[g3dvl] add skeleton and incomplete idct

5 files changed, 616 insertions, 76 deletions

diff --git a/src/gallium/auxiliary/Makefile b/src/gallium/auxiliary/Makefile
index 49ff1653e0e..07b3372c914 100644
--- a/src/gallium/auxiliary/Makefile
+++ b/src/gallium/auxiliary/Makefile
@@ -147,7 +147,8 @@ C_SOURCES = \
 	vl/vl_bitstream_parser.c \
 	vl/vl_mpeg12_mc_renderer.c \
 	vl/vl_compositor.c \
-	vl/vl_csc.c
+	vl/vl_csc.c \
+        vl/vl_idct.c
 
 GALLIVM_SOURCES = \
         gallivm/lp_bld_arit.c \
diff --git a/src/gallium/auxiliary/vl/vl_idct.c b/src/gallium/auxiliary/vl/vl_idct.c
new file mode 100644
index 00000000000..ce535ad3862
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_idct.c
@@ -0,0 +1,505 @@
+/**************************************************************************
+ *
+ * Copyright 2010 Christian König
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include "vl_idct.h"
+#include <assert.h>
+#include <pipe/p_context.h>
+#include <pipe/p_screen.h>
+#include <util/u_inlines.h>
+#include <util/u_sampler.h>
+#include <util/u_format.h>
+#include <tgsi/tgsi_ureg.h>
+#include "vl_types.h"
+
+#define BLOCK_WIDTH 8
+#define BLOCK_HEIGHT 8
+#define SCALE_FACTOR_16_TO_12 (32768.0f / 2048.0f)
+#define SCALE_FACTOR_9_TO_16 (256.0f / 32768.0f)
+
+struct vertex_shader_consts
+{
+   struct vertex4f norm;
+};
+
+enum VS_INPUT
+{
+   VS_I_RECT,
+   VS_I_VPOS,
+
+   NUM_VS_INPUTS
+};
+
+enum VS_OUTPUT
+{
+   VS_O_VPOS,
+   VS_O_BLOCK,
+   VS_O_TEX,
+   VS_O_START,
+   VS_O_STEP
+};
+
+
+const float const_matrix[8][8] = {
+   {  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.353553f,  0.3535530f },
+   {  0.4903930f,  0.4157350f,  0.2777850f,  0.0975451f, -0.0975452f, -0.2777850f, -0.415735f, -0.4903930f },
+   {  0.4619400f,  0.1913420f, -0.1913420f, -0.4619400f, -0.4619400f, -0.1913420f,  0.191342f,  0.4619400f },
+   {  0.4157350f, -0.0975452f, -0.4903930f, -0.2777850f,  0.2777850f,  0.4903930f,  0.097545f, -0.4157350f },
+   {  0.3535530f, -0.3535530f, -0.3535530f,  0.3535540f,  0.3535530f, -0.3535540f, -0.353553f,  0.3535530f },
+   {  0.2777850f, -0.4903930f,  0.0975452f,  0.4157350f, -0.4157350f, -0.0975451f,  0.490393f, -0.2777850f },
+   {  0.1913420f, -0.4619400f,  0.4619400f, -0.1913420f, -0.1913410f,  0.4619400f, -0.461940f,  0.1913420f },
+   {  0.0975451f, -0.2777850f,  0.4157350f, -0.4903930f,  0.4903930f, -0.4157350f,  0.277786f, -0.0975458f }
+};
+
+const float const_transpose[8][8] = {
+   {  0.3535530f,  0.4903930f,  0.4619400f,  0.4157350f,  0.3535530f,  0.2777850f,  0.191342f,  0.0975451f },
+   {  0.3535530f,  0.4157350f,  0.1913420f, -0.0975452f, -0.3535530f, -0.4903930f, -0.461940f, -0.2777850f },
+   {  0.3535530f,  0.2777850f, -0.1913420f, -0.4903930f, -0.3535530f,  0.0975452f,  0.461940f,  0.4157350f },
+   {  0.3535530f,  0.0975451f, -0.4619400f, -0.2777850f,  0.3535540f,  0.4157350f, -0.191342f, -0.4903930f },
+   {  0.3535530f, -0.0975452f, -0.4619400f,  0.2777850f,  0.3535530f, -0.4157350f, -0.191341f,  0.4903930f },
+   {  0.3535530f, -0.2777850f, -0.1913420f,  0.4903930f, -0.3535540f, -0.0975451f,  0.461940f, -0.4157350f },
+   {  0.3535530f, -0.4157350f,  0.1913420f,  0.0975450f, -0.3535530f,  0.4903930f, -0.461940f,  0.2777860f },
+   {  0.3535530f, -0.4903930f,  0.4619400f, -0.4157350f,  0.3535530f, -0.2777850f,  0.191342f, -0.0975458f }
+};
+
+static void *
+create_vert_shader(struct vl_idct *idct)
+{
+   struct ureg_program *shader;
+   struct ureg_src norm, bs;
+   struct ureg_src vrect, vpos;
+   struct ureg_dst scale, t_vpos;
+   struct ureg_dst o_vpos, o_block, o_tex, o_start, o_step;
+
+   shader = ureg_create(TGSI_PROCESSOR_VERTEX);
+   if (!shader)
+      return NULL;
+
+   norm = ureg_DECL_constant(shader, 0);
+   bs = ureg_imm2f(shader, BLOCK_WIDTH, BLOCK_HEIGHT);
+
+   scale = ureg_DECL_temporary(shader);
+   t_vpos = ureg_DECL_temporary(shader);
+
+   vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
+   vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
+
+   o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
+   o_block = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK);
+   o_tex = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_TEX);
+   o_start = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_START);
+   o_step = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_STEP);
+
+   /*
+    * scale = norm * mbs;
+    *
+    * t_vpos = vpos + vrect
+    * o_vpos.xy = t_vpos * scale
+    * o_vpos.zw = vpos
+    *
+    * o_block = vrect
+    * o_tex = t_pos
+    * o_start = vpos * scale
+    * o_step = norm
