llvmpipe: more wipping

This commit is contained in:
Keith Whitwell 2009-10-08 11:47:33 +01:00
parent 89498d0153
commit e529170c11
9 changed files with 486 additions and 71 deletions

View File

@ -138,7 +138,7 @@ lp_vbuf_set_primitive(struct vbuf_render *vbr, unsigned prim)
struct llvmpipe_vbuf_render *cvbr = llvmpipe_vbuf_render(vbr);
struct setup_context *setup_ctx = cvbr->setup;
llvmpipe_setup_prepare( setup_ctx );
llvmpipe_update_state( setup_ctx->llvmpipe );
cvbr->llvmpipe->reduced_prim = u_reduced_prim(prim);
cvbr->prim = prim;

View File

@ -1,29 +1,4 @@
struct lp_rasterizer {
/* We can choose whatever layout for the internal tile storage we
* prefer:
*/
struct {
unsigned color[TILESIZE][TILESIZE];
unsigned depth[TILESIZE][TILESIZE];
char stencil[TILESIZE][TILESIZE];
} tile;
unsigned x;
unsigned y;
struct {
struct pipe_surface *color;
struct pipe_surface *zstencil;
unsigned clear_color;
unsigned clear_depth;
char clear_stencil;
} state;
};
struct lp_rasterizer *lp_rast_create( void )
{
return CALLOC_STRUCT(lp_rasterizer);
@ -96,31 +71,18 @@ void lp_rast_set_state( struct lp_rasterizer *rast,
const struct lp_rast_state *state )
{
rast->shader_state = state;
}
void lp_rast_triangle( struct lp_rasterizer *rast,
const struct lp_rast_triangle *inputs )
{
/* Set up the silly quad coef pointers
*/
for (i = 0; i < 4; i++) {
rast->quads[i].posCoef = inputs->posCoef;
rast->quads[i].coef = inputs->coef;
}
/* Scan the tile in 4x4 chunks (?) and figure out which bits to
* rasterize:
*/
lp->quad.first->begin( lp->quad.first );
}
void lp_rast_shade_tile( struct lp_rasterizer *rast,
const struct lp_rast_shader_inputs *inputs )
{
/* Set up the silly quad coef pointers
*/
for (i = 0; i < 4; i++) {
rast->quads[i].posCoef = inputs->posCoef;
rast->quads[i].posCoef = &inputs->posCoef;
rast->quads[i].coef = inputs->coef;
}

View File

@ -1,4 +1,7 @@
#ifndef LP_RAST_H
#define LP_RAST_H
/* Initially create and program a single rasterizer directly. Later
* will want multiple of these, one or two per core. At that stage
* will probably pass command buffers into the rasterizers rather than
@ -7,12 +10,16 @@
struct lp_rasterizer;
struct lp_rast_state {
/* State:
/* State for the shader:
*/
struct lp_jit_context jc;
/* Shader itself:
/* The shader itself. Probably we also need to pass a pointer to
* the tile color/z/stencil data somehow:
*/
void (*run)( struct lp_jit_context *jc,
struct quad_header **quads,
unsigned nr );
};
/* Coefficients necessary to run the shader at a given location:
@ -25,10 +32,6 @@ struct lp_rast_shader_inputs {
/* Attribute interpolation:
*/
float oneoverarea;
float x1;
float y1;
struct tgsi_interp_coef position_coef;
struct tgsi_interp_coef *coef;
};
@ -79,34 +82,48 @@ void lp_rast_start_tile( struct lp_rasterizer *,
unsigned x,
unsigned y );
void lp_rast_clear_color( struct lp_rasterizer * );
void lp_rast_clear_zstencil( struct lp_rasterizer * );
void lp_rast_load_color( struct lp_rasterizer * );
void lp_rast_load_zstencil( struct lp_rasterizer * );
/* Within a tile:
union lp_rast_cmd_arg {
const struct lp_rast_shader_inputs *shade_tile;
const struct lp_rast_triangle *triangle;
const struct lp_rast_state *set_state;
};
/* Binnable Commands:
*/
void lp_rast_set_state( struct lp_rasterizer *,
const struct lp_rast_state * );
void lp_rast_clear_color( struct lp_rasterizer *,
const union lp_rast_cmd_arg *);
void lp_rast_triangle( struct lp_rasterizer *,
const struct lp_rast_triangle * );
void lp_rast_clear_zstencil( struct lp_rasterizer *,
const union lp_rast_cmd_arg *);
void lp_rast_load_color( struct lp_rasterizer *,
const union lp_rast_cmd_arg *);
void lp_rast_load_zstencil( struct lp_rasterizer *,
const union lp_rast_cmd_arg *);
void lp_rast_set_state( struct lp_rasterizer *,
const union lp_rast_cmd_arg * );
void lp_rast_triangle( struct lp_rasterizer *,
const union lp_rast_cmd_arg * );
void lp_rast_shade_tile( struct lp_rasterizer *,
const struct lp_rast_shader_inputs * );
const union lp_rast_cmd_arg * );
/* End of tile:
*/
void lp_rast_store_color( struct lp_rasterizer * );
void lp_rast_store_color( struct lp_rasterizer *,
const union lp_rast_cmd_arg *);
void lp_rast_store_zstencil( struct lp_rasterizer * );
void lp_rast_store_zstencil( struct lp_rasterizer *,
const union lp_rast_cmd_arg *);
/* Shutdown:
*/
void lp_rast_destroy( struct lp_rasterizer * );
#endif

