diff options
Diffstat (limited to 'src/gallium/drivers/cell/spu/spu_tri.c')
| -rw-r--r-- | src/gallium/drivers/cell/spu/spu_tri.c | 789 |
1 files changed, 322 insertions, 467 deletions
diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c index 8b93878192..d727268475 100644 --- a/src/gallium/drivers/cell/spu/spu_tri.c +++ b/src/gallium/drivers/cell/spu/spu_tri.c @@ -29,12 +29,12 @@ * Triangle rendering within a tile. */ -#include <transpose_matrix4x4.h> #include "pipe/p_compiler.h" #include "pipe/p_format.h" #include "util/u_math.h" #include "spu_colorpack.h" #include "spu_main.h" +#include "spu_shuffle.h" #include "spu_texture.h" #include "spu_tile.h" #include "spu_tri.h" @@ -43,11 +43,6 @@ /** Masks are uint[4] vectors with each element being 0 or 0xffffffff */ typedef vector unsigned int mask_t; -typedef union -{ - vector float v; - float f[4]; -} float4; /** @@ -61,7 +56,7 @@ struct vertex_header { /* XXX fix this */ #undef CEILF -#define CEILF(X) ((float) (int) ((X) + 0.99999)) +#define CEILF(X) ((float) (int) ((X) + 0.99999f)) #define QUAD_TOP_LEFT 0 @@ -75,14 +70,25 @@ struct vertex_header { #define MASK_ALL 0xf +#define CHAN0 0 +#define CHAN1 1 +#define CHAN2 2 +#define CHAN3 3 + + #define DEBUG_VERTS 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 */ + union { + struct { + float dx; /**< X(v1) - X(v0), used only during setup */ + float dy; /**< Y(v1) - Y(v0), used only during setup */ + }; + vec_float4 ds; /**< vector accessor for dx and dy */ + }; float dxdy; /**< dx/dy */ float sx, sy; /**< first sample point coord */ int lines; /**< number of lines on this edge */ @@ -91,9 +97,9 @@ struct edge { struct interp_coef { - float4 a0; - float4 dadx; - float4 dady; + vector float a0; + vector float dadx; + vector float dady; }; @@ -107,34 +113,32 @@ struct setup_stage { * turn. Currently fixed at 4 floats, but should change in time. * Codegen will help cope with this. */ - const struct vertex_header *vmax; - const struct vertex_header *vmid; - const struct vertex_header *vmin; - const struct vertex_header *vprovoke; + union { + struct { + const struct vertex_header *vmin; + const struct vertex_header *vmid; + const struct vertex_header *vmax; + const struct vertex_header *vprovoke; + }; + qword vertex_headers; + }; struct edge ebot; struct edge etop; struct edge emaj; - float oneoverarea; + float oneOverArea; /* XXX maybe make into vector? */ - uint tx, ty; + uint facing; + + uint tx, ty; /**< position of current tile (x, y) */ int cliprect_minx, cliprect_maxx, cliprect_miny, cliprect_maxy; -#if 0 - struct tgsi_interp_coef coef[PIPE_MAX_SHADER_INPUTS]; -#else struct interp_coef coef[PIPE_MAX_SHADER_INPUTS]; -#endif - -#if 0 - struct quad_header quad; -#endif struct { - int left[2]; /**< [0] = row0, [1] = row1 */ - int right[2]; + vec_int4 quad; /**< [0] = row0, [1] = row1; {left[0],left[1],right[0],right[1]} */ int y; unsigned y_flags; unsigned mask; /**< mask of MASK_BOTTOM/TOP_LEFT/RIGHT bits */ @@ -142,118 +146,103 @@ struct setup_stage { }; - static struct setup_stage setup; +static INLINE vector float +splatx(vector float v) +{ + return spu_splats(spu_extract(v, CHAN0)); +} - -#if 0 -/** - * Basically a cast wrapper. - */ -static INLINE struct setup_stage *setup_stage( struct draw_stage *stage ) +static INLINE vector float +splaty(vector float v) { - return (struct setup_stage *)stage; + return spu_splats(spu_extract(v, CHAN1)); } -#endif -#if 0 -/** - * Clip setup.quad against the scissor/surface bounds. - */ -static INLINE void -quad_clip(struct setup_stage *setup) +static INLINE vector float +splatz(vector float v) { - const struct pipe_scissor_state *cliprect = &setup.softpipe->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 (setup.quad.x0 >= maxx || - setup.quad.y0 >= maxy || - setup.quad.x0 + 1 < minx || - setup.quad.y0 + 1 < miny) { - /* totally clipped */ - setup.quad.mask = 0x0; - return; - } - if (setup.quad.x0 < minx) - setup.quad.mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT); - if (setup.quad.y0 < miny) - setup.quad.mask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT); - if (setup.quad.x0 == maxx - 1) - setup.quad.mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT); - if (setup.quad.y0 == maxy - 1) - setup.quad.mask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT); + return spu_splats(spu_extract(v, CHAN2)); } -#endif -#if 0 -/** - * Emit a quad (pass to next stage) with clipping. - */ -static INLINE void -clip_emit_quad(struct setup_stage *setup) +static INLINE vector float +splatw(vector float v) { - quad_clip(setup); - if (setup.quad.mask) { - struct softpipe_context *sp = setup.softpipe; - sp->quad.first->run(sp->quad.first, &setup.quad); - } + return spu_splats(spu_extract(v, CHAN3)); } -#endif + /** - * Evaluate attribute coefficients (plane equations) to compute - * attribute values for the four fragments in a quad. - * Eg: four colors will be computed (in AoS format). + * Setup fragment shader inputs by evaluating triangle's vertex + * attribute coefficient info. + * \param x quad x pos + * \param y quad y pos + * \param fragZ returns quad Z values + * \param fragInputs returns fragment program inputs + * Note: this code could be incorporated into the fragment program + * itself to avoid the loop and switch. */ -static INLINE void -eval_coeff(uint slot, float x, float y, vector float result[4]) +static void +eval_inputs(float x, float y, vector float *fragZ, vector float fragInputs[]) { - switch (spu.vertex_info.interp_mode[slot]) { - case INTERP_CONSTANT: - result[QUAD_TOP_LEFT] = - result[QUAD_TOP_RIGHT] = - result[QUAD_BOTTOM_LEFT] = - result[QUAD_BOTTOM_RIGHT] = setup.coef[slot].a0.v; - break; + static const vector float deltaX = (const vector float) {0, 1, 0, 1}; + static const vector float deltaY = (const vector float) {0, 0, 1, 1}; + + const uint posSlot = 0; + const vector float pos = setup.coef[posSlot].a0; + const vector float dposdx = setup.coef[posSlot].dadx; + const vector float dposdy = setup.coef[posSlot].dady; + const vector float fragX = spu_splats(x) + deltaX; + const vector float fragY = spu_splats(y) + deltaY; + vector float fragW, wInv; + uint i; - case INTERP_LINEAR: - /* fall-through, for now */ - default: - { - register vector float dadx = setup.coef[slot].dadx.v; - register vector float dady = setup.coef[slot].dady.v; - register vector float topLeft - = spu_add(setup.coef[slot].a0.v, - spu_add(spu_mul(spu_splats(x), dadx), - spu_mul(spu_splats(y), dady))); - - result[QUAD_TOP_LEFT] = topLeft; - result[QUAD_TOP_RIGHT] = spu_add(topLeft, dadx); - result[QUAD_BOTTOM_LEFT] = spu_add(topLeft, dady); - result[QUAD_BOTTOM_RIGHT] = spu_add(spu_add(topLeft, dadx), dady); + *fragZ = splatz(pos) + fragX * splatz(dposdx) + fragY * splatz(dposdy); + fragW = splatw(pos) + fragX * splatw(dposdx) + fragY * splatw(dposdy); + wInv = spu_re(fragW); /* 1 / w */ + + /* loop over fragment program inputs */ + for (i = 0; i < spu.vertex_info.num_attribs; i++) { + uint attr = i + 1; + enum interp_mode interp = spu.vertex_info.attrib[attr].interp_mode; + + /* constant term */ + vector float a0 = setup.coef[attr].a0; + vector float r0 = splatx(a0); + vector float r1 = splaty(a0); + vector float r2 = splatz(a0); + vector float r3 = splatw(a0); + + if (interp == INTERP_LINEAR || interp == INTERP_PERSPECTIVE) { + /* linear term */ + vector float dadx = setup.coef[attr].dadx; + vector float dady = setup.coef[attr].dady; + /* Use SPU intrinsics here to get slightly better code. + * originally: r0 += fragX * splatx(dadx) + fragY * splatx(dady); + */ + r0 = spu_madd(fragX, splatx(dadx), spu_madd(fragY, splatx(dady), r0)); + r1 = spu_madd(fragX, splaty(dadx), spu_madd(fragY, splaty(dady), r1)); + r2 = spu_madd(fragX, splatz(dadx), spu_madd(fragY, splatz(dady), r2)); + r3 = spu_madd(fragX, splatw(dadx), spu_madd(fragY, splatw(dady), r3)); + if (interp == INTERP_PERSPECTIVE) { + /* perspective term */ + r0 *= wInv; + r1 *= wInv; + r2 *= wInv; + r3 *= wInv; + } } + fragInputs[CHAN0] = r0; + fragInputs[CHAN1] = r1; + fragInputs[CHAN2] = r2; + fragInputs[CHAN3] = r3; + fragInputs += 4; } } -static INLINE vector float -eval_z(float x, float y) -{ - const uint slot = 0; - const float dzdx = setup.coef[slot].dadx.f[2]; - const float dzdy = setup.coef[slot].dady.f[2]; - const float topLeft = setup.coef[slot].a0.f[2] + x * dzdx + y * dzdy; - const vector float topLeftv = spu_splats(topLeft); - const vector float derivs = (vector float) { 0.0, dzdx, dzdy, dzdx + dzdy }; - return spu_add(topLeftv, derivs); -} - - /** * Emit a quad (pass to next stage). No clipping is done. * Note: about 1/5 to 1/7 of the time, mask is zero and this function @@ -261,120 +250,51 @@ eval_z(float x, float y) * overall. */ static INLINE void -emit_quad( int x, int y, mask_t mask ) +emit_quad( int x, int y, mask_t mask) { /* If any bits in mask are set... */ if (spu_extract(spu_orx(mask), 0)) { const int ix = x - setup.cliprect_minx; const int iy = y - setup.cliprect_miny; - vector float colors[4]; spu.cur_ctile_status = TILE_STATUS_DIRTY; spu.cur_ztile_status = TILE_STATUS_DIRTY; - if (spu.texture[0].start) { - /* texture mapping */ - const uint unit = 0; - vector float texcoords[4]; - eval_coeff(2, (float) x, (float) y, texcoords); - - if (spu_extract(mask, 0)) - colors[0] = spu.sample_texture[unit](unit, texcoords[0]); - if (spu_extract(mask, 1)) - colors[1] = spu.sample_texture[unit](unit, texcoords[1]); - if (spu_extract(mask, 2)) - colors[2] = spu.sample_texture[unit](unit, texcoords[2]); - if (spu_extract(mask, 3)) - colors[3] = spu.sample_texture[unit](unit, texcoords[3]); - - - if (spu.texture[1].start) { - /* multi-texture mapping */ - const uint unit = 1; - vector float colors1[4]; - - eval_coeff(2, (float) x, (float) y, texcoords); - - if (spu_extract(mask, 0)) - colors1[0] = spu.sample_texture[unit](unit, texcoords[0]); - if (spu_extract(mask, 1)) - colors1[1] = spu.sample_texture[unit](unit, texcoords[1]); - if (spu_extract(mask, 2)) - colors1[2] = spu.sample_texture[unit](unit, texcoords[2]); - if (spu_extract(mask, 3)) - colors1[3] = spu.sample_texture[unit](unit, texcoords[3]); - - /* hack: modulate first texture by second */ - colors[0] = spu_mul(colors[0], colors1[0]); - colors[1] = spu_mul(colors[1], colors1[1]); - colors[2] = spu_mul(colors[2], colors1[2]); - colors[3] = spu_mul(colors[3], colors1[3]); - } - - } - else { - /* simple shading */ -#if 0 - eval_coeff(1, (float) x, (float) y, colors); - -#else - /* XXX new fragment program code */ - - if (spu.fragment_program) { - vector float inputs[4*4], outputs[2*4]; - - /* setup inputs */ - eval_coeff(1, (float) x, (float) y, inputs); - - /* Execute the current fragment program */ - spu.fragment_program(inputs, outputs, spu.constants); - - /* Copy outputs */ - colors[0] = outputs[0*4+0]; - colors[1] = outputs[0*4+1]; - colors[2] = outputs[0*4+2]; - colors[3] = outputs[0*4+3]; - - if (0 && spu.init.id==0 && y == 48) { - printf("colors[0] = %f %f %f %f\n", - spu_extract(colors[0], 0), - spu_extract(colors[0], 1), - spu_extract(colors[0], 2), - spu_extract(colors[0], 3)); - printf("colors[1] = %f %f %f %f\n", - spu_extract(colors[1], 0), - spu_extract(colors[1], 1), - spu_extract(colors[1], 2), - spu_extract(colors[1], 3)); - } - - } -#endif - } - - { - /* Convert