+    *
+    */
+   ureg_MUL(shader, ureg_writemask(scale, TGSI_WRITEMASK_XY), norm, bs);
+
+   ureg_ADD(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), vpos, vrect);
+   ureg_MUL(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos), ureg_src(scale));
+   ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
+   ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), vpos);
+
+   ureg_MOV(shader, ureg_writemask(o_tex, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
+   ureg_MOV(shader, ureg_writemask(o_block, TGSI_WRITEMASK_XY), vrect);
+   ureg_MUL(shader, ureg_writemask(o_start, TGSI_WRITEMASK_XY), vpos, ureg_src(scale));
+   ureg_MOV(shader, ureg_writemask(o_step, TGSI_WRITEMASK_XY), norm);
+
+   ureg_release_temporary(shader, t_vpos);
+   ureg_release_temporary(shader, scale);
+
+   ureg_END(shader);
+
+   return ureg_create_shader_and_destroy(shader, idct->pipe);
+}
+
+static void
+matrix_mul(struct ureg_program *shader, struct ureg_dst dst,
+           struct ureg_src tc[2], struct ureg_src sampler[2],
+           struct ureg_src start[2], struct ureg_src step[2],
+           float scale[2])
+{
+   struct ureg_dst t_tc[2], m[2][2], tmp[2];
+   unsigned i, j;
+
+   for(i = 0; i < 2; ++i) {
+      t_tc[i] = ureg_DECL_temporary(shader);
+      for(j = 0; j < 2; ++j)
+         m[i][j] = ureg_DECL_temporary(shader);
+      tmp[i] = ureg_DECL_temporary(shader);
+   }
+
+   /*
+    * m[0..1][0] = ?
+    * tmp[0..1] = dot4(m[0..1][0], m[0..1][1])
+    * fragment = tmp[0] + tmp[1]
+    */
+   ureg_MOV(shader, ureg_writemask(t_tc[0], TGSI_WRITEMASK_X), start[0]);
+   ureg_MOV(shader, ureg_writemask(t_tc[0], TGSI_WRITEMASK_Y), tc[0]);
+
+   ureg_MOV(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_X), tc[1]);
+   ureg_MOV(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_Y), start[1]);
+
+   for(i = 0; i < 2; ++i) {
+      for(j = 0; j < 4; ++j) {
+         /* Nouveau and r600g can't writemask tex dst regs (yet?), do in two steps */
+         ureg_TEX(shader, tmp[0], TGSI_TEXTURE_2D, ureg_src(t_tc[0]), sampler[0]);
+         ureg_MOV(shader, ureg_writemask(m[i][0], TGSI_WRITEMASK_X << j), ureg_scalar(ureg_src(tmp[0]), TGSI_SWIZZLE_X));
+
+         ureg_TEX(shader, tmp[1], TGSI_TEXTURE_2D, ureg_src(t_tc[1]), sampler[1]);
+         ureg_MOV(shader, ureg_writemask(m[i][1], TGSI_WRITEMASK_X << j), ureg_scalar(ureg_src(tmp[1]), TGSI_SWIZZLE_X));
+
+         ureg_ADD(shader, ureg_writemask(t_tc[0], TGSI_WRITEMASK_X), ureg_src(t_tc[0]), step[0]);
+         ureg_ADD(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_Y), ureg_src(t_tc[1]), step[1]);
+      }
+
+      if(scale[0] != 1.0f)
+         ureg_MUL(shader, m[i][0], ureg_src(m[i][0]), ureg_scalar(ureg_imm1f(shader, scale[0]), TGSI_SWIZZLE_X));
+
+      if(scale[1] != 1.0f)
+         ureg_MUL(shader, m[i][1], ureg_src(m[i][1]), ureg_scalar(ureg_imm1f(shader, scale[1]), TGSI_SWIZZLE_X));
+   }
+
+   ureg_DP4(shader, ureg_writemask(tmp[0], TGSI_WRITEMASK_X), ureg_src(m[0][0]), ureg_src(m[0][1]));
+   ureg_DP4(shader, ureg_writemask(tmp[1], TGSI_WRITEMASK_X), ureg_src(m[1][0]), ureg_src(m[1][1]));
+   ureg_ADD(shader, ureg_writemask(dst, TGSI_WRITEMASK_X), ureg_src(tmp[0]), ureg_src(tmp[1]));
+
+   for(i = 0; i < 2; ++i) {
+      ureg_release_temporary(shader, t_tc[i]);
+      for(j = 0; j < 2; ++j)
+         ureg_release_temporary(shader, m[i][j]);
+      ureg_release_temporary(shader, tmp[i]);
+   }
+}
+
+static void *
+create_transpose_frag_shader(struct vl_idct *idct)
+{
+   struct ureg_program *shader;
+   struct ureg_src tc[2], sampler[2];
+   struct ureg_src start[2], step[2];
+   struct ureg_dst fragment;
+   float scale[2];
+
+   shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
+   if (!shader)
+      return NULL;
+
+   tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
+   tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR);
+
+   start[0] = ureg_imm1f(shader, 0.0f);
+   start[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
+
+   step[0] = ureg_imm1f(shader, 1.0f / BLOCK_HEIGHT);
+   step[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_STEP, TGSI_INTERPOLATE_CONSTANT);
+
+   sampler[0] = ureg_DECL_sampler(shader, 0);
+   sampler[1] = ureg_DECL_sampler(shader, 1);
+
+   scale[0] = 1.0f;
+   scale[1] = SCALE_FACTOR_16_TO_12;
+
+   fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
+
+   matrix_mul(shader, fragment, tc, sampler, start, step, scale);
+
+   ureg_END(shader);
+
+   return ureg_create_shader_and_destroy(shader, idct->pipe);
+}
+
+static void *