View File

@ -0,0 +1,31 @@
#ifndef LP_RAST_PRIV_H
#define LP_RAST_PRIV_H
#include "lp_rast.h"
struct lp_rasterizer {
/* We can choose whatever layout for the internal tile storage we
* prefer:
*/
struct {
unsigned color[TILESIZE][TILESIZE];
unsigned depth[TILESIZE][TILESIZE];
char stencil[TILESIZE][TILESIZE];
} tile;
unsigned x;
unsigned y;
struct {
struct pipe_surface *color;
struct pipe_surface *zstencil;
unsigned clear_color;
unsigned clear_depth;
char clear_stencil;
} state;
};
#endif

View File

@ -0,0 +1,348 @@
/**************************************************************************
*
* Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
* 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.
*
**************************************************************************/
/*
* Rasterization for binned triangles within a tile
*/
#include "lp_context.h"
#include "lp_quad.h"
#include "lp_quad_pipe.h"
#include "lp_setup.h"
#include "lp_state.h"
#include "draw/draw_context.h"
#include "draw/draw_private.h"
#include "draw/draw_vertex.h"
#include "pipe/p_shader_tokens.h"
#include "pipe/p_thread.h"
#include "util/u_math.h"
#include "util/u_memory.h"
#define BLOCKSIZE 4
/* Convert 8x8 block into four runs of quads and render each in turn.
*/
#if (BLOCKSIZE == 8)
static void block_full( struct triangle *tri, int x, int y )
{
struct quad_header *ptrs[4];
int i;
tri->quad[0].input.x0 = x + 0;
tri->quad[1].input.x0 = x + 2;
tri->quad[2].input.x0 = x + 4;
tri->quad[3].input.x0 = x + 6;
for (i = 0; i < 4; i++, y += 2) {
tri->quad[0].inout.mask = 0xf;
tri->quad[1].inout.mask = 0xf;
tri->quad[2].inout.mask = 0xf;
tri->quad[3].inout.mask = 0xf;
tri->quad[0].input.y0 = y;
tri->quad[1].input.y0 = y;
tri->quad[2].input.y0 = y;
tri->quad[3].input.y0 = y;
/* XXX: don't bother with this ptrs business */
ptrs[0] = &tri->quad[0];
ptrs[1] = &tri->quad[1];
ptrs[2] = &tri->quad[2];
ptrs[3] = &tri->quad[3];
tri->llvmpipe->quad.first->run( tri->llvmpipe->quad.first, ptrs, 4 );
}
}
#else
static void block_full( struct triangle *tri, int x, int y )
{
struct quad_header *ptrs[4];
int iy;
tri->quad[0].input.x0 = x + 0;
tri->quad[1].input.x0 = x + 2;
for (iy = 0; iy < 4; iy += 2) {
tri->quad[0].inout.mask = 0xf;
tri->quad[1].inout.mask = 0xf;
tri->quad[0].input.y0 = y + iy;
tri->quad[1].input.y0 = y + iy;
/* XXX: don't bother with this ptrs business */
ptrs[0] = &tri->quad[0];
ptrs[1] = &tri->quad[1];
tri->llvmpipe->quad.first->run( tri->llvmpipe->quad.first, ptrs, 2 );
}
}
#endif
static void
do_quad( struct lp_rasterizer *rast,
int x, int y,
float c1, float c2, float c3 )
{
struct triangle *tri = rast->tri;
struct quad_header *quad = &rast->quad[0];
float xstep1 = -tri->dy12;
float xstep2 = -tri->dy23;
float xstep3 = -tri->dy31;
float ystep1 = tri->dx12;
float ystep2 = tri->dx23;
float ystep3 = tri->dx31;
quad->input.x0 = x;
quad->input.y0 = y;
quad->inout.mask = 0;
if (c1 > 0 &&
c2 > 0 &&
c3 > 0)
quad->inout.mask |= 1;
if (c1 + xstep1 > 0 &&
c2 + xstep2 > 0 &&
c3 + xstep3 > 0)
quad->inout.mask |= 2;
if (c1 + ystep1 > 0 &&
c2 + ystep2 > 0 &&
c3 + ystep3 > 0)
quad->inout.mask |= 4;
if (c1 + ystep1 + xstep1 > 0 &&
c2 + ystep2 + xstep2 > 0 &&
c3 + ystep3 + xstep3 > 0)
quad->inout.mask |= 8;
if (quad->inout.mask)
rast->state->run( rast->state->state, &quad, 1 );
}
