diff options
Diffstat (limited to 'src/gallium/state_trackers/g3dvl')
23 files changed, 6815 insertions, 0 deletions
diff --git a/src/gallium/state_trackers/g3dvl/Makefile b/src/gallium/state_trackers/g3dvl/Makefile new file mode 100644 index 0000000000..4f7a953484 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/Makefile @@ -0,0 +1,18 @@ +TARGET = libg3dvl.a +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 vl_r16snorm_mc_buf.o +GALLIUMDIR = ../.. + +CFLAGS += -g -Wall -fPIC -I${GALLIUMDIR}/include -I${GALLIUMDIR}/auxiliary -I${GALLIUMDIR}/winsys/g3dvl + +############################################# + +.PHONY = all clean + +all: ${TARGET} + +${TARGET}: ${OBJECTS} + ar rcs $@ $^ + +clean: + rm -rf ${OBJECTS} ${TARGET} diff --git a/src/gallium/state_trackers/g3dvl/vl_basic_csc.c b/src/gallium/state_trackers/g3dvl/vl_basic_csc.c new file mode 100644 index 0000000000..9f9dafc8a9 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_basic_csc.c @@ -0,0 +1,707 @@ +#define VL_INTERNAL +#include "vl_basic_csc.h" +#include <assert.h> +#include <stdlib.h> +#include <pipe/p_context.h> +#include <pipe/p_winsys.h> +#include <pipe/p_state.h> +#include <tgsi/tgsi_parse.h> +#include <tgsi/tgsi_build.h> +#include "vl_csc.h" +#include "vl_surface.h" +#include "vl_shader_build.h" +#include "vl_types.h" + +struct vlVertexShaderConsts +{ + struct vlVertex4f dst_scale; + struct vlVertex4f dst_trans; + struct vlVertex4f src_scale; + struct vlVertex4f src_trans; +}; + +struct vlFragmentShaderConsts +{ + struct vlVertex4f bias; + float matrix[16]; +}; + +struct vlBasicCSC +{ + struct vlCSC base; + + struct pipe_context *pipe; + struct pipe_viewport_state viewport; + struct pipe_framebuffer_state framebuffer; + void *sampler; + void *vertex_shader, *fragment_shader; + struct pipe_vertex_buffer vertex_bufs[2]; + struct pipe_vertex_element vertex_elems[2]; + struct pipe_constant_buffer vs_const_buf, fs_const_buf; +}; + +static int vlResizeFrameBuffer +( + struct vlCSC *csc, + unsigned int width, + unsigned int height +) +{ + struct vlBasicCSC *basic_csc; + struct pipe_context *pipe; + + assert(csc); + + basic_csc = (struct vlBasicCSC*)csc; + pipe = basic_csc->pipe; + + if (basic_csc->framebuffer.width == width && basic_csc->framebuffer.height == height) + return 0; + + if (basic_csc->framebuffer.cbufs[0]) + pipe->winsys->surface_release + ( + pipe->winsys, + &basic_csc->framebuffer.cbufs[0] + ); + + basic_csc->viewport.scale[0] = width; + basic_csc->viewport.scale[1] = height; + basic_csc->viewport.scale[2] = 1; + basic_csc->viewport.scale[3] = 1; + basic_csc->viewport.translate[0] = 0; + basic_csc->viewport.translate[1] = 0; + basic_csc->viewport.translate[2] = 0; + basic_csc->viewport.translate[3] = 0; + + basic_csc->framebuffer.width = width; + basic_csc->framebuffer.height = height; + basic_csc->framebuffer.cbufs[0] = pipe->winsys->surface_alloc(pipe->winsys); + pipe->winsys->surface_alloc_storage + ( + pipe->winsys, + basic_csc->framebuffer.cbufs[0], + width, + height, + PIPE_FORMAT_A8R8G8B8_UNORM, + /* XXX: SoftPipe doesn't change GPU usage to CPU like it does for textures */ + PIPE_BUFFER_USAGE_CPU_READ | PIPE_BUFFER_USAGE_CPU_WRITE, + 0 + ); + + /* Clear to black, in case video doesn't fill the entire window */ + pipe->clear(pipe, basic_csc->framebuffer.cbufs[0], 0); + + return 0; +} + +static int vlBegin +( + struct vlCSC *csc +) +{ + struct vlBasicCSC *basic_csc; + struct pipe_context *pipe; + + assert(csc); + + basic_csc = (struct vlBasicCSC*)csc; + pipe = basic_csc->pipe; + + pipe->set_framebuffer_state(pipe, &basic_csc->framebuffer); + pipe->set_viewport_state(pipe, &basic_csc->viewport); + pipe->bind_sampler_states(pipe, 1, (void**)&basic_csc->sampler); + /* Source texture set in vlPutSurface() */ + pipe->bind_vs_state(pipe, basic_csc->vertex_shader); + pipe->bind_fs_state(pipe, basic_csc->fragment_shader); + pipe->set_vertex_buffers(pipe, 2, basic_csc->vertex_bufs); + pipe->set_vertex_elements(pipe, 2, basic_csc->vertex_elems); + pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 0, &basic_csc->vs_const_buf); + pipe->set_constant_buffer(pipe, PIPE_SHADER_FRAGMENT, 0, &basic_csc->fs_const_buf); + + return 0; +} + +static int vlPutPictureCSC +( + struct vlCSC *csc, + struct vlSurface *surface, + int srcx, + int srcy, + int srcw, + int srch, + int destx, + int desty, + int destw, + int desth, + enum vlPictureType picture_type +) +{ + struct vlBasicCSC *basic_csc; + struct pipe_context *pipe; + struct vlVertexShaderConsts *vs_consts; + + assert(csc); + assert(surface); + + basic_csc = (struct vlBasicCSC*)csc; + pipe = basic_csc->pipe; + + vs_consts = pipe->winsys->buffer_map + ( + pipe->winsys, + basic_csc->vs_const_buf.buffer, + PIPE_BUFFER_USAGE_CPU_WRITE + ); + + vs_consts->dst_scale.x = destw / (float)basic_csc->framebuffer.cbufs[0]->width; + vs_consts->dst_scale.y = desth / (float)basic_csc->framebuffer.cbufs[0]->height; + vs_consts->dst_scale.z = 1; + vs_consts->dst_scale.w = 1; + vs_consts->dst_trans.x = destx / (float)basic_csc->framebuffer.cbufs[0]->width; + vs_consts->dst_trans.y = desty / (float)basic_csc->framebuffer.cbufs[0]->height; + vs_consts->dst_trans.z = 0; + vs_consts->dst_trans.w = 0; + + vs_consts->src_scale.x = srcw / (float)surface->texture->width[0]; + vs_consts->src_scale.y = srch / (float)surface->texture->height[0]; + vs_consts->src_scale.z = 1; + vs_consts->src_scale.w = 1; + vs_consts->src_trans.x = srcx / (float)surface->texture->width[0]; + vs_consts->src_trans.y = srcy / (float)surface->texture->height[0]; + vs_consts->src_trans.z = 0; + vs_consts->src_trans.w = 0; + + pipe->winsys->buffer_unmap(pipe->winsys, basic_csc->vs_const_buf.buffer); + + pipe->set_sampler_textures(pipe, 1, &surface->texture); + pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLE_STRIP, 0, 4); + + return 0; +} + +static int vlEnd +( + struct vlCSC *csc +) +{ + assert(csc); + + return 0; +} + +static struct pipe_surface* vlGetFrameBuffer +( + struct vlCSC *csc +) +{ + struct vlBasicCSC *basic_csc; + + assert(csc); + + basic_csc = (struct vlBasicCSC*)csc; + + return basic_csc->framebuffer.cbufs[0]; +} + +static int vlDestroy +( + struct vlCSC *csc +) +{ + struct vlBasicCSC *basic_csc; + struct pipe_context *pipe; + unsigned int i; + + assert(csc); + + basic_csc = (struct vlBasicCSC*)csc; + pipe = basic_csc->pipe; + + if (basic_csc->framebuffer.cbufs[0]) + pipe->winsys->surface_release + ( + pipe->winsys, + &basic_csc->framebuffer.cbufs[0] + ); + + pipe->delete_sampler_state(pipe, basic_csc->sampler); + pipe->delete_vs_state(pipe, basic_csc->vertex_shader); + pipe->delete_fs_state(pipe, basic_csc->fragment_shader); + + for (i = 0; i < 2; ++i) + pipe->winsys->buffer_destroy(pipe->winsys, basic_csc->vertex_bufs[i].buffer); + + pipe->winsys->buffer_destroy(pipe->winsys, basic_csc->vs_const_buf.buffer); + pipe->winsys->buffer_destroy(pipe->winsys, basic_csc->fs_const_buf.buffer); + + free(basic_csc); + + return 0; +} + +/* + * Represents 2 triangles in a strip in normalized coords. + * Used to render the surface onto the frame buffer. + */ +static const struct vlVertex2f 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 vlVertex2f *surface_texcoords = surface_verts; + +/* + * Identity color conversion constants, for debugging + */ +static const struct vlFragmentShaderConsts 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 vlFragmentShaderConsts 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 vlFragmentShaderConsts 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 vlFragmentShaderConsts 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 vlFragmentShaderConsts 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 int vlCreateVertexShader +( + struct vlBasicCSC *csc +) +{ + const unsigned int max_tokens = 50; + + struct pipe_context *pipe; + struct pipe_shader_state vs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(context); + + pipe = csc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header); + + ti = 3; + + /* + * decl i0 ; Vertex pos + * decl i1 ; Vertex texcoords + */ + for (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 (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); + + /* + * madd o0, i0, c0, c1 ; Scale and translate unit output rect to destination size and pos + * madd o1, i1, c2, c3 ; Scale and translate unit texcoord rect to source size and pos + */ + for (i = 0; i < 2; ++i) + { + inst = vl_inst4(TGSI_OPCODE_MADD, 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); + + vs.tokens = tokens; + csc->vertex_shader = pipe->create_vs_state(pipe, &vs); + free(tokens); + + return 0; +} + +static int vlCreateFragmentShader +( + struct vlBasicCSC *csc +) +{ + const unsigned int max_tokens = 50; + + struct pipe_context *pipe; + struct pipe_shader_state fs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(context); + + pipe = csc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 (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); + + fs.tokens = tokens; + csc->fragment_shader = pipe->create_fs_state(pipe, &fs); + free(tokens); + + return 0; +} + +static int vlCreateDataBufs +( + struct vlBasicCSC *csc +) +{ + struct pipe_context *pipe; + + assert(csc); + + pipe = csc->pipe; + + /* + * 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 + */ + csc->vertex_bufs[0].pitch = sizeof(struct vlVertex2f); + csc->vertex_bufs[0].max_index = 3; + csc->vertex_bufs[0].buffer_offset = 0; + csc->vertex_bufs[0].buffer = pipe->winsys->buffer_create + ( + pipe->winsys, + 1, + PIPE_BUFFER_USAGE_VERTEX, + sizeof(struct vlVertex2f) * 4 + ); + + memcpy + ( + pipe->winsys->buffer_map(pipe->winsys, csc->vertex_bufs[0].buffer, PIPE_BUFFER_USAGE_CPU_WRITE), + surface_verts, + sizeof(struct vlVertex2f) * 4 + ); + + pipe->winsys->buffer_unmap(pipe->winsys, csc->vertex_bufs[0].buffer); + + csc->vertex_elems[0].src_offset = 0; + csc->vertex_elems[0].vertex_buffer_index = 0; + csc->vertex_elems[0].nr_components = 2; + csc->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 + */ + csc->vertex_bufs[1].pitch = sizeof(struct vlVertex2f); + csc->vertex_bufs[1].max_index = 3; + csc->vertex_bufs[1].buffer_offset = 0; + csc->vertex_bufs[1].buffer = pipe->winsys->buffer_create + ( + pipe->winsys, + 1, + PIPE_BUFFER_USAGE_VERTEX, + sizeof(struct vlVertex2f) * 4 + ); + + memcpy + ( + pipe->winsys->buffer_map(pipe->winsys, csc->vertex_bufs[1].buffer, PIPE_BUFFER_USAGE_CPU_WRITE), + surface_texcoords, + sizeof(struct vlVertex2f) * 4 + ); + + pipe->winsys->buffer_unmap(pipe->winsys, csc->vertex_bufs[1].buffer); + + csc->vertex_elems[1].src_offset = 0; + csc->vertex_elems[1].vertex_buffer_index = 1; + csc->vertex_elems[1].nr_components = 2; + csc->vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT; + + /* + * Create our vertex shader's constant buffer + * Const buffer contains scaling and translation vectors + */ + csc->vs_const_buf.size = sizeof(struct vlVertexShaderConsts); + csc->vs_const_buf.buffer = pipe->winsys->buffer_create + ( + pipe->winsys, + 1, + PIPE_BUFFER_USAGE_CONSTANT, + csc->vs_const_buf.size + ); + + /* + * Create our fragment shader's constant buffer + * Const buffer contains the color conversion matrix and bias vectors + */ + csc->fs_const_buf.size = sizeof(struct vlFragmentShaderConsts); + csc->fs_const_buf.buffer = pipe->winsys->buffer_create + ( + pipe->winsys, + 1, + PIPE_BUFFER_USAGE_CONSTANT, + csc->fs_const_buf.size + ); + + /* + * TODO: Refactor this into a seperate function, + * allow changing the CSC matrix at runtime to switch between regular & full versions + */ + memcpy + ( + pipe->winsys->buffer_map(pipe->winsys, csc->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE), + &bt_601, + sizeof(struct vlFragmentShaderConsts) + ); + + pipe->winsys->buffer_unmap(pipe->winsys, csc->fs_const_buf.buffer); + + return 0; +} + +static int vlInit +( + struct vlBasicCSC *csc +) +{ + struct pipe_context *pipe; + struct pipe_sampler_state sampler; + + assert(csc); + + pipe = csc->pipe; + + /* Delay creating the FB until vlPutSurface() so we know window size */ + csc->framebuffer.num_cbufs = 1; + csc->framebuffer.cbufs[0] = NULL; + csc->framebuffer.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.shadow_ambient = ;*/ + /*sampler.lod_bias = ;*/ + /*sampler.min_lod = ;*/ + /*sampler.max_lod = ;*/ + /*sampler.border_color[i] = ;*/ + /*sampler.max_anisotropy = ;*/ + csc->sampler = pipe->create_sampler_state(pipe, &sampler); + + vlCreateVertexShader(csc); + vlCreateFragmentShader(csc); + vlCreateDataBufs(csc); + + return 0; +} + +int vlCreateBasicCSC +( + struct pipe_context *pipe, + struct vlCSC **csc +) +{ + struct vlBasicCSC *basic_csc; + + assert(pipe); + assert(csc); + + basic_csc = calloc(1, sizeof(struct vlBasicCSC)); + + if (!basic_csc) + return 1; + + basic_csc->base.vlResizeFrameBuffer = &vlResizeFrameBuffer; + basic_csc->base.vlBegin = &vlBegin; + basic_csc->base.vlPutPicture = &vlPutPictureCSC; + basic_csc->base.vlEnd = &vlEnd; + basic_csc->base.vlGetFrameBuffer = &vlGetFrameBuffer; + basic_csc->base.vlDestroy = &vlDestroy; + basic_csc->pipe = pipe; + + vlInit(basic_csc); + + *csc = &basic_csc->base; + + return 0; +} diff --git a/src/gallium/state_trackers/g3dvl/vl_basic_csc.h b/src/gallium/state_trackers/g3dvl/vl_basic_csc.h new file mode 100644 index 0000000000..2e17f1d814 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_basic_csc.h @@ -0,0 +1,13 @@ +#ifndef vl_basic_csc_h +#define vl_basic_csc_h + +struct pipe_context; +struct vlCSC; + +int vlCreateBasicCSC +( + struct pipe_context *pipe, + struct vlCSC **csc +); + +#endif diff --git a/src/gallium/state_trackers/g3dvl/vl_context.c b/src/gallium/state_trackers/g3dvl/vl_context.c new file mode 100644 index 0000000000..fe107e406d --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_context.c @@ -0,0 +1,210 @@ +#define VL_INTERNAL +#include "vl_context.h" +#include <assert.h> +#include <stdlib.h> +#include <pipe/p_context.h> +#include <pipe/p_state.h> +#include "vl_render.h" +#include "vl_r16snorm_mc.h" +#include "vl_r16snorm_mc_buf.h" +#include "vl_csc.h" +#include "vl_basic_csc.h" + +static int vlInitCommon(struct vlContext *context) +{ + struct pipe_context *pipe; + struct pipe_rasterizer_state rast; + struct pipe_blend_state blend; + struct pipe_depth_stencil_alpha_state dsa; + unsigned int i; + + assert(context); + + pipe = context->pipe; + + 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; + /* Don't need clipping, but viewport mapping done here */ + rast.bypass_clipping = 0; + rast.bypass_vs = 0; + rast.origin_lower_left = 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] = ;*/ + context->raster = pipe->create_rasterizer_state(pipe, &rast); + pipe->bind_rasterizer_state(pipe, context->raster); + + 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; + context->blend = pipe->create_blend_state(pipe, &blend); + pipe->bind_blend_state(pipe, context->blend); + + dsa.depth.enabled = 0; + dsa.depth.writemask = 0; + dsa.depth.func = PIPE_FUNC_ALWAYS; + dsa.depth.occlusion_count = 0; + for (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].value_mask = 0; + dsa.stencil[i].write_mask = 0; + } + dsa.alpha.enabled = 0; + dsa.alpha.func = PIPE_FUNC_ALWAYS; + dsa.alpha.ref = 0; + context->dsa = pipe->create_depth_stencil_alpha_state(pipe, &dsa); + pipe->bind_depth_stencil_alpha_state(pipe, context->dsa); + + return 0; +} + +int vlCreateContext +( + struct vlScreen *screen, + struct pipe_context *pipe, + unsigned int picture_width, + unsigned int picture_height, + enum vlFormat picture_format, + enum vlProfile profile, + enum vlEntryPoint