+create_matrix_frag_shader(struct vl_idct *idct)
+{
+   struct ureg_program *shader;
+   struct ureg_src tc[2], sampler[2];
+   struct ureg_src start[2], step[2];
+   struct ureg_dst tmp, fragment;
+   float scale[2];
+
+   shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
+   if (!shader)
+      return NULL;
+
+   tmp = ureg_DECL_temporary(shader);
+
+   tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR);
+   tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
+
+   start[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
+   start[1] = ureg_imm1f(shader, 0.0f);
+
+   step[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_STEP, TGSI_INTERPOLATE_CONSTANT);
+   step[1] = ureg_imm1f(shader, 1.0f / BLOCK_WIDTH);
+
+   sampler[0] = ureg_DECL_sampler(shader, 0);
+   sampler[1] = ureg_DECL_sampler(shader, 1);
+
+   scale[0] = 1.0f;
+   scale[1] = 1.0f;
+
+   fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
+
+   matrix_mul(shader, tmp, tc, sampler, start, step, scale);
+   ureg_MUL(shader, fragment, ureg_src(tmp), ureg_scalar(ureg_imm1f(shader, SCALE_FACTOR_9_TO_16), TGSI_SWIZZLE_X));
+
+   ureg_END(shader);
+
+   return ureg_create_shader_and_destroy(shader, idct->pipe);
+}
+
+static void
+xfer_buffers_map(struct vl_idct *idct)
+{
+   struct pipe_box rect =
+   {
+      0, 0, 0,
+      idct->destination->width0,
+      idct->destination->height0,
+      1
+   };
+
+   idct->tex_transfer = idct->pipe->get_transfer
+   (
+      idct->pipe, idct->destination,
+      u_subresource(0, 0),
+      PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
+      &rect
+   );
+
+   idct->texels = idct->pipe->transfer_map(idct->pipe, idct->tex_transfer);
+}
+
+static void
+xfer_buffers_unmap(struct vl_idct *idct)
+{
+   idct->pipe->transfer_unmap(idct->pipe, idct->tex_transfer);
+   idct->pipe->transfer_destroy(idct->pipe, idct->tex_transfer);
+}
+
+static bool
+init_shaders(struct vl_idct *idct)
+{
+   assert(idct);
+
+   assert(idct->vs = create_vert_shader(idct));
+   assert(idct->transpose_fs = create_transpose_frag_shader(idct));
+   assert(idct->matrix_fs = create_matrix_frag_shader(idct));
+
+   return true;
+}
+
+static void
+cleanup_shaders(struct vl_idct *idct)
+{
+   assert(idct);
+
+   idct->pipe->delete_vs_state(idct->pipe, idct->vs);
+   idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs);
+   idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
+}
+
+static bool
+init_buffers(struct vl_idct *idct)
+{
+   struct pipe_resource template;
+   struct pipe_sampler_view sampler_view;
+   struct pipe_vertex_element vertex_elems[2];
+
+   const unsigned max_blocks =
+      align(idct->destination->width0, BLOCK_WIDTH) / BLOCK_WIDTH *
+      align(idct->destination->height0, BLOCK_HEIGHT) / BLOCK_HEIGHT *
+      idct->destination->depth0;
+
+   unsigned i;
+
+   memset(&template, 0, sizeof(struct pipe_resource));
+   template.target = PIPE_TEXTURE_2D;
+   template.format = PIPE_FORMAT_R32_FLOAT;
+   template.last_level = 0;
+   template.width0 = 8;
+   template.height0 = 8;
+   template.depth0 = 1;
+   template.usage = PIPE_USAGE_IMMUTABLE;
+   template.bind = PIPE_BIND_SAMPLER_VIEW;
+   template.flags = 0;
+
+   idct->textures.individual.matrix = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
+   idct->textures.individual.transpose = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
+
+   template.format = idct->destination->format;
+   template.width0 = idct->destination->width0;
+   template.height0 = idct->destination->height0;
+   template.depth0 = idct->destination->depth0;
+   template.usage = PIPE_USAGE_DYNAMIC;
+   idct->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
+
+   template.format = PIPE_FORMAT_R32_FLOAT;
+   template.usage = PIPE_USAGE_STATIC;
+   idct->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
+
+   for (i = 0; i < 4; ++i) {
+      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);
+   }
+
+   idct->quad.stride = sizeof(struct vertex2f);
+   idct->quad.max_index = 4 * max_blocks - 1;
+   idct->quad.buffer_offset = 0;
+   idct->quad.buffer = pipe_buffer_create
+   (
+      idct->pipe->screen,
+      PIPE_BIND_VERTEX_BUFFER,
+      sizeof(struct vertex2f) * 4 * max_blocks
+   );
+
+   idct->pos.stride = sizeof(struct vertex2f);
+   idct->pos.max_index = 4 * max_blocks - 1;
+   idct->pos.buffer_offset = 0;
+   idct->pos.buffer = pipe_buffer_create
+   (
+      idct->pipe->screen,
+      PIPE_BIND_VERTEX_BUFFER,
+      sizeof(struct vertex2f) * 4 * max_blocks
+   );
+
+   /* Rect element */
+   vertex_elems[0].src_offset = 0;
