X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fllvmpipe%2Flp_setup_tri.c;h=4e2e17f77b1eb03b4a4a43c444f07fc8badb46bb;hb=392a2515c0967c395be098cac6a37f325dd66b90;hp=018d254c765f67eac3e9d6e3f544575688c6e97c;hpb=ba5d600c90f7075cf2c33a0c5c679ef822e8746e;p=mesa.git diff --git a/src/gallium/drivers/llvmpipe/lp_setup_tri.c b/src/gallium/drivers/llvmpipe/lp_setup_tri.c index 018d254c765..4e2e17f77b1 100644 --- a/src/gallium/drivers/llvmpipe/lp_setup_tri.c +++ b/src/gallium/drivers/llvmpipe/lp_setup_tri.c @@ -29,10 +29,12 @@ * Binning code for triangles */ -#include "lp_setup_context.h" -#include "lp_rast.h" #include "util/u_math.h" #include "util/u_memory.h" +#include "lp_perf.h" +#include "lp_setup_context.h" +#include "lp_rast.h" +#include "lp_state_fs.h" #define NUM_CHANNELS 4 @@ -40,7 +42,8 @@ /** * Compute a0 for a constant-valued coefficient (GL_FLAT shading). */ -static void constant_coef( struct lp_rast_triangle *tri, +static void constant_coef( struct lp_setup_context *setup, + struct lp_rast_triangle *tri, unsigned slot, const float value, unsigned i ) @@ -55,7 +58,8 @@ static void constant_coef( struct lp_rast_triangle *tri, * Compute a0, dadx and dady for a linearly interpolated coefficient, * for a triangle. */ -static void linear_coef( struct lp_rast_triangle *tri, +static void linear_coef( struct lp_setup_context *setup, + struct lp_rast_triangle *tri, float oneoverarea, unsigned slot, const float (*v1)[4], @@ -88,9 +92,9 @@ static void linear_coef( struct lp_rast_triangle *tri, * to define a0 as the sample at a pixel center somewhere near vmin * instead - i'll switch to this later. */ - tri->inputs.a0[slot][i] = (v1[vert_attr][i] - - (dadx * (v1[0][0] - 0.5f) + - dady * (v1[0][1] - 0.5f))); + tri->inputs.a0[slot][i] = (a1 - + (dadx * (v1[0][0] - setup->pixel_offset) + + dady * (v1[0][1] - setup->pixel_offset))); } @@ -102,7 +106,8 @@ static void linear_coef( struct lp_rast_triangle *tri, * 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 perspective_coef( struct lp_rast_triangle *tri, +static void perspective_coef( struct lp_setup_context *setup, + struct lp_rast_triangle *tri, float oneoverarea, unsigned slot, const float (*v1)[4], @@ -124,8 +129,8 @@ static void perspective_coef( struct lp_rast_triangle *tri, tri->inputs.dadx[slot][i] = dadx; tri->inputs.dady[slot][i] = dady; tri->inputs.a0[slot][i] = (a1 - - (dadx * (v1[0][0] - 0.5f) + - dady * (v1[0][1] - 0.5f))); + (dadx * (v1[0][0] - setup->pixel_offset) + + dady * (v1[0][1] - setup->pixel_offset))); } @@ -136,43 +141,70 @@ static void perspective_coef( struct lp_rast_triangle *tri, * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask. */ static void -setup_fragcoord_coef(struct lp_rast_triangle *tri, +setup_fragcoord_coef(struct lp_setup_context *setup, + struct lp_rast_triangle *tri, float oneoverarea, unsigned slot, const float (*v1)[4], const float (*v2)[4], - const float (*v3)[4]) + const float (*v3)[4], + unsigned usage_mask) { /*X*/ - tri->inputs.a0[slot][0] = 0.0; - tri->inputs.dadx[slot][0] = 1.0; - tri->inputs.dady[slot][0] = 0.0; + if (usage_mask & TGSI_WRITEMASK_X) { + tri->inputs.a0[slot][0] = 0.0; + tri->inputs.dadx[slot][0] = 1.0; + tri->inputs.dady[slot][0] = 0.0; + } + /*Y*/ - tri->inputs.a0[slot][1] = 0.0; - tri->inputs.dadx[slot][1] = 0.0; - tri->inputs.dady[slot][1] = 1.0; + if (usage_mask & TGSI_WRITEMASK_Y) { + tri->inputs.a0[slot][1] = 0.0; + tri->inputs.dadx[slot][1] = 