g3dvl: pipe_video_context interface, softpipe impl, auxiliary libs

24 files changed, 3561 insertions, 5 deletions

diff --git a/configs/default b/configs/default
index cb3ca1046f4..f1e2aa3ef5f 100644
--- a/configs/default
+++ b/configs/default
@@ -94,7 +94,7 @@ EGL_DRIVERS_DIRS = demo
 
 # Gallium directories and 
 GALLIUM_DIRS = auxiliary drivers state_trackers
-GALLIUM_AUXILIARY_DIRS = rbug draw translate cso_cache pipebuffer tgsi sct rtasm util indices
+GALLIUM_AUXILIARY_DIRS = rbug draw translate cso_cache pipebuffer tgsi sct rtasm util indices vl
 GALLIUM_AUXILIARIES = $(foreach DIR,$(GALLIUM_AUXILIARY_DIRS),$(TOP)/src/gallium/auxiliary/$(DIR)/lib$(DIR).a)
 GALLIUM_DRIVERS_DIRS = softpipe i915simple failover trace identity
 GALLIUM_DRIVERS = $(foreach DIR,$(GALLIUM_DRIVERS_DIRS),$(TOP)/src/gallium/drivers/$(DIR)/lib$(DIR).a)
diff --git a/configure.ac b/configure.ac
index 2881bb6bc25..143fd31a022 100644
--- a/configure.ac
+++ b/configure.ac
@@ -417,7 +417,7 @@ WINDOW_SYSTEM=""
 GALLIUM_DIRS="auxiliary drivers state_trackers"
 GALLIUM_WINSYS_DIRS=""
 GALLIUM_WINSYS_DRM_DIRS=""
-GALLIUM_AUXILIARY_DIRS="rbug draw translate cso_cache pipebuffer tgsi sct rtasm util indices"
+GALLIUM_AUXILIARY_DIRS="rbug draw translate cso_cache pipebuffer tgsi sct rtasm util indices vl"
 GALLIUM_DRIVERS_DIRS="softpipe failover trace identity"
 GALLIUM_STATE_TRACKERS_DIRS=""
 
diff --git a/src/gallium/SConscript b/src/gallium/SConscript
index 89c69d7205e..8be84cddbe7 100644
--- a/src/gallium/SConscript
+++ b/src/gallium/SConscript
@@ -23,6 +23,7 @@ SConscript([
 	'auxiliary/pipebuffer/SConscript',
 	'auxiliary/indices/SConscript',
 	'auxiliary/rbug/SConscript',
+	'auxiliary/vl/SConscript',
 ])
 
