void GGLAssembler::store(const pointer_t& addr, const pixel_t& s, uint32_t flags)
{
const int bits = addr.size;
const int inc = (flags & WRITE_BACK)?1:0;
switch (bits) {
case 32:
if (inc) STR(AL, s.reg, addr.reg, immed12_post(4));
else STR(AL, s.reg, addr.reg);
break;
case 24:
// 24 bits formats are a little special and used only for RGB
// 0x00BBGGRR is unpacked as R,G,B
STRB(AL, s.reg, addr.reg, immed12_pre(0));
MOV(AL, 0, s.reg, reg_imm(s.reg, ROR, 8));
STRB(AL, s.reg, addr.reg, immed12_pre(1));
MOV(AL, 0, s.reg, reg_imm(s.reg, ROR, 8));
STRB(AL, s.reg, addr.reg, immed12_pre(2));
if (!(s.flags & CORRUPTIBLE)) {
MOV(AL, 0, s.reg, reg_imm(s.reg, ROR, 16));
}
if (inc)
ADD(AL, 0, addr.reg, addr.reg, imm(3));
break;
case 16:
if (inc) STRH(AL, s.reg, addr.reg, immed8_post(2));
else STRH(AL, s.reg, addr.reg);
break;
case 8:
if (inc) STRB(AL, s.reg, addr.reg, immed12_post(1));
else STRB(AL, s.reg, addr.reg);
break;
}
}
void GGLAssembler::build_fog(
component_t& temp, // incomming fragment / output
int component,
Scratch& regs)
{
if (mInfo[component].fog) {
Scratch scratches(registerFile());
comment("fog");
integer_t fragment(temp.reg, temp.h, temp.flags);
if (!(temp.flags & CORRUPTIBLE)) {
temp.reg = regs.obtain();
temp.flags |= CORRUPTIBLE;
}
integer_t fogColor(scratches.obtain(), 8, CORRUPTIBLE);
LDRB(AL, fogColor.reg, mBuilderContext.Rctx,
immed12_pre(GGL_OFFSETOF(state.fog.color[component])));
integer_t factor(scratches.obtain(), 16, CORRUPTIBLE);
CONTEXT_LOAD(factor.reg, generated_vars.f);
// clamp fog factor (TODO: see if there is a way to guarantee
// we won't overflow, when setting the iterators)
BIC(AL, 0, factor.reg, factor.reg, reg_imm(factor.reg, ASR, 31));
CMP(AL, factor.reg, imm( 0x10000 ));
MOV(HS, 0, factor.reg, imm( 0x10000 ));
build_blendFOneMinusF(temp, factor, fragment, fogColor);
}
}
void GGLAssembler::load(const pointer_t& addr, const pixel_t& s, uint32_t flags)
{
Scratch scratches(registerFile());
int s0;
const int bits = addr.size;
const int inc = (flags & WRITE_BACK)?1:0;
switch (bits) {
case 32:
if (inc) LDR(AL, s.reg, addr.reg, immed12_post(4));
else LDR(AL, s.reg, addr.reg);
break;
case 24:
// 24 bits formats are a little special and used only for RGB
// R,G,B is packed as 0x00BBGGRR
s0 = scratches.obtain();
if (s.reg != addr.reg) {
LDRB(AL, s.reg, addr.reg, immed12_pre(0)); // R
LDRB(AL, s0, addr.reg, immed12_pre(1)); // G
ORR(AL, 0, s.reg, s.reg, reg_imm(s0, LSL, 8));
LDRB(AL, s0, addr.reg, immed12_pre(2)); // B
ORR(AL, 0, s.reg, s.reg, reg_imm(s0, LSL, 16));
} else {
int s1 = scratches.obtain();
LDRB(AL, s1, addr.reg, immed12_pre(0)); // R
LDRB(AL, s0, addr.reg, immed12_pre(1)); // G
ORR(AL, 0, s1, s1, reg_imm(s0, LSL, 8));
LDRB(AL, s0, addr.reg, immed12_pre(2)); // B
ORR(AL, 0, s.reg, s1, reg_imm(s0, LSL, 16));
}
if (inc)
ADD(AL, 0, addr.reg, addr.reg, imm(3));
break;
case 16:
if (inc) LDRH(AL, s.reg, addr.reg, immed8_post(2));
else LDRH(AL, s.reg, addr.reg);
break;
case 8:
if (inc) LDRB(AL, s.reg, addr.reg, immed12_post(1));
else LDRB(AL, s.reg, addr.reg);
break;
}
}
int GGLAssembler::scanline_core(const needs_t& needs, context_t const* c)
{
int64_t duration = ggl_system_time();
mBlendFactorCached = 0;
mBlending = 0;
mMasking = 0;
mAA = GGL_READ_NEEDS(P_AA, needs.p);
mDithering = GGL_READ_NEEDS(P_DITHER, needs.p);
mAlphaTest = GGL_READ_NEEDS(P_ALPHA_TEST, needs.p) + GGL_NEVER;
mDepthTest = GGL_READ_NEEDS(P_DEPTH_TEST, needs.p) + GGL_NEVER;
