct *idct)
+init_state(struct vl_idct *idct)
+{
+   struct pipe_vertex_element vertex_elems[NUM_VS_INPUTS];
+   struct pipe_sampler_state sampler;
+   struct pipe_rasterizer_state rs_state;
+   unsigned i;
+
+   assert(idct);
+
+   idct->quad = vl_vb_upload_quads(idct->pipe, idct->max_blocks);
+
+   if(idct->quad.buffer == NULL)
+      return false;
+
+   for (i = 0; i < 4; ++i) {
+      memset(&sampler, 0, sizeof(sampler));
+      sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+      sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+      sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+      sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
+      sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
+      sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
+      sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
+      sampler.compare_func = PIPE_FUNC_ALWAYS;
+      sampler.normalized_coords = 1;
+      /*sampler.shadow_ambient = ; */
+      /*sampler.lod_bias = ; */
+      sampler.min_lod = 0;
+      /*sampler.max_lod = ; */
+      /*sampler.border_color[0] = ; */
+      /*sampler.max_anisotropy = ; */
+      idct->samplers.all[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
+   }
+
+   memset(&rs_state, 0, sizeof(rs_state));
+   /*rs_state.sprite_coord_enable */
+   rs_state.sprite_coord_mode = PIPE_SPRITE_COORD_UPPER_LEFT;
+   rs_state.point_quad_rasterization = true;
+   rs_state.point_size = BLOCK_WIDTH;
+   rs_state.gl_rasterization_rules = false;
+   idct->rs_state = idct->pipe->create_rasterizer_state(idct->pipe, &rs_state);
+
+   vertex_elems[VS_I_RECT] = vl_vb_get_quad_vertex_element();
+
+   /* Pos element */
+   vertex_elems[VS_I_VPOS].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   idct->vertex_buffer_stride = vl_vb_element_helper(&vertex_elems[VS_I_VPOS], 1, 1);
+   idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems);
+
+   return true;
+}
+
+static void
+cleanup_state(struct vl_idct *idct)
+{
+   unsigned i;
+
+   for (i = 0; i < 4; ++i)
+      idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]);
+
+   idct->pipe->delete_rasterizer_state(idct->pipe, idct->rs_state);
+   idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
+}
+
+static bool
+init_textures(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 {
    struct pipe_resource template;
    struct pipe_sampler_view sampler_view;
-   struct pipe_vertex_element vertex_elems[2];
    unsigned i;
 
+   assert(idct && buffer);
+
+   /* create textures */
    memset(&template, 0, sizeof(struct pipe_resource));
    template.last_level = 0;
    template.depth0 = 1;
@@ -354,149 +416,74 @@ init_buffers(struct vl_idct *idct)
 
    template.target = PIPE_TEXTURE_2D;
    template.format = PIPE_FORMAT_R16G16B16A16_SNORM;
-   template.width0 = idct->destination->width0 / 4;
-   template.height0 = idct->destination->height0;
+   template.width0 = idct->buffer_width / 4;
+   template.height0 = idct->buffer_height;
    template.depth0 = 1;
    template.usage = PIPE_USAGE_STREAM;
-   idct->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
+   buffer->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 
    template.target = PIPE_TEXTURE_3D;
    template.format = PIPE_FORMAT_R16G16B16A16_SNORM;
-   template.width0 = idct->destination->width0 / NR_RENDER_TARGETS;
-   template.height0 = idct->destination->height0 / 4;
+   template.width0 = idct->buffer_width / NR_RENDER_TARGETS;
+   template.height0 = idct->buffer_height / 4;
    template.depth0 = NR_RENDER_TARGETS;
    template.usage = PIPE_USAGE_STATIC;
-   idct->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
+   buffer->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 
    for (i = 0; i < 4; ++i) {
-      if(idct->textures.all[i] == NULL)
+      if(buffer->textures.all[i] == NULL)
          return false; /* a texture failed to allocate */
 
-      u_sampler_view_default_template(&sampler_view, idct->textures.all[i], idct->textures.all[i]->format);
-      idct->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, idct->textures.all[i], &sampler_view);
+      u_sampler_view_default_template(&sampler_view, buffer->textures.all[i], buffer->textures.all[i]->format);
+      buffer->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, buffer->textures.all[i], &sampler_view);
    }
 
-   idct->vertex_bufs.individual.quad = vl_vb_upload_quads(idct->pipe, idct->max_blocks, &vertex_elems[VS_I_RECT]);
-
-   if(idct->vertex_bufs.individual.quad.buffer == NULL)
-      return false;
-
-   /* Pos element */
-   vertex_elems[VS_I_VPOS].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-   idct->vertex_bufs.individual.pos = vl_vb_create_buffer(idct->pipe, idct->max_blocks, &vertex_elems[VS_I_VPOS], 1, 1);
-
-   if(idct->vertex_bufs.individual.pos.buffer == NULL)
-      return false;
-
-   idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems);
-
    return true;
 }
 
 static void
-cleanup_buffers(struct vl_idct *idct)
+cleanup_textures(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 {
    unsigned i;
 
