scanline.cpp

Android 一些工具

        for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
            texture_t& tmu = c->state.texture[i];
            if (!tmu.enable) continue;
            texture_iterators_t& ti = tmu.iterators;

            for (int j=0 ; j<2 ; j++) {
                int32_t v = tc[i].st[j].v;
                if (span)   v += (tmu.shade.st[j].dx)*span;
                else        v += (tmu.shade.st[j].dx)<<SPAN_BITS;
                const int32_t v0 = tc[i].st[j].q;
                const int32_t v1 = gglMulx(v, q1, iwscale);
                int32_t dvdx = v1 - v0;
                if (span)   dvdx /= span;
                else        dvdx >>= SPAN_BITS;
                tc[i].st[j].v = v;
                tc[i].st[j].q = v1;

                const int scale = ti.st[j].scale + (iwscale - 30);
                if (scale >= 0) {
                    ti.st[j].ydvdy = v0   << scale;
                    ti.st[j].dvdx  = dvdx << scale;
                } else {
                    ti.st[j].ydvdy = v0   >> -scale;
                    ti.st[j].dvdx  = dvdx >> -scale;
                }
            }
            generated_tex_vars_t& gen = c->generated_vars.texture[i];
            gen.dsdx = ti.st[0].dvdx;
            gen.dtdx = ti.st[1].dvdx;
        }
        c->iterators.xl = xl;
        c->iterators.xr = xl = xl + (span ? span : (1<<SPAN_BITS));
        w0 = w1;
        q0 = q1;
        c->span(c);
    } while(numSpans--);
}

void scanline_perspective_single(context_t* c)
{
    // 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;

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

    const int i = 31 - gglClz(c->state.enabled_tmu);
    generated_tex_vars_t& gen = c->generated_vars.texture[i];
    texture_t& tmu = c->state.texture[i];
    texture_iterators_t& ti = tmu.iterators;
    const int sscale = ti.sscale + (iwscale - 30);
    const int tscale = ti.tscale + (iwscale - 30);
    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);
    int32_t s0 = gglMulx(s, iw, iwscale);
    int32_t t0 = gglMulx(t, iw, iwscale);
    int32_t xl = c->iterators.xl;

    int32_t sq, tq, dsdx, dtdx;
    int32_t premainder = xc & ((1<<SPAN_BITS)-1);
    uint32_t numSpans = xc >> SPAN_BITS;
    if (c->shade.dwdx == 0) {
        // XXX: we could choose to do this if the error is small enough
        numSpans = 0;
        premainder = xc;
        goto no_perspective;
    }

    if (premainder) {
        w += c->shade.dwdx   * premainder;
        iw = gglRecipQ(w, 30);
no_perspective:        
        s += tmu.shade.idsdx * premainder;
        t += tmu.shade.idtdx * premainder;
        sq = gglMulx(s, iw, iwscale);
        tq = gglMulx(t, iw, iwscale);
        dsdx = (sq - s0) / premainder;
        dtdx = (tq - t0) / premainder;
        c->iterators.xl = xl;
        c->iterators.xr = xl = xl + premainder;
        goto finish;
    }

    while (numSpans--) {
        w += c->shade.dwdx   << SPAN_BITS;
        s += tmu.shade.idsdx << SPAN_BITS;
        t += tmu.shade.idtdx << SPAN_BITS;
        iw = gglRecipQ(w, 30);
        sq = gglMulx(s, iw, iwscale);
        tq = gglMulx(t, iw, iwscale);
        dsdx = (sq - s0) >> SPAN_BITS;
        dtdx = (tq - t0) >> SPAN_BITS;
        c->iterators.xl = xl;
        c->iterators.xr = xl = xl + (1<<SPAN_BITS);
finish:
        if (sscale >= 0) {
            ti.ydsdy = s0   << sscale;
            ti.dsdx  = dsdx << sscale;
        } else {
            ti.ydsdy = s0   >>-sscale;
            ti.dsdx  = dsdx >>-sscale;
        }
        if (tscale >= 0) {
            ti.ydtdy = t0   << tscale;
            ti.dtdx  = dtdx << tscale;
        } else {
            ti.ydtdy = t0   >>-tscale;
            ti.dtdx  = dtdx >>-tscale;
        }
        s0 = sq;
        t0 = tq;
        gen.dsdx = ti.dsdx;
        gen.dtdx = ti.dtdx;
        c->span(c);
    }
}

