int fb_clearscissor[4];
int fb_viewport[4];
int fb_viewportscissor[4];
- float fb_viewportcenter[2];
- float fb_viewportscale[2];
+ ALIGN(float fb_viewportcenter[4]);
+ ALIGN(float fb_viewportscale[4]);
// derived values (DPSOFTRAST_VALIDATE_DEPTHFUNC)
int fb_depthfunc;
dpsoftrast.fb_viewportscissor[1] = y5;
dpsoftrast.fb_viewportscissor[2] = x6 - x5;
dpsoftrast.fb_viewportscissor[3] = y6 - y5;
- dpsoftrast.fb_viewportcenter[0] = dpsoftrast.user.viewport[0] + 0.5f * dpsoftrast.user.viewport[2] - 0.5f;
- dpsoftrast.fb_viewportcenter[1] = dpsoftrast.fb_height - dpsoftrast.user.viewport[1] - 0.5f * dpsoftrast.user.viewport[3] - 0.5f;
- dpsoftrast.fb_viewportscale[0] = 0.5f * dpsoftrast.user.viewport[2];
- dpsoftrast.fb_viewportscale[1] = -0.5f * dpsoftrast.user.viewport[3];
+ dpsoftrast.fb_viewportcenter[1] = dpsoftrast.user.viewport[0] + 0.5f * dpsoftrast.user.viewport[2] - 0.5f;
+ dpsoftrast.fb_viewportcenter[2] = dpsoftrast.fb_height - dpsoftrast.user.viewport[1] - 0.5f * dpsoftrast.user.viewport[3] - 0.5f;
+ dpsoftrast.fb_viewportcenter[3] = 0.5f;
+ dpsoftrast.fb_viewportcenter[0] = 0.0f;
+ dpsoftrast.fb_viewportscale[1] = 0.5f * dpsoftrast.user.viewport[2];
+ dpsoftrast.fb_viewportscale[2] = -0.5f * dpsoftrast.user.viewport[3];
+ dpsoftrast.fb_viewportscale[3] = 0.5f;
+ dpsoftrast.fb_viewportscale[0] = 0.0f;
}
void DPSOFTRAST_RecalcDepthFunc(void)
memcpy(out4f, in4f, numitems * sizeof(float[4]));
}
+#ifdef SSE2_PRESENT
+__m128 DPSOFTRAST_Draw_ProjectVertex(__m128 v)
+{
+ __m128 viewportcenter = _mm_load_ps(dpsoftrast.fb_viewportcenter), viewportscale = _mm_load_ps(dpsoftrast.fb_viewportscale);
+ __m128 w = _mm_rcp_ss(_mm_shuffle_ps(v, v, _MM_SHUFFLE(2, 1, 0, 3)));
+ v = _mm_add_ps(viewportcenter, _mm_mul_ps(_mm_mul_ps(viewportscale, w), _mm_shuffle_ps(w, w, _MM_SHUFFLE(0, 0, 0, 0))));
+ v = _mm_move_ss(v, w);
+ v = _mm_shuffle_ps(v, v, _MM_SHUFFLE(0, 3, 2, 1));
+ return v;
+}
+#endif
+
void DPSOFTRAST_Draw_ProjectVertices(float *out4f, const float *in4f, int numitems)
{
+#ifdef SSE2_PRESENT
// NOTE: this is used both as a whole mesh transform function and a
// per-triangle transform function (for clipped triangles), accordingly
// it should not crash on divide by 0 but the result of divide by 0 is
// unimportant...
// TODO: SIMD
int i;
- float w;
- float viewportcenter[4];
- float viewportscale[4];
- viewportscale[0] = dpsoftrast.fb_viewportscale[0];
- viewportscale[1] = dpsoftrast.fb_viewportscale[1];
- viewportscale[2] = 0.5f;
- viewportscale[3] = 0.0f;
- viewportcenter[0] = dpsoftrast.fb_viewportcenter[0];
- viewportcenter[1] = dpsoftrast.fb_viewportcenter[1];
- viewportcenter[2] = 0.5f;
- viewportcenter[3] = 0.0f;
+ __m128 viewportcenter = _mm_load_ps(dpsoftrast.fb_viewportcenter), viewportscale = _mm_load_ps(dpsoftrast.fb_viewportscale);
for (i = 0;i < numitems;i++)
{
- if (!in4f[3])
- {
- out4f[0] = 0.0f;
- out4f[1] = 0.0f;
- out4f[2] = 0.0f;
- out4f[3] = 0.0f;
- continue;
- }
- w = 1.0f / in4f[3];
- out4f[0] = viewportcenter[0] + viewportscale[0] * in4f[0] * w;
- out4f[1] = viewportcenter[1] + viewportscale[1] * in4f[1] * w;
- out4f[2] = viewportcenter[2] + viewportscale[2] * in4f[2] * w;
- out4f[3] = viewportcenter[3] + viewportscale[3] * in4f[3] * w;
- out4f[3] = w;
+ __m128 v = _mm_load_ps(in4f), w = _mm_rcp_ss(_mm_shuffle_ps(v, v, _MM_SHUFFLE(2, 1, 0, 3)));
+ v = _mm_add_ps(viewportcenter, _mm_mul_ps(_mm_mul_ps(viewportscale, w), _mm_shuffle_ps(w, w, _MM_SHUFFLE(0, 0, 0, 0))));
+ v = _mm_move_ss(v, w);
+ _mm_store_ps(out4f, _mm_shuffle_ps(v, v, _MM_SHUFFLE(0, 3, 2, 1)));
in4f += 4;
out4f += 4;
}
+#endif
}
void DPSOFTRAST_Draw_DebugEdgePoints(const float *screen0, const float *screen1)
endsubtc = _mm_cvtps_epi32(_mm_mul_ps(endtc, _mm_set1_ps(65536.0f)));
if (filter)
{