entry_point, + struct vlContext **context +) +{ + struct vlContext *ctx; + + assert(screen); + assert(context); + assert(pipe); + + ctx = calloc(1, sizeof(struct vlContext)); + + if (!ctx) + return 1; + + ctx->screen = screen; + ctx->pipe = pipe; + ctx->picture_width = picture_width; + ctx->picture_height = picture_height; + ctx->picture_format = picture_format; + ctx->profile = profile; + ctx->entry_point = entry_point; + + vlInitCommon(ctx); + + /*vlCreateR16SNormMC(pipe, picture_width, picture_height, picture_format, &ctx->render);*/ + vlCreateR16SNormBufferedMC(pipe, picture_width, picture_height, picture_format, &ctx->render); + vlCreateBasicCSC(pipe, &ctx->csc); + + *context = ctx; + + return 0; +} + +int vlDestroyContext +( + struct vlContext *context +) +{ + assert(context); + + /* XXX: Must unbind shaders before we can delete them for some reason */ + context->pipe->bind_vs_state(context->pipe, NULL); + context->pipe->bind_fs_state(context->pipe, NULL); + + context->render->vlDestroy(context->render); + context->csc->vlDestroy(context->csc); + + context->pipe->delete_blend_state(context->pipe, context->blend); + context->pipe->delete_rasterizer_state(context->pipe, context->raster); + context->pipe->delete_depth_stencil_alpha_state(context->pipe, context->dsa); + + free(context); + + return 0; +} + +struct vlScreen* vlContextGetScreen +( + struct vlContext *context +) +{ + assert(context); + + return context->screen; +} + +struct pipe_context* vlGetPipeContext +( + struct vlContext *context +) +{ + assert(context); + + return context->pipe; +} + +unsigned int vlGetPictureWidth +( + struct vlContext *context +) +{ + assert(context); + + return context->picture_width; +} + +unsigned int vlGetPictureHeight +( + struct vlContext *context +) +{ + assert(context); + + return context->picture_height; +} + +enum vlFormat vlGetPictureFormat +( + struct vlContext *context +) +{ + assert(context); + + return context->picture_format; +} diff --git a/src/gallium/state_trackers/g3dvl/vl_context.h b/src/gallium/state_trackers/g3dvl/vl_context.h new file mode 100644 index 0000000000..3d14634c44 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_context.h @@ -0,0 +1,73 @@ +#ifndef vl_context_h +#define vl_context_h + +#include "vl_types.h" + +struct pipe_context; + +#ifdef VL_INTERNAL +struct vlRender; +struct vlCSC; + +struct vlContext +{ + struct vlScreen *screen; + struct pipe_context *pipe; + unsigned int picture_width; + unsigned int picture_height; + enum vlFormat picture_format; + enum vlProfile profile; + enum vlEntryPoint entry_point; + + void *raster; + void *dsa; + void *blend; + + struct vlRender *render; + struct vlCSC *csc; +}; +#endif + +int vlCreateContext +( + struct vlScreen *screen, + struct pipe_context *pipe, + unsigned int picture_width, + unsigned int picture_height, + enum vlFormat picture_format, + enum vlProfile profile, + enum vlEntryPoint entry_point, + struct vlContext **context +); + +int vlDestroyContext +( + struct vlContext *context +); + +struct vlScreen* vlContextGetScreen +( + struct vlContext *context +); + +struct pipe_context* vlGetPipeContext +( + struct vlContext *context +); + +unsigned int vlGetPictureWidth +( + struct vlContext *context +); + +unsigned int vlGetPictureHeight +( + struct vlContext *context +); + +enum vlFormat vlGetPictureFormat +( + struct vlContext *context +); + +#endif diff --git a/src/gallium/state_trackers/g3dvl/vl_csc.h b/src/gallium/state_trackers/g3dvl/vl_csc.h new file mode 100644 index 0000000000..36417a2792 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_csc.h @@ -0,0 +1,53 @@ +#ifndef vl_csc_h +#define vl_csc_h + +#include "vl_types.h" + +struct pipe_surface; + +struct vlCSC +{ + int (*vlResizeFrameBuffer) + ( + struct vlCSC *csc, + unsigned int width, + unsigned int height + ); + + int (*vlBegin) + ( + struct vlCSC *csc + ); + + int (*vlPutPicture) + ( + struct vlCSC *csc, + struct vlSurface *surface, + int srcx, + int srcy, + int srcw, + int srch, + int destx, + int desty, + int destw, + int desth, + enum vlPictureType picture_type + ); + + int (*vlEnd) + ( + struct vlCSC *csc + ); + + struct pipe_surface* (*vlGetFrameBuffer) + ( + struct vlCSC *csc + ); + + int (*vlDestroy) + ( + struct vlCSC *csc + ); +}; + +#endif diff --git a/src/gallium/state_trackers/g3dvl/vl_defs.h b/src/gallium/state_trackers/g3dvl/vl_defs.h new file mode 100644 index 0000000000..d612d02502 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_defs.h @@ -0,0 +1,11 @@ +#ifndef vl_defs_h +#define vl_defs_h + +#define VL_BLOCK_WIDTH 8 +#define VL_BLOCK_HEIGHT 8 +#define VL_BLOCK_SIZE (VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT) +#define VL_MACROBLOCK_WIDTH 16 +#define VL_MACROBLOCK_HEIGHT 16 +#define VL_MACROBLOCK_SIZE (VL_MACROBLOCK_WIDTH * VL_MACROBLOCK_HEIGHT) + +#endif diff --git a/src/gallium/state_trackers/g3dvl/vl_display.c b/src/gallium/state_trackers/g3dvl/vl_display.c new file mode 100644 index 0000000000..af80faa7f5 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_display.c @@ -0,0 +1,48 @@ +#define VL_INTERNAL +#include "vl_display.h" +#include <assert.h> +#include <stdlib.h> + +int vlCreateDisplay +( + vlNativeDisplay native_display, + struct vlDisplay **display +) +{ + struct vlDisplay *dpy; + + assert(native_display); + assert(display); + + dpy = calloc(1, sizeof(struct vlDisplay)); + + if (!dpy) + return 1; + + dpy->native = native_display; + *display = dpy; + + return 0; +} + +int vlDestroyDisplay +( + struct vlDisplay *display +) +{ + assert(display); + + free(display); + + return 0; +} + +vlNativeDisplay vlGetNativeDisplay +( + struct vlDisplay *display +) +{ + assert(display); + + return display->native; +} diff --git a/src/gallium/state_trackers/g3dvl/vl_display.h b/src/gallium/state_trackers/g3dvl/vl_display.h new file mode 100644 index 0000000000..e11fd40799 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_display.h @@ -0,0 +1,29 @@ +#ifndef vl_display_h +#define vl_display_h + +#include "vl_types.h" + +#ifdef VL_INTERNAL +struct vlDisplay +{ + vlNativeDisplay native; +}; +#endif + +int vlCreateDisplay +( + vlNativeDisplay native_display, + struct vlDisplay **display +); + +int vlDestroyDisplay +( + struct vlDisplay *display +); + +vlNativeDisplay vlGetNativeDisplay +( + struct vlDisplay *display +); + +#endif diff --git a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.c b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.c new file mode 100644 index 0000000000..3272220ef8 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.c @@ -0,0 +1,2344 @@ +#define VL_INTERNAL +#include "vl_r16snorm_mc.h" +#include <assert.h> +#include <stdlib.h> +#include <pipe/p_context.h> +#include <pipe/p_winsys.h> +#include <pipe/p_state.h> +#include <pipe/p_inlines.h> +#include <tgsi/tgsi_parse.h> +#include <tgsi/tgsi_build.h> +#include "vl_render.h" +#include "vl_shader_build.h" +#include "vl_surface.h" +#include "vl_util.h" +#include "vl_types.h" +#include "vl_defs.h" + +#define NUM_BUFS 4 /* Number of rotating buffers to use */ + +struct vlVertexShaderConsts +{ + /*struct vlVertex4f scale; + struct vlVertex4f denorm;*/ + struct vlVertex4f scale; + struct vlVertex4f mb_pos_trans; + struct vlVertex4f denorm; + struct + { + struct vlVertex4f top_field; + struct vlVertex4f bottom_field; + } mb_tc_trans[2]; +}; + +struct vlFragmentShaderConsts +{ + struct vlVertex4f multiplier; + struct vlVertex4f div; +}; + +struct vlR16SnormMC +{ + struct vlRender base; + + 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[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]; + struct pipe_vertex_element vertex_elems[3]; + struct pipe_constant_buffer vs_const_buf, fs_const_buf; +}; + +static int vlBegin +( + struct vlRender *render +) +{ + struct vlR16SnormMC *mc; + struct pipe_context *pipe; + + assert(render); + + mc = (struct vlR16SnormMC*)render; + pipe = mc->pipe; + + /* Frame buffer set in vlRender*Macroblock() */ + /* Shaders, samplers, textures set in vlRender*Macroblock() */ + pipe->set_vertex_buffers(pipe, 3, mc->vertex_bufs); + pipe->set_vertex_elements(pipe, 3, mc->vertex_elems); + pipe->set_viewport_state(pipe, &mc->viewport); + pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 0, &mc->vs_const_buf); + pipe->set_constant_buffer(pipe, PIPE_SHADER_FRAGMENT, 0, &mc->fs_const_buf); + + return 0; +} + +/*static int vlGrabMacroBlock +( + struct vlR16SnormMC *mc, + struct vlMpeg2MacroBlock *macroblock +) +{ + assert(mc); + assert(macroblock); + + + + return 0; +}*/ + +/*#define DO_IDCT*/ + +#ifdef DO_IDCT +static int vlTransformBlock(short *src, short *dst, short bias) +{ + static const float basis[8][8] = + { + {0.3536, 0.4904, 0.4619, 0.4157, 0.3536, 0.2778, 0.1913, 0.0975}, + {0.3536, 0.4157, 0.1913, -0.0975, -0.3536, -0.4904, -0.4619, -0.2778}, + {0.3536, 0.2778, -0.1913, -0.4904, -0.3536, 0.0975, 0.4619, 0.4157}, + {0.3536, 0.0975, -0.4619, -0.2778, 0.3536, 0.4157, -0.1913, -0.4904}, + {0.3536, -0.0975, -0.4619, 0.2778, 0.3536, -0.4157, -0.1913, 0.4904}, + {0.3536, -0.2778, -0.1913, 0.4904, -0.3536, -0.0975, 0.4619, -0.4157}, + {0.3536, -0.4157, 0.1913, 0.0975, -0.3536, 0.4904, -0.4619, 0.2778}, + {0.3536, -0.4904, 0.4619, -0.4157, 0.3536, -0.2778, 0.1913, -0.0975} + }; + + unsigned int x, y; + short tmp[64]; + + for (y = 0; y < VL_BLOCK_HEIGHT; ++y) + for (x = 0; x < VL_BLOCK_WIDTH; ++x) + tmp[y * VL_BLOCK_WIDTH + x] = (short) + ( + src[y * VL_BLOCK_WIDTH + 0] * basis[x][0] + + src[y * VL_BLOCK_WIDTH + 1] * basis[x][1] + + src[y * VL_BLOCK_WIDTH + 2] * basis[x][2] + + src[y * VL_BLOCK_WIDTH + 3] * basis[x][3] + + src[y * VL_BLOCK_WIDTH + 4] * basis[x][4] + + src[y * VL_BLOCK_WIDTH + 5] * basis[x][5] + + src[y * VL_BLOCK_WIDTH + 6] * basis[x][6] + + src[y * VL_BLOCK_WIDTH + 7] * basis[x][7] + ); + + for (x = 0; x < VL_BLOCK_WIDTH; ++x) + for (y = 0; y < VL_BLOCK_HEIGHT; ++y) + { + dst[y * VL_BLOCK_WIDTH + x] = bias + (short) + ( + tmp[0 * VL_BLOCK_WIDTH + x] * basis[y][0] + + tmp[1 * VL_BLOCK_WIDTH + x] * basis[y][1] + + tmp[2 * VL_BLOCK_WIDTH + x] * basis[y][2] + + tmp[3 * VL_BLOCK_WIDTH + x] * basis[y][3] + + tmp[4 * VL_BLOCK_WIDTH + x] * basis[y][4] + + tmp[5 * VL_BLOCK_WIDTH + x] * basis[y][5] + + tmp[6 * VL_BLOCK_WIDTH + x] * basis[y][6] + + tmp[7 * VL_BLOCK_WIDTH + x] * basis[y][7] + ); + if (dst[y * VL_BLOCK_WIDTH + x] > 255) + dst[y * VL_BLOCK_WIDTH + x] = 255; + else if (bias > 0 && dst[y * VL_BLOCK_WIDTH + x] < 0) + dst[y * VL_BLOCK_WIDTH + x] = 0; + } + return 0; +} +#endif + +static int vlGrabFrameCodedBlock(short *src, short *dst, unsigned int dst_pitch) +{ + unsigned int y; + + for (y = 0; y < VL_BLOCK_HEIGHT; ++y) + memcpy + ( + dst + y * dst_pitch, + src + y * VL_BLOCK_WIDTH, + VL_BLOCK_WIDTH * 2 + ); + + return 0; +} + +static int vlGrabFieldCodedBlock(short *src, short *dst, unsigned int dst_pitch) +{ + unsigned int y; + + for (y = 0; y < VL_BLOCK_HEIGHT / 2; ++y) + memcpy + ( + dst + y * dst_pitch * 2, + src + y * VL_BLOCK_WIDTH, + VL_BLOCK_WIDTH * 2 + ); + + dst += VL_BLOCK_HEIGHT * dst_pitch; + + for (; y < VL_BLOCK_HEIGHT; ++y) + memcpy + ( + dst + y * dst_pitch * 2, + src + y * VL_BLOCK_WIDTH, + VL_BLOCK_WIDTH * 2 + ); + + return 0; +} + +static int vlGrabNoBlock(short *dst, unsigned int dst_pitch) +{ + unsigned int y; + + for (y = 0; y < VL_BLOCK_HEIGHT; ++y) + memset + ( + dst + y * dst_pitch, + 0, + VL_BLOCK_WIDTH * 2 + ); + + return 0; +} + +enum vlSampleType +{ + vlSampleTypeFull, + vlSampleTypeDiff +}; + +static int vlGrabBlocks +( + struct vlR16SnormMC *mc, + unsigned int coded_block_pattern, + enum vlDCTType dct_type, + enum vlSampleType sample_type, + short *blocks +) +{ + struct pipe_surface *tex_surface; + short *texels; + unsigned int tex_pitch; + unsigned int tb, sb = 0; + + assert(mc); + assert(blocks); + + tex_surface = mc->pipe->screen->get_tex_surface + ( + mc->pipe->screen, + mc->textures[mc->cur_buf % NUM_BUFS][0], + 0, 0, 0, PIPE_BUFFER_USAGE_CPU_WRITE + ); + + texels = pipe_surface_map(tex_surface, PIPE_BUFFER_USAGE_CPU_WRITE); + tex_pitch = tex_surface->stride / tex_surface->block.size; + + for (tb = 0; tb < 4; ++tb) + { + if ((coded_block_pattern >> (5 - tb)) & 1) + { + short *cur_block = blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT; + +#ifdef DO_IDCT + vlTransformBlock(cur_block, cur_block, sample_type == vlSampleTypeFull ? 128 : 0); +#endif + + if (dct_type == vlDCTTypeFrameCoded) + vlGrabFrameCodedBlock + ( + cur_block, + texels + tb * tex_pitch * VL_BLOCK_HEIGHT, + tex_pitch + ); + else + vlGrabFieldCodedBlock + ( + cur_block, + texels + (tb % 2) * tex_pitch * VL_BLOCK_HEIGHT + (tb / 2) * tex_pitch, + tex_pitch + ); + + ++sb; + } + else + vlGrabNoBlock(texels + tb * tex_pitch * VL_BLOCK_HEIGHT, tex_pitch); + } + + pipe_surface_unmap(tex_surface); + + /* TODO: Implement 422, 444 */ + for (tb = 0; tb < 2; ++tb) + { + tex_surface = mc->pipe->screen->get_tex_surface + ( + mc->pipe->screen, + mc->textures[mc->cur_buf % NUM_BUFS][tb + 1], + 0, 0, 0, PIPE_BUFFER_USAGE_CPU_WRITE + ); + + texels = pipe_surface_map(tex_surface, PIPE_BUFFER_USAGE_CPU_WRITE); + tex_pitch = tex_surface->stride / tex_surface->block.size; + + if ((coded_block_pattern >> (1 - tb)) & 1) + { + short *cur_block = blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT; + +#ifdef DO_IDCT + vlTransformBlock(cur_block, cur_block, sample_type == vlSampleTypeFull ? 