+   vertex_elems[0].instance_divisor = 0;
+   vertex_elems[0].vertex_buffer_index = 0;
+   vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Pos element */
+   vertex_elems[1].src_offset = 0;
+   vertex_elems[1].instance_divisor = 0;
+   vertex_elems[1].vertex_buffer_index = 1;
+   vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems);
+
+   idct->vs_const_buf = pipe_buffer_create
+   (
+      idct->pipe->screen,
+      PIPE_BIND_CONSTANT_BUFFER,
+      sizeof(struct vertex_shader_consts)
+   );
+
+   return true;
+}
+
+static void
+cleanup_buffers(struct vl_idct *idct)
+{
+   unsigned i;
+
+   assert(idct);
+
+   pipe_resource_reference(&idct->vs_const_buf, NULL);
+
+   for (i = 0; i < 4; ++i) {
+      pipe_sampler_view_reference(&idct->sampler_views.all[i], NULL);
+      pipe_resource_reference(&idct->textures.all[i], NULL);
+   }
+
+   idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
+   pipe_resource_reference(&idct->quad.buffer, NULL);
+   pipe_resource_reference(&idct->pos.buffer, NULL);
+}
+
+bool vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resource *dst)
+{
+   assert(idct && pipe && dst);
+
+   idct->pipe = pipe;
+
+   idct->viewport.scale[0] = dst->width0;
+   idct->viewport.scale[1] = dst->height0;
+   idct->viewport.scale[2] = 1;
+   idct->viewport.scale[3] = 1;
+   idct->viewport.translate[0] = 0;
+   idct->viewport.translate[1] = 0;
+   idct->viewport.translate[2] = 0;
+   idct->viewport.translate[3] = 0;
+
+   idct->fb_state.width = dst->width0;
+   idct->fb_state.height = dst->height0;
+   idct->fb_state.nr_cbufs = 1;
+   idct->fb_state.zsbuf = NULL;
+
+   pipe_resource_reference(&idct->destination, dst);
+
+   if(!init_shaders(idct))
+      return false;
+
+   if(!init_buffers(idct)) {
+      cleanup_shaders(idct);
+      return false;
+   }
+
+   xfer_buffers_map(idct);
+
+   return true;
+}
+
+void vl_idct_cleanup(struct vl_idct *idct)
+{
+   cleanup_shaders(idct);
+   cleanup_buffers(idct);
+}
+
+void vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block)
+{
+   unsigned tex_pitch;
+   short *texels;
+   unsigned i;
+
+   assert(idct);
+   assert(block);
+
+   tex_pitch = idct->tex_transfer->stride / util_format_get_blocksize(idct->tex_transfer->resource->format);
+   texels = idct->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 * 2);
+}
+
+void vl_idct_flush(struct vl_idct *idct)
+{
+   xfer_buffers_unmap(idct);
+   // TODO
+   xfer_buffers_map(idct);
+}
diff --git a/src/gallium/auxiliary/vl/vl_idct.h b/src/gallium/auxiliary/vl/vl_idct.h
new file mode 100644
index 00000000000..02a3250399f
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_idct.h
@@ -0,0 +1,90 @@
+/**************************************************************************
+ *
+ * Copyright 2010 Christian König
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#ifndef vl_idct_h
+#define vl_idct_h
+
+#include <pipe/p_state.h>
+
+struct vl_idct
+{
+   struct pipe_context *pipe;
+
+   struct pipe_viewport_state viewport;
+   struct pipe_resource *vs_const_buf;
+   struct pipe_framebuffer_state fb_state;
+
+   struct pipe_resource *destination;
+
+   void *vertex_elems_state;
+
+   union
+   {
+      void *all[4];
+      struct {
+         void *matrix, *transpose;
+         void *source, *intermediate;
+      } individual;
+   } samplers;
+
+   union
+   {
+      struct pipe_sampler_view *all[4];
+      struct {
+         struct pipe_sampler_view *matrix, *transpose;
+         struct pipe_sampler_view *source, *intermediate;
+      } individual;
+   } sampler_views;
+
+   void *vs;
+   void *transpose_fs, *matrix_fs;
+
+   union
+   {
+      struct pipe_resource *all[4];
+      struct {
+         struct pipe_resource *matrix, *transpose;
+         struct pipe_resource *source, *intermediate;
+      } individual;
+   } textures;
+
+   struct pipe_vertex_buffer quad;
+   struct pipe_vertex_buffer pos;
+
+   struct pipe_transfer *tex_transfer;
+   short *texels;
+};
+
+bool vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resource *dst);
+
+void vl_idct_cleanup(struct vl_idct *idct);
+
+void vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block);
+
+void vl_idct_flush(struct vl_idct *idct);
+
+#endif
diff --git a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
index 5c0404b4b07..7bc7ba91f94 100644
--- a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
+++ b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
@@ -572,47 +572,6 @@ create_field_bi_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
    return ureg_create_shader_and_destroy(shader, r->pipe);
 }
 
