pixelflinger: back-port AOSP master version.

Pixelflinger doesn't have an x86 JIT, and the generic scanline()
routine is _extremely_ slow. This makes UI performance very slow.
when running full_x86-eng under emulation.

This patch back-ports the AOSP master pixelflinger into the
gingerbread branch. This essentially contains more than a dozen
new "shortcut" scanline functions used to speed-up the pixel
operations that are mostly used with the normal widget UI.

GL applications (e.g. Gallery3D) will still be very slow, but
this makes the home screen, app launcher, settings and many
simple applications work with very usable performance.

Change-Id: Ia64d8dd2e46820f07e98c4c9a4ec831b59fc61e4

14 files changed, 975 insertions, 119 deletions

diff --git a/libpixelflinger/Android.mk b/libpixelflinger/Android.mk
index 6491d24..ed2ab5e 100644
--- a/libpixelflinger/Android.mk
+++ b/libpixelflinger/Android.mk
@@ -2,17 +2,6 @@ LOCAL_PATH:= $(call my-dir)
 include $(CLEAR_VARS)
 
 #
-# ARMv6 specific objects
-#
-
-ifeq ($(TARGET_ARCH),arm)
-LOCAL_ASFLAGS := -march=armv6
-LOCAL_SRC_FILES := rotate90CW_4x4_16v6.S
-LOCAL_MODULE := libpixelflinger_armv6
-include $(BUILD_STATIC_LIBRARY)
-endif
-
-#
 # C/C++ and ARMv5 objects
 #
 
@@ -77,10 +66,6 @@ ifneq ($(BUILD_TINY_ANDROID),true)
 LOCAL_SHARED_LIBRARIES += libhardware_legacy
 LOCAL_CFLAGS += -DWITH_LIB_HARDWARE
 endif
-
-ifeq ($(TARGET_ARCH),arm)
-LOCAL_WHOLE_STATIC_LIBRARIES := libpixelflinger_armv6
-endif
 include $(BUILD_SHARED_LIBRARY)
 
 #
@@ -91,9 +76,6 @@ include $(CLEAR_VARS)
 LOCAL_MODULE:= libpixelflinger_static
 LOCAL_SRC_FILES := $(PIXELFLINGER_SRC_FILES)
 LOCAL_CFLAGS := $(PIXELFLINGER_CFLAGS) 
-ifeq ($(TARGET_ARCH),arm)
-LOCAL_WHOLE_STATIC_LIBRARIES := libpixelflinger_armv6
-endif
 include $(BUILD_STATIC_LIBRARY)
 
 
diff --git a/libpixelflinger/codeflinger/ARMAssembler.cpp b/libpixelflinger/codeflinger/ARMAssembler.cpp
index d3720c3..4726a08 100644
--- a/libpixelflinger/codeflinger/ARMAssembler.cpp
+++ b/libpixelflinger/codeflinger/ARMAssembler.cpp
@@ -334,7 +334,7 @@ void ARMAssembler::LDM(int cc, int dir,
 
 void ARMAssembler::STM(int cc, int dir,
         int Rn, int W, uint32_t reg_list)
-{   //                    FA EA FD ED      IB IA DB DA
+{   //                    ED FD EA FA      IB IA DB DA
     const uint8_t P[8] = { 0, 1, 0, 1,      1, 0, 1, 0 };
     const uint8_t U[8] = { 0, 0, 1, 1,      1, 1, 0, 0 };
     *mPC++ = (cc<<28) | (4<<25) | (uint32_t(P[dir])<<24) |
@@ -433,6 +433,16 @@ void ARMAssembler::UXTB16(int cc, int Rd, int Rm, int rotate)
 {
     *mPC++ = (cc<<28) | 0x6CF0070 | (Rd<<12) | ((rotate >> 3) << 10) | Rm;
 }
+#if 0
+#pragma mark -
+#pragma mark Bit manipulation (ARMv7+ only)...
+#endif
+
+// Bit manipulation (ARMv7+ only)...
+void ARMAssembler::UBFX(int cc, int Rd, int Rn, int lsb, int width)
+{
+    *mPC++ = (cc<<28) | 0x7E00000 | ((width-1)<<16) | (Rd<<12) | (lsb<<7) | 0x50 | Rn;
+}
 