 for driver in env['drivers']:
diff --git a/src/gallium/auxiliary/vl/Makefile b/src/gallium/auxiliary/vl/Makefile
new file mode 100644
index 00000000000..71bfb937ad7
--- /dev/null
+++ b/src/gallium/auxiliary/vl/Makefile
@@ -0,0 +1,12 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = vl
+
+C_SOURCES = \
+	vl_bitstream_parser.c \
+	vl_mpeg12_mc_renderer.c \
+	vl_compositor.c \
+	vl_shader_build.c
+
+include ../../Makefile.template
diff --git a/src/gallium/auxiliary/vl/SConscript b/src/gallium/auxiliary/vl/SConscript
new file mode 100644
index 00000000000..eb50940c359
--- /dev/null
+++ b/src/gallium/auxiliary/vl/SConscript
@@ -0,0 +1,12 @@
+Import('*')
+
+vl = env.ConvenienceLibrary(
+	target = 'vl',
+	source = [
+		'vl_bitstream_parser.c',
+		'vl_mpeg12_mc_renderer.c',
+		'vl_compositor.c',
+		'vl_shader_build.c',
+	])
+
+auxiliaries.insert(0, vl)
diff --git a/src/gallium/auxiliary/vl/vl_bitstream_parser.c b/src/gallium/auxiliary/vl/vl_bitstream_parser.c
new file mode 100644
index 00000000000..356faa13481
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_bitstream_parser.c
@@ -0,0 +1,144 @@
+#include "vl_bitstream_parser.h"
+#include <assert.h>
+#include <limits.h>
+#include <util/u_memory.h>
+
+static unsigned
+grab_bits(unsigned cursor, unsigned how_many_bits, unsigned bitstream_elt)
+{
+   unsigned excess_bits = sizeof(unsigned) * CHAR_BIT - how_many_bits - cursor;
+	
+   assert(cursor < sizeof(unsigned) * CHAR_BIT);
+   assert(how_many_bits > 0 && how_many_bits <= sizeof(unsigned) * CHAR_BIT);
+   assert(cursor + how_many_bits <= sizeof(unsigned) * CHAR_BIT);
+
+   return (bitstream_elt << excess_bits) >> (excess_bits + cursor);
+}
+
+static unsigned
+show_bits(unsigned cursor, unsigned how_many_bits, const unsigned *bitstream)
+{	
+   unsigned cur_int = cursor / (sizeof(unsigned) * CHAR_BIT);
+   unsigned cur_bit = cursor % (sizeof(unsigned) * CHAR_BIT);
+	
+   assert(bitstream);
+	
+   if (cur_bit + how_many_bits > sizeof(unsigned) * CHAR_BIT)
+   {
+      return grab_bits(cur_bit, sizeof(unsigned) * CHAR_BIT - cur_bit,
+                       bitstream[cur_int]) |
+             grab_bits(0, cur_bit + how_many_bits - sizeof(unsigned) * CHAR_BIT,
+                       bitstream[cur_int + 1]) << (sizeof(unsigned) * CHAR_BIT - cur_bit);
+   }
+   else
+      return grab_bits(cur_bit, how_many_bits, bitstream[cur_int]);
+}
+
+bool vl_bitstream_parser_init(struct vl_bitstream_parser *parser,
+                              unsigned num_bitstreams,
+                              const void **bitstreams,
+                              const unsigned *sizes)
+{
+   assert(parser);
+   assert(num_bitstreams);
+   assert(bitstreams);
+   assert(sizes);
+
+   parser->num_bitstreams = num_bitstreams;
+   parser->bitstreams = (const unsigned**)bitstreams;
+   parser->sizes = sizes;
+   parser->cur_bitstream = 0;
+   parser->cursor = 0;
+
+   return true;
+}
+
+void vl_bitstream_parser_cleanup(struct vl_bitstream_parser *parser)
+{
+   assert(parser);
+}
+
+unsigned
+vl_bitstream_parser_get_bits(struct vl_bitstream_parser *parser,
+                             unsigned how_many_bits)
+{
+   unsigned bits;
+
+   assert(parser);
+
+   bits = vl_bitstream_parser_show_bits(parser, how_many_bits);
+
+   vl_bitstream_parser_forward(parser, how_many_bits);
+
+   return bits;
+}
+
+unsigned
+vl_bitstream_parser_show_bits(struct vl_bitstream_parser *parser,
+                              unsigned how_many_bits)
+{	
+   unsigned bits = 0;
+   unsigned shift = 0;
+   unsigned cursor;
+   unsigned cur_bitstream;
+
+   assert(parser);
+
+   cursor = parser->cursor;
+   cur_bitstream = parser->cur_bitstream;
+
+   while (1)
+   {
+      unsigned bits_left = parser->sizes[cur_bitstream] * CHAR_BIT - cursor;
+      unsigned bits_to_show = how_many_bits > bits_left ? bits_left : how_many_bits;
+
+      bits |= show_bits(cursor, bits_to_show,
+                        parser->bitstreams[cur_bitstream]) << shift;
+		
+      if (how_many_bits > bits_to_show)
+      {
+         how_many_bits -= bits_to_show;
+         cursor = 0;
+         ++cur_bitstream;
+         shift += bits_to_show;
+      }
+      else
+         break;
+   }
+
+   return bits;
+}
+
+void vl_bitstream_parser_forward(struct vl_bitstream_parser *parser,
+                                 unsigned how_many_bits)
+{
+   assert(parser);
+   assert(how_many_bits);
+
+   parser->cursor += how_many_bits;
+
+   while (parser->cursor > parser->sizes[parser->cur_bitstream] * CHAR_BIT)
+   {
+      parser->cursor -= parser->sizes[parser->cur_bitstream++] * CHAR_BIT;
+      assert(parser->cur_bitstream < parser->num_bitstreams);
+   }
+}
+
+void vl_bitstream_parser_rewind(struct vl_bitstream_parser *parser,
+                                unsigned how_many_bits)
+{
+   signed c;
+	
+   assert(parser);
+   assert(how_many_bits);
+	
+   c = parser->cursor - how_many_bits;
+
+   while (c < 0)
+   {
+      c += parser->sizes[parser->cur_bitstream--] * CHAR_BIT;
+      assert(parser->cur_bitstream < parser->num_bitstreams);
+   }
+
+   parser->cursor = (unsigned)c;
+}
diff --git a/src/gallium/auxiliary/vl/vl_bitstream_parser.h b/src/gallium/auxiliary/vl/vl_bitstream_parser.h
new file mode 100644
index 00000000000..46bebf470ff
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_bitstream_parser.h
@@ -0,0 +1,36 @@
+#ifndef vl_bitstream_parser_h
+#define vl_bitstream_parser_h
+
+#include <stdbool.h>
+
+struct vl_bitstream_parser
+{
+   unsigned num_bitstreams;
+   const unsigned **bitstreams;
+   const unsigned *sizes;
+   unsigned cur_bitstream;
+   unsigned cursor;
+};
+
+bool vl_bitstream_parser_init(struct vl_bitstream_parser *parser,
+                              unsigned num_bitstreams,
+                              const void **bitstreams,
+                              const unsigned *sizes);
+
+void vl_bitstream_parser_cleanup(struct vl_bitstream_parser *parser);
+
+unsigned
+vl_bitstream_parser_get_bits(struct vl_bitstream_parser *parser,
+                             unsigned how_many_bits);
+
+unsigned
+vl_bitstream_parser_show_bits(struct vl_bitstream_parser *parser,
+                              unsigned how_many_bits);
+
+void vl_bitstream_parser_forward(struct vl_bitstream_parser *parser,
+                                 unsigned how_many_bits);
+
+void vl_bitstream_parser_rewind(struct vl_bitstream_parser *parser,
+                                unsigned how_many_bits);
+
+#endif /* vl_bitstream_parser_h */
diff --git a/src/gallium/auxiliary/vl/vl_compositor.c b/src/gallium/auxiliary/vl/vl_compositor.c
new file mode 100644
index 00000000000..0894421c0b8
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_compositor.c
@@ -0,0 +1,590 @@
+#include "vl_compositor.h"
+#include <assert.h>
+#include <pipe/p_context.h>
+#include <pipe/p_inlines.h>
+#include <tgsi/tgsi_parse.h>
+#include <tgsi/tgsi_build.h>
+#include <util/u_memory.h>
+#include "vl_shader_build.h"
+
+struct vertex2f
+{
+   float x, y;
+};
+
+struct vertex4f
+{
+   float x, y, z, w;
+};
+
+struct vertex_shader_consts
+{
+   struct vertex4f dst_scale;
+   struct vertex4f dst_trans;
+   struct vertex4f src_scale;
+   struct vertex4f src_trans;
+};
+
+struct fragment_shader_consts
+{
+   struct vertex4f bias;
+   float matrix[16];
+};
+
+/*
+ * Represents 2 triangles in a strip in normalized coords.
+ * Used to render the surface onto the frame buffer.
+ */
+static const struct vertex2f surface_verts[4] =
+{
+   {0.0f, 0.0f},
+   {0.0f, 1.0f},
+   {1.0f, 0.0f},
+   {1.0f, 1.0f}
+};
+
+/*
+ * Represents texcoords for the above. We can use the position values directly.
+ * TODO: Duplicate these in the shader, no need to create a buffer.
+ */
+static const struct vertex2f *surface_texcoords = surface_verts;
+
+/*
+ * Identity color conversion constants, for debugging
+ */
+static const struct fragment_shader_consts identity =
+{
+   {
+      0.0f, 0.0f, 0.0f, 0.0f
+   },
+   {
+      1.0f, 0.0f, 0.0f, 0.0f,
+      0.0f, 1.0f, 0.0f, 0.0f,
+      0.0f, 0.0f, 1.0f, 0.0f,
+      0.0f, 0.0f, 0.0f, 1.0f
+   }
+};
+
+/*
+ * Converts ITU-R BT.601 YCbCr pixels to RGB pixels where:
+ * Y is in [16,235], Cb and Cr are in [16,240]
+ * R, G, and B are in [16,235]
+ */
+static const struct fragment_shader_consts bt_601 =
+{
+   {
+      0.0f, 0.501960784f, 0.501960784f,	0.0f
+   },
+   {
+      1.0f, 0.0f,   1.371f,   0.0f,
+      1.0f, -0.336f, -0.698f, 0.0f,
+      1.0f, 1.732f, 0.0f,     0.0f,
+      0.0f, 0.0f,   0.0f,     1.0f
+   }
+};
+
+/*
+ * Converts ITU-R BT.601 YCbCr pixels to RGB pixels where:
+ * Y is in [16,235], Cb and Cr are in [16,240]
+ * R, G, and B are in [0,255]
+ */
+static const struct fragment_shader_consts bt_601_full =
+{
+   {
+      0.062745098f, 0.501960784f, 0.501960784f,	0.0f
+   },
+   {
+      1.164f, 0.0f,    1.596f,  0.0f,
+      1.164f, -0.391f, -0.813f, 0.0f,
+      1.164f, 2.018f,  0.0f,    0.0f,
+      0.0f,   0.0f,    0.0f,    1.0f
+   }
+};
+
+/*
+ * Converts ITU-R BT.709 YCbCr pixels to RGB pixels where:
+ * Y is in [16,235], Cb and Cr are in [16,240]
+ * R, G, and B are in [16,235]
+ */
+static const struct fragment_shader_consts bt_709 =
+{
+   {
+      0.0f, 0.501960784f, 0.501960784f,	0.0f
+   },
+   {
+      1.0f, 0.0f,    1.540f,  0.0f,
+      1.0f, -0.183f, -0.459f, 0.0f,
+      1.0f, 1.816f,  0.0f,    0.0f,
+      0.0f, 0.0f,    0.0f,    1.0f
+   }
+};
+
+/*
+ * Converts ITU-R BT.709 YCbCr pixels to RGB pixels where:
+ * Y is in [16,235], Cb and Cr are in [16,240]
+ * R, G, and B are in [0,255]
+ */
+const struct fragment_shader_consts bt_709_full =
+{
+   {
+      0.062745098f, 0.501960784f, 0.501960784f,	0.0f
+   },
+   {
+      1.164f, 0.0f,    1.793f,  0.0f,
+      1.164f, -0.213f, -0.534f,	0.0f,
+      1.164f, 2.115f,  0.0f,    0.0f,
+      0.0f,   0.0f,    0.0f,    1.0f
+   }
+};
+
+static void
+create_vert_shader(struct vl_compositor *c)
+{
+   const unsigned max_tokens = 50;
+
+   struct pipe_shader_state vs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   assert(c);
+
+   tokens = (struct tgsi_token*)MALLOC(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version*)&tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header*)&tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
+
+   ti = 3;
+
+   /*
+    * decl i0		; Vertex pos
+    * decl i1		; Vertex texcoords
+    */
+   for (unsigned i = 0; i < 2; i++)
+   {
+      decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * decl c0		; Scaling vector to scale vertex pos rect to destination size
+    * decl c1		; Translation vector to move vertex pos rect into position
+    * decl c2		; Scaling vector to scale texcoord rect to source size
+    * decl c3		; Translation vector to move texcoord rect into position
+    */
+   decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 3);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * decl o0		; Vertex pos
+    * decl o1		; Vertex texcoords
+    */
+   for (unsigned i = 0; i < 2; i++)
+   {
+      decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* decl t0, t1 */
+   decl = vl_decl_temps(0, 1);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * mad o0, i0, c0, c1	; Scale and translate unit output rect to destination size and pos
+    * mad o1, i1, c2, c3	; Scale and translate unit texcoord rect to source size and pos
+    */
+   for (unsigned i = 0; i < 2; ++i)
+   {
+      inst = vl_inst4(TGSI_OPCODE_MAD, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i, TGSI_FILE_CONSTANT, i * 2, TGSI_FILE_CONSTANT, i * 2 + 1);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   assert(ti <= max_tokens);
+
+   vs.tokens = tokens;
+   c->vertex_shader = c->pipe->create_vs_state(c->pipe, &vs);
+   FREE(tokens);
+}
+
+static void
+create_frag_shader(struct vl_compositor *c)
+{
+   const unsigned max_tokens = 50;
+
+   struct pipe_shader_state fs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   assert(c);
+
+   tokens = (struct tgsi_token*)MALLOC(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version*)&tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header*)&tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
+
+   ti = 3;
+
+   /* decl i0		; Texcoords for s0 */
+   decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, 1, 0, 0, TGSI_INTERPOLATE_LINEAR);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * decl c0		; Bias vector for CSC
+    * decl c1-c4		; CSC matrix c1-c4
+    */
+   decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 4);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl o0		; Fragment color */
+   decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl t0 */
+   decl = vl_decl_temps(0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl s0		; Sampler for tex containing picture to display */
+   decl = vl_decl_samplers(0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* tex2d t0, i0, s0	; Read src pixel */
+   inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, 0, TGSI_FILE_SAMPLER, 0);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /* sub t0, t0, c0	; Subtract bias vector from pixel */
+   inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * dp4 o0.x, t0, c1	; Multiply pixel by the color conversion matrix
+    * dp4 o0.y, t0, c2
+    * dp4 o0.z, t0, c3
+    */
+   for (unsigned i = 0; i < 3; ++i)
+   {
+      inst = vl_inst3(TGSI_OPCODE_DP4, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, i + 1);
+      inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+	
+   assert(ti <= max_tokens);
+
+   fs.tokens = tokens;
+   c->fragment_shader = c->pipe->create_fs_state(c->pipe, &fs);
+   FREE(tokens);
+}
+
+static bool
+init_pipe_state(struct vl_compositor *c)
+{
+   struct pipe_sampler_state sampler;
+
+   assert(c);
+
+   c->fb_state.nr_cbufs = 1;
+   c->fb_state.zsbuf = NULL;
+
+   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_LINEAR;
+   sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
+   sampler.mag_img_filter = PIPE_TEX_FILTER_LINEAR;
+   sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
+   sampler.compare_func = PIPE_FUNC_ALWAYS;
+   sampler.normalized_coords = 1;
+   /*sampler.prefilter = ;*/
+   /*sampler.lod_bias = ;*/
+   /*sampler.min_lod = ;*/
+   /*sampler.max_lod = ;*/
+   /*sampler.border_color[i] = ;*/
+   /*sampler.max_anisotropy = ;*/
+   c->sampler = c->pipe->create_sampler_state(c->pipe, &sampler);
+	
+   return true;
+}
+
+static void cleanup_pipe_state(struct vl_compositor *c)
+{
+   assert(c);
+	
+   c->pipe->delete_sampler_state(c->pipe, c->sampler);
+}
+
+static bool
+init_shaders(struct vl_compositor *c)
+{
+   assert(c);
+
+   create_vert_shader(c);
+   create_frag_shader(c);
+
+   return true;
+}
+
+static void cleanup_shaders(struct vl_compositor *c)
+{
+   assert(c);
+	
+   c->pipe->delete_vs_state(c->pipe, c->vertex_shader);
+   c->pipe->delete_fs_state(c->pipe, c->fragment_shader);
+}
+
+static bool
+init_buffers(struct vl_compositor *c)
+{
+   assert(c);
+	
+   /*
+    * Create our vertex buffer and vertex buffer element
+    * VB contains 4 vertices that render a quad covering the entire window
+    * to display a rendered surface
+    * Quad is rendered as a tri strip
+    */
+   c->vertex_bufs[0].stride = sizeof(struct vertex2f);
+   c->vertex_bufs[0].max_index = 3;
+   c->vertex_bufs[0].buffer_offset = 0;
+   c->vertex_bufs[0].buffer = pipe_buffer_create
+   (
+   c->pipe->screen,
+   1,
+   PIPE_BUFFER_USAGE_VERTEX,
+   sizeof(struct vertex2f) * 4
+   );
+
+   memcpy
+   (
+      pipe_buffer_map(c->pipe->screen, c->vertex_bufs[0].buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
+      surface_verts,
+      sizeof(struct vertex2f) * 4
+   );
+
+   pipe_buffer_unmap(c->pipe->screen, c->vertex_bufs[0].buffer);
+
+   c->vertex_elems[0].src_offset = 0;
+   c->vertex_elems[0].vertex_buffer_index = 0;
+   c->vertex_elems[0].nr_components = 2;
+   c->vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /*
+    * Create our texcoord buffer and texcoord buffer element
+    * Texcoord buffer contains the TCs for mapping the rendered surface to the 4 vertices
+    */
+   c->vertex_bufs[1].stride = sizeof(struct vertex2f);
+   c->vertex_bufs[1].max_index = 3;
+   c->vertex_bufs[1].buffer_offset = 0;
+   c->vertex_bufs[1].buffer = pipe_buffer_create
+   (
+      c->pipe->screen,
+      1,
+      PIPE_BUFFER_USAGE_VERTEX,
+      sizeof(struct vertex2f) * 4
+   );
+
+   memcpy
+   (
+      pipe_buffer_map(c->pipe->screen, c->vertex_bufs[1].buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
+      surface_texcoords,
+      sizeof(struct vertex2f) * 4
+   );
+
+   pipe_buffer_unmap(c->pipe->screen, c->vertex_bufs[1].buffer);
+
+   c->vertex_elems[1].src_offset = 0;
+   c->vertex_elems[1].vertex_buffer_index = 1;
+   c->vertex_elems[1].nr_components = 2;
+   c->vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /*
+    * Create our vertex shader's constant buffer
+    * Const buffer contains scaling and translation vectors
+    */
+   c->vs_const_buf.buffer = pipe_buffer_create
+   (
+      c->pipe->screen,
+      1,
+      PIPE_BUFFER_USAGE_CONSTANT | PIPE_BUFFER_USAGE_DISCARD,
+      sizeof(struct vertex_shader_consts)
+   );
+
+   /*
+    * Create our fragment shader's constant buffer
+    * Const buffer contains the color conversion matrix and bias vectors
+    */
+   c->fs_const_buf.buffer = pipe_buffer_create
+   (
+      c->pipe->screen,
+      1,
+      PIPE_BUFFER_USAGE_CONSTANT,
+      sizeof(struct fragment_shader_consts)
+   );
+
+   /*
+    * TODO: Refactor this into a seperate function,
+    * allow changing the CSC matrix at runtime to switch between regular & full versions
+    */
+   memcpy
+   (
+      pipe_buffer_map(c->pipe->screen, c->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
+      &bt_601_full,
+      sizeof(struct fragment_shader_consts)
+   );
+
+   pipe_buffer_unmap(c->pipe->screen, c->fs_const_buf.buffer);
+
+   return true;
+}
+
+static void
+cleanup_buffers(struct vl_compositor *c)
+{
+   assert(c);
+	
+   for (unsigned i = 0; i < 2; ++i)
+      pipe_buffer_reference(&c->vertex_bufs[i].buffer, NULL);
+
+   pipe_buffer_reference(&c->vs_const_buf.buffer, NULL);
+   pipe_buffer_reference(&c->fs_const_buf.buffer, NULL);
+}
+
+bool vl_compositor_init(struct vl_compositor *compositor, struct pipe_context *pipe)
+{
+   assert(compositor);
+
+   memset(compositor, 0, sizeof(struct vl_compositor));
+
+   compositor->pipe = pipe;
+
+   if (!init_pipe_state(compositor))
+      return false;
+   if (!init_shaders(compositor))
+   {
+      cleanup_pipe_state(compositor);
+      return false;
+   }
+   if (!init_buffers(compositor))
+   {
+      cleanup_shaders(compositor);
+      cleanup_pipe_state(compositor);
+      return false;
+   }
+
+   return true;
+}
+
+void vl_compositor_cleanup(struct vl_compositor *compositor)
+{
+   assert(compositor);
+	
+   cleanup_buffers(compositor);
+   cleanup_shaders(compositor);
+   cleanup_pipe_state(compositor);
+}
+
+void vl_compositor_render(struct vl_compositor          *compositor,
+                          /*struct pipe_texture         *backround,
+                          struct pipe_video_rect        *backround_area,*/
+                          struct pipe_texture           *src_surface,
+                          enum pipe_mpeg12_picture_type picture_type,
+                          /*unsigned                    num_past_surfaces,
+                          struct pipe_texture           *past_surfaces,
+                          unsigned                      num_future_surfaces,
+                          struct pipe_texture           *future_surfaces,*/
+                          struct pipe_video_rect        *src_area,
+                          struct pipe_texture           *dst_surface,
+                          struct pipe_video_rect        *dst_area,
+                          /*unsigned                      num_layers,
+                          struct pipe_texture           *layers,
+                          struct pipe_video_rect        *layer_src_areas,
+                          struct pipe_video_rect        *layer_dst_areas*/
+                          struct pipe_fence_handle      **fence)
+{
+   struct vertex_shader_consts *vs_consts;
+
+   assert(compositor);
+   assert(src_surface);
+   assert(src_area);
+   assert(dst_surface);
+   assert(dst_area);
+   assert(picture_type == PIPE_MPEG12_PICTURE_TYPE_FRAME);
+
+   compositor->fb_state.width = dst_surface->width[0];
+   compositor->fb_state.height = dst_surface->height[0];
+   compositor->fb_state.cbufs[0] = compositor->pipe->screen->get_tex_surface
+   (
+      compositor->pipe->screen,
+      dst_surface,
+      0, 0, 0, PIPE_BUFFER_USAGE_GPU_READ | PIPE_BUFFER_USAGE_GPU_WRITE
+   );
+
+   compositor->viewport.scale[0] = compositor->fb_state.width;
+   compositor->viewport.scale[1] = compositor->fb_state.height;
+   compositor->viewport.scale[2] = 1;
+   compositor->viewport.scale[3] = 1;
+   compositor->viewport.translate[0] = 0;
+   compositor->viewport.translate[1] = 0;
+   compositor->viewport.translate[2] = 0;
+   compositor->viewport.translate[3] = 0;
+
+   compositor->pipe->set_framebuffer_state(compositor->pipe, &compositor->fb_state);
+   compositor->pipe->set_viewport_state(compositor->pipe, &compositor->viewport);
+   compositor->pipe->bind_sampler_states(compositor->pipe, 1, &compositor->sampler);
+   compositor->pipe->set_sampler_textures(compositor->pipe, 1, &src_surface);
+   compositor->pipe->bind_vs_state(compositor->pipe, compositor->vertex_shader);
+   compositor->pipe->bind_fs_state(compositor->pipe, compositor->fragment_shader);
+   compositor->pipe->set_vertex_buffers(compositor->pipe, 2, compositor->vertex_bufs);
+   compositor->pipe->set_vertex_elements(compositor->pipe, 2, compositor->vertex_elems);
+   compositor->pipe->set_constant_buffer(compositor->pipe, PIPE_SHADER_VERTEX, 0, &compositor->vs_const_buf);
+   compositor->pipe->set_constant_buffer(compositor->pipe, PIPE_SHADER_FRAGMENT, 0, &compositor->fs_const_buf);
+
+   vs_consts = pipe_buffer_map
+   (
+      compositor->pipe->screen,
+      compositor->vs_const_buf.buffer,
+      PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
+   );
+
+   vs_consts->dst_scale.x = dst_area->w / (float)compositor->fb_state.cbufs[0]->width;
+   vs_consts->dst_scale.y = dst_area->h / (float)compositor->fb_state.cbufs[0]->height;
+   vs_consts->dst_scale.z = 1;
+   vs_consts->dst_scale.w = 1;
+   vs_consts->dst_trans.x = dst_area->x / (float)compositor->fb_state.cbufs[0]->width;
+   vs_consts->dst_trans.y = dst_area->y / (float)compositor->fb_state.cbufs[0]->height;
+   vs_consts->dst_trans.z = 0;
+   vs_consts->dst_trans.w = 0;
+
+   vs_consts->src_scale.x = src_area->w / (float)src_surface->width[0];