/* Evaluate each pixel in a block, generate a mask and possibly render
* the quad:
*/
static void
do_block( struct triangle *tri,
int x, int y,
float c1,
float c2,
float c3 )
{
const int step = 2;
float xstep1 = -step * tri->dy12;
float xstep2 = -step * tri->dy23;
float xstep3 = -step * tri->dy31;
float ystep1 = step * tri->dx12;
float ystep2 = step * tri->dx23;
float ystep3 = step * tri->dx31;
int ix, iy;
for (iy = 0; iy < BLOCKSIZE; iy += 2) {
float cx1 = c1;
float cx2 = c2;
float cx3 = c3;
for (ix = 0; ix < BLOCKSIZE; ix += 2) {
do_quad(tri, x+ix, y+iy, cx1, cx2, cx3);
cx1 += xstep1;
cx2 += xstep2;
cx3 += xstep3;
}
c1 += ystep1;
c2 += ystep2;
c3 += ystep3;
}
}
/* Scan the tile in chunks and figure out which pixels to rasterize
* for this triangle:
*/
void lp_rast_triangle( struct lp_rasterizer *rast,
const struct lp_rast_triangle *tri )
{
int minx, maxx, miny, maxy;
/* Clamp to tile dimensions:
*/
minx = MAX2(tri->maxx, rast->x);
miny = MAX2(tri->miny, rast->y);
maxx = MIN2(tri->maxx, rast->x + TILESIZE);
maxy = MIN2(tri->maxy, rast->y + TILESIZE);
if (miny == maxy ||
minx == maxx) {
debug_printf("%s: non-intersecting triangle in bin\n", __FUNCTION__);
//assert(0);
return;
}
/* Bind parameter interpolants:
*/
for (i = 0; i < Elements(rast->quad); i++) {
rast->quad[i].coef = tri->coef;
rast->quad[i].posCoef = &tri->position_coef;
}
/* Small area?
*/
if (miny + 16 > maxy &&
minx + 16 > maxx)
{
const int step = 2;
float xstep1 = -step * tri->dy12;
float xstep2 = -step * tri->dy23;
float xstep3 = -step * tri->dy31;
float ystep1 = step * tri->dx12;
float ystep2 = step * tri->dx23;
float ystep3 = step * tri->dx31;
float eo1 = tri->eo1 * step;
float eo2 = tri->eo2 * step;
float eo3 = tri->eo3 * step;
int x, y;
minx &= ~(step-1);
maxx &= ~(step-1);
/* Subdivide space into NxM blocks, where each block is square and
* power-of-four in dimension.
*
* Trivially accept or reject blocks, else jump to per-pixel
* examination above.
*/
for (y = miny; y < maxy; y += step)
{
float cx1 = c1;
float cx2 = c2;
float cx3 = c3;
for (x = minx; x < maxx; x += step)
{
if (cx1 + eo1 < 0 ||
cx2 + eo2 < 0 ||
cx3 + eo3 < 0)
{
}
else
{
do_quad(&tri, x, y, cx1, cx2, cx3);
}
/* Iterate cx values across the region:
*/
cx1 += xstep1;
cx2 += xstep2;
cx3 += xstep3;
}
/* Iterate c values down the region:
*/
c1 += ystep1;
c2 += ystep2;
c3 += ystep3;
}
}
else
{
const int step = BLOCKSIZE;
float ei1 = tri->ei1 * step;
float ei2 = tri->ei2 * step;
float ei3 = tri->ei3 * step;
float eo1 = tri->eo1 * step;
float eo2 = tri->eo2 * step;
float eo3 = tri->eo3 * step;
float xstep1 = -step * tri->dy12;
float xstep2 = -step * tri->dy23;
float xstep3 = -step * tri->dy31;
float ystep1 = step * tri->dx12;
float ystep2 = step * tri->dx23;
float ystep3 = step * tri->dx31;
int x, y;
minx &= ~(step-1);
miny &= ~(step-1);
for (y = miny; y < maxy; y += step)
{
float cx1 = c1;
float cx2 = c2;
float cx3 = c3;
for (x = minx; x < maxx; x += step)
{
if (cx1 + eo1 < 0 ||
cx2 + eo2 < 0 ||
cx3 + eo3 < 0)
{
}
else if (cx1 + ei1 > 0 &&
cx2 + ei2 > 0 &&
cx3 + ei3 > 0)
{
block_full(&tri, x, y); /* trivial accept */
}
else
{
do_block(&tri, x, y, cx1, cx2, cx3);
}
/* Iterate cx values across the region:
*/
cx1 += xstep1;
cx2 += xstep2;
cx3 += xstep3;
}
/* Iterate c values down the region:
*/
c1 += ystep1;
c2 += ystep2;
c3 += ystep3;
}
}
}

