**************************************************************************/
/**
- * \brief Primitive rasterization/rendering (points, lines, triangles)
+ * Tiling engine.
*
- * \author Keith Whitwell <keith@tungstengraphics.com>
- * \author Brian Paul
+ * Builds per-tile display lists and executes them on calls to
+ * lp_setup_flush().
*/
-#include "lp_context.h"
-#include "lp_quad.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_format.h"
-#include "util/u_math.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
#include "util/u_memory.h"
-#include "lp_bld_debug.h"
-#include "lp_tile_cache.h"
-#include "lp_tile_soa.h"
-
-
-#define DEBUG_VERTS 0
-#define DEBUG_FRAGS 0
-
-/**
- * Triangle edge info
- */
-struct edge {
- float dx; /**< X(v1) - X(v0), used only during setup */
- float dy; /**< Y(v1) - Y(v0), used only during setup */
- float dxdy; /**< dx/dy */
- float sx, sy; /**< first sample point coord */
- int lines; /**< number of lines on this edge */
-};
-
-
-#define MAX_QUADS 16
-
-
-/**
- * Triangle setup info (derived from draw_stage).
- * Also used for line drawing (taking some liberties).
- */
-struct setup_context {
- struct llvmpipe_context *llvmpipe;
-
- /* Vertices are just an array of floats making up each attribute in
- * turn. Currently fixed at 4 floats, but should change in time.
- * Codegen will help cope with this.
- */
- const float (*vmax)[4];
- const float (*vmid)[4];
- const float (*vmin)[4];
- const float (*vprovoke)[4];
-
- struct edge ebot;
- struct edge etop;
- struct edge emaj;
-
- float oneoverarea;
- int facing;
-
- float pixel_offset;
-
- struct quad_header quad[MAX_QUADS];
- struct quad_header *quad_ptrs[MAX_QUADS];
- unsigned count;
+#include "util/u_pack_color.h"
+#include "util/u_surface.h"
+#include "lp_scene.h"
+#include "lp_scene_queue.h"
+#include "lp_buffer.h"
+#include "lp_texture.h"
+#include "lp_debug.h"
+#include "lp_fence.h"
+#include "lp_rast.h"
+#include "lp_setup_context.h"
- struct quad_interp_coef coef;
+#include "draw/draw_context.h"
+#include "draw/draw_vbuf.h"
- struct {
- int left[2]; /**< [0] = row0, [1] = row1 */
- int right[2];
- int y;
- } span;
-#if DEBUG_FRAGS
- uint numFragsEmitted; /**< per primitive */
- uint numFragsWritten; /**< per primitive */
-#endif
+/** XXX temporary value, temporary here */
+#define MAX_SCENES 2
- unsigned winding; /* which winding to cull */
-};
+static void set_scene_state( struct setup_context *, unsigned );
-/**
- * Execute fragment shader for the four fragments in the quad.
- */
-ALIGN_STACK
-static void
-shade_quads(struct llvmpipe_context *llvmpipe,
- struct quad_header *quads[],
- unsigned nr)
+struct lp_scene *
+lp_setup_get_current_scene(struct setup_context *setup)
{
- struct lp_fragment_shader *fs = llvmpipe->fs;
- struct quad_header *quad = quads[0];
- const unsigned x = quad->input.x0;
- const unsigned y = quad->input.y0;
- uint8_t *tile;
- uint8_t *color;
- void *depth;
- uint32_t ALIGN16_ATTRIB mask[4][NUM_CHANNELS];
- unsigned chan_index;
- unsigned q;
-
- assert(fs->current);
- if(!fs->current)
- return;
+ if (!setup->scene) {
+ /* wait for a free/empty bin */
+ setup->scene = lp_scene_dequeue(setup->empty_scenes);
+ if(0)lp_scene_reset( setup->scene ); /* XXX temporary? */
- /* Sanity checks */
- assert(nr * QUAD_SIZE == TILE_VECTOR_HEIGHT * TILE_VECTOR_WIDTH);
- assert(x % TILE_VECTOR_WIDTH == 0);
- assert(y % TILE_VECTOR_HEIGHT == 0);
- for (q = 0; q < nr; ++q) {
- assert(quads[q]->input.x0 == x + q*2);
- assert(quads[q]->input.y0 == y);
- }
-
- /* mask */
- for (q = 0; q < 4; ++q)
- for (chan_index = 0; chan_index < NUM_CHANNELS; ++chan_index)
- mask[q][chan_index] = quads[q]->inout.mask & (1 << chan_index) ? ~0 : 0;
-
- /* color buffer */
- if(llvmpipe->framebuffer.nr_cbufs >= 1 &&
- llvmpipe->framebuffer.cbufs[0]) {
- tile = lp_get_cached_tile(llvmpipe->cbuf_cache[0], x, y);
- color = &TILE_PIXEL(tile, x & (TILE_SIZE-1), y & (TILE_SIZE-1), 0);
- }
- else
- color = NULL;
-
- /* depth buffer */
- if(llvmpipe->zsbuf_map) {
- assert((x % 2) == 0);
- assert((y % 2) == 0);
- depth = llvmpipe->zsbuf_map +
- y*llvmpipe->zsbuf_transfer->stride +
- 2*x*util_format_get_blocksize(llvmpipe->zsbuf_transfer->texture->format);
+ lp_scene_set_framebuffer_size(setup->scene,
+ setup->fb.width,
+ setup->fb.height);
}
- else
- depth = NULL;
-
- /* XXX: This will most likely fail on 32bit x86 without -mstackrealign */
- assert(lp_check_alignment(mask, 16));
-
- assert(lp_check_alignment(depth, 16));
- assert(lp_check_alignment(color, 16));
- assert(lp_check_alignment(llvmpipe->jit_context.blend_color, 16));
-
- /* run shader */
- fs->current->jit_function( &llvmpipe->jit_context,
- x, y,
- quad->coef->a0,
- quad->coef->dadx,
- quad->coef->dady,
- &mask[0][0],
- color,
- depth);
+ return setup->scene;
}
-
-
-/**
- * Do triangle cull test using tri determinant (sign indicates orientation)
- * \return true if triangle is to be culled.
- */
-static INLINE boolean
-cull_tri(const struct setup_context *setup, float det)
+static void
+first_triangle( struct setup_context *setup,
+ const float (*v0)[4],
+ const float (*v1)[4],
+ const float (*v2)[4])
{
- if (det != 0) {
- /* if (det < 0 then Z points toward camera and triangle is
- * counter-clockwise winding.
- */
- unsigned winding = (det < 0) ? PIPE_WINDING_CCW : PIPE_WINDING_CW;
-
- if ((winding & setup->winding) == 0)
- return FALSE;
- }
-
- /* Culled:
- */
- return TRUE;
+ set_scene_state( setup, SETUP_ACTIVE );
+ lp_setup_choose_triangle( setup );
+ setup->triangle( setup, v0, v1, v2 );
}
-
-
-/**
- * Clip setup->quad against the scissor/surface bounds.
- */
-static INLINE void
-quad_clip( struct setup_context *setup, struct quad_header *quad )
+static void
+first_line( struct setup_context *setup,
+ const float (*v0)[4],
+ const float (*v1)[4])
{
- const struct pipe_scissor_state *cliprect = &setup->llvmpipe->cliprect;
- const int minx = (int) cliprect->minx;
- const int maxx = (int) cliprect->maxx;
- const int miny = (int) cliprect->miny;
- const int maxy = (int) cliprect->maxy;
-
- if (quad->input.x0 >= maxx ||
- quad->input.y0 >= maxy ||
- quad->input.x0 + 1 < minx ||
- quad->input.y0 + 1 < miny) {
- /* totally clipped */
- quad->inout.mask = 0x0;
- return;
- }
- if (quad->input.x0 < minx)
- quad->inout.mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT);
- if (quad->input.y0 < miny)
- quad->inout.mask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT);
- if (quad->input.x0 == maxx - 1)
- quad->inout.mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
- if (quad->input.y0 == maxy - 1)
- quad->inout.mask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT);
+ set_scene_state( setup, SETUP_ACTIVE );
+ lp_setup_choose_line( setup );
+ setup->line( setup, v0, v1 );
}
-
-
-/**
- * Given an X or Y coordinate, return the block/quad coordinate that it
- * belongs to.