+   vs_consts->src_scale.y = src_area->h / (float)src_surface->height[0];
+   vs_consts->src_scale.z = 1;
+   vs_consts->src_scale.w = 1;
+   vs_consts->src_trans.x = src_area->x / (float)src_surface->width[0];
+   vs_consts->src_trans.y = src_area->y / (float)src_surface->height[0];
+   vs_consts->src_trans.z = 0;
+   vs_consts->src_trans.w = 0;
+
+   pipe_buffer_unmap(compositor->pipe->screen, compositor->vs_const_buf.buffer);
+
+   compositor->pipe->draw_arrays(compositor->pipe, PIPE_PRIM_TRIANGLE_STRIP, 0, 4);
+   compositor->pipe->flush(compositor->pipe, PIPE_FLUSH_RENDER_CACHE, fence);
+
+   pipe_surface_reference(&compositor->fb_state.cbufs[0], NULL);
+}
diff --git a/src/gallium/auxiliary/vl/vl_compositor.h b/src/gallium/auxiliary/vl/vl_compositor.h
new file mode 100644
index 00000000000..2af41e1981f
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_compositor.h
@@ -0,0 +1,47 @@
+#ifndef vl_compositor_h
+#define vl_compositor_h
+
+#include <stdbool.h>
+#include <pipe/p_state.h>
+#include <pipe/p_video_state.h>
+
+struct pipe_context;
+struct pipe_texture;
+
+struct vl_compositor
+{
+   struct pipe_context *pipe;
+
+   struct pipe_framebuffer_state fb_state;
+   void *sampler;
+   void *vertex_shader;
+   void *fragment_shader;
+   struct pipe_viewport_state viewport;
+   struct pipe_vertex_buffer vertex_bufs[2];
+   struct pipe_vertex_element vertex_elems[2];
+   struct pipe_constant_buffer vs_const_buf, fs_const_buf;
+};
+
+bool vl_compositor_init(struct vl_compositor *compositor, struct pipe_context *pipe);
+
+void vl_compositor_cleanup(struct vl_compositor *compositor);
+
+void vl_compositor_render(struct vl_compositor          *compositor,
+                          /*struct pipe_texture         *backround,
+                          struct pipe_video_rect        *backround_area,*/
+                          struct pipe_texture           *src_surface,
+                          enum pipe_mpeg12_picture_type picture_type,
+                          /*unsigned                    num_past_surfaces,
+                          struct pipe_texture           *past_surfaces,
+                          unsigned                      num_future_surfaces,
+                          struct pipe_texture           *future_surfaces,*/
+                          struct pipe_video_rect        *src_area,
+                          struct pipe_texture           *dst_surface,
+                          struct pipe_video_rect        *dst_area,
+                          /*unsigned                      num_layers,
+                          struct pipe_texture           *layers,
+                          struct pipe_video_rect        *layer_src_areas,
+                          struct pipe_video_rect        *layer_dst_areas,*/
+                          struct pipe_fence_handle      **fence);
+
+#endif /* vl_compositor_h */
diff --git a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
new file mode 100644
index 00000000000..7e73c5ced9f
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
@@ -0,0 +1,1662 @@
+#include "vl_mpeg12_mc_renderer.h"
+#include <assert.h>
+#include <pipe/p_context.h>
+#include <pipe/p_inlines.h>
+#include <util/u_math.h>
+#include <util/u_memory.h>
+#include <tgsi/tgsi_parse.h>
+#include <tgsi/tgsi_build.h>
+#include "vl_shader_build.h"
+
+#define DEFAULT_BUF_ALIGNMENT 1
+#define MACROBLOCK_WIDTH 16
+#define MACROBLOCK_HEIGHT 16
+#define BLOCK_WIDTH 8
+#define BLOCK_HEIGHT 8
+#define ZERO_BLOCK_NIL -1.0f
+#define ZERO_BLOCK_IS_NIL(zb) ((zb).x < 0.0f)
+
+struct vertex2f
+{
+   float x, y;
+};
+
+struct vertex4f
+{
+   float x, y, z, w;
+};
+
+struct vertex_shader_consts
+{
+   struct vertex4f denorm;
+};
+
+struct fragment_shader_consts
+{
+   struct vertex4f multiplier;
+   struct vertex4f div;
+};
+
+/*
+ * Muliplier renormalizes block samples from 16 bits to 12 bits.
+ * Divider is used when calculating Y % 2 for choosing top or bottom
+ * field for P or B macroblocks.
+ * TODO: Use immediates.
+ */
+static const struct fragment_shader_consts fs_consts = {
+   {32767.0f / 255.0f, 32767.0f / 255.0f, 32767.0f / 255.0f, 0.0f},
+   {0.5f, 2.0f, 0.0f, 0.0f}
+};
+
+struct vert_stream_0
+{
+   struct vertex2f pos;
+   struct vertex2f luma_tc;
+   struct vertex2f cb_tc;
+   struct vertex2f cr_tc;
+};
+
+enum MACROBLOCK_TYPE
+{
+   MACROBLOCK_TYPE_INTRA,
+   MACROBLOCK_TYPE_FWD_FRAME_PRED,
+   MACROBLOCK_TYPE_FWD_FIELD_PRED,
+   MACROBLOCK_TYPE_BKWD_FRAME_PRED,
+   MACROBLOCK_TYPE_BKWD_FIELD_PRED,
+   MACROBLOCK_TYPE_BI_FRAME_PRED,
+   MACROBLOCK_TYPE_BI_FIELD_PRED,
+
+   NUM_MACROBLOCK_TYPES
+};
+
+static void
+create_intra_vert_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   const unsigned max_tokens = 50;
+
+   struct pipe_shader_state vs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   assert(r);
+
+   tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
+
+   ti = 3;
+
+   /*
+    * decl i0              ; Vertex pos
+    * decl i1              ; Luma texcoords
+    * decl i2              ; Chroma Cb texcoords
+    * decl i3              ; Chroma Cr texcoords
+    */
+   for (unsigned i = 0; i < 4; i++)
+   {
+      decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * decl o0              ; Vertex pos
+    * decl o1              ; Luma texcoords
+    * decl o2              ; Chroma Cb texcoords
+    * decl o3              ; Chroma Cr texcoords
+    */
+   for (unsigned i = 0; i < 4; i++)
+   {
+      decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * mov o0, i0           ; Move input vertex pos to output
+    * mov o1, i1           ; Move input luma texcoords to output
+    * mov o2, i2           ; Move input chroma Cb texcoords to output
+    * mov o3, i3           ; Move input chroma Cr texcoords to output
+    */
+   for (unsigned i = 0; i < 4; ++i)
+   {
+      inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   assert(ti <= max_tokens);
+
+   vs.tokens = tokens;
+   r->i_vs = r->pipe->create_vs_state(r->pipe, &vs);
+   free(tokens);
+}
+
+static void
+create_intra_frag_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   const unsigned max_tokens = 100;
+
+   struct pipe_shader_state fs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   assert(r);
+
+   tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
+
+   ti = 3;
+
+   /*
+    * decl i0                      ; Luma texcoords
+    * decl i1                      ; Chroma Cb texcoords
+    * decl i2                      ; Chroma Cr texcoords
+    */
+   for (unsigned i = 0; i < 3; ++i)
+   {
+      decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* decl c0                      ; Scaling factor, rescales 16-bit snorm to 9-bit snorm */
+   decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl o0                      ; Fragment color */
+   decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl t0, t1 */
+   decl = vl_decl_temps(0, 1);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * decl s0                      ; Sampler for luma texture
+    * decl s1                      ; Sampler for chroma Cb texture
+    * decl s2                      ; Sampler for chroma Cr texture
+    */
+   for (unsigned i = 0; i < 3; ++i)
+   {
+      decl = vl_decl_samplers(i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * tex2d t1, i0, s0             ; Read texel from luma texture
+    * mov t0.x, t1.x               ; Move luma sample into .x component
+    * tex2d t1, i1, s1             ; Read texel from chroma Cb texture
+    * mov t0.y, t1.x               ; Move Cb sample into .y component
+    * tex2d t1, i2, s2             ; Read texel from chroma Cr texture
+    * mov t0.z, t1.x               ; Move Cr sample into .z component
+    */
+   for (unsigned i = 0; i < 3; ++i)
+   {
+      inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+      inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
+      inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* mul o0, t0, c0               ; Rescale texel to correct range */
+   inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   assert(ti <= max_tokens);
+
+   fs.tokens = tokens;
+   r->i_fs = r->pipe->create_fs_state(r->pipe, &fs);
+   free(tokens);
+}
+
+static void
+create_frame_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   const unsigned max_tokens = 100;
+
+   struct pipe_shader_state vs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   assert(r);
+
+   tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
+
+   ti = 3;
+
+   /*
+    * decl i0              ; Vertex pos
+    * decl i1              ; Luma texcoords
+    * decl i2              ; Chroma Cb texcoords
+    * decl i3              ; Chroma Cr texcoords
+    * decl i4              ; Ref surface top field texcoords
+    * decl i5              ; Ref surface bottom field texcoords (unused, packed in the same stream)
+    */
+   for (unsigned i = 0; i < 6; i++)
+   {
+      decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * decl o0              ; Vertex pos
+    * decl o1              ; Luma texcoords
+    * decl o2              ; Chroma Cb texcoords
+    * decl o3              ; Chroma Cr texcoords
+    * decl o4              ; Ref macroblock texcoords
+    */
+   for (unsigned i = 0; i < 5; i++)
+   {
+      decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * mov o0, i0           ; Move input vertex pos to output
+    * mov o1, i1           ; Move input luma texcoords to output
+    * mov o2, i2           ; Move input chroma Cb texcoords to output
+    * mov o3, i3           ; Move input chroma Cr texcoords to output
+    */
+   for (unsigned i = 0; i < 4; ++i)
+   {
+        inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
+        ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* add o4, i0, i4       ; Translate vertex pos by motion vec to form ref macroblock texcoords */
+   inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 4, TGSI_FILE_INPUT, 0, TGSI_FILE_INPUT, 4);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   assert(ti <= max_tokens);
+
+   vs.tokens = tokens;
+   r->p_vs[0] = r->pipe->create_vs_state(r->pipe, &vs);
+   free(tokens);
+}
+
+static void
+create_field_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(false);
+}
+
+static void
+create_frame_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   const unsigned max_tokens = 100;
+
+   struct pipe_shader_state fs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   assert(r);
+
+   tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
+
+   ti = 3;
+
+   /*
+    * decl i0                      ; Luma texcoords
+    * decl i1                      ; Chroma Cb texcoords
+    * decl i2                      ; Chroma Cr texcoords
+    * decl i3                      ; Ref macroblock texcoords
+    */
+   for (unsigned i = 0; i < 4; ++i)
+   {
+      decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* decl c0                      ; Scaling factor, rescales 16-bit snorm to 9-bit snorm */
+   decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl o0                      ; Fragment color */
+   decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl t0, t1 */
+   decl = vl_decl_temps(0, 1);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * decl s0                      ; Sampler for luma texture
+    * decl s1                      ; Sampler for chroma Cb texture
+    * decl s2                      ; Sampler for chroma Cr texture
+    * decl s3                      ; Sampler for ref surface texture
+    */
+   for (unsigned i = 0; i < 4; ++i)
+   {
+      decl = vl_decl_samplers(i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * tex2d t1, i0, s0             ; Read texel from luma texture
+    * mov t0.x, t1.x               ; Move luma sample into .x component
+    * tex2d t1, i1, s1             ; Read texel from chroma Cb texture
+    * mov t0.y, t1.x               ; Move Cb sample into .y component
+    * tex2d t1, i2, s2             ; Read texel from chroma Cr texture
+    * mov t0.z, t1.x               ; Move Cr sample into .z component
+    */
+   for (unsigned i = 0; i < 3; ++i)
+   {
+      inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+      inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
+      inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* mul t0, t0, c0               ; Rescale texel to correct range */
+   inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /* tex2d t1, i3, s3             ; Read texel from ref macroblock */
+   inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, 3, TGSI_FILE_SAMPLER, 3);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /* add o0, t0, t1               ; Add ref and differential to form final output */
+   inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   assert(ti <= max_tokens);
+
+   fs.tokens = tokens;
+   r->p_fs[0] = r->pipe->create_fs_state(r->pipe, &fs);
+   free(tokens);
+}
+
+static void
+create_field_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(false);
+}
+
+static void
+create_frame_bi_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   const unsigned max_tokens = 100;
+
+   struct pipe_shader_state vs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   assert(r);
+
+   tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
+
+   ti = 3;
+
+   /*
+    * decl i0              ; Vertex pos
+    * decl i1              ; Luma texcoords
+    * decl i2              ; Chroma Cb texcoords
+    * decl i3              ; Chroma Cr texcoords
+    * decl i4              ; First ref macroblock top field texcoords
+    * decl i5              ; First ref macroblock bottom field texcoords (unused, packed in the same stream)
+    * decl i6              ; Second ref macroblock top field texcoords
+    * decl i7              ; Second ref macroblock bottom field texcoords (unused, packed in the same stream)
+    */
+   for (unsigned i = 0; i < 8; i++)
+   {
+      decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * decl o0              ; Vertex pos
+    * decl o1              ; Luma texcoords
+    * decl o2              ; Chroma Cb texcoords
+    * decl o3              ; Chroma Cr texcoords
+    * decl o4              ; First ref macroblock texcoords
+    * decl o5              ; Second ref macroblock texcoords
+    */
+   for (unsigned i = 0; i < 6; i++)
+   {
+      decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * mov o0, i0           ; Move input vertex pos to output
+    * mov o1, i1           ; Move input luma texcoords to output
+    * mov o2, i2           ; Move input chroma Cb texcoords to output
+    * mov o3, i3           ; Move input chroma Cr texcoords to output
+    */
+   for (unsigned i = 0; i < 4; ++i)
+   {
+      inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * add o4, i0, i4       ; Translate vertex pos by motion vec to form first ref macroblock texcoords
+    * add o5, i0, i6       ; Translate vertex pos by motion vec to form second ref macroblock texcoords
+    */
+   for (unsigned i = 0; i < 2; ++i)
+   {
+      inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, i + 4, TGSI_FILE_INPUT, 0, TGSI_FILE_INPUT, (i + 2) * 2);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   assert(ti <= max_tokens);
+
+   vs.tokens = tokens;
+   r->b_vs[0] = r->pipe->create_vs_state(r->pipe, &vs);
+   free(tokens);
+}
+
+static void
+create_field_bi_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(false);
+}
+
+static void
+create_frame_bi_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   const unsigned max_tokens = 100;
+
+   struct pipe_shader_state fs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   assert(r);
+
+   tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
+
+   ti = 3;
+
+   /*
+    * decl i0                      ; Luma texcoords
+    * decl i1                      ; Chroma Cb texcoords
+    * decl i2                      ; Chroma Cr texcoords
+    * decl i3                      ; First ref macroblock texcoords
+    * decl i4                      ; Second ref macroblock texcoords
+    */
+   for (unsigned i = 0; i < 5; ++i)
+   {
+      decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * decl c0                      ; Scaling factor, rescales 16-bit snorm to 9-bit snorm
+    * decl c1                      ; Constant 1/2 in .x channel to use as weight to blend past and future texels
+    */
+   decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 1);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl o0                      ; Fragment color */
+   decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl t0-t2 */
+   decl = vl_decl_temps(0, 2);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * decl s0                      ; Sampler for luma texture
+    * decl s1                      ; Sampler for chroma Cb texture
+    * decl s2                      ; Sampler for chroma Cr texture
+    * decl s3                      ; Sampler for first ref surface texture
+    * decl s4                      ; Sampler for second ref surface texture
+    */
+   for (unsigned i = 0; i < 5; ++i)
+   {
+      decl = vl_decl_samplers(i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * tex2d t1, i0, s0             ; Read texel from luma texture
+    * mov t0.x, t1.x               ; Move luma sample into .x component
+    * tex2d t1, i1, s1             ; Read texel from chroma Cb texture
+    * mov t0.y, t1.x               ; Move Cb sample into .y component
+    * tex2d t1, i2, s2             ; Read texel from chroma Cr texture
+    * mov t0.z, t1.x               ; Move Cr sample into .z component
+    */
+   for (unsigned i = 0; i < 3; ++i)
+   {
+      inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+      inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
+      inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* mul t0, t0, c0               ; Rescale texel to correct range */
+   inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * tex2d t1, i3, s3             ; Read texel from first ref macroblock
+    * tex2d t2, i4, s4             ; Read texel from second ref macroblock
+    */
+   for (unsigned i = 0; i < 2; ++i)
+   {
+      inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 1, TGSI_FILE_INPUT, i + 3, TGSI_FILE_SAMPLER, i + 3);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* lerp t1, c1.x, t1, t2        ; Blend past and future texels */
+   inst = vl_inst4(TGSI_OPCODE_LRP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
+   inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+   inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+   inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
+   inst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_X;
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /* add o0, t0, t1               ; Add past/future ref and differential to form final output */
+   inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   assert(ti <= max_tokens);
+
+   fs.tokens = tokens;
+   r->b_fs[0] = r->pipe->create_fs_state(r->pipe, &fs);
+   free(tokens);
+}
+
+static void
+create_field_bi_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(false);
+}
+
+static void
+xfer_buffers_map(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(r);
+
+   for (unsigned i = 0; i < 3; ++i)
+   {
+      r->tex_transfer[i] = r->pipe->screen->get_tex_transfer
+      (
+         r->pipe->screen, r->textures.all[i],
+         0, 0, 0, PIPE_TRANSFER_WRITE, 0, 0,
+         r->textures.all[i]->width[0], r->textures.all[i]->height[0]
+      );
+
+      r->texels[i] = r->pipe->screen->transfer_map(r->pipe->screen, r->tex_transfer[i]);
+   }
+}
+
+static void
+xfer_buffers_unmap(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(r);
+
+   for (unsigned i = 0; i < 3; ++i)
+   {
+      r->pipe->screen->transfer_unmap(r->pipe->screen, r->tex_transfer[i]);
+      r->pipe->screen->tex_transfer_destroy(r->tex_transfer[i]);
+   }
+}
+
+static bool
+init_pipe_state(struct vl_mpeg12_mc_renderer *r)
+{
+   struct pipe_sampler_state sampler;
+   unsigned filters[5];
+
+   assert(r);
+
+   r->viewport.scale[0] = r->pot_buffers ?
+      util_next_power_of_two(r->picture_width) : r->picture_width;
+   r->viewport.scale[1] = r->pot_buffers ?
+      util_next_power_of_two(r->picture_height) : r->picture_height;
+   r->viewport.scale[2] = 1;
+   r->viewport.scale[3] = 1;
+   r->viewport.translate[0] = 0;
+   r->viewport.translate[1] = 0;
+   r->viewport.translate[2] = 0;
+   r->viewport.translate[3] = 0;
+
+   r->fb_state.width = r->pot_buffers ?
+      util_next_power_of_two(r->picture_width) : r->picture_width;
+   r->fb_state.height = r->pot_buffers ?
+      util_next_power_of_two(r->picture_height) : r->picture_height;
+   r->fb_state.nr_cbufs = 1;
+   r->fb_state.zsbuf = NULL;
+
+   /* Luma filter */
+   filters[0] = PIPE_TEX_FILTER_NEAREST;
+   /* Chroma filters */
+   if (r->chroma_format == PIPE_VIDEO_CHROMA_FORMAT_444 ||
+       r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE)
+   {
+      filters[1] = PIPE_TEX_FILTER_NEAREST;
+      filters[2] = PIPE_TEX_FILTER_NEAREST;
+   }
+   else
+   {
+      filters[1] = PIPE_TEX_FILTER_LINEAR;
+      filters[2] = PIPE_TEX_FILTER_LINEAR;
+   }
+   /* Fwd, bkwd ref filters */
+   filters[3] = PIPE_TEX_FILTER_LINEAR;
+   filters[4] = PIPE_TEX_FILTER_LINEAR;
+
+   for (unsigned i = 0; i < 5; ++i)
+   {
+      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 = filters[i];
+      sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
+      sampler.mag_img_filter = filters[i];
+      sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
+      sampler.compare_func = PIPE_FUNC_ALWAYS;
+      sampler.normalized_coords = 1;
+      /*sampler.prefilter = ; */
+      /*sampler.shadow_ambient = ; */
+      /*sampler.lod_bias = ; */
+      sampler.min_lod = 0;
+      /*sampler.max_lod = ; */
+      /*sampler.border_color[i] = ; */
+      /*sampler.max_anisotropy = ; */
+      r->samplers.all[i] = r->pipe->create_sampler_state(r->pipe, &sampler);
+   }
+
+   return true;
+}
+
+static void
+cleanup_pipe_state(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(r);
+
+   for (unsigned i = 0; i < 5; ++i)
+      r->pipe->delete_sampler_state(r->pipe, r->samplers.all[i]);
+}
+
+static bool
+init_shaders(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(r);
+
+   create_intra_vert_shader(r);
+   create_intra_frag_shader(r);
+   create_frame_pred_vert_shader(r);
+   create_frame_pred_frag_shader(r);
+   create_frame_bi_pred_vert_shader(r);
+   create_frame_bi_pred_frag_shader(r);
+
+   return true;
+}
+
+static void
+cleanup_shaders(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(r);
+
+   r->pipe->delete_vs_state(r->pipe, r->i_vs);
+   r->pipe->delete_fs_state(r->pipe, r->i_fs);
+   r->pipe->delete_vs_state(r->pipe, r->p_vs[0]);