- __m128i tcrange = _mm_srai_epi32(_mm_unpacklo_epi64(subtc, endsubtc), 16);
+ __m128i tcrange = _mm_srai_epi32(_mm_unpacklo_epi64(subtc, _mm_add_epi32(endsubtc, substep)), 16);
if (_mm_movemask_epi8(_mm_andnot_si128(_mm_cmplt_epi32(tcrange, _mm_setzero_si128()), _mm_cmplt_epi32(tcrange, tcmask))) == 0xFFFF)
{
subtc = _mm_unpacklo_epi64(subtc, _mm_add_epi32(subtc, substep));
int k;
int y;
int e[3];
- ALIGN(int screeny[4]);
+ __m128i screeny;
int starty, endy;
- int screenyless[4];
int numpoints;
- int clipflags;
int edge0p;
int edge0n;
int edge1p;
int edge1n;
int startx;
int endx;
- float mip_edge0tc[2];
- float mip_edge1tc[2];
- float mip_edge0xy[2];
- float mip_edge1xy[2];
- float mip_edge0xymul;
- float mip_edge1xymul;
- float mip_edge0mip;
- float mip_edge1mip;
float mipdensity;
unsigned char mip[DPSOFTRAST_MAXTEXTUREUNITS];
- float startxf;
- float endxf;
- float edge0ylerp;
- float edge0yilerp;
- float edge1ylerp;
- float edge1yilerp;
- float edge0xf;
- float edge1xf;
- float spanilength;
- float startxlerp;
- float yc;
- float frac;
- float ifrac;
- //float trianglearea2;
- __m128 triangleedge[2];
- __m128 trianglenormal;
- ALIGN(float clipdist[4]);
- ALIGN(float clipped[DPSOFTRAST_ARRAY_TOTAL][4][4]);
- ALIGN(float screen[4][4]);
+ __m128 triangleedge[2], triangleorigin, trianglenormal;
+ __m128 mipedgescale;
+ float clipdist[4];
+ __m128 clipped[DPSOFTRAST_ARRAY_TOTAL][4];
+ __m128 screen[4];
__m128 proj[DPSOFTRAST_ARRAY_TOTAL][4];
DPSOFTRAST_Texture *texture;
DPSOFTRAST_State_Draw_Span *span;
e[1] = i*3+1;
e[2] = i*3+2;
}
- {
- __m128 v1 = _mm_load_ps(&dpsoftrast.draw.post_array4f[DPSOFTRAST_ARRAY_POSITION][e[1]*4]);
- triangleedge[0] = _mm_sub_ps(_mm_load_ps(&dpsoftrast.draw.post_array4f[DPSOFTRAST_ARRAY_POSITION][e[0]*4]), v1);
- triangleedge[1] = _mm_sub_ps(_mm_load_ps(&dpsoftrast.draw.post_array4f[DPSOFTRAST_ARRAY_POSITION][e[2]*4]), v1);
- }
+ triangleorigin = _mm_load_ps(&dpsoftrast.draw.post_array4f[DPSOFTRAST_ARRAY_POSITION][e[1]*4]);
+ triangleedge[0] = _mm_sub_ps(_mm_load_ps(&dpsoftrast.draw.post_array4f[DPSOFTRAST_ARRAY_POSITION][e[0]*4]), triangleorigin);
+ triangleedge[1] = _mm_sub_ps(_mm_load_ps(&dpsoftrast.draw.post_array4f[DPSOFTRAST_ARRAY_POSITION][e[2]*4]), triangleorigin);
// store normal in 2, 0, 1 order instead of 0, 1, 2 as it requires fewer shuffles and leaves z component accessible as scalar
trianglenormal = _mm_sub_ps(_mm_mul_ps(triangleedge[0], _mm_shuffle_ps(triangleedge[1], triangleedge[1], _MM_SHUFFLE(3, 0, 2, 1))),
_mm_mul_ps(_mm_shuffle_ps(triangleedge[0], triangleedge[0], _MM_SHUFFLE(3, 0, 2, 1)), triangleedge[1]));
continue;
#endif
- // calculate distance from nearplane
- clipdist[0] = dpsoftrast.draw.post_array4f[DPSOFTRAST_ARRAY_POSITION][e[0]*4+2] + 1.0f;
- clipdist[1] = dpsoftrast.draw.post_array4f[DPSOFTRAST_ARRAY_POSITION][e[1]*4+2] + 1.0f;
- clipdist[2] = dpsoftrast.draw.post_array4f[DPSOFTRAST_ARRAY_POSITION][e[2]*4+2] + 1.0f;
- clipflags = 0;
- if (clipdist[0] < 0.0f)
- clipflags |= 1;
- if (clipdist[1] < 0.0f)
- clipflags |= 2;
- if (clipdist[2] < 0.0f)
- clipflags |= 4;
- // clip triangle if necessary
- switch(clipflags)
- {
- case 0: /*000*/
- // triangle is entirely in front of nearplane
-
// macros for clipping vertices
-#define CLIPPEDVERTEXLERP(k,p1,p2) \
- frac = clipdist[p1] / (clipdist[p1] - clipdist[p2]);\
- ifrac = 1.0f - frac;\
+#define CLIPPEDVERTEXLERP(k,p1,p2) { \
+ __m128 frac = _mm_set1_ps(clipdist[p1] / (clipdist[p1] - clipdist[p2]));\
for (j = 0;j < DPSOFTRAST_ARRAY_TOTAL;j++)\
{\
/*if (arraymask[j])*/\
{\
- clipped[j][k][0] = dpsoftrast.draw.post_array4f[j][e[p1]*4+0]*ifrac+dpsoftrast.draw.post_array4f[j][e[p2]*4+0]*frac;\
- clipped[j][k][1] = dpsoftrast.draw.post_array4f[j][e[p1]*4+1]*ifrac+dpsoftrast.draw.post_array4f[j][e[p2]*4+1]*frac;\
