From: Christian König Date: Sun, 28 Nov 2010 13:48:31 +0000 (+0100) Subject: use a shadow buffer for vertex data to optimize memory access X-Git-Tag: 062012170305~4979^2~326 X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=4abe7382882a451a7750ccc451b8568768d122cb;p=profile%2Fivi%2Fmesa.git use a shadow buffer for vertex data to optimize memory access --- diff --git a/src/gallium/auxiliary/Makefile b/src/gallium/auxiliary/Makefile index bf90a70..997478f 100644 --- a/src/gallium/auxiliary/Makefile +++ b/src/gallium/auxiliary/Makefile @@ -147,7 +147,8 @@ C_SOURCES = \ vl/vl_mpeg12_mc_renderer.c \ vl/vl_compositor.c \ vl/vl_csc.c \ - vl/vl_idct.c + vl/vl_idct.c \ + vl/vl_vertex_buffers.c GALLIVM_SOURCES = \ gallivm/lp_bld_arit.c \ diff --git a/src/gallium/auxiliary/vl/vl_idct.c b/src/gallium/auxiliary/vl/vl_idct.c index 2466d5b..22feff8 100644 --- a/src/gallium/auxiliary/vl/vl_idct.c +++ b/src/gallium/auxiliary/vl/vl_idct.c @@ -26,6 +26,7 @@ **************************************************************************/ #include "vl_idct.h" +#include "vl_vertex_buffers.h" #include "util/u_draw.h" #include #include @@ -78,11 +79,6 @@ static const float const_matrix[8][8] = { { 0.0975451f, -0.2777850f, 0.4157350f, -0.4903930f, 0.4903930f, -0.4157350f, 0.277786f, -0.0975458f } }; -/* vertices for a quad covering a block */ -static const struct vertex2f const_quad[4] = { - {0.0f, 0.0f}, {1.0f, 0.0f}, {1.0f, 1.0f}, {0.0f, 1.0f} -}; - static void * create_vert_shader(struct vl_idct *idct, bool calc_src_cords) { @@ -409,11 +405,6 @@ init_buffers(struct vl_idct *idct) struct pipe_vertex_element vertex_elems[2]; unsigned i; - idct->max_blocks = - align(idct->destination->width0, BLOCK_WIDTH) / BLOCK_WIDTH * - align(idct->destination->height0, BLOCK_HEIGHT) / BLOCK_HEIGHT * - idct->destination->depth0; - memset(&template, 0, sizeof(struct pipe_resource)); template.last_level = 0; template.depth0 = 1; @@ -443,15 +434,7 @@ init_buffers(struct vl_idct *idct) idct->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, idct->textures.all[i], &sampler_view); } - idct->vertex_bufs.individual.quad.stride = sizeof(struct vertex2f); - idct->vertex_bufs.individual.quad.max_index = 4 * idct->max_blocks - 1; - idct->vertex_bufs.individual.quad.buffer_offset = 0; - idct->vertex_bufs.individual.quad.buffer = pipe_buffer_create - ( - idct->pipe->screen, - PIPE_BIND_VERTEX_BUFFER, - sizeof(struct vertex2f) * 4 * idct->max_blocks - ); + idct->vertex_bufs.individual.quad = vl_vb_upload_quads(idct->pipe, idct->max_blocks); if(idct->vertex_bufs.individual.quad.buffer == NULL) return false; @@ -504,35 +487,11 @@ cleanup_buffers(struct vl_idct *idct) } static void -init_constants(struct vl_idct *idct) -{ - struct pipe_transfer *buf_transfer; - struct vertex2f *v; - - unsigned i; - - /* quad vectors */ - v = pipe_buffer_map - ( - idct->pipe, - idct->vertex_bufs.individual.quad.buffer, - PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD, - &buf_transfer - ); - for ( i = 0; i < idct->max_blocks; ++i) - memcpy(v + i * 4, &const_quad, sizeof(const_quad)); - pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.quad.buffer, buf_transfer); -} - -static void init_state(struct vl_idct *idct) { struct pipe_sampler_state sampler; unsigned i; - idct->num_blocks = 0; - idct->num_empty_blocks = 0; - idct->viewport[0].scale[0] = idct->textures.individual.intermediate->width0; idct->viewport[0].scale[1] = idct->textures.individual.intermediate->height0; @@ -674,26 +633,11 @@ xfer_buffers_map(struct vl_idct *idct) ); idct->texels = idct->pipe->transfer_map(idct->pipe, idct->tex_transfer); - - idct->vectors = pipe_buffer_map - ( - idct->pipe, - idct->vertex_bufs.individual.pos.buffer, - PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD, - &idct->vec_transfer - ); - - idct->next_empty_block.l_x = ~1; - idct->next_empty_block.l_y = ~1; - idct->next_empty_block.r_x = ~1; - idct->next_empty_block.r_y = ~1; } static void xfer_buffers_unmap(struct vl_idct *idct) { - pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.pos.buffer, idct->vec_transfer); - idct->pipe->transfer_unmap(idct->pipe, idct->tex_transfer); idct->pipe->transfer_destroy(idct->pipe, idct->tex_transfer); } @@ -708,6 +652,11 @@ vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resour pipe_resource_reference(&idct->textures.individual.transpose, matrix); pipe_resource_reference(&idct->destination, dst); + idct->max_blocks = + align(idct->destination->width0, BLOCK_WIDTH) / BLOCK_WIDTH * + align(idct->destination->height0, BLOCK_HEIGHT) / BLOCK_HEIGHT * + idct->destination->depth0; + if(!init_buffers(idct)) return false; @@ -716,9 +665,21 @@ vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resour return false; } + if(!vl_vb_init(&idct->blocks, idct->max_blocks)) { + cleanup_shaders(idct); + cleanup_buffers(idct); + return false; + } + + if(!vl_vb_init(&idct->empty_blocks, idct->max_blocks)) { + vl_vb_cleanup(&idct->blocks); + cleanup_shaders(idct); + cleanup_buffers(idct); + return false; + } + init_state(idct); - init_constants(idct); xfer_buffers_map(idct); return true; @@ -727,6 +688,8 @@ vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resour void vl_idct_cleanup(struct vl_idct *idct) { + vl_vb_cleanup(&idct->blocks); + vl_vb_cleanup(&idct->empty_blocks); cleanup_shaders(idct); cleanup_buffers(idct); @@ -735,43 +698,9 @@ vl_idct_cleanup(struct vl_idct *idct) pipe_resource_reference(&idct->destination, NULL); } -static void -flush_empty_block(struct vl_idct *idct, unsigned new_x, unsigned new_y) -{ - if (idct->next_empty_block.l_x == ~1 || - idct->next_empty_block.l_y == ~1) { - - idct->next_empty_block.l_x = new_x; - idct->next_empty_block.l_y = new_y; - - } else if (idct->next_empty_block.r_x != (new_x - 1) || - idct->next_empty_block.r_y != new_y) { - - struct vertex2f l, r, *v_dst; - - v_dst = idct->vectors + (idct->max_blocks - idct->num_empty_blocks) * 4 - 4; - - l.x = idct->next_empty_block.l_x; - l.y = idct->next_empty_block.l_y; - r.x = idct->next_empty_block.r_x; - r.y = idct->next_empty_block.r_y; - v_dst[0] = v_dst[3] = l; - v_dst[1] = v_dst[2] = r; - - idct->next_empty_block.l_x = new_x; - idct->next_empty_block.l_y = new_y; - idct->num_empty_blocks++; - } - - idct->next_empty_block.r_x = new_x; - idct->next_empty_block.r_y = new_y; -} - void vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block) { - struct vertex2f v, *v_dst; - unsigned tex_pitch; short *texels; @@ -786,32 +715,38 @@ vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block) for (i = 0; i < BLOCK_HEIGHT; ++i) memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * sizeof(short)); - /* non empty blocks fills the vector buffer from left to right */ - v_dst = idct->vectors + idct->num_blocks * 4; - - idct->num_blocks++; - - v.x = x; - v.y = y; - - for (i = 0; i < 4; ++i) { - v_dst[i] = v; - } - + vl_vb_add_block(&idct->blocks, false, x, y); } else { - /* while empty blocks fills the vector buffer from right to left */ - flush_empty_block(idct, x, y); + vl_vb_add_block(&idct->empty_blocks, true, x, y); } } void vl_idct_flush(struct vl_idct *idct) { - flush_empty_block(idct, ~1, ~1); + struct pipe_transfer *vec_transfer; + struct quadf *vectors; + unsigned num_blocks, num_empty_blocks; + + assert(idct); + + vectors = pipe_buffer_map + ( + idct->pipe, + idct->vertex_bufs.individual.pos.buffer, + PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD, + &vec_transfer + ); + + num_blocks = vl_vb_upload(&idct->blocks, vectors); + num_empty_blocks = vl_vb_upload(&idct->empty_blocks, vectors + num_blocks); + + pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.pos.buffer, vec_transfer); + xfer_buffers_unmap(idct); - if(idct->num_blocks > 0) { + if(num_blocks > 0) { /* first stage */ idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[0]); @@ -824,7 +759,7 @@ vl_idct_flush(struct vl_idct *idct) idct->pipe->bind_vs_state(idct->pipe, idct->matrix_vs); idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs); - util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, idct->num_blocks * 4); + util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_blocks * 4); /* second stage */ idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[1]); @@ -837,10 +772,10 @@ vl_idct_flush(struct vl_idct *idct) idct->pipe->bind_vs_state(idct->pipe, idct->transpose_vs); idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs); - util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, idct->num_blocks * 4); + util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_blocks * 4); } - if(idct->num_empty_blocks > 0) { + if(num_empty_blocks > 0) { /* empty block handling */ idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[1]); @@ -851,12 +786,8 @@ vl_idct_flush(struct vl_idct *idct) idct->pipe->bind_vs_state(idct->pipe, idct->eb_vs); idct->pipe->bind_fs_state(idct->pipe, idct->eb_fs); - util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, - (idct->max_blocks - idct->num_empty_blocks) * 4, - idct->num_empty_blocks * 4); + util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, num_blocks * 4, num_empty_blocks * 4); } - idct->num_blocks = 0; - idct->num_empty_blocks = 0; xfer_buffers_map(idct); } diff --git a/src/gallium/auxiliary/vl/vl_idct.h b/src/gallium/auxiliary/vl/vl_idct.h index c26f5cb..94a5c73 100644 --- a/src/gallium/auxiliary/vl/vl_idct.h +++ b/src/gallium/auxiliary/vl/vl_idct.h @@ -29,6 +29,7 @@ #define vl_idct_h #include +#include "vl_vertex_buffers.h" struct vl_idct { @@ -82,20 +83,11 @@ struct vl_idct struct { struct pipe_vertex_buffer quad, pos; } individual; } vertex_bufs; - unsigned num_blocks; - - struct - { - unsigned l_x, l_y, r_x, r_y; - } next_empty_block; - - unsigned num_empty_blocks; + struct vl_vertex_buffer blocks; + struct vl_vertex_buffer empty_blocks; struct pipe_transfer *tex_transfer; short *texels; - - struct pipe_transfer *vec_transfer; - struct vertex2f *vectors; }; struct pipe_resource *vl_idct_upload_matrix(struct pipe_context *pipe); diff --git a/src/gallium/auxiliary/vl/vl_types.h b/src/gallium/auxiliary/vl/vl_types.h index ce17554..eeabd43 100644 --- a/src/gallium/auxiliary/vl/vl_types.h +++ b/src/gallium/auxiliary/vl/vl_types.h @@ -38,4 +38,9 @@ struct vertex4f float x, y, z, w; }; +struct quadf +{ + struct vertex2f bl, tl, tr, br; +}; + #endif /* vl_types_h */ diff --git a/src/gallium/auxiliary/vl/vl_vertex_buffers.c b/src/gallium/auxiliary/vl/vl_vertex_buffers.c new file mode 100644 index 0000000..6df11db --- /dev/null +++ b/src/gallium/auxiliary/vl/vl_vertex_buffers.c @@ -0,0 +1,116 @@ +/************************************************************************** + * + * Copyright 2010 Christian König + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include +#include +#include +#include +#include +#include "vl_vertex_buffers.h" +#include "vl_types.h" + +/* vertices for a quad covering a block */ +static const struct quadf const_quad = { + {0.0f, 1.0f}, {0.0f, 0.0f}, {1.0f, 0.0f}, {1.0f, 1.0f} +}; + +struct pipe_vertex_buffer +vl_vb_upload_quads(struct pipe_context *pipe, unsigned max_blocks) +{ + struct pipe_vertex_buffer quad; + struct pipe_transfer *buf_transfer; + struct quadf *v; + + unsigned i; + + assert(pipe); + assert(max_blocks); + + /* create buffer */ + quad.stride = sizeof(struct vertex2f); + quad.max_index = 4 * max_blocks - 1; + quad.buffer_offset = 0; + quad.buffer = pipe_buffer_create + ( + pipe->screen, + PIPE_BIND_VERTEX_BUFFER, + sizeof(struct vertex2f) * 4 * max_blocks + ); + + if(!quad.buffer) + return quad; + + /* and fill it */ + v = pipe_buffer_map + ( + pipe, + quad.buffer, + PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD, + &buf_transfer + ); + + for ( i = 0; i < max_blocks; ++i) + memcpy(v + i, &const_quad, sizeof(const_quad)); + + pipe_buffer_unmap(pipe, quad.buffer, buf_transfer); + + return quad; +} + +bool +vl_vb_init(struct vl_vertex_buffer *buffer, unsigned max_blocks) +{ + assert(buffer); + + buffer->num_blocks = 0; + buffer->blocks = MALLOC(max_blocks * sizeof(struct quadf)); + return buffer->blocks != NULL; +} + +unsigned +vl_vb_upload(struct vl_vertex_buffer *buffer, struct quadf *dst) +{ + unsigned todo; + + assert(buffer); + + todo = buffer->num_blocks; + buffer->num_blocks = 0; + + if(todo) + memcpy(dst, buffer->blocks, sizeof(struct quadf) * todo); + + return todo; +} + +void +vl_vb_cleanup(struct vl_vertex_buffer *buffer) +{ + assert(buffer); + + FREE(buffer->blocks); +} diff --git a/src/gallium/auxiliary/vl/vl_vertex_buffers.h b/src/gallium/auxiliary/vl/vl_vertex_buffers.h new file mode 100644 index 0000000..43ddc34 --- /dev/null +++ b/src/gallium/auxiliary/vl/vl_vertex_buffers.h @@ -0,0 +1,75 @@ +/************************************************************************** + * + * Copyright 2010 Christian König + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ +#ifndef vl_vertex_buffers_h +#define vl_vertex_buffers_h + +#include +#include +#include "vl_types.h" + +struct vl_vertex_buffer +{ + unsigned num_blocks; + struct quadf *blocks; +}; + +struct pipe_vertex_buffer vl_vb_upload_quads(struct pipe_context *pipe, unsigned max_blocks); + +bool vl_vb_init(struct vl_vertex_buffer *buffer, unsigned max_blocks); + +static inline bool +vl_vb_add_block(struct vl_vertex_buffer *buffer, bool allow_merge, signed x, signed y) +{ + struct quadf *quad; + + assert(buffer); + + allow_merge &= buffer->num_blocks > 0; + if (allow_merge) { + + quad = buffer->blocks + buffer->num_blocks - 1; + if(quad->tr.x == (x - 1) && quad->br.x == (x - 1) && + quad->tr.y == y && quad->br.y == y) { + + quad->tr.x = quad->br.x = x; + quad->tr.y = quad->br.y = y; + return true; + } + } + + quad = buffer->blocks + buffer->num_blocks; + quad->bl.x = quad->tl.x = quad->tr.x = quad->br.x = x; + quad->bl.y = quad->tl.y = quad->tr.y = quad->br.y = y; + buffer->num_blocks++; + return false; +} + +unsigned vl_vb_upload(struct vl_vertex_buffer *buffer, struct quadf *dst); + +void vl_vb_cleanup(struct vl_vertex_buffer *buffer); + +#endif