+   r->pipe->delete_fs_state(r->pipe, r->p_fs[0]);
+   r->pipe->delete_vs_state(r->pipe, r->b_vs[0]);
+   r->pipe->delete_fs_state(r->pipe, r->b_fs[0]);
+}
+
+static bool
+init_buffers(struct vl_mpeg12_mc_renderer *r)
+{
+   struct pipe_texture template;
+
+   const unsigned mbw =
+      align(r->picture_width, MACROBLOCK_WIDTH) / MACROBLOCK_WIDTH;
+   const unsigned mbh =
+      align(r->picture_height, MACROBLOCK_HEIGHT) / MACROBLOCK_HEIGHT;
+
+   assert(r);
+
+   r->macroblocks_per_batch =
+      mbw * (r->bufmode == VL_MPEG12_MC_RENDERER_BUFFER_PICTURE ? mbh : 1);
+   r->num_macroblocks = 0;
+   r->macroblock_buf = MALLOC(r->macroblocks_per_batch * sizeof(struct pipe_mpeg12_macroblock));
+
+   memset(&template, 0, sizeof(struct pipe_texture));
+   template.target = PIPE_TEXTURE_2D;
+   /* TODO: Accomodate HW that can't do this and also for cases when this isn't precise enough */
+   template.format = PIPE_FORMAT_R16_SNORM;
+   template.last_level = 0;
+   template.width[0] = r->pot_buffers ?
+      util_next_power_of_two(r->picture_width) : r->picture_width;
+   template.height[0] = r->pot_buffers ?
+      util_next_power_of_two(r->picture_height) : r->picture_height;
+   template.depth[0] = 1;
+   pf_get_block(template.format, &template.block);
+   template.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER | PIPE_TEXTURE_USAGE_DYNAMIC;
+
+   r->textures.individual.y = r->pipe->screen->texture_create(r->pipe->screen, &template);
+
+   if (r->chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420)
+   {
+      template.width[0] = r->pot_buffers ?
+         util_next_power_of_two(r->picture_width / 2) :
+         r->picture_width / 2;
+      template.height[0] = r->pot_buffers ?
+         util_next_power_of_two(r->picture_height / 2) :
+         r->picture_height / 2;
+   }
+   else if (r->chroma_format == PIPE_VIDEO_CHROMA_FORMAT_422)
+      template.height[0] = r->pot_buffers ?
+         util_next_power_of_two(r->picture_height / 2) :
+         r->picture_height / 2;
+
+   r->textures.individual.cb =
+      r->pipe->screen->texture_create(r->pipe->screen, &template);
+   r->textures.individual.cr =
+      r->pipe->screen->texture_create(r->pipe->screen, &template);
+
+   r->vertex_bufs.individual.ycbcr.stride = sizeof(struct vertex2f) * 4;
+   r->vertex_bufs.individual.ycbcr.max_index = 24 * r->macroblocks_per_batch - 1;
+   r->vertex_bufs.individual.ycbcr.buffer_offset = 0;
+   r->vertex_bufs.individual.ycbcr.buffer = pipe_buffer_create
+   (
+      r->pipe->screen,
+      DEFAULT_BUF_ALIGNMENT,
+      PIPE_BUFFER_USAGE_VERTEX | PIPE_BUFFER_USAGE_DISCARD,
+      sizeof(struct vertex2f) * 4 * 24 * r->macroblocks_per_batch
+   );
+
+   for (unsigned i = 1; i < 3; ++i)
+   {
+      r->vertex_bufs.all[i].stride = sizeof(struct vertex2f) * 2;
+      r->vertex_bufs.all[i].max_index = 24 * r->macroblocks_per_batch - 1;
+      r->vertex_bufs.all[i].buffer_offset = 0;
+      r->vertex_bufs.all[i].buffer = pipe_buffer_create
+      (
+         r->pipe->screen,
+         DEFAULT_BUF_ALIGNMENT,
+         PIPE_BUFFER_USAGE_VERTEX | PIPE_BUFFER_USAGE_DISCARD,
+         sizeof(struct vertex2f) * 2 * 24 * r->macroblocks_per_batch
+      );
+   }
+
+   /* Position element */
+   r->vertex_elems[0].src_offset = 0;
+   r->vertex_elems[0].vertex_buffer_index = 0;
+   r->vertex_elems[0].nr_components = 2;
+   r->vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Luma, texcoord element */
+   r->vertex_elems[1].src_offset = sizeof(struct vertex2f);
+   r->vertex_elems[1].vertex_buffer_index = 0;
+   r->vertex_elems[1].nr_components = 2;
+   r->vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Chroma Cr texcoord element */
+   r->vertex_elems[2].src_offset = sizeof(struct vertex2f) * 2;
+   r->vertex_elems[2].vertex_buffer_index = 0;
+   r->vertex_elems[2].nr_components = 2;
+   r->vertex_elems[2].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Chroma Cb texcoord element */
+   r->vertex_elems[3].src_offset = sizeof(struct vertex2f) * 3;
+   r->vertex_elems[3].vertex_buffer_index = 0;
+   r->vertex_elems[3].nr_components = 2;
+   r->vertex_elems[3].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* First ref surface top field texcoord element */
+   r->vertex_elems[4].src_offset = 0;
+   r->vertex_elems[4].vertex_buffer_index = 1;
+   r->vertex_elems[4].nr_components = 2;
+   r->vertex_elems[4].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* First ref surface bottom field texcoord element */
+   r->vertex_elems[5].src_offset = sizeof(struct vertex2f);
+   r->vertex_elems[5].vertex_buffer_index = 1;
+   r->vertex_elems[5].nr_components = 2;
+   r->vertex_elems[5].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Second ref surface top field texcoord element */
+   r->vertex_elems[6].src_offset = 0;
+   r->vertex_elems[6].vertex_buffer_index = 2;
+   r->vertex_elems[6].nr_components = 2;
+   r->vertex_elems[6].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Second ref surface bottom field texcoord element */
+   r->vertex_elems[7].src_offset = sizeof(struct vertex2f);
+   r->vertex_elems[7].vertex_buffer_index = 2;
+   r->vertex_elems[7].nr_components = 2;
+   r->vertex_elems[7].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   r->vs_const_buf.buffer = pipe_buffer_create
+   (
+      r->pipe->screen,
+      DEFAULT_BUF_ALIGNMENT,
+      PIPE_BUFFER_USAGE_CONSTANT | PIPE_BUFFER_USAGE_DISCARD,
+      sizeof(struct vertex_shader_consts)
+   );
+
+   r->fs_const_buf.buffer = pipe_buffer_create
+   (
+      r->pipe->screen,
+      DEFAULT_BUF_ALIGNMENT,
+      PIPE_BUFFER_USAGE_CONSTANT, sizeof(struct fragment_shader_consts)
+   );
+
+   memcpy
+   (
+      pipe_buffer_map(r->pipe->screen, r->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
+      &fs_consts, sizeof(struct fragment_shader_consts)
+   );
+
+   pipe_buffer_unmap(r->pipe->screen, r->fs_const_buf.buffer);
+
+   return true;
+}
+
+static void
+cleanup_buffers(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(r);
+
+   pipe_buffer_reference(&r->vs_const_buf.buffer, NULL);
+   pipe_buffer_reference(&r->fs_const_buf.buffer, NULL);
+
+   for (unsigned i = 0; i < 3; ++i)
+      pipe_buffer_reference(&r->vertex_bufs.all[i].buffer, NULL);
+
+   for (unsigned i = 0; i < 3; ++i)
+      pipe_texture_reference(&r->textures.all[i], NULL);
+
+   FREE(r->macroblock_buf);
+}
+
+static enum MACROBLOCK_TYPE
+get_macroblock_type(struct pipe_mpeg12_macroblock *mb)
+{
+   assert(mb);
+
+   switch (mb->mb_type)
+   {
+      case PIPE_MPEG12_MACROBLOCK_TYPE_INTRA:
+         return MACROBLOCK_TYPE_INTRA;
+      case PIPE_MPEG12_MACROBLOCK_TYPE_FWD:
+         return mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FRAME ?
+            MACROBLOCK_TYPE_FWD_FRAME_PRED : MACROBLOCK_TYPE_FWD_FIELD_PRED;
+      case PIPE_MPEG12_MACROBLOCK_TYPE_BKWD:
+         return mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FRAME ?
+            MACROBLOCK_TYPE_BKWD_FRAME_PRED : MACROBLOCK_TYPE_BKWD_FIELD_PRED;
+      case PIPE_MPEG12_MACROBLOCK_TYPE_BI:
+         return mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FRAME ?
+            MACROBLOCK_TYPE_BI_FRAME_PRED : MACROBLOCK_TYPE_BI_FIELD_PRED;
+      default:
+         assert(0);
+   }
+
+   /* Unreachable */
+   return -1;
+}
+
+/* XXX: One of these days this will have to be killed with fire */
+#define SET_BLOCK(vb, cbp, mbx, mby, unitx, unity, ofsx, ofsy, hx, hy, lm, cbm, crm, use_zb, zb)				\
+	do {															\
+	(vb)[0].pos.x = (mbx) * (unitx) + (ofsx);		(vb)[0].pos.y = (mby) * (unity) + (ofsy);			\
+	(vb)[1].pos.x = (mbx) * (unitx) + (ofsx);		(vb)[1].pos.y = (mby) * (unity) + (ofsy) + (hy);		\
+	(vb)[2].pos.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[2].pos.y = (mby) * (unity) + (ofsy);			\
+	(vb)[3].pos.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[3].pos.y = (mby) * (unity) + (ofsy);			\
+	(vb)[4].pos.x = (mbx) * (unitx) + (ofsx);		(vb)[4].pos.y = (mby) * (unity) + (ofsy) + (hy);		\
+	(vb)[5].pos.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[5].pos.y = (mby) * (unity) + (ofsy) + (hy);		\
+																\
+	if (!use_zb || (cbp) & (lm))												\
+	{															\
+		(vb)[0].luma_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[0].luma_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[1].luma_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[1].luma_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+		(vb)[2].luma_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[2].luma_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[3].luma_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[3].luma_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[4].luma_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[4].luma_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+		(vb)[5].luma_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[5].luma_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+	}															\
+	else															\
+	{															\
+		(vb)[0].luma_tc.x = (zb)[0].x;		(vb)[0].luma_tc.y = (zb)[0].y;						\
+		(vb)[1].luma_tc.x = (zb)[0].x;		(vb)[1].luma_tc.y = (zb)[0].y + (hy);					\
+		(vb)[2].luma_tc.x = (zb)[0].x + (hx);	(vb)[2].luma_tc.y = (zb)[0].y;						\
+		(vb)[3].luma_tc.x = (zb)[0].x + (hx);	(vb)[3].luma_tc.y = (zb)[0].y;						\
+		(vb)[4].luma_tc.x = (zb)[0].x;		(vb)[4].luma_tc.y = (zb)[0].y + (hy);					\
+		(vb)[5].luma_tc.x = (zb)[0].x + (hx);	(vb)[5].luma_tc.y = (zb)[0].y + (hy);					\
+	}															\
+																\
+	if (!use_zb || (cbp) & (cbm))												\
+	{															\
+		(vb)[0].cb_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[0].cb_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[1].cb_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[1].cb_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+		(vb)[2].cb_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[2].cb_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[3].cb_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[3].cb_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[4].cb_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[4].cb_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+		(vb)[5].cb_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[5].cb_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+	}															\
+	else															\
+	{															\
+		(vb)[0].cb_tc.x = (zb)[1].x;		(vb)[0].cb_tc.y = (zb)[1].y;						\
+		(vb)[1].cb_tc.x = (zb)[1].x;		(vb)[1].cb_tc.y = (zb)[1].y + (hy);					\
+		(vb)[2].cb_tc.x = (zb)[1].x + (hx);	(vb)[2].cb_tc.y = (zb)[1].y;						\
+		(vb)[3].cb_tc.x = (zb)[1].x + (hx);	(vb)[3].cb_tc.y = (zb)[1].y;						\
+		(vb)[4].cb_tc.x = (zb)[1].x;		(vb)[4].cb_tc.y = (zb)[1].y + (hy);					\
+		(vb)[5].cb_tc.x = (zb)[1].x + (hx);	(vb)[5].cb_tc.y = (zb)[1].y + (hy);					\
+	}															\
+																\
+	if (!use_zb || (cbp) & (crm))												\
+	{															\
+		(vb)[0].cr_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[0].cr_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[1].cr_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[1].cr_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+		(vb)[2].cr_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[2].cr_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[3].cr_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[3].cr_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[4].cr_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[4].cr_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+		(vb)[5].cr_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[5].cr_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+	}															\
+	else															\
+	{															\
+		(vb)[0].cr_tc.x = (zb)[2].x;		(vb)[0].cr_tc.y = (zb)[2].y;						\
+		(vb)[1].cr_tc.x = (zb)[2].x;		(vb)[1].cr_tc.y = (zb)[2].y + (hy);					\
+		(vb)[2].cr_tc.x = (zb)[2].x + (hx);	(vb)[2].cr_tc.y = (zb)[2].y;						\
+		(vb)[3].cr_tc.x = (zb)[2].x + (hx);	(vb)[3].cr_tc.y = (zb)[2].y;						\
+		(vb)[4].cr_tc.x = (zb)[2].x;		(vb)[4].cr_tc.y = (zb)[2].y + (hy);					\
+		(vb)[5].cr_tc.x = (zb)[2].x + (hx);	(vb)[5].cr_tc.y = (zb)[2].y + (hy);					\
+	}															\
+	} while (0)
+
+static void
+gen_macroblock_verts(struct vl_mpeg12_mc_renderer *r,
+                     struct pipe_mpeg12_macroblock *mb, unsigned pos,
+                     struct vert_stream_0 *ycbcr_vb, struct vertex2f **ref_vb)
+{
+   struct vertex2f mo_vec[2];
+
+   assert(r);
+   assert(mb);
+   assert(ycbcr_vb);
+   assert(pos < r->macroblocks_per_batch);
+
+   switch (mb->mb_type)
+   {
+      case PIPE_MPEG12_MACROBLOCK_TYPE_BI:
+      {
+         struct vertex2f *vb;
+
+         assert(ref_vb && ref_vb[1]);
+
+         vb = ref_vb[1] + pos * 2 * 24;
+
+         mo_vec[0].x = mb->pmv[0][1][0] * 0.5f * r->surface_tex_inv_size.x;
+         mo_vec[0].y = mb->pmv[0][1][1] * 0.5f * r->surface_tex_inv_size.y;
+
+         if (mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FRAME)
+         {
+            for (unsigned i = 0; i < 24 * 2; i += 2)
+            {
+               vb[i].x = mo_vec[0].x;
+               vb[i].y = mo_vec[0].y;
+            }
+         }
+         else
+         {
+            mo_vec[1].x = mb->pmv[1][1][0] * 0.5f * r->surface_tex_inv_size.x;
+            mo_vec[1].y = mb->pmv[1][1][1] * 0.5f * r->surface_tex_inv_size.y;
+
+            for (unsigned i = 0; i < 24 * 2; i += 2)
+            {
+               vb[i].x = mo_vec[0].x;
+               vb[i].y = mo_vec[0].y;
+               vb[i + 1].x = mo_vec[1].x;
+               vb[i + 1].y = mo_vec[1].y;
+            }
+         }
+
+         /* fall-through */
+      }
+      case PIPE_MPEG12_MACROBLOCK_TYPE_FWD:
+      case PIPE_MPEG12_MACROBLOCK_TYPE_BKWD:
+      {
+         struct vertex2f *vb;
+
+         assert(ref_vb && ref_vb[0]);
+
+         vb = ref_vb[0] + pos * 2 * 24;
+
+         if (mb->mb_type == PIPE_MPEG12_MACROBLOCK_TYPE_BKWD)
+         {
+             mo_vec[0].x = mb->pmv[0][1][0] * 0.5f * r->surface_tex_inv_size.x;
+             mo_vec[0].y = mb->pmv[0][1][1] * 0.5f * r->surface_tex_inv_size.y;
+
+             if (mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FIELD)
+             {
+                mo_vec[1].x = mb->pmv[1][1][0] * 0.5f * r->surface_tex_inv_size.x;
+                mo_vec[1].y = mb->pmv[1][1][1] * 0.5f * r->surface_tex_inv_size.y;
+             }
+         }
+         else
+         {
+            mo_vec[0].x = mb->pmv[0][0][0] * 0.5f * r->surface_tex_inv_size.x;
+            mo_vec[0].y = mb->pmv[0][0][1] * 0.5f * r->surface_tex_inv_size.y;
+
+            if (mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FIELD)
+            {
+               mo_vec[1].x = mb->pmv[1][0][0] * 0.5f * r->surface_tex_inv_size.x;
+               mo_vec[1].y = mb->pmv[1][0][1] * 0.5f * r->surface_tex_inv_size.y;
+            }
+         }
+
+         if (mb->mb_type == PIPE_MPEG12_MOTION_TYPE_FRAME)
+         {
+            for (unsigned i = 0; i < 24 * 2; i += 2)
+            {
+               vb[i].x = mo_vec[0].x;
+               vb[i].y = mo_vec[0].y;
+            }
+         }
+         else
+         {
+            for (unsigned i = 0; i < 24 * 2; i += 2)
+            {
+               vb[i].x = mo_vec[0].x;
+               vb[i].y = mo_vec[0].y;
+               vb[i + 1].x = mo_vec[1].x;
+               vb[i + 1].y = mo_vec[1].y;
+            }
+         }
+
+         /* fall-through */
+      }
+      case PIPE_MPEG12_MACROBLOCK_TYPE_INTRA:
+      {
+         const struct vertex2f unit =
+         {
+            r->surface_tex_inv_size.x * MACROBLOCK_WIDTH,
+            r->surface_tex_inv_size.y * MACROBLOCK_HEIGHT
+         };
+         const struct vertex2f half =
+         {
+            r->surface_tex_inv_size.x * (MACROBLOCK_WIDTH / 2),
+            r->surface_tex_inv_size.y * (MACROBLOCK_HEIGHT / 2)
+         };
+         const bool use_zb = r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE;
+
+         struct vert_stream_0 *vb = ycbcr_vb + pos * 24;
+
+         SET_BLOCK(vb, mb->cbp, mb->mbx, mb->mby,
+                   unit.x, unit.y, 0, 0, half.x, half.y,
+                   32, 2, 1, use_zb, r->zero_block);
+
+         SET_BLOCK(vb + 6, mb->cbp, mb->mbx, mb->mby,
+                   unit.x, unit.y, half.x, 0, half.x, half.y,
+                   16, 2, 1, use_zb, r->zero_block);
+
+         SET_BLOCK(vb + 12, mb->cbp, mb->mbx, mb->mby,
+                   unit.x, unit.y, 0, half.y, half.x, half.y,
+                   8, 2, 1, use_zb, r->zero_block);
+
+         SET_BLOCK(vb + 18, mb->cbp, mb->mbx, mb->mby,
+                   unit.x, unit.y, half.x, half.y, half.x, half.y,
+                   4, 2, 1, use_zb, r->zero_block);
+
+         break;
+      }
+      default:
+         assert(0);
+   }
+}
+
+static void
+gen_macroblock_stream(struct vl_mpeg12_mc_renderer *r,
+                      unsigned *num_macroblocks)
+{
+   unsigned offset[NUM_MACROBLOCK_TYPES];
+   struct vert_stream_0 *ycbcr_vb;
+   struct vertex2f *ref_vb[2];
+
+   assert(r);
+   assert(num_macroblocks);
+
+   for (unsigned i = 0; i < r->num_macroblocks; ++i)
+   {
+      enum MACROBLOCK_TYPE mb_type = get_macroblock_type(&r->macroblock_buf[i]);
+      ++num_macroblocks[mb_type];
+   }
+
+   offset[0] = 0;
+
+   for (unsigned i = 1; i < NUM_MACROBLOCK_TYPES; ++i)
+      offset[i] = offset[i - 1] + num_macroblocks[i - 1];
+
+   ycbcr_vb = (struct vert_stream_0 *)pipe_buffer_map
+   (
+      r->pipe->screen,
+      r->vertex_bufs.individual.ycbcr.buffer,
+      PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
+   );
+
+   for (unsigned i = 0; i < 2; ++i)
+      ref_vb[i] = (struct vertex2f *)pipe_buffer_map
+      (
+         r->pipe->screen,
+         r->vertex_bufs.individual.ref[i].buffer,
+         PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
+      );
+
+   for (unsigned i = 0; i < r->num_macroblocks; ++i)
+   {
+      enum MACROBLOCK_TYPE mb_type = get_macroblock_type(&r->macroblock_buf[i]);
+
+      gen_macroblock_verts(r, &r->macroblock_buf[i], offset[mb_type],
+                           ycbcr_vb, ref_vb);
+
+      ++offset[mb_type];
+   }
+
+   pipe_buffer_unmap(r->pipe->screen, r->vertex_bufs.individual.ycbcr.buffer);
+   for (unsigned i = 0; i < 2; ++i)
+      pipe_buffer_unmap(r->pipe->screen, r->vertex_bufs.individual.ref[i].buffer);
+}
+
+static void
+flush(struct vl_mpeg12_mc_renderer *r)
+{
+   unsigned num_macroblocks[NUM_MACROBLOCK_TYPES] = { 0 };
+   unsigned vb_start = 0;
+   struct vertex_shader_consts *vs_consts;
+
+   assert(r);
+   assert(r->num_macroblocks == r->macroblocks_per_batch);
+
+   gen_macroblock_stream(r, num_macroblocks);
+
+   r->fb_state.cbufs[0] = r->pipe->screen->get_tex_surface
+   (
+      r->pipe->screen, r->surface,
+      0, 0, 0, PIPE_BUFFER_USAGE_GPU_WRITE
+   );
+
+   r->pipe->set_framebuffer_state(r->pipe, &r->fb_state);
+   r->pipe->set_viewport_state(r->pipe, &r->viewport);
+
+   vs_consts = pipe_buffer_map
+   (
+      r->pipe->screen, r->vs_const_buf.buffer,
+      PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
+   );
+
+   vs_consts->denorm.x = r->surface->width[0];
+   vs_consts->denorm.y = r->surface->height[0];
+
+   pipe_buffer_unmap(r->pipe->screen, r->vs_const_buf.buffer);
+
+   r->pipe->set_constant_buffer(r->pipe, PIPE_SHADER_VERTEX, 0,
+                                &r->vs_const_buf);
+   r->pipe->set_constant_buffer(r->pipe, PIPE_SHADER_FRAGMENT, 0,
+                                &r->fs_const_buf);
+
+   if (num_macroblocks[MACROBLOCK_TYPE_INTRA] > 0)
+   {
+      r->pipe->set_vertex_buffers(r->pipe, 1, r->vertex_bufs.all);
+      r->pipe->set_vertex_elements(r->pipe, 4, r->vertex_elems);
+      r->pipe->set_sampler_textures(r->pipe, 3, r->textures.all);
+      r->pipe->bind_sampler_states(r->pipe, 3, r->samplers.all);
+      r->pipe->bind_vs_state(r->pipe, r->i_vs);
+      r->pipe->bind_fs_state(r->pipe, r->i_fs);
+
+      r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
+                           num_macroblocks[MACROBLOCK_TYPE_INTRA] * 24);
+      vb_start += num_macroblocks[MACROBLOCK_TYPE_INTRA] * 24;
+   }
+
+   if (num_macroblocks[MACROBLOCK_TYPE_FWD_FRAME_PRED] > 0)
+   {
+      r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
+      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->textures.individual.ref[0] = r->past;
+      r->pipe->set_sampler_textures(r->pipe, 4, r->textures.all);
+      r->pipe->bind_sampler_states(r->pipe, 4, r->samplers.all);
+      r->pipe->bind_vs_state(r->pipe, r->p_vs[0]);
+      r->pipe->bind_fs_state(r->pipe, r->p_fs[0]);
+
+      r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
+                           num_macroblocks[MACROBLOCK_TYPE_FWD_FRAME_PRED] * 24);
+      vb_start += num_macroblocks[MACROBLOCK_TYPE_FWD_FRAME_PRED] * 24;
+   }
+
+   if (false /*num_macroblocks[MACROBLOCK_TYPE_FWD_FIELD_PRED] > 0 */ )
+   {
+      r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
+      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->textures.individual.ref[0] = r->past;
+      r->pipe->set_sampler_textures(r->pipe, 4, r->textures.all);
+      r->pipe->bind_sampler_states(r->pipe, 4, r->samplers.all);
+      r->pipe->bind_vs_state(r->pipe, r->p_vs[1]);
+      r->pipe->bind_fs_state(r->pipe, r->p_fs[1]);
+
+      r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
+                           num_macroblocks[MACROBLOCK_TYPE_FWD_FIELD_PRED] * 24);
+      vb_start += num_macroblocks[MACROBLOCK_TYPE_FWD_FIELD_PRED] * 24;
+   }
+
+   if (num_macroblocks[MACROBLOCK_TYPE_BKWD_FRAME_PRED] > 0)
+   {
+      r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
+      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->textures.individual.ref[0] = r->future;
+      r->pipe->set_sampler_textures(r->pipe, 4, r->textures.all);
+      r->pipe->bind_sampler_states(r->pipe, 4, r->samplers.all);
+      r->pipe->bind_vs_state(r->pipe, r->p_vs[0]);
+      r->pipe->bind_fs_state(r->pipe, r->p_fs[0]);
+
+      r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
+                           num_macroblocks[MACROBLOCK_TYPE_BKWD_FRAME_PRED] * 24);
+      vb_start += num_macroblocks[MACROBLOCK_TYPE_BKWD_FRAME_PRED] * 24;
+   }
+
+   if (false /*num_macroblocks[MACROBLOCK_TYPE_BKWD_FIELD_PRED] > 0 */ )
+   {
+      r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
+      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->textures.individual.ref[0] = r->future;
+      r->pipe->set_sampler_textures(r->pipe, 4, r->textures.all);
+      r->pipe->bind_sampler_states(r->pipe, 4, r->samplers.all);
+      r->pipe->bind_vs_state(r->pipe, r->p_vs[1]);
+      r->pipe->bind_fs_state(r->pipe, r->p_fs[1]);
+
+      r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