0.0; + tri->inputs.dady[slot][1] = 1.0; + } + /*Z*/ - linear_coef(tri, oneoverarea, slot, v1, v2, v3, 0, 2); + if (usage_mask & TGSI_WRITEMASK_Z) { + linear_coef(setup, tri, oneoverarea, slot, v1, v2, v3, 0, 2); + } + /*W*/ - linear_coef(tri, oneoverarea, slot, v1, v2, v3, 0, 3); + if (usage_mask & TGSI_WRITEMASK_W) { + linear_coef(setup, tri, oneoverarea, slot, v1, v2, v3, 0, 3); + } } -static void setup_facing_coef( struct lp_rast_triangle *tri, +/** + * Setup the fragment input attribute with the front-facing value. + * \param frontface is the triangle front facing? + */ +static void setup_facing_coef( struct lp_setup_context *setup, + struct lp_rast_triangle *tri, unsigned slot, - boolean frontface ) + boolean frontface, + unsigned usage_mask) { - constant_coef( tri, slot, 1.0f - frontface, 0 ); - constant_coef( tri, slot, 0.0f, 1 ); /* wasted */ - constant_coef( tri, slot, 0.0f, 2 ); /* wasted */ - constant_coef( tri, slot, 0.0f, 3 ); /* wasted */ + /* convert TRUE to 1.0 and FALSE to -1.0 */ + if (usage_mask & TGSI_WRITEMASK_X) + constant_coef( setup, tri, slot, 2.0f * frontface - 1.0f, 0 ); + + if (usage_mask & TGSI_WRITEMASK_Y) + constant_coef( setup, tri, slot, 0.0f, 1 ); /* wasted */ + + if (usage_mask & TGSI_WRITEMASK_Z) + constant_coef( setup, tri, slot, 0.0f, 2 ); /* wasted */ + + if (usage_mask & TGSI_WRITEMASK_W) + constant_coef( setup, tri, slot, 0.0f, 3 ); /* wasted */ } /** * Compute the tri->coef[] array dadx, dady, a0 values. */ -static void setup_tri_coefficients( struct setup_context *setup, +static void setup_tri_coefficients( struct lp_setup_context *setup, struct lp_rast_triangle *tri, float oneoverarea, const float (*v1)[4], @@ -180,92 +212,180 @@ static void setup_tri_coefficients( struct setup_context *setup, const float (*v3)[4], boolean frontface) { - struct lp_scene *scene = lp_setup_get_current_scene(setup); + unsigned fragcoord_usage_mask = TGSI_WRITEMASK_XYZ; unsigned slot; - /* Allocate space for the a0, dadx and dady arrays - */ - { - unsigned bytes = (setup->fs.nr_inputs + 1) * 4 * sizeof(float); - tri->inputs.a0 = lp_scene_alloc_aligned( scene, bytes, 16 ); - tri->inputs.dadx = lp_scene_alloc_aligned( scene, bytes, 16 ); - tri->inputs.dady = lp_scene_alloc_aligned( scene, bytes, 16 ); - } - - /* The internal position input is in slot zero: - */ - setup_fragcoord_coef(tri, oneoverarea, 0, v1, v2, v3); - /* setup interpolation for all the remaining attributes: */ for (slot = 0; slot < setup->fs.nr_inputs; slot++) { unsigned vert_attr = setup->fs.input[slot].src_index; + unsigned usage_mask = setup->fs.input[slot].usage_mask; unsigned i; switch (setup->fs.input[slot].interp) { case LP_INTERP_CONSTANT: - for (i = 0; i < NUM_CHANNELS; i++) - constant_coef(tri, slot+1, v3[vert_attr][i], i); + if (setup->flatshade_first) { + for (i = 0; i < NUM_CHANNELS; i++) + if (usage_mask & (1 << i)) + constant_coef(setup, tri, slot+1, v1[vert_attr][i], i); + } + else { + for (i = 0; i < NUM_CHANNELS; i++) + if (usage_mask & (1 << i)) + constant_coef(setup, tri, slot+1, v3[vert_attr][i], i); + } break; case LP_INTERP_LINEAR: for (i = 0; i < NUM_CHANNELS; i++) - linear_coef(tri, oneoverarea, slot+1, v1, v2, v3, vert_attr, i); + if (usage_mask & (1 << i)) + linear_coef(setup, tri, oneoverarea, slot+1, v1, v2, v3, vert_attr, i); break; case LP_INTERP_PERSPECTIVE: for (i = 0; i < NUM_CHANNELS; i++) - perspective_coef(tri, oneoverarea, slot+1, v1, v2, v3, vert_attr, i); + if (usage_mask & (1 << i)) + perspective_coef(setup, tri, oneoverarea, slot+1, v1, v2, v3, vert_attr, i); + fragcoord_usage_mask |= TGSI_WRITEMASK_W; break; case LP_INTERP_POSITION: - /* XXX: fix me - duplicates the values in slot zero. + /* + * The generated pixel interpolators will pick up the coeffs from + * slot 0, so all need to ensure that the usage mask is covers all + * usages. */ - setup_fragcoord_coef(tri, oneoverarea, slot+1, v1, v2, v3); + fragcoord_usage_mask |= usage_mask; break; case LP_INTERP_FACING: - setup_facing_coef(tri, slot+1, frontface); + setup_facing_coef(setup, tri, slot+1, frontface, usage_mask); break; default: assert(0); } } + + /* The internal position input is in slot zero: + */ + setup_fragcoord_coef(setup, tri, oneoverarea, 0, v1, v2, v3, + fragcoord_usage_mask); } -static inline int subpixel_snap( float a ) +static INLINE int subpixel_snap( float a ) { return util_iround(FIXED_ONE * a - (FIXED_ONE / 2)); } + +/** + * Alloc space for a new triangle plus the input.a0/dadx/dady arrays + * immediately after it. + * The memory is allocated from the per-scene pool, not per-tile. + * \param tri_size returns number of bytes allocated + * \param nr_inputs number of fragment shader inputs + * \return pointer to triangle space + */ +static INLINE struct lp_rast_triangle * +alloc_triangle(struct lp_scene *scene, unsigned nr_inputs, unsigned *tri_size) +{ + unsigned input_array_sz = NUM_CHANNELS * (nr_inputs + 1) * sizeof(float); + struct lp_rast_triangle *tri; + unsigned bytes; + char *inputs; + + assert(sizeof(*tri) % 16 == 0); + + bytes = sizeof(*tri) + (3 * input_array_sz); + + tri = lp_scene_alloc_aligned( scene, bytes, 16 ); + + if (tri) { + inputs = (char *) (tri + 1); + tri->inputs.a0 = (float (*)[4]) inputs; + tri->inputs.dadx = (float (*)[4]) (inputs + input_array_sz); + tri->inputs.dady = (float (*)[4]) (inputs + 2 * input_array_sz); + + *tri_size = bytes; + } + + return tri; +} + + +/** + * Print triangle vertex attribs (for debug). + */ +static void +print_triangle(struct lp_setup_context *setup, + const float (*v1)[4], + const float (*v2)[4], + const float (*v3)[4]) +{ + uint i; + + debug_printf("llvmpipe triangle\n"); + for (i = 0; i < setup->fs.nr_inputs; i++) { + debug_printf(" v1[%d]: %f %f %f %f\n", i, + v1[i][0], v1[i][1], v1[i][2], v1[i][3]); + } + for (i = 0; i < setup->fs.nr_inputs; i++) { + debug_printf(" v2[%d]: %f %f %f %f\n", i, + v2[i][0], v2[i][1], v2[i][2], v2[i][3]); + } + for (i = 0; i < setup->fs.nr_inputs; i++) { + debug_printf(" v3[%d]: %f %f %f %f\n", i, + v3[i][0], v3[i][1], v3[i][2], v3[i][3]); + } +} + + /** * Do basic setup for triangle rasterization and determine which * framebuffer tiles are touched. Put the triangle in the scene's * bins for the tiles which we overlap. */ static void -do_triangle_ccw(struct setup_context *setup, +do_triangle_ccw(struct lp_setup_context *setup, const float (*v1)[4], const float (*v2)[4], const float (*v3)[4], boolean frontfacing ) { /* x/y positions in fixed point */ - const int x1 = subpixel_snap(v1[0][0]); - const int x2 = subpixel_snap(v2[0][0]); - const int x3 = subpixel_snap(v3[0][0]); - const int y1 = subpixel_snap(v1[0][1]); - const int y2 = subpixel_snap(v2[0][1]); - const int y3 = subpixel_snap(v3[0][1]); + const int x1 = subpixel_snap(v1[0][0] + 0.5 - setup->pixel_offset); + const int x2 = subpixel_snap(v2[0][0] + 0.5 - setup->pixel_offset); + const int x3 = subpixel_snap(v3[0][0] + 0.5 - setup->pixel_offset); + const int y1 = subpixel_snap(v1[0][1] + 0.5 - setup->pixel_offset); + const int y2 = subpixel_snap(v2[0][1] + 0.5 - setup->pixel_offset); + const int y3 = subpixel_snap(v3[0][1] + 0.5 - setup->pixel_offset); struct lp_scene *scene = lp_setup_get_current_scene(setup); - struct lp_rast_triangle *tri = lp_scene_alloc_aligned( scene, sizeof *tri, 16 ); - float area, oneoverarea; + struct lp_rast_triangle *tri; + int area; + float oneoverarea; int minx, maxx, miny, maxy; + unsigned tri_bytes; + + if (0) + print_triangle(setup, v1, v2, v3); + + tri = alloc_triangle(scene, setup->fs.nr_inputs, &tri_bytes); + if (!tri) + return; + +#ifdef DEBUG + tri->v[0][0] = v1[0][0]; + tri->v[1][0] = v2[0][0]; + tri->v[2][0] = v3[0][0]; + tri->v[0][1] = v1[0][1]; + tri->v[1][1] = v2[0][1]; + tri->v[2][1] = v3[0][1]; +#endif tri->dx12 = x1 - x2; tri->dx23 = x2 - x3; @@ -275,15 +395,17 @@ do_triangle_ccw(struct setup_context *setup, tri->dy23 = y2 - y3; tri->dy31 = y3 - y1; - area = (tri->dx12 * tri->dy31 - - tri->dx31 * tri->dy12); + area = (tri->dx12 * tri->dy31 - tri->dx31 * tri->dy12); + + LP_COUNT(nr_tris); /* Cull non-ccw and zero-sized triangles. * * XXX: subject to overflow?? */ - if (area <= 0.0f) { - lp_scene_putback_data( scene, sizeof *tri ); + if (area <= 0) { + lp_scene_putback_data( scene, tri_bytes ); + LP_COUNT(nr_culled_tris); return; } @@ -302,7 +424,8 @@ do_triangle_ccw(struct setup_context *setup, if (miny == maxy || minx == maxx) { - lp_scene_putback_data( scene, sizeof *tri ); + lp_scene_putback_data( scene, tri_bytes ); + LP_COUNT(nr_culled_tris); return; } @@ -314,6 +437,8 @@ do_triangle_ccw(struct setup_context *setup, */ setup_tri_coefficients( setup, tri, oneoverarea, v1, v2, v3, frontfacing ); + tri->inputs.facing = frontfacing ? 1.0F : -1.0F; + /* half-edge constants, will be interated over the whole render target. */ tri->c1 = tri->dy12 * x1 - tri->dx12 * y1; @@ -357,31 +482,44 @@ do_triangle_ccw(struct setup_context *setup, tri->ei2 = tri->dx23 - tri->dy23 - tri->eo2; tri->ei3 = tri->dx31 - tri->dy31 - tri->eo3; + /* Fill in the inputs.step[][] arrays. + * We've manually unrolled some loops here. + */ { const int xstep1 = -tri->dy12; const int xstep2 = -tri->dy23; const int xstep3 = -tri->dy31; - const int ystep1 = tri->dx12; const int ystep2 = tri->dx23; const int ystep3 = tri->dx31; - - int qx, qy, ix, iy; - int i = 0; - - for (qy = 0; qy < 2; qy++) { - for (qx = 0; qx < 2; qx++) { - for (iy = 0; iy < 2; iy++) { - for (ix = 0; ix < 2; ix++, i++) { - int x = qx * 2 + ix; - int y = qy * 2 + iy; - tri->inputs.step[0][i] = x * xstep1 + y * ystep1; - tri->inputs.step[1][i] = x * xstep2 + y * ystep2; - tri->inputs.step[2][i] = x * xstep3 + y * ystep3; - } - } - } - } + +#define SETUP_STEP(i, x, y) \ + do { \ + tri->inputs.step[0][i] = x * xstep1 + y * ystep1; \ + tri->inputs.step[1][i] = x * xstep2 + y * ystep2; \ + tri->inputs.step[2][i] = x * xstep3 + y * ystep3; \ + } while (0) + + SETUP_STEP(0, 0, 0); + SETUP_STEP(1, 1, 0); + SETUP_STEP(2, 0, 1); + SETUP_STEP(3, 1, 1); + + SETUP_STEP(4, 2, 0); + SETUP_STEP(5, 3, 0); + SETUP_STEP(6, 2, 1); + SETUP_STEP(7, 3, 1); + + SETUP_STEP(8, 0, 2); + SETUP_STEP(9, 1, 2); + SETUP_STEP(10, 0, 3); + SETUP_STEP(11, 1, 3); + + SETUP_STEP(12, 2, 2); + SETUP_STEP(13, 3, 2); + SETUP_STEP(14, 2, 3); + SETUP_STEP(15, 3, 3); +#undef STEP } /* @@ -396,8 +534,11 @@ do_triangle_ccw(struct