 }; // namespace android
 
diff --git a/libpixelflinger/codeflinger/ARMAssembler.h b/libpixelflinger/codeflinger/ARMAssembler.h
index a667cb5..e7f038a 100644
--- a/libpixelflinger/codeflinger/ARMAssembler.h
+++ b/libpixelflinger/codeflinger/ARMAssembler.h
@@ -124,6 +124,7 @@ public:
     virtual void SMLAW(int cc, int y,
                 int Rd, int Rm, int Rs, int Rn);
     virtual void UXTB16(int cc, int Rd, int Rm, int rotate);
+    virtual void UBFX(int cc, int Rd, int Rn, int lsb, int width);
 
 private:
                 ARMAssembler(const ARMAssembler& rhs);
diff --git a/libpixelflinger/codeflinger/ARMAssemblerInterface.h b/libpixelflinger/codeflinger/ARMAssemblerInterface.h
index ff6af2a..796342a 100644
--- a/libpixelflinger/codeflinger/ARMAssemblerInterface.h
+++ b/libpixelflinger/codeflinger/ARMAssemblerInterface.h
@@ -206,6 +206,9 @@ public:
     // byte/half word extract...
     virtual void UXTB16(int cc, int Rd, int Rm, int rotate) = 0;
 
+    // bit manipulation...
+    virtual void UBFX(int cc, int Rd, int Rn, int lsb, int width) = 0;
+
     // -----------------------------------------------------------------------
     // convenience...
     // -----------------------------------------------------------------------
diff --git a/libpixelflinger/codeflinger/ARMAssemblerProxy.cpp b/libpixelflinger/codeflinger/ARMAssemblerProxy.cpp
index 7c422db..c57d7da 100644
--- a/libpixelflinger/codeflinger/ARMAssemblerProxy.cpp
+++ b/libpixelflinger/codeflinger/ARMAssemblerProxy.cpp
@@ -199,5 +199,9 @@ void ARMAssemblerProxy::UXTB16(int cc, int Rd, int Rm, int rotate) {
     mTarget->UXTB16(cc, Rd, Rm, rotate);
 }
 
+void ARMAssemblerProxy::UBFX(int cc, int Rd, int Rn, int lsb, int width) {
+    mTarget->UBFX(cc, Rd, Rn, lsb, width);
+}
+
 }; // namespace android
 
diff --git a/libpixelflinger/codeflinger/ARMAssemblerProxy.h b/libpixelflinger/codeflinger/ARMAssemblerProxy.h
index 9134cce..8c7f270 100644
--- a/libpixelflinger/codeflinger/ARMAssemblerProxy.h
+++ b/libpixelflinger/codeflinger/ARMAssemblerProxy.h
@@ -115,6 +115,7 @@ public:
                 int Rd, int Rm, int Rs, int Rn);
 
     virtual void UXTB16(int cc, int Rd, int Rm, int rotate);
+    virtual void UBFX(int cc, int Rd, int Rn, int lsb, int width);
 
