Code drop from //branches/cupcake/...@124589
diff --git a/libpixelflinger/codeflinger/GGLAssembler.cpp b/libpixelflinger/codeflinger/GGLAssembler.cpp
index 90c275e..1cd189c 100644
--- a/libpixelflinger/codeflinger/GGLAssembler.cpp
+++ b/libpixelflinger/codeflinger/GGLAssembler.cpp
@@ -151,6 +151,7 @@
// 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;
@@ -176,9 +177,14 @@
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;
// ------------------------------------------------------------------------
@@ -226,8 +232,10 @@
build_textures(parts, regs);
}
- if ((blending & (FACTOR_DST|BLEND_DST)) || mMasking ||
- (mLogicOp & LOGIC_OP_DST)) {
+ if ((blending & (FACTOR_DST|BLEND_DST)) ||
+ (mMasking && !mAllMasked) ||
+ (mLogicOp & LOGIC_OP_DST))
+ {
// blending / logic_op / masking need the framebuffer
mDstPixel.setTo(regs.obtain(), &mCbFormat);
@@ -284,14 +292,16 @@
pixel = mDstPixel;
}
- // logic operation
- build_logic_op(pixel, regs);
-
- // masking
- build_masking(pixel, regs);
-
- comment("store");
- store(parts.cbPtr, pixel, WRITE_BACK);
+ 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())
@@ -322,7 +332,9 @@
build_smooth_shade(parts);
build_iterate_z(parts);
build_iterate_f(parts);
- ADD(AL, 0, parts.cbPtr.reg, parts.cbPtr.reg, imm(parts.cbPtr.size>>3));
+ if (!mAllMasked) {
+ 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());
@@ -370,16 +382,18 @@
MOV(AL, 0, parts.count.reg, reg_imm(parts.count.reg, LSL, 16));
}
- // compute dst ptr
- comment("compute color-buffer pointer");
- const int cb_bits = mCbFormat.size*8;
- int Rs = scratches.obtain();
- parts.cbPtr.setTo(obtainReg(), cb_bits);
- CONTEXT_LOAD(Rs, state.buffers.color.stride);
- CONTEXT_LOAD(parts.cbPtr.reg, state.buffers.color.data);
- SMLABB(AL, Rs, Ry, Rs, Rx); // Rs = Rx + Ry*Rs
- base_offset(parts.cbPtr, parts.cbPtr, Rs);
- scratches.recycle(Rs);
+ if (!mAllMasked) {
+ // compute dst ptr
+ comment("compute color-buffer pointer");
+ const int cb_bits = mCbFormat.size*8;
+ int Rs = scratches.obtain();
+ parts.cbPtr.setTo(obtainReg(), cb_bits);
+ CONTEXT_LOAD(Rs, state.buffers.color.stride);
+ CONTEXT_LOAD(parts.cbPtr.reg, state.buffers.color.data);
+ SMLABB(AL, Rs, Ry, Rs, Rx); // Rs = Rx + Ry*Rs
+ base_offset(parts.cbPtr, parts.cbPtr, Rs);
+ scratches.recycle(Rs);
+ }
// init fog
const int need_fog = GGL_READ_NEEDS(P_FOG, needs.p);
@@ -904,8 +918,9 @@
void GGLAssembler::build_masking(pixel_t& pixel, Scratch& regs)
{
- if (!mMasking)
+ if (!mMasking || mAllMasked) {
return;
+ }
comment("color mask");
@@ -928,7 +943,7 @@
// There is no need to clear the masked components of the source
// (unless we applied a logic op), because they're already zeroed
- // by contruction (masked components are not computed)
+ // by construction (masked components are not computed)
if (mLogicOp) {
const needs_t& needs = mBuilderContext.needs;
diff --git a/libpixelflinger/codeflinger/GGLAssembler.h b/libpixelflinger/codeflinger/GGLAssembler.h
index ccaf43d..d1d29f0 100644