-   assert(idct);
+   assert(idct && buffer);
 
    for (i = 0; i < 4; ++i) {
-      pipe_sampler_view_reference(&idct->sampler_views.all[i], NULL);
-      pipe_resource_reference(&idct->textures.all[i], NULL);
+      pipe_sampler_view_reference(&buffer->sampler_views.all[i], NULL);
+      pipe_resource_reference(&buffer->textures.all[i], NULL);
    }
-
-   idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
-   pipe_resource_reference(&idct->vertex_bufs.individual.quad.buffer, NULL);
-   pipe_resource_reference(&idct->vertex_bufs.individual.pos.buffer, NULL);
 }
 
-static void
-init_state(struct vl_idct *idct)
+static bool
+init_vertex_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 {
-   struct pipe_sampler_state sampler;
-   struct pipe_rasterizer_state rs_state;
-   unsigned i;
-
-   idct->viewport[0].scale[0] = idct->textures.individual.intermediate->width0;
-   idct->viewport[0].scale[1] = idct->textures.individual.intermediate->height0;
+   assert(idct && buffer);
 
-   idct->viewport[1].scale[0] = idct->destination->width0;
-   idct->viewport[1].scale[1] = idct->destination->height0;
+   buffer->vertex_bufs.individual.quad.stride = idct->quad.stride;
+   buffer->vertex_bufs.individual.quad.max_index = idct->quad.max_index;
+   buffer->vertex_bufs.individual.quad.buffer_offset = idct->quad.buffer_offset;
+   pipe_resource_reference(&buffer->vertex_bufs.individual.quad.buffer, idct->quad.buffer);
 
-   idct->fb_state[0].width = idct->textures.individual.intermediate->width0;
-   idct->fb_state[0].height = idct->textures.individual.intermediate->height0;
+   buffer->vertex_bufs.individual.pos = vl_vb_create_buffer(idct->pipe, idct->max_blocks, idct->vertex_buffer_stride);
 
-   idct->fb_state[0].nr_cbufs = NR_RENDER_TARGETS;
-   for(i = 0; i < NR_RENDER_TARGETS; ++i) {
-      idct->fb_state[0].cbufs[i] = idct->pipe->screen->get_tex_surface(
-         idct->pipe->screen, idct->textures.individual.intermediate, 0, 0, i,
-         PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
-   }
-
-   idct->fb_state[1].width = idct->destination->width0;
-   idct->fb_state[1].height = idct->destination->height0;
-
-   idct->fb_state[1].nr_cbufs = 1;
-   idct->fb_state[1].cbufs[0] = idct->pipe->screen->get_tex_surface(
-      idct->pipe->screen, idct->destination, 0, 0, 0,
-      PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
-
-   for(i = 0; i < 2; ++i) {
-      idct->viewport[i].scale[2] = 1;
-      idct->viewport[i].scale[3] = 1;
-      idct->viewport[i].translate[0] = 0;
-      idct->viewport[i].translate[1] = 0;
-      idct->viewport[i].translate[2] = 0;
-      idct->viewport[i].translate[3] = 0;
-
-      idct->fb_state[i].zsbuf = NULL;
-   }
+   if(buffer->vertex_bufs.individual.pos.buffer == NULL)
+      return false;
 
-   for (i = 0; i < 4; ++i) {
-      memset(&sampler, 0, sizeof(sampler));
-      sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-      sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-      sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
-      sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
-      sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
-      sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
-      sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
-      sampler.compare_func = PIPE_FUNC_ALWAYS;
-      sampler.normalized_coords = 1;
-      /*sampler.shadow_ambient = ; */
-      /*sampler.lod_bias = ; */
-      sampler.min_lod = 0;
-      /*sampler.max_lod = ; */
-      /*sampler.border_color[0] = ; */
-      /*sampler.max_anisotropy = ; */
-      idct->samplers.all[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
-   }
+   if (!vl_vb_init(&buffer->blocks, idct->max_blocks, 2))
+      return false;
 
-   memset(&rs_state, 0, sizeof(rs_state));
-   /*rs_state.sprite_coord_enable */
-   rs_state.sprite_coord_mode = PIPE_SPRITE_COORD_UPPER_LEFT;
-   rs_state.point_quad_rasterization = true;
-   rs_state.point_size = BLOCK_WIDTH;
-   rs_state.gl_rasterization_rules = false;
-   idct->rs_state = idct->pipe->create_rasterizer_state(idct->pipe, &rs_state);
+   return true;
 }
 
 static void
-cleanup_state(struct vl_idct *idct)
+cleanup_vertex_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 {
-   unsigned i;
+   assert(idct && buffer);
 