-static void
-xfer_buffers_map(struct vl_mpeg12_mc_renderer *r)
-{
-   unsigned i;
-
-   assert(r);
-
-   for (i = 0; i < 3; ++i) {
-      struct pipe_box rect =
-      {
-         0, 0, 0,
-         r->textures.all[i]->width0,
-         r->textures.all[i]->height0,
-         1
-      };
-
-      r->tex_transfer[i] = r->pipe->get_transfer
-      (
-         r->pipe, r->textures.all[i],
-         u_subresource(0, 0),
-         PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
-         &rect
-      );
-
-      r->texels[i] = r->pipe->transfer_map(r->pipe, r->tex_transfer[i]);
-   }
-}
-
-static void
-xfer_buffers_unmap(struct vl_mpeg12_mc_renderer *r)
-{
-   unsigned i;
-
-   assert(r);
-
-   for (i = 0; i < 3; ++i) {
-      r->pipe->transfer_unmap(r->pipe, r->tex_transfer[i]);
-      r->pipe->transfer_destroy(r->pipe, r->tex_transfer[i]);
-   }
-}
-
 static bool
 init_pipe_state(struct vl_mpeg12_mc_renderer *r)
 {
@@ -1209,7 +1168,10 @@ flush(struct vl_mpeg12_mc_renderer *r)
    assert(r);
    assert(r->num_macroblocks == r->macroblocks_per_batch);
 
