scanline.cpp

Android 一些工具

/* libs/pixelflinger/scanline.cpp
**
** Copyright 2006, The Android Open Source Project
**
** Licensed under the Apache License, Version 2.0 (the "License"); 
** you may not use this file except in compliance with the License. 
** You may obtain a copy of the License at 
**
**     http://www.apache.org/licenses/LICENSE-2.0 
**
** Unless required by applicable law or agreed to in writing, software 
** distributed under the License is distributed on an "AS IS" BASIS, 
** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 
** See the License for the specific language governing permissions and 
** limitations under the License.
*/


#define LOG_TAG "pixelflinger"

#include <assert.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>

#include <cutils/memory.h>
#include <cutils/log.h>

#include "buffer.h"
#include "scanline.h"

#include "codeflinger/CodeCache.h"
#include "codeflinger/GGLAssembler.h"
#include "codeflinger/ARMAssembler.h"
//#include "codeflinger/ARMAssemblerOptimizer.h"

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

#define ANDROID_CODEGEN_GENERIC     0   // force generic pixel pipeline
#define ANDROID_CODEGEN_C           1   // hand-written C, fallback generic 
#define ANDROID_CODEGEN_ASM         2   // hand-written asm, fallback generic
#define ANDROID_CODEGEN_GENERATED   3   // hand-written asm, fallback codegen

#ifdef NDEBUG
#   define ANDROID_RELEASE
#   define ANDROID_CODEGEN      ANDROID_CODEGEN_GENERATED
#else
#   define ANDROID_DEBUG
#   define ANDROID_CODEGEN      ANDROID_CODEGEN_GENERATED
#endif

#if defined(__arm__)
#   define ANDROID_ARM_CODEGEN  1
#else
#   define ANDROID_ARM_CODEGEN  0
#endif


#define DEBUG__CODEGEN_ONLY     0

// ----------------------------------------------------------------------------
namespace android {
// ----------------------------------------------------------------------------

static void init_y(context_t*, int32_t);
static void init_y_noop(context_t*, int32_t);
static void init_y_packed(context_t*, int32_t);
static void init_y_error(context_t*, int32_t);

static void step_y__generic(context_t* c);
static void step_y__nop(context_t*);
static void step_y__smooth(context_t* c);
static void step_y__tmu(context_t* c);
static void step_y__w(context_t* c);

static void scanline(context_t* c);
static void scanline_perspective(context_t* c);
static void scanline_perspective_single(context_t* c);
static void scanline_t32cb16blend(context_t* c);
static void scanline_t32cb16(context_t* c);
static void scanline_memcpy(context_t* c);
static void scanline_memset8(context_t* c);
static void scanline_memset16(context_t* c);
static void scanline_memset32(context_t* c);
static void scanline_noop(context_t* c);
static void scanline_set(context_t* c);
static void scanline_clear(context_t* c);

static void rect_generic(context_t* c, size_t yc);
static void rect_memcpy(context_t* c, size_t yc);

extern "C" void scanline_t32cb16blend_arm(uint16_t*, uint32_t*, size_t);
extern "C" void scanline_t32cb16_arm(uint16_t *dst, uint32_t *src, size_t ct);

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

struct shortcut_t {
    needs_filter_t  filter;
    const char*     desc;
    void            (*scanline)(context_t*);
    void            (*init_y)(context_t*, int32_t);
};

// Keep in sync with needs
static shortcut_t shortcuts[] = {
    { { { 0x03515104, 0x00000077, { 0x00000A01, 0x00000000 } },
        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
        "565 fb, 8888 tx, blend", scanline_t32cb16blend, init_y_noop },
    { { { 0x03010104, 0x00000077, { 0x00000A01, 0x00000000 } },
        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
        "565 fb, 8888 tx", scanline_t32cb16, init_y_noop  },  
    { { { 0x00000000, 0x00000000, { 0x00000000, 0x00000000 } },
        { 0x00000000, 0x00000007, { 0x00000000, 0x00000000 } } },
        "(nop) alpha test", scanline_noop, init_y_noop },
    { { { 0x00000000, 0x00000000, { 0x00000000, 0x00000000 } },
        { 0x00000000, 0x00000070, { 0x00000000, 0x00000000 } } },
        "(nop) depth test", scanline_noop, init_y_noop },
    { { { 0x05000000, 0x00000000, { 0x00000000, 0x00000000 } },
        { 0x0F000000, 0x00000080, { 0x00000000, 0x00000000 } } },
        "(nop) logic_op", scanline_noop, init_y_noop },
    { { { 0xF0000000, 0x00000000, { 0x00000000, 0x00000000 } },
        { 0xF0000000, 0x00000080, { 0x00000000, 0x00000000 } } },
        "(nop) color mask", scanline_noop, init_y_noop },
    { { { 0x0F000000, 0x00000077, { 0x00000000, 0x00000000 } },
        { 0xFF000000, 0x000000F7, { 0x00000000, 0x00000000 } } },
        "(set) logic_op", scanline_set, init_y_noop },
    { { { 0x00000000, 0x00000077, { 0x00000000, 0x00000000 } },
        { 0xFF000000, 0x000000F7, { 0x00000000, 0x00000000 } } },
        "(clear) logic_op", scanline_clear, init_y_noop },
    { { { 0x03000000, 0x00000077, { 0x00000000, 0x00000000 } },
        { 0xFFFFFF00, 0x000000F7, { 0x00000000, 0x00000000 } } },
        "(clear) blending 0/0", scanline_clear, init_y_noop },
    { { { 0x00000000, 0x00000000, { 0x00000000, 0x00000000 } },
        { 0x0000003F, 0x00000000, { 0x00000000, 0x00000000 } } },
        "(error) invalid color-buffer format", scanline_noop, init_y_error },
};
static const needs_filter_t noblend1to1 = {
        // (disregard dithering, see below)
        { 0x03010100, 0x00000077, { 0x00000A00, 0x00000000 } },
        { 0xFFFFFFC0, 0xFFFFFEFF, { 0xFFFFFFC0, 0x0000003F } }
};
static  const needs_filter_t fill16noblend = {
        { 0x03010100, 0x00000077, { 0x00000000, 0x00000000 } },
        { 0xFFFFFFC0, 0xFFFFFFFF, { 0x0000003F, 0x0000003F } }
};