-   for(i = 0; i < NR_RENDER_TARGETS; ++i) {
-      idct->pipe->screen->tex_surface_destroy(idct->fb_state[0].cbufs[i]);
-   }
+   pipe_resource_reference(&buffer->vertex_bufs.individual.quad.buffer, NULL);
+   pipe_resource_reference(&buffer->vertex_bufs.individual.pos.buffer, NULL);
 
-   idct->pipe->screen->tex_surface_destroy(idct->fb_state[1].cbufs[0]);
-
-   for (i = 0; i < 4; ++i)
-      idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]);
-
-   idct->pipe->delete_rasterizer_state(idct->pipe, idct->rs_state);
+   vl_vb_cleanup(&buffer->blocks);
 }
 
 struct pipe_resource *
@@ -549,85 +536,142 @@ vl_idct_upload_matrix(struct pipe_context *pipe)
    return matrix;
 }
 
-bool
-vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resource *dst, struct pipe_resource *matrix)
+bool vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, 
+                  unsigned buffer_width, unsigned buffer_height,
+                  struct pipe_resource *matrix)
 {
-   assert(idct && pipe && dst);
+   assert(idct && pipe && matrix);
 
    idct->pipe = pipe;
-   idct->buffer_width = dst->width0;
-   idct->buffer_height = dst->height0;
-
-   pipe_resource_reference(&idct->textures.individual.matrix, matrix);
-   pipe_resource_reference(&idct->textures.individual.transpose, matrix);
-   pipe_resource_reference(&idct->destination, dst);
+   idct->buffer_width = buffer_width;
+   idct->buffer_height = buffer_height;
+   pipe_resource_reference(&idct->matrix, matrix);
 
    idct->max_blocks =
-      align(idct->destination->width0, BLOCK_WIDTH) / BLOCK_WIDTH *
-      align(idct->destination->height0, BLOCK_HEIGHT) / BLOCK_HEIGHT *
-      idct->destination->depth0;
+      align(buffer_width, BLOCK_WIDTH) / BLOCK_WIDTH *
+      align(buffer_height, BLOCK_HEIGHT) / BLOCK_HEIGHT;
 
-   if(!init_buffers(idct))
+   if(!init_shaders(idct))
       return false;
 
-   if(!init_shaders(idct)) {
-      cleanup_buffers(idct);
+   if(!init_state(idct)) {
+      cleanup_shaders(idct);
       return false;
    }
 
-   if(!vl_vb_init(&idct->blocks, idct->max_blocks, 2)) {
-      cleanup_shaders(idct);
-      cleanup_buffers(idct);
+   return true;
+}
+
+void
+vl_idct_cleanup(struct vl_idct *idct)
+{
+   cleanup_shaders(idct);
+   cleanup_state(idct);
+
+   pipe_resource_reference(&idct->matrix, NULL);
+}
+
+bool
+vl_idct_init_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer, struct pipe_resource *dst)
+{
+   unsigned i;
+
+   assert(buffer);
+   assert(idct);
+   assert(dst);
+
+   pipe_resource_reference(&buffer->textures.individual.matrix, idct->matrix);
+   pipe_resource_reference(&buffer->textures.individual.transpose, idct->matrix);
+   pipe_resource_reference(&buffer->destination, dst);
+
+   if (!init_textures(idct, buffer))
       return false;
+
+   if (!init_vertex_buffers(idct, buffer))
+      return false;
+
+   /* init state */
+   buffer->viewport[0].scale[0] = buffer->textures.individual.intermediate->width0;
+   buffer->viewport[0].scale[1] = buffer->textures.individual.intermediate->height0;
+
+   buffer->viewport[1].scale[0] = buffer->destination->width0;
+   buffer->viewport[1].scale[1] = buffer->destination->height0;
+
+   buffer->fb_state[0].width = buffer->textures.individual.intermediate->width0;
+   buffer->fb_state[0].height = buffer->textures.individual.intermediate->height0;
+
+   buffer->fb_state[0].nr_cbufs = NR_RENDER_TARGETS;
+   for(i = 0; i < NR_RENDER_TARGETS; ++i) {
+      buffer->fb_state[0].cbufs[i] = idct->pipe->screen->get_tex_surface(
+         idct->pipe->screen, buffer->textures.individual.intermediate, 0, 0, i,
+         PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
    }
 
-   init_state(idct);
+   buffer->fb_state[1].width = buffer->destination->width0;
+   buffer->fb_state[1].height = buffer->destination->height0;
 
-   vl_idct_map_buffers(idct);
+   buffer->fb_state[1].nr_cbufs = 1;
+   buffer->fb_state[1].cbufs[0] = idct->pipe->screen->get_tex_surface(
+      idct->pipe->screen, buffer->destination, 0, 0, 0,
+      PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
+
+   for(i = 0; i < 2; ++i) {
+      buffer->viewport[i].scale[2] = 1;
+      buffer->viewport[i].scale[3] = 1;
+      buffer->viewport[i].translate[0] = 0;
+      buffer->viewport[i].translate[1] = 0;
+      buffer->viewport[i].translate[2] = 0;
+      buffer->viewport[i].translate[3] = 0;
+
+      buffer->fb_state[i].zsbuf = NULL;
+   }
 
    return true;
 }
 
 void
-vl_idct_cleanup(struct vl_idct *idct)
+vl_idct_cleanup_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 {
-   vl_idct_unmap_buffers(idct);
+   unsigned i;
 