- clipped[j][k][2] = dpsoftrast.draw.post_array4f[j][e[p1]*4+2]*ifrac+dpsoftrast.draw.post_array4f[j][e[p2]*4+2]*frac;\
- clipped[j][k][3] = dpsoftrast.draw.post_array4f[j][e[p1]*4+3]*ifrac+dpsoftrast.draw.post_array4f[j][e[p2]*4+3]*frac;\
+ __m128 v1 = _mm_load_ps(&dpsoftrast.draw.post_array4f[j][e[p1]*4]), v2 = _mm_load_ps(&dpsoftrast.draw.post_array4f[j][e[p2]*4]); \
+ clipped[j][k] = _mm_add_ps(v1, _mm_mul_ps(_mm_sub_ps(v2, v1), frac)); \
}\
}\
- DPSOFTRAST_Draw_ProjectVertices(screen[k], clipped[DPSOFTRAST_ARRAY_POSITION][k], 1)
+ screen[k] = DPSOFTRAST_Draw_ProjectVertex(clipped[DPSOFTRAST_ARRAY_POSITION][k]); \
+ }
#define CLIPPEDVERTEXCOPY(k,p1) \
for (j = 0;j < DPSOFTRAST_ARRAY_TOTAL;j++)\
{\
/*if (arraymask[j])*/\
{\
- clipped[j][k][0] = dpsoftrast.draw.post_array4f[j][e[p1]*4+0];\
- clipped[j][k][1] = dpsoftrast.draw.post_array4f[j][e[p1]*4+1];\
- clipped[j][k][2] = dpsoftrast.draw.post_array4f[j][e[p1]*4+2];\
- clipped[j][k][3] = dpsoftrast.draw.post_array4f[j][e[p1]*4+3];\
+ clipped[j][k] = _mm_load_ps(&dpsoftrast.draw.post_array4f[j][e[p1]*4]); \
}\
}\
- screen[k][0] = dpsoftrast.draw.screencoord4f[e[p1]*4+0];\
- screen[k][1] = dpsoftrast.draw.screencoord4f[e[p1]*4+1];\
- screen[k][2] = dpsoftrast.draw.screencoord4f[e[p1]*4+2];\
- screen[k][3] = dpsoftrast.draw.screencoord4f[e[p1]*4+3];
+ screen[k] = _mm_load_ps(&dpsoftrast.draw.screencoord4f[e[p1]*4]);
- CLIPPEDVERTEXCOPY(0,0);
- CLIPPEDVERTEXCOPY(1,1);
- CLIPPEDVERTEXCOPY(2,2);
- numpoints = 3;
- break;
- case 1: /*100*/
+ // calculate distance from nearplane
+ clipdist[0] = dpsoftrast.draw.post_array4f[DPSOFTRAST_ARRAY_POSITION][e[0]*4+2] + 1.0f;
+ clipdist[1] = dpsoftrast.draw.post_array4f[DPSOFTRAST_ARRAY_POSITION][e[1]*4+2] + 1.0f;
+ clipdist[2] = dpsoftrast.draw.post_array4f[DPSOFTRAST_ARRAY_POSITION][e[2]*4+2] + 1.0f;
+ if (clipdist[0] >= 0.0f)
+ {
+ if (clipdist[1] >= 0.0f)
+ {
+ CLIPPEDVERTEXCOPY(0,0);
+ CLIPPEDVERTEXCOPY(1,1);
+ if (clipdist[2] >= 0.0f)
+ {
+ // triangle is entirely in front of nearplane
+ CLIPPEDVERTEXCOPY(2,2);
+ numpoints = 3;
+ }
+ else
+ {
+ CLIPPEDVERTEXLERP(2,1,2);
+ CLIPPEDVERTEXLERP(3,2,0);
+ numpoints = 4;
+ }
+ }
+ else
+ {
+ CLIPPEDVERTEXCOPY(0,0);
+ CLIPPEDVERTEXLERP(1,0,1);
+ if (clipdist[2] >= 0.0f)
+ {
+ CLIPPEDVERTEXLERP(2,1,2);
+ CLIPPEDVERTEXCOPY(3,2);
+ numpoints = 4;
+ }
+ else
+ {
+ CLIPPEDVERTEXLERP(2,2,0);
+ numpoints = 3;
+ }
+ }
+ }
+ else if (clipdist[1] >= 0.0f)
+ {
CLIPPEDVERTEXLERP(0,0,1);
CLIPPEDVERTEXCOPY(1,1);
- CLIPPEDVERTEXCOPY(2,2);
- CLIPPEDVERTEXLERP(3,2,0);
- numpoints = 4;
- break;
- case 2: /*010*/
- CLIPPEDVERTEXCOPY(0,0);
- CLIPPEDVERTEXLERP(1,0,1);
- CLIPPEDVERTEXLERP(2,1,2);
- CLIPPEDVERTEXCOPY(3,2);
- numpoints = 4;
- break;
- case 3: /*110*/
+ if (clipdist[2] >= 0.0f)
+ {
+ CLIPPEDVERTEXCOPY(2,2);
+ CLIPPEDVERTEXLERP(3,2,0);
+ numpoints = 4;
+ }
+ else
+ {
+ CLIPPEDVERTEXLERP(2,1,2);
+ numpoints = 3;
+ }
+ }
+ else if (clipdist[2] >= 0.0f)
+ {
CLIPPEDVERTEXLERP(0,1,2);
CLIPPEDVERTEXCOPY(1,2);
CLIPPEDVERTEXLERP(2,2,0);
numpoints = 3;
- break;
- case 4: /*001*/
- CLIPPEDVERTEXCOPY(0,0);
- CLIPPEDVERTEXCOPY(1,1);
- CLIPPEDVERTEXLERP(2,1,2);
- CLIPPEDVERTEXLERP(3,2,0);
- numpoints = 4;
- break;
- case 5: /*101*/
- CLIPPEDVERTEXLERP(0,0,1);
- CLIPPEDVERTEXCOPY(1,1);
- CLIPPEDVERTEXLERP(2,1,2);
- numpoints = 3;
- break;
- case 6: /*011*/
- CLIPPEDVERTEXCOPY(0,0);
- CLIPPEDVERTEXLERP(1,0,1);
- CLIPPEDVERTEXLERP(2,2,0);
- numpoints = 3;
- break;
- case 7: /*111*/
- // triangle is entirely behind nearplane
- continue;
}
+ else continue; // triangle is entirely behind nearpla
{
// calculate integer y coords for triangle points
- __m128i screeni = _mm_packs_epi32(_mm_cvttps_epi32(_mm_shuffle_ps(_mm_load_ps(screen[0]), _mm_load_ps(screen[1]), _MM_SHUFFLE(1, 0, 1, 0))),