- */
-static INLINE int block( int x )
+static void
+first_point( struct setup_context *setup,
+ const float (*v0)[4])
{
- return x & ~(2-1);
+ set_scene_state( setup, SETUP_ACTIVE );
+ lp_setup_choose_point( setup );
+ setup->point( setup, v0 );
}
-static INLINE int block_x( int x )
+static void reset_context( struct setup_context *setup )
{
- return x & ~(TILE_VECTOR_WIDTH - 1);
-}
+ LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
+ /* Reset derived state */
+ setup->constants.stored_size = 0;
+ setup->constants.stored_data = NULL;
+ setup->fs.stored = NULL;
+ setup->dirty = ~0;
-/**
- * Emit a quad (pass to next stage) with clipping.
- */
-static INLINE void
-clip_emit_quad( struct setup_context *setup, struct quad_header *quad )
-{
- quad_clip( setup, quad );
-
- if (quad->inout.mask) {
- struct llvmpipe_context *lp = setup->llvmpipe;
-
-#if 1
- /* XXX: The blender expects 4 quads. This is far from efficient, but
- * until we codegenerate single-quad variants of the fragment pipeline
- * we need this hack. */
- const unsigned nr_quads = TILE_VECTOR_HEIGHT*TILE_VECTOR_WIDTH/QUAD_SIZE;
- struct quad_header quads[4];
- struct quad_header *quad_ptrs[4];
- int x0 = block_x(quad->input.x0);
- unsigned i;
-
- assert(nr_quads == 4);
-
- for(i = 0; i < nr_quads; ++i) {
- int x = x0 + 2*i;
- if(x == quad->input.x0)
- memcpy(&quads[i], quad, sizeof quads[i]);
- else {
- memset(&quads[i], 0, sizeof quads[i]);
- quads[i].input.x0 = x;
- quads[i].input.y0 = quad->input.y0;
- quads[i].coef = quad->coef;
- }
- quad_ptrs[i] = &quads[i];
- }
+ /* no current bin */
+ setup->scene = NULL;
- shade_quads( lp, quad_ptrs, nr_quads );
-#else
- shade_quads( lp, &quad, 1 );
-#endif
- }
+ /* Reset some state:
+ */
+ setup->clear.flags = 0;
+
+ /* Have an explicit "start-binning" call and get rid of this
+ * pointer twiddling?
+ */
+ setup->line = first_line;
+ setup->point = first_point;
+ setup->triangle = first_triangle;
}
-/**
- * Render a horizontal span of quads
- */
-static void flush_spans( struct setup_context *setup )
+/** Rasterize all scene's bins */
+static void
+lp_setup_rasterize_scene( struct setup_context *setup,
+ boolean write_depth )
{
- const int step = TILE_VECTOR_WIDTH;
- const int xleft0 = setup->span.left[0];
- const int xleft1 = setup->span.left[1];
- const int xright0 = setup->span.right[0];
- const int xright1 = setup->span.right[1];
-
-
- int minleft = block_x(MIN2(xleft0, xleft1));
- int maxright = MAX2(xright0, xright1);
- int x;
-
- for (x = minleft; x < maxright; x += step) {
- unsigned skip_left0 = CLAMP(xleft0 - x, 0, step);
- unsigned skip_left1 = CLAMP(xleft1 - x, 0, step);
- unsigned skip_right0 = CLAMP(x + step - xright0, 0, step);
- unsigned skip_right1 = CLAMP(x + step - xright1, 0, step);
- unsigned lx = x;
- const unsigned nr_quads = TILE_VECTOR_HEIGHT*TILE_VECTOR_WIDTH/QUAD_SIZE;
- unsigned q = 0;
-
- unsigned skipmask_left0 = (1U << skip_left0) - 1U;
- unsigned skipmask_left1 = (1U << skip_left1) - 1U;
-
- /* These calculations fail when step == 32 and skip_right == 0.
- */
- unsigned skipmask_right0 = ~0U << (unsigned)(step - skip_right0);
- unsigned skipmask_right1 = ~0U << (unsigned)(step - skip_right1);
-
- unsigned mask0 = ~skipmask_left0 & ~skipmask_right0;
- unsigned mask1 = ~skipmask_left1 & ~skipmask_right1;
-
- if (mask0 | mask1) {
- for(q = 0; q < nr_quads; ++q) {
- unsigned quadmask = (mask0 & 3) | ((mask1 & 3) << 2);
- setup->quad[q].input.x0 = lx;
- setup->quad[q].input.y0 = setup->span.y;
- setup->quad[q].inout.mask = quadmask;
- setup->quad_ptrs[q] = &setup->quad[q];
- mask0 >>= 2;
- mask1 >>= 2;
- lx += 2;
- }
- assert(!(mask0 | mask1));
+ struct lp_scene *scene = lp_setup_get_current_scene(setup);
- shade_quads(setup->llvmpipe, setup->quad_ptrs, nr_quads );
- }
- }
+ lp_rasterize_scene(setup->rast,
+ scene,
+ &setup->fb,
+ write_depth);
+ reset_context( setup );
- setup->span.y = 0;
- setup->span.right[0] = 0;
- setup->span.right[1] = 0;
- setup->span.left[0] = 1000000; /* greater than right[0] */
- setup->span.left[1] = 1000000; /* greater than right[1] */
+ LP_DBG(DEBUG_SETUP, "%s done \n", __FUNCTION__);
}
-#if DEBUG_VERTS
-static void print_vertex(const struct setup_context *setup,
- const float (*v)[4])
-{
- int i;
- debug_printf(" Vertex: (%p)\n", v);
- for (i = 0; i < setup->quad[0].nr_attrs; i++) {
- debug_printf(" %d: %f %f %f %f\n", i,
- v[i][0], v[i][1], v[i][2], v[i][3]);
- if (util_is_inf_or_nan(v[i][0])) {
- debug_printf(" NaN!\n");
- }
- }
-}
-#endif
-/**
- * Sort the vertices from top to bottom order, setting up the triangle
- * edge fields (ebot, emaj, etop).
- * \return FALSE if coords are inf/nan (cull the tri), TRUE otherwise
- */
-static boolean setup_sort_vertices( struct setup_context *setup,
- float det,
- const float (*v0)[4],
- const float (*v1)[4],
- const float (*v2)[4] )
+static void
+begin_binning( struct setup_context *setup )
{
- setup->vprovoke = v2;
-
- /* determine bottom to top order of vertices */
- {
- float y0 = v0[0][1];
- float y1 = v1[0][1];
- float y2 = v2[0][1];
- if (y0 <= y1) {
- if (y1 <= y2) {
- /* y0<=y1<=y2 */
- setup->vmin = v0;
- setup->vmid = v1;
- setup->vmax = v2;
- }
- else if (y2 <= y0) {
- /* y2<=y0<=y1 */
- setup->vmin = v2;
- setup->vmid = v0;
- setup->vmax = v1;
- }
- else {
- /* y0<=y2<=y1 */
- setup->vmin = v0;
- setup->vmid = v2;
- setup->vmax = v1;
- }
- }
- else {
- if (y0 <= y2) {
- /* y1<=y0<=y2 */
- setup->vmin = v1;
- setup->vmid = v0;
- setup->vmax = v2;
- }
- else if (y2 <= y1) {
- /* y2<=y1<=y0 */
- setup->vmin = v2;
- setup->vmid = v1;
- setup->vmax = v0;
- }
- else {
- /* y1<=y2<=y0 */
- setup->vmin = v1;
- setup->vmid = v2;
- setup->vmax = v0;
- }
- }
- }
-
- setup->ebot.dx = setup->vmid[0][0] - setup->vmin[0][0];
- setup->ebot.dy = setup->vmid[0][1] - setup->vmin[0][1];
- setup->emaj.dx = setup->vmax[0][0] - setup->vmin[0][0];
- setup->emaj.dy = setup->vmax[0][1] - setup->vmin[0][1];
- setup->etop.dx = setup->vmax[0][0] - setup->vmid[0][0];
- setup->etop.dy = setup->vmax[0][1] - setup->vmid[0][1];
-
- /*
- * Compute triangle's area. Use 1/area to compute partial
- * derivatives of attributes later.