-   xfer_buffers_unmap(r);
+   vl_idct_flush(&r->idct_y);
+   vl_idct_flush(&r->idct_cr);
+   vl_idct_flush(&r->idct_cb);
+
    gen_macroblock_stream(r, num_macroblocks);
 
    if (num_macroblocks[MACROBLOCK_TYPE_INTRA] > 0) {
@@ -1322,7 +1284,6 @@ flush(struct vl_mpeg12_mc_renderer *r)
    r->pipe->flush(r->pipe, PIPE_FLUSH_RENDER_CACHE, r->fence);
 
    r->num_macroblocks = 0;
-   xfer_buffers_map(r);
 }
 
 static void
@@ -1353,39 +1314,19 @@ update_render_target(struct vl_mpeg12_mc_renderer *r)
 }
 
 static void
-grab_coded_block(short *src, short *dst, unsigned dst_pitch)
-{
-   unsigned y;
-
-   assert(src);
-   assert(dst);
-
-   for (y = 0; y < BLOCK_HEIGHT; ++y)
-      memcpy(dst + y * dst_pitch, src + y * BLOCK_WIDTH, BLOCK_WIDTH * 2);
-}
-
-static void
 grab_blocks(struct vl_mpeg12_mc_renderer *r, unsigned mbx, unsigned mby,
             enum pipe_mpeg12_dct_type dct_type, unsigned cbp, short *blocks)
 {
-   unsigned tex_pitch;
-   short *texels;
    unsigned tb = 0, sb = 0;
-   unsigned mbpx = mbx * MACROBLOCK_WIDTH, mbpy = mby * MACROBLOCK_HEIGHT;
    unsigned x, y;
 
    assert(r);
    assert(blocks);
 