128 : 0); +#endif + + vlGrabFrameCodedBlock + ( + cur_block, + texels, + tex_pitch + ); + + ++sb; + } + else + vlGrabNoBlock(texels, tex_pitch); + + pipe_surface_unmap(tex_surface); + } + + return 0; +} + +static int vlRenderIMacroBlock +( + struct vlR16SnormMC *mc, + enum vlPictureType picture_type, + enum vlFieldOrder field_order, + unsigned int mbx, + unsigned int mby, + unsigned int coded_block_pattern, + enum vlDCTType dct_type, + short *blocks, + struct vlSurface *surface +) +{ + struct pipe_context *pipe; + struct vlVertexShaderConsts *vs_consts; + + assert(blocks); + assert(surface); + + /* TODO: Implement interlaced rendering */ + if (picture_type != vlPictureTypeFrame) + return 0; + + vlGrabBlocks(mc, coded_block_pattern, dct_type, vlSampleTypeFull, blocks); + + pipe = mc->pipe; + + vs_consts = pipe->winsys->buffer_map + ( + pipe->winsys, + mc->vs_const_buf.buffer, + PIPE_BUFFER_USAGE_CPU_WRITE + ); + + vs_consts->scale.x = VL_MACROBLOCK_WIDTH / (float)surface->texture->width[0]; + vs_consts->scale.y = VL_MACROBLOCK_HEIGHT / (float)surface->texture->height[0]; + vs_consts->scale.z = 1.0f; + vs_consts->scale.w = 1.0f; + vs_consts->mb_pos_trans.x = (mbx * VL_MACROBLOCK_WIDTH) / (float)surface->texture->width[0]; + vs_consts->mb_pos_trans.y = (mby * VL_MACROBLOCK_HEIGHT) / (float)surface->texture->height[0]; + vs_consts->mb_pos_trans.z = 0.0f; + vs_consts->mb_pos_trans.w = 0.0f; + + pipe->winsys->buffer_unmap(pipe->winsys, mc->vs_const_buf.buffer); + + mc->render_target.cbufs[0] = pipe->screen->get_tex_surface + ( + pipe->screen, + surface->texture, + 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[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; +} + +static int vlRenderPMacroBlock +( + struct vlR16SnormMC *mc, + enum vlPictureType picture_type, + enum vlFieldOrder field_order, + unsigned int mbx, + unsigned int mby, + enum vlMotionType mc_type, + short top_x, + short top_y, + short bottom_x, + short bottom_y, + unsigned int coded_block_pattern, + enum vlDCTType dct_type, + short *blocks, + struct vlSurface *ref_surface, + struct vlSurface *surface +) +{ + struct pipe_context *pipe; + struct vlVertexShaderConsts *vs_consts; + + assert(motion_vectors); + assert(blocks); + assert(ref_surface); + assert(surface); + + /* TODO: Implement interlaced rendering */ + if (picture_type != vlPictureTypeFrame) + return 0; + /* TODO: Implement other MC types */ + if (mc_type != vlMotionTypeFrame && mc_type != vlMotionTypeField) + return 0; + + vlGrabBlocks(mc, coded_block_pattern, dct_type, vlSampleTypeDiff, blocks); + + pipe = mc->pipe; + + vs_consts = pipe->winsys->buffer_map + ( + pipe->winsys, + mc->vs_const_buf.buffer, + PIPE_BUFFER_USAGE_CPU_WRITE + ); + + vs_consts->scale.x = VL_MACROBLOCK_WIDTH / (float)surface->texture->width[0]; + vs_consts->scale.y = VL_MACROBLOCK_HEIGHT / (float)surface->texture->height[0]; + vs_consts->scale.z = 1.0f; + vs_consts->scale.w = 1.0f; + vs_consts->mb_pos_trans.x = (mbx * VL_MACROBLOCK_WIDTH) / (float)surface->texture->width[0]; + vs_consts->mb_pos_trans.y = (mby * VL_MACROBLOCK_HEIGHT) / (float)surface->texture->height[0]; + vs_consts->mb_pos_trans.z = 0.0f; + vs_consts->mb_pos_trans.w = 0.0f; + vs_consts->mb_tc_trans[0].top_field.x = (mbx * VL_MACROBLOCK_WIDTH + top_x * 0.5f) / (float)surface->texture->width[0]; + vs_consts->mb_tc_trans[0].top_field.y = (mby * VL_MACROBLOCK_HEIGHT + top_y * 0.5f) / (float)surface->texture->height[0]; + vs_consts->mb_tc_trans[0].top_field.z = 0.0f; + vs_consts->mb_tc_trans[0].top_field.w = 0.0f; + + if (mc_type == vlMotionTypeField) + { + vs_consts->denorm.x = (float)surface->texture->width[0]; + vs_consts->denorm.y = (float)surface->texture->height[0]; + + vs_consts->mb_tc_trans[0].bottom_field.x = (mbx * VL_MACROBLOCK_WIDTH + bottom_x * 0.5f) / (float)surface->texture->width[0]; + vs_consts->mb_tc_trans[0].bottom_field.y = (mby * VL_MACROBLOCK_HEIGHT + bottom_y * 0.5f) / (float)surface->texture->height[0]; + vs_consts->mb_tc_trans[0].bottom_field.z = 0.0f; + vs_consts->mb_tc_trans[0].bottom_field.w = 0.0f; + + pipe->bind_vs_state(pipe, mc->p_vs[1]); + pipe->bind_fs_state(pipe, mc->p_fs[1]); + } + else + { + pipe->bind_vs_state(pipe, mc->p_vs[0]); + pipe->bind_fs_state(pipe, mc->p_fs[0]); + } + + pipe->winsys->buffer_unmap(pipe->winsys, mc->vs_const_buf.buffer); + + mc->render_target.cbufs[0] = pipe->screen->get_tex_surface + ( + pipe->screen, + surface->texture, + 0, 0, 0, PIPE_BUFFER_USAGE_GPU_READ | PIPE_BUFFER_USAGE_GPU_WRITE + ); + pipe->set_framebuffer_state(pipe, &mc->render_target); + + 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; +} + +static int vlRenderBMacroBlock +( + struct vlR16SnormMC *mc, + enum vlPictureType picture_type, + enum vlFieldOrder field_order, + unsigned int mbx, + unsigned int mby, + enum vlMotionType mc_type, + short top_past_x, + short top_past_y, + short bottom_past_x, + short bottom_past_y, + short top_future_x, + short top_future_y, + short bottom_future_x, + short bottom_future_y, + unsigned int coded_block_pattern, + enum vlDCTType dct_type, + short *blocks, + struct vlSurface *past_surface, + struct vlSurface *future_surface, + struct vlSurface *surface +) +{ + struct pipe_context *pipe; + struct vlVertexShaderConsts *vs_consts; + + assert(motion_vectors); + assert(blocks); + assert(ref_surface); + assert(surface); + + /* TODO: Implement interlaced rendering */ + if (picture_type != vlPictureTypeFrame) + return 0; + /* TODO: Implement other MC types */ + if (mc_type != vlMotionTypeFrame && mc_type != vlMotionTypeField) + return 0; + + vlGrabBlocks(mc, coded_block_pattern, dct_type, vlSampleTypeDiff, blocks); + + pipe = mc->pipe; + + vs_consts = pipe->winsys->buffer_map + ( + pipe->winsys, + mc->vs_const_buf.buffer, + PIPE_BUFFER_USAGE_CPU_WRITE + ); + + vs_consts->scale.x = VL_MACROBLOCK_WIDTH / (float)surface->texture->width[0]; + vs_consts->scale.y = VL_MACROBLOCK_HEIGHT / (float)surface->texture->height[0]; + vs_consts->scale.z = 1.0f; + vs_consts->scale.w = 1.0f; + vs_consts->mb_pos_trans.x = (mbx * VL_MACROBLOCK_WIDTH) / (float)surface->texture->width[0]; + vs_consts->mb_pos_trans.y = (mby * VL_MACROBLOCK_HEIGHT) / (float)surface->texture->height[0]; + vs_consts->mb_pos_trans.z = 0.0f; + vs_consts->mb_pos_trans.w = 0.0f; + vs_consts->mb_tc_trans[0].top_field.x = (mbx * VL_MACROBLOCK_WIDTH + top_past_x * 0.5f) / (float)surface->texture->width[0]; + vs_consts->mb_tc_trans[0].top_field.y = (mby * VL_MACROBLOCK_HEIGHT + top_past_y * 0.5f) / (float)surface->texture->height[0]; + vs_consts->mb_tc_trans[0].top_field.z = 0.0f; + vs_consts->mb_tc_trans[0].top_field.w = 0.0f; + vs_consts->mb_tc_trans[1].top_field.x = (mbx * VL_MACROBLOCK_WIDTH + top_future_x * 0.5f) / (float)surface->texture->width[0]; + vs_consts->mb_tc_trans[1].top_field.y = (mby * VL_MACROBLOCK_HEIGHT + top_future_y * 0.5f) / (float)surface->texture->height[0]; + vs_consts->mb_tc_trans[1].top_field.z = 0.0f; + vs_consts->mb_tc_trans[1].top_field.w = 0.0f; + + if (mc_type == vlMotionTypeField) + { + vs_consts->denorm.x = (float)surface->texture->width[0]; + vs_consts->denorm.y = (float)surface->texture->height[0]; + + vs_consts->mb_tc_trans[0].bottom_field.x = (mbx * VL_MACROBLOCK_WIDTH + bottom_past_x * 0.5f) / (float)surface->texture->width[0]; + vs_consts->mb_tc_trans[0].bottom_field.y = (mby * VL_MACROBLOCK_HEIGHT + bottom_past_y * 0.5f) / (float)surface->texture->height[0]; + vs_consts->mb_tc_trans[0].bottom_field.z = 0.0f; + vs_consts->mb_tc_trans[0].bottom_field.w = 0.0f; + vs_consts->mb_tc_trans[1].bottom_field.x = (mbx * VL_MACROBLOCK_WIDTH + bottom_future_x * 0.5f) / (float)surface->texture->width[0]; + vs_consts->mb_tc_trans[1].bottom_field.y = (mby * VL_MACROBLOCK_HEIGHT + bottom_future_y * 0.5f) / (float)surface->texture->height[0]; + vs_consts->mb_tc_trans[1].bottom_field.z = 0.0f; + vs_consts->mb_tc_trans[1].bottom_field.w = 0.0f; + + pipe->bind_vs_state(pipe, mc->b_vs[1]); + pipe->bind_fs_state(pipe, mc->b_fs[1]); + } + else + { + pipe->bind_vs_state(pipe, mc->b_vs[0]); + pipe->bind_fs_state(pipe, mc->b_fs[0]); + } + + pipe->winsys->buffer_unmap(pipe->winsys, mc->vs_const_buf.buffer); + + mc->render_target.cbufs[0] = pipe->screen->get_tex_surface + ( + pipe->screen, + surface->texture, + 0, 0, 0, PIPE_BUFFER_USAGE_GPU_READ | PIPE_BUFFER_USAGE_GPU_WRITE + ); + pipe->set_framebuffer_state(pipe, &mc->render_target); + + 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; +} + +static int vlRenderMacroBlocksMpeg2R16Snorm +( + struct vlRender *render, + struct vlMpeg2MacroBlockBatch *batch, + struct vlSurface *surface +) +{ + struct vlR16SnormMC *mc; + unsigned int i; + + assert(render); + + mc = (struct vlR16SnormMC*)render; + + /*for (i = 0; i < batch->num_macroblocks; ++i) + vlGrabMacroBlock(batch->macroblocks[i]);*/ + + for (i = 0; i < batch->num_macroblocks; ++i) + { + switch (batch->macroblocks[i].mb_type) + { + case vlMacroBlockTypeIntra: + { + vlRenderIMacroBlock + ( + mc, + batch->picture_type, + batch->field_order, + batch->macroblocks[i].mbx, + batch->macroblocks[i].mby, + batch->macroblocks[i].cbp, + batch->macroblocks[i].dct_type, + batch->macroblocks[i].blocks, + surface + ); + break; + } + case vlMacroBlockTypeFwdPredicted: + { + vlRenderPMacroBlock + ( + mc, + batch->picture_type, + batch->field_order, + batch->macroblocks[i].mbx, + batch->macroblocks[i].mby, + batch->macroblocks[i].mo_type, + batch->macroblocks[i].PMV[0][0][0], + batch->macroblocks[i].PMV[0][0][1], + batch->macroblocks[i].PMV[1][0][0], + batch->macroblocks[i].PMV[1][0][1], + batch->macroblocks[i].cbp, + batch->macroblocks[i].dct_type, + batch->macroblocks[i].blocks, + batch->past_surface, + surface + ); + break; + } + case vlMacroBlockTypeBkwdPredicted: + { + vlRenderPMacroBlock + ( + mc, + batch->picture_type, + batch->field_order, + batch->macroblocks[i].mbx, + batch->macroblocks[i].mby, + batch->macroblocks[i].mo_type, + batch->macroblocks[i].PMV[0][1][0], + batch->macroblocks[i].PMV[0][1][1], + batch->macroblocks[i].PMV[1][1][0], + batch->macroblocks[i].PMV[1][1][1], + batch->macroblocks[i].cbp, + batch->macroblocks[i].dct_type, + batch->macroblocks[i].blocks, + batch->future_surface, + surface + ); + break; + } + case vlMacroBlockTypeBiPredicted: + { + vlRenderBMacroBlock + ( + mc, + batch->picture_type, + batch->field_order, + batch->macroblocks[i].mbx, + batch->macroblocks[i].mby, + batch->macroblocks[i].mo_type, + batch->macroblocks[i].PMV[0][0][0], + batch->macroblocks[i].PMV[0][0][1], + batch->macroblocks[i].PMV[1][0][0], + batch->macroblocks[i].PMV[1][0][1], + batch->macroblocks[i].PMV[0][1][0], + batch->macroblocks[i].PMV[0][1][1], + batch->macroblocks[i].PMV[1][1][0], + batch->macroblocks[i].PMV[1][1][1], + batch->macroblocks[i].cbp, + batch->macroblocks[i].dct_type, + batch->macroblocks[i].blocks, + batch->past_surface, + batch->future_surface, + surface + ); + break; + } + default: + assert(0); + } + } + + return 0; +} + +static int vlEnd +( + struct vlRender *render +) +{ + assert(render); + + return 0; +} + +static int vlFlush +( + struct vlRender *render +) +{ + assert(render); + + return 0; +} + +static int vlDestroy +( + struct vlRender *render +) +{ + struct vlR16SnormMC *mc; + struct pipe_context *pipe; + unsigned int i; + + assert(render); + + mc = (struct vlR16SnormMC*)render; + pipe = mc->pipe; + + for (i = 0; i < 5; ++i) + pipe->delete_sampler_state(pipe, mc->samplers[i]); + + for (i = 0; i < 3; ++i) + 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 < 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); + + for (i = 0; i < 2; ++i) + { + pipe->delete_vs_state(pipe, mc->p_vs[i]); + pipe->delete_fs_state(pipe, mc->p_fs[i]); + pipe->delete_vs_state(pipe, mc->b_vs[i]); + pipe->delete_fs_state(pipe, mc->b_fs[i]); + } + + pipe->winsys->buffer_destroy(pipe->winsys, mc->vs_const_buf.buffer); + pipe->winsys->buffer_destroy(pipe->winsys, mc->fs_const_buf.buffer); + + free(mc); + + return 0; +} + +/* + * 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. + */ +static 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. + */ +static 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. + */ +static 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. + */ +static 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. + */ +static 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). + */ +static const struct vlFragmentShaderConsts fs_consts = +{ + {32767.0f / 255.0f, 32767.0f / 255.0f, 32767.0f / 255.0f, 0.0f}, + {0.5f, 2.0f, 0.0f, 0.0f} +}; + +static int vlCreateVertexShaderIMB +( + struct vlR16SnormMC *mc +) +{ + const unsigned int max_tokens = 50; + + struct pipe_context *pipe; + struct pipe_shader_state vs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 texcoords + */ + for (i = 0; i < 3; 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 unit rect to macroblock size + * decl c1 ; Translation vector to move macroblock into position + */ + decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 1); + ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti); + + /* + * decl o0 ; Vertex pos + * decl o1 ; Luma texcoords + * decl o2 ; Chroma texcoords + */ + for (i = 0; i < 3; 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 */ + decl = vl_decl_temps(0, 0); + ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti); + + /* mul t0, i0, c0 ; Scale unit rect to normalized MB size */ + inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, 0, TGSI_FILE_CONSTANT, 0); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* add o0, t0, c1 ; Translate rect into position */ + inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 1); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* + * mov o1, i1 ; Move input luma texcoords to output + * mov o2, i2 ; Move input chroma texcoords to output + */ + for (i = 1; i < 3; ++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); + + vs.tokens = tokens; + mc->i_vs = pipe->create_vs_state(pipe, &vs); + free(tokens); + + return 0; +} + +static int vlCreateFragmentShaderIMB +( + struct vlR16SnormMC *mc +) +{ + const unsigned int max_tokens = 100; + + struct pipe_context *pipe; + struct pipe_shader_state fs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header); + + ti = 3; + + /* + * decl i0 ; Texcoords for s0 + * decl i1 ; Texcoords for s1, s2 + */ + for (i = 0; i < 2; ++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 (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, i1, s2 ; Read texel from chroma Cr texture + * mov t0.z, t1.x ; Move Cr sample into .z component + */ + for (i = 0; i < 3; ++i) + { + inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i > 0 ? 1 : 0, 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); + + fs.tokens = tokens; + mc->i_fs = pipe->create_fs_state(pipe, &fs); + free(tokens); + + return 0; +} + +static int vlCreateVertexShaderFramePMB +( + struct vlR16SnormMC *mc +) +{ + const unsigned int max_tokens = 100; + + struct pipe_context *pipe; + struct pipe_shader_state vs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 texcoords + */ + for (i = 0; i < 3; 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 unit rect to macroblock size + * decl c1 ; Translation vector to move macroblock into position + * decl c2 ; Unused + * decl c3 ; Translation vector to move ref macroblock texcoords 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 ; Luma texcoords + * decl o2 ; Chroma texcoords + * decl o3 ; Ref macroblock texcoords + */ + for (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); + } + + /* decl t0 */ + decl = vl_decl_temps(0, 0); + ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti); + + /* mul t0, i0, c0 ; Scale unit rect to normalized MB size */ + inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, 0, TGSI_FILE_CONSTANT, 0); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* add o0, t0, c1 ; Translate rect into position */ + inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 1); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* + * mov o1, i1 ; Move input luma texcoords to output + * mov o2, i2 ; Move input chroma texcoords to output + */ + for (i = 1; i < 3; ++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 o3, t0, c3 ; Translate rect into position on ref macroblock */ + inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 3, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 3); + 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); + + vs.tokens = tokens; + mc->p_vs[0] = pipe->create_vs_state(pipe, &vs); + free(tokens); + + return 0; +} + +static int vlCreateVertexShaderFieldPMB +( + struct vlR16SnormMC *mc +) +{ + const unsigned int max_tokens = 100; + + struct pipe_context *pipe; + struct pipe_shader_state vs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 texcoords + */ + for (i = 0; i < 3; 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 unit rect to macroblock size + * decl c1 ; Translation vector to move macroblock into position + * decl c2 ; Denorm coefficients + * decl c3 ; Translation vector to move top field ref macroblock texcoords into position + * decl c4 ; Translation vector to move bottom field ref macroblock texcoords into position + */ + decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 4); + ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti); + + /* + * decl o0 ; Vertex pos + * decl o1 ; Luma texcoords + * decl o2 ; Chroma texcoords + * decl o3 ; Top field ref macroblock texcoords + * decl o4 ; Bottom field ref macroblock texcoords + * decl o5 ; Denormalized vertex pos + */ + for (i = 0; i < 6; i++) + { + decl = vl_decl_output((i == 0 || i == 5) ? 