- *
- * The area will be the same as prim->det, but the sign may be
- * different depending on how the vertices get sorted above.
- *
- * To determine whether the primitive is front or back facing we
- * use the prim->det value because its sign is correct.
- */
- {
- const float area = (setup->emaj.dx * setup->ebot.dy -
- setup->ebot.dx * setup->emaj.dy);
-
- setup->oneoverarea = 1.0f / area;
-
- /*
- debug_printf("%s one-over-area %f area %f det %f\n",
- __FUNCTION__, setup->oneoverarea, area, det );
- */
- if (util_is_inf_or_nan(setup->oneoverarea))
- return FALSE;
- }
+ struct lp_scene *scene = lp_setup_get_current_scene(setup);
- /* We need to know if this is a front or back-facing triangle for:
- * - the GLSL gl_FrontFacing fragment attribute (bool)
- * - two-sided stencil test
- */
- setup->facing =
- ((det > 0.0) ^
- (setup->llvmpipe->rasterizer->front_winding == PIPE_WINDING_CW));
+ LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
- /* Prepare pixel offset for rasterisation:
- * - pixel center (0.5, 0.5) for GL, or
- * - assume (0.0, 0.0) for other APIs.
- */
- if (setup->llvmpipe->rasterizer->gl_rasterization_rules) {
- setup->pixel_offset = 0.5f;
- } else {
- setup->pixel_offset = 0.0f;
+ if (setup->fb.cbufs[0]) {
+ if (setup->clear.flags & PIPE_CLEAR_COLOR)
+ lp_scene_bin_everywhere( scene,
+ lp_rast_clear_color,
+ setup->clear.color );
+ else
+ lp_scene_bin_everywhere( scene,
+ lp_rast_load_color,
+ lp_rast_arg_null() );
}
- return TRUE;
-}
-
+ if (setup->fb.zsbuf) {
+ if (setup->clear.flags & PIPE_CLEAR_DEPTHSTENCIL)
+ lp_scene_bin_everywhere( scene,
+ lp_rast_clear_zstencil,
+ setup->clear.zstencil );
+ else
+ lp_scene_bin_everywhere( scene,
+ lp_rast_load_zstencil,
+ lp_rast_arg_null() );
+ }
-/**
- * Compute a0, dadx and dady for a linearly interpolated coefficient,
- * for a triangle.
- */
-static void tri_pos_coeff( struct setup_context *setup,
- uint vertSlot, unsigned i)
-{
- float botda = setup->vmid[vertSlot][i] - setup->vmin[vertSlot][i];
- float majda = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i];
- float a = setup->ebot.dy * majda - botda * setup->emaj.dy;
- float b = setup->emaj.dx * botda - majda * setup->ebot.dx;
- float dadx = a * setup->oneoverarea;
- float dady = b * setup->oneoverarea;
-
- assert(i <= 3);
-
- setup->coef.dadx[0][i] = dadx;
- setup->coef.dady[0][i] = dady;
-
- /* calculate a0 as the value which would be sampled for the
- * fragment at (0,0), taking into account that we want to sample at
- * pixel centers, in other words (pixel_offset, pixel_offset).
- *
- * this is neat but unfortunately not a good way to do things for
- * triangles with very large values of dadx or dady as it will
- * result in the subtraction and re-addition from a0 of a very
- * large number, which means we'll end up loosing a lot of the
- * fractional bits and precision from a0. the way to fix this is
- * to define a0 as the sample at a pixel center somewhere near vmin
- * instead - i'll switch to this later.
- */
- setup->coef.a0[0][i] = (setup->vmin[vertSlot][i] -
- (dadx * (setup->vmin[0][0] - setup->pixel_offset) +
- dady * (setup->vmin[0][1] - setup->pixel_offset)));
-
- /*
- debug_printf("attr[%d].%c: %f dx:%f dy:%f\n",
- slot, "xyzw"[i],
- setup->coef[slot].a0[i],
- setup->coef[slot].dadx[i],
- setup->coef[slot].dady[i]);
- */
+ LP_DBG(DEBUG_SETUP, "%s done\n", __FUNCTION__);
}
-/**
- * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
- * The value value comes from vertex[slot][i].
- * The result will be put into setup->coef[slot].a0[i].
- * \param slot which attribute slot
- * \param i which component of the slot (0..3)
+/* This basically bins and then flushes any outstanding full-screen
+ * clears.
+ *
+ * TODO: fast path for fullscreen clears and no triangles.
*/
-static void const_pos_coeff( struct setup_context *setup,
- uint vertSlot, unsigned i)
+static void
+execute_clears( struct setup_context *setup )
{
- setup->coef.dadx[0][i] = 0;
- setup->coef.dady[0][i] = 0;
+ LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
- /* need provoking vertex info!
- */
- setup->coef.a0[0][i] = setup->vprovoke[vertSlot][i];
+ begin_binning( setup );
+ lp_setup_rasterize_scene( setup, TRUE );
}
-/**
- * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
- * The value value comes from vertex[slot][i].
- * The result will be put into setup->coef[slot].a0[i].
- * \param slot which attribute slot
- * \param i which component of the slot (0..3)
- */
-static void const_coeff( struct setup_context *setup,
- unsigned attrib,
- uint vertSlot)
+static void
+set_scene_state( struct setup_context *setup,
+ unsigned new_state )
{
- unsigned i;
- for (i = 0; i < NUM_CHANNELS; ++i) {
- setup->coef.dadx[1 + attrib][i] = 0;
- setup->coef.dady[1 + attrib][i] = 0;
+ unsigned old_state = setup->state;
- /* need provoking vertex info!
- */
- setup->coef.a0[1 + attrib][i] = setup->vprovoke[vertSlot][i];
- }
-}
+ if (old_state == new_state)
+ return;
+
+ LP_DBG(DEBUG_SETUP, "%s old %d new %d\n", __FUNCTION__, old_state, new_state);
+ switch (new_state) {
+ case SETUP_ACTIVE:
+ begin_binning( setup );
+ break;
-/**
- * Compute a0, dadx and dady for a linearly interpolated coefficient,
- * for a triangle.
- */
-static void tri_linear_coeff( struct setup_context *setup,
- unsigned attrib,
- uint vertSlot)
-{
- unsigned i;
- for (i = 0; i < NUM_CHANNELS; ++i) {
- float botda = setup->vmid[vertSlot][i] - setup->vmin[vertSlot][i];
- float majda = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i];
- float a = setup->ebot.dy * majda - botda * setup->emaj.dy;
- float b = setup->emaj.dx * botda - majda * setup->ebot.dx;
- float dadx = a * setup->oneoverarea;
- float dady = b * setup->oneoverarea;
-
- assert(i <= 3);
-
- setup->coef.dadx[1 + attrib][i] = dadx;
- setup->coef.dady[1 + attrib][i] = dady;
-
- /* calculate a0 as the value which would be sampled for the
- * fragment at (0,0), taking into account that we want to sample at
- * pixel centers, in other words (0.5, 0.5).
- *
- * this is neat but unfortunately not a good way to do things for
- * triangles with very large values of dadx or dady as it will
- * result in the subtraction and re-addition from a0 of a very
- * large number, which means we'll end up loosing a lot of the
- * fractional bits and precision from a0. the way to fix this is
- * to define a0 as the sample at a pixel center somewhere near vmin
- * instead - i'll switch to this later.
- */
- setup->coef.a0[1 + attrib][i] = (setup->vmin[vertSlot][i] -
- (dadx * (setup->vmin[0][0] - setup->pixel_offset) +
- dady * (setup->vmin[0][1] - setup->pixel_offset)));
-
- /*
- debug_printf("attr[%d].%c: %f dx:%f dy:%f\n",
- slot, "xyzw"[i],
- setup->coef[slot].a0[i],
- setup->coef[slot].dadx[i],
- setup->coef[slot].dady[i]);
- */
+ case SETUP_CLEARED:
+ if (old_state == SETUP_ACTIVE) {
+ assert(0);
+ return;
+ }
+ break;
+
+ case SETUP_FLUSHED:
+ if (old_state == SETUP_CLEARED)
+ execute_clears( setup );
+ else
+ lp_setup_rasterize_scene( setup, TRUE );
+ break;
}
-}
-
-/**
- * Compute a0, dadx and dady for a perspective-corrected interpolant,
- * for a triangle.