mFog = GGL_READ_NEEDS(P_FOG, needs.p) != 0;
mSmooth = GGL_READ_NEEDS(SHADE, needs.n) != 0;
mBuilderContext.needs = needs;
mBuilderContext.c = c;
mBuilderContext.Rctx = reserveReg(R0); // context always in R0
mCbFormat = c->formats[ GGL_READ_NEEDS(CB_FORMAT, needs.n) ];
// ------------------------------------------------------------------------
decodeLogicOpNeeds(needs);
decodeTMUNeeds(needs, c);
mBlendSrc = ggl_needs_to_blendfactor(GGL_READ_NEEDS(BLEND_SRC, needs.n));
mBlendDst = ggl_needs_to_blendfactor(GGL_READ_NEEDS(BLEND_DST, needs.n));
mBlendSrcA = ggl_needs_to_blendfactor(GGL_READ_NEEDS(BLEND_SRCA, needs.n));
mBlendDstA = ggl_needs_to_blendfactor(GGL_READ_NEEDS(BLEND_DSTA, needs.n));
if (!mCbFormat.c[GGLFormat::ALPHA].h) {
if ((mBlendSrc == GGL_ONE_MINUS_DST_ALPHA) ||
(mBlendSrc == GGL_DST_ALPHA)) {
mBlendSrc = GGL_ONE;
}
if ((mBlendSrcA == GGL_ONE_MINUS_DST_ALPHA) ||
(mBlendSrcA == GGL_DST_ALPHA)) {
mBlendSrcA = GGL_ONE;
}
if ((mBlendDst == GGL_ONE_MINUS_DST_ALPHA) ||
(mBlendDst == GGL_DST_ALPHA)) {
mBlendDst = GGL_ONE;
}
if ((mBlendDstA == GGL_ONE_MINUS_DST_ALPHA) ||
(mBlendDstA == GGL_DST_ALPHA)) {
mBlendDstA = GGL_ONE;
}
}
// if we need the framebuffer, read it now
const int blending = blending_codes(mBlendSrc, mBlendDst) |
blending_codes(mBlendSrcA, mBlendDstA);
// XXX: handle special cases, destination not modified...
if ((mBlendSrc==GGL_ZERO) && (mBlendSrcA==GGL_ZERO) &&
(mBlendDst==GGL_ONE) && (mBlendDstA==GGL_ONE)) {
// Destination unmodified (beware of logic ops)
} else if ((mBlendSrc==GGL_ZERO) && (mBlendSrcA==GGL_ZERO) &&
(mBlendDst==GGL_ZERO) && (mBlendDstA==GGL_ZERO)) {
// Destination is zero (beware of logic ops)
}
int fbComponents = 0;
const int masking = GGL_READ_NEEDS(MASK_ARGB, needs.n);
for (int i=0 ; i<4 ; i++) {
const int mask = 1<<i;
component_info_t& info = mInfo[i];
int fs = i==GGLFormat::ALPHA ? mBlendSrcA : mBlendSrc;
int fd = i==GGLFormat::ALPHA ? mBlendDstA : mBlendDst;
if (fs==GGL_SRC_ALPHA_SATURATE && i==GGLFormat::ALPHA)
fs = GGL_ONE;
info.masked = !!(masking & mask);
info.inDest = !info.masked && mCbFormat.c[i].h &&
((mLogicOp & LOGIC_OP_SRC) || (!mLogicOp));
if (mCbFormat.components >= GGL_LUMINANCE &&
(i==GGLFormat::GREEN || i==GGLFormat::BLUE)) {
info.inDest = false;
}
info.needed = (i==GGLFormat::ALPHA) &&
(isAlphaSourceNeeded() || mAlphaTest != GGL_ALWAYS);
info.replaced = !!(mTextureMachine.replaced & mask);
info.iterated = (!info.replaced && (info.inDest || info.needed));
info.smooth = mSmooth && info.iterated;
info.fog = mFog && info.inDest && (i != GGLFormat::ALPHA);
info.blend = (fs != int(GGL_ONE)) || (fd > int(GGL_ZERO));
mBlending |= (info.blend ? mask : 0);
mMasking |= (mCbFormat.c[i].h && info.masked) ? mask : 0;
fbComponents |= mCbFormat.c[i].h ? mask : 0;
}
mAllMasked = (mMasking == fbComponents);
if (mAllMasked) {
mDithering = 0;
}
fragment_parts_t parts;
// ------------------------------------------------------------------------
prolog();
// ------------------------------------------------------------------------
build_scanline_prolog(parts, needs);
if (registerFile().status())
return registerFile().status();
// ------------------------------------------------------------------------
label("fragment_loop");
// ------------------------------------------------------------------------
{
Scratch regs(registerFile());
if (mDithering) {
// update the dither index.
