* Triangle rendering within a tile.
*/
-#include <transpose_matrix4x4.h>
#include "pipe/p_compiler.h"
#include "pipe/p_format.h"
#include "util/u_math.h"
#define MASK_ALL 0xf
+#define CHAN0 0
+#define CHAN1 1
+#define CHAN2 2
+#define CHAN3 3
+
+
#define DEBUG_VERTS 0
/**
uint tx, ty; /**< position of current tile (x, y) */
- int cliprect_minx, cliprect_maxx, cliprect_miny, cliprect_maxy;
+ union {
+ struct {
+ int cliprect_minx;
+ int cliprect_miny;
+ int cliprect_maxx;
+ int cliprect_maxy;
+ };
+ qword cliprect;
+ };
struct interp_coef coef[PIPE_MAX_SHADER_INPUTS];
static struct setup_stage setup;
-/**
- * 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).
- */
-static INLINE void
-eval_coeff(uint slot, float x, float y, vector float w, vector float result[4])
+static INLINE vector float
+splatx(vector float v)
{
- switch (spu.vertex_info.attrib[slot].interp_mode) {
- case INTERP_CONSTANT:
- result[QUAD_TOP_LEFT] =
- result[QUAD_TOP_RIGHT] =
- result[QUAD_BOTTOM_LEFT] =
- result[QUAD_BOTTOM_RIGHT] = setup.coef[slot].a0;
- break;
- case INTERP_LINEAR:
- {
- vector float dadx = setup.coef[slot].dadx;
- vector float dady = setup.coef[slot].dady;
- vector float topLeft =
- spu_add(setup.coef[slot].a0,
- 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);
- }
- break;
- case INTERP_PERSPECTIVE:
- {
- vector float dadx = setup.coef[slot].dadx;
- vector float dady = setup.coef[slot].dady;
- vector float topLeft =
- spu_add(setup.coef[slot].a0,
- spu_add(spu_mul(spu_splats(x), dadx),
- spu_mul(spu_splats(y), dady)));
-
- vector float wInv = spu_re(w); /* 1.0 / w */
-
- result[QUAD_TOP_LEFT] = spu_mul(topLeft, wInv);
- result[QUAD_TOP_RIGHT] = spu_mul(spu_add(topLeft, dadx), wInv);
- result[QUAD_BOTTOM_LEFT] = spu_mul(spu_add(topLeft, dady), wInv);
- result[QUAD_BOTTOM_RIGHT] = spu_mul(spu_add(spu_add(topLeft, dadx), dady), wInv);
- }
- break;
- case INTERP_POS:
- case INTERP_NONE:
- break;
- default:
- ASSERT(0);
- }
+ return spu_splats(spu_extract(v, CHAN0));
}
-
-/**
- * As above, but return 4 vectors in SOA format.
- * XXX this will all be re-written someday.
- */
-static INLINE void
-eval_coeff_soa(uint slot, float x, float y, vector float w, vector float result[4])
+static INLINE vector float
+splaty(vector float v)
{
- eval_coeff(slot, x, y, w, result);
- _transpose_matrix4x4(result, result);
+ return spu_splats(spu_extract(v, CHAN1));
}
+static INLINE vector float
+splatz(vector float v)
+{
+ return spu_splats(spu_extract(v, CHAN2));
+}
-/** Evalute coefficients to get Z for four pixels in a quad */
static INLINE vector float
-eval_z(float x, float y)
+splatw(vector float v)
{
- const uint slot = 0;
- const float dzdx = spu_extract(setup.coef[slot].dadx, 2);
- const float dzdy = spu_extract(setup.coef[slot].dady, 2);
- const float topLeft = spu_extract(setup.coef[slot].a0, 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);
+ return spu_splats(spu_extract(v, CHAN3));
}
-/** Evalute coefficients to get W for four pixels in a quad */
-static INLINE vector float
-eval_w(float x, float y)
+/**
+ * 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 void
+eval_inputs(float x, float y, vector float *fragZ, vector float fragInputs[])
{
- const uint slot = 0;
- const float dwdx = spu_extract(setup.coef[slot].dadx, 3);
- const float dwdy = spu_extract(setup.coef[slot].dady, 3);
- const float topLeft = spu_extract(setup.coef[slot].a0, 3) + x * dwdx + y * dwdy;
- const vector float topLeftv = spu_splats(topLeft);
- const vector float derivs = (vector float) { 0.0, dwdx, dwdy, dwdx + dwdy };
- return spu_add(topLeftv, derivs);
+ 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;
+
+ *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;
+ }
}
* Run fragment shader, execute per-fragment ops, update fb/tile.