- * We basically multiply the vertex value by 1/w before computing
- * the plane coefficients (a0, dadx, dady).
- * Later, when we compute the value at a particular fragment position we'll
- * divide the interpolated value by the interpolated W at that fragment.
- */
-static void tri_persp_coeff( struct setup_context *setup,
- unsigned attrib,
- uint vertSlot)
-{
- unsigned i;
- for (i = 0; i < NUM_CHANNELS; ++i) {
- /* premultiply by 1/w (v[0][3] is always W):
- */
- float mina = setup->vmin[vertSlot][i] * setup->vmin[0][3];
- float mida = setup->vmid[vertSlot][i] * setup->vmid[0][3];
- float maxa = setup->vmax[vertSlot][i] * setup->vmax[0][3];
- float botda = mida - mina;
- float majda = maxa - mina;
- float a = setup->ebot.dy * majda - botda * setup->emaj.dy;
- float b = setup->emaj.dx * botda - majda * setup->ebot.dx;
- float dadx = a * setup->oneoverarea;
- float dady = b * setup->oneoverarea;
-
- /*
- debug_printf("tri persp %d,%d: %f %f %f\n", vertSlot, i,
- setup->vmin[vertSlot][i],
- setup->vmid[vertSlot][i],
- setup->vmax[vertSlot][i]
- );
- */
- assert(i <= 3);
-
- setup->coef.dadx[1 + attrib][i] = dadx;
- setup->coef.dady[1 + attrib][i] = dady;
- setup->coef.a0[1 + attrib][i] = (mina -
- (dadx * (setup->vmin[0][0] - setup->pixel_offset) +
- dady * (setup->vmin[0][1] - setup->pixel_offset)));
- }
+ setup->state = new_state;
}
-/**
- * Special coefficient setup for gl_FragCoord.
- * X and Y are trivial, though Y has to be inverted for OpenGL.
- * Z and W are copied from posCoef which should have already been computed.
- * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
- */
-static void
-setup_fragcoord_coeff(struct setup_context *setup, uint slot)
+void
+lp_setup_flush( struct setup_context *setup,
+ unsigned flags )
{
- /*X*/
- setup->coef.a0[1 + slot][0] = 0;
- setup->coef.dadx[1 + slot][0] = 1.0;
- setup->coef.dady[1 + slot][0] = 0.0;
- /*Y*/
- setup->coef.a0[1 + slot][1] = 0.0;
- setup->coef.dadx[1 + slot][1] = 0.0;
- setup->coef.dady[1 + slot][1] = 1.0;
- /*Z*/
- setup->coef.a0[1 + slot][2] = setup->coef.a0[0][2];
- setup->coef.dadx[1 + slot][2] = setup->coef.dadx[0][2];
- setup->coef.dady[1 + slot][2] = setup->coef.dady[0][2];
- /*W*/
- setup->coef.a0[1 + slot][3] = setup->coef.a0[0][3];
- setup->coef.dadx[1 + slot][3] = setup->coef.dadx[0][3];
- setup->coef.dady[1 + slot][3] = setup->coef.dady[0][3];
-}
+ LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
+ set_scene_state( setup, SETUP_FLUSHED );
+}
-/**
- * Compute the setup->coef[] array dadx, dady, a0 values.
- * Must be called after setup->vmin,vmid,vmax,vprovoke are initialized.
- */
-static void setup_tri_coefficients( struct setup_context *setup )
+void
+lp_setup_bind_framebuffer( struct setup_context *setup,
+ const struct pipe_framebuffer_state *fb )
{
- struct llvmpipe_context *llvmpipe = setup->llvmpipe;
- const struct lp_fragment_shader *lpfs = llvmpipe->fs;
- const struct vertex_info *vinfo = llvmpipe_get_vertex_info(llvmpipe);
- uint fragSlot;
+ struct lp_scene *scene = lp_setup_get_current_scene(setup);
- /* z and w are done by linear interpolation:
- */
- tri_pos_coeff(setup, 0, 2);
- tri_pos_coeff(setup, 0, 3);
+ LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
- /* setup interpolation for all the remaining attributes:
- */
- for (fragSlot = 0; fragSlot < lpfs->info.num_inputs; fragSlot++) {
- const uint vertSlot = vinfo->attrib[fragSlot].src_index;
+ set_scene_state( setup, SETUP_FLUSHED );
- switch (vinfo->attrib[fragSlot].interp_mode) {
- case INTERP_CONSTANT:
- const_coeff(setup, fragSlot, vertSlot);
- break;
- case INTERP_LINEAR:
- tri_linear_coeff(setup, fragSlot, vertSlot);
- break;
- case INTERP_PERSPECTIVE:
- tri_persp_coeff(setup, fragSlot, vertSlot);
- break;
- case INTERP_POS:
- setup_fragcoord_coeff(setup, fragSlot);
- break;
- default:
- assert(0);
- }
+ util_copy_framebuffer_state(&setup->fb, fb);
- if (lpfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) {
- setup->coef.a0[1 + fragSlot][0] = 1.0f - setup->facing;
- setup->coef.dadx[1 + fragSlot][0] = 0.0;
- setup->coef.dady[1 + fragSlot][0] = 0.0;
- }
- }
+ lp_scene_set_framebuffer_size(scene, setup->fb.width, setup->fb.height);
}
-
-static void setup_tri_edges( struct setup_context *setup )
+void
+lp_setup_clear( struct setup_context *setup,
+ const float *color,
+ double depth,
+ unsigned stencil,
+ unsigned flags )
{
- float vmin_x = setup->vmin[0][0] + setup->pixel_offset;
- float vmid_x = setup->vmid[0][0] + setup->pixel_offset;
-
- float vmin_y = setup->vmin[0][1] - setup->pixel_offset;
- float vmid_y = setup->vmid[0][1] - setup->pixel_offset;
- float vmax_y = setup->vmax[0][1] - setup->pixel_offset;
-
- setup->emaj.sy = ceilf(vmin_y);
- setup->emaj.lines = (int) ceilf(vmax_y - setup->emaj.sy);
- setup->emaj.dxdy = setup->emaj.dx / setup->emaj.dy;
- setup->emaj.sx = vmin_x + (setup->emaj.sy - vmin_y) * setup->emaj.dxdy;
-
- setup->etop.sy = ceilf(vmid_y);
- setup->etop.lines = (int) ceilf(vmax_y - setup->etop.sy);
- setup->etop.dxdy = setup->etop.dx / setup->etop.dy;
- setup->etop.sx = vmid_x + (setup->etop.sy - vmid_y) * setup->etop.dxdy;
-
- setup->ebot.sy = ceilf(vmin_y);
- setup->ebot.lines = (int) ceilf(vmid_y - setup->ebot.sy);
- setup->ebot.dxdy = setup->ebot.dx / setup->ebot.dy;
- setup->ebot.sx = vmin_x + (setup->ebot.sy - vmin_y) * setup->ebot.dxdy;
-}
+ struct lp_scene *scene = lp_setup_get_current_scene(setup);
+ unsigned i;
+ LP_DBG(DEBUG_SETUP, "%s state %d\n", __FUNCTION__, setup->state);
-/**
- * Render the upper or lower half of a triangle.
- * Scissoring/cliprect is applied here too.
- */
-static void subtriangle( struct setup_context *setup,
- struct edge *eleft,
- struct edge *eright,
- unsigned lines )
-{
- const struct pipe_scissor_state *cliprect = &setup->llvmpipe->cliprect;
- const int minx = (int) cliprect->minx;
- const int maxx = (int) cliprect->maxx;
- const int miny = (int) cliprect->miny;
- const int maxy = (int) cliprect->maxy;
- int y, start_y, finish_y;
- int sy = (int)eleft->sy;
-
- assert((int)eleft->sy == (int) eright->sy);
-
- /* clip top/bottom */
- start_y = sy;
- if (start_y < miny)
- start_y = miny;
-
- finish_y = sy + lines;
- if (finish_y > maxy)
- finish_y = maxy;
-
- start_y -= sy;
- finish_y -= sy;
-
- /*
- debug_printf("%s %d %d\n", __FUNCTION__, start_y, finish_y);
- */
-
- for (y = start_y; y < finish_y; y++) {
-
- /* avoid accumulating adds as floats don't have the precision to
- * accurately iterate large triangle edges that way. luckily we
- * can just multiply these days.