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

#if ANDROID_ARM_CODEGEN
static CodeCache gCodeCache(12 * 1024);

class ScanlineAssembly : public Assembly {
    AssemblyKey<needs_t> mKey;
public:
    ScanlineAssembly(needs_t needs, size_t size)
        : Assembly(size), mKey(needs) { }
    const AssemblyKey<needs_t>& key() const { return mKey; }
};
#endif

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

void ggl_init_scanline(context_t* c)
{
    c->init_y = init_y;
    c->step_y = step_y__generic;
    c->scanline = scanline;
}

void ggl_uninit_scanline(context_t* c)
{
    if (c->state.buffers.coverage)
        free(c->state.buffers.coverage);
#if ANDROID_ARM_CODEGEN
    if (c->scanline_as)
        c->scanline_as->decStrong(c);
#endif
}

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

static void pick_scanline(context_t* c)
{
#if (!defined(DEBUG__CODEGEN_ONLY) || (DEBUG__CODEGEN_ONLY == 0))

#if ANDROID_CODEGEN == ANDROID_CODEGEN_GENERIC
    c->init_y = init_y;
    c->step_y = step_y__generic;
    c->scanline = scanline;
    return;
#endif

    //printf("*** needs [%08lx:%08lx:%08lx:%08lx]\n",
    //    c->state.needs.n, c->state.needs.p,
    //    c->state.needs.t[0], c->state.needs.t[1]);

    // first handle the special case that we cannot test with a filter
    const uint32_t cb_format = GGL_READ_NEEDS(CB_FORMAT, c->state.needs.n);
    if (GGL_READ_NEEDS(T_FORMAT, c->state.needs.t[0]) == cb_format) {
        if (c->state.needs.match(noblend1to1)) {
            // this will match regardless of dithering state, since both
            // src and dest have the same format anyway, there is no dithering
            // to be done.
            const GGLFormat* f =
                &(c->formats[GGL_READ_NEEDS(T_FORMAT, c->state.needs.t[0])]);
            if ((f->components == GGL_RGB) ||
                (f->components == GGL_RGBA) ||
                (f->components == GGL_LUMINANCE) ||
                (f->components == GGL_LUMINANCE_ALPHA))
            {
                // format must have all of RGB components
                // (so the current color doesn't show through)
                c->scanline = scanline_memcpy;
                c->init_y = init_y_noop;
                return;
            }
        }
    }

    if (c->state.needs.match(fill16noblend)) {
        c->init_y = init_y_packed;
        switch (c->formats[cb_format].size) {
        case 1: c->scanline = scanline_memset8;  return;
        case 2: c->scanline = scanline_memset16; return;
        case 4: c->scanline = scanline_memset32; return;
        }
    }

    const int numFilters = sizeof(shortcuts)/sizeof(shortcut_t);
    for (int i=0 ; i<numFilters ; i++) {
        if (c->state.needs.match(shortcuts[i].filter)) {
            c->scanline = shortcuts[i].scanline;
            c->init_y = shortcuts[i].init_y;
            return;
        }
    }

#endif // DEBUG__CODEGEN_ONLY

    c->init_y = init_y;
    c->step_y = step_y__generic;