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); + + /* mul t0, i0, c0 ; Scale unit rect to normalized MB size */ + inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, 0, TGSI_FILE_CONSTANT, 0); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* add t1, t0, c1 ; Translate rect into position */ + inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 1); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* mov o0, t1 ; Move vertex pos to output */ + inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 1); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* + mov o1, i1 ; Move input luma texcoords to output + mov o2, i2 ; Move input chroma texcoords to output + */ + for (i = 1; i < 3; ++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 o3, t0, c3 ; Translate top field rect into position on ref macroblock + add o4, t0, c4 ; Translate bottom field rect into position on ref macroblock */ + for (i = 0; i < 2; ++i) + { + inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, i + 3, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, i + 3); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + } + + /* mul o5, t1, c2 ; Denorm vertex pos */ + inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_OUTPUT, 5, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 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); + + vs.tokens = tokens; + mc->p_vs[1] = pipe->create_vs_state(pipe, &vs); + free(tokens); + + return 0; +} + +static int vlCreateFragmentShaderFramePMB +( + struct vlR16SnormMC *mc +) +{ + const unsigned int max_tokens = 100; + + struct pipe_context *pipe; + struct pipe_shader_state fs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header); + + ti = 3; + + /* + * decl i0 ; Texcoords for s0 + * decl i1 ; Texcoords for s1, s2 + * decl i2 ; Texcoords for s3 + */ + for (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 + * decl s3 ; Sampler for ref surface texture + */ + for (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, i1, s2 ; Read texel from chroma Cr texture + * mov t0.z, t1.x ; Move Cr sample into .z component + */ + for (i = 0; i < 3; ++i) + { + inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i > 0 ? 1 : 0, 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, i2, s3 ; Read texel from ref macroblock */ + inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, 2, 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); + + fs.tokens = tokens; + mc->p_fs[0] = pipe->create_fs_state(pipe, &fs); + free(tokens); + + return 0; +} + +static int vlCreateFragmentShaderFieldPMB +( + struct vlR16SnormMC *mc +) +{ + const unsigned int max_tokens = 200; + + struct pipe_context *pipe; + struct pipe_shader_state fs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header); + + ti = 3; + + /* + * decl i0 ; Texcoords for s0 + * decl i1 ; Texcoords for s1, s2 + * decl i2 ; Texcoords for s3 + * decl i3 ; Texcoords for s3 + * decl i4 ; Denormalized vertex pos + */ + for (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 ; Constants 1/2 & 2 in .x, .y channels for Y-mod-2 top/bottom field selection + */ + 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-t4 */ + decl = vl_decl_temps(0, 4); + 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 (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, i1, s2 ; Read texel from chroma Cr texture + * mov t0.z, t1.x ; Move Cr sample into .z component + */ + for (i = 0; i < 3; ++i) + { + inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i > 0 ? 1 : 0, 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, i2, s3 ; Read texel from ref macroblock top field + * tex2d t2, i3, s3 ; Read texel from ref macroblock bottom field + */ + for (i = 0; i < 2; ++i) + { + inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 1, TGSI_FILE_INPUT, i + 2, TGSI_FILE_SAMPLER, 3); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + } + + /* XXX: Pos values off by 0.5? */ + /* sub t4, i4.y, c1.x ; Sub 0.5 from denormalized pos */ + inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_INPUT, 4, TGSI_FILE_CONSTANT, 1); + inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_X; + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* mul t3, t4, c1.x ; Multiply pos Y-coord by 1/2 */ + inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_CONSTANT, 1); + inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_X; + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* floor t3, t3 ; Get rid of fractional part */ + inst = vl_inst2(TGSI_OPCODE_FLOOR, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* mul t3, t3, c1.y ; Multiply by 2 */ + inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_CONSTANT, 1); + inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y; + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* sub t3, t4, t3 ; Subtract from original Y to get Y % 2 */ + inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_TEMPORARY, 3); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* TODO: Move to conditional tex fetch on t3 instead of lerp */ + /* lerp t1, t3, t1, t2 ; Choose between top and bottom fields based on Y % 2 */ + inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2); + 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); + + fs.tokens = tokens; + mc->p_fs[1] = pipe->create_fs_state(pipe, &fs); + free(tokens); + + return 0; +} + +static int vlCreateVertexShaderFrameBMB +( + struct vlR16SnormMC *mc +) +{ + const unsigned int max_tokens = 100; + + struct pipe_context *pipe; + struct pipe_shader_state vs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 texcoords + */ + for (i = 0; i < 3; 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 unit rect to macroblock size + * decl c1 ; Translation vector to move macroblock into position + * decl c2 ; Unused + * decl c3 ; Translation vector to move past ref macroblock texcoords into position + * decl c4 ; Unused + * decl c5 ; Translation vector to move future ref macroblock texcoords into position + */ + decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 5); + ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti); + + /* + * decl o0 ; Vertex pos + * decl o1 ; Luma texcoords + * decl o2 ; Chroma texcoords + * decl o3 ; Past ref macroblock texcoords + * decl o4 ; Future ref macroblock texcoords + */ + for (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); + } + + /* decl t0 */ + decl = vl_decl_temps(0, 0); + ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti); + + /* mul t0, i0, c0 ; Scale unit rect to normalized MB size */ + inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, 0, TGSI_FILE_CONSTANT, 0); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* add o0, t0, c1 ; Translate rect into position */ + inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 1); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* + * mov o1, i1 ; Move input luma texcoords to output + * mov o2, i2 ; Move input chroma texcoords to output + */ + for (i = 1; i < 3; ++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 o3, t0, c3 ; Translate rect into position on past ref macroblock + add o4, t0, c5 ; Translate rect into position on future ref macroblock */ + for (i = 0; i < 2; ++i) + { + inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, i + 3, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, i * 2 + 3); + 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); + + vs.tokens = tokens; + mc->b_vs[0] = pipe->create_vs_state(pipe, &vs); + free(tokens); + + return 0; +} + +static int vlCreateVertexShaderFieldBMB +( + struct vlR16SnormMC *mc +) +{ + const unsigned int max_tokens = 100; + + struct pipe_context *pipe; + struct pipe_shader_state vs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 texcoords + */ + for (i = 0; i < 3; 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 unit rect to macroblock size + * decl c1 ; Translation vector to move macroblock into position + * decl c2 ; Denorm coefficients + * decl c3 ; Translation vector to move top field past ref macroblock texcoords into position + * decl c4 ; Translation vector to move bottom field past ref macroblock texcoords into position + * decl c5 ; Translation vector to move top field future ref macroblock texcoords into position + * decl c6 ; Translation vector to move bottom field future ref macroblock texcoords into position + */ + decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 6); + ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti); + + /* + * decl o0 ; Vertex pos + * decl o1 ; Luma texcoords + * decl o2 ; Chroma texcoords + * decl o3 ; Top field past ref macroblock texcoords + * decl o4 ; Bottom field past ref macroblock texcoords + * decl o5 ; Top field future ref macroblock texcoords + * decl o6 ; Bottom field future ref macroblock texcoords + * decl o7 ; Denormalized vertex pos + */ + for (i = 0; i < 8; i++) + { + decl = vl_decl_output((i == 0 || i == 7) ? 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); + + /* mul t0, i0, c0 ; Scale unit rect to normalized MB size */ + inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, 0, TGSI_FILE_CONSTANT, 0); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* add t1, t0, c1 ; Translate rect into position */ + inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 1); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* mov o0, t1 ; Move vertex pos to output */ + inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 1); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* + * mov o1, i1 ; Move input luma texcoords to output + * mov o2, i2 ; Move input chroma texcoords to output + */ + for (i = 1; i < 3; ++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 o3, t0, c3 ; Translate top field rect into position on past ref macroblock + * add o4, t0, c4 ; Translate bottom field rect into position on past ref macroblock + * add o5, t0, c5 ; Translate top field rect into position on future ref macroblock + * add o6, t0, c6 ; Translate bottom field rect into position on future ref macroblock + */ + for (i = 0; i < 4; ++i) + { + inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, i + 3, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, i + 3); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + } + + /* mul o7, t1, c2 ; Denorm vertex pos */ + inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_OUTPUT, 7, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 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); + + vs.tokens = tokens; + mc->b_vs[1] = pipe->create_vs_state(pipe, &vs); + free(tokens); + + return 0; +} + +static int vlCreateFragmentShaderFrameBMB +( + struct vlR16SnormMC *mc +) +{ + const unsigned int max_tokens = 100; + + struct pipe_context *pipe; + struct pipe_shader_state fs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header); + + ti = 3; + + /* + * decl i0 ; Texcoords for s0 + * decl i1 ; Texcoords for s1, s2 + * decl i2 ; Texcoords for s3 + * decl i3 ; Texcoords for s4 + */ + for (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 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 past ref surface texture + * decl s4 ; Sampler for future ref surface texture + */ + for (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, i1, s2 ; Read texel from chroma Cr texture + * mov t0.z, t1.x ; Move Cr sample into .z component + */ + for (i = 0; i < 3; ++i) + { + inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i > 0 ? 1 : 0, 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, i2, s3 ; Read texel from past ref macroblock + * tex2d t2, i3, s4 ; Read texel from future ref macroblock + */ + for (i = 0; i < 2; ++i) + { + inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 1, TGSI_FILE_INPUT, i + 2, 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_LERP, 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); + + fs.tokens = tokens; + mc->b_fs[0] = pipe->create_fs_state(pipe, &fs); + free(tokens); + + return 0; +} + +static int vlCreateFragmentShaderFieldBMB +( + struct vlR16SnormMC *mc +) +{ + const unsigned int max_tokens = 200; + + struct pipe_context *pipe; + struct pipe_shader_state fs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header); + + ti = 3; + + /* + * decl i0 ; Texcoords for s0 + * decl i1 ; Texcoords for s1, s2 + * decl i2 ; Texcoords for s3 + * decl i3 ; Texcoords for s3 + * decl i4 ; Texcoords for s4 + * decl i5 ; Texcoords for s4 + * decl i6 ; Denormalized vertex pos + */ + for (i = 0; i < 7; ++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 ; Constants 1/2 & 2 in .x, .y channels to use as weight to blend past and future texels + * ; and for Y-mod-2 top/bottom field selection + */ + 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-t5 */ + decl = vl_decl_temps(0, 5); + 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 past ref surface texture + * decl s4 ; Sampler for future ref surface texture + */ + for (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, i1, s2 ; Read texel from chroma Cr texture + * mov t0.z, t1.x ; Move Cr sample into .z component + */ + for (i = 0; i < 3; ++i) + { + inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i > 0 ? 1 : 0, 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); + + /* XXX: Pos values off by 0.5? */ + /* sub t4, i6.y, c1.x ; Sub 0.5 from denormalized pos */ + inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_INPUT, 6, TGSI_FILE_CONSTANT, 1); + inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_X; + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* mul t3, t4, c1.x ; Multiply pos Y-coord by 1/2 */ + inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_CONSTANT, 1); + inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_X; + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* floor t3, t3 ; Get rid of fractional part */ + inst = vl_inst2(TGSI_OPCODE_FLOOR, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* mul t3, t3, c1.y ; Multiply by 2 */ + inst = vl_inst3( TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_CONSTANT, 1); + inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y; + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* sub t3, t4, t3 ; Subtract from original Y to get Y % 2 */ + inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_TEMPORARY, 3); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* + * tex2d t1, i2, s3 ; Read texel from past ref macroblock top field + * tex2d t2, i3, s3 ; Read texel from past ref macroblock bottom field + */ + for (i = 0; i < 2; ++i) + { + inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 1, TGSI_FILE_INPUT, i + 2, TGSI_FILE_SAMPLER, 3); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + } + + /* TODO: Move to conditional tex fetch on t3 instead of lerp */ + /* lerp t1, t3, t1, t2 ; Choose between top and bottom fields based on Y % 2 */ + inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* + * tex2d t4, i4, s4 ; Read texel from future ref macroblock top field + * tex2d t5, i5, s4 ; Read texel from future ref macroblock bottom field + */ + for (i = 0; i < 2; ++i) + { + inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 4, TGSI_FILE_INPUT, i + 4, TGSI_FILE_SAMPLER, 4); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + } + + /* TODO: Move to conditional tex fetch on t3 instead of lerp */ + /* lerp t2, t3, t4, t5 ; Choose between top and bottom fields based on Y % 2 */ + inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 2, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_TEMPORARY, 5); + 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_LERP, 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); + + fs.tokens = tokens; + mc->b_fs[1] = pipe->create_fs_state(pipe, &fs); + free(tokens); + + return 0; +} + +static int vlCreateDataBufs +( + struct vlR16SnormMC *mc +) +{ + struct pipe_context *pipe; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + + /* Create our vertex buffer and vertex buffer element */ + mc->vertex_bufs[0].pitch = sizeof(struct vlVertex2f); + mc->vertex_bufs[0].max_index = 23; + mc->vertex_bufs[0].buffer_offset = 0; + mc->vertex_bufs[0].buffer = pipe->winsys->buffer_create + ( + pipe->winsys, + 1, + PIPE_BUFFER_USAGE_VERTEX, + sizeof(struct vlVertex2f) * 24 + ); + + mc->vertex_elems[0].src_offset = 0; + mc->vertex_elems[0].vertex_buffer_index = 0; + mc->vertex_elems[0].nr_components = 2; + mc->vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT; + + /* Create our texcoord buffers and texcoord buffer elements */ + for (i = 1; i < 3; ++i) + { + mc->vertex_bufs[i].pitch = sizeof(struct vlVertex2f); + mc->vertex_bufs[i].max_index = 23; + mc->vertex_bufs[i].buffer_offset = 0; + mc->vertex_bufs[i].buffer = pipe->winsys->buffer_create + ( + pipe->winsys, + 1, + PIPE_BUFFER_USAGE_VERTEX, + sizeof(struct vlVertex2f) * 24 + ); + + mc->vertex_elems[i].src_offset = 0; + mc->vertex_elems[i].vertex_buffer_index = i; + mc->vertex_elems[i].nr_components = 2; + mc->vertex_elems[i].src_format = PIPE_FORMAT_R32G32_FLOAT; + } + + /* Fill buffers */ + memcpy + ( + pipe->winsys->buffer_map(pipe->winsys, mc->vertex_bufs[0].buffer, PIPE_BUFFER_USAGE_CPU_WRITE), + macroblock_verts, + sizeof(struct vlVertex2f) * 24 + ); + memcpy + ( + pipe->winsys->buffer_map(pipe->winsys, mc->vertex_bufs[1].buffer, PIPE_BUFFER_USAGE_CPU_WRITE), + macroblock_luma_texcoords, + sizeof(struct vlVertex2f) * 24 + ); + /* TODO: Accomodate 422, 444 */ + memcpy + ( + pipe->winsys->buffer_map(pipe->winsys, mc->vertex_bufs[2].buffer, PIPE_BUFFER_USAGE_CPU_WRITE), + macroblock_chroma_420_texcoords, + sizeof(struct vlVertex2f) * 24 + ); + + for (i = 0; i < 3; ++i) + pipe->winsys->buffer_unmap(pipe->winsys, mc->vertex_bufs[i].buffer); + + /* Create our constant buffer */ + mc->vs_const_buf.size = sizeof(struct vlVertexShaderConsts); + mc->vs_const_buf.buffer = pipe->winsys->buffer_create + ( + pipe->winsys, + 1, + PIPE_BUFFER_USAGE_CONSTANT, + mc->vs_const_buf.size + ); + + mc->fs_const_buf.size = sizeof(struct vlFragmentShaderConsts); + mc->fs_const_buf.buffer = pipe->winsys->buffer_create + ( + pipe->winsys, + 1, + PIPE_BUFFER_USAGE_CONSTANT, + mc->fs_const_buf.size + ); + + memcpy + ( + pipe->winsys->buffer_map(pipe->winsys, mc->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE), + &fs_consts, + sizeof(struct vlFragmentShaderConsts) + ); + + pipe->winsys->buffer_unmap(pipe->winsys, mc->fs_const_buf.buffer); + + return 0; +} + +static int vlInit +( + struct vlR16SnormMC *mc +) +{ + struct pipe_context *pipe; + struct pipe_sampler_state sampler; + struct pipe_texture template; + unsigned int filters[5]; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + + /* For MC we render to textures, which are rounded up to nearest POT */ + mc->viewport.scale[0] = vlRoundUpPOT(mc->video_width); + mc->viewport.scale[1] = vlRoundUpPOT(mc->video_height); + mc->viewport.scale[2] = 1; + mc->viewport.scale[3] = 1; + mc->viewport.translate[0] = 0; + mc->viewport.translate[1] = 0; + mc->viewport.translate[2] = 0; + mc->viewport.translate[3] = 0; + + mc->render_target.width = vlRoundUpPOT(mc->video_width); + mc->render_target.height = vlRoundUpPOT(mc->video_height); + mc->render_target.num_cbufs = 1; + /* FB for MC stage is a vlSurface, set in vlSetRenderSurface() */ + mc->render_target.zsbuf = NULL; + + filters[0] = PIPE_TEX_FILTER_NEAREST; + filters[1] = mc->video_format == vlFormatYCbCr444 ? PIPE_TEX_FILTER_NEAREST : PIPE_TEX_FILTER_LINEAR; + filters[2] = mc->video_format == vlFormatYCbCr444 ? PIPE_TEX_FILTER_NEAREST : PIPE_TEX_FILTER_LINEAR; + filters[3] = PIPE_TEX_FILTER_LINEAR; + filters[4] = PIPE_TEX_FILTER_LINEAR; + + for (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 = ;*/ + mc->samplers[i] = pipe->create_sampler_state(pipe, &sampler); + } + + memset(&template, 0, sizeof(struct pipe_texture)); + template.target = PIPE_TEXTURE_2D; + template.format = PIPE_FORMAT_R16_SNORM; + template.last_level = 0; + template.width[0] = 8; + template.height[0] = 8 * 4; + template.depth[0] = 1; + template.compressed = 0; + pf_get_block(template.format, &template.block); + + 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; + else if (mc->video_format == vlFormatYCbCr422) + template.height[0] = 8 * 2; + else if (mc->video_format == vlFormatYCbCr444) + template.height[0] = 8 * 4; + else + assert(0); + + 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 */ + + vlCreateVertexShaderIMB(mc); + vlCreateFragmentShaderIMB(mc); + vlCreateVertexShaderFramePMB(mc); + vlCreateVertexShaderFieldPMB(mc); + vlCreateFragmentShaderFramePMB(mc); + vlCreateFragmentShaderFieldPMB(mc); + vlCreateVertexShaderFrameBMB(mc); + vlCreateVertexShaderFieldBMB(mc); + vlCreateFragmentShaderFrameBMB(mc); + vlCreateFragmentShaderFieldBMB(mc); + vlCreateDataBufs(mc); + + return 0; +} + +int vlCreateR16SNormMC +( + struct pipe_context *pipe, + unsigned int video_width, + unsigned int video_height, + enum vlFormat video_format, + struct vlRender **render +) +{ + struct vlR16SnormMC *mc; + + assert(pipe); + assert(render); + + mc = calloc(1, sizeof(struct vlR16SnormMC)); + + mc->base.vlBegin = &vlBegin; + mc->base.vlRenderMacroBlocksMpeg2 = &vlRenderMacroBlocksMpeg2R16Snorm; + mc->base.vlEnd = &vlEnd; + mc->base.vlFlush = &vlFlush; + mc->base.vlDestroy = &vlDestroy; + mc->pipe = pipe; + mc->video_width = video_width; + mc->video_height = video_height; + mc->cur_buf = 0; + + vlInit(mc); + + *render = &mc->base; + + return 0; +} diff --git a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.h b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.h new file mode 100644 index 0000000000..9842926bf7 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.h @@ -0,0 +1,18 @@ +#ifndef vl_r16snorm_mc_h +#define vl_r16snorm_mc_h + +#include "vl_types.h" + +struct pipe_context; +struct vlRender; + +int vlCreateR16SNormMC +( + struct pipe_context *pipe, + unsigned int video_width, + unsigned int video_height, + enum vlFormat video_format, + struct vlRender **render +); + +#endif diff --git a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c new file mode 100644 index 0000000000..e7a070ef4d --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c @@ -0,0 +1,2321 @@ +#define VL_INTERNAL +#include "vl_r16snorm_mc_buf.h" +#include <assert.h> +#include <stdlib.h> +#include <pipe/p_context.h> +#include <pipe/p_winsys.h> +#include <pipe/p_screen.h> +#include <pipe/p_state.h> +#include <pipe/p_inlines.h> +#include <tgsi/tgsi_parse.h> +#include <tgsi/tgsi_build.h> +#include <util/u_math.h> +#include "vl_render.h" +#include "vl_shader_build.h" +#include "vl_surface.h" +#include "vl_util.h" +#include "vl_types.h" +#include "vl_defs.h" + +/* + * TODO: Dynamically determine number of buf sets to use, based on + * video size and available mem, since we can easily run out of memory + * for high res videos. + * Note: Destroying previous frame's buffers and creating new ones + * doesn't work, since the buffer are not actually destroyed until their + * fence is signalled, and if we render fast enough we will create faster + * than we destroy. + */ +#define NUM_BUF_SETS 4 /* Number of rotating buffer sets to use */ + +enum vlMacroBlockTypeEx +{ + vlMacroBlockExTypeIntra, + vlMacroBlockExTypeFwdPredictedFrame, + vlMacroBlockExTypeFwdPredictedField, + vlMacroBlockExTypeBkwdPredictedFrame, + vlMacroBlockExTypeBkwdPredictedField, + vlMacroBlockExTypeBiPredictedFrame, + vlMacroBlockExTypeBiPredictedField, + + vlNumMacroBlockExTypes +}; + +struct vlVertexShaderConsts +{ + struct vlVertex4f denorm; +}; + +struct vlFragmentShaderConsts +{ + struct vlVertex4f multiplier; + struct vlVertex4f div; +}; + +struct vlR16SnormBufferedMC +{ + struct vlRender base; + + unsigned int picture_width, picture_height; + enum vlFormat picture_format; + + unsigned int cur_buf; + struct vlSurface *buffered_surface; + struct vlSurface *past_surface, *future_surface; + struct vlVertex2f surface_tex_inv_size; + struct vlVertex2f zero_block[3]; + unsigned int num_macroblocks; + struct vlMpeg2MacroBlock *macroblocks; + + 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[NUM_BUF_SETS][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[NUM_BUF_SETS][3]; + struct pipe_vertex_element vertex_elems[8]; + struct pipe_constant_buffer vs_const_buf, fs_const_buf; +}; + +static int vlBegin +( + struct vlRender *render +) +{ + assert(render); + + return 0; +} + +static inline int vlGrabFrameCodedBlock(short *src, short *dst, unsigned int dst_pitch) +{ + unsigned int y; + + for (y = 0; y < VL_BLOCK_HEIGHT; ++y) + memcpy + ( + dst + y * dst_pitch, + src + y * VL_BLOCK_WIDTH, + VL_BLOCK_WIDTH * 2 + ); + + return 0; +} + +static inline int vlGrabFieldCodedBlock(short *src, short *dst, unsigned int dst_pitch) +{ + unsigned int y; + + for (y = 0; y < VL_BLOCK_HEIGHT; ++y) + memcpy + ( + dst + y * dst_pitch * 2, + src + y * VL_BLOCK_WIDTH, + VL_BLOCK_WIDTH * 2 + ); + + return 0; +} + +static inline int vlGrabNoBlock(short *dst, unsigned int dst_pitch) +{ + unsigned int y; + + for (y = 0; y < VL_BLOCK_HEIGHT; ++y) + memset + ( + dst + y * dst_pitch, + 0, + VL_BLOCK_WIDTH * 2 + ); + + return 0; +} + +static inline int vlGrabBlocks +( + struct vlR16SnormBufferedMC *mc, + unsigned int mbx, + unsigned int mby, + enum vlDCTType dct_type, + unsigned int coded_block_pattern, + short *blocks +) +{ + struct pipe_surface *tex_surface; + short *texels; + unsigned int tex_pitch; + unsigned int x, y, tb = 0, sb = 0; + unsigned int mbpx = mbx * VL_MACROBLOCK_WIDTH, mbpy = mby * VL_MACROBLOCK_HEIGHT; + + assert(mc); + assert(blocks); + + tex_surface = mc->pipe->screen->get_tex_surface + ( + mc->pipe->screen, + mc->textures[mc->cur_buf % NUM_BUF_SETS][0], + 0, 0, 0, PIPE_BUFFER_USAGE_CPU_WRITE + ); + + texels = pipe_surface_map(tex_surface, PIPE_BUFFER_USAGE_CPU_WRITE); + tex_pitch = tex_surface->stride / tex_surface->block.size; + + texels += mbpy * tex_pitch + mbpx; + + for (y = 0; y < 2; ++y) + { + for (x = 0; x < 2; ++x, ++tb) + { + if ((coded_block_pattern >> (5 - tb)) & 1) + { + short *cur_block = blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT; + + if (dct_type == vlDCTTypeFrameCoded) + { + vlGrabFrameCodedBlock + ( + cur_block, + texels + y * tex_pitch * VL_BLOCK_HEIGHT + x * VL_BLOCK_WIDTH, + tex_pitch + ); + } + else + { + vlGrabFieldCodedBlock + ( + cur_block, + texels + y * tex_pitch + x * VL_BLOCK_WIDTH, + tex_pitch + ); + } + + ++sb; + } + else if (mc->zero_block[0].x < 0.0f) + { + vlGrabNoBlock(texels + y * tex_pitch * VL_BLOCK_HEIGHT + x * VL_BLOCK_WIDTH, tex_pitch); + + mc->zero_block[0].x = (mbpx + x * 8) * mc->surface_tex_inv_size.x; + mc->zero_block[0].y = (mbpy + y * 8) * mc->surface_tex_inv_size.y; + } + } + } + + pipe_surface_unmap(tex_surface); + + /* TODO: Implement 422, 444 */ + mbpx >>= 1; + mbpy >>= 1; + + for (tb = 0; tb < 2; ++tb) + { + tex_surface = mc->pipe->screen->get_tex_surface + ( + mc->pipe->screen, + mc->textures[mc->cur_buf % NUM_BUF_SETS][tb + 1], + 0, 0, 0, PIPE_BUFFER_USAGE_CPU_WRITE + ); + + texels = pipe_surface_map(tex_surface, PIPE_BUFFER_USAGE_CPU_WRITE); + tex_pitch = tex_surface->stride / tex_surface->block.size; + + texels += mbpy * tex_pitch + mbpx; + + if ((coded_block_pattern >> (1 - tb)) & 1) + { + short *cur_block = blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT; + + vlGrabFrameCodedBlock + ( + cur_block, + texels, + tex_pitch + ); + + ++sb; + } + else if (mc->zero_block[tb + 1].x < 0.0f) + { + vlGrabNoBlock(texels, tex_pitch); + + mc->zero_block[tb + 1].x = (mbpx << 1) * mc->surface_tex_inv_size.x; + mc->zero_block[tb + 1].y = (mbpy << 1) * mc->surface_tex_inv_size.y; + } + + pipe_surface_unmap(tex_surface); + } + + return 0; +} + +static inline enum vlMacroBlockTypeEx vlGetMacroBlockTypeEx(struct vlMpeg2MacroBlock *mb) +{ + assert(mb); + + switch (mb->mb_type) + { + case vlMacroBlockTypeIntra: + return vlMacroBlockExTypeIntra; + case vlMacroBlockTypeFwdPredicted: + return mb->mo_type == vlMotionTypeFrame ? + vlMacroBlockExTypeFwdPredictedFrame : vlMacroBlockExTypeFwdPredictedField; + case vlMacroBlockTypeBkwdPredicted: + return mb->mo_type == vlMotionTypeFrame ? + vlMacroBlockExTypeBkwdPredictedFrame : vlMacroBlockExTypeBkwdPredictedField; + case vlMacroBlockTypeBiPredicted: + return mb->mo_type == vlMotionTypeFrame ? + vlMacroBlockExTypeBiPredictedFrame : vlMacroBlockExTypeBiPredictedField; + default: + assert(0); + } + + /* Unreachable */ + return -1; +} + +static inline int vlGrabMacroBlock +( + struct vlR16SnormBufferedMC *mc, + struct vlMpeg2MacroBlock *macroblock +) +{ + assert(mc); + assert(macroblock); + + mc->macroblocks[mc->num_macroblocks].mbx = macroblock->mbx; + mc->macroblocks[mc->num_macroblocks].mby = macroblock->mby; + mc->macroblocks[mc->num_macroblocks].mb_type = macroblock->mb_type; + mc->macroblocks[mc->num_macroblocks].mo_type = macroblock->mo_type; + mc->macroblocks[mc->num_macroblocks].dct_type = macroblock->dct_type; + mc->macroblocks[mc->num_macroblocks].PMV[0][0][0] = macroblock->PMV[0][0][0]; + mc->macroblocks[mc->num_macroblocks].PMV[0][0][1] = macroblock->PMV[0][0][1]; + mc->macroblocks[mc->num_macroblocks].PMV[0][1][0] = macroblock->PMV[0][1][0]; + mc->macroblocks[mc->num_macroblocks].PMV[0][1][1] = macroblock->PMV[0][1][1]; + mc->macroblocks[mc->num_macroblocks].PMV[1][0][0] = macroblock->PMV[1][0][0]; + mc->macroblocks[mc->num_macroblocks].PMV[1][0][1] = macroblock->PMV[1][0][1]; + mc->macroblocks[mc->num_macroblocks].PMV[1][1][0] = macroblock->PMV[1][1][0]; + mc->macroblocks[mc->num_macroblocks].PMV[1][1][1] = macroblock->PMV[1][1][1]; + mc->macroblocks[mc->num_macroblocks].cbp = macroblock->cbp; + mc->macroblocks[mc->num_macroblocks].blocks = macroblock->blocks; + + vlGrabBlocks + ( + mc, + macroblock->mbx, + macroblock->mby, + macroblock->dct_type, + macroblock->cbp, + macroblock->blocks + ); + + mc->num_macroblocks++; + + return 0; +} + +#define SET_BLOCK(vb, cbp, mbx, mby, unitx, unity, ofsx, ofsy, hx, hy, lm, cbm, crm, zb) \ + (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 ((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 ((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 ((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); \ + } + +static inline int vlGrabMacroBlockVB +( + struct vlR16SnormBufferedMC *mc, + struct vlMpeg2MacroBlock *macroblock, + unsigned int pos +) +{ + struct vlVertex2f mo_vec[2]; + unsigned int i; + + assert(mc); + assert(macroblock); + + switch (macroblock->mb_type) + { + case vlMacroBlockTypeBiPredicted: + { + struct vlVertex2f *vb; + + vb = (struct vlVertex2f*)mc->pipe->winsys->buffer_map + ( + mc->pipe->winsys, + mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS][2].buffer, + PIPE_BUFFER_USAGE_CPU_WRITE + ) + pos * 2 * 24; + + mo_vec[0].x = macroblock->PMV[0][1][0] * 0.5f * mc->surface_tex_inv_size.x; + mo_vec[0].y = macroblock->PMV[0][1][1] * 0.5f * mc->surface_tex_inv_size.y; + + if (macroblock->mo_type == vlMotionTypeFrame) + { + for (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 = macroblock->PMV[1][1][0] * 0.5f * mc->surface_tex_inv_size.x; + mo_vec[1].y = macroblock->PMV[1][1][1] * 0.5f * mc->surface_tex_inv_size.y; + + for (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; + } + } + + mc->pipe->winsys->buffer_unmap(mc->pipe->winsys, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS][2].buffer); + + /* fall-through */ + } + case vlMacroBlockTypeFwdPredicted: + case vlMacroBlockTypeBkwdPredicted: + { + struct vlVertex2f *vb; + + vb = (struct