g3dvl: Use rotating buffers to avoid waiting for map().

4 files changed, 34 insertions, 119 deletions

diff --git a/src/gallium/state_trackers/g3dvl/Makefile b/src/gallium/state_trackers/g3dvl/Makefile
index 9995c554ab..bd77c62bc5 100644
--- a/src/gallium/state_trackers/g3dvl/Makefile
+++ b/src/gallium/state_trackers/g3dvl/Makefile
@@ -1,9 +1,9 @@
 TARGET		= libg3dvl.a
-OBJECTS		= vl_display.o vl_screen.o vl_context.o vl_surface.o vl_data.o vl_shader_build.o vl_util.o vl_basic_csc.o \
+OBJECTS		= vl_display.o vl_screen.o vl_context.o vl_surface.o vl_shader_build.o vl_util.o vl_basic_csc.o	\
 		  vl_r16snorm_mc.o
 GALLIUMDIR	= ../..
 
-CFLAGS		+= -g -Wall -fPIC -Werror -I${GALLIUMDIR}/include -I${GALLIUMDIR}/auxiliary -I${GALLIUMDIR}/winsys/g3dvl
+CFLAGS		+= -g -Wall -fPIC -I${GALLIUMDIR}/include -I${GALLIUMDIR}/auxiliary -I${GALLIUMDIR}/winsys/g3dvl
 
 #############################################
 
diff --git a/src/gallium/state_trackers/g3dvl/vl_data.c b/src/gallium/state_trackers/g3dvl/vl_data.c
deleted file mode 100644
index f2476dbf1e..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_data.c
+++ /dev/null
@@ -1,76 +0,0 @@
-#include "vl_data.h"
-
-/*
- * Represents 8 triangles (4 quads, 1 per block) in noormalized coords
- * that render a macroblock.
- * Need to be scaled to cover mbW*mbH macroblock pixels and translated into
- * position on target surface.
- */
-const struct vlVertex2f macroblock_verts[24] =
-{
-	{0.0f, 0.0f}, {0.0f, 0.5f}, {0.5f, 0.0f},
-	{0.5f, 0.0f}, {0.0f, 0.5f}, {0.5f, 0.5f},
-
-	{0.5f, 0.0f}, {0.5f, 0.5f}, {1.0f, 0.0f},
-	{1.0f, 0.0f}, {0.5f, 0.5f}, {1.0f, 0.5f},
-
-	{0.0f, 0.5f}, {0.0f, 1.0f}, {0.5f, 0.5f},
-	{0.5f, 0.5f}, {0.0f, 1.0f}, {0.5f, 1.0f},
-
-	{0.5f, 0.5f}, {0.5f, 1.0f}, {1.0f, 0.5f},
-	{1.0f, 0.5f}, {0.5f, 1.0f}, {1.0f, 1.0f}
-};
-
-/*
- * Represents texcoords for the above for rendering 4 luma blocks arranged
- * in a bW*(bH*4) texture. First luma block located at 0,0->bW,bH; second at
- * 0,bH->bW,2bH; third at 0,2bH->bW,3bH; fourth at 0,3bH->bW,4bH.
- */
-const struct vlVertex2f macroblock_luma_texcoords[24] =
-{
-	{0.0f, 0.0f}, {0.0f, 0.25f}, {1.0f, 0.0f},
-	{1.0f, 0.0f}, {0.0f, 0.25f}, {1.0f, 0.25f},
-
-	{0.0f, 0.25f}, {0.0f, 0.5f}, {1.0f, 0.25f},
-	{1.0f, 0.25f}, {0.0f, 0.5f}, {1.0f, 0.5f},
-
-	{0.0f, 0.5f}, {0.0f, 0.75f}, {1.0f, 0.5f},
-	{1.0f, 0.5f}, {0.0f, 0.75f}, {1.0f, 0.75f},
-
-	{0.0f, 0.75f}, {0.0f, 1.0f}, {1.0f, 0.75f},
-	{1.0f, 0.75f}, {0.0f, 1.0f}, {1.0f, 1.0f}
-};
-
-/*
- * Represents texcoords for the above for rendering 1 chroma block.
- * Straight forward 0,0->1,1 mapping so we can reuse the MB pos vectors.
- */
-const struct vlVertex2f *macroblock_chroma_420_texcoords = macroblock_verts;
-
-/*
- * Represents texcoords for the above for rendering 2 chroma blocks arranged
- * in a bW*(bH*2) texture. First chroma block located at 0,0->bW,bH; second at
- * 0,bH->bW,2bH. We can render this with 0,0->1,1 mapping.
- * Straight forward 0,0->1,1 mapping so we can reuse MB pos vectors.
- */
-const struct vlVertex2f *macroblock_chroma_422_texcoords = macroblock_verts;
-
-/*
- * Represents texcoords for the above for rendering 4 chroma blocks.
- * Same case as 4 luma blocks.
- */
-const struct vlVertex2f *macroblock_chroma_444_texcoords = macroblock_luma_texcoords;
-
-/*
- * Used when rendering P and B macroblocks, multiplier is applied to the A channel,
- * which is then added to the L channel, then the bias is subtracted from that to
- * get back the differential. The differential is then added to the samples from the
- * reference surface(s).
- */
-#if 0
-const struct VL_MC_FS_CONSTS vl_mc_fs_consts =
-{
-	{32767.0f / 255.0f, 32767.0f / 255.0f, 32767.0f / 255.0f, 0.0f},
-	{0.5f, 2.0f, 0.0f, 0.0f}
-};
-#endif
diff --git a/src/gallium/state_trackers/g3dvl/vl_data.h b/src/gallium/state_trackers/g3dvl/vl_data.h
deleted file mode 100644
index f0de2e976c..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_data.h
+++ /dev/null
@@ -1,27 +0,0 @@
-#ifndef vl_data_h
-#define vl_data_h
-
-#include "vl_types.h"
-
-/* TODO: Needs to be rolled into the appropriate stage */
-
-extern const struct vlVertex2f macroblock_verts[24];
-extern const struct vlVertex2f macroblock_luma_texcoords[24];
-extern const struct vlVertex2f *macroblock_chroma_420_texcoords;
-extern const struct vlVertex2f *macroblock_chroma_422_texcoords;
-extern const struct vlVertex2f *macroblock_chroma_444_texcoords;
-
-extern const struct vlVertex2f surface_verts[4];
-extern const struct vlVertex2f *surface_texcoords;
-
-/*
-extern const struct VL_MC_FS_CONSTS vl_mc_fs_consts;
-
-extern const struct VL_CSC_FS_CONSTS vl_csc_fs_consts_identity;
-extern const struct VL_CSC_FS_CONSTS vl_csc_fs_consts_601;
-extern const struct VL_CSC_FS_CONSTS vl_csc_fs_consts_601_full;
-extern const struct VL_CSC_FS_CONSTS vl_csc_fs_consts_709;
-extern const struct VL_CSC_FS_CONSTS vl_csc_fs_consts_709_full;
-*/
-
-#endif
diff --git a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.c b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.c
index 4fae224431..80b09a6d1d 100644
--- a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.c
+++ b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.c
@@ -15,6 +15,8 @@
 #include "vl_types.h"
 #include "vl_defs.h"
 