- *
- * this is all drowned out by the attribute interpolation anyway.
- */
- int left = (int)(eleft->sx + y * eleft->dxdy);
- int right = (int)(eright->sx + y * eright->dxdy);
-
- /* clip left/right */
- if (left < minx)
- left = minx;
- if (right > maxx)
- right = maxx;
-
- if (left < right) {
- int _y = sy + y;
- if (block(_y) != setup->span.y) {
- flush_spans(setup);
- setup->span.y = block(_y);
- }
- setup->span.left[_y&1] = left;
- setup->span.right[_y&1] = right;
- }
+ if (flags & PIPE_CLEAR_COLOR) {
+ for (i = 0; i < 4; ++i)
+ setup->clear.color.clear_color[i] = float_to_ubyte(color[i]);
}
+ if (flags & PIPE_CLEAR_DEPTHSTENCIL) {
+ setup->clear.zstencil.clear_zstencil =
+ util_pack_z_stencil(setup->fb.zsbuf->format,
+ depth,
+ stencil);
+ }
- /* save the values so that emaj can be restarted:
- */
- eleft->sx += lines * eleft->dxdy;
- eright->sx += lines * eright->dxdy;
- eleft->sy += lines;
- eright->sy += lines;
-}
+ if (setup->state == SETUP_ACTIVE) {
+ /* Add the clear to existing scene. In the unusual case where
+ * both color and depth-stencil are being cleared when there's
+ * already been some rendering, we could discard the currently
+ * binned scene and start again, but I don't see that as being
+ * a common usage.
+ */
+ if (flags & PIPE_CLEAR_COLOR)
+ lp_scene_bin_everywhere( scene,
+ lp_rast_clear_color,
+ setup->clear.color );
+ if (setup->clear.flags & PIPE_CLEAR_DEPTHSTENCIL)
+ lp_scene_bin_everywhere( scene,
+ lp_rast_clear_zstencil,
+ setup->clear.zstencil );
+ }
+ else {
+ /* Put ourselves into the 'pre-clear' state, specifically to try
+ * and accumulate multiple clears to color and depth_stencil
+ * buffers which the app or state-tracker might issue
+ * separately.
+ */
+ set_scene_state( setup, SETUP_CLEARED );
-/**
- * Recalculate prim's determinant. This is needed as we don't have
- * get this information through the vbuf_render interface & we must
- * calculate it here.
- */
-static float
-calc_det( const float (*v0)[4],
- const float (*v1)[4],
- const float (*v2)[4] )
-{
- /* edge vectors e = v0 - v2, f = v1 - v2 */
- const float ex = v0[0][0] - v2[0][0];
- const float ey = v0[0][1] - v2[0][1];
- const float fx = v1[0][0] - v2[0][0];
- const float fy = v1[0][1] - v2[0][1];
-
- /* det = cross(e,f).z */
- return ex * fy - ey * fx;
+ setup->clear.flags |= flags;
+ }
}
/**
- * Do setup for triangle rasterization, then render the triangle.
+ * Emit a fence.
*/
-void llvmpipe_setup_tri( struct setup_context *setup,
- const float (*v0)[4],
- const float (*v1)[4],
- const float (*v2)[4] )
+struct pipe_fence_handle *
+lp_setup_fence( struct setup_context *setup )
{
- float det;
+ struct lp_scene *scene = lp_setup_get_current_scene(setup);
+ const unsigned rank = lp_scene_get_num_bins( scene ); /* xxx */
+ struct lp_fence *fence = lp_fence_create(rank);
-#if DEBUG_VERTS
- debug_printf("Setup triangle:\n");
- print_vertex(setup, v0);
- print_vertex(setup, v1);
- print_vertex(setup, v2);
-#endif
+ LP_DBG(DEBUG_SETUP, "%s rank %u\n", __FUNCTION__, rank);
- if (setup->llvmpipe->no_rast)
- return;
-
- det = calc_det(v0, v1, v2);
- /*
- debug_printf("%s\n", __FUNCTION__ );
- */
+ set_scene_state( setup, SETUP_ACTIVE );
-#if DEBUG_FRAGS
- setup->numFragsEmitted = 0;
- setup->numFragsWritten = 0;
-#endif
+ /* insert the fence into all command bins */
+ lp_scene_bin_everywhere( scene,
+ lp_rast_fence,
+ lp_rast_arg_fence(fence) );
- if (cull_tri( setup, det ))
- return;
+ return (struct pipe_fence_handle *) fence;
+}
- if (!setup_sort_vertices( setup, det, v0, v1, v2 ))
- return;
- setup_tri_coefficients( setup );
- setup_tri_edges( setup );
- assert(setup->llvmpipe->reduced_prim == PIPE_PRIM_TRIANGLES);
+void
+lp_setup_set_triangle_state( struct setup_context *setup,
+ unsigned cull_mode,
+ boolean ccw_is_frontface)
+{
+ LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
- setup->span.y = 0;
- setup->span.right[0] = 0;
- setup->span.right[1] = 0;
- /* setup->span.z_mode = tri_z_mode( setup->ctx ); */
+ setup->ccw_is_frontface = ccw_is_frontface;
+ setup->cullmode = cull_mode;
+ setup->triangle = first_triangle;
+}
- /* init_constant_attribs( setup ); */
- if (setup->oneoverarea < 0.0) {
- /* emaj on left:
- */
- subtriangle( setup, &setup->emaj, &setup->ebot, setup->ebot.lines );
- subtriangle( setup, &setup->emaj, &setup->etop, setup->etop.lines );
- }
- else {
- /* emaj on right:
- */
- subtriangle( setup, &setup->ebot, &setup->emaj, setup->ebot.lines );
- subtriangle( setup, &setup->etop, &setup->emaj, setup->etop.lines );
- }
- flush_spans( setup );
+void
+lp_setup_set_fs_inputs( struct setup_context *setup,
+ const struct lp_shader_input *input,
+ unsigned nr )
+{
+ LP_DBG(DEBUG_SETUP, "%s %p %u\n", __FUNCTION__, (void *) input, nr);
-#if DEBUG_FRAGS
- printf("Tri: %u frags emitted, %u written\n",
- setup->numFragsEmitted,
- setup->numFragsWritten);
-#endif
+ memcpy( setup->fs.input, input, nr * sizeof input[0] );
+ setup->fs.nr_inputs = nr;
}
+void
+lp_setup_set_fs_function( struct setup_context *setup,
+ lp_jit_frag_func jit_function )
+{
+ LP_DBG(DEBUG_SETUP, "%s %p\n", __FUNCTION__, (void *) jit_function);
+ /* FIXME: reference count */
+ setup->fs.current.jit_function = jit_function;
+ setup->dirty |= LP_SETUP_NEW_FS;
+}
-/**
- * Compute a0, dadx and dady for a linearly interpolated coefficient,
- * for a line.
- */
-static void
-linear_pos_coeff(struct setup_context *setup,
- uint vertSlot, uint i)
+void
+lp_setup_set_fs_constants(struct setup_context *setup,
+ struct pipe_buffer *buffer)
{
- const float da = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i];
- const float dadx = da * setup->emaj.dx * setup->oneoverarea;
- const float dady = da * setup->emaj.dy * setup->oneoverarea;
- setup->coef.dadx[0][i] = dadx;
- setup->coef.dady[0][i] = dady;
- setup->coef.a0[0][i] = (setup->vmin[vertSlot][i] -
- (dadx * (setup->vmin[0][0] - setup->pixel_offset) +
- dady * (setup->vmin[0][1] - setup->pixel_offset)));
-}
+ LP_DBG(DEBUG_SETUP, "%s %p\n", __FUNCTION__, (void *) buffer);
+ pipe_buffer_reference(&setup->constants.current, buffer);
-/**
- * Compute a0, dadx and dady for a linearly interpolated coefficient,
- * for a line.