 private:
     ARMAssemblerInterface*  mTarget;
diff --git a/libpixelflinger/codeflinger/disassem.c b/libpixelflinger/codeflinger/disassem.c
index c17f3ec..aeb8034 100644
--- a/libpixelflinger/codeflinger/disassem.c
+++ b/libpixelflinger/codeflinger/disassem.c
@@ -81,6 +81,8 @@
  * g - 2nd fp operand (register) (bits 16-18)
  * h - 3rd fp operand (register/immediate) (bits 0-4)
  * j - xtb rotate literal (bits 10-11)
+ * i - bfx lsb literal (bits 7-11)
+ * w - bfx width literal (bits 16-20)
  * b - branch address
  * t - thumb branch address (bits 24, 0-23)
  * k - breakpoint comment (bits 0-3, 8-19)
@@ -124,6 +126,7 @@ static const struct arm32_insn arm32_i[] = {
     { 0x0fe000f0, 0x00a00090, "umlal",	"Sdnms" },
     { 0x0fe000f0, 0x00e00090, "smlal",	"Sdnms" },
     { 0x0fff03f0, 0x06cf0070, "uxtb16", "dmj" },
+    { 0x0fe00070, 0x07e00050, "ubfx",   "dmiw" },
     { 0x0d700000, 0x04200000, "strt",	"daW" },
     { 0x0d700000, 0x04300000, "ldrt",	"daW" },
     { 0x0d700000, 0x04600000, "strbt",	"daW" },
@@ -412,6 +415,14 @@ disasm(const disasm_interface_t *di, u_int loc, int altfmt)
 		case 'j':
 			di->di_printf("ror #%d", ((insn >> 10) & 3) << 3);
 			break;
+        /* i - bfx lsb literal (bits 7-11) */
+        case 'i':
+            di->di_printf("#%d", (insn >> 7) & 31);
+            break;
+        /* w - bfx width literal (bits 16-20) */
+        case 'w':
+            di->di_printf("#%d", 1 + ((insn >> 16) & 31));
+            break;
 		/* b - branch address */
 		case 'b':
 			branch = ((insn << 2) & 0x03ffffff);
diff --git a/libpixelflinger/codeflinger/load_store.cpp b/libpixelflinger/codeflinger/load_store.cpp
index 93c5825..ed20a00 100644
--- a/libpixelflinger/codeflinger/load_store.cpp
+++ b/libpixelflinger/codeflinger/load_store.cpp
@@ -18,9 +18,12 @@
 #include <assert.h>
 #include <stdio.h>
 #include <cutils/log.h>
-
 #include "codeflinger/GGLAssembler.h"
 
+#ifdef __ARM_ARCH__
+#include <machine/cpu-features.h>
+#endif
+
 namespace android {
 
 // ----------------------------------------------------------------------------
@@ -110,6 +113,20 @@ void GGLAssembler::extract(integer_t& d, int s, int h, int l, int bits)
     assert(maskLen<=8);
     assert(h);
     
+#if __ARM_ARCH__ >= 7
+    const int mask = (1<<maskLen)-1;
+    if ((h == bits) && !l && (s != d.reg)) {
+        MOV(AL, 0, d.reg, s);                   // component = packed;
+    } else if ((h == bits) && l) {
+        MOV(AL, 0, d.reg, reg_imm(s, LSR, l));  // component = packed >> l;
+    } else if (!l && isValidImmediate(mask)) {
+        AND(AL, 0, d.reg, s, imm(mask));        // component = packed & mask;
+    } else if (!l && isValidImmediate(~mask)) {
+        BIC(AL, 0, d.reg, s, imm(~mask));       // component = packed & mask;
+    } else {
+        UBFX(AL, d.reg, s, l, maskLen);         // component = (packed & mask) >> l;
+    }
+#else
     if (h != bits) {
         const int mask = ((1<<maskLen)-1) << l;
         if (isValidImmediate(mask)) {
@@ -132,6 +149,7 @@ void GGLAssembler::extract(integer_t& d, int s, int h, int l, int bits)
     if (s != d.reg) {
         MOV(AL, 0, d.reg, s);
     }
+#endif
 
     d.s = maskLen;
 }
diff --git a/libpixelflinger/col32cb16blend.S b/libpixelflinger/col32cb16blend.S
index 1450bde..1831255 100644
--- a/libpixelflinger/col32cb16blend.S
+++ b/libpixelflinger/col32cb16blend.S
@@ -1,20 +1,19 @@
 /* libs/pixelflinger/col32cb16blend.S
-**
-** (C) COPYRIGHT 2009 ARM Limited.
-**
-** 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.
-**
-*/
+ *
+ * Copyright (C) 2009 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.
+ */
 
     .text
     .align
diff --git a/libpixelflinger/col32cb16blend_neon.S b/libpixelflinger/col32cb16blend_neon.S
index 17b0d01..cbd54d1 100644
--- a/libpixelflinger/col32cb16blend_neon.S
+++ b/libpixelflinger/col32cb16blend_neon.S
@@ -1,20 +1,20 @@
 /* libs/pixelflinger/col32cb16blend_neon.S
-**
-** (C) COPYRIGHT 2009 ARM Limited.
-**
-** 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.
-**
-*/
+ *
+ * Copyright (C) 2009 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.
+ */
+
 