*/
vector float inputs[4*4], outputs[2*4];
- vector float fragZ = eval_z((float) x, (float) y);
- vector float fragW = eval_w((float) x, (float) y);
vector unsigned int kill_mask;
+ vector float fragZ;
+
+ eval_inputs((float) x, (float) y, &fragZ, inputs);
- /* setup inputs */
-#if 0
- eval_coeff_soa(1, (float) x, (float) y, fragW, inputs);
-#else
- uint i;
- for (i = 0; i < spu.vertex_info.num_attribs; i++) {
- eval_coeff_soa(i+1, (float) x, (float) y, fragW, inputs + i * 4);
- }
-#endif
ASSERT(spu.fragment_program);
ASSERT(spu.fragment_ops);
}
#endif
+/* Returns the minimum of each slot of two vec_float4s as qwords.
+ * i.e. return[n] = min(q0[n],q1[n]);
+ */
+static qword
+minfq(qword q0, qword q1)
+{
+ const qword q0q1m = si_fcgt(q0, q1);
+ return si_selb(q0, q1, q0q1m);
+}
+
+/* Returns the minimum of each slot of three vec_float4s as qwords.
+ * i.e. return[n] = min(q0[n],q1[n],q2[n]);
+ */
+static qword
+min3fq(qword q0, qword q1, qword q2)
+{
+ return minfq(minfq(q0, q1), q2);
+}
+
+/* Returns the maximum of each slot of two vec_float4s as qwords.
+ * i.e. return[n] = min(q0[n],q1[n],q2[n]);
+ */
+static qword
+maxfq(qword q0, qword q1) {
+ const qword q0q1m = si_fcgt(q0, q1);
+ return si_selb(q1, q0, q0q1m);
+}
+
+/* Returns the maximum of each slot of three vec_float4s as qwords.
+ * i.e. return[n] = min(q0[n],q1[n],q2[n]);
+ */
+static qword
+max3fq(qword q0, qword q1, qword q2) {
+ return maxfq(maxfq(q0, q1), q2);
+}
/**
* Sort vertices from top to bottom.