// ----------------------------------------------------------------------------

void scanline_t32cb16(context_t* c)
{
    int32_t x = c->iterators.xl;
    size_t ct = c->iterators.xr - x;
    int32_t y = c->iterators.y;
    surface_t* cb = &(c->state.buffers.color);
    union {
        uint16_t* dst;
        uint32_t* dst32;
    };
    dst = reinterpret_cast<uint16_t*>(cb->data) + (x+(cb->stride*y));

    surface_t* tex = &(c->state.texture[0].surface);
    const int32_t u = (c->state.texture[0].shade.is0>>16) + x;
    const int32_t v = (c->state.texture[0].shade.it0>>16) + y;
    uint32_t *src = reinterpret_cast<uint32_t*>(tex->data)+(u+(tex->stride*v));
    int sR, sG, sB;
    uint32_t s, d;

    if (ct==1 || uint32_t(dst)&2) {
last_one:
        s = GGL_RGBA_TO_HOST( *src++ );
        sR = (s >> (   3))&0x1F;
        sG = (s >> ( 8+2))&0x3F;
        sB = (s >> (16+3))&0x1F;
        *dst++ = uint16_t((sR<<11)|(sG<<5)|sB);
        ct--;
    }

    while (ct > 0) {
        s = GGL_RGBA_TO_HOST( *src++ );
        sR = (s >> (   3))&0x1F;
        sG = (s >> ( 8+2))&0x3F;
        sB = (s >> (16+3))&0x1F;
        d = (sR<<11)|(sG<<5)|sB;
        
        s = GGL_RGBA_TO_HOST( *src++ );
        sR = (s >> (   3))&0x1F;
        sG = (s >> ( 8+2))&0x3F;
        sB = (s >> (16+3))&0x1F;        
        d |= ((sR<<11)|(sG<<5)|sB)<<16;

#if BYTE_ORDER == BIG_ENDIAN
        d = (d>>16) | (d<<16);
#endif

        *dst32++ = d;
        ct -= 2;
    }
    
    if (ct > 0) {
        goto last_one;
    }
}

void scanline_t32cb16blend(context_t* c)
{
    int32_t x = c->iterators.xl;
    size_t ct = c->iterators.xr - x;
    int32_t y = c->iterators.y;
    surface_t* cb = &(c->state.buffers.color);
    uint16_t* dst = reinterpret_cast<uint16_t*>(cb->data) + (x+(cb->stride*y));

    surface_t* tex = &(c->state.texture[0].surface);
    const int32_t u = (c->state.texture[0].shade.is0>>16) + x;
    const int32_t v = (c->state.texture[0].shade.it0>>16) + y;
    uint32_t *src = reinterpret_cast<uint32_t*>(tex->data)+(u+(tex->stride*v));

#if ((ANDROID_CODEGEN >= ANDROID_CODEGEN_ASM) && defined(__arm__))
    scanline_t32cb16blend_arm(dst, src, ct);
#else
    while (ct--) {
        uint32_t s = *src++;
        if (!s) {
            dst++;
            continue;
        }
        uint16_t d = *dst;
        s = GGL_RGBA_TO_HOST(s);
        int sR = (s >> (   3))&0x1F;
        int sG = (s >> ( 8+2))&0x3F;
        int sB = (s >> (16+3))&0x1F;
        int sA = (s>>24);
        int f = 0x100 - (sA + (sA>>7));
        int dR = (d>>11)&0x1f;
        int dG = (d>>5)&0x3f;
        int dB = (d)&0x1f;
        sR += (f*dR)>>8;
        sG += (f*dG)>>8;
        sB += (f*dB)>>8;
        *dst++ = uint16_t((sR<<11)|(sG<<5)|sB);
    }
#endif
}

void scanline_memcpy(context_t* c)
{
    int32_t x = c->iterators.xl;
    size_t ct = c->iterators.xr - x;
    int32_t y = c->iterators.y;
    surface_t* cb = &(c->state.buffers.color);
    const GGLFormat* fp = &(c->formats[cb->format]);
    uint8_t* dst = reinterpret_cast<uint8_t*>(cb->data) +
                            (x + (cb->stride * y)) * fp->size;

    surface_t* tex = &(c->state.texture[0].surface);
    const int32_t u = (c->state.texture[0].shade.is0>>16) + x;
    const int32_t v = (c->state.texture[0].shade.it0>>16) + y;
    uint8_t *src = reinterpret_cast<uint8_t*>(tex->data) +
                            (u + (tex->stride * v)) * fp->size;