--- a/libpixelflinger/codeflinger/GGLAssembler.h
+++ b/libpixelflinger/codeflinger/GGLAssembler.h
@@ -363,6 +363,10 @@
const component_t& incoming,
const pixel_t& texel, int component, int tmu);
+ void add( component_t& dest,
+ const component_t& incoming,
+ const pixel_t& texel, int component);
+
// load/store stuff
void store(const pointer_t& addr, const pixel_t& src, uint32_t flags=0);
void load(const pointer_t& addr, const pixel_t& dest, uint32_t flags=0);
@@ -517,6 +521,7 @@
component_info_t mInfo[4];
int mBlending;
int mMasking;
+ int mAllMasked;
int mLogicOp;
int mAlphaTest;
int mAA;
diff --git a/libpixelflinger/codeflinger/blending.cpp b/libpixelflinger/codeflinger/blending.cpp
index 6d3b282..f10217b 100644
--- a/libpixelflinger/codeflinger/blending.cpp
+++ b/libpixelflinger/codeflinger/blending.cpp
@@ -50,6 +50,12 @@
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);
}
}
diff --git a/libpixelflinger/codeflinger/load_store.cpp b/libpixelflinger/codeflinger/load_store.cpp
index 514ce07..93c5825 100644
--- a/libpixelflinger/codeflinger/load_store.cpp
+++ b/libpixelflinger/codeflinger/load_store.cpp
@@ -168,7 +168,7 @@
void GGLAssembler::expand(component_t& d, const component_t& s, int dbits)
{
integer_t r(d.reg, 32, d.flags);
- expand(r, d, dbits);
+ expand(r, s, dbits);
d = component_t(r);
}
diff --git a/libpixelflinger/codeflinger/texturing.cpp b/libpixelflinger/codeflinger/texturing.cpp
index 269b6c0..90e6584 100644
--- a/libpixelflinger/codeflinger/texturing.cpp
+++ b/libpixelflinger/codeflinger/texturing.cpp
@@ -1000,6 +1000,9 @@
case GGL_BLEND:
blend(fragment, incoming, texel, component, i);
break;
+ case GGL_ADD:
+ add(fragment, incoming, texel, component);
+ break;
}
}
}
@@ -1202,6 +1205,46 @@
build_blendOneMinusFF(dest, factor, incomingNorm, color);
}
+void GGLAssembler::add(
+ component_t& dest,
+ const component_t& incoming,
+ const pixel_t& incomingTexel, int component)
+{
+ // RGBA:
+ // Cv = Cf + Ct;
+ Scratch locals(registerFile());
+
+ component_t incomingTemp(incoming);
+
+ // use "dest" as a temporary for extracting the texel, unless "dest"
+ // overlaps "incoming".
+ integer_t texel(dest.reg, 32, CORRUPTIBLE);
+ if (dest.reg == incomingTemp.reg)
+ texel.reg = locals.obtain();
+ extract(texel, incomingTexel, component);
+
+ if (texel.s < incomingTemp.size()) {
+ expand(texel, texel, incomingTemp.size());
+ } else if (texel.s > incomingTemp.size()) {
+ if (incomingTemp.flags & CORRUPTIBLE) {
+ expand(incomingTemp, incomingTemp, texel.s);
+ } else {
+ incomingTemp.reg = locals.obtain();
+ expand(incomingTemp, incoming, texel.s);
+ }
+ }
+
+ if (incomingTemp.l) {
+ ADD(AL, 0, dest.reg, texel.reg,
+ reg_imm(incomingTemp.reg, LSR, incomingTemp.l));
+ } else {
+ ADD(AL, 0, dest.reg, texel.reg, incomingTemp.reg);
+ }
+ dest.l = 0;
+ dest.h = texel.size();
+ component_sat(dest);
+}
+
// ----------------------------------------------------------------------------
}; // namespace android