+                           num_macroblocks[MACROBLOCK_TYPE_BKWD_FIELD_PRED] * 24);
+      vb_start += num_macroblocks[MACROBLOCK_TYPE_BKWD_FIELD_PRED] * 24;
+   }
+
+   if (num_macroblocks[MACROBLOCK_TYPE_BI_FRAME_PRED] > 0)
+   {
+      r->pipe->set_vertex_buffers(r->pipe, 3, r->vertex_bufs.all);
+      r->pipe->set_vertex_elements(r->pipe, 8, r->vertex_elems);
+      r->textures.individual.ref[0] = r->past;
+      r->textures.individual.ref[1] = r->future;
+      r->pipe->set_sampler_textures(r->pipe, 5, r->textures.all);
+      r->pipe->bind_sampler_states(r->pipe, 5, r->samplers.all);
+      r->pipe->bind_vs_state(r->pipe, r->b_vs[0]);
+      r->pipe->bind_fs_state(r->pipe, r->b_fs[0]);
+
+      r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
+                           num_macroblocks[MACROBLOCK_TYPE_BI_FRAME_PRED] * 24);
+      vb_start += num_macroblocks[MACROBLOCK_TYPE_BI_FRAME_PRED] * 24;
+   }
+
+   if (false /*num_macroblocks[MACROBLOCK_TYPE_BI_FIELD_PRED] > 0 */ )
+   {
+      r->pipe->set_vertex_buffers(r->pipe, 3, r->vertex_bufs.all);
+      r->pipe->set_vertex_elements(r->pipe, 8, r->vertex_elems);
+      r->textures.individual.ref[0] = r->past;
+      r->textures.individual.ref[1] = r->future;
+      r->pipe->set_sampler_textures(r->pipe, 5, r->textures.all);
+      r->pipe->bind_sampler_states(r->pipe, 5, r->samplers.all);
+      r->pipe->bind_vs_state(r->pipe, r->b_vs[1]);
+      r->pipe->bind_fs_state(r->pipe, r->b_fs[1]);
+
+      r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
+                           num_macroblocks[MACROBLOCK_TYPE_BI_FIELD_PRED] * 24);
+      vb_start += num_macroblocks[MACROBLOCK_TYPE_BI_FIELD_PRED] * 24;
+   }
+
+   r->pipe->flush(r->pipe, PIPE_FLUSH_RENDER_CACHE, r->fence);
+   pipe_surface_reference(&r->fb_state.cbufs[0], NULL);
+
+   if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE)
+      for (unsigned i = 0; i < 3; ++i)
+         r->zero_block[i].x = ZERO_BLOCK_NIL;
+
+   r->num_macroblocks = 0;
+}
+
+static void
+grab_frame_coded_block(short *src, short *dst, unsigned dst_pitch)
+{
+   assert(src);
+   assert(dst);
+
+   for (unsigned y = 0; y < BLOCK_HEIGHT; ++y)
+      memcpy(dst + y * dst_pitch, src + y * BLOCK_WIDTH, BLOCK_WIDTH * 2);
+}
+
+static void
+grab_field_coded_block(short *src, short *dst, unsigned dst_pitch)
+{
+   assert(src);
+   assert(dst);
+
+   for (unsigned y = 0; y < BLOCK_HEIGHT; ++y)
+      memcpy(dst + y * dst_pitch * 2, src + y * BLOCK_WIDTH, BLOCK_WIDTH * 2);
+}
+
+static void
+fill_zero_block(short *dst, unsigned dst_pitch)
+{
+   assert(dst);
+
+   for (unsigned y = 0; y < BLOCK_HEIGHT; ++y)
+      memset(dst + y * dst_pitch, 0, 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;
+
+   assert(r);
+   assert(blocks);
+
+   tex_pitch = r->tex_transfer[0]->stride / r->tex_transfer[0]->block.size;
+   texels = r->texels[0] + mbpy * tex_pitch + mbpx;
+
+   for (unsigned y = 0; y < 2; ++y)
+   {
+      for (unsigned x = 0; x < 2; ++x, ++tb)
+      {
+         if ((cbp >> (5 - tb)) & 1)
+         {
+            if (dct_type == PIPE_MPEG12_DCT_TYPE_FRAME)
+            {
+               grab_frame_coded_block(blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT,
+                                      texels + y * tex_pitch * BLOCK_WIDTH +
+                                      x * BLOCK_WIDTH, tex_pitch);
+            }
+            else
+            {
+               grab_field_coded_block(blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT,
+                                      texels + y * tex_pitch + x * BLOCK_WIDTH,
+                                      tex_pitch);
+            }
+
+            ++sb;
+         }
+         else if (r->eb_handling != VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_NONE)
+         {
+            if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ALL ||
+                ZERO_BLOCK_IS_NIL(r->zero_block[0]))
+            {
+               fill_zero_block(texels + y * tex_pitch * BLOCK_WIDTH + x * BLOCK_WIDTH, tex_pitch);
+               if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE)
+               {
+                  r->zero_block[0].x = (mbpx + x * 8) * r->surface_tex_inv_size.x;
+                  r->zero_block[0].y = (mbpy + y * 8) * r->surface_tex_inv_size.y;
+               }
+            }
+         }
+      }
+   }
+
+   /* 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 / r->tex_transfer[tb + 1]->block.size;
+      texels = r->texels[tb + 1] + mbpy * tex_pitch + mbpx;
+
+      if ((cbp >> (1 - tb)) & 1)
+      {
+         grab_frame_coded_block(blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT, texels, tex_pitch);
+         ++sb;
+      }
+      else if (r->eb_handling != VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_NONE)
+      {
+         if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ALL ||
+             ZERO_BLOCK_IS_NIL(r->zero_block[tb + 1]))
+         {
+            fill_zero_block(texels, tex_pitch);
+            if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE)
+            {
+               r->zero_block[tb + 1].x = (mbpx << 1) * r->surface_tex_inv_size.x;
+               r->zero_block[tb + 1].y = (mbpy << 1) * r->surface_tex_inv_size.y;
+            }
+         }
+      }
+   }
+}
+
+static void
+grab_macroblock(struct vl_mpeg12_mc_renderer *r,
+                struct pipe_mpeg12_macroblock *mb)
+{
+   assert(r);
+   assert(mb);
+   assert(r->num_macroblocks < r->macroblocks_per_batch);
+
+   memcpy(&r->macroblock_buf[r->num_macroblocks], mb,
+          sizeof(struct pipe_mpeg12_macroblock));
+
+   grab_blocks(r, mb->mbx, mb->mby, mb->dct_type, mb->cbp, mb->blocks);
+
+   ++r->num_macroblocks;
+}
+
+bool
+vl_mpeg12_mc_renderer_init(struct vl_mpeg12_mc_renderer *renderer,
+                           struct pipe_context *pipe,
+                           unsigned picture_width,
+                           unsigned picture_height,
+                           enum pipe_video_chroma_format chroma_format,
+                           enum VL_MPEG12_MC_RENDERER_BUFFER_MODE bufmode,
+                           enum VL_MPEG12_MC_RENDERER_EMPTY_BLOCK eb_handling,
+                           bool pot_buffers)
+{
+   assert(renderer);
+   assert(pipe);
+   /* TODO: Implement other policies */
+   assert(bufmode == VL_MPEG12_MC_RENDERER_BUFFER_PICTURE);
+   /* TODO: Implement this */
+   /* XXX: XFER_ALL sampling issue at block edges when using bilinear filtering */
+   assert(eb_handling != VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_NONE);
+   /* TODO: Non-pot buffers untested, probably doesn't work without changes to texcoord generation, vert shader, etc */
+   assert(pot_buffers);
+
+   memset(renderer, 0, sizeof(struct vl_mpeg12_mc_renderer));
+
+   renderer->pipe = pipe;
+   renderer->picture_width = picture_width;
+   renderer->picture_height = picture_height;
+   renderer->chroma_format = chroma_format;
+   renderer->bufmode = bufmode;
+   renderer->eb_handling = eb_handling;
+   renderer->pot_buffers = pot_buffers;
+
+   if (!init_pipe_state(renderer))
+      return false;
+   if (!init_shaders(renderer))
+   {
+      cleanup_pipe_state(renderer);
+      return false;
+   }
+   if (!init_buffers(renderer))
+   {
+      cleanup_shaders(renderer);
+      cleanup_pipe_state(renderer);
+      return false;
+   }
+
+   renderer->surface = NULL;
+   renderer->past = NULL;
+   renderer->future = NULL;
+   for (unsigned i = 0; i < 3; ++i)
+      renderer->zero_block[i].x = ZERO_BLOCK_NIL;
+   renderer->num_macroblocks = 0;
+
+   xfer_buffers_map(renderer);
+
+   return true;
+}
+
+void
+vl_mpeg12_mc_renderer_cleanup(struct vl_mpeg12_mc_renderer *renderer)
+{
+   assert(renderer);
+
+   xfer_buffers_unmap(renderer);
+
+   cleanup_pipe_state(renderer);
+   cleanup_shaders(renderer);
+   cleanup_buffers(renderer);
+}
+
+void
+vl_mpeg12_mc_renderer_render_macroblocks(struct vl_mpeg12_mc_renderer
+                                         *renderer,
+                                         struct pipe_texture *surface,
+                                         struct pipe_texture *past,
+                                         struct pipe_texture *future,
+                                         unsigned num_macroblocks,
+                                         struct pipe_mpeg12_macroblock
+                                         *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)
+      {
+         if (renderer->num_macroblocks > 0)
+         {
+            xfer_buffers_unmap(renderer);
+            flush(renderer);
+         }
+         
+         new_surface = true;
+      }
+
+      /* If the surface we're rendering hasn't changed the ref frames shouldn't change. */
+      assert(surface != renderer->surface || renderer->past == past);
+      assert(surface != renderer->surface || renderer->future == future);
+   }
+   else
+      new_surface = true;
+
+   if (new_surface)
+   {
+      renderer->surface = surface;
+      renderer->past = past;
+      renderer->future = future;
+      renderer->fence = fence;
+      renderer->surface_tex_inv_size.x = 1.0f / surface->width[0];
+      renderer->surface_tex_inv_size.y = 1.0f / surface->height[0];
+   }
+
+   while (num_macroblocks)
+   {
+      unsigned left_in_batch = renderer->macroblocks_per_batch - renderer->num_macroblocks;
+      unsigned num_to_submit = MIN2(num_macroblocks, left_in_batch);
+
+      for (unsigned i = 0; i < num_to_submit; ++i)
+      {
+         assert(mpeg12_macroblocks[i].base.codec == PIPE_VIDEO_CODEC_MPEG12);
+         grab_macroblock(renderer, &mpeg12_macroblocks[i]);
+      }
+
+      num_macroblocks -= num_to_submit;
+
+      if (renderer->num_macroblocks == renderer->macroblocks_per_batch)
+      {
+         xfer_buffers_unmap(renderer);
+         flush(renderer);
+         xfer_buffers_map(renderer);
+         /* Next time we get this surface it may have new ref frames */
+         renderer->surface = NULL;
+      }
+   }
+}
diff --git a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
new file mode 100644
index 00000000000..dfe0f7a24b8
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
@@ -0,0 +1,93 @@
+#ifndef vl_mpeg12_mc_renderer_h
+#define vl_mpeg12_mc_renderer_h
+
+#include <stdbool.h>
+#include <pipe/p_state.h>
+#include <pipe/p_video_state.h>
+
+struct pipe_context;
+struct pipe_video_surface;
+struct pipe_macroblock;
+
+/* A slice is video-width (rounded up to a multiple of macroblock width) x macroblock height */
+enum VL_MPEG12_MC_RENDERER_BUFFER_MODE
+{
+   VL_MPEG12_MC_RENDERER_BUFFER_SLICE,  /* Saves memory at the cost of smaller batches */
+   VL_MPEG12_MC_RENDERER_BUFFER_PICTURE /* Larger batches, more memory */
+};
+
+enum VL_MPEG12_MC_RENDERER_EMPTY_BLOCK
+{
+   VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ALL, /* Waste of memory bandwidth */
+   VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE, /* Can only do point-filtering when interpolating subsampled chroma channels */
+   VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_NONE /* Needs conditional texel fetch! */
+};
+
+struct vl_mpeg12_mc_renderer
+{
+   struct pipe_context *pipe;
+   unsigned picture_width;
+   unsigned picture_height;
+   enum pipe_video_chroma_format chroma_format;
+   enum VL_MPEG12_MC_RENDERER_BUFFER_MODE bufmode;
+   enum VL_MPEG12_MC_RENDERER_EMPTY_BLOCK eb_handling;
+   bool pot_buffers;
+   unsigned macroblocks_per_batch;
+
+   struct pipe_viewport_state viewport;
+   struct pipe_constant_buffer vs_const_buf;
+   struct pipe_constant_buffer fs_const_buf;
+   struct pipe_framebuffer_state fb_state;
+   struct pipe_vertex_element vertex_elems[8];
+	
+   union
+   {
+      void *all[5];
+      struct { void *y, *cb, *cr, *ref[2]; } individual;
+   } samplers;
+	
+   void *i_vs, *p_vs[2], *b_vs[2];
+   void *i_fs, *p_fs[2], *b_fs[2];
+	
+   union
+   {
+      struct pipe_texture *all[5];
+      struct { struct pipe_texture *y, *cb, *cr, *ref[2]; } individual;
+   } textures;
+
+   union
+   {
+      struct pipe_vertex_buffer all[3];
+      struct { struct pipe_vertex_buffer ycbcr, ref[2]; } individual;
+   } vertex_bufs;
+	
+   struct pipe_texture *surface, *past, *future;
+   struct pipe_fence_handle **fence;
+   unsigned num_macroblocks;
+   struct pipe_mpeg12_macroblock *macroblock_buf;
+   struct pipe_transfer *tex_transfer[3];
+   short *texels[3];
+   struct { float x, y; } surface_tex_inv_size;
+   struct { float x, y; } zero_block[3];
+};
+
+bool vl_mpeg12_mc_renderer_init(struct vl_mpeg12_mc_renderer *renderer,
+                                struct pipe_context *pipe,
+                                unsigned picture_width,
+                                unsigned picture_height,
+                                enum pipe_video_chroma_format chroma_format,
+                                enum VL_MPEG12_MC_RENDERER_BUFFER_MODE bufmode,
+                                enum VL_MPEG12_MC_RENDERER_EMPTY_BLOCK eb_handling,
+                                bool pot_buffers);
+
+void vl_mpeg12_mc_renderer_cleanup(struct vl_mpeg12_mc_renderer *renderer);
+
+void vl_mpeg12_mc_renderer_render_macroblocks(struct vl_mpeg12_mc_renderer *renderer,
+                                              struct pipe_texture *surface,
+                                              struct pipe_texture *past,
+                                              struct pipe_texture *future,
+                                              unsigned num_macroblocks,
+                                              struct pipe_mpeg12_macroblock *mpeg12_macroblocks,
+                                              struct pipe_fence_handle **fence);
+
+#endif /* vl_mpeg12_mc_renderer_h */
diff --git a/src/gallium/auxiliary/vl/vl_shader_build.c b/src/gallium/auxiliary/vl/vl_shader_build.c
new file mode 100644
index 00000000000..5a4a5ab72cc
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_shader_build.c
@@ -0,0 +1,215 @@
+#include "vl_shader_build.h"
+#include <assert.h>
+#include <tgsi/tgsi_parse.h>
+#include <tgsi/tgsi_build.h>
+
+struct tgsi_full_declaration vl_decl_input(unsigned int name, unsigned int index, unsigned int first, unsigned int last)
+{
+   struct tgsi_full_declaration decl = tgsi_default_full_declaration();
+
+   decl.Declaration.File = TGSI_FILE_INPUT;
+   decl.Declaration.Semantic = 1;
+   decl.Semantic.SemanticName = name;
+   decl.Semantic.SemanticIndex = index;
+   decl.DeclarationRange.First = first;
+   decl.DeclarationRange.Last = last;
+
+   return decl;
+}
+
+struct tgsi_full_declaration vl_decl_interpolated_input
+(
+   unsigned int name,
+   unsigned int index,
+   unsigned int first,
+   unsigned int last,
+   int interpolation
+)
+{
+   struct tgsi_full_declaration decl = tgsi_default_full_declaration();
+
+   assert
+   (
+   interpolation == TGSI_INTERPOLATE_CONSTANT ||
+   interpolation == TGSI_INTERPOLATE_LINEAR ||
+   interpolation == TGSI_INTERPOLATE_PERSPECTIVE
+   );
+
+   decl.Declaration.File = TGSI_FILE_INPUT;
+   decl.Declaration.Semantic = 1;
+   decl.Semantic.SemanticName = name;
+   decl.Semantic.SemanticIndex = index;
+   decl.Declaration.Interpolate = interpolation;;
+   decl.DeclarationRange.First = first;
+   decl.DeclarationRange.Last = last;
+
+   return decl;
+}
+
+struct tgsi_full_declaration vl_decl_constants(unsigned int name, unsigned int index, unsigned int first, unsigned int last)
+{
+   struct tgsi_full_declaration decl = tgsi_default_full_declaration();
+
+   decl.Declaration.File = TGSI_FILE_CONSTANT;
+   decl.Declaration.Semantic = 1;
+   decl.Semantic.SemanticName = name;
+   decl.Semantic.SemanticIndex = index;
+   decl.DeclarationRange.First = first;
+   decl.DeclarationRange.Last = last;
+
+   return decl;
+}
+
+struct tgsi_full_declaration vl_decl_output(unsigned int name, unsigned int index, unsigned int first, unsigned int last)
+{
+   struct tgsi_full_declaration decl = tgsi_default_full_declaration();
+
+   decl.Declaration.File = TGSI_FILE_OUTPUT;
+   decl.Declaration.Semantic = 1;
+   decl.Semantic.SemanticName = name;
+   decl.Semantic.SemanticIndex = index;
+   decl.DeclarationRange.First = first;
+   decl.DeclarationRange.Last = last;
+
+   return decl;
+}
+
+struct tgsi_full_declaration vl_decl_temps(unsigned int first, unsigned int last)
+{
+   struct tgsi_full_declaration decl = tgsi_default_full_declaration();
+
+   decl = tgsi_default_full_declaration();
+   decl.Declaration.File = TGSI_FILE_TEMPORARY;
+   decl.DeclarationRange.First = first;
+   decl.DeclarationRange.Last = last;
+
+   return decl;
+}
+
+struct tgsi_full_declaration vl_decl_samplers(unsigned int first, unsigned int last)
+{
+   struct tgsi_full_declaration decl = tgsi_default_full_declaration();
+
+   decl = tgsi_default_full_declaration();
+   decl.Declaration.File = TGSI_FILE_SAMPLER;
+   decl.DeclarationRange.First = first;
+   decl.DeclarationRange.Last = last;
+
+   return decl;
+}
+
+struct tgsi_full_instruction vl_inst2
+(
+   int opcode,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src_file,
+   unsigned int src_index
+)
+{
+   struct tgsi_full_instruction inst = tgsi_default_full_instruction();
+
+   inst.Instruction.Opcode = opcode;
+   inst.Instruction.NumDstRegs = 1;
+   inst.FullDstRegisters[0].DstRegister.File = dst_file;
+   inst.FullDstRegisters[0].DstRegister.Index = dst_index;
+   inst.Instruction.NumSrcRegs = 1;
+   inst.FullSrcRegisters[0].SrcRegister.File = src_file;
+   inst.FullSrcRegisters[0].SrcRegister.Index = src_index;
+
+   return inst;
+}
+
+struct tgsi_full_instruction vl_inst3
+(
+   int opcode,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src1_file,
+   unsigned int src1_index,
+   enum tgsi_file_type src2_file,
+   unsigned int src2_index
+)
+{
+   struct tgsi_full_instruction inst = tgsi_default_full_instruction();
+
+   inst.Instruction.Opcode = opcode;
+   inst.Instruction.NumDstRegs = 1;
+   inst.FullDstRegisters[0].DstRegister.File = dst_file;
+   inst.FullDstRegisters[0].DstRegister.Index = dst_index;
+   inst.Instruction.NumSrcRegs = 2;
+   inst.FullSrcRegisters[0].SrcRegister.File = src1_file;
+   inst.FullSrcRegisters[0].SrcRegister.Index = src1_index;
+   inst.FullSrcRegisters[1].SrcRegister.File = src2_file;
+   inst.FullSrcRegisters[1].SrcRegister.Index = src2_index;
+
+   return inst;
+}
+
+struct tgsi_full_instruction vl_tex
+(
+   int tex,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src1_file,
+   unsigned int src1_index,
+   enum tgsi_file_type src2_file,
+   unsigned int src2_index
+)
+{
+   struct tgsi_full_instruction inst = tgsi_default_full_instruction();
+
+   inst.Instruction.Opcode = TGSI_OPCODE_TEX;
+   inst.Instruction.NumDstRegs = 1;
+   inst.FullDstRegisters[0].DstRegister.File = dst_file;
+   inst.FullDstRegisters[0].DstRegister.Index = dst_index;
+   inst.Instruction.NumSrcRegs = 2;
+   inst.InstructionExtTexture.Texture = tex;
+   inst.FullSrcRegisters[0].SrcRegister.File = src1_file;
+   inst.FullSrcRegisters[0].SrcRegister.Index = src1_index;
+   inst.FullSrcRegisters[1].SrcRegister.File = src2_file;
+   inst.FullSrcRegisters[1].SrcRegister.Index = src2_index;
+
+   return inst;
+}
+
+struct tgsi_full_instruction vl_inst4
+(
+   int opcode,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src1_file,
+   unsigned int src1_index,
+   enum tgsi_file_type src2_file,
+   unsigned int src2_index,
+   enum tgsi_file_type src3_file,
+   unsigned int src3_index
+)
+{
+   struct tgsi_full_instruction inst = tgsi_default_full_instruction();
+
+   inst.Instruction.Opcode = opcode;
+   inst.Instruction.NumDstRegs = 1;
+   inst.FullDstRegisters[0].DstRegister.File = dst_file;
+   inst.FullDstRegisters[0].DstRegister.Index = dst_index;
+   inst.Instruction.NumSrcRegs = 3;
+   inst.FullSrcRegisters[0].SrcRegister.File = src1_file;
+   inst.FullSrcRegisters[0].SrcRegister.Index = src1_index;
+   inst.FullSrcRegisters[1].SrcRegister.File = src2_file;
+   inst.FullSrcRegisters[1].SrcRegister.Index = src2_index;
+   inst.FullSrcRegisters[2].SrcRegister.File = src3_file;
+   inst.FullSrcRegisters[2].SrcRegister.Index = src3_index;
+
+   return inst;
+}
+
+struct tgsi_full_instruction vl_end(void)
+{
+   struct tgsi_full_instruction inst = tgsi_default_full_instruction();
+
+   inst.Instruction.Opcode = TGSI_OPCODE_END;
+   inst.Instruction.NumDstRegs = 0;
+   inst.Instruction.NumSrcRegs = 0;
+
+   return inst;
+}
diff --git a/src/gallium/auxiliary/vl/vl_shader_build.h b/src/gallium/auxiliary/vl/vl_shader_build.h
new file mode 100644
index 00000000000..c6c60b55526
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_shader_build.h
@@ -0,0 +1,61 @@
+#ifndef vl_shader_build_h
+#define vl_shader_build_h
+
+#include <pipe/p_shader_tokens.h>
+
+struct tgsi_full_declaration vl_decl_input(unsigned int name, unsigned int index, unsigned int first, unsigned int last);
+struct tgsi_full_declaration vl_decl_interpolated_input
+(
+   unsigned int name,
+   unsigned int index,
+   unsigned int first,
+   unsigned int last,
+   int interpolation
+);
+struct tgsi_full_declaration vl_decl_constants(unsigned int name, unsigned int index, unsigned int first, unsigned int last);
+struct tgsi_full_declaration vl_decl_output(unsigned int name, unsigned int index, unsigned int first, unsigned int last);
+struct tgsi_full_declaration vl_decl_temps(unsigned int first, unsigned int last);
+struct tgsi_full_declaration vl_decl_samplers(unsigned int first, unsigned int last);
+struct tgsi_full_instruction vl_inst2
+(
+   int opcode,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src_file,
+   unsigned int src_index
+);
+struct tgsi_full_instruction vl_inst3
+(
+   int opcode,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src1_file,
+   unsigned int src1_index,
+   enum tgsi_file_type src2_file,
+   unsigned int src2_index
+);
+struct tgsi_full_instruction vl_tex
+(
+   int tex,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src1_file,
+   unsigned int src1_index,
+   enum tgsi_file_type src2_file,
+   unsigned int src2_index
+);
+struct tgsi_full_instruction vl_inst4
+(
+   int opcode,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src1_file,
+   unsigned int src1_index,
+   enum tgsi_file_type src2_file,
+   unsigned int src2_index,
+   enum tgsi_file_type src3_file,
+   unsigned int src3_index
+);
+struct tgsi_full_instruction vl_end(void);
+
+#endif
diff --git a/src/gallium/drivers/softpipe/Makefile b/src/gallium/drivers/softpipe/Makefile
index 6ab37537628..bcb887a0b26 100644
--- a/src/gallium/drivers/softpipe/Makefile
+++ b/src/gallium/drivers/softpipe/Makefile
@@ -31,6 +31,7 @@ C_SOURCES = \
 	sp_tex_sample.c \
 	sp_tex_tile_cache.c \
 	sp_tile_cache.c \
-	sp_surface.c 
+	sp_surface.c \
+	sp_video_context.c
 
