scanline.cpp

Android 一些工具

                s = 16;
            }
        }

        // blending
        if (enables & GGL_ENABLE_BLENDING) {
            fb.c[1] = fb.c[2] = fb.c[3] = fb.c[0] = 0; // placate valgrind
            fb.s[1] = fb.s[2] = fb.s[3] = fb.s[0] = 0;
            c->state.buffers.color.read(
                    &(c->state.buffers.color), c, x, y, &fb);
            blending( c, &fragment, &fb );
        }

		// write
        c->state.buffers.color.write(
                &(c->state.buffers.color), c, x, y, &fragment);
        }

discard:
		// iterate...
        x += 1;
        if (enables & GGL_ENABLE_SMOOTH) {
            r += c->shade.drdx;
            g += c->shade.dgdx;
            b += c->shade.dbdx;
            a += c->shade.dadx;
        }
        z += c->shade.dzdx;
        f += c->shade.dfdx;
	}
}

#endif // ANDROID_ARM_CODEGEN && (ANDROID_CODEGEN == ANDROID_CODEGEN_GENERATED)

// ----------------------------------------------------------------------------
#if 0
#pragma mark -
#pragma mark Scanline
#endif

template <typename T, typename U>
static inline __attribute__((const))
T interpolate(int y, T v0, U dvdx, U dvdy) {
    // interpolates in pixel's centers
    // v = v0 + (y + 0.5) * dvdy + (0.5 * dvdx)
    return (y * dvdy) + (v0 + ((dvdy + dvdx) >> 1));
}

// ----------------------------------------------------------------------------
#if 0
#pragma mark -
#endif

void init_y(context_t* c, int32_t ys)
{
    const uint32_t enables = c->state.enables;

    // compute iterators...
    iterators_t& ci = c->iterators;
    
    // sample in the center
    ci.y = ys;

    if (enables & (GGL_ENABLE_DEPTH_TEST|GGL_ENABLE_W|GGL_ENABLE_FOG)) {
        ci.ydzdy = interpolate(ys, c->shade.z0, c->shade.dzdx, c->shade.dzdy);
        ci.ydwdy = interpolate(ys, c->shade.w0, c->shade.dwdx, c->shade.dwdy);
        ci.ydfdy = interpolate(ys, c->shade.f0, c->shade.dfdx, c->shade.dfdy);
    }

    if (ggl_unlikely(enables & GGL_ENABLE_SMOOTH)) {
        ci.ydrdy = interpolate(ys, c->shade.r0, c->shade.drdx, c->shade.drdy);
        ci.ydgdy = interpolate(ys, c->shade.g0, c->shade.dgdx, c->shade.dgdy);
        ci.ydbdy = interpolate(ys, c->shade.b0, c->shade.dbdx, c->shade.dbdy);
        ci.ydady = interpolate(ys, c->shade.a0, c->shade.dadx, c->shade.dady);
        c->step_y = step_y__smooth;
    } else {
        ci.ydrdy = c->shade.r0;
        ci.ydgdy = c->shade.g0;
        ci.ydbdy = c->shade.b0;
        ci.ydady = c->shade.a0;
        // XXX: do only if needed, or make sure this is fast
        c->packed = ggl_pack_color(c, c->state.buffers.color.format,
                ci.ydrdy, ci.ydgdy, ci.ydbdy, ci.ydady);
        c->packed8888 = ggl_pack_color(c, GGL_PIXEL_FORMAT_RGBA_8888, 
                ci.ydrdy, ci.ydgdy, ci.ydbdy, ci.ydady);
    }

    // initialize the variables we need in the shader
    generated_vars_t& gen = c->generated_vars;
    gen.argb[GGLFormat::ALPHA].c  = ci.ydady;
    gen.argb[GGLFormat::ALPHA].dx = c->shade.dadx;
    gen.argb[GGLFormat::RED  ].c  = ci.ydrdy;
    gen.argb[GGLFormat::RED  ].dx = c->shade.drdx;
    gen.argb[GGLFormat::GREEN].c  = ci.ydgdy;
    gen.argb[GGLFormat::GREEN].dx = c->shade.dgdx;
    gen.argb[GGLFormat::BLUE ].c  = ci.ydbdy;
    gen.argb[GGLFormat::BLUE ].dx = c->shade.dbdx;
    gen.dzdx = c->shade.dzdx;
    gen.f    = ci.ydfdy;
    gen.dfdx = c->shade.dfdx;

    if (enables & GGL_ENABLE_TMUS) {
        for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
            texture_t& t = c->state.texture[i];
            if (!t.enable) continue;