- _mm_cvttps_epi32(_mm_shuffle_ps(_mm_load_ps(screen[2]), _mm_load_ps(screen[3]), _MM_SHUFFLE(1, 0, 1, 0)))),
+ __m128i screeni = _mm_packs_epi32(_mm_cvttps_epi32(_mm_shuffle_ps(screen[0], screen[1], _MM_SHUFFLE(1, 0, 1, 0))),
+ _mm_cvttps_epi32(_mm_shuffle_ps(screen[2], screen[3], _MM_SHUFFLE(1, 0, 1, 0)))),
screenir, screenmin, screenmax;
if (numpoints <= 3) screeni = _mm_shuffle_epi32(screeni, _MM_SHUFFLE(2, 2, 1, 0));
screenir = _mm_shuffle_epi32(screeni, _MM_SHUFFLE(1, 0, 3, 2)),
}
starty = _mm_extract_epi16(screenmin, 1);
endy = _mm_extract_epi16(screenmax, 1)+1;
- _mm_store_si128((__m128i *)screeny, _mm_srai_epi32(screeni, 16));
+ screeny = _mm_srai_epi32(screeni, 16);
}
// okay, this triangle is going to produce spans, we'd better project
{
for (k = 0;k < numpoints;k++)
{
- proj[j][k] = _mm_mul_ps(_mm_load_ps(clipped[j][k]), _mm_set1_ps(screen[k][3]));
+ proj[j][k] = _mm_mul_ps(clipped[j][k], _mm_shuffle_ps(screen[k], screen[k], _MM_SHUFFLE(3, 3, 3, 3)));
}
}
}
// adjust texture LOD by texture density, in the simplest way possible...
- mip_edge0xy[0] = screen[0][0] - screen[1][0];
- mip_edge0xy[1] = screen[0][1] - screen[1][1];
- mip_edge1xy[0] = screen[2][0] - screen[1][0];
- mip_edge1xy[1] = screen[2][1] - screen[1][1];
- mip_edge0xymul = 1.0f / (mip_edge0xy[0]*mip_edge0xy[0]+mip_edge0xy[1]*mip_edge0xy[1]);
- mip_edge1xymul = 1.0f / (mip_edge1xy[0]*mip_edge1xy[0]+mip_edge1xy[1]*mip_edge1xy[1]);
+ mipedgescale = _mm_sub_ps(_mm_shuffle_ps(screen[0], screen[2], _MM_SHUFFLE(1, 0, 1, 0)), _mm_shuffle_ps(screen[1], screen[1], _MM_SHUFFLE(1, 0, 1, 0)));
+ mipedgescale = _mm_mul_ps(mipedgescale, mipedgescale);
+ mipedgescale = _mm_rcp_ps(_mm_add_ps(mipedgescale, _mm_shuffle_ps(mipedgescale, mipedgescale, _MM_SHUFFLE(2, 3, 0, 1))));
for (j = 0;j < DPSOFTRAST_MAXTEXTUREUNITS;j++)
{
texture = dpsoftrast.texbound[j];
if (texture)
{
+ __m128 mipedgetc;
if (texture->filter <= DPSOFTRAST_TEXTURE_FILTER_LINEAR)
{
mip[j] = 0;
continue;
}
k = DPSOFTRAST_ShaderModeTable[dpsoftrast.shader_mode].lodarrayindex;
- mip_edge0tc[0] = (clipped[k][0][0] - clipped[k][1][0]) * texture->mipmap[0][2];
- mip_edge0tc[1] = (clipped[k][0][1] - clipped[k][1][1]) * texture->mipmap[0][3];
- mip_edge1tc[0] = (clipped[k][2][0] - clipped[k][1][0]) * texture->mipmap[0][2];
- mip_edge1tc[1] = (clipped[k][2][1] - clipped[k][1][1]) * texture->mipmap[0][3];
- mip_edge0mip = (mip_edge0tc[0]*mip_edge0tc[0]+mip_edge0tc[1]*mip_edge0tc[1]) * mip_edge0xymul;
- mip_edge1mip = (mip_edge1tc[0]*mip_edge1tc[0]+mip_edge1tc[1]*mip_edge1tc[1]) * mip_edge1xymul;
+ mipedgetc = _mm_sub_ps(_mm_shuffle_ps(clipped[k][0], clipped[k][2], _MM_SHUFFLE(1, 0, 1, 0)),
+ _mm_shuffle_ps(clipped[k][1], clipped[k][1], _MM_SHUFFLE(1, 0, 1, 0)));
+ mipedgetc = _mm_mul_ps(mipedgetc, _mm_cvtepi32_ps(_mm_shuffle_epi32(_mm_loadl_epi64((const __m128i *)&texture->mipmap[0][2]), _MM_SHUFFLE(1, 0, 1, 0))));
+ mipedgetc = _mm_mul_ps(mipedgetc, mipedgetc);
+ mipedgetc = _mm_add_ps(mipedgetc, _mm_shuffle_ps(mipedgetc, mipedgetc, _MM_SHUFFLE(2, 3, 0, 1)));
+ mipedgetc = _mm_mul_ps(mipedgetc, mipedgescale);
+ mipedgetc = _mm_min_ss(mipedgetc, _mm_shuffle_ps(mipedgetc, mipedgetc, _MM_SHUFFLE(2, 2, 2, 2)));
+ _mm_store_ss(&mipdensity, mipedgetc);
// this will be multiplied in the texturing routine by the texture resolution
- mipdensity = mip_edge0mip < mip_edge1mip ? mip_edge0mip : mip_edge1mip;
y = (int)(log(mipdensity)/log(2.0f));
if (y < 0)
y = 0;
}
// iterate potential spans
// TODO: optimize? if we figured out the edge order beforehand, this
- // could do loops over the edges in the proper order rather than
- // selecting them for each span
+ // could do loops over the edges in the proper order rather than