 include ../../Makefile.template
diff --git a/src/gallium/drivers/softpipe/SConscript b/src/gallium/drivers/softpipe/SConscript
index 950c3d9955b..aac9edf44e6 100644
--- a/src/gallium/drivers/softpipe/SConscript
+++ b/src/gallium/drivers/softpipe/SConscript
@@ -33,6 +33,7 @@ softpipe = env.ConvenienceLibrary(
 		'sp_tex_tile_cache.c',
 		'sp_texture.c',
 		'sp_tile_cache.c',
+		'sp_video_context.c',
 	])
 
-Export('softpipe')
\ No newline at end of file
+Export('softpipe')
diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c
index 49b51afda7e..45289380d08 100644
--- a/src/gallium/drivers/softpipe/sp_texture.c
+++ b/src/gallium/drivers/softpipe/sp_texture.c
@@ -381,6 +381,59 @@ softpipe_transfer_unmap(struct pipe_screen *screen,
    }
 }
 
+static struct pipe_video_surface*
+softpipe_video_surface_create(struct pipe_screen *screen,
+                              enum pipe_video_chroma_format chroma_format,
+                              unsigned width, unsigned height)
+{
+   struct softpipe_video_surface *sp_vsfc;
+   struct pipe_texture template;
+
+   assert(screen);
+   assert(width && height);
+
+   sp_vsfc = CALLOC_STRUCT(softpipe_video_surface);
+   if (!sp_vsfc)
+      return NULL;
+
+   pipe_reference_init(&sp_vsfc->base.reference, 1);
+   sp_vsfc->base.screen = screen;
+   sp_vsfc->base.chroma_format = chroma_format;
+   /*sp_vsfc->base.surface_format = PIPE_VIDEO_SURFACE_FORMAT_VUYA;*/
+   sp_vsfc->base.width = width;
+   sp_vsfc->base.height = height;
+
+   memset(&template, 0, sizeof(struct pipe_texture));
+   template.target = PIPE_TEXTURE_2D;
+   template.format = PIPE_FORMAT_X8R8G8B8_UNORM;
+   template.last_level = 0;
+   /* vl_mpeg12_mc_renderer expects this when it's initialized with pot_buffers=true */
+   template.width[0] = util_next_power_of_two(width);
+   template.height[0] = util_next_power_of_two(height);
+   template.depth[0] = 1;
+   pf_get_block(template.format, &template.block);
+   template.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER | PIPE_TEXTURE_USAGE_RENDER_TARGET;
+
+   sp_vsfc->tex = screen->texture_create(screen, &template);
+   if (!sp_vsfc->tex)
+   {
+      FREE(sp_vsfc);
+      return NULL;
+   }
+
+   return &sp_vsfc->base;
+}
+
+
+static void
+softpipe_video_surface_destroy(struct pipe_video_surface *vsfc)
+{
+   struct softpipe_video_surface *sp_vsfc = softpipe_video_surface(vsfc);
+
+   pipe_texture_reference(&sp_vsfc->tex, NULL);
+   FREE(sp_vsfc);
+}
+
 