- */
-static void
-line_linear_coeff(struct setup_context *setup,
- unsigned attrib,
- uint vertSlot)
-{
- unsigned i;
- for (i = 0; i < NUM_CHANNELS; ++i) {
- const float da = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i];
- const float dadx = da * setup->emaj.dx * setup->oneoverarea;
- const float dady = da * setup->emaj.dy * setup->oneoverarea;
- setup->coef.dadx[1 + attrib][i] = dadx;
- setup->coef.dady[1 + attrib][i] = dady;
- setup->coef.a0[1 + attrib][i] = (setup->vmin[vertSlot][i] -
- (dadx * (setup->vmin[0][0] - setup->pixel_offset) +
- dady * (setup->vmin[0][1] - setup->pixel_offset)));
- }
+ setup->dirty |= LP_SETUP_NEW_CONSTANTS;
}
-/**
- * Compute a0, dadx and dady for a perspective-corrected interpolant,
- * for a line.
- */
-static void
-line_persp_coeff(struct setup_context *setup,
- unsigned attrib,
- uint vertSlot)
+void
+lp_setup_set_alpha_ref_value( struct setup_context *setup,
+ float alpha_ref_value )
{
- unsigned i;
- for (i = 0; i < NUM_CHANNELS; ++i) {
- /* XXX double-check/verify this arithmetic */
- const float a0 = setup->vmin[vertSlot][i] * setup->vmin[0][3];
- const float a1 = setup->vmax[vertSlot][i] * setup->vmax[0][3];
- const float da = a1 - a0;
- const float dadx = da * setup->emaj.dx * setup->oneoverarea;
- const float dady = da * setup->emaj.dy * setup->oneoverarea;
- setup->coef.dadx[1 + attrib][i] = dadx;
- setup->coef.dady[1 + attrib][i] = dady;
- setup->coef.a0[1 + attrib][i] = (setup->vmin[vertSlot][i] -
- (dadx * (setup->vmin[0][0] - setup->pixel_offset) +
- dady * (setup->vmin[0][1] - setup->pixel_offset)));
+ LP_DBG(DEBUG_SETUP, "%s %f\n", __FUNCTION__, alpha_ref_value);
+
+ if(setup->fs.current.jit_context.alpha_ref_value != alpha_ref_value) {
+ setup->fs.current.jit_context.alpha_ref_value = alpha_ref_value;
+ setup->dirty |= LP_SETUP_NEW_FS;
}
}
-
-/**
- * Compute the setup->coef[] array dadx, dady, a0 values.
- * Must be called after setup->vmin,vmax are initialized.
- */
-static INLINE boolean
-setup_line_coefficients(struct setup_context *setup,
- const float (*v0)[4],
- const float (*v1)[4])
+void
+lp_setup_set_blend_color( struct setup_context *setup,
+ const struct pipe_blend_color *blend_color )
{
- struct llvmpipe_context *llvmpipe = setup->llvmpipe;
- const struct lp_fragment_shader *lpfs = llvmpipe->fs;
- const struct vertex_info *vinfo = llvmpipe_get_vertex_info(llvmpipe);
- uint fragSlot;
- float area;
-
- /* use setup->vmin, vmax to point to vertices */
- if (llvmpipe->rasterizer->flatshade_first)
- setup->vprovoke = v0;
- else
- setup->vprovoke = v1;
- setup->vmin = v0;
- setup->vmax = v1;
-
- setup->emaj.dx = setup->vmax[0][0] - setup->vmin[0][0];
- setup->emaj.dy = setup->vmax[0][1] - setup->vmin[0][1];
-
- /* NOTE: this is not really area but something proportional to it */
- area = setup->emaj.dx * setup->emaj.dx + setup->emaj.dy * setup->emaj.dy;
- if (area == 0.0f || util_is_inf_or_nan(area))
- return FALSE;
- setup->oneoverarea = 1.0f / area;
-
- /* z and w are done by linear interpolation:
- */
- linear_pos_coeff(setup, 0, 2);
- linear_pos_coeff(setup, 0, 3);
+ LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
- /* setup interpolation for all the remaining attributes:
- */
- for (fragSlot = 0; fragSlot < lpfs->info.num_inputs; fragSlot++) {
- const uint vertSlot = vinfo->attrib[fragSlot].src_index;
+ assert(blend_color);
- switch (vinfo->attrib[fragSlot].interp_mode) {
- case INTERP_CONSTANT:
- const_coeff(setup, fragSlot, vertSlot);
- break;
- case INTERP_LINEAR:
- line_linear_coeff(setup, fragSlot, vertSlot);
- break;
- case INTERP_PERSPECTIVE:
- line_persp_coeff(setup, fragSlot, vertSlot);
- break;
- case INTERP_POS:
- setup_fragcoord_coeff(setup, fragSlot);
- break;
- default:
- assert(0);
- }
-
- if (lpfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) {
- setup->coef.a0[1 + fragSlot][0] = 1.0f - setup->facing;
- setup->coef.dadx[1 + fragSlot][0] = 0.0;
- setup->coef.dady[1 + fragSlot][0] = 0.0;
- }
+ if(memcmp(&setup->blend_color.current, blend_color, sizeof *blend_color) != 0) {
+ memcpy(&setup->blend_color.current, blend_color, sizeof *blend_color);
+ setup->dirty |= LP_SETUP_NEW_BLEND_COLOR;
}
- return TRUE;
}
-/**
- * Plot a pixel in a line segment.
- */
-static INLINE void
-plot(struct setup_context *setup, int x, int y)
+void
+lp_setup_set_flatshade_first( struct setup_context *setup,
+ boolean flatshade_first )
{
- const int iy = y & 1;
- const int ix = x & 1;
- const int quadX = x - ix;
- const int quadY = y - iy;
- const int mask = (1 << ix) << (2 * iy);
-
- if (quadX != setup->quad[0].input.x0 ||
- quadY != setup->quad[0].input.y0)
- {
- /* flush prev quad, start new quad */
-
- if (setup->quad[0].input.x0 != -1)
- clip_emit_quad( setup, &setup->quad[0] );
-
- setup->quad[0].input.x0 = quadX;
- setup->quad[0].input.y0 = quadY;
- setup->quad[0].inout.mask = 0x0;
- }
-
- setup->quad[0].inout.mask |= mask;
+ setup->flatshade_first = flatshade_first;
}
-/**
- * Do setup for line rasterization, then render the line.
- * Single-pixel width, no stipple, etc. We rely on the 'draw' module
- * to handle stippling and wide lines.
- */
-void
-llvmpipe_setup_line(struct setup_context *setup,
- const float (*v0)[4],
- const float (*v1)[4])
+void
+lp_setup_set_vertex_info( struct setup_context *setup,
+ struct vertex_info *vertex_info )
{
- int x0 = (int) v0[0][0];
- int x1 = (int) v1[0][0];
- int y0 = (int) v0[0][1];
- int y1 = (int) v1[0][1];
- int dx = x1 - x0;
- int dy = y1 - y0;
- int xstep, ystep;
-
-#if DEBUG_VERTS
- debug_printf("Setup line:\n");
- print_vertex(setup, v0);
- print_vertex(setup, v1);
-#endif
-
- if (setup->llvmpipe->no_rast)
- return;
+ /* XXX: just silently holding onto the pointer:
+ */
+ setup->vertex_info = vertex_info;
+}
- if (dx == 0 && dy == 0)
- return;
- if (!setup_line_coefficients(setup, v0, v1))
- return;
+void
+lp_setup_set_sampler_textures( struct setup_context *setup,
+ unsigned num, struct pipe_texture **texture)
+{
+ struct pipe_texture *dummy;
+ unsigned i;
- assert(v0[0][0] < 1.0e9);
- assert(v0[0][1] < 1.0e9);
- assert(v1[0][0] < 1.0e9);
- assert(v1[0][1] < 1.0e9);
+ LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
- if (dx < 0) {
- dx = -dx; /* make positive */
- xstep = -1;
- }
- else {
- xstep = 1;
- }
- if (dy < 0) {
- dy = -dy; /* make positive */
- ystep = -1;
- }
- else {
- ystep = 1;
- }
+ assert(num <= PIPE_MAX_SAMPLERS);
- assert(dx >= 0);
- assert(dy >= 0);
- assert(setup->llvmpipe->reduced_prim == PIPE_PRIM_LINES);
+ for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+ struct pipe_texture *tex = i < num ? texture[i] : NULL;
- setup->quad[0].input.x0 = setup->quad[0].input.y0 = -1;
- setup->quad[0].inout.mask = 0x0;
+ /* FIXME: hold on to the reference */
+ dummy = NULL;
+ pipe_texture_reference(&dummy, tex);
- /* XXX temporary: set coverage to 1.0 so the line appears
- * if AA mode happens to be enabled.