vlVertex2f*)mc->pipe->winsys->buffer_map + ( + mc->pipe->winsys, + mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS][1].buffer, + PIPE_BUFFER_USAGE_CPU_WRITE + ) + pos * 2 * 24; + + if (macroblock->mb_type == vlMacroBlockTypeBkwdPredicted) + { + mo_vec[0].x = macroblock->PMV[0][1][0] * 0.5f * mc->surface_tex_inv_size.x; + mo_vec[0].y = macroblock->PMV[0][1][1] * 0.5f * mc->surface_tex_inv_size.y; + + if (macroblock->mo_type == vlMotionTypeField) + { + mo_vec[1].x = macroblock->PMV[1][1][0] * 0.5f * mc->surface_tex_inv_size.x; + mo_vec[1].y = macroblock->PMV[1][1][1] * 0.5f * mc->surface_tex_inv_size.y; + } + } + else + { + mo_vec[0].x = macroblock->PMV[0][0][0] * 0.5f * mc->surface_tex_inv_size.x; + mo_vec[0].y = macroblock->PMV[0][0][1] * 0.5f * mc->surface_tex_inv_size.y; + + if (macroblock->mo_type == vlMotionTypeField) + { + mo_vec[1].x = macroblock->PMV[1][0][0] * 0.5f * mc->surface_tex_inv_size.x; + mo_vec[1].y = macroblock->PMV[1][0][1] * 0.5f * mc->surface_tex_inv_size.y; + } + } + + if (macroblock->mo_type == vlMotionTypeFrame) + { + for (i = 0; i < 24 * 2; i += 2) + { + vb[i].x = mo_vec[0].x; + vb[i].y = mo_vec[0].y; + } + } + else + { + for (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; + } + } + + mc->pipe->winsys->buffer_unmap(mc->pipe->winsys, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS][1].buffer); + + /* fall-through */ + } + case vlMacroBlockTypeIntra: + { + const struct vlVertex2f unit = + { + mc->surface_tex_inv_size.x * VL_MACROBLOCK_WIDTH, + mc->surface_tex_inv_size.y * VL_MACROBLOCK_HEIGHT + }; + const struct vlVertex2f half = + { + mc->surface_tex_inv_size.x * (VL_MACROBLOCK_WIDTH / 2), + mc->surface_tex_inv_size.y * (VL_MACROBLOCK_HEIGHT / 2) + }; + + struct vlMacroBlockVertexStream0 + { + struct vlVertex2f pos; + struct vlVertex2f luma_tc; + struct vlVertex2f cb_tc; + struct vlVertex2f cr_tc; + } *vb; + + vb = (struct vlMacroBlockVertexStream0*)mc->pipe->winsys->buffer_map + ( + mc->pipe->winsys, + mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS][0].buffer, + PIPE_BUFFER_USAGE_CPU_WRITE + ) + pos * 24; + + SET_BLOCK + ( + vb, + macroblock->cbp, macroblock->mbx, macroblock->mby, + unit.x, unit.y, 0, 0, half.x, half.y, + 32, 2, 1, mc->zero_block + ); + + SET_BLOCK + ( + vb + 6, + macroblock->cbp, macroblock->mbx, macroblock->mby, + unit.x, unit.y, half.x, 0, half.x, half.y, + 16, 2, 1, mc->zero_block + ); + + SET_BLOCK + ( + vb + 12, + macroblock->cbp, macroblock->mbx, macroblock->mby, + unit.x, unit.y, 0, half.y, half.x, half.y, + 8, 2, 1, mc->zero_block + ); + + SET_BLOCK + ( + vb + 18, + macroblock->cbp, macroblock->mbx, macroblock->mby, + unit.x, unit.y, half.x, half.y, half.x, half.y, + 4, 2, 1, mc->zero_block + ); + + mc->pipe->winsys->buffer_unmap(mc->pipe->winsys, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS][0].buffer); + + break; + } + default: + assert(0); + } + + return 0; +} + +static int vlFlush +( + struct vlRender *render +) +{ + struct vlR16SnormBufferedMC *mc; + struct pipe_context *pipe; + struct vlVertexShaderConsts *vs_consts; + unsigned int num_macroblocks[vlNumMacroBlockExTypes] = {0}; + unsigned int offset[vlNumMacroBlockExTypes]; + unsigned int vb_start = 0; + unsigned int mbw; + unsigned int mbh; + unsigned int num_mb_per_frame; + unsigned int i; + + assert(render); + + mc = (struct vlR16SnormBufferedMC*)render; + + if (!mc->buffered_surface) + return 0; + + mbw = align(mc->picture_width, VL_MACROBLOCK_WIDTH) / VL_MACROBLOCK_WIDTH; + mbh = align(mc->picture_height, VL_MACROBLOCK_HEIGHT) / VL_MACROBLOCK_HEIGHT; + num_mb_per_frame = mbw * mbh; + + if (mc->num_macroblocks < num_mb_per_frame) + return 0; + + pipe = mc->pipe; + + for (i = 0; i < mc->num_macroblocks; ++i) + { + enum vlMacroBlockTypeEx mb_type_ex = vlGetMacroBlockTypeEx(&mc->macroblocks[i]); + + num_macroblocks[mb_type_ex]++; + } + + offset[0] = 0; + + for (i = 1; i < vlNumMacroBlockExTypes; ++i) + offset[i] = offset[i - 1] + num_macroblocks[i - 1]; + + for (i = 0; i < mc->num_macroblocks; ++i) + { + enum vlMacroBlockTypeEx mb_type_ex = vlGetMacroBlockTypeEx(&mc->macroblocks[i]); + + vlGrabMacroBlockVB(mc, &mc->macroblocks[i], offset[mb_type_ex]); + + offset[mb_type_ex]++; + } + + mc->render_target.cbufs[0] = pipe->screen->get_tex_surface + ( + pipe->screen, + mc->buffered_surface->texture, + 0, 0, 0, PIPE_BUFFER_USAGE_GPU_READ | PIPE_BUFFER_USAGE_GPU_WRITE + ); + + pipe->set_framebuffer_state(pipe, &mc->render_target); + pipe->set_viewport_state(pipe, &mc->viewport); + vs_consts = pipe->winsys->buffer_map + ( + pipe->winsys, + mc->vs_const_buf.buffer, + PIPE_BUFFER_USAGE_CPU_WRITE + ); + + vs_consts->denorm.x = mc->buffered_surface->texture->width[0]; + vs_consts->denorm.y = mc->buffered_surface->texture->height[0]; + + pipe->winsys->buffer_unmap(pipe->winsys, mc->vs_const_buf.buffer); + pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 0, &mc->vs_const_buf); + pipe->set_constant_buffer(pipe, PIPE_SHADER_FRAGMENT, 0, &mc->fs_const_buf); + + if (num_macroblocks[vlMacroBlockExTypeIntra] > 0) + { + pipe->set_vertex_buffers(pipe, 1, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS]); + pipe->set_vertex_elements(pipe, 4, mc->vertex_elems); + pipe->set_sampler_textures(pipe, 3, mc->textures[mc->cur_buf % NUM_BUF_SETS]); + 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, vb_start, num_macroblocks[vlMacroBlockExTypeIntra] * 24); + vb_start += num_macroblocks[vlMacroBlockExTypeIntra] * 24; + } + + if (num_macroblocks[vlMacroBlockExTypeFwdPredictedFrame] > 0) + { + pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS]); + pipe->set_vertex_elements(pipe, 6, mc->vertex_elems); + mc->textures[mc->cur_buf % NUM_BUF_SETS][3] = mc->past_surface->texture; + pipe->set_sampler_textures(pipe, 4, mc->textures[mc->cur_buf % NUM_BUF_SETS]); + pipe->bind_sampler_states(pipe, 4, (void**)mc->samplers); + pipe->bind_vs_state(pipe, mc->p_vs[0]); + pipe->bind_fs_state(pipe, mc->p_fs[0]); + + pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, vb_start, num_macroblocks[vlMacroBlockExTypeFwdPredictedFrame] * 24); + vb_start += num_macroblocks[vlMacroBlockExTypeFwdPredictedFrame] * 24; + } + + if (num_macroblocks[vlMacroBlockExTypeFwdPredictedField] > 0) + { + pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS]); + pipe->set_vertex_elements(pipe, 6, mc->vertex_elems); + mc->textures[mc->cur_buf % NUM_BUF_SETS][3] = mc->past_surface->texture; + pipe->set_sampler_textures(pipe, 4, mc->textures[mc->cur_buf % NUM_BUF_SETS]); + pipe->bind_sampler_states(pipe, 4, (void**)mc->samplers); + pipe->bind_vs_state(pipe, mc->p_vs[1]); + pipe->bind_fs_state(pipe, mc->p_fs[1]); + + pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, vb_start, num_macroblocks[vlMacroBlockExTypeFwdPredictedField] * 24); + vb_start += num_macroblocks[vlMacroBlockExTypeFwdPredictedField] * 24; + } + + if (num_macroblocks[vlMacroBlockExTypeBkwdPredictedFrame] > 0) + { + pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS]); + pipe->set_vertex_elements(pipe, 6, mc->vertex_elems); + mc->textures[mc->cur_buf % NUM_BUF_SETS][3] = mc->future_surface->texture; + pipe->set_sampler_textures(pipe, 4, mc->textures[mc->cur_buf % NUM_BUF_SETS]); + pipe->bind_sampler_states(pipe, 4, (void**)mc->samplers); + pipe->bind_vs_state(pipe, mc->p_vs[0]); + pipe->bind_fs_state(pipe, mc->p_fs[0]); + + pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, vb_start, num_macroblocks[vlMacroBlockExTypeBkwdPredictedFrame] * 24); + vb_start += num_macroblocks[vlMacroBlockExTypeBkwdPredictedFrame] * 24; + } + + if (num_macroblocks[vlMacroBlockExTypeBkwdPredictedField] > 0) + { + pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS]); + pipe->set_vertex_elements(pipe, 6, mc->vertex_elems); + mc->textures[mc->cur_buf % NUM_BUF_SETS][3] = mc->future_surface->texture; + pipe->set_sampler_textures(pipe, 4, mc->textures[mc->cur_buf % NUM_BUF_SETS]); + pipe->bind_sampler_states(pipe, 4, (void**)mc->samplers); + pipe->bind_vs_state(pipe, mc->p_vs[1]); + pipe->bind_fs_state(pipe, mc->p_fs[1]); + + pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, vb_start, num_macroblocks[vlMacroBlockExTypeBkwdPredictedField] * 24); + vb_start += num_macroblocks[vlMacroBlockExTypeBkwdPredictedField] * 24; + } + + if (num_macroblocks[vlMacroBlockExTypeBiPredictedFrame] > 0) + { + pipe->set_vertex_buffers(pipe, 3, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS]); + pipe->set_vertex_elements(pipe, 8, mc->vertex_elems); + mc->textures[mc->cur_buf % NUM_BUF_SETS][3] = mc->past_surface->texture; + mc->textures[mc->cur_buf % NUM_BUF_SETS][4] = mc->future_surface->texture; + pipe->set_sampler_textures(pipe, 5, mc->textures[mc->cur_buf % NUM_BUF_SETS]); + pipe->bind_sampler_states(pipe, 5, (void**)mc->samplers); + pipe->bind_vs_state(pipe, mc->b_vs[0]); + pipe->bind_fs_state(pipe, mc->b_fs[0]); + + pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, vb_start, num_macroblocks[vlMacroBlockExTypeBiPredictedFrame] * 24); + vb_start += num_macroblocks[vlMacroBlockExTypeBiPredictedFrame] * 24; + } + + if (num_macroblocks[vlMacroBlockExTypeBiPredictedField] > 0) + { + pipe->set_vertex_buffers(pipe, 3, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS]); + pipe->set_vertex_elements(pipe, 8, mc->vertex_elems); + mc->textures[mc->cur_buf % NUM_BUF_SETS][3] = mc->past_surface->texture; + mc->textures[mc->cur_buf % NUM_BUF_SETS][4] = mc->future_surface->texture; + pipe->set_sampler_textures(pipe, 5, mc->textures[mc->cur_buf % NUM_BUF_SETS]); + pipe->bind_sampler_states(pipe, 5, (void**)mc->samplers); + pipe->bind_vs_state(pipe, mc->b_vs[1]); + pipe->bind_fs_state(pipe, mc->b_fs[1]); + + pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, vb_start, num_macroblocks[vlMacroBlockExTypeBiPredictedField] * 24); + vb_start += num_macroblocks[vlMacroBlockExTypeBiPredictedField] * 24; + } + + pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, &mc->buffered_surface->render_fence); + + for (i = 0; i < 3; ++i) + mc->zero_block[i].x = -1.0f; + + mc->buffered_surface = NULL; + mc->num_macroblocks = 0; + mc->cur_buf++; + + return 0; +} + +static int vlRenderMacroBlocksMpeg2R16SnormBuffered +( + struct vlRender *render, + struct vlMpeg2MacroBlockBatch *batch, + struct vlSurface *surface +) +{ + struct vlR16SnormBufferedMC *mc; + unsigned int i; + + assert(render); + + mc = (struct vlR16SnormBufferedMC*)render; + + if (mc->buffered_surface) + { + if (mc->buffered_surface != surface) + { + vlFlush(&mc->base); + mc->buffered_surface = surface; + mc->past_surface = batch->past_surface; + mc->future_surface = batch->future_surface; + mc->surface_tex_inv_size.x = 1.0f / surface->texture->width[0]; + mc->surface_tex_inv_size.y = 1.0f / surface->texture->height[0]; + } + } + else + { + mc->buffered_surface = surface; + mc->past_surface = batch->past_surface; + mc->future_surface = batch->future_surface; + mc->surface_tex_inv_size.x = 1.0f / surface->texture->width[0]; + mc->surface_tex_inv_size.y = 1.0f / surface->texture->height[0]; + } + + for (i = 0; i < batch->num_macroblocks; ++i) + vlGrabMacroBlock(mc, &batch->macroblocks[i]); + + return 0; +} + +static int vlEnd +( + struct vlRender *render +) +{ + assert(render); + + return 0; +} + +static int vlDestroy +( + struct vlRender *render +) +{ + struct vlR16SnormBufferedMC *mc; + struct pipe_context *pipe; + unsigned int h, i; + + assert(render); + + mc = (struct vlR16SnormBufferedMC*)render; + pipe = mc->pipe; + + for (i = 0; i < 5; ++i) + pipe->delete_sampler_state(pipe, mc->samplers[i]); + + for (h = 0; h < NUM_BUF_SETS; ++h) + for (i = 0; i < 3; ++i) + pipe->winsys->buffer_destroy(pipe->winsys, mc->vertex_bufs[h][i].buffer); + + /* Textures 3 & 4 are not created directly, no need to release them here */ + for (i = 0; i < NUM_BUF_SETS; ++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); + + for (i = 0; i < 2; ++i) + { + pipe->delete_vs_state(pipe, mc->p_vs[i]); + pipe->delete_fs_state(pipe, mc->p_fs[i]); + pipe->delete_vs_state(pipe, mc->b_vs[i]); + pipe->delete_fs_state(pipe, mc->b_fs[i]); + } + + pipe->winsys->buffer_destroy(pipe->winsys, mc->vs_const_buf.buffer); + pipe->winsys->buffer_destroy(pipe->winsys, mc->fs_const_buf.buffer); + + free(mc->macroblocks); + free(mc); + + return 0; +} + +/* + * 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 vlFragmentShaderConsts fs_consts = +{ + {32767.0f / 255.0f, 32767.0f / 255.0f, 32767.0f / 255.0f, 0.0f}, + {0.5f, 2.0f, 0.0f, 0.0f} +}; + +static int vlCreateVertexShaderIMB +( + struct vlR16SnormBufferedMC *mc +) +{ + const unsigned int max_tokens = 50; + + struct pipe_context *pipe; + struct pipe_shader_state vs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 (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 (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 (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); + + vs.tokens = tokens; + mc->i_vs = pipe->create_vs_state(pipe, &vs); + free(tokens); + + return 0; +} + +static int vlCreateFragmentShaderIMB +( + struct vlR16SnormBufferedMC *mc +) +{ + const unsigned int max_tokens = 100; + + struct pipe_context *pipe; + struct pipe_shader_state fs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 (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 (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 (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); + + fs.tokens = tokens; + mc->i_fs = pipe->create_fs_state(pipe, &fs); + free(tokens); + + return 0; +} + +static int vlCreateVertexShaderFramePMB +( + struct vlR16SnormBufferedMC *mc +) +{ + const unsigned int max_tokens = 100; + + struct pipe_context *pipe; + struct pipe_shader_state vs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 (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 (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 (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); + + vs.tokens = tokens; + mc->p_vs[0] = pipe->create_vs_state(pipe, &vs); + free(tokens); + + return 0; +} + +static int vlCreateVertexShaderFieldPMB +( + struct vlR16SnormBufferedMC *mc +) +{ + const unsigned int max_tokens = 100; + + struct pipe_context *pipe; + struct pipe_shader_state vs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 macroblock top field texcoords + * decl i5 ; Ref macroblock bottom field texcoords + */ + for (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 c0 ; Render target dimensions */ + decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 0); + 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 top field texcoords + * decl o5 ; Ref macroblock bottom field texcoords + * decl o6 ; Denormalized vertex pos + */ + for (i = 0; i < 7; 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 (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 top field macroblock texcoords + * add o5, i0, i5 ; Translate vertex pos by motion vec to form bottom field macroblock texcoords + */ + for (i = 0; i < 2; ++i) + { + inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, i + 4, TGSI_FILE_INPUT, 0, TGSI_FILE_INPUT, i + 4); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + } + + /* mul o6, i0, c0 ; Denorm vertex pos */ + inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_OUTPUT, 6, TGSI_FILE_INPUT, 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); + + vs.tokens = tokens; + mc->p_vs[1] = pipe->create_vs_state(pipe, &vs); + free(tokens); + + return 0; +} + +static int vlCreateFragmentShaderFramePMB +( + struct vlR16SnormBufferedMC *mc +) +{ + const unsigned int max_tokens = 100; + + struct pipe_context *pipe; + struct pipe_shader_state fs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 (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 (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 (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); + + fs.tokens = tokens; + mc->p_fs[0] = pipe->create_fs_state(pipe, &fs); + free(tokens); + + return 0; +} + +static int vlCreateFragmentShaderFieldPMB +( + struct vlR16SnormBufferedMC *mc +) +{ + const unsigned int max_tokens = 200; + + struct pipe_context *pipe; + struct pipe_shader_state fs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 top field texcoords + * decl i4 ; Ref macroblock bottom field texcoords + * decl i5 ; Denormalized vertex pos + */ + for (i = 0; i < 6; ++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 ; Constants 1/2 & 2 in .x, .y channels for Y-mod-2 top/bottom field selection + */ + 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-t4 */ + decl = vl_decl_temps(0, 4); + 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 (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 (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 top field + * tex2d t2, i4, s3 ; Read texel from ref macroblock bottom field + */ + for (i = 0; i < 2; ++i) + { + inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 1, TGSI_FILE_INPUT, i + 3, TGSI_FILE_SAMPLER, 3); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + } + + /* XXX: Pos values off by 0.5? */ + /* sub t4, i5.y, c1.x ; Sub 0.5 from denormalized pos */ + inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_INPUT, 5, TGSI_FILE_CONSTANT, 1); + inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_X; + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* mul t3, t4, c1.x ; Multiply pos Y-coord by 1/2 */ + inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_CONSTANT, 1); + inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_X; + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* floor t3, t3 ; Get rid of fractional part */ + inst = vl_inst2(TGSI_OPCODE_FLOOR, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* mul t3, t3, c1.y ; Multiply by 2 */ + inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_CONSTANT, 1); + inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y; + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* sub t3, t4, t3 ; Subtract from original Y to get Y % 2 */ + inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_TEMPORARY, 3); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* TODO: Move to conditional tex fetch on t3 instead of lerp */ + /* lerp t1, t3, t1, t2 ; Choose between top and bottom fields based on Y % 2 */ + inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2); + 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); + + fs.tokens = tokens; + mc->p_fs[1] = pipe->create_fs_state(pipe, &fs); + free(tokens); + + return 0; +} + +static int vlCreateVertexShaderFrameBMB +( + struct vlR16SnormBufferedMC *mc +) +{ + const unsigned int max_tokens = 100; + + struct pipe_context *pipe; + struct pipe_shader_state vs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 (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 (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 (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 (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); + + vs.tokens = tokens; + mc->b_vs[0] = pipe->create_vs_state(pipe, &vs); + free(tokens); + + return 0; +} + +static int vlCreateVertexShaderFieldBMB +( + struct vlR16SnormBufferedMC *mc +) +{ + const unsigned int max_tokens = 100; + + struct pipe_context *pipe; + struct pipe_shader_state vs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 + * decl i6 ; Second ref macroblock top field texcoords + * decl i7 ; Second ref macroblock bottom field texcoords + */ + for (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 c0 ; Render target dimensions */ + decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 0); + 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 top field texcoords + * decl o5 ; First ref macroblock Bottom field texcoords + * decl o6 ; Second ref macroblock top field texcoords + * decl o7 ; Second ref macroblock Bottom field texcoords + * decl o8 ; Denormalized vertex pos + */ + for (i = 0; i < 9; 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); + + /* + * 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 (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 top field macroblock texcoords + * add o5, i0, i5 ; Translate vertex pos by motion vec to form first bottom field macroblock texcoords + * add o6, i0, i6 ; Translate vertex pos by motion vec to form second top field macroblock texcoords + * add o7, i0, i7 ; Translate vertex pos by motion vec to form second bottom field macroblock texcoords + */ + for (i = 0; i < 4; ++i) + { + inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, i + 4, TGSI_FILE_INPUT, 0, TGSI_FILE_INPUT, i + 4); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + } + + /* mul o8, i0, c0 ; Denorm vertex pos */ + inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_OUTPUT, 8, TGSI_FILE_INPUT, 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); + + vs.tokens = tokens; + mc->b_vs[1] = pipe->create_vs_state(pipe, &vs); + free(tokens); + + return 0; +} + +static int vlCreateFragmentShaderFrameBMB +( + struct vlR16SnormBufferedMC *mc +) +{ + const unsigned int max_tokens = 100; + + struct pipe_context *pipe; + struct pipe_shader_state fs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 (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 (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 (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 (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_LERP, 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); + + fs.tokens = tokens; + mc->b_fs[0] = pipe->create_fs_state(pipe, &fs); + free(tokens); + + return 0; +} + +static int vlCreateFragmentShaderFieldBMB +( + struct vlR16SnormBufferedMC *mc +) +{ + const unsigned int max_tokens = 200; + + struct pipe_context *pipe; + struct pipe_shader_state fs; + struct tgsi_token *tokens; + struct tgsi_header *header; + + struct tgsi_full_declaration decl; + struct tgsi_full_instruction inst; + + unsigned int ti; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token)); + + /* Version */ + *(struct tgsi_version*)&tokens[0] = tgsi_build_version(); + /* Header */ + header = (struct tgsi_header*)&tokens[1]; + *header = tgsi_build_header(); + /* Processor */ + *(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 top field texcoords + * decl i4 ; First ref macroblock bottom field texcoords + * decl i5 ; Second ref macroblock top field texcoords + * decl i6 ; Second ref macroblock bottom field texcoords + * decl i7 ; Denormalized vertex pos + */ + for (i = 0; i < 8; ++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 ; Constants 1/2 & 2 in .x, .y channels to use as weight to blend past and future texels + * ; and for Y-mod-2 top/bottom field selection + */ + 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-t5 */ + decl = vl_decl_temps(0, 5); + 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 (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 (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); + + /* XXX: Pos values off by 0.5? */ + /* sub t4, i7.y, c1.x ; Sub 0.5 from denormalized pos */ + inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_INPUT, 7, TGSI_FILE_CONSTANT, 1); + inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_X; + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* mul t3, t4, c1.x ; Multiply pos Y-coord by 1/2 */ + inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_CONSTANT, 1); + inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X; + inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_X; + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* floor t3, t3 ; Get rid of fractional part */ + inst = vl_inst2(TGSI_OPCODE_FLOOR, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* mul t3, t3, c1.y ; Multiply by 2 */ + inst = vl_inst3( TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_CONSTANT, 1); + inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y; + inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y; + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* sub t3, t4, t3 ; Subtract from original Y to get Y % 2 */ + inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_TEMPORARY, 3); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* + * tex2d t1, i3, s3 ; Read texel from past ref macroblock top field + * tex2d t2, i4, s3 ; Read texel from past ref macroblock bottom field + */ + for (i = 0; i < 2; ++i) + { + inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 1, TGSI_FILE_INPUT, i + 3, TGSI_FILE_SAMPLER, 3); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + } + + /* TODO: Move to conditional tex fetch on t3 instead of lerp */ + /* lerp t1, t3, t1, t2 ; Choose between top and bottom fields based on Y % 2 */ + inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + + /* + * tex2d t4, i5, s4 ; Read texel from future ref macroblock top field + * tex2d t5, i6, s4 ; Read texel from future ref macroblock bottom field + */ + for (i = 0; i < 2; ++i) + { + inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 4, TGSI_FILE_INPUT, i + 5, TGSI_FILE_SAMPLER, 4); + ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti); + } + + /* TODO: Move to conditional tex fetch on t3 instead of lerp */ + /* lerp t2, t3, t4, t5 ; Choose between top and bottom fields based on Y % 2 */ + inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 2, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_TEMPORARY, 5); + 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_LERP, 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); + + fs.tokens = tokens; + mc->b_fs[1] = pipe->create_fs_state(pipe, &fs); + free(tokens); + + return 0; +} + +static int vlCreateDataBufs +( + struct vlR16SnormBufferedMC *mc +) +{ + const unsigned int mbw = align(mc->picture_width, VL_MACROBLOCK_WIDTH) / VL_MACROBLOCK_WIDTH; + const unsigned int mbh = align(mc->picture_height, VL_MACROBLOCK_HEIGHT) / VL_MACROBLOCK_HEIGHT; + const unsigned int num_mb_per_frame = mbw * mbh; + + struct pipe_context *pipe; + unsigned int h, i; + + assert(mc); + + pipe = mc->pipe; + + /* Create our vertex buffers */ + for (h = 0; h < NUM_BUF_SETS; ++h) + { + mc->vertex_bufs[h][0].pitch = sizeof(struct vlVertex2f) * 4; + mc->vertex_bufs[h][0].max_index = 24 * num_mb_per_frame - 1; + mc->vertex_bufs[h][0].buffer_offset = 0; + mc->vertex_bufs[h][0].buffer = pipe->winsys->buffer_create + ( + pipe->winsys, + 1, + PIPE_BUFFER_USAGE_VERTEX, + sizeof(struct vlVertex2f) * 4 * 24 * num_mb_per_frame + ); + + for (i = 1; i < 3; ++i) + { + mc->vertex_bufs[h][i].pitch = sizeof(struct vlVertex2f) * 2; + mc->vertex_bufs[h][i].max_index = 24 * num_mb_per_frame - 1; + mc->vertex_bufs[h][i].buffer_offset = 0; + mc->vertex_bufs[h][i].buffer = pipe->winsys->buffer_create + ( + pipe->winsys, + 1, + PIPE_BUFFER_USAGE_VERTEX, + sizeof(struct vlVertex2f) * 2 * 24 * num_mb_per_frame + ); + } + } + + /* Position element */ + mc->vertex_elems[0].src_offset = 0; + mc->vertex_elems[0].vertex_buffer_index = 0; + mc->vertex_elems[0].nr_components = 2; + mc->vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT; + + /* Luma, texcoord element */ + mc->vertex_elems[1].src_offset = sizeof(struct vlVertex2f); + mc->vertex_elems[1].vertex_buffer_index = 0; + mc->vertex_elems[1].nr_components = 2; + mc->vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT; + + /* Chroma Cr texcoord element */ + mc->vertex_elems[2].src_offset = sizeof(struct vlVertex2f) * 2; + mc->vertex_elems[2].vertex_buffer_index = 0; + mc->vertex_elems[2].nr_components = 2; + mc->vertex_elems[2].src_format = PIPE_FORMAT_R32G32_FLOAT; + + /* Chroma Cb texcoord element */ + mc->vertex_elems[3].src_offset = sizeof(struct vlVertex2f) * 3; + mc->vertex_elems[3].vertex_buffer_index = 0; + mc->vertex_elems[3].nr_components = 2; + mc->vertex_elems[3].src_format = PIPE_FORMAT_R32G32_FLOAT; + + /* First ref surface top field texcoord element */ + mc->vertex_elems[4].src_offset = 0; + mc->vertex_elems[4].vertex_buffer_index = 1; + mc->vertex_elems[4].nr_components = 2; + mc->vertex_elems[4].src_format = PIPE_FORMAT_R32G32_FLOAT; + + /* First ref surface bottom field texcoord element */ + mc->vertex_elems[5].src_offset = sizeof(struct vlVertex2f); + mc->vertex_elems[5].vertex_buffer_index = 1; + mc->vertex_elems[5].nr_components = 2; + mc->vertex_elems[5].src_format = PIPE_FORMAT_R32G32_FLOAT; + + /* Second ref surface top field texcoord element */ + mc->vertex_elems[6].src_offset = 0; + mc->vertex_elems[6].vertex_buffer_index = 2; + mc->vertex_elems[6].nr_components = 2; + mc->vertex_elems[6].src_format = PIPE_FORMAT_R32G32_FLOAT; + + /* Second ref surface bottom field texcoord element */ + mc->vertex_elems[7].src_offset = sizeof(struct vlVertex2f); + mc->vertex_elems[7].vertex_buffer_index = 2; + mc->vertex_elems[7].nr_components = 2; + mc->vertex_elems[7].src_format = PIPE_FORMAT_R32G32_FLOAT; + + /* Create our constant buffer */ + mc->vs_const_buf.size = sizeof(struct vlVertexShaderConsts); + mc->vs_const_buf.buffer = pipe->winsys->buffer_create + ( + pipe->winsys, + 1, + PIPE_BUFFER_USAGE_CONSTANT, + mc->vs_const_buf.size + ); + + mc->fs_const_buf.size = sizeof(struct vlFragmentShaderConsts); + mc->fs_const_buf.buffer = pipe->winsys->buffer_create + ( + pipe->winsys, + 1, + PIPE_BUFFER_USAGE_CONSTANT, + mc->fs_const_buf.size + ); + + memcpy + ( + pipe->winsys->buffer_map(pipe->winsys, mc->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE), + &fs_consts, + sizeof(struct vlFragmentShaderConsts) + ); + + pipe->winsys->buffer_unmap(pipe->winsys, mc->fs_const_buf.buffer); + + mc->macroblocks = malloc(sizeof(struct vlMpeg2MacroBlock) * num_mb_per_frame); + + return 0; +} + +static int vlInit +( + struct vlR16SnormBufferedMC *mc +) +{ + struct pipe_context *pipe; + struct pipe_sampler_state sampler; + struct pipe_texture template; + unsigned int filters[5]; + unsigned int i; + + assert(mc); + + pipe = mc->pipe; + + /* For MC we render to textures, which are rounded up to nearest POT */ + mc->viewport.scale[0] = vlRoundUpPOT(mc->picture_width); + mc->viewport.scale[1] = vlRoundUpPOT(mc->picture_height); + mc->viewport.scale[2] = 1; + mc->viewport.scale[3] = 1; + mc->viewport.translate[0] = 0; + mc->viewport.translate[1] = 0; + mc->viewport.translate[2] = 0; + mc->viewport.translate[3] = 0; + + mc->render_target.width = vlRoundUpPOT(mc->picture_width); + mc->render_target.height = vlRoundUpPOT(mc->picture_height); + mc->render_target.num_cbufs = 1; + /* FB for MC stage is a vlSurface created by the user, set at render time */ + mc->render_target.zsbuf = NULL; + + filters[0] = PIPE_TEX_FILTER_NEAREST; + /* FIXME: Linear causes discoloration around block edges */ + filters[1] = /*mc->picture_format == vlFormatYCbCr444 ?*/ PIPE_TEX_FILTER_NEAREST /*: PIPE_TEX_FILTER_LINEAR*/; + filters[2] = /*mc->picture_format == vlFormatYCbCr444 ?