-   vl_vb_cleanup(&idct->blocks);
-   cleanup_shaders(idct);
-   cleanup_buffers(idct);
+   assert(buffer);
 
-   cleanup_state(idct);
+   for(i = 0; i < NR_RENDER_TARGETS; ++i) {
+      idct->pipe->screen->tex_surface_destroy(buffer->fb_state[0].cbufs[i]);
+   }
+
+   idct->pipe->screen->tex_surface_destroy(buffer->fb_state[1].cbufs[0]);
 
-   pipe_resource_reference(&idct->destination, NULL);
+   cleanup_textures(idct, buffer);
+   cleanup_vertex_buffers(idct, buffer);
 }
 
 void
-vl_idct_map_buffers(struct vl_idct *idct)
+vl_idct_map_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 {
    assert(idct);
 
    struct pipe_box rect =
    {
       0, 0, 0,
-      idct->textures.individual.source->width0,
-      idct->textures.individual.source->height0,
+      buffer->textures.individual.source->width0,
+      buffer->textures.individual.source->height0,
       1
    };
 
-   idct->tex_transfer = idct->pipe->get_transfer
+   buffer->tex_transfer = idct->pipe->get_transfer
    (
-      idct->pipe, idct->textures.individual.source,
+      idct->pipe, buffer->textures.individual.source,
       u_subresource(0, 0),
       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
       &rect
    );
 
-   idct->texels = idct->pipe->transfer_map(idct->pipe, idct->tex_transfer);
+   buffer->texels = idct->pipe->transfer_map(idct->pipe, buffer->tex_transfer);
 }
 
 void
-vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block)
+vl_idct_add_block(struct vl_idct_buffer *buffer, unsigned x, unsigned y, short *block)
 {
    struct vertex2f v;
    unsigned tex_pitch;
@@ -635,30 +679,30 @@ vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block)
 
    unsigned i;
 
-   assert(idct);
+   assert(buffer);
 
-   tex_pitch = idct->tex_transfer->stride / sizeof(short);
-   texels = idct->texels + y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH;
+   tex_pitch = buffer->tex_transfer->stride / sizeof(short);
+   texels = buffer->texels + y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH;
 
    for (i = 0; i < BLOCK_HEIGHT; ++i)
       memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * sizeof(short));
 
    v.x = x;
    v.y = y;
-   vl_vb_add_block(&idct->blocks, (float*)&v);
+   vl_vb_add_block(&buffer->blocks, (float*)&v);
 }
 
 void
-vl_idct_unmap_buffers(struct vl_idct *idct)
+vl_idct_unmap_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 {
-   assert(idct);
+   assert(idct && buffer);
 
-   idct->pipe->transfer_unmap(idct->pipe, idct->tex_transfer);
-   idct->pipe->transfer_destroy(idct->pipe, idct->tex_transfer);
+   idct->pipe->transfer_unmap(idct->pipe, buffer->tex_transfer);
+   idct->pipe->transfer_destroy(idct->pipe, buffer->tex_transfer);
 }
 
 void
-vl_idct_flush(struct vl_idct *idct)
+vl_idct_flush(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 {
    struct pipe_transfer *vec_transfer;
    void *vectors;
@@ -669,34 +713,34 @@ vl_idct_flush(struct vl_idct *idct)
    vectors = pipe_buffer_map
    (
       idct->pipe,
-      idct->vertex_bufs.individual.pos.buffer,
+      buffer->vertex_bufs.individual.pos.buffer,
       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
       &vec_transfer
    );
 
-   num_verts = vl_vb_upload(&idct->blocks, vectors);
+   num_verts = vl_vb_upload(&buffer->blocks, vectors);
 
-   pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.pos.buffer, vec_transfer);
+   pipe_buffer_unmap(idct->pipe, buffer->vertex_bufs.individual.pos.buffer, vec_transfer);
 
    if(num_verts > 0) {
 
       idct->pipe->bind_rasterizer_state(idct->pipe, idct->rs_state);
-      idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
+      idct->pipe->set_vertex_buffers(idct->pipe, 2, buffer->vertex_bufs.all);
       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
       idct->pipe->bind_vs_state(idct->pipe, idct->vs);
 
       /* first stage */
-      idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[0]);
-      idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[0]);
-      idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[0]);
+      idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state[0]);
+      idct->pipe->set_viewport_state(idct->pipe, &buffer->viewport[0]);
+      idct->pipe->set_fragment_sampler_views(idct->pipe, 2, buffer->sampler_views.stage[0]);
       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[0]);
       idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs);
       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_verts);
 