MOV(AL, 0, parts.count.reg,
reg_imm(parts.count.reg, ROR, GGL_DITHER_ORDER_SHIFT));
ADD(AL, 0, parts.count.reg, parts.count.reg,
imm( 1 << (32 - GGL_DITHER_ORDER_SHIFT)));
MOV(AL, 0, parts.count.reg,
reg_imm(parts.count.reg, ROR, 32 - GGL_DITHER_ORDER_SHIFT));
}
// XXX: could we do an early alpha-test here in some cases?
// It would probaly be used only with smooth-alpha and no texture
// (or no alpha component in the texture).
// Early z-test
if (mAlphaTest==GGL_ALWAYS) {
build_depth_test(parts, Z_TEST|Z_WRITE);
} else {
// we cannot do the z-write here, because
// it might be killed by the alpha-test later
build_depth_test(parts, Z_TEST);
}
{ // texture coordinates
Scratch scratches(registerFile());
// texel generation
build_textures(parts, regs);
if (registerFile().status())
return registerFile().status();
}
if ((blending & (FACTOR_DST|BLEND_DST)) ||
(mMasking && !mAllMasked) ||
(mLogicOp & LOGIC_OP_DST))
{
// blending / logic_op / masking need the framebuffer
mDstPixel.setTo(regs.obtain(), &mCbFormat);
// load the framebuffer pixel
comment("fetch color-buffer");
load(parts.cbPtr, mDstPixel);
}
if (registerFile().status())
return registerFile().status();
pixel_t pixel;
int directTex = mTextureMachine.directTexture;
if (directTex | parts.packed) {
// note: we can't have both here
// iterated color or direct texture
pixel = directTex ? parts.texel[directTex-1] : parts.iterated;
pixel.flags &= ~CORRUPTIBLE;
} else {
if (mDithering) {
const int ctxtReg = mBuilderContext.Rctx;
const int mask = GGL_DITHER_SIZE-1;
parts.dither = reg_t(regs.obtain());
AND(AL, 0, parts.dither.reg, parts.count.reg, imm(mask));
ADDR_ADD(AL, 0, parts.dither.reg, ctxtReg, parts.dither.reg);
LDRB(AL, parts.dither.reg, parts.dither.reg,
immed12_pre(GGL_OFFSETOF(ditherMatrix)));
}
// allocate a register for the resulting pixel
pixel.setTo(regs.obtain(), &mCbFormat, FIRST);
build_component(pixel, parts, GGLFormat::ALPHA, regs);
if (mAlphaTest!=GGL_ALWAYS) {
// only handle the z-write part here. We know z-test
// was successful, as well as alpha-test.
build_depth_test(parts, Z_WRITE);
}
build_component(pixel, parts, GGLFormat::RED, regs);
build_component(pixel, parts, GGLFormat::GREEN, regs);
build_component(pixel, parts, GGLFormat::BLUE, regs);
pixel.flags |= CORRUPTIBLE;
}
if (registerFile().status())
return registerFile().status();
if (pixel.reg == -1) {
// be defensive here. if we're here it's probably
// that this whole fragment is a no-op.
pixel = mDstPixel;
}
if (!mAllMasked) {
// logic operation
build_logic_op(pixel, regs);
// masking
build_masking(pixel, regs);
comment("store");
store(parts.cbPtr, pixel, WRITE_BACK);
}
}
if (registerFile().status())
return registerFile().status();
// update the iterated color...
if (parts.reload != 3) {
build_smooth_shade(parts);
}
// update iterated z
build_iterate_z(parts);
// update iterated fog
build_iterate_f(parts);
SUB(AL, S, parts.count.reg, parts.count.reg, imm(1<<16));
B(PL, "fragment_loop");
label("epilog");
epilog(registerFile().touched());
if ((mAlphaTest!=GGL_ALWAYS) || (mDepthTest!=GGL_ALWAYS)) {
if (mDepthTest!=GGL_ALWAYS) {
label("discard_before_textures");
build_iterate_texture_coordinates(parts);
}
label("discard_after_textures");
build_smooth_shade(parts);
build_iterate_z(parts);
build_iterate_f(parts);
if (!mAllMasked) {
ADDR_ADD(AL, 0, parts.cbPtr.reg, parts.cbPtr.reg, imm(parts.cbPtr.size>>3));
}
SUB(AL, S, parts.count.reg, parts.count.reg, imm(1<<16));
B(PL, "fragment_loop");
epilog(registerFile().touched());
}