fragment data from AoS to SoA format. - * I.e. (RGBA,RGBA,RGBA,RGBA) -> (RRRR,GGGG,BBBB,AAAA) - * This is temporary! + /* + * Run fragment shader, execute per-fragment ops, update fb/tile. */ - vector float soa_frag[4]; - _transpose_matrix4x4(soa_frag, colors); + vector float inputs[4*4], outputs[2*4]; + vector unsigned int kill_mask; + vector float fragZ; + + eval_inputs((float) x, (float) y, &fragZ, inputs); - float4 fragZ; + ASSERT(spu.fragment_program); + ASSERT(spu.fragment_ops); - fragZ.v = eval_z((float) x, (float) y); + /* Execute the current fragment program */ + kill_mask = spu.fragment_program(inputs, outputs, spu.constants); - /* Do all per-fragment/quad operations here, including: - * alpha test, z test, stencil test, blend and framebuffer writing. + mask = spu_andc(mask, kill_mask); + + /* Execute per-fragment/quad operations, including: + * alpha test, z test, stencil test, blend and framebuffer writing. + * Note that there are two different fragment operations functions + * that can be called, one for front-facing fragments, and one + * for back-facing fragments. (Often the two are the same; + * but in some cases, like two-sided stenciling, they can be + * very different.) So choose the correct function depending + * on the calculated facing. */ - spu.fragment_ops(ix, iy, &spu.ctile, &spu.ztile, - fragZ.v, - soa_frag[0], soa_frag[1], - soa_frag[2], soa_frag[3], + spu.fragment_ops[setup.facing](ix, iy, &spu.ctile, &spu.ztile, + fragZ, + outputs[0*4+0], + outputs[0*4+1], + outputs[0*4+2], + outputs[0*4+3], mask); } - } } @@ -383,64 +303,49 @@ emit_quad( int x, int y, mask_t mask ) * Given an X or Y coordinate, return the block/quad coordinate that it * belongs to. */ -static INLINE int block( int x ) +static INLINE int +block(int x) { return x & ~1; } /** - * Compute mask which indicates which pixels in the 2x2 quad are actually inside - * the triangle's bounds. - * The mask is a uint4 vector and each element will be 0 or 0xffffffff. - */ -static INLINE mask_t calculate_mask( int x ) -{ - /* This is a little tricky. - * Use & instead of && to avoid branches. - * Use negation to convert true/false to ~0/0 values. - */ - mask_t mask; - mask = spu_insert(-((x >= setup.span.left[0]) & (x < setup.span.right[0])), mask, 0); - mask = spu_insert(-((x+1 >= setup.span.left[0]) & (x+1 < setup.span.right[0])), mask, 1); - mask = spu_insert(-((x >= setup.span.left[1]) & (x < setup.span.right[1])), mask, 2); - mask = spu_insert(-((x+1 >= setup.span.left[1]) & (x+1 < setup.span.right[1])), mask, 3); - return mask; -} - - -/** * Render a horizontal span of quads */ -static void flush_spans( void ) +static void +flush_spans(void) { int minleft, maxright; - int x; + + const int l0 = spu_extract(setup.span.quad, 0); + const int l1 = spu_extract(setup.span.quad, 1); + const int r0 = spu_extract(setup.span.quad, 2); + const int r1 = spu_extract(setup.span.quad, 3); switch (setup.span.y_flags) { case 0x3: /* both odd and even lines written (both quad rows) */ - minleft = MIN2(setup.span.left[0], setup.span.left[1]); - maxright = MAX2(setup.span.right[0], setup.span.right[1]); + minleft = MIN2(l0, l1); + maxright = MAX2(r0, r1); break; case 0x1: /* only even line written (quad top row) */ - minleft = setup.span.left[0]; - maxright = setup.span.right[0]; + minleft = l0; + maxright = r0; break; case 0x2: /* only odd line written (quad bottom row) */ - minleft = setup.span.left[1]; - maxright = setup.span.right[1]; + minleft = l1; + maxright = r1; break; default: return; } - /* OK, we're very likely to need the tile data now. * clear or finish waiting if needed. */ @@ -457,7 +362,7 @@ static void flush_spans( void ) } ASSERT(spu.cur_ctile_status != TILE_STATUS_DEFINED); - if (spu.read_depth) { + if (spu.read_depth_stencil) { if (spu.cur_ztile_status == TILE_STATUS_GETTING) { /* wait for mfc_get() to complete */ //printf("SPU: %u: waiting for ztile\n", spu.init.id); @@ -472,93 +377,119 @@ static void flush_spans( void ) ASSERT(spu.cur_ztile_status != TILE_STATUS_DEFINED); } - /* XXX this loop could be moved into the above switch cases and - * calculate_mask() could be simplified a bit... - */ - for (x = block(minleft); x <= block(maxright); x += 2) { -#if 1 - emit_quad( x, setup.span.y, calculate_mask( x ) ); -#endif + /* XXX this loop could be moved into the above switch cases... */ + + /* Setup for mask calculation */ + const vec_int4 quad_LlRr = setup.span.quad; + const vec_int4 quad_RrLl = spu_rlqwbyte(quad_LlRr, 8); + const vec_int4 quad_LLll = spu_shuffle(quad_LlRr, quad_LlRr, SHUFFLE4(A,A,B,B)); + const vec_int4 quad_RRrr = spu_shuffle(quad_RrLl, quad_RrLl, SHUFFLE4(A,A,B,B)); + + const vec_int4 twos = spu_splats(2); + + const int x = block(minleft); + vec_int4 xs = {x, x+1, x, x+1}; + + for (; spu_extract(xs, 0) <= block(maxright); xs += twos) { + /** + * Computes mask to indicate which pixels in the 2x2 quad are actually + * inside the triangle's bounds. + */ + + /* Calculate ({x,x+1,x,x+1} >= {l[0],l[0],l[1],l[1]}) */ + const mask_t gt_LLll_xs = spu_cmpgt(quad_LLll, xs); + const mask_t gte_xs_LLll = spu_nand(gt_LLll_xs, gt_LLll_xs); + + /* Calculate ({r[0],r[0],r[1],r[1]} > {x,x+1,x,x+1}) */ + const mask_t gt_RRrr_xs = spu_cmpgt(quad_RRrr, xs); + + /* Combine results to create mask */ + const mask_t mask = spu_and(gte_xs_LLll, gt_RRrr_xs); + + emit_quad(spu_extract(xs, 0), setup.span.y, mask); } setup.span.y = 0; setup.span.y_flags = 0; - setup.span.right[0] = 0; - setup.span.right[1] = 0; + /* Zero right elements */ + setup.span.quad = spu_shuffle(setup.span.quad, setup.span.quad, SHUFFLE4(A,B,0,0)); } + #if DEBUG_VERTS -static void print_vertex(const struct vertex_header *v) +static void +print_vertex(const struct vertex_header *v) { - int i; - fprintf(stderr, "Vertex: (%p)\n", v); - for (i = 0; i < setup.quad.nr_attrs; i++) { - fprintf(stderr, " %d: %f %f %f %f\n", i, - v->data[i][0], v->data[i][1], v->data[i][2], v->data[i][3]); + uint i; + fprintf(stderr, " Vertex: (%p)\n", v); + for (i = 0; i < spu.vertex_info.num_attribs; i++) { + fprintf(stderr, " %d: %f %f %f %f\n", i, + spu_extract(v->data[i], 0), + spu_extract(v->data[i], 1), + spu_extract(v->data[i], 2), + spu_extract(v->data[i], 3)); } } #endif -static boolean setup_sort_vertices(const struct vertex_header *v0, - const struct vertex_header *v1, - const struct vertex_header *v2) +/** + * Sort vertices from top to bottom. + * Compute area and determine front vs. back facing. + * Do coarse clip test against tile bounds + * \return FALSE if tri is totally outside tile, TRUE otherwise + */ +static boolean +setup_sort_vertices(const struct vertex_header *v0, + const struct vertex_header *v1, + const struct vertex_header *v2) { + float area, sign; #if DEBUG_VERTS - fprintf(stderr, "Triangle:\n"); - print_vertex(v0); - print_vertex(v1); - print_vertex(v2); + if (spu.init.id==0) { + fprintf(stderr, "SPU %u: Triangle:\n", spu.init.id); + print_vertex(v0); + print_vertex(v1); + print_vertex(v2); + } #endif - setup.vprovoke = v2; - /* determine bottom to top order of vertices */ { - float y0 = spu_extract(v0->data[0], 1); - float y1 = spu_extract(v1->data[0], 1); - float y2 = spu_extract(v2->data[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; - } - } + /* A table of shuffle patterns for putting vertex_header pointers into + correct order. Quite magical. */ + const vec_uchar16 sort_order_patterns[] = { + SHUFFLE4(A,B,C,C), + SHUFFLE4(C,A,B,C), + SHUFFLE4(A,C,B,C), + SHUFFLE4(B,C,A,C), + SHUFFLE4(B,A,C,C), + SHUFFLE4(C,B,A,C) }; + + /* The