+ // selecting them for each span
// TODO: optimize? the edges could have data slopes calculated
// TODO: optimize? the data slopes could be calculated as a plane
- // (2D slopes) to avoid any interpolation along edges at all
- for (y = starty;y < endy;y++)
+ // (2D slopes) to avoid any interpolation along edges at all
+ for (y = starty+1;y < endy;)
{
- // get center of pixel y
- yc = y;
- // do the compares all at once
- screenyless[0] = y <= screeny[0];
- screenyless[1] = y <= screeny[1];
- screenyless[2] = y <= screeny[2];
- screenyless[3] = y <= screeny[3];
+ int nexty = -1;
+ __m128 edge0lerp, edge1lerp, edge0scale, edge1scale;
+ __m128i screenycc = _mm_cmpgt_epi32(_mm_set1_epi32(y), screeny);
+ int screenymask = _mm_movemask_epi8(screenycc);
if (numpoints == 4)
{
- switch(screenyless[0] + screenyless[1] * 2 + screenyless[2] * 4 + screenyless[3] * 8)
+ switch(screenymask)
{
- case 0: /*0000*/ continue;
- case 1: /*1000*/ edge0p = 3;edge0n = 0;edge1p = 0;edge1n = 1;break;
- case 2: /*0100*/ edge0p = 0;edge0n = 1;edge1p = 1;edge1n = 2;break;
- case 3: /*1100*/ edge0p = 3;edge0n = 0;edge1p = 1;edge1n = 2;break;
- case 4: /*0010*/ edge0p = 1;edge0n = 2;edge1p = 2;edge1n = 3;break;
- case 5: /*1010*/ edge0p = 1;edge0n = 2;edge1p = 2;edge1n = 3;break; // concave - nonsense
- case 6: /*0110*/ edge0p = 0;edge0n = 1;edge1p = 2;edge1n = 3;break;
- case 7: /*1110*/ edge0p = 3;edge0n = 0;edge1p = 2;edge1n = 3;break;
- case 8: /*0001*/ edge0p = 2;edge0n = 3;edge1p = 3;edge1n = 0;break;
- case 9: /*1001*/ edge0p = 2;edge0n = 3;edge1p = 0;edge1n = 1;break;
- case 10: /*0101*/ edge0p = 2;edge0n = 3;edge1p = 1;edge1n = 2;break; // concave - nonsense
- case 11: /*1101*/ edge0p = 2;edge0n = 3;edge1p = 1;edge1n = 2;break;
- case 12: /*0011*/ edge0p = 1;edge0n = 2;edge1p = 3;edge1n = 0;break;
- case 13: /*1011*/ edge0p = 1;edge0n = 2;edge1p = 0;edge1n = 1;break;
- case 14: /*0111*/ edge0p = 0;edge0n = 1;edge1p = 3;edge1n = 0;break;
- case 15: /*1111*/ continue;
+ default:
+ case 0xFFFF: /*0000*/ y++; continue;
+ case 0xFFF0: /*1000*/ edge0p = 3;edge0n = 0;edge1p = 1;edge1n = 0;break;
+ case 0xFF0F: /*0100*/ edge0p = 0;edge0n = 1;edge1p = 2;edge1n = 1;break;
+ case 0xFF00: /*1100*/ edge0p = 3;edge0n = 0;edge1p = 2;edge1n = 1;break;
+ case 0xF0FF: /*0010*/ edge0p = 1;edge0n = 2;edge1p = 3;edge1n = 2;break;
+ case 0xF0F0: /*1010*/ edge0p = 1;edge0n = 2;edge1p = 3;edge1n = 2;break; // concave - nonsense
+ case 0xF00F: /*0110*/ edge0p = 0;edge0n = 1;edge1p = 3;edge1n = 2;break;
+ case 0xF000: /*1110*/ edge0p = 3;edge0n = 0;edge1p = 3;edge1n = 2;break;
+ case 0x0FFF: /*0001*/ edge0p = 2;edge0n = 3;edge1p = 0;edge1n = 3;break;
+ case 0x0FF0: /*1001*/ edge0p = 2;edge0n = 3;edge1p = 1;edge1n = 0;break;
+ case 0x0F0F: /*0101*/ edge0p = 2;edge0n = 3;edge1p = 1;edge1n = 2;break; // concave - nonsense
+ case 0x0F00: /*1101*/ edge0p = 2;edge0n = 3;edge1p = 2;edge1n = 1;break;
+ case 0x00FF: /*0011*/ edge0p = 1;edge0n = 2;edge1p = 0;edge1n = 3;break;
+ case 0x00F0: /*1011*/ edge0p = 1;edge0n = 2;edge1p = 1;edge1n = 0;break;
+ case 0x000F: /*0111*/ edge0p = 0;edge0n = 1;edge1p = 0;edge1n = 3;break;
+ case 0x0000: /*1111*/ y++; continue;
}
}
else
{
- switch(screenyless[0] + screenyless[1] * 2 + screenyless[2] * 4)
+ switch(screenymask)
{
- case 0: /*000*/ continue;
- case 1: /*100*/ edge0p = 2;edge0n = 0;edge1p = 0;edge1n = 1;break;
- case 2: /*010*/ edge0p = 0;edge0n = 1;edge1p = 1;edge1n = 2;break;
- case 3: /*110*/ edge0p = 2;edge0n = 0;edge1p = 1;edge1n = 2;break;
- case 4: /*001*/ edge0p = 1;edge0n = 2;edge1p = 2;edge1n = 0;break;
- case 5: /*101*/ edge0p = 1;edge0n = 2;edge1p = 0;edge1n = 1;break;
- case 6: /*011*/ edge0p = 0;edge0n = 1;edge1p = 2;edge1n = 0;break;
- case 7: /*111*/ continue;
+ default:
+ case 0xFFFF: /*000*/ y++; continue;