*/ PIPE_TEX_FILTER_NEAREST /*: PIPE_TEX_FILTER_LINEAR*/; + filters[3] = PIPE_TEX_FILTER_LINEAR; + filters[4] = PIPE_TEX_FILTER_LINEAR; + + for (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 = ;*/ + mc->samplers[i] = pipe->create_sampler_state(pipe, &sampler); + } + + memset(&template, 0, sizeof(struct pipe_texture)); + template.target = PIPE_TEXTURE_2D; + template.format = PIPE_FORMAT_R16_SNORM; + template.last_level = 0; + template.width[0] = vlRoundUpPOT(mc->picture_width); + template.height[0] = vlRoundUpPOT(mc->picture_height); + template.depth[0] = 1; + template.compressed = 0; + pf_get_block(template.format, &template.block); + + for (i = 0; i < NUM_BUF_SETS; ++i) + mc->textures[i][0] = pipe->screen->texture_create(pipe->screen, &template); + + if (mc->picture_format == vlFormatYCbCr420) + { + template.width[0] = vlRoundUpPOT(mc->picture_width / 2); + template.height[0] = vlRoundUpPOT(mc->picture_height / 2); + } + else if (mc->picture_format == vlFormatYCbCr422) + template.height[0] = vlRoundUpPOT(mc->picture_height / 2); + + for (i = 0; i < NUM_BUF_SETS; ++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 */ + + vlCreateVertexShaderIMB(mc); + vlCreateFragmentShaderIMB(mc); + vlCreateVertexShaderFramePMB(mc); + vlCreateVertexShaderFieldPMB(mc); + vlCreateFragmentShaderFramePMB(mc); + vlCreateFragmentShaderFieldPMB(mc); + vlCreateVertexShaderFrameBMB(mc); + vlCreateVertexShaderFieldBMB(mc); + vlCreateFragmentShaderFrameBMB(mc); + vlCreateFragmentShaderFieldBMB(mc); + vlCreateDataBufs(mc); + + return 0; +} + +int vlCreateR16SNormBufferedMC +( + struct pipe_context *pipe, + unsigned int picture_width, + unsigned int picture_height, + enum vlFormat picture_format, + struct vlRender **render +) +{ + struct vlR16SnormBufferedMC *mc; + unsigned int i; + + assert(pipe); + assert(render); + + mc = calloc(1, sizeof(struct vlR16SnormBufferedMC)); + + mc->base.vlBegin = &vlBegin; + mc->base.vlRenderMacroBlocksMpeg2 = &vlRenderMacroBlocksMpeg2R16SnormBuffered; + mc->base.vlEnd = &vlEnd; + mc->base.vlFlush = &vlFlush; + mc->base.vlDestroy = &vlDestroy; + mc->pipe = pipe; + mc->picture_width = picture_width; + mc->picture_height = picture_height; + + mc->cur_buf = 0; + mc->buffered_surface = NULL; + mc->past_surface = NULL; + mc->future_surface = NULL; + for (i = 0; i < 3; ++i) + mc->zero_block[i].x = -1.0f; + mc->num_macroblocks = 0; + + vlInit(mc); + + *render = &mc->base; + + return 0; +} diff --git a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.h b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.h new file mode 100644 index 0000000000..27177d64ca --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.h @@ -0,0 +1,18 @@ +#ifndef vl_r16snorm_mc_buf_h +#define vl_r16snorm_mc_buf_h + +#include "vl_types.h" + +struct pipe_context; +struct vlRender; + +int vlCreateR16SNormBufferedMC +( + struct pipe_context *pipe, + unsigned int picture_width, + unsigned int picture_height, + enum vlFormat picture_format, + struct vlRender **render +); + +#endif diff --git a/src/gallium/state_trackers/g3dvl/vl_render.h b/src/gallium/state_trackers/g3dvl/vl_render.h new file mode 100644 index 0000000000..166030b498 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_render.h @@ -0,0 +1,38 @@ +#ifndef vl_render_h +#define vl_render_h + +#include "vl_types.h" + +struct pipe_surface; + +struct vlRender +{ + int (*vlBegin) + ( + struct vlRender *render + ); + + int (*vlRenderMacroBlocksMpeg2) + ( + struct vlRender *render, + struct vlMpeg2MacroBlockBatch *batch, + struct vlSurface *surface + ); + + int (*vlEnd) + ( + struct vlRender *render + ); + + int (*vlFlush) + ( + struct vlRender *render + ); + + int (*vlDestroy) + ( + struct vlRender *render + ); +}; + +#endif diff --git a/src/gallium/state_trackers/g3dvl/vl_screen.c b/src/gallium/state_trackers/g3dvl/vl_screen.c new file mode 100644 index 0000000000..484f63b0d4 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_screen.c @@ -0,0 +1,115 @@ +#define VL_INTERNAL +#include "vl_screen.h" +#include <assert.h> +#include <stdlib.h> + +int vlCreateScreen +( + struct vlDisplay *display, + int screen, + struct pipe_screen *pscreen, + struct vlScreen **vl_screen +) +{ + struct vlScreen *scrn; + + assert(display); + assert(pscreen); + assert(vl_screen); + + scrn = calloc(1, sizeof(struct vlScreen)); + + if (!scrn) + return 1; + + scrn->display = display; + scrn->ordinal = screen; + scrn->pscreen = pscreen; + *vl_screen = scrn; + + return 0; +} + +int vlDestroyScreen +( + struct vlScreen *screen +) +{ + assert(screen); + + free(screen); + + return 0; +} + +struct vlDisplay* vlGetDisplay +( + struct vlScreen *screen +) +{ + assert(screen); + + return screen->display; +} + +struct pipe_screen* vlGetPipeScreen +( + struct vlScreen *screen +) +{ + assert(screen); + + return screen->pscreen; +} + +unsigned int vlGetMaxProfiles +( + struct vlScreen *screen +) +{ + assert(screen); + + return vlProfileCount; +} + +int vlQueryProfiles +( + struct vlScreen *screen, + enum vlProfile *profiles +) +{ + assert(screen); + assert(profiles); + + profiles[0] = vlProfileMpeg2Simple; + profiles[1] = vlProfileMpeg2Main; + + return 0; +} + +unsigned int vlGetMaxEntryPoints +( + struct vlScreen *screen +) +{ + assert(screen); + + return vlEntryPointCount; +} + +int vlQueryEntryPoints +( + struct vlScreen *screen, + enum vlProfile profile, + enum vlEntryPoint *entry_points +) +{ + assert(screen); + assert(entry_points); + + entry_points[0] = vlEntryPointIDCT; + entry_points[1] = vlEntryPointMC; + entry_points[2] = vlEntryPointCSC; + + return 0; +} diff --git a/src/gallium/state_trackers/g3dvl/vl_screen.h b/src/gallium/state_trackers/g3dvl/vl_screen.h new file mode 100644 index 0000000000..98f3d429b6 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_screen.h @@ -0,0 +1,63 @@ +#ifndef vl_screen_h +#define vl_screen_h + +#include "vl_types.h" + +struct pipe_screen; + +#ifdef VL_INTERNAL +struct vlScreen +{ + struct vlDisplay *display; + unsigned int ordinal; + struct pipe_screen *pscreen; +}; +#endif + +int vlCreateScreen +( + struct vlDisplay *display, + int screen, + struct pipe_screen *pscreen, + struct vlScreen **vl_screen +); + +int vlDestroyScreen +( + struct vlScreen *screen +); + +struct vlDisplay* vlGetDisplay +( + struct vlScreen *screen +); + +struct pipe_screen* vlGetPipeScreen +( + struct vlScreen *screen +); + +unsigned int vlGetMaxProfiles +( + struct vlScreen *screen +); + +int vlQueryProfiles +( + struct vlScreen *screen, + enum vlProfile *profiles +); + +unsigned int vlGetMaxEntryPoints +( + struct vlScreen *screen +); + +int vlQueryEntryPoints +( + struct vlScreen *screen, + enum vlProfile profile, + enum vlEntryPoint *entry_points +); + +#endif diff --git a/src/gallium/state_trackers/g3dvl/vl_shader_build.c b/src/gallium/state_trackers/g3dvl/vl_shader_build.c new file mode 100644 index 0000000000..51f1721a33 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/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/state_trackers/g3dvl/vl_shader_build.h b/src/gallium/state_trackers/g3dvl/vl_shader_build.h new file mode 100644 index 0000000000..dc615cb156 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/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/state_trackers/g3dvl/vl_surface.c b/src/gallium/state_trackers/g3dvl/vl_surface.c new file mode 100644 index 0000000000..076bd40d41 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_surface.c @@ -0,0 +1,237 @@ +#define VL_INTERNAL +#include "vl_surface.h" +#include <assert.h> +#include <stdlib.h> +#include <string.h> +#include <pipe/p_screen.h> +#include <pipe/p_state.h> +#include <pipe/p_inlines.h> +#include <vl_winsys.h> +#include "vl_screen.h" +#include "vl_context.h" +#include "vl_render.h" +#include "vl_csc.h" +#include "vl_util.h" + +int vlCreateSurface +( + struct vlScreen *screen, + unsigned int width, + unsigned int height, + enum vlFormat format, + struct vlSurface **surface +) +{ + struct vlSurface *sfc; + struct pipe_texture template; + + assert(screen); + assert(surface); + + sfc = calloc(1, sizeof(struct vlSurface)); + + if (!sfc) + return 1; + + sfc->screen = screen; + sfc->width = width; + sfc->height = height; + sfc->format = format; + + memset(&template, 0, sizeof(struct pipe_texture)); + template.target = PIPE_TEXTURE_2D; + template.format = PIPE_FORMAT_A8R8G8B8_UNORM; + template.last_level = 0; + template.width[0] = vlRoundUpPOT(sfc->width); + template.height[0] = vlRoundUpPOT(sfc->height); + template.depth[0] = 1; + template.compressed = 0; + pf_get_block(template.format, &template.block); + template.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER | PIPE_TEXTURE_USAGE_RENDER_TARGET; + + sfc->texture = vlGetPipeScreen(screen)->texture_create(vlGetPipeScreen(screen), &template); + + *surface = sfc; + + return 0; +} + +int vlDestroySurface +( + struct vlSurface *surface +) +{ + assert(surface); + + pipe_texture_release(&surface->texture); + free(surface); + + return 0; +} + +int vlRenderMacroBlocksMpeg2 +( + struct vlMpeg2MacroBlockBatch *batch, + struct vlSurface *surface +) +{ + assert(batch); + assert(surface); + assert(surface->context); + + surface->context->render->vlBegin(surface->context->render); + + surface->context->render->vlRenderMacroBlocksMpeg2 + ( + surface->context->render, + batch, + surface + ); + + surface->context->render->vlEnd(surface->context->render); + + return 0; +} + +int vlPutPicture +( + struct vlSurface *surface, + vlNativeDrawable drawable, + int srcx, + int srcy, + int srcw, + int srch, + int destx, + int desty, + int destw, + int desth, + int drawable_w, + int drawable_h, + enum vlPictureType picture_type +) +{ + struct vlCSC *csc; + struct pipe_context *pipe; + + assert(surface); + assert(surface->context); + + surface->context->render->vlFlush(surface->context->render); + + csc = surface->context->csc; + pipe = surface->context->pipe; + + csc->vlResizeFrameBuffer(csc, drawable_w, drawable_h); + + csc->vlBegin(csc); + + csc->vlPutPicture + ( + csc, + surface, + srcx, + srcy, + srcw, + srch, + destx, + desty, + destw, + desth, + picture_type + ); + + csc->vlEnd(csc); + + pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, &surface->disp_fence); + + bind_pipe_drawable(pipe, drawable); + + pipe->winsys->flush_frontbuffer + ( + pipe->winsys, + csc->vlGetFrameBuffer(csc), + pipe->priv + ); + + return 0; +} + +int vlSurfaceGetStatus +( + struct vlSurface *surface, + enum vlResourceStatus *status +) +{ + assert(surface); + assert(surface->context); + assert(status); + + if (surface->render_fence && !surface->context->pipe->winsys->fence_signalled(surface->context->pipe->winsys, surface->render_fence, 0)) + { + *status = vlResourceStatusRendering; + return 0; + } + + if (surface->disp_fence && !surface->context->pipe->winsys->fence_signalled(surface->context->pipe->winsys, surface->disp_fence, 0)) + { + *status = vlResourceStatusDisplaying; + return 0; + } + + *status = vlResourceStatusFree; + + return 0; +} + +int vlSurfaceFlush +( + struct vlSurface *surface +) +{ + assert(surface); + assert(surface->context); + + surface->context->render->vlFlush(surface->context->render); + + return 0; +} + +int vlSurfaceSync +( + struct vlSurface *surface +) +{ + assert(surface); + assert(surface->context); + assert(surface->render_fence); + + surface->context->pipe->winsys->fence_finish(surface->context->pipe->winsys, surface->render_fence, 0); + + return 0; +} + +struct vlScreen* vlSurfaceGetScreen +( + struct vlSurface *surface +) +{ + assert(surface); + + return surface->screen; +} + +struct vlContext* vlBindToContext +( + struct vlSurface *surface, + struct vlContext *context +) +{ + struct vlContext *old; + + assert(surface); + + old = surface->context; + surface->context = context; + + return old; +} diff --git a/src/gallium/state_trackers/g3dvl/vl_surface.h b/src/gallium/state_trackers/g3dvl/vl_surface.h new file mode 100644 index 0000000000..133e1515ef --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_surface.h @@ -0,0 +1,86 @@ +#ifndef vl_surface_h +#define vl_surface_h + +#include "vl_types.h" + +#ifdef VL_INTERNAL +struct pipe_texture; + +struct vlSurface +{ + struct vlScreen *screen; + struct vlContext *context; + unsigned int width; + unsigned int height; + enum vlFormat format; + struct pipe_texture *texture; + struct pipe_fence_handle *render_fence; + struct pipe_fence_handle *disp_fence; +}; +#endif + +int vlCreateSurface +( + struct vlScreen *screen, + unsigned int width, + unsigned int height, + enum vlFormat format, + struct vlSurface **surface +); + +int vlDestroySurface +( + struct vlSurface *surface +); + +int vlRenderMacroBlocksMpeg2 +( + struct vlMpeg2MacroBlockBatch *batch, + struct vlSurface *surface +); + +int vlPutPicture +( + struct vlSurface *surface, + vlNativeDrawable drawable, + int srcx, + int srcy, + int srcw, + int srch, + int destx, + int desty, + int destw, + int desth, + int drawable_w, + int drawable_h, + enum vlPictureType picture_type +); + +int vlSurfaceGetStatus +( + struct vlSurface *surface, + enum vlResourceStatus *status +); + +int vlSurfaceFlush +( + struct vlSurface *surface +); + +int vlSurfaceSync +( + struct vlSurface *surface +); + +struct vlScreen* vlSurfaceGetScreen +( + struct vlSurface *surface +); + +struct vlContext* vlBindToContext +( + struct vlSurface *surface, + struct vlContext *context +); + +#endif diff --git a/src/gallium/state_trackers/g3dvl/vl_types.h b/src/gallium/state_trackers/g3dvl/vl_types.h new file mode 100644 index 0000000000..274e1f7437 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_types.h @@ -0,0 +1,115 @@ +#ifndef vl_types_h +#define vl_types_h + +#if 1 /*#ifdef X11*/ +#include <X11/Xlib.h> + +typedef Display* vlNativeDisplay; +typedef Drawable vlNativeDrawable; +#endif + +struct vlDisplay; +struct vlScreen; +struct vlContext; +struct vlSurface; + +enum vlResourceStatus +{ + vlResourceStatusFree, + vlResourceStatusRendering, + vlResourceStatusDisplaying +}; + +enum vlProfile +{ + vlProfileMpeg2Simple, + vlProfileMpeg2Main, + + vlProfileCount +}; + +enum vlEntryPoint +{ + vlEntryPointIDCT, + vlEntryPointMC, + vlEntryPointCSC, + + vlEntryPointCount +}; + +enum vlFormat +{ + vlFormatYCbCr420, + vlFormatYCbCr422, + vlFormatYCbCr444 +}; + +enum vlPictureType +{ + vlPictureTypeTopField, + vlPictureTypeBottomField, + vlPictureTypeFrame +}; + +enum vlMotionType +{ + vlMotionTypeField, + vlMotionTypeFrame, + vlMotionTypeDualPrime, + vlMotionType16x8 +}; + +enum vlFieldOrder +{ + vlFieldOrderFirst, + vlFieldOrderSecond +}; + +enum vlDCTType +{ + vlDCTTypeFrameCoded, + vlDCTTypeFieldCoded +}; + +struct vlVertex2f +{ + float x, y; +}; + +struct vlVertex4f +{ + float x, y, z, w; +}; + +enum vlMacroBlockType +{ + vlMacroBlockTypeIntra, + vlMacroBlockTypeFwdPredicted, + vlMacroBlockTypeBkwdPredicted, + vlMacroBlockTypeBiPredicted, + + vlNumMacroBlockTypes +}; + +struct vlMpeg2MacroBlock +{ + unsigned int mbx, mby; + enum vlMacroBlockType mb_type; + enum vlMotionType mo_type; + enum vlDCTType dct_type; + int PMV[2][2][2]; + unsigned int cbp; + short *blocks; +}; + +struct vlMpeg2MacroBlockBatch +{ + struct vlSurface *past_surface; + struct vlSurface *future_surface; + enum vlPictureType picture_type; + enum vlFieldOrder field_order; + unsigned int num_macroblocks; + struct vlMpeg2MacroBlock *macroblocks; +}; + +#endif diff --git a/src/gallium/state_trackers/g3dvl/vl_util.c b/src/gallium/state_trackers/g3dvl/vl_util.c new file mode 100644 index 0000000000..50aa9af66f --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_util.c @@ -0,0 +1,16 @@ +#include "vl_util.h" +#include <assert.h> + +unsigned int vlRoundUpPOT(unsigned int x) +{ + unsigned int i; + + assert(x > 0); + + --x; + + for (i = 1; i < sizeof(unsigned int) * 8; i <<= 1) + x |= x >> i; + + return x + 1; +} diff --git a/src/gallium/state_trackers/g3dvl/vl_util.h b/src/gallium/state_trackers/g3dvl/vl_util.h new file mode 100644 index 0000000000..bc98e79df4 --- /dev/null +++ b/src/gallium/state_trackers/g3dvl/vl_util.h @@ -0,0 +1,6 @@ +#ifndef vl_util_h +#define vl_util_h + +unsigned int vlRoundUpPOT(unsigned int x); + +#endif |