+#define NUM_BUFS 4	/* Number of rotating buffers to use */
+
 struct vlVertexShaderConsts
 {
 	/*struct vlVertex4f scale;
@@ -41,12 +43,13 @@ struct vlR16SnormMC
 
 	unsigned int				video_width, video_height;
 	enum vlFormat				video_format;
+	unsigned int				cur_buf;
 
 	struct pipe_context			*pipe;
 	struct pipe_viewport_state		viewport;
 	struct pipe_framebuffer_state		render_target;
 	struct pipe_sampler_state		*samplers[5];
-	struct pipe_texture			*textures[5];
+	struct pipe_texture			*textures[NUM_BUFS][5];
 	void					*i_vs, *p_vs[2], *b_vs[2];
 	void					*i_fs, *p_fs[2], *b_fs[2];
 	struct pipe_vertex_buffer 		vertex_bufs[3];
@@ -230,7 +233,7 @@ static int vlGrabBlocks
 	tex_surface = mc->pipe->screen->get_tex_surface
 	(
 		mc->pipe->screen,
-		mc->textures[0],
+		mc->textures[mc->cur_buf % NUM_BUFS][0],
 		0, 0, 0, PIPE_BUFFER_USAGE_CPU_WRITE
 	);
 
@@ -276,7 +279,7 @@ static int vlGrabBlocks
 		tex_surface = mc->pipe->screen->get_tex_surface
 		(
 			mc->pipe->screen,
-			mc->textures[tb + 1],
+			mc->textures[mc->cur_buf % NUM_BUFS][tb + 1],
 			0, 0, 0, PIPE_BUFFER_USAGE_CPU_WRITE
 		);
 