#if ANDROID_ARM_CODEGEN
    // we're going to have to generate some code...
    // here, generate code for our pixel pipeline
    const AssemblyKey<needs_t> key(c->state.needs);
    sp<Assembly> assembly = gCodeCache.lookup(key);
    if (assembly == 0) {
        // create a new assembly region
        sp<ScanlineAssembly> a = new ScanlineAssembly(c->state.needs, 1024);
        // initialize our assembler
        GGLAssembler assembler( new ARMAssembler(a) );
        //GGLAssembler assembler(
        //        new ARMAssemblerOptimizer(new ARMAssembler(a)) );
        // generate the scanline code for the given needs
        int err = assembler.scanline(c->state.needs, c);
        if (ggl_likely(!err)) {
            // finally, cache this assembly
            err = gCodeCache.cache(a->key(), a);
        }
        if (ggl_unlikely(err)) {
            LOGE("error generating or caching assembly. Reverting to NOP.");
            c->scanline = scanline_noop;
            c->init_y = init_y_noop;
            c->step_y = step_y__nop;
            return;
        }
        assembly = a;
    }

    // release the previous assembly
    if (c->scanline_as) {
        c->scanline_as->decStrong(c);
    }

    //LOGI("using generated pixel-pipeline");
    c->scanline_as = assembly.get();
    c->scanline_as->incStrong(c); //  hold on to assembly
    c->scanline = (void(*)(context_t* c))assembly->base();
#else
//    LOGW("using generic (slow) pixel-pipeline");
    c->scanline = scanline;
#endif
}

void ggl_pick_scanline(context_t* c)
{
    pick_scanline(c);
    if ((c->state.enables & GGL_ENABLE_W) &&
        (c->state.enables & GGL_ENABLE_TMUS))
    {
        c->span = c->scanline;
        c->scanline = scanline_perspective;
        if (!(c->state.enabled_tmu & (c->state.enabled_tmu - 1))) {
            // only one TMU enabled
            c->scanline = scanline_perspective_single;
        }
    }
}

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

static void blending(context_t* c, pixel_t* fragment, pixel_t* fb);
static void blend_factor(context_t* c, pixel_t* r, uint32_t factor,
        const pixel_t* src, const pixel_t* dst);
static void rescale(uint32_t& u, uint8_t& su, uint32_t& v, uint8_t& sv);

#if ANDROID_ARM_CODEGEN && (ANDROID_CODEGEN == ANDROID_CODEGEN_GENERATED)

// no need to compile the generic-pipeline, it can't be reached
void scanline(context_t*)
{
}

#else

void rescale(uint32_t& u, uint8_t& su, uint32_t& v, uint8_t& sv)
{
    if (su && sv) {
        if (su > sv) {
            v = ggl_expand(v, sv, su);
            sv = su;
        } else if (su < sv) {
            u = ggl_expand(u, su, sv);
            su = sv;
        }
    }
}

void blending(context_t* c, pixel_t* fragment, pixel_t* fb)
{
    rescale(fragment->c[0], fragment->s[0], fb->c[0], fb->s[0]);
    rescale(fragment->c[1], fragment->s[1], fb->c[1], fb->s[1]);
    rescale(fragment->c[2], fragment->s[2], fb->c[2], fb->s[2]);
    rescale(fragment->c[3], fragment->s[3], fb->c[3], fb->s[3]);

    pixel_t sf, df;
    blend_factor(c, &sf, c->state.blend.src, fragment, fb);
    blend_factor(c, &df, c->state.blend.dst, fragment, fb);

    fragment->c[1] =
            gglMulAddx(fragment->c[1], sf.c[1], gglMulx(fb->c[1], df.c[1]));
    fragment->c[2] =
            gglMulAddx(fragment->c[2], sf.c[2], gglMulx(fb->c[2], df.c[2]));
    fragment->c[3] =
            gglMulAddx(fragment->c[3], sf.c[3], gglMulx(fb->c[3], df.c[3]));

    if (c->state.blend.alpha_separate) {
        blend_factor(c, &sf, c->state.blend.src_alpha, fragment, fb);
        blend_factor(c, &df, c->state.blend.dst_alpha, fragment, fb);
    }

    fragment->c[0] =
            gglMulAddx(fragment->c[0], sf.c[0], gglMulx(fb->c[0], df.c[0]));

    // clamp to 1.0
    if (fragment->c[0] >= (1LU<<fragment->s[0]))
        fragment->c[0] = (1<<fragment->s[0])-1;
    if (fragment->c[1] >= (1LU<<fragment->s[1]))
        fragment->c[1] = (1<<fragment->s[1])-1;
    if (fragment->c[2] >= (1LU<<fragment->s[2]))
        fragment->c[2] = (1<<fragment->s[2])-1;
    if (fragment->c[3] >= (1LU<<fragment->s[3]))
        fragment->c[3] = (1<<fragment->s[3])-1;
}

static inline int blendfactor(uint32_t x, uint32_t size, uint32_t def = 0)
{
    if (!size)
        return def;

    // scale to 16 bits