vertex_header pointers, packed for easy shuffling later */ + const vec_uint4 vs = {(unsigned)v0, (unsigned)v1, (unsigned)v2}; + + /* Collate y values into two vectors for comparison. + Using only one shuffle constant! ;) */ + const vec_float4 y_02_ = spu_shuffle(v0->data[0], v2->data[0], SHUFFLE4(0,B,b,C)); + const vec_float4 y_10_ = spu_shuffle(v1->data[0], v0->data[0], SHUFFLE4(0,B,b,C)); + const vec_float4 y_012 = spu_shuffle(y_02_, v1->data[0], SHUFFLE4(0,B,b,C)); + const vec_float4 y_120 = spu_shuffle(y_10_, v2->data[0], SHUFFLE4(0,B,b,C)); + + /* Perform comparison: {y0,y1,y2} > {y1,y2,y0} */ + const vec_uint4 compare = spu_cmpgt(y_012, y_120); + /* Compress the result of the comparison into 4 bits */ + const vec_uint4 gather = spu_gather(compare); + /* Subtract one to attain the index into the LUT. Magical. */ + const unsigned int index = spu_extract(gather, 0) - 1; + + /* Load the appropriate pattern and construct the desired vector. */ + setup.vertex_headers = (qword)spu_shuffle(vs, vs, sort_order_patterns[index]); + + /* Using the result of the comparison, set sign. + Very magical. */ + sign = ((si_to_uint(si_cntb((qword)gather)) == 2) ? 1.0f : -1.0f); } /* Check if triangle is completely outside the tile bounds */ @@ -575,41 +506,28 @@ static boolean setup_sort_vertices(const struct vertex_header *v0, spu_extract(setup.vmax->data[0], 0) > setup.cliprect_maxx) return FALSE; - setup.ebot.dx = spu_extract(setup.vmid->data[0], 0) - spu_extract(setup.vmin->data[0], 0); - setup.ebot.dy = spu_extract(setup.vmid->data[0], 1) - spu_extract(setup.vmin->data[0], 1); - setup.emaj.dx = spu_extract(setup.vmax->data[0], 0) - spu_extract(setup.vmin->data[0], 0); - setup.emaj.dy = spu_extract(setup.vmax->data[0], 1) - spu_extract(setup.vmin->data[0], 1); - setup.etop.dx = spu_extract(setup.vmax->data[0], 0) - spu_extract(setup.vmid->data[0], 0); - setup.etop.dy = spu_extract(setup.vmax->data[0], 1) - spu_extract(setup.vmid->data[0], 1); + setup.ebot.ds = spu_sub(setup.vmid->data[0], setup.vmin->data[0]); + setup.emaj.ds = spu_sub(setup.vmax->data[0], setup.vmin->data[0]); + setup.etop.ds = spu_sub(setup.vmax->data[0], setup.vmid->data[0]); /* * 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; - /* - _mesa_printf("%s one-over-area %f area %f det %f\n", - __FUNCTION__, setup.oneoverarea, area, prim->det ); - */ - } + area = setup.emaj.dx * setup.ebot.dy - setup.ebot.dx * setup.emaj.dy; + + setup.oneOverArea = 1.0f / area; -#if 0 - /* 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 + /* The product of area * sign indicates front/back orientation (0/1). + * Just in case someone gets the bright idea of switching the front + * and back constants without noticing that we're assuming their + * values in this operation, also assert that the values are + * what we think they are. */ - setup.quad.facing = (prim->det > 0.0) ^ (setup.softpipe->rasterizer->front_winding == PIPE_WINDING_CW); -#endif + ASSERT(CELL_FACING_FRONT == 0); + ASSERT(CELL_FACING_BACK == 1); + setup.facing = (area * sign > 0.0f) + ^ (spu.rasterizer.front_winding == PIPE_WINDING_CW); return TRUE; } @@ -622,63 +540,11 @@ static boolean setup_sort_vertices(const struct vertex_header *v0, * \param slot which attribute slot */ static INLINE void -const_coeff(uint slot) +const_coeff4(uint slot) { - setup.coef[slot].dadx.v = (vector float) {0.0, 0.0, 0.0, 0.0}; - setup.coef[slot].dady.v = (vector float) {0.0, 0.0, 0.0, 0.0}; - setup.coef[slot].a0.v = setup.vprovoke->data[slot]; -} - - -/** - * Compute a0, dadx and dady for a linearly interpolated coefficient, - * for a triangle. - */ -static INLINE void -tri_linear_coeff(uint slot, uint firstComp, uint lastComp) -{ - uint i; - const float *vmin_d = (float *) &setup.vmin->data[slot]; - const float *vmid_d = (float *) &setup.vmid->data[slot]; - const float *vmax_d = (float *) &setup.vmax->data[slot]; - const float x = spu_extract(setup.vmin->data[0], 0) - 0.5f; - const float y = spu_extract(setup.vmin->data[0], 1) - 0.5f; - - for (i = firstComp; i < lastComp; i++) { - float botda = vmid_d[i] - vmin_d[i]; - float majda = vmax_d[i] - vmin_d[i]; - float a = setup.ebot.dy * majda - botda * setup.emaj.dy; - float b = setup.emaj.dx * botda - majda * setup.ebot.dx; - - ASSERT(slot < PIPE_MAX_SHADER_INPUTS); - - setup.coef[slot].dadx.f[i] = a * setup.oneoverarea; - setup.coef[slot].dady.f[i] = b * setup.oneoverarea; - - /* 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[slot].a0.f[i] = (vmin_d[i] - - (setup.coef[slot].dadx.f[i] * x + - setup.coef[slot].dady.f[i] * y)); - } - - /* - _mesa_printf("attr[%d].%c: %f dx:%f dy:%f\n", - slot, "xyzw"[i], - setup.coef[slot].a0[i], - setup.coef[slot].dadx.f[i], - setup.coef[slot].dady.f[i]); - */ + setup.coef[slot].dadx = (vector float) {0.0, 0.0, 0.0, 0.0}; + setup.coef[slot].dady = (vector float) {0.0, 0.0, 0.0, 0.0}; + setup.coef[slot].a0 = setup.vprovoke->data[slot]; } @@ -702,18 +568,16 @@ tri_linear_coeff4(uint slot) vector float b = spu_sub(spu_mul(spu_splats(setup.emaj.dx), botda), spu_mul(majda, spu_splats(setup.ebot.dx))); - setup.coef[slot].dadx.v = spu_mul(a, spu_splats(setup.oneoverarea)); - setup.coef[slot].dady.v = spu_mul(b, spu_splats(setup.oneoverarea)); + setup.coef[slot].dadx = spu_mul(a, spu_splats(setup.oneOverArea)); + setup.coef[slot].dady = spu_mul(b, spu_splats(setup.oneOverArea)); - vector float tempx = spu_mul(setup.coef[slot].dadx.v, xxxx); - vector float tempy = spu_mul(setup.coef[slot].dady.v, yyyy); + vector float tempx = spu_mul(setup.coef[slot].dadx, xxxx); + vector float tempy = spu_mul(setup.coef[slot].dady, yyyy); - setup.coef[slot].a0.v = spu_sub(vmin_d, spu_add(tempx, tempy)); + setup.coef[slot].a0 = spu_sub(vmin_d, spu_add(tempx, tempy)); } - -#if 0 /** * Compute a0, dadx and dady for a perspective-corrected interpolant, * for a triangle. @@ -722,82 +586,76 @@ tri_linear_coeff4(uint slot) * 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( unsigned slot, - unsigned i ) +static void +tri_persp_coeff4(uint slot) { - /* premultiply by 1/w: - */ - float mina = setup.vmin->data[slot][i] * setup.vmin->data[0][3]; - float mida = setup.vmid->data[slot][i] * setup.vmid->data[0][3]; - float maxa = setup.vmax->data[slot][i] * setup.vmax->data[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; - - /* - printf("tri persp %d,%d: %f %f %f\n", slot, i, - setup.vmin->data[slot][i], - setup.vmid->data[slot][i], - setup.vmax->data[slot][i] - ); - */ + const vector float xxxx = spu_splats(spu_extract(setup.vmin->data[0], 0) - 0.5f); + const vector float yyyy = spu_splats(spu_extract(setup.vmin->data[0], 1) - 0.5f); + + const vector float vmin_w = spu_splats(spu_extract(setup.vmin->data[0], 3)); + const vector float vmid_w = spu_splats(spu_extract(setup.vmid->data[0], 3)); + const vector float vmax_w = spu_splats(spu_extract(setup.vmax->data[0], 3)); + + vector float vmin_d = setup.vmin->data[slot]; + vector float vmid_d = setup.vmid->data[slot]; + vector float vmax_d = setup.vmax->data[slot]; + + vmin_d = spu_mul(vmin_d, vmin_w); + vmid_d = spu_mul(vmid_d, vmid_w); + vmax_d = spu_mul(vmax_d, vmax_w); + + vector float botda = vmid_d - vmin_d; + vector float majda = vmax_d - vmin_d; - assert(slot < PIPE_MAX_SHADER_INPUTS); - assert(i <= 3); + vector float a = spu_sub(spu_mul(spu_splats(setup.ebot.dy), majda), + spu_mul(botda, spu_splats(setup.emaj.dy))); + vector float b = spu_sub(spu_mul(spu_splats(setup.emaj.dx), botda), + spu_mul(majda, spu_splats(setup.ebot.dx))); - setup.coef[slot].dadx.f[i] = a * setup.oneoverarea; - setup.coef[slot].dady.f[i] = b * setup.oneoverarea; - setup.coef[slot].a0.f[i] = (mina - - (setup.coef[slot].dadx.f[i] * (setup.vmin->data[0][0] - 0.5f) + - setup.coef[slot].dady.f[i] * (setup.vmin->data[0][1] - 0.5f))); + setup.coef[slot].dadx = spu_mul(a, spu_splats(setup.oneOverArea)); + setup.coef[slot].dady = spu_mul(b, spu_splats(setup.oneOverArea)); + + vector float tempx = spu_mul(setup.coef[slot].dadx, xxxx); + vector float tempy = spu_mul(setup.coef[slot].dady, yyyy); + + setup.coef[slot].a0 = spu_sub(vmin_d, spu_add(tempx, tempy)); } -#endif + /** * 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(void) +static void +setup_tri_coefficients(void) { -#if 1 uint i; for (i = 0; i < spu.vertex_info.num_attribs; i++) { - switch (spu.vertex_info.interp_mode[i]) { + switch (spu.vertex_info.attrib[i].interp_mode) { case INTERP_NONE: break; - case INTERP_POS: - /*tri_linear_coeff(i, 2, 3);*/ - /* XXX interp W if PERSPECTIVE... */ - tri_linear_coeff4(i); - break; case INTERP_CONSTANT: - const_coeff(i); + const_coeff4(i); break; + case INTERP_POS: + /* fall-through */ case INTERP_LINEAR: tri_linear_coeff4(i); break; case INTERP_PERSPECTIVE: - tri_linear_coeff4(i); /* temporary */ + tri_persp_coeff4(i); break; default: ASSERT(0); } } -#else - ASSERT(spu.vertex_info.interp_mode[0] == INTERP_POS); - ASSERT(spu.vertex_info.interp_mode[1] == INTERP_LINEAR || - spu.vertex_info.interp_mode[1] == INTERP_CONSTANT); - tri_linear_coeff(0, 2, 3); /* slot 0, z */ - tri_linear_coeff(1, 0, 4); /* slot 1, color */ -#endif } -static void setup_tri_edges(void) +static void +setup_tri_edges(void) { float vmin_x = spu_extract(setup.vmin->data[0], 0) + 0.5f; float vmid_x = spu_extract(setup.vmid->data[0], 0) + 0.5f; @@ -827,9 +685,8 @@ static void setup_tri_edges(void) * Render the upper or lower half of a triangle. * Scissoring/cliprect is applied here too. */ -static void subtriangle( struct edge *eleft, - struct edge *eright, - unsigned lines ) +static void +subtriangle(struct edge *eleft, struct edge *eright, unsigned lines) { const int minx = setup.cliprect_minx; const int maxx = setup.cliprect_maxx; @@ -881,9 +738,11 @@ static void subtriangle( struct edge *eleft, setup.span.y = block(_y); } - setup.span.left[_y&1] = left; - setup.span.right[_y&1] = right; - setup.span.y_flags |= 1<<(_y&1); + int offset = _y&1; + vec_int4 quad_LlRr = {left, left, right, right}; + /* Store left and right in 0 or 1 row of quad based on offset */ + setup.span.quad = spu_sel(quad_LlRr, setup.span.quad, spu_maskw(5<<offset)); + setup.span.y_flags |= 1<<offset; } } @@ -902,7 +761,8 @@ static void subtriangle( struct edge *eleft, * The tile data should have already been fetched. */ boolean -tri_draw(const float *v0, const float *v1, const float *v2, uint tx, uint ty) +tri_draw(const float *v0, const float *v1, const float *v2, + uint tx, uint ty) { setup.tx = tx; setup.ty = ty; @@ -924,21 +784,16 @@ tri_draw(const float *v0, const float *v1, const float *v2, uint tx, uint ty) setup.span.y = 0; setup.span.y_flags = 0; - setup.span.right[0] = 0; - setup.span.right[1] = 0; - /* setup.span.z_mode = tri_z_mode( setup.ctx ); */ + /* Zero right elements */ + setup.span.quad = spu_shuffle(setup.span.quad, setup.span.quad, SHUFFLE4(A,B,0,0)); - /* init_constant_attribs( setup ); */ - - if (setup.oneoverarea < 0.0) { - /* emaj on left: - */ + if (setup.oneOverArea < 0.0) { + /* emaj on left */ subtriangle( &setup.emaj, &setup.ebot, setup.ebot.lines ); subtriangle( &setup.emaj, &setup.etop, setup.etop.lines ); } else { - /* emaj on right: - */ + /* emaj on right */ subtriangle( &setup.ebot, &setup.emaj, setup.ebot.lines ); subtriangle( &setup.etop, &setup.emaj, setup.etop.lines ); } |