* \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)
+setup_sort_vertices(const qword vs)
{
float area, sign;
}
#endif
- /* determine bottom to top order of vertices */
{
+ /* Load the float values for various processing... */
+ const qword f0 = (qword)(((const struct vertex_header*)si_to_ptr(vs))->data[0]);
+ const qword f1 = (qword)(((const struct vertex_header*)si_to_ptr(si_rotqbyi(vs, 4)))->data[0]);
+ const qword f2 = (qword)(((const struct vertex_header*)si_to_ptr(si_rotqbyi(vs, 8)))->data[0]);
+
+ /* Check if triangle is completely outside the tile bounds
+ * Find the min and max x and y positions of the three poits */
+ const qword minf = min3fq(f0, f1, f2);
+ const qword maxf = max3fq(f0, f1, f2);
+
+ /* Compare min and max against cliprect vals */
+ const qword maxsmins = si_shufb(maxf, minf, SHUFB4(A,B,a,b));
+ const qword outside = si_fcgt(maxsmins, si_csflt(setup.cliprect, 0));
+
+ /* Use a little magic to work out of the tri is visible or not */
+ if(si_to_uint(si_xori(si_gb(outside), 0xc))) return FALSE;
+
+ /* determine bottom to top order of vertices */
/* 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};
+ const qword sort_order_patterns[] = {
+ SHUFB4(A,B,C,C),
+ SHUFB4(C,A,B,C),
+ SHUFB4(A,C,B,C),
+ SHUFB4(B,C,A,C),
+ SHUFB4(B,A,C,C),
+ SHUFB4(C,B,A,C) };
/* 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));
+ const qword y_02_ = si_shufb(f0, f2, SHUFB4(0,B,b,C));
+ const qword y_10_ = si_shufb(f1, f0, SHUFB4(0,B,b,C));
+ const qword y_012 = si_shufb(y_02_, f1, SHUFB4(0,B,b,C));
+ const qword y_120 = si_shufb(y_10_, f2, SHUFB4(0,B,b,C));
/* Perform comparison: {y0,y1,y2} > {y1,y2,y0} */
- const vec_uint4 compare = spu_cmpgt(y_012, y_120);
+ const qword compare = si_fcgt(y_012, y_120);
/* Compress the result of the comparison into 4 bits */
- const vec_uint4 gather = spu_gather(compare);
+ const qword gather = si_gb(compare);
/* Subtract one to attain the index into the LUT. Magical. */
- const unsigned int index = spu_extract(gather, 0) - 1;
+ const unsigned int index = si_to_uint(gather) - 1;
/* Load the appropriate pattern and construct the desired vector. */
- setup.vertex_headers = (qword)spu_shuffle(vs, vs, sort_order_patterns[index]);
+ setup.vertex_headers = si_shufb(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);
+ sign = ((si_to_uint(si_cntb(gather)) == 2) ? 1.0f : -1.0f);
}
- /* Check if triangle is completely outside the tile bounds */
- if (spu_extract(setup.vmin->data[0], 1) > setup.cliprect_maxy)
- return FALSE;
- if (spu_extract(setup.vmax->data[0], 1) < setup.cliprect_miny)
- return FALSE;
- if (spu_extract(setup.vmin->data[0], 0) < setup.cliprect_minx &&
- spu_extract(setup.vmid->data[0], 0) < setup.cliprect_minx &&
- spu_extract(setup.vmax->data[0], 0) < setup.cliprect_minx)
- return FALSE;
- if (spu_extract(setup.vmin->data[0], 0) > setup.cliprect_maxx &&
- spu_extract(setup.vmid->data[0], 0) > setup.cliprect_maxx &&
- spu_extract(setup.vmax->data[0], 0) > setup.cliprect_maxx)
- return FALSE;
-
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]);
* The tile data should have already been fetched.
*/
boolean
-tri_draw(const float *v0, const float *v1, const float *v2,
+tri_draw(const qword vs,
uint tx, uint ty)
{
setup.tx = tx;
setup.ty = ty;
/* set clipping bounds to tile bounds */
- setup.cliprect_minx = tx * TILE_SIZE;
- setup.cliprect_miny = ty * TILE_SIZE;
- setup.cliprect_maxx = (tx + 1) * TILE_SIZE;
- setup.cliprect_maxy = (ty + 1) * TILE_SIZE;
-
- if (!setup_sort_vertices((struct vertex_header *) v0,
- (struct vertex_header *) v1,
- (struct vertex_header *) v2)) {
+ const qword clipbase = (qword)((vec_uint4){tx, ty});
+ const qword clipmin = si_mpyui(clipbase, TILE_SIZE);
+ const qword clipmax = si_ai(clipmin, TILE_SIZE);
+ setup.cliprect = si_shufb(clipmin, clipmax, SHUFB4(A,B,a,b));
+
+ if(!setup_sort_vertices(vs)) {
return FALSE; /* totally clipped */
}