       /* second stage */
-      idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[1]);
-      idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[1]);
-      idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[1]);
+      idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state[1]);
+      idct->pipe->set_viewport_state(idct->pipe, &buffer->viewport[1]);
+      idct->pipe->set_fragment_sampler_views(idct->pipe, 2, buffer->sampler_views.stage[1]);
       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[1]);
       idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs);
       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_verts);
diff --git a/src/gallium/auxiliary/vl/vl_idct.h b/src/gallium/auxiliary/vl/vl_idct.h
index 2e98f22bea2..6076bdec46c 100644
--- a/src/gallium/auxiliary/vl/vl_idct.h
+++ b/src/gallium/auxiliary/vl/vl_idct.h
@@ -31,10 +31,6 @@
 #include <pipe/p_state.h>
 #include "vl_vertex_buffers.h"
 
-struct vl_idct_buffer
-{
-};
-
 struct vl_idct
 {
    struct pipe_context *pipe;
@@ -44,13 +40,7 @@ struct vl_idct
 
    unsigned max_blocks;
 
-   struct pipe_viewport_state viewport[2];
-   struct pipe_framebuffer_state fb_state[2];
-
-   struct pipe_resource *destination;
-
    void *rs_state;
-
    void *vertex_elems_state;
 
    union
@@ -63,6 +53,22 @@ struct vl_idct
       } individual;
    } samplers;
 
+   void *vs;
+   void *matrix_fs, *transpose_fs;
+
+   struct pipe_resource *matrix;
+   struct pipe_vertex_buffer quad;
+
+   unsigned vertex_buffer_stride;
+};
+
+struct vl_idct_buffer
+{
+   struct pipe_viewport_state viewport[2];
+   struct pipe_framebuffer_state fb_state[2];
+
+   struct pipe_resource *destination;
+
    union
    {
       struct pipe_sampler_view *all[4];
@@ -73,9 +79,6 @@ struct vl_idct
       } individual;
    } sampler_views;
 
-   void *vs;
-   void *matrix_fs, *transpose_fs;
-
    union
    {
       struct pipe_resource *all[4];
@@ -100,16 +103,22 @@ struct vl_idct
 
 struct pipe_resource *vl_idct_upload_matrix(struct pipe_context *pipe);
 
-bool vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resource *dst, struct pipe_resource *matrix);
+bool vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, 
+                  unsigned buffer_width, unsigned buffer_height, 
+                  struct pipe_resource *matrix);
 
 void vl_idct_cleanup(struct vl_idct *idct);
 
-void vl_idct_map_buffers(struct vl_idct *idct);
+bool vl_idct_init_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer, struct pipe_resource *dst);
+
+void vl_idct_cleanup_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer);
+
+void vl_idct_map_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer);
 
-void vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block);
+void vl_idct_add_block(struct vl_idct_buffer *buffer, unsigned x, unsigned y, short *block);
 
-void vl_idct_unmap_buffers(struct vl_idct *idct);
+void vl_idct_unmap_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer);
 
-void vl_idct_flush(struct vl_idct *idct);
+void vl_idct_flush(struct vl_idct *idct, struct vl_idct_buffer *buffer);
 
 #endif
diff --git a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
index 5ee9836f293..28de8c9a81a 100644
--- a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
+++ b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
@@ -593,6 +593,7 @@ cleanup_pipe_state(struct vl_mpeg12_mc_renderer *r)
 static bool
 init_buffers(struct vl_mpeg12_mc_renderer *r)
 {
+   struct pipe_resource *idct_matrix;
    struct pipe_resource template;
    struct pipe_vertex_element vertex_elems[NUM_VS_INPUTS];
    struct pipe_sampler_view sampler_view;
@@ -602,7 +603,7 @@ init_buffers(struct vl_mpeg12_mc_renderer *r)
    const unsigned mbh =
       align(r->buffer_height, MACROBLOCK_HEIGHT) / MACROBLOCK_HEIGHT;
 
-   unsigned i;
+   unsigned i, stride;
 
    assert(r);
 
@@ -624,6 +625,17 @@ init_buffers(struct vl_mpeg12_mc_renderer *r)
 
    r->textures.individual.y = r->pipe->screen->resource_create(r->pipe->screen, &template);
 
+   if (!(idct_matrix = vl_idct_upload_matrix(r->pipe)))
+      return false;
+
+   if (!vl_idct_init(&r->idct_luma, r->pipe, r->buffer_width, r->buffer_height, idct_matrix))
+      return false;
+
+   if (!vl_idct_init_buffer(&r->idct_luma, &r->idct_y, r->textures.individual.y))
+      return false;
+
+   vl_idct_map_buffers(&r->idct_luma, &r->idct_y);
+
    if (r->chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420) {
       template.width0 = r->buffer_width / 2;
       template.height0 = r->buffer_height / 2;
@@ -636,6 +648,19 @@ init_buffers(struct vl_mpeg12_mc_renderer *r)
    r->textures.individual.cr =
       r->pipe->screen->resource_create(r->pipe->screen, &template);
 