-   tex_pitch = r->tex_transfer[0]->stride / util_format_get_blocksize(r->tex_transfer[0]->resource->format);
-   texels = r->texels[0] + mbpy * tex_pitch + mbpx;
-
    for (y = 0; y < 2; ++y) {
       for (x = 0; x < 2; ++x, ++tb) {
          if ((cbp >> (5 - tb)) & 1) {
-            grab_coded_block(blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT,
-                             texels + y * tex_pitch * BLOCK_HEIGHT +
-                             x * BLOCK_WIDTH, tex_pitch);
+            vl_idct_add_block(&r->idct_y, mbx * 2 + x, mby * 2 + y, blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT);
             ++sb;
          }
       }
@@ -1394,15 +1335,12 @@ grab_blocks(struct vl_mpeg12_mc_renderer *r, unsigned mbx, unsigned mby,
    /* TODO: Implement 422, 444 */
    assert(r->chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420);
 
-   mbpx /= 2;
-   mbpy /= 2;
-
    for (tb = 0; tb < 2; ++tb) {
-      tex_pitch = r->tex_transfer[tb + 1]->stride / util_format_get_blocksize(r->tex_transfer[tb + 1]->resource->format);
-      texels = r->texels[tb + 1] + mbpy * tex_pitch + mbpx;
-
       if ((cbp >> (1 - tb)) & 1) {
-         grab_coded_block(blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT, texels, tex_pitch);
+         if(tb == 0)
+            vl_idct_add_block(&r->idct_cb, mbx, mby, blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT);
+         else
+            vl_idct_add_block(&r->idct_cr, mbx, mby, blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT);
          ++sb;
       }
    }
@@ -1499,7 +1437,9 @@ vl_mpeg12_mc_renderer_init(struct vl_mpeg12_mc_renderer *renderer,
    renderer->future = NULL;
    renderer->num_macroblocks = 0;
 
-   xfer_buffers_map(renderer);
+   vl_idct_init(&renderer->idct_y, pipe, renderer->textures.individual.y);
+   vl_idct_init(&renderer->idct_cr, pipe, renderer->textures.individual.cr);
+   vl_idct_init(&renderer->idct_cb, pipe, renderer->textures.individual.cb);
 
    return true;
 }
@@ -1509,7 +1449,9 @@ vl_mpeg12_mc_renderer_cleanup(struct vl_mpeg12_mc_renderer *renderer)
 {
    assert(renderer);
 
-   xfer_buffers_unmap(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);
diff --git a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
index 50c38f9ff10..ed48b5b6b45 100644
--- a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
+++ b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
@@ -32,6 +32,7 @@
 #include <pipe/p_state.h>
 #include <pipe/p_video_state.h>
 #include "vl_types.h"
+#include "vl_idct.h"
 
 struct pipe_context;
 struct pipe_macroblock;
@@ -57,6 +58,9 @@ struct vl_mpeg12_mc_renderer
    struct pipe_viewport_state viewport;
    struct pipe_resource *vs_const_buf;
    struct pipe_framebuffer_state fb_state;
+
+   struct vl_idct idct_y, idct_cb, idct_cr;
+
    union
    {
       void *all[3];
@@ -94,8 +98,6 @@ struct vl_mpeg12_mc_renderer
    struct pipe_fence_handle **fence;
    unsigned num_macroblocks;
    struct pipe_mpeg12_macroblock *macroblock_buf;
-   struct pipe_transfer *tex_transfer[3];
-   short *texels[3];
 
    struct keymap *texview_map;
 };