+ case 0xFFF0: /*100*/ edge0p = 2;edge0n = 0;edge1p = 1;edge1n = 0;break;
+ case 0xFF0F: /*010*/ edge0p = 0;edge0n = 1;edge1p = 2;edge1n = 1;break;
+ case 0xFF00: /*110*/ edge0p = 2;edge0n = 0;edge1p = 2;edge1n = 1;break;
+ case 0x00FF: /*001*/ edge0p = 1;edge0n = 2;edge1p = 0;edge1n = 2;break;
+ case 0x00F0: /*101*/ edge0p = 1;edge0n = 2;edge1p = 1;edge1n = 0;break;
+ case 0x000F: /*011*/ edge0p = 0;edge0n = 1;edge1p = 0;edge1n = 2;break;
+ case 0x0000: /*111*/ y++; continue;
}
}
-#if 0
- {
- int foundedges = 0;
- int cedge0p = 0;
- int cedge0n = 0;
- int cedge1p = 0;
- int cedge1n = 0;
- for (j = 0, k = numpoints-1;j < numpoints;k = j, j++)
+ screenycc = _mm_max_epi16(_mm_srli_epi16(screenycc, 1), screeny);
+ screenycc = _mm_min_epi16(screenycc, _mm_shuffle_epi32(screenycc, _MM_SHUFFLE(1, 0, 3, 2)));
+ screenycc = _mm_min_epi16(screenycc, _mm_shuffle_epi32(screenycc, _MM_SHUFFLE(2, 3, 0, 1)));
+ nexty = _mm_extract_epi16(screenycc, 0);
+ if(nexty >= endy) nexty = endy-1;
+ if (_mm_ucomigt_ss(_mm_max_ps(screen[edge0n], screen[edge0p]), _mm_min_ps(screen[edge1n], screen[edge1p])))
{
- if (screenyless[k] && !screenyless[j])
- {
- cedge1p = k;
- cedge1n = j;
- foundedges |= 1;
- }
- else if (screenyless[j] && !screenyless[k])
- {
- cedge0p = k;
- cedge0n = j;
- foundedges |= 2;
- }
- }
- if (foundedges != 3)
- continue;
- if (cedge0p != edge0p || cedge0n != edge0n || cedge1p != edge1p || cedge1n != edge1n)
+ int tmp = edge0n;
+ edge0n = edge1n;
+ edge1n = tmp;
+ tmp = edge0p;
+ edge0p = edge1p;
+ edge1p = tmp;
+ }
+ edge0lerp = _mm_shuffle_ps(screen[edge0p], screen[edge0p], _MM_SHUFFLE(1, 1, 1, 1));
+ edge0scale = _mm_div_ss(_mm_set1_ps(1.0f), _mm_sub_ss(_mm_shuffle_ps(screen[edge0n], screen[edge0n], _MM_SHUFFLE(1, 1, 1, 1)), edge0lerp));
+ edge0scale = _mm_shuffle_ps(edge0scale, edge0scale, _MM_SHUFFLE(0, 0, 0, 0));
+ edge0lerp = _mm_mul_ps(_mm_sub_ps(_mm_set1_ps(y), edge0lerp), edge0scale);
+ edge1lerp = _mm_shuffle_ps(screen[edge1p], screen[edge1p], _MM_SHUFFLE(1, 1, 1, 1));
+ edge1scale = _mm_div_ss(_mm_set1_ps(1.0f), _mm_sub_ss(_mm_shuffle_ps(screen[edge1n], screen[edge1n], _MM_SHUFFLE(1, 1, 1, 1)), edge1lerp));
+ edge1scale = _mm_shuffle_ps(edge1scale, edge1scale, _MM_SHUFFLE(0, 0, 0, 0));
+ edge1lerp = _mm_mul_ps(_mm_sub_ps(_mm_set1_ps(y), edge1lerp), edge1scale);
+ for(; y <= nexty; y++, edge0lerp = _mm_add_ps(edge0lerp, edge0scale), edge1lerp = _mm_add_ps(edge1lerp, edge1scale))
{
- if (numpoints == 4)
- printf("case %i%i%i%i is broken %i %i %i %i != %i %i %i %i\n", screenyless[0], screenyless[1], screenyless[2], screenyless[3], cedge0p, cedge0n, cedge1p, cedge1n, edge0p, edge0n, edge1p, edge1n);
- else
- printf("case %i%i%i is broken %i %i %i %i != %i %i %i %i\n", screenyless[0], screenyless[1], screenyless[2], cedge0p, cedge0n, cedge1p, cedge1n, edge0p, edge0n, edge1p, edge1n);
- }
- }
+ __m128 data0, data1, spanilength, startxlerp;
+#if 0
+ if (_mm_ucomilt_ss(edge0lerp, _mm_setzero_ps()) || _mm_ucomilt_ss(edge1lerp, _mm_setzero_ps()) ||
+ _mm_ucomigt_ss(edge0lerp, _mm_set1_ps(1)) || _mm_ucomigt_ss(edge1lerp, _mm_set1_ps(1)))
+ continue;
#endif
- edge0ylerp = (yc - screen[edge0p][1]) / (screen[edge0n][1] - screen[edge0p][1]);
- edge1ylerp = (yc - screen[edge1p][1]) / (screen[edge1n][1] - screen[edge1p][1]);
- if (edge0ylerp < 0 || edge0ylerp > 1 || edge1ylerp < 0 || edge1ylerp > 1)
- continue;
- edge0yilerp = 1.0f - edge0ylerp;
- edge1yilerp = 1.0f - edge1ylerp;
- edge0xf = screen[edge0p][0] * edge0yilerp + screen[edge0n][0] * edge0ylerp;
- edge1xf = screen[edge1p][0] * edge1yilerp + screen[edge1n][0] * edge1ylerp;
- if (edge0xf < edge1xf)
- {
- startxf = edge0xf;
- endxf = edge1xf;
- }
- else
- {
- startxf = edge1xf;
- endxf = edge0xf;
- }
- startx = (int)ceil(startxf);