    const size_t size = ct * fp->size;
    memcpy(dst, src, size);
}

void scanline_memset8(context_t* c)
{
    int32_t x = c->iterators.xl;
    size_t ct = c->iterators.xr - x;
    int32_t y = c->iterators.y;
    surface_t* cb = &(c->state.buffers.color);
    uint8_t* dst = reinterpret_cast<uint8_t*>(cb->data) + (x+(cb->stride*y));
    uint32_t packed = c->packed;
    memset(dst, packed, ct);
}

void scanline_memset16(context_t* c)
{
    int32_t x = c->iterators.xl;
    size_t ct = c->iterators.xr - x;
    int32_t y = c->iterators.y;
    surface_t* cb = &(c->state.buffers.color);
    uint16_t* dst = reinterpret_cast<uint16_t*>(cb->data) + (x+(cb->stride*y));
    uint32_t packed = c->packed;
    android_memset16(dst, packed, ct*2);
}

void scanline_memset32(context_t* c)
{
    int32_t x = c->iterators.xl;
    size_t ct = c->iterators.xr - x;
    int32_t y = c->iterators.y;
    surface_t* cb = &(c->state.buffers.color);
    uint32_t* dst = reinterpret_cast<uint32_t*>(cb->data) + (x+(cb->stride*y));
    uint32_t packed = GGL_HOST_TO_RGBA(c->packed);
    android_memset32(dst, packed, ct*4);
}

void scanline_clear(context_t* c)
{
    int32_t x = c->iterators.xl;
    size_t ct = c->iterators.xr - x;
    int32_t y = c->iterators.y;
    surface_t* cb = &(c->state.buffers.color);
    const GGLFormat* fp = &(c->formats[cb->format]);
    uint8_t* dst = reinterpret_cast<uint8_t*>(cb->data) +
                            (x + (cb->stride * y)) * fp->size;
    const size_t size = ct * fp->size;
    memset(dst, 0, size);
}

void scanline_set(context_t* c)
{
    int32_t x = c->iterators.xl;
    size_t ct = c->iterators.xr - x;
    int32_t y = c->iterators.y;
    surface_t* cb = &(c->state.buffers.color);
    const GGLFormat* fp = &(c->formats[cb->format]);
    uint8_t* dst = reinterpret_cast<uint8_t*>(cb->data) +
                            (x + (cb->stride * y)) * fp->size;
    const size_t size = ct * fp->size;
    memset(dst, 0xFF, size);
}

void scanline_noop(context_t* c)
{
}

void rect_generic(context_t* c, size_t yc)
{
    do {
        c->scanline(c);
        c->step_y(c);
    } while (--yc);
}

void rect_memcpy(context_t* c, size_t yc)
{
    int32_t x = c->iterators.xl;
    size_t ct = c->iterators.xr - x;
    int32_t y = c->iterators.y;
    surface_t* cb = &(c->state.buffers.color);
    const GGLFormat* fp = &(c->formats[cb->format]);
    uint8_t* dst = reinterpret_cast<uint8_t*>(cb->data) +
                            (x + (cb->stride * y)) * fp->size;

    surface_t* tex = &(c->state.texture[0].surface);
    const int32_t u = (c->state.texture[0].shade.is0>>16) + x;
    const int32_t v = (c->state.texture[0].shade.it0>>16) + y;
    uint8_t *src = reinterpret_cast<uint8_t*>(tex->data) +
                            (u + (tex->stride * v)) * fp->size;

    if (cb->stride == tex->stride && ct == size_t(cb->stride)) {
        memcpy(dst, src, ct * fp->size * yc);
    } else {
        const size_t size = ct * fp->size;
        const size_t dbpr = cb->stride  * fp->size;
        const size_t sbpr = tex->stride * fp->size;
        do {
            memcpy(dst, src, size);
            dst += dbpr;
            src += sbpr;        
        } while (--yc);
    }
}
// ----------------------------------------------------------------------------
}; // namespace android

using namespace android;
extern "C" void ggl_test_codegen(uint32_t n, uint32_t p, uint32_t t0, uint32_t t1)
{
#if ANDROID_ARM_CODEGEN
    GGLContext* c;
    gglInit(&c);
    needs_t needs;
    needs.n = n;
    needs.p = p;
    needs.t[0] = t0;
    needs.t[1] = t1;
    sp<ScanlineAssembly> a(new ScanlineAssembly(needs, 1024));
    GGLAssembler assembler( new ARMAssembler(a) );
    int err = assembler.scanline(needs, (context_t*)c);
    if (err != 0) {
        printf("error %08x (%s)\n", err, strerror(-err));
    }
    gglUninit(c);
#else
    printf("This test runs only on ARM\n");
#endif
}