View File

@ -48,6 +48,23 @@
#define DEBUG_VERTS 0
void
llvmpipe_setup_flush()
{
}
void
llvmpipe_setup_bind_framebuffer()
{
}
void
llvmpipe_setup_clear()
{
}
/* Stubs for lines & points for now:
*/
void

View File

@ -1,7 +1,20 @@
void
rasterize( struct llvmpipe_context *llvmpipe,
struct binned_scene *scene )
lp_setup_rasterize( struct llvmpipe_context *llvmpipe,
struct binned_scene *scene )
{
lp_rast_bind_surfaces( rast, scene->framebuffer );
for (i = 0; i < scene->tiles_x; i++) {
for (j = 0; j < scene->tiles_y; j++) {
lp_rast_start_tile( rast, i * TILESIZE, j * TILESIZE );
for (block = scene->tile[i][j].first; block; block = block->next) {
for (k = 0; k < block->nr_cmds; k++) {
block->cmd[k].func( rast, block->cmd[k].arg );
}
}
}
}
}

View File

@ -728,9 +728,9 @@ static void triangle_nop( struct llvmpipe_context *llvmpipe,
{
}
/**
* Do setup for triangle rasterization, then render the triangle.
*/
void setup_prepare_tri( struct llvmpipe_context *llvmpipe )
{
llvmpipe->ccw_is_frontface = (llvmpipe->rasterizer->front_winding ==

View File

@ -232,6 +232,22 @@ update_tgsi_samplers( struct llvmpipe_context *llvmpipe )
llvmpipe->jit_context.samplers = (struct tgsi_sampler **)llvmpipe->tgsi.frag_samplers_list;
}
static void
update_culling()
{
if (lp->reduced_api_prim == PIPE_PRIM_TRIANGLES &&
lp->rasterizer->fill_cw == PIPE_POLYGON_MODE_FILL &&
lp->rasterizer->fill_ccw == PIPE_POLYGON_MODE_FILL) {
/* we'll do culling */
setup->winding = lp->rasterizer->cull_mode;
}
else {
/* 'draw' will do culling */
setup->winding = PIPE_WINDING_NONE;
}
}
/* Hopefully this will remain quite simple, otherwise need to pull in
* something like the state tracker mechanism.
*/
@ -270,3 +286,14 @@ void llvmpipe_update_derived( struct llvmpipe_context *llvmpipe )
llvmpipe->dirty = 0;
}
void llvmpipe_prepare( )
{
struct llvmpipe_context *lp = setup->llvmpipe;
if (lp->dirty) {
llvmpipe_update_derived(lp);
}
}