- */
- setup->quad[0].input.coverage[0] =
- setup->quad[0].input.coverage[1] =
- setup->quad[0].input.coverage[2] =
- setup->quad[0].input.coverage[3] = 1.0;
-
- if (dx > dy) {
- /*** X-major line ***/
- int i;
- const int errorInc = dy + dy;
- int error = errorInc - dx;
- const int errorDec = error - dx;
-
- for (i = 0; i < dx; i++) {
- plot(setup, x0, y0);
-
- x0 += xstep;
- if (error < 0) {
- error += errorInc;
- }
- else {
- error += errorDec;
- y0 += ystep;
- }
- }
- }
- else {
- /*** Y-major line ***/
- int i;
- const int errorInc = dx + dx;
- int error = errorInc - dy;
- const int errorDec = error - dy;
-
- for (i = 0; i < dy; i++) {
- plot(setup, x0, y0);
-
- y0 += ystep;
- if (error < 0) {
- error += errorInc;
- }
- else {
- error += errorDec;
- x0 += xstep;
- }
+ if(tex) {
+ struct llvmpipe_texture *lp_tex = llvmpipe_texture(tex);
+ struct lp_jit_texture *jit_tex;
+ jit_tex = &setup->fs.current.jit_context.textures[i];
+ jit_tex->width = tex->width0;
+ jit_tex->height = tex->height0;
+ jit_tex->stride = lp_tex->stride[0];
+ if(!lp_tex->dt)
+ jit_tex->data = lp_tex->data;
+ else
+ /* FIXME: map the rendertarget */
+ assert(0);
}
}
- /* draw final quad */
- if (setup->quad[0].inout.mask) {
- clip_emit_quad( setup, &setup->quad[0] );
- }
+ setup->dirty |= LP_SETUP_NEW_FS;
}
-
-static void
-point_persp_coeff(struct setup_context *setup,
- const float (*vert)[4],
- unsigned attrib,
- uint vertSlot)
+boolean
+lp_setup_is_texture_referenced( struct setup_context *setup,
+ const struct pipe_texture *texture )
{
- unsigned i;
- for(i = 0; i < NUM_CHANNELS; ++i) {
- setup->coef.dadx[1 + attrib][i] = 0.0F;
- setup->coef.dady[1 + attrib][i] = 0.0F;
- setup->coef.a0[1 + attrib][i] = vert[vertSlot][i] * vert[0][3];
- }
+ /* FIXME */
+ return PIPE_UNREFERENCED;
}
-/**
- * Do setup for point rasterization, then render the point.
- * Round or square points...
- * XXX could optimize a lot for 1-pixel points.
- */
void
-llvmpipe_setup_point( struct setup_context *setup,
- const float (*v0)[4] )
+lp_setup_update_state( struct setup_context *setup )
{
- struct llvmpipe_context *llvmpipe = setup->llvmpipe;
- const struct lp_fragment_shader *lpfs = llvmpipe->fs;
- const int sizeAttr = setup->llvmpipe->psize_slot;
- const float size
- = sizeAttr > 0 ? v0[sizeAttr][0]
- : setup->llvmpipe->rasterizer->point_size;
- const float halfSize = 0.5F * size;
- const boolean round = (boolean) setup->llvmpipe->rasterizer->point_smooth;
- const float x = v0[0][0]; /* Note: data[0] is always position */
- const float y = v0[0][1];
- const struct vertex_info *vinfo = llvmpipe_get_vertex_info(llvmpipe);
- uint fragSlot;
-
-#if DEBUG_VERTS
- debug_printf("Setup point:\n");
- print_vertex(setup, v0);
-#endif
-
- if (llvmpipe->no_rast)
- return;
+ struct lp_scene *scene = lp_setup_get_current_scene(setup);
- assert(setup->llvmpipe->reduced_prim == PIPE_PRIM_POINTS);
-
- /* For points, all interpolants are constant-valued.
- * However, for point sprites, we'll need to setup texcoords appropriately.
- * XXX: which coefficients are the texcoords???
- * We may do point sprites as textured quads...
- *
- * KW: We don't know which coefficients are texcoords - ultimately
- * the choice of what interpolation mode to use for each attribute
- * should be determined by the fragment program, using
- * per-attribute declaration statements that include interpolation
- * mode as a parameter. So either the fragment program will have
- * to be adjusted for pointsprite vs normal point behaviour, or
- * otherwise a special interpolation mode will have to be defined
- * which matches the required behaviour for point sprites. But -
- * the latter is not a feature of normal hardware, and as such
- * probably should be ruled out on that basis.
- */
- setup->vprovoke = v0;
+ LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
- /* setup Z, W */
- const_pos_coeff(setup, 0, 2);
- const_pos_coeff(setup, 0, 3);
+ assert(setup->fs.current.jit_function);
- for (fragSlot = 0; fragSlot < lpfs->info.num_inputs; fragSlot++) {
- const uint vertSlot = vinfo->attrib[fragSlot].src_index;
+ if(setup->dirty & LP_SETUP_NEW_BLEND_COLOR) {
+ uint8_t *stored;
+ unsigned i, j;
- switch (vinfo->attrib[fragSlot].interp_mode) {
- case INTERP_CONSTANT:
- /* fall-through */
- case INTERP_LINEAR:
- const_coeff(setup, fragSlot, vertSlot);
- break;
- case INTERP_PERSPECTIVE:
- point_persp_coeff(setup, setup->vprovoke, fragSlot, vertSlot);
- break;
- case INTERP_POS:
- setup_fragcoord_coeff(setup, fragSlot);
- break;
- default:
- assert(0);
- }
+ stored = lp_scene_alloc_aligned(scene, 4 * 16, 16);
- if (lpfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) {
- setup->coef.a0[1 + fragSlot][0] = 1.0f - setup->facing;
- setup->coef.dadx[1 + fragSlot][0] = 0.0;
- setup->coef.dady[1 + fragSlot][0] = 0.0;
+ /* smear each blend color component across 16 ubyte elements */
+ for (i = 0; i < 4; ++i) {
+ uint8_t c = float_to_ubyte(setup->blend_color.current.color[i]);
+ for (j = 0; j < 16; ++j)
+ stored[i*16 + j] = c;
}
- }
+ setup->blend_color.stored = stored;
- if (halfSize <= 0.5 && !round) {
- /* special case for 1-pixel points */
- const int ix = ((int) x) & 1;
- const int iy = ((int) y) & 1;
- setup->quad[0].input.x0 = (int) x - ix;
- setup->quad[0].input.y0 = (int) y - iy;
- setup->quad[0].inout.mask = (1 << ix) << (2 * iy);
- clip_emit_quad( setup, &setup->quad[0] );
+ setup->fs.current.jit_context.blend_color = setup->blend_color.stored;
+ setup->dirty |= LP_SETUP_NEW_FS;
}
- else {
- if (round) {
- /* rounded points */
- const int ixmin = block((int) (x - halfSize));
- const int ixmax = block((int) (x + halfSize));
- const int iymin = block((int) (y - halfSize));
- const int iymax = block((int) (y + halfSize));
- const float rmin = halfSize - 0.7071F; /* 0.7071 = sqrt(2)/2 */
- const float rmax = halfSize + 0.7071F;
- const float rmin2 = MAX2(0.0F, rmin * rmin);
- const float rmax2 = rmax * rmax;
- const float cscale = 1.0F / (rmax2 - rmin2);
- int ix, iy;
-
- for (iy = iymin; iy <= iymax; iy += 2) {
- for (ix = ixmin; ix <= ixmax; ix += 2) {
- float dx, dy, dist2, cover;
-
- setup->quad[0].inout.mask = 0x0;
-
- dx = (ix + 0.5f) - x;
- dy = (iy + 0.5f) - y;
- dist2 = dx * dx + dy * dy;
- if (dist2 <= rmax2) {
- cover = 1.0F - (dist2 - rmin2) * cscale;
- setup->quad[0].input.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f);
- setup->quad[0].inout.mask |= MASK_TOP_LEFT;
- }
-