+   if(!vl_idct_init(&r->idct_chroma, r->pipe, template.width0, template.height0, idct_matrix))
+      return false;
+
+   if (!vl_idct_init_buffer(&r->idct_chroma, &r->idct_cb, r->textures.individual.cb))
+      return false;
+
+   vl_idct_map_buffers(&r->idct_chroma, &r->idct_cb);
+
+   if (!vl_idct_init_buffer(&r->idct_chroma, &r->idct_cr, r->textures.individual.cr))
+      return false;
+
+   vl_idct_map_buffers(&r->idct_chroma, &r->idct_cr);
+
    for (i = 0; i < 3; ++i) {
       u_sampler_view_default_template(&sampler_view,
                                       r->textures.all[i],
@@ -645,7 +670,8 @@ init_buffers(struct vl_mpeg12_mc_renderer *r)
 
    memset(&vertex_elems, 0, sizeof(vertex_elems));
 
-   r->vertex_bufs.individual.quad = vl_vb_upload_quads(r->pipe, r->macroblocks_per_batch, &vertex_elems[VS_I_RECT]);
+   vertex_elems[VS_I_RECT] = vl_vb_get_quad_vertex_element();
+   r->vertex_bufs.individual.quad = vl_vb_upload_quads(r->pipe, r->macroblocks_per_batch);
 
    /* Position element */
    vertex_elems[VS_I_VPOS].src_format = PIPE_FORMAT_R32G32_FLOAT;
@@ -665,16 +691,17 @@ init_buffers(struct vl_mpeg12_mc_renderer *r)
    /* progressive=0.0f interlaced=1.0f */
    vertex_elems[VS_I_INTERLACED].src_format = PIPE_FORMAT_R32_FLOAT;
 
+   stride = vl_vb_element_helper(&vertex_elems[VS_I_VPOS], 6, 1);
+
    r->vertex_bufs.individual.pos = vl_vb_create_buffer(
-      r->pipe, r->macroblocks_per_batch,
-      &vertex_elems[VS_I_VPOS], 6, 1);
+      r->pipe, r->macroblocks_per_batch, stride);
 
    for (i = 0; i < 4; ++i) {
       /* motion vector 0..4 element */
       vertex_elems[VS_I_MV0 + i].src_format = PIPE_FORMAT_R32G32_FLOAT;
+      stride = vl_vb_element_helper(&vertex_elems[VS_I_MV0 + i], 1, i + 2);
       r->vertex_bufs.individual.mv[i] = vl_vb_create_buffer(
-         r->pipe, r->macroblocks_per_batch,
-         &vertex_elems[VS_I_MV0 + i], 1, i + 2);
+         r->pipe, r->macroblocks_per_batch, stride);
    }
 
    for(i = 0; i < VL_NUM_MACROBLOCK_TYPES; ++i)
@@ -698,6 +725,17 @@ cleanup_buffers(struct vl_mpeg12_mc_renderer *r)
 
    for(i = 0; i<VL_NUM_MACROBLOCK_TYPES; ++i)
       cleanup_mbtype_handler(r, i);
+
+   vl_idct_unmap_buffers(&r->idct_luma, &r->idct_y);
+   vl_idct_unmap_buffers(&r->idct_chroma, &r->idct_cb);
+   vl_idct_unmap_buffers(&r->idct_chroma, &r->idct_cr);
+
+   vl_idct_cleanup_buffer(&r->idct_luma, &r->idct_y);
+   vl_idct_cleanup_buffer(&r->idct_chroma, &r->idct_cb);
+   vl_idct_cleanup_buffer(&r->idct_chroma, &r->idct_cr);
+
+   vl_idct_cleanup(&r->idct_luma);
+   vl_idct_cleanup(&r->idct_chroma);
 }
 
 static enum VL_MACROBLOCK_TYPE
@@ -1048,8 +1086,6 @@ vl_mpeg12_mc_renderer_init(struct vl_mpeg12_mc_renderer *renderer,
                            enum pipe_video_chroma_format chroma_format,
                            enum VL_MPEG12_MC_RENDERER_BUFFER_MODE bufmode)
 {
-   struct pipe_resource *idct_matrix;
-
    assert(renderer);
    assert(pipe);
 
@@ -1080,30 +1116,8 @@ vl_mpeg12_mc_renderer_init(struct vl_mpeg12_mc_renderer *renderer,
    renderer->future = NULL;
    renderer->num_macroblocks = 0;
 
-   if(!(idct_matrix = vl_idct_upload_matrix(pipe)))
-      goto error_idct_matrix;
-
-   if(!vl_idct_init(&renderer->idct_y, pipe, renderer->textures.individual.y, idct_matrix))
-      goto error_idct_y;
-
-   if(!vl_idct_init(&renderer->idct_cr, pipe, renderer->textures.individual.cr, idct_matrix))
-      goto error_idct_cr;
-
-   if(!vl_idct_init(&renderer->idct_cb, pipe, renderer->textures.individual.cb, idct_matrix))
-      goto error_idct_cb;
-
    return true;
 