- endx = (int)ceil(endxf);
- if (startx < 0)
- startx = 0;
- if (endx > width)
- endx = width;
- if (startx >= endx)
- continue;
- if (startxf > startx || endxf < endx-1) { printf("%s:%i X wrong (%i to %i is outside %f to %f)\n", __FILE__, __LINE__, startx, endx, startxf, endxf); }
- spanilength = 1.0f / (endxf - startxf);
- startxlerp = startx - startxf;
- span = &dpsoftrast.draw.spanqueue[dpsoftrast.draw.numspans++];
- memcpy(span->mip, mip, sizeof(span->mip));
- span->start = y * width + startx;
- span->length = endx - startx;
- {
- __m128 edge0ylerpm = _mm_set1_ps(edge0ylerp), edge0yilerpm = _mm_set1_ps(edge0yilerp),
- edge1ylerpm = _mm_set1_ps(edge1ylerp), edge1yilerpm = _mm_set1_ps(edge1yilerp),
- spanilengthm = _mm_set1_ps(spanilength), startxlerpm = _mm_set1_ps(startxlerp),
- data0, data1;
+ data0 = _mm_add_ps(_mm_mul_ps(_mm_sub_ps(screen[edge0n], screen[edge0p]), edge0lerp), screen[edge0p]);
+ data1 = _mm_add_ps(_mm_mul_ps(_mm_sub_ps(screen[edge1n], screen[edge1p]), edge1lerp), screen[edge1p]);
+ startx = _mm_cvtss_si32(_mm_add_ss(data0, _mm_set1_ps(0.5f)));
+ endx = _mm_cvtss_si32(_mm_add_ss(data1, _mm_set1_ps(0.5f)));
+ if (startx < 0) startx = 0;
+ if (endx > width) endx = width;
+ if (startx >= endx) continue;
+#if 0
+ _mm_store_ss(&startxf, data0);
+ _mm_store_ss(&endxf, data1);
+ if (startxf > startx || endxf < endx-1) { printf("%s:%i X wrong (%i to %i is outside %f to %f)\n", __FILE__, __LINE__, startx, endx, startxf, endxf); }
+#endif
+ spanilength = _mm_rcp_ss(_mm_sub_ss(data1, data0));
+ spanilength = _mm_shuffle_ps(spanilength, spanilength, _MM_SHUFFLE(0, 0, 0, 0));
+ startxlerp = _mm_sub_ps(_mm_set1_ps(startx), _mm_shuffle_ps(data0, data0, _MM_SHUFFLE(0, 0, 0, 0)));
+ span = &dpsoftrast.draw.spanqueue[dpsoftrast.draw.numspans++];
+ memcpy(span->mip, mip, sizeof(span->mip));
+ span->start = y * width + startx;
+ span->length = endx - startx;
j = DPSOFTRAST_ARRAY_TOTAL;
- if (edge0xf < edge1xf)
+ data1 = _mm_mul_ps(_mm_sub_ps(data1, data0), spanilength);
+ data0 = _mm_add_ps(data0, _mm_mul_ps(data1, startxlerp));
+ _mm_store_ps(span->data[0][j], data0);
+ _mm_store_ps(span->data[1][j], data1);
+ for (j = 0;j < DPSOFTRAST_ARRAY_TOTAL;j++)
{
- data0 = _mm_add_ps(_mm_mul_ps(_mm_load_ps(screen[edge0p]), edge0yilerpm), _mm_mul_ps(_mm_load_ps(screen[edge0n]), edge0ylerpm));
- data1 = _mm_add_ps(_mm_mul_ps(_mm_load_ps(screen[edge1p]), edge1yilerpm), _mm_mul_ps(_mm_load_ps(screen[edge1n]), edge1ylerpm));
- data1 = _mm_mul_ps(_mm_sub_ps(data1, data0), spanilengthm);
- data0 = _mm_add_ps(data0, _mm_mul_ps(data1, startxlerpm));
- _mm_store_ps(span->data[0][j], data0);
- _mm_store_ps(span->data[1][j], data1);
- for (j = 0;j < DPSOFTRAST_ARRAY_TOTAL;j++)
+ //if (arraymask[j])
{
- //if (arraymask[j])
- {
- data0 = _mm_add_ps(_mm_mul_ps(proj[j][edge0p], edge0yilerpm), _mm_mul_ps(proj[j][edge0n], edge0ylerpm));
- data1 = _mm_add_ps(_mm_mul_ps(proj[j][edge1p], edge1yilerpm), _mm_mul_ps(proj[j][edge1n], edge1ylerpm));
- data1 = _mm_mul_ps(_mm_sub_ps(data1, data0), spanilengthm);
- data0 = _mm_add_ps(data0, _mm_mul_ps(data1, startxlerpm));
- _mm_store_ps(span->data[0][j], data0);
- _mm_store_ps(span->data[1][j], data1);
- }
+ data0 = _mm_add_ps(_mm_mul_ps(_mm_sub_ps(proj[j][edge0n], proj[j][edge0p]), edge0lerp), proj[j][edge0p]);
+ data1 = _mm_add_ps(_mm_mul_ps(_mm_sub_ps(proj[j][edge1n], proj[j][edge1p]), edge1lerp), proj[j][edge1p]);
+ data1 = _mm_mul_ps(_mm_sub_ps(data1, data0), spanilength);
+ data0 = _mm_add_ps(data0, _mm_mul_ps(data1, startxlerp));
+ _mm_store_ps(span->data[0][j], data0);
+ _mm_store_ps(span->data[1][j], data1);
}
}
- else
+ // to keep the shader routines from needing more than a small
+ // buffer for pixel intermediate data, we split long spans...