- dx = (ix + 1.5f) - x;
- dy = (iy + 0.5f) - y;
- dist2 = dx * dx + dy * dy;
- if (dist2 <= rmax2) {
- cover = 1.0F - (dist2 - rmin2) * cscale;
- setup->quad[0].input.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f);
- setup->quad[0].inout.mask |= MASK_TOP_RIGHT;
- }
-
- dx = (ix + 0.5f) - x;
- dy = (iy + 1.5f) - y;
- dist2 = dx * dx + dy * dy;
- if (dist2 <= rmax2) {
- cover = 1.0F - (dist2 - rmin2) * cscale;
- setup->quad[0].input.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f);
- setup->quad[0].inout.mask |= MASK_BOTTOM_LEFT;
- }
-
- dx = (ix + 1.5f) - x;
- dy = (iy + 1.5f) - y;
- dist2 = dx * dx + dy * dy;
- if (dist2 <= rmax2) {
- cover = 1.0F - (dist2 - rmin2) * cscale;
- setup->quad[0].input.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f);
- setup->quad[0].inout.mask |= MASK_BOTTOM_RIGHT;
- }
-
- if (setup->quad[0].inout.mask) {
- setup->quad[0].input.x0 = ix;
- setup->quad[0].input.y0 = iy;
- clip_emit_quad( setup, &setup->quad[0] );
- }
+
+
+ if(setup->dirty & LP_SETUP_NEW_CONSTANTS) {
+ struct pipe_buffer *buffer = setup->constants.current;
+
+ if(buffer) {
+ unsigned current_size = buffer->size;
+ const void *current_data = llvmpipe_buffer(buffer)->data;
+
+ /* TODO: copy only the actually used constants? */
+
+ if(setup->constants.stored_size != current_size ||
+ !setup->constants.stored_data ||
+ memcmp(setup->constants.stored_data,
+ current_data,
+ current_size) != 0) {
+ void *stored;
+
+ stored = lp_scene_alloc(scene, current_size);
+ if(stored) {
+ memcpy(stored,
+ current_data,
+ current_size);
+ setup->constants.stored_size = current_size;
+ setup->constants.stored_data = stored;
}
}
}
else {
- /* square points */
- const int xmin = (int) (x + 0.75 - halfSize);
- const int ymin = (int) (y + 0.25 - halfSize);
- const int xmax = xmin + (int) size;
- const int ymax = ymin + (int) size;
- /* XXX could apply scissor to xmin,ymin,xmax,ymax now */
- const int ixmin = block(xmin);
- const int ixmax = block(xmax - 1);
- const int iymin = block(ymin);
- const int iymax = block(ymax - 1);
- int ix, iy;
-
- /*
- debug_printf("(%f, %f) -> X:%d..%d Y:%d..%d\n", x, y, xmin, xmax,ymin,ymax);
- */
- for (iy = iymin; iy <= iymax; iy += 2) {
- uint rowMask = 0xf;
- if (iy < ymin) {
- /* above the top edge */
- rowMask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT);
- }
- if (iy + 1 >= ymax) {
- /* below the bottom edge */
- rowMask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT);
- }
+ setup->constants.stored_size = 0;
+ setup->constants.stored_data = NULL;
+ }
- for (ix = ixmin; ix <= ixmax; ix += 2) {
- uint mask = rowMask;
-
- if (ix < xmin) {
- /* fragment is past left edge of point, turn off left bits */
- mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT);
- }
- if (ix + 1 >= xmax) {
- /* past the right edge */
- mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
- }
-
- setup->quad[0].inout.mask = mask;
- setup->quad[0].input.x0 = ix;
- setup->quad[0].input.y0 = iy;
- clip_emit_quad( setup, &setup->quad[0] );
- }
+ setup->fs.current.jit_context.constants = setup->constants.stored_data;
+ setup->dirty |= LP_SETUP_NEW_FS;
+ }
+
+
+ if(setup->dirty & LP_SETUP_NEW_FS) {
+ if(!setup->fs.stored ||
+ memcmp(setup->fs.stored,
+ &setup->fs.current,
+ sizeof setup->fs.current) != 0) {
+ /* The fs state that's been stored in the scene is different from
+ * the new, current state. So allocate a new lp_rast_state object
+ * and append it to the bin's setup data buffer.
+ */
+ struct lp_rast_state *stored =
+ (struct lp_rast_state *) lp_scene_alloc(scene, sizeof *stored);
+ if(stored) {
+ memcpy(stored,
+ &setup->fs.current,
+ sizeof setup->fs.current);
+ setup->fs.stored = stored;
+
+ /* put the state-set command into all bins */
+ lp_scene_bin_state_command( scene,
+ lp_rast_set_state,
+ lp_rast_arg_state(setup->fs.stored) );
}
}
}
+
+ setup->dirty = 0;
+
+ assert(setup->fs.stored);
}
-void llvmpipe_setup_prepare( struct setup_context *setup )
+
+
+/* Only caller is lp_setup_vbuf_destroy()
+ */
+void
+lp_setup_destroy( struct setup_context *setup )
{
- struct llvmpipe_context *lp = setup->llvmpipe;
+ reset_context( setup );
- if (lp->dirty) {
- llvmpipe_update_derived(lp);
- }
+ pipe_buffer_reference(&setup->constants.current, NULL);
- 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;
+ /* free the scenes in the 'empty' queue */
+ while (lp_scene_queue_count(setup->empty_scenes) > 0) {
+ struct lp_scene *scene = lp_scene_dequeue(setup->empty_scenes);
+ if (!scene)
+ break;
+ lp_scene_destroy(scene);
}
-}
-
+ lp_rast_destroy( setup->rast );
-void llvmpipe_setup_destroy_context( struct setup_context *setup )
-{
- align_free( setup );
+ FREE( setup );
}
/**
- * Create a new primitive setup/render stage.
+ * Create a new primitive tiling engine. Plug it into the backend of
+ * the draw module. Currently also creates a rasterizer to use with
+ * it.
*/
-struct setup_context *llvmpipe_setup_create_context( struct llvmpipe_context *llvmpipe )
+struct setup_context *
+lp_setup_create( struct pipe_screen *screen,
+ struct draw_context *draw )
{
- struct setup_context *setup;
unsigned i;
+ struct setup_context *setup = CALLOC_STRUCT(setup_context);
- setup = align_malloc(sizeof(struct setup_context), 16);
if (!setup)
return NULL;
- memset(setup, 0, sizeof *setup);
- setup->llvmpipe = llvmpipe;
+ lp_setup_init_vbuf(setup);
+
+ setup->empty_scenes = lp_scene_queue_create();
+ if (!setup->empty_scenes)
+ goto fail;
- for (i = 0; i < MAX_QUADS; i++) {
- setup->quad[i].coef = &setup->coef;
+ setup->rast = lp_rast_create( screen, setup->empty_scenes );
+ if (!setup->rast)
+ goto fail;
+
+ setup->vbuf = draw_vbuf_stage(draw, &setup->base);
+ if (!setup->vbuf)
+ goto fail;
+
+ draw_set_rasterize_stage(draw, setup->vbuf);
+ draw_set_render(draw, &setup->base);
+
+ /* create some empty scenes */
+ for (i = 0; i < MAX_SCENES; i++) {
+ struct lp_scene *scene = lp_scene_create();
+ lp_scene_enqueue(setup->empty_scenes, scene);
}
- setup->span.left[0] = 1000000; /* greater than right[0] */
- setup->span.left[1] = 1000000; /* greater than right[1] */
+ setup->triangle = first_triangle;
+ setup->line = first_line;
+ setup->point = first_point;
+
+ setup->dirty = ~0;
return setup;
+
+fail:
+ if (setup->rast)
+ lp_rast_destroy( setup->rast );
+
+ if (setup->vbuf)
+ ;
+
+ if (setup->empty_scenes)
+ lp_scene_queue_destroy(setup->empty_scenes);
+
+ FREE(setup);
+ return NULL;
}
projects / mesa.git / blobdiff