            texture_iterators_t& ti = t.iterators;
            if (t.s_coord == GGL_ONE_TO_ONE && t.t_coord == GGL_ONE_TO_ONE) {
                // we need to set all of these to 0 because in some cases
                // step_y__generic() or step_y__tmu() will be used and
                // therefore will update dtdy, however, in 1:1 mode
                // this is always done by the scanline rasterizer.
                ti.dsdx = ti.dsdy = ti.dtdx = ti.dtdy = 0;
                ti.ydsdy = t.shade.is0;
                ti.ydtdy = t.shade.it0;
            } else {
                const int adjustSWrap = ((t.s_wrap==GGL_CLAMP)?0:16);
                const int adjustTWrap = ((t.t_wrap==GGL_CLAMP)?0:16);
                ti.sscale = t.shade.sscale + adjustSWrap;
                ti.tscale = t.shade.tscale + adjustTWrap;
                if (!(enables & GGL_ENABLE_W)) {
                    // S coordinate
                    const int32_t sscale = ti.sscale;
                    const int32_t sy = interpolate(ys,
                            t.shade.is0, t.shade.idsdx, t.shade.idsdy);
                    if (sscale>=0) {
                        ti.ydsdy= sy            << sscale;
                        ti.dsdx = t.shade.idsdx << sscale; 
                        ti.dsdy = t.shade.idsdy << sscale;
                    } else {
                        ti.ydsdy= sy            >> -sscale;
                        ti.dsdx = t.shade.idsdx >> -sscale; 
                        ti.dsdy = t.shade.idsdy >> -sscale;
                    }
                    // T coordinate
                    const int32_t tscale = ti.tscale;
                    const int32_t ty = interpolate(ys,
                            t.shade.it0, t.shade.idtdx, t.shade.idtdy);
                    if (tscale>=0) {
                        ti.ydtdy= ty            << tscale;
                        ti.dtdx = t.shade.idtdx << tscale; 
                        ti.dtdy = t.shade.idtdy << tscale;
                    } else {
                        ti.ydtdy= ty            >> -tscale;
                        ti.dtdx = t.shade.idtdx >> -tscale; 
                        ti.dtdy = t.shade.idtdy >> -tscale;
                    }
                }
            }
            // mirror for generated code...
            generated_tex_vars_t& gen = c->generated_vars.texture[i];
            gen.width   = t.surface.width;
            gen.height  = t.surface.height;
            gen.stride  = t.surface.stride;
            gen.data    = int32_t(t.surface.data);
            gen.dsdx = ti.dsdx;
            gen.dtdx = ti.dtdx;
        }
    }

    // choose the y-stepper
    c->step_y = step_y__nop;
    if (enables & GGL_ENABLE_FOG) {
        c->step_y = step_y__generic;
    } else if (enables & GGL_ENABLE_TMUS) {
        if (enables & GGL_ENABLE_SMOOTH) {
            c->step_y = step_y__generic;
        } else if (enables & GGL_ENABLE_W) {
            c->step_y = step_y__w;
        } else {
            c->step_y = step_y__tmu;
        }
    } else {
        if (enables & GGL_ENABLE_SMOOTH) {
            c->step_y = step_y__smooth;
        }
    }
    
    // choose the rectangle blitter
    c->rect = rect_generic;
    if ((c->step_y == step_y__nop) &&
        (c->scanline == scanline_memcpy))
    {
        c->rect = rect_memcpy;
    }
}

void init_y_packed(context_t* c, int32_t y0)
{
    uint8_t f = c->state.buffers.color.format;
    c->packed = ggl_pack_color(c, f,
            c->shade.r0, c->shade.g0, c->shade.b0, c->shade.a0);
    c->iterators.y = y0;
    c->step_y = step_y__nop;
    // choose the rectangle blitter
    c->rect = rect_generic;
    if (c->scanline == scanline_memcpy) {
        c->rect = rect_memcpy;
    }
}

void init_y_noop(context_t* c, int32_t y0)
{
    c->iterators.y = y0;
    c->step_y = step_y__nop;
    // choose the rectangle blitter
    c->rect = rect_generic;
    if (c->scanline == scanline_memcpy) {
        c->rect = rect_memcpy;
    }
}

void init_y_error(context_t* c, int32_t y0)
{
    // woooops, shoud never happen,
    // fail gracefully (don't display anything)
    init_y_noop(c, y0);
    LOGE("color-buffer has an invalid format!");
}

// ----------------------------------------------------------------------------
#if 0
#pragma mark -
#endif

void step_y__generic(context_t* c)
{
    const uint32_t enables = c->state.enables;