 void
 softpipe_init_screen_texture_funcs(struct pipe_screen *screen)
@@ -396,6 +449,9 @@ softpipe_init_screen_texture_funcs(struct pipe_screen *screen)
    screen->tex_transfer_destroy = softpipe_tex_transfer_destroy;
    screen->transfer_map = softpipe_transfer_map;
    screen->transfer_unmap = softpipe_transfer_unmap;
+
+   screen->video_surface_create = softpipe_video_surface_create;
+   screen->video_surface_destroy = softpipe_video_surface_destroy;
 }
 
 
diff --git a/src/gallium/drivers/softpipe/sp_texture.h b/src/gallium/drivers/softpipe/sp_texture.h
index 2537ab6a40d..2ef64e1e7c3 100644
--- a/src/gallium/drivers/softpipe/sp_texture.h
+++ b/src/gallium/drivers/softpipe/sp_texture.h
@@ -30,6 +30,7 @@
 
 
 #include "pipe/p_state.h"
+#include "pipe/p_video_state.h"
 
 
 struct pipe_context;
@@ -62,6 +63,15 @@ struct softpipe_transfer
    unsigned long offset;
 };
 
+struct softpipe_video_surface
+{
+   struct pipe_video_surface base;
+
+   /* The data is held here:
+    */
+   struct pipe_texture *tex;
+};
+
 
 /** cast wrappers */
 static INLINE struct softpipe_texture *
@@ -76,6 +86,12 @@ softpipe_transfer(struct pipe_transfer *pt)
    return (struct softpipe_transfer *) pt;
 }
 
+static INLINE struct softpipe_video_surface *
+softpipe_video_surface(struct pipe_video_surface *pvs)
+{
+   return (struct softpipe_video_surface *) pvs;
+}
+
 
 extern void
 softpipe_init_screen_texture_funcs(struct pipe_screen *screen);
diff --git a/src/gallium/drivers/softpipe/sp_video_context.c b/src/gallium/drivers/softpipe/sp_video_context.c
new file mode 100644
index 00000000000..1b47bbede26
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_video_context.c
@@ -0,0 +1,273 @@
+#include "sp_video_context.h"
+#include <pipe/p_inlines.h>
+#include <util/u_memory.h>
+#include "softpipe/sp_winsys.h"
+#include "softpipe/sp_texture.h"
+
+static void
+sp_mpeg12_destroy(struct pipe_video_context *vpipe)
+{
+	struct sp_mpeg12_context *ctx = (struct sp_mpeg12_context*)vpipe;
+
+	assert(vpipe);
+	
+	/* Asserted in softpipe_delete_fs_state() for some reason */
+	ctx->pipe->bind_vs_state(ctx->pipe, NULL);
+	ctx->pipe->bind_fs_state(ctx->pipe, NULL);
+
+	ctx->pipe->delete_blend_state(ctx->pipe, ctx->blend);
+	ctx->pipe->delete_rasterizer_state(ctx->pipe, ctx->rast);
+	ctx->pipe->delete_depth_stencil_alpha_state(ctx->pipe, ctx->dsa);
+
+	pipe_video_surface_reference(&ctx->decode_target, NULL);
+	vl_compositor_cleanup(&ctx->compositor);
+	vl_mpeg12_mc_renderer_cleanup(&ctx->mc_renderer);
+	ctx->pipe->destroy(ctx->pipe);
+
+	FREE(ctx);
+}
+
+static void
+sp_mpeg12_decode_macroblocks(struct pipe_video_context *vpipe,
+                             struct pipe_video_surface *past,
+                             struct pipe_video_surface *future,
+                             unsigned num_macroblocks,
+                             struct pipe_macroblock *macroblocks,
+                             struct pipe_fence_handle **fence)
+{
+	struct sp_mpeg12_context *ctx = (struct sp_mpeg12_context*)vpipe;
+	struct pipe_mpeg12_macroblock *mpeg12_macroblocks = (struct pipe_mpeg12_macroblock*)macroblocks;
+
+	assert(vpipe);
+	assert(num_macroblocks);
+	assert(macroblocks);
+	assert(macroblocks->codec == PIPE_VIDEO_CODEC_MPEG12);
+	assert(ctx->decode_target);
+
+	vl_mpeg12_mc_renderer_render_macroblocks(&ctx->mc_renderer,
+	                                         softpipe_video_surface(ctx->decode_target)->tex,
+	                                         past ? softpipe_video_surface(past)->tex : NULL,
+	                                         future ? softpipe_video_surface(future)->tex : NULL,
+	                                         num_macroblocks, mpeg12_macroblocks, fence);
+}
+
+static void
+sp_mpeg12_clear_surface(struct pipe_video_context *vpipe,
+                        unsigned x, unsigned y,
+                        unsigned width, unsigned height,
+                        unsigned value,
+                        struct pipe_surface *surface)
+{
+	struct sp_mpeg12_context *ctx = (struct sp_mpeg12_context*)vpipe;
+
+	assert(vpipe);
+	assert(surface);
+
+	ctx->pipe->surface_fill(ctx->pipe, surface, x, y, width, height, value);
+}
+
+static void
+sp_mpeg12_render_picture(struct pipe_video_context     *vpipe,
+                         /*struct pipe_surface         *backround,
+                         struct pipe_video_rect        *backround_area,*/
+                         struct pipe_video_surface     *src_surface,
+                         enum pipe_mpeg12_picture_type picture_type,
+                         /*unsigned                    num_past_surfaces,
+                         struct pipe_video_surface     *past_surfaces,
+                         unsigned                      num_future_surfaces,
+                         struct pipe_video_surface     *future_surfaces,*/
+                         struct pipe_video_rect        *src_area,
+                         struct pipe_surface           *dst_surface,
+                         struct pipe_video_rect        *dst_area,
+                         /*unsigned                      num_layers,
+                         struct pipe_surface           *layers,
+                         struct pipe_video_rect        *layer_src_areas,
+                         struct pipe_video_rect        *layer_dst_areas*/
+                         struct pipe_fence_handle      **fence)
+{
+	struct sp_mpeg12_context *ctx = (struct sp_mpeg12_context*)vpipe;
+	
+	assert(vpipe);
+	assert(src_surface);
+	assert(src_area);
+	assert(dst_surface);
+	assert(dst_area);
+	
+	vl_compositor_render(&ctx->compositor, softpipe_video_surface(src_surface)->tex,
+	                     picture_type, src_area, dst_surface->texture, dst_area, fence);
+}
+
+static void
+sp_mpeg12_set_decode_target(struct pipe_video_context *vpipe,
+                            struct pipe_video_surface *dt)
+{
+	struct sp_mpeg12_context *ctx = (struct sp_mpeg12_context*)vpipe;
+
+	assert(vpipe);
+	assert(dt);
+
+	pipe_video_surface_reference(&ctx->decode_target, dt);
+}
+
+static bool
+init_pipe_state(struct sp_mpeg12_context *ctx)
+{
+	struct pipe_rasterizer_state rast;
+	struct pipe_blend_state blend;
+	struct pipe_depth_stencil_alpha_state dsa;
+
+	assert(ctx);
+	
+	rast.flatshade = 1;
+	rast.flatshade_first = 0;
+	rast.light_twoside = 0;
+	rast.front_winding = PIPE_WINDING_CCW;
+	rast.cull_mode = PIPE_WINDING_CW;
+	rast.fill_cw = PIPE_POLYGON_MODE_FILL;
+	rast.fill_ccw = PIPE_POLYGON_MODE_FILL;
+	rast.offset_cw = 0;
+	rast.offset_ccw = 0;
+	rast.scissor = 0;
+	rast.poly_smooth = 0;
+	rast.poly_stipple_enable = 0;
+	rast.point_sprite = 0;
+	rast.point_size_per_vertex = 0;
+	rast.multisample = 0;
+	rast.line_smooth = 0;
+	rast.line_stipple_enable = 0;
+	rast.line_stipple_factor = 0;
+	rast.line_stipple_pattern = 0;
+	rast.line_last_pixel = 0;
+	rast.bypass_vs_clip_and_viewport = 0;
+	rast.line_width = 1;
+	rast.point_smooth = 0;
+	rast.point_size = 1;
+	rast.offset_units = 1;
+	rast.offset_scale = 1;
+	/*rast.sprite_coord_mode[i] = ;*/
+	ctx->rast = ctx->pipe->create_rasterizer_state(ctx->pipe, &rast);
+	ctx->pipe->bind_rasterizer_state(ctx->pipe, ctx->rast);
+
+	blend.blend_enable = 0;
+	blend.rgb_func = PIPE_BLEND_ADD;
+	blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+	blend.rgb_dst_factor = PIPE_BLENDFACTOR_ONE;
+	blend.alpha_func = PIPE_BLEND_ADD;
+	blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+	blend.alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
+	blend.logicop_enable = 0;
+	blend.logicop_func = PIPE_LOGICOP_CLEAR;
+	/* Needed to allow color writes to FB, even if blending disabled */
+	blend.colormask = PIPE_MASK_RGBA;
+	blend.dither = 0;
+	ctx->blend = ctx->pipe->create_blend_state(ctx->pipe, &blend);
+	ctx->pipe->bind_blend_state(ctx->pipe, ctx->blend);
+
+	dsa.depth.enabled = 0;
+	dsa.depth.writemask = 0;
+	dsa.depth.func = PIPE_FUNC_ALWAYS;
+	dsa.depth.occlusion_count = 0;
+	for (unsigned i = 0; i < 2; ++i)
+	{
+		dsa.stencil[i].enabled = 0;
+		dsa.stencil[i].func = PIPE_FUNC_ALWAYS;
+		dsa.stencil[i].fail_op = PIPE_STENCIL_OP_KEEP;
+		dsa.stencil[i].zpass_op = PIPE_STENCIL_OP_KEEP;
+		dsa.stencil[i].zfail_op = PIPE_STENCIL_OP_KEEP;
+		dsa.stencil[i].ref_value = 0;
+		dsa.stencil[i].valuemask = 0;
+		dsa.stencil[i].writemask = 0;
+	}
+	dsa.alpha.enabled = 0;
+	dsa.alpha.func = PIPE_FUNC_ALWAYS;
+	dsa.alpha.ref_value = 0;
+	ctx->dsa = ctx->pipe->create_depth_stencil_alpha_state(ctx->pipe, &dsa);
+	ctx->pipe->bind_depth_stencil_alpha_state(ctx->pipe, ctx->dsa);
+	
+	return true;
+}
+
+static struct pipe_video_context *
+sp_mpeg12_create(struct pipe_screen *screen, enum pipe_video_profile profile,
+                 enum pipe_video_chroma_format chroma_format,
+                 unsigned width, unsigned height)
+{
+	struct sp_mpeg12_context *ctx;
+
+	assert(u_reduce_video_profile(profile) == PIPE_VIDEO_CODEC_MPEG12);
+
+	ctx = CALLOC_STRUCT(sp_mpeg12_context);
+
+	if (!ctx)
+		return NULL;
+
+	ctx->base.profile = profile;
+	ctx->base.chroma_format = chroma_format;
+	ctx->base.width = width;
+	ctx->base.height = height;
+
+	ctx->base.screen = screen;
+	ctx->base.destroy = sp_mpeg12_destroy;
+	ctx->base.decode_macroblocks = sp_mpeg12_decode_macroblocks;
+	ctx->base.clear_surface = sp_mpeg12_clear_surface;
+	ctx->base.render_picture = sp_mpeg12_render_picture;
+	ctx->base.set_decode_target = sp_mpeg12_set_decode_target;
+
+	ctx->pipe = softpipe_create(screen);
+	if (!ctx->pipe)
+	{
+		FREE(ctx);
+		return NULL;
+	}
+
+	/* TODO: Use slice buffering for softpipe when implemented, no advantage to buffering an entire picture */
+	if (!vl_mpeg12_mc_renderer_init(&ctx->mc_renderer, ctx->pipe,
+	                                width, height, chroma_format,
+	                                VL_MPEG12_MC_RENDERER_BUFFER_PICTURE,
+	                                /* TODO: Use XFER_NONE when implemented */
+	                                VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE,
+	                                true))
+	{
+		ctx->pipe->destroy(ctx->pipe);
+		FREE(ctx);
+		return NULL;
+	}
+	
+	if (!vl_compositor_init(&ctx->compositor, ctx->pipe))
+	{
+		vl_mpeg12_mc_renderer_cleanup(&ctx->mc_renderer);
+		ctx->pipe->destroy(ctx->pipe);
+		FREE(ctx);
+		return NULL;
+	}
+	
+	if (!init_pipe_state(ctx))
+	{
+		vl_compositor_cleanup(&ctx->compositor);
+		vl_mpeg12_mc_renderer_cleanup(&ctx->mc_renderer);
+		ctx->pipe->destroy(ctx->pipe);
+		FREE(ctx);
+		return NULL;
+	}
+
+	return &ctx->base;
+}
+
+struct pipe_video_context *
+sp_video_create(struct pipe_screen *screen, enum pipe_video_profile profile,
+                enum pipe_video_chroma_format chroma_format,
+                unsigned width, unsigned height)
+{
+	assert(screen);
+	assert(width && height);
+
+	switch (u_reduce_video_profile(profile))
+	{
+		case PIPE_VIDEO_CODEC_MPEG12:
+			return sp_mpeg12_create(screen, profile,
+			                        chroma_format,
+			                        width, height);
+		default:
+			return NULL;
+	}
+}
diff --git a/src/gallium/drivers/softpipe/sp_video_context.h b/src/gallium/drivers/softpipe/sp_video_context.h
new file mode 100644
index 00000000000..a70ce9f476a
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_video_context.h
@@ -0,0 +1,30 @@
+#ifndef SP_VIDEO_CONTEXT_H
+#define SP_VIDEO_CONTEXT_H
+
+#include <pipe/p_video_context.h>
+#include <vl/vl_mpeg12_mc_renderer.h>
+#include <vl/vl_compositor.h>
+
+struct pipe_screen;
+struct pipe_context;
+struct pipe_video_surface;
+
+struct sp_mpeg12_context
+{
+	struct pipe_video_context base;
+	struct pipe_context *pipe;
+	struct pipe_video_surface *decode_target;
+	struct vl_mpeg12_mc_renderer mc_renderer;
+	struct vl_compositor compositor;
+
+	void *rast;
+	void *dsa;
+	void *blend;
+};
+
+struct pipe_video_context *
+sp_video_create(struct pipe_screen *screen, enum pipe_video_profile profile,
+                enum pipe_video_chroma_format chroma_format,
+                unsigned width, unsigned height);
+
+#endif /* SP_VIDEO_CONTEXT_H */
diff --git a/src/gallium/include/pipe/p_defines.h b/src/gallium/include/pipe/p_defines.h
index f252d6df00d..1980831dd9d 100644
--- a/src/gallium/include/pipe/p_defines.h
+++ b/src/gallium/include/pipe/p_defines.h
@@ -315,6 +315,30 @@ enum pipe_transfer_usage {
 #define PIPE_REFERENCED_FOR_READ  (1 << 0)
 #define PIPE_REFERENCED_FOR_WRITE (1 << 1)
 