     .text
     .align
diff --git a/libpixelflinger/raster.cpp b/libpixelflinger/raster.cpp
index d751202..32b2a97 100644
--- a/libpixelflinger/raster.cpp
+++ b/libpixelflinger/raster.cpp
@@ -143,7 +143,7 @@ void ggl_copyPixels(void* con, GGLint xs, GGLint ys,
 
 using namespace android;
 
-GGLint gglBitBlti(GGLContext* con, int tmu, GGLint crop[4], GGLint where[4])
+GGLint gglBitBlit(GGLContext* con, int tmu, GGLint crop[4], GGLint where[4])
 {
     GGL_CONTEXT(c, (void*)con);
 
diff --git a/libpixelflinger/scanline.cpp b/libpixelflinger/scanline.cpp
index a2f43eb..8fba147 100644
--- a/libpixelflinger/scanline.cpp
+++ b/libpixelflinger/scanline.cpp
@@ -1,6 +1,6 @@
 /* libs/pixelflinger/scanline.cpp
 **
-** Copyright 2006, The Android Open Source Project
+** Copyright 2006-2011, 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. 
@@ -57,6 +57,11 @@
 
 #define DEBUG__CODEGEN_ONLY     0
 
+/* Set to 1 to dump to the log the states that need a new
+ * code-generated scanline callback, i.e. those that don't
+ * have a corresponding shortcut function.
+ */
+#define DEBUG_NEEDS  0
 
 #define ASSEMBLY_SCRATCH_SIZE   2048
 
@@ -79,8 +84,21 @@ 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_t32cb16blend_dither(context_t* c);
+static void scanline_t32cb16blend_srca(context_t* c);
+static void scanline_t32cb16blend_clamp(context_t* c);
+static void scanline_t32cb16blend_clamp_dither(context_t* c);
+static void scanline_t32cb16blend_clamp_mod(context_t* c);
+static void scanline_x32cb16blend_clamp_mod(context_t* c);
+static void scanline_t32cb16blend_clamp_mod_dither(context_t* c);
+static void scanline_x32cb16blend_clamp_mod_dither(context_t* c);
 static void scanline_t32cb16(context_t* c);
+static void scanline_t32cb16_dither(context_t* c);
+static void scanline_t32cb16_clamp(context_t* c);
+static void scanline_t32cb16_clamp_dither(context_t* c);
 static void scanline_col32cb16blend(context_t* c);
+static void scanline_t16cb16_clamp(context_t* c);
+static void scanline_t16cb16blend_clamp_mod(context_t* c);
 static void scanline_memcpy(context_t* c);
 static void scanline_memset8(context_t* c);
 static void scanline_memset16(context_t* c);
@@ -99,6 +117,13 @@ extern "C" void scanline_col32cb16blend_arm(uint16_t *dst, uint32_t col, size_t
 
 // ----------------------------------------------------------------------------
 
+static inline uint16_t  convertAbgr8888ToRgb565(uint32_t  pix)
+{
+    return uint16_t( ((pix << 8) & 0xf800) |
+                      ((pix >> 5) & 0x07e0) |
+                      ((pix >> 19) & 0x001f) );
+}
+
 struct shortcut_t {
     needs_filter_t  filter;
     const char*     desc;
@@ -107,13 +132,95 @@ struct shortcut_t {
 };
 
 // Keep in sync with needs
+
+/* To understand the values here, have a look at:
+ *     system/core/include/private/pixelflinger/ggl_context.h
+ *
+ * Especially the lines defining and using GGL_RESERVE_NEEDS
+ *
+ * Quick reminders:
+ *   - the last nibble of the first value is the destination buffer format.
+ *   - the last nibble of the third value is the source texture format
+ *   - formats: 4=rgb565 1=abgr8888 2=xbgr8888
+ *
+ * In the descriptions below:
+ *
+ *   SRC      means we copy the source pixels to the destination
+ *
+ *   SRC_OVER means we blend the source pixels to the destination
+ *            with dstFactor = 1-srcA, srcFactor=1  (premultiplied source).
+ *