    // iterate...
    iterators_t& ci = c->iterators;
    ci.y += 1;
                
    if (enables & GGL_ENABLE_SMOOTH) {
        ci.ydrdy += c->shade.drdy;
        ci.ydgdy += c->shade.dgdy;
        ci.ydbdy += c->shade.dbdy;
        ci.ydady += c->shade.dady;
    }

    const uint32_t mask =
            GGL_ENABLE_DEPTH_TEST |
            GGL_ENABLE_W |
            GGL_ENABLE_FOG;
    if (enables & mask) {
        ci.ydzdy += c->shade.dzdy;
        ci.ydwdy += c->shade.dwdy;
        ci.ydfdy += c->shade.dfdy;
    }

    if ((enables & GGL_ENABLE_TMUS) && (!(enables & GGL_ENABLE_W))) {
        for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
            if (c->state.texture[i].enable) {
                texture_iterators_t& ti = c->state.texture[i].iterators;
                ti.ydsdy += ti.dsdy;
                ti.ydtdy += ti.dtdy;
            }
        }
    }
}

void step_y__nop(context_t* c)
{
    c->iterators.y += 1;
    c->iterators.ydzdy += c->shade.dzdy;
}

void step_y__smooth(context_t* c)
{
    iterators_t& ci = c->iterators;
    ci.y += 1;
    ci.ydrdy += c->shade.drdy;
    ci.ydgdy += c->shade.dgdy;
    ci.ydbdy += c->shade.dbdy;
    ci.ydady += c->shade.dady;
    ci.ydzdy += c->shade.dzdy;
}

void step_y__w(context_t* c)
{
    iterators_t& ci = c->iterators;
    ci.y += 1;
    ci.ydzdy += c->shade.dzdy;
    ci.ydwdy += c->shade.dwdy;
}

void step_y__tmu(context_t* c)
{
    iterators_t& ci = c->iterators;
    ci.y += 1;
    ci.ydzdy += c->shade.dzdy;
    for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
        if (c->state.texture[i].enable) {
            texture_iterators_t& ti = c->state.texture[i].iterators;
            ti.ydsdy += ti.dsdy;
            ti.ydtdy += ti.dtdy;
        }
    }
}

// ----------------------------------------------------------------------------
#if 0
#pragma mark -
#endif

void scanline_perspective(context_t* c)
{
    struct {
        union {
            struct {
                int32_t s, sq;
                int32_t t, tq;
            };
            struct {
                int32_t v, q;
            } st[2];
        };
    } tc[GGL_TEXTURE_UNIT_COUNT] __attribute__((aligned(16)));

    // XXX: we should have a special case when dwdx = 0

    // 32 pixels spans works okay. 16 is a lot better,
    // but hey, it's a software renderer...
    const uint32_t SPAN_BITS = 5; 
    const uint32_t ys = c->iterators.y;
    const uint32_t xs = c->iterators.xl;
    const uint32_t x1 = c->iterators.xr;
	const uint32_t xc = x1 - xs;
    uint32_t remainder = xc & ((1<<SPAN_BITS)-1);
    uint32_t numSpans = xc >> SPAN_BITS;

    const iterators_t& ci = c->iterators;
    int32_t w0 = (xs * c->shade.dwdx) + ci.ydwdy;
    int32_t q0 = gglRecipQ(w0, 30);
    const int iwscale = 32 - gglClz(q0);

    const int32_t dwdx = c->shade.dwdx << SPAN_BITS;
    int32_t xl = c->iterators.xl;

    // We process s & t with a loop to reduce the code size
    // (and i-cache pressure).

    for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
        const texture_t& tmu = c->state.texture[i];
        if (!tmu.enable) continue;
        int32_t s =   tmu.shade.is0 +
                     (tmu.shade.idsdy * ys) + (tmu.shade.idsdx * xs) +
                     ((tmu.shade.idsdx + tmu.shade.idsdy)>>1);
        int32_t t =   tmu.shade.it0 +
                     (tmu.shade.idtdy * ys) + (tmu.shade.idtdx * xs) +
                     ((tmu.shade.idtdx + tmu.shade.idtdy)>>1);
        tc[i].s  = s;
        tc[i].t  = t;
        tc[i].sq = gglMulx(s, q0, iwscale);
        tc[i].tq = gglMulx(t, q0, iwscale);
    }

    int32_t span = 0;
    do {
        int32_t w1;
        if (ggl_likely(numSpans)) {
            w1 = w0 + dwdx;
        } else {
            if (remainder) {
                // finish off the scanline...
                span = remainder;
                w1 = (c->shade.dwdx * span) + w0;
            } else {
                break;
            }
        }
        int32_t q1 = gglRecipQ(w1, 30);