+
+enum pipe_video_codec
+{
+   PIPE_VIDEO_CODEC_MPEG12,   /**< MPEG1, MPEG2 */
+   PIPE_VIDEO_CODEC_MPEG4,    /**< DIVX, XVID */
+   PIPE_VIDEO_CODEC_VC1,      /**< WMV */
+   PIPE_VIDEO_CODEC_MPEG4_AVC /**< H.264 */
+};
+
+enum pipe_video_profile
+{
+   PIPE_VIDEO_PROFILE_MPEG1,
+   PIPE_VIDEO_PROFILE_MPEG2_SIMPLE,
+   PIPE_VIDEO_PROFILE_MPEG2_MAIN,
+   PIPE_VIDEO_PROFILE_MPEG4_SIMPLE,
+   PIPE_VIDEO_PROFILE_MPEG4_ADVANCED_SIMPLE,
+   PIPE_VIDEO_PROFILE_VC1_SIMPLE,
+   PIPE_VIDEO_PROFILE_VC1_MAIN,
+   PIPE_VIDEO_PROFILE_VC1_ADVANCED,
+   PIPE_VIDEO_PROFILE_MPEG4_AVC_BASELINE,
+   PIPE_VIDEO_PROFILE_MPEG4_AVC_MAIN,
+   PIPE_VIDEO_PROFILE_MPEG4_AVC_HIGH
+};
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/gallium/include/pipe/p_format.h b/src/gallium/include/pipe/p_format.h
index c4469d4a9e9..af230809201 100644
--- a/src/gallium/include/pipe/p_format.h
+++ b/src/gallium/include/pipe/p_format.h
@@ -613,6 +613,24 @@ pf_has_alpha( enum pipe_format format )
    }
 }
 
+enum pipe_video_chroma_format
+{
+   PIPE_VIDEO_CHROMA_FORMAT_420,
+   PIPE_VIDEO_CHROMA_FORMAT_422,
+   PIPE_VIDEO_CHROMA_FORMAT_444
+};
+
+#if 0
+enum pipe_video_surface_format
+{
+   PIPE_VIDEO_SURFACE_FORMAT_NV12,  /**< Planar; Y plane, UV plane */
+   PIPE_VIDEO_SURFACE_FORMAT_YV12,  /**< Planar; Y plane, U plane, V plane */
+   PIPE_VIDEO_SURFACE_FORMAT_YUYV,  /**< Interleaved; Y,U,Y,V,Y,U,Y,V */
+   PIPE_VIDEO_SURFACE_FORMAT_UYVY,  /**< Interleaved; U,Y,V,Y,U,Y,V,Y */
+   PIPE_VIDEO_SURFACE_FORMAT_VUYA   /**< Packed; A31-24|Y23-16|U15-8|V7-0 */
+};
+#endif
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/gallium/include/pipe/p_screen.h b/src/gallium/include/pipe/p_screen.h
index 3f30c52a169..f0a4de5df33 100644
--- a/src/gallium/include/pipe/p_screen.h
+++ b/src/gallium/include/pipe/p_screen.h
@@ -53,7 +53,10 @@ extern "C" {
 struct pipe_fence_handle;
 struct pipe_winsys;
 struct pipe_buffer;
-
+struct pipe_texture;
+struct pipe_surface;
+struct pipe_video_surface;
+struct pipe_transfer;
 
 
 /**
@@ -252,6 +255,17 @@ struct pipe_screen {
 
    void (*buffer_destroy)( struct pipe_buffer *buf );
 
+   /**
+    * Create a video surface suitable for use as a decoding target by the
+    * driver's pipe_video_context.
+    */
+   struct pipe_video_surface*
+   (*video_surface_create)( struct pipe_screen *screen,
+                            enum pipe_video_chroma_format chroma_format,
+                            unsigned width, unsigned height );
+
+   void (*video_surface_destroy)( struct pipe_video_surface *vsfc );
+
 
    /**
     * Do any special operations to ensure frontbuffer contents are
diff --git a/src/gallium/include/pipe/p_video_context.h b/src/gallium/include/pipe/p_video_context.h
new file mode 100644
index 00000000000..937705ac505
--- /dev/null
+++ b/src/gallium/include/pipe/p_video_context.h
@@ -0,0 +1,92 @@
+#ifndef PIPE_VIDEO_CONTEXT_H
+#define PIPE_VIDEO_CONTEXT_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <pipe/p_video_state.h>
+
+struct pipe_screen;
+struct pipe_buffer;
+struct pipe_surface;
+struct pipe_video_surface;
+struct pipe_macroblock;
+struct pipe_picture_desc;
+struct pipe_fence_handle;
+
+/**
+ * Gallium video rendering context
+ */
+struct pipe_video_context
+{
+   struct pipe_screen *screen;
+   enum pipe_video_profile profile;
+   enum pipe_video_chroma_format chroma_format;
+   unsigned width;
+   unsigned height;
+
+   void *priv; /**< context private data (for DRI for example) */
+
+   void (*destroy)(struct pipe_video_context *vpipe);
+
+   /**
+    * Picture decoding and displaying
+    */
+   /*@{*/
+   void (*decode_bitstream)(struct pipe_video_context *vpipe,
+                            unsigned num_bufs,
+                            struct pipe_buffer **bitstream_buf);
+
+   void (*decode_macroblocks)(struct pipe_video_context *vpipe,
+                              struct pipe_video_surface *past,
+                              struct pipe_video_surface *future,
+                              unsigned num_macroblocks,
+                              struct pipe_macroblock *macroblocks,
+                              struct pipe_fence_handle **fence);
+
+   void (*clear_surface)(struct pipe_video_context *vpipe,
+                         unsigned x, unsigned y,
+                         unsigned width, unsigned height,
+                         unsigned value,
+                         struct pipe_surface *surface);
+
+   void (*render_picture)(struct pipe_video_context     *vpipe,
+                          /*struct pipe_surface         *backround,
+                          struct pipe_video_rect        *backround_area,*/
+                          struct pipe_video_surface     *src_surface,
+                          enum pipe_mpeg12_picture_type picture_type,
+                          /*unsigned                    num_past_surfaces,
+                          struct pipe_video_surface     *past_surfaces,
+                          unsigned                      num_future_surfaces,
+                          struct pipe_video_surface     *future_surfaces,*/
+                          struct pipe_video_rect        *src_area,
+                          struct pipe_surface           *dst_surface,
+                          struct pipe_video_rect        *dst_area,
+                          /*unsigned                      num_layers,
+                          struct pipe_texture           *layers,
+                          struct pipe_video_rect        *layer_src_areas,
+                          struct pipe_video_rect        *layer_dst_areas,*/
+                          struct pipe_fence_handle      **fence);
+   /*@}*/
+
+   /**
+    * Parameter-like states (or properties)
+    */
+   /*@{*/
+   void (*set_picture_desc)(struct pipe_video_context *vpipe,
+                            const struct pipe_picture_desc *desc);
+
+   void (*set_decode_target)(struct pipe_video_context *vpipe,
+                             struct pipe_video_surface *dt);
+
+   /* TODO: Interface for CSC matrix, scaling modes, post-processing, etc. */
+   /*@}*/
+};
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* PIPE_VIDEO_CONTEXT_H */
diff --git a/src/gallium/include/pipe/p_video_state.h b/src/gallium/include/pipe/p_video_state.h
new file mode 100644
index 00000000000..a0128fbd48f
--- /dev/null
+++ b/src/gallium/include/pipe/p_video_state.h
@@ -0,0 +1,158 @@
+#ifndef PIPE_VIDEO_STATE_H
+#define PIPE_VIDEO_STATE_H
+
+/* u_reduce_video_profile() needs these */
+#include <assert.h>
+#include <pipe/p_compiler.h>
+
+#include <pipe/p_defines.h>
+#include <pipe/p_format.h>
+#include <pipe/p_refcnt.h>
+#include <pipe/p_screen.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+struct pipe_video_surface
+{
+   struct pipe_reference reference;
+   struct pipe_screen *screen;
+   enum pipe_video_chroma_format chroma_format;
+   /*enum pipe_video_surface_format surface_format;*/
+   unsigned width;
+   unsigned height;
+};
+
+static INLINE void
+pipe_video_surface_reference(struct pipe_video_surface **ptr, struct pipe_video_surface *surf)
+{
+   struct pipe_video_surface *old_surf = *ptr;
+
+   if (pipe_reference((struct pipe_reference **)ptr, &surf->reference))
+      old_surf->screen->video_surface_destroy(old_surf);
+}
+
+struct pipe_video_rect
+{
+   unsigned x, y, w, h;
+};
+
+static INLINE enum pipe_video_codec
+u_reduce_video_profile(enum pipe_video_profile profile)
+{
+   switch (profile)
+   {
+      case PIPE_VIDEO_PROFILE_MPEG1:
+      case PIPE_VIDEO_PROFILE_MPEG2_SIMPLE:
+      case PIPE_VIDEO_PROFILE_MPEG2_MAIN:
+         return PIPE_VIDEO_CODEC_MPEG12;
+
+      case PIPE_VIDEO_PROFILE_MPEG4_SIMPLE:
+      case PIPE_VIDEO_PROFILE_MPEG4_ADVANCED_SIMPLE:
+         return PIPE_VIDEO_CODEC_MPEG4;
+
+      case PIPE_VIDEO_PROFILE_VC1_SIMPLE:
+      case PIPE_VIDEO_PROFILE_VC1_MAIN:
+      case PIPE_VIDEO_PROFILE_VC1_ADVANCED:
+         return PIPE_VIDEO_CODEC_VC1;
+
+      case PIPE_VIDEO_PROFILE_MPEG4_AVC_BASELINE:
+      case PIPE_VIDEO_PROFILE_MPEG4_AVC_MAIN:
+      case PIPE_VIDEO_PROFILE_MPEG4_AVC_HIGH:
+         return PIPE_VIDEO_CODEC_MPEG4_AVC;
+
+      default:
+         assert(false);
+   }
+
+   return -1;
+}
+
+enum pipe_mpeg12_picture_type
+{
+   PIPE_MPEG12_PICTURE_TYPE_FIELD_TOP,
+   PIPE_MPEG12_PICTURE_TYPE_FIELD_BOTTOM,
+   PIPE_MPEG12_PICTURE_TYPE_FRAME
+};
+
+enum pipe_mpeg12_macroblock_type
+{
+   PIPE_MPEG12_MACROBLOCK_TYPE_INTRA,
+   PIPE_MPEG12_MACROBLOCK_TYPE_FWD,
+   PIPE_MPEG12_MACROBLOCK_TYPE_BKWD,
+   PIPE_MPEG12_MACROBLOCK_TYPE_BI,
+	
+   PIPE_MPEG12_MACROBLOCK_NUM_TYPES
+};
+
+enum pipe_mpeg12_motion_type
+{
+   PIPE_MPEG12_MOTION_TYPE_FIELD,
+   PIPE_MPEG12_MOTION_TYPE_FRAME,
+   PIPE_MPEG12_MOTION_TYPE_DUALPRIME,
+   PIPE_MPEG12_MOTION_TYPE_16x8
+};
+
+enum pipe_mpeg12_dct_type
+{
+   PIPE_MPEG12_DCT_TYPE_FIELD,
+   PIPE_MPEG12_DCT_TYPE_FRAME
+};
+
+struct pipe_macroblock
+{
+   enum pipe_video_codec codec;
+};
+
+struct pipe_mpeg12_macroblock
+{
+   struct pipe_macroblock base;
+
+   unsigned mbx;
+   unsigned mby;
+   enum pipe_mpeg12_macroblock_type mb_type;
+   enum pipe_mpeg12_motion_type mo_type;
+   enum pipe_mpeg12_dct_type dct_type;
+   signed pmv[2][2][2];
+   unsigned cbp;
+   void *blocks;
+};
+
+#if 0
+struct pipe_picture_desc
+{
+   enum pipe_video_format format;
+};
+
+struct pipe_mpeg12_picture_desc
+{
+   struct pipe_picture_desc base;
+
+   /* TODO: Use bitfields where possible? */
+   struct pipe_surface *forward_reference;
+   struct pipe_surface *backward_reference;
+   unsigned picture_coding_type;
+   unsigned fcode;
+   unsigned intra_dc_precision;
+   unsigned picture_structure;
+   unsigned top_field_first;
+   unsigned frame_pred_frame_dct;
+   unsigned concealment_motion_vectors;
+   unsigned q_scale_type;
+   unsigned intra_vlc_format;
+   unsigned alternate_scan;
+   unsigned full_pel_forward_vector;
+   unsigned full_pel_backward_vector;
+   struct pipe_buffer *intra_quantizer_matrix;
+   struct pipe_buffer *non_intra_quantizer_matrix;
+   struct pipe_buffer *chroma_intra_quantizer_matrix;
+   struct pipe_buffer *chroma_non_intra_quantizer_matrix;
+};
+#endif
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* PIPE_VIDEO_STATE_H */