@@ -361,12 +364,14 @@ int vlRenderIMacroBlock
 		0, 0, 0, PIPE_BUFFER_USAGE_GPU_READ | PIPE_BUFFER_USAGE_GPU_WRITE
 	);
 	pipe->set_framebuffer_state(pipe, &mc->render_target);
-	pipe->set_sampler_textures(pipe, 3, mc->textures);
+	pipe->set_sampler_textures(pipe, 3, mc->textures[mc->cur_buf % NUM_BUFS]);
 	pipe->bind_sampler_states(pipe, 3, (void**)mc->samplers);
 	pipe->bind_vs_state(pipe, mc->i_vs);
 	pipe->bind_fs_state(pipe, mc->i_fs);
 
 	pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, 0, 24);
+	
+	mc->cur_buf++;
 
 	return 0;
 }
@@ -458,11 +463,13 @@ int vlRenderPMacroBlock
 	);
 	pipe->set_framebuffer_state(pipe, &mc->render_target);
 
-	mc->textures[3] = ref_surface->texture;
-	pipe->set_sampler_textures(pipe, 4, mc->textures);
+	mc->textures[mc->cur_buf % NUM_BUFS][3] = ref_surface->texture;
+	pipe->set_sampler_textures(pipe, 4, mc->textures[mc->cur_buf % NUM_BUFS]);
 	pipe->bind_sampler_states(pipe, 4, (void**)mc->samplers);
 
 	pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, 0, 24);
+	
+	mc->cur_buf++;
 
 	return 0;
 }
@@ -567,12 +574,14 @@ int vlRenderBMacroBlock
 	);
 	pipe->set_framebuffer_state(pipe, &mc->render_target);
 
-	mc->textures[3] = past_surface->texture;
-	mc->textures[4] = future_surface->texture;
-	pipe->set_sampler_textures(pipe, 5, mc->textures);
+	mc->textures[mc->cur_buf % NUM_BUFS][3] = past_surface->texture;
+	mc->textures[mc->cur_buf % NUM_BUFS][4] = future_surface->texture;
+	pipe->set_sampler_textures(pipe, 5, mc->textures[mc->cur_buf % NUM_BUFS]);
 	pipe->bind_sampler_states(pipe, 5, (void**)mc->samplers);
 
 	pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, 0, 24);
+	
+	mc->cur_buf++;
 
 	return 0;
 }
@@ -724,8 +733,12 @@ int vlDestroy
 		pipe->winsys->buffer_destroy(pipe->winsys, mc->vertex_bufs[i].buffer);
 
 	/* Textures 3 & 4 are not created directly, no need to release them here */
-	for (i = 0; i < 3; ++i)
-		pipe_texture_release(&mc->textures[i]);
+	for (i = 0; i < NUM_BUFS; ++i)
+	{
+		pipe_texture_release(&mc->textures[i][0]);
+		pipe_texture_release(&mc->textures[i][1]);
+		pipe_texture_release(&mc->textures[i][2]);
+	}
 
 	pipe->delete_vs_state(pipe, mc->i_vs);
 	pipe->delete_fs_state(pipe, mc->i_fs);
@@ -2252,7 +2265,8 @@ static int vlInit
 	template.compressed = 0;
 	pf_get_block(template.format, &template.block);
 
-	mc->textures[0] = pipe->screen->texture_create(pipe->screen, &template);
+	for (i = 0; i < NUM_BUFS; ++i)
+		mc->textures[i][0] = pipe->screen->texture_create(pipe->screen, &template);
 
 	if (mc->video_format == vlFormatYCbCr420)
 		template.height[0] = 8;
@@ -2263,8 +2277,11 @@ static int vlInit
 	else
 		assert(0);
 
-	mc->textures[1] = pipe->screen->texture_create(pipe->screen, &template);
-	mc->textures[2] = pipe->screen->texture_create(pipe->screen, &template);
+	for (i = 0; i < NUM_BUFS; ++i)
+	{
+		mc->textures[i][1] = pipe->screen->texture_create(pipe->screen, &template);
+		mc->textures[i][2] = pipe->screen->texture_create(pipe->screen, &template);
+	}
 
 	/* textures[3] & textures[4] are assigned from vlSurfaces for P and B macroblocks at render time */
 
@@ -2306,6 +2323,7 @@ int vlCreateR16SNormMC
 	mc->pipe = pipe;
 	mc->video_width = video_width;
 	mc->video_height = video_height;
+	mc->cur_buf = 0;
 
 	vlInit(mc);