-error_idct_cb:
-   vl_idct_cleanup(&renderer->idct_cr);
-
-error_idct_cr:
-   vl_idct_cleanup(&renderer->idct_y);
-
-error_idct_y:
-error_idct_matrix:
-   cleanup_buffers(renderer);
-
 error_buffers:
    cleanup_pipe_state(renderer);
 
@@ -1117,10 +1131,6 @@ vl_mpeg12_mc_renderer_cleanup(struct vl_mpeg12_mc_renderer *renderer)
 {
    assert(renderer);
 
-   vl_idct_cleanup(&renderer->idct_y);
-   vl_idct_cleanup(&renderer->idct_cr);
-   vl_idct_cleanup(&renderer->idct_cb);
-
    util_delete_keymap(renderer->texview_map, renderer->pipe);
    cleanup_pipe_state(renderer);
    cleanup_buffers(renderer);
@@ -1141,27 +1151,12 @@ vl_mpeg12_mc_renderer_render_macroblocks(struct vl_mpeg12_mc_renderer
                                          *mpeg12_macroblocks,
                                          struct pipe_fence_handle **fence)
 {
-   bool new_surface = false;
-
    assert(renderer);
    assert(surface);
    assert(num_macroblocks);
    assert(mpeg12_macroblocks);
 
-   if (renderer->surface) {
-      if (surface != renderer->surface) {
-         new_surface = true;
-      } else {
-
-         /* If the surface we're rendering hasn't changed the ref frames shouldn't change. */
-         assert(renderer->past == past);
-         assert(renderer->future == future);
-      }
-   }
-   else
-      new_surface = true;
-
-   if (new_surface) {
+   if (surface != renderer->surface) {
       pipe_surface_reference(&renderer->surface, surface);
       pipe_surface_reference(&renderer->past, past);
       pipe_surface_reference(&renderer->future, future);
@@ -1203,13 +1198,13 @@ vl_mpeg12_mc_renderer_flush(struct vl_mpeg12_mc_renderer *renderer)
    if (renderer->num_macroblocks == 0)
       return;
 
-   vl_idct_unmap_buffers(&renderer->idct_y);
-   vl_idct_unmap_buffers(&renderer->idct_cr);
-   vl_idct_unmap_buffers(&renderer->idct_cb);
+   vl_idct_unmap_buffers(&renderer->idct_luma, &renderer->idct_y);
+   vl_idct_unmap_buffers(&renderer->idct_chroma, &renderer->idct_cr);
+   vl_idct_unmap_buffers(&renderer->idct_chroma, &renderer->idct_cb);
 
-   vl_idct_flush(&renderer->idct_y);
-   vl_idct_flush(&renderer->idct_cr);
-   vl_idct_flush(&renderer->idct_cb);
+   vl_idct_flush(&renderer->idct_luma, &renderer->idct_y);
+   vl_idct_flush(&renderer->idct_chroma, &renderer->idct_cr);
+   vl_idct_flush(&renderer->idct_chroma, &renderer->idct_cb);
 
    upload_vertex_stream(renderer, num_verts);
 
@@ -1225,9 +1220,9 @@ vl_mpeg12_mc_renderer_flush(struct vl_mpeg12_mc_renderer *renderer)
 
    renderer->pipe->flush(renderer->pipe, PIPE_FLUSH_RENDER_CACHE, renderer->fence);
 
-   vl_idct_map_buffers(&renderer->idct_y);
-   vl_idct_map_buffers(&renderer->idct_cr);
-   vl_idct_map_buffers(&renderer->idct_cb);
+   vl_idct_map_buffers(&renderer->idct_luma, &renderer->idct_y);
+   vl_idct_map_buffers(&renderer->idct_chroma, &renderer->idct_cr);
+   vl_idct_map_buffers(&renderer->idct_chroma, &renderer->idct_cb);
 
    renderer->num_macroblocks = 0;
 }
diff --git a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
index 6ce9cbe390c..f1c7a2a4012 100644
--- a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
+++ b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
@@ -80,7 +80,8 @@ struct vl_mpeg12_mc_renderer
    struct pipe_viewport_state viewport;
    struct pipe_framebuffer_state fb_state;
 