+ while (span->length > DPSOFTRAST_DRAW_MAXSPANLENGTH)
{
- data0 = _mm_add_ps(_mm_mul_ps(_mm_load_ps(screen[edge1p]), edge1yilerpm), _mm_mul_ps(_mm_load_ps(screen[edge1n]), edge1ylerpm));
- data1 = _mm_add_ps(_mm_mul_ps(_mm_load_ps(screen[edge0p]), edge0yilerpm), _mm_mul_ps(_mm_load_ps(screen[edge0n]), edge0ylerpm));
- data1 = _mm_mul_ps(_mm_sub_ps(data1, data0), spanilengthm);
- data0 = _mm_add_ps(data0, _mm_mul_ps(data1, startxlerpm));
- _mm_store_ps(span->data[0][j], data0);
- _mm_store_ps(span->data[1][j], data1);
+ __m128 maxspanlengthm;
+ span->length = DPSOFTRAST_DRAW_MAXSPANLENGTH;
+ if (dpsoftrast.draw.numspans >= DPSOFTRAST_DRAW_MAXSPANQUEUE)
+ {
+ DPSOFTRAST_Draw_ProcessSpans();
+ dpsoftrast.draw.numspans = 0;
+ }
+ oldspan = span;
+ span = &dpsoftrast.draw.spanqueue[dpsoftrast.draw.numspans++];
+ *span = *oldspan;
+ startx += DPSOFTRAST_DRAW_MAXSPANLENGTH;
+ span->start = y * width + startx;
+ span->length = endx - startx;
+ j = DPSOFTRAST_ARRAY_TOTAL;
+ maxspanlengthm = _mm_set1_ps(DPSOFTRAST_DRAW_MAXSPANLENGTH);
+ _mm_store_ps(span->data[0][j], _mm_add_ps(_mm_load_ps(span->data[0][j]), _mm_mul_ps(_mm_load_ps(span->data[1][j]), maxspanlengthm)));
for (j = 0;j < DPSOFTRAST_ARRAY_TOTAL;j++)
{
//if (arraymask[j])
{
- data0 = _mm_add_ps(_mm_mul_ps(proj[j][edge1p], edge1yilerpm), _mm_mul_ps(proj[j][edge1n], edge1ylerpm));
- data1 = _mm_add_ps(_mm_mul_ps(proj[j][edge0p], edge0yilerpm), _mm_mul_ps(proj[j][edge0n], edge0ylerpm));
- data1 = _mm_mul_ps(_mm_sub_ps(data1, data0), spanilengthm);
- data0 = _mm_add_ps(data0, _mm_mul_ps(data1, startxlerpm));
- _mm_store_ps(span->data[0][j], data0);
- _mm_store_ps(span->data[1][j], data1);
+ _mm_store_ps(span->data[0][j], _mm_add_ps(_mm_load_ps(span->data[0][j]), _mm_mul_ps(_mm_load_ps(span->data[1][j]), maxspanlengthm)));
}
}
}
- }
- // to keep the shader routines from needing more than a small
- // buffer for pixel intermediate data, we split long spans...
- while (span->length > DPSOFTRAST_DRAW_MAXSPANLENGTH)
- {
- __m128 maxspanlengthm;
- span->length = DPSOFTRAST_DRAW_MAXSPANLENGTH;
+ // after all that, we have a span suitable for the pixel shader...
if (dpsoftrast.draw.numspans >= DPSOFTRAST_DRAW_MAXSPANQUEUE)
{
DPSOFTRAST_Draw_ProcessSpans();
dpsoftrast.draw.numspans = 0;
}
- oldspan = span;
- span = &dpsoftrast.draw.spanqueue[dpsoftrast.draw.numspans++];
- *span = *oldspan;
- startx += DPSOFTRAST_DRAW_MAXSPANLENGTH;
- span->start = y * width + startx;
- span->length = endx - startx;
- j = DPSOFTRAST_ARRAY_TOTAL;
- maxspanlengthm = _mm_set1_ps(DPSOFTRAST_DRAW_MAXSPANLENGTH);
- _mm_store_ps(span->data[0][j], _mm_add_ps(_mm_load_ps(span->data[0][j]), _mm_mul_ps(_mm_load_ps(span->data[1][j]), maxspanlengthm)));
- for (j = 0;j < DPSOFTRAST_ARRAY_TOTAL;j++)
- {
- //if (arraymask[j])
- {
- _mm_store_ps(span->data[0][j], _mm_add_ps(_mm_load_ps(span->data[0][j]), _mm_mul_ps(_mm_load_ps(span->data[1][j]), maxspanlengthm)));
- }
- }
- }
- // after all that, we have a span suitable for the pixel shader...
- if (dpsoftrast.draw.numspans >= DPSOFTRAST_DRAW_MAXSPANQUEUE)
- {
- DPSOFTRAST_Draw_ProcessSpans();
- dpsoftrast.draw.numspans = 0;
}
}
// draw outlines over triangle for debugging
projects / xonotic / darkplaces.git / commitdiff