setup_context *setup, maxx = maxx / TILE_SIZE; maxy = maxy / TILE_SIZE; - /* Clamp maxx, maxy to framebuffer size + /* + * Clamp to framebuffer size */ + minx = MAX2(minx, 0); + miny = MAX2(miny, 0); maxx = MIN2(maxx, scene->tiles_x - 1); maxy = MIN2(maxy, scene->tiles_y - 1); @@ -440,8 +581,10 @@ do_triangle_ccw(struct setup_context *setup, int x, y; - /* Trivially accept or reject blocks, else jump to per-pixel - * examination above. + /* Test tile-sized blocks against the triangle. + * Discard blocks fully outside the tri. If the block is fully + * contained inside the tri, bin an lp_rast_shade_tile command. + * Else, bin a lp_rast_triangle command. */ for (y = miny; y <= maxy; y++) { @@ -457,6 +600,7 @@ do_triangle_ccw(struct setup_context *setup, cx3 + eo3 < 0) { /* do nothing */ + LP_COUNT(nr_empty_64); if (in) break; /* exiting triangle, all done with this row */ } @@ -464,9 +608,11 @@ do_triangle_ccw(struct setup_context *setup, cx2 + ei2 > 0 && cx3 + ei3 > 0) { - in = TRUE; /* triangle covers the whole tile- shade whole tile */ - if(setup->fs.current.opaque) { + LP_COUNT(nr_fully_covered_64); + in = TRUE; + if (setup->fs.current.variant->opaque && + !setup->fb.zsbuf) { lp_scene_bin_reset( scene, x, y ); lp_scene_bin_command( scene, x, y, lp_rast_set_state, @@ -478,8 +624,9 @@ do_triangle_ccw(struct setup_context *setup, } else { + /* rasterizer/shade partial tile */ + LP_COUNT(nr_partially_covered_64); in = TRUE; - /* shade partial tile */ lp_scene_bin_command( scene, x, y, lp_rast_triangle, lp_rast_arg_triangle(tri) ); @@ -502,7 +649,10 @@ do_triangle_ccw(struct setup_context *setup, } -static void triangle_cw( struct setup_context *setup, +/** + * Draw triangle if it's CW, cull otherwise. + */ +static void triangle_cw( struct lp_setup_context *setup, const float (*v0)[4], const float (*v1)[4], const float (*v2)[4] ) @@ -511,7 +661,10 @@ static void triangle_cw( struct setup_context *setup, } -static void triangle_ccw( struct setup_context *setup, +/** + * Draw triangle if it's CCW, cull otherwise. + */ +static void triangle_ccw( struct lp_setup_context *setup, const float (*v0)[4], const float (*v1)[4], const float (*v2)[4] ) @@ -520,7 +673,11 @@ static void triangle_ccw( struct setup_context *setup, } -static void triangle_both( struct setup_context *setup, + +/** + * Draw triangle whether it's CW or CCW. + */ +static void triangle_both( struct lp_setup_context *setup, const float (*v0)[4], const float (*v1)[4], const float (*v2)[4] ) @@ -539,7 +696,7 @@ static void triangle_both( struct setup_context *setup, } -static void triangle_nop( struct setup_context *setup, +static void triangle_nop( struct lp_setup_context *setup, const float (*v0)[4], const float (*v1)[4], const float (*v2)[4] ) @@ -548,17 +705,17 @@ static void triangle_nop( struct setup_context *setup, void -lp_setup_choose_triangle( struct setup_context *setup ) +lp_setup_choose_triangle( struct lp_setup_context *setup ) { switch (setup->cullmode) { - case PIPE_WINDING_NONE: + case PIPE_FACE_NONE: setup->triangle = triangle_both; break; - case PIPE_WINDING_CCW: - setup->triangle = triangle_cw; + case PIPE_FACE_BACK: + setup->triangle = setup->ccw_is_frontface ? triangle_ccw : triangle_cw; break; - case PIPE_WINDING_CW: - setup->triangle = triangle_ccw; + case PIPE_FACE_FRONT: + setup->triangle = setup->ccw_is_frontface ? triangle_cw : triangle_ccw; break; default: setup->triangle = triangle_nop;