-   struct vl_idct idct_y, idct_cb, idct_cr;
+   struct vl_idct idct_luma, idct_chroma;
+   struct vl_idct_buffer idct_y, idct_cb, idct_cr;
 
    void *rs_state;
 
diff --git a/src/gallium/auxiliary/vl/vl_vertex_buffers.c b/src/gallium/auxiliary/vl/vl_vertex_buffers.c
index 4e0d704a23d..4647685053b 100644
--- a/src/gallium/auxiliary/vl/vl_vertex_buffers.c
+++ b/src/gallium/auxiliary/vl/vl_vertex_buffers.c
@@ -40,7 +40,7 @@ static const struct quadf const_quad = {
 };
 
 struct pipe_vertex_buffer
-vl_vb_upload_quads(struct pipe_context *pipe, unsigned max_blocks, struct pipe_vertex_element* element)
+vl_vb_upload_quads(struct pipe_context *pipe, unsigned max_blocks)
 {
    struct pipe_vertex_buffer quad;
    struct pipe_transfer *buf_transfer;
@@ -50,13 +50,6 @@ vl_vb_upload_quads(struct pipe_context *pipe, unsigned max_blocks, struct pipe_v
 
    assert(pipe);
    assert(max_blocks);
-   assert(element);
-
-   /* setup rectangle element */
-   element->src_offset = 0;
-   element->instance_divisor = 0;
-   element->vertex_buffer_index = 0;
-   element->src_format = PIPE_FORMAT_R32G32_FLOAT;
 
    /* create buffer */
    quad.stride = sizeof(struct vertex2f);
@@ -89,34 +82,56 @@ vl_vb_upload_quads(struct pipe_context *pipe, unsigned max_blocks, struct pipe_v
    return quad;
 }
 
+struct pipe_vertex_element
+vl_vb_get_quad_vertex_element()
+{
+   struct pipe_vertex_element element;
+
+   /* setup rectangle element */
+   element.src_offset = 0;
+   element.instance_divisor = 0;
+   element.vertex_buffer_index = 0;
+   element.src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   return element;
+}
+
 struct pipe_vertex_buffer
-vl_vb_create_buffer(struct pipe_context *pipe, unsigned max_blocks,
-                    struct pipe_vertex_element *elements, unsigned num_elements,
-                    unsigned vertex_buffer_index)
+vl_vb_create_buffer(struct pipe_context *pipe, unsigned max_blocks, unsigned stride)
 {
    struct pipe_vertex_buffer buf;
-   unsigned i, size = 0;
-
-   for ( i = 0; i < num_elements; ++i ) {
-      elements[i].src_offset = size;
-      elements[i].instance_divisor = 0;
-      elements[i].vertex_buffer_index = vertex_buffer_index;
-      size += util_format_get_blocksize(elements[i].src_format);
-   }
 
-   buf.stride = size;
+   buf.stride = stride;
    buf.max_index = 4 * max_blocks - 1;
    buf.buffer_offset = 0;
    buf.buffer = pipe_buffer_create
    (
       pipe->screen,
       PIPE_BIND_VERTEX_BUFFER,
-      size * 4 * max_blocks
+      stride * 4 * max_blocks
    );
 
    return buf;
 }
 
+unsigned
+vl_vb_element_helper(struct pipe_vertex_element* elements, unsigned num_elements,
+                              unsigned vertex_buffer_index)
+{
+   unsigned i, offset = 0;
+
+   assert(elements && num_elements);
+
+   for ( i = 0; i < num_elements; ++i ) {
+      elements[i].src_offset = offset;
+      elements[i].instance_divisor = 0;
+      elements[i].vertex_buffer_index = vertex_buffer_index;
+      offset += util_format_get_blocksize(elements[i].src_format);
+   }
+
+   return offset;
+}
+
 bool
 vl_vb_init(struct vl_vertex_buffer *buffer, unsigned max_blocks, unsigned num_elements)
 {
diff --git a/src/gallium/auxiliary/vl/vl_vertex_buffers.h b/src/gallium/auxiliary/vl/vl_vertex_buffers.h
index cf71f029a79..dae80fa767b 100644
--- a/src/gallium/auxiliary/vl/vl_vertex_buffers.h
+++ b/src/gallium/auxiliary/vl/vl_vertex_buffers.h
@@ -38,12 +38,14 @@ struct vl_vertex_buffer
    float *buffer;
 };
 
-struct pipe_vertex_buffer vl_vb_upload_quads(struct pipe_context *pipe, unsigned max_blocks,
-                                             struct pipe_vertex_element* element);
+struct pipe_vertex_buffer vl_vb_upload_quads(struct pipe_context *pipe, unsigned max_blocks);
 
-struct pipe_vertex_buffer vl_vb_create_buffer(struct pipe_context *pipe, unsigned max_blocks,
-                                              struct pipe_vertex_element* elements, unsigned num_elements,
-                                              unsigned vertex_buffer_index);
+struct pipe_vertex_element vl_vb_get_quad_vertex_element();
+
+struct pipe_vertex_buffer vl_vb_create_buffer(struct pipe_context *pipe, unsigned max_blocks, unsigned stride);
+
+unsigned vl_vb_element_helper(struct pipe_vertex_element* elements, unsigned num_elements,
+                              unsigned vertex_buffer_index);
 
 bool vl_vb_init(struct vl_vertex_buffer *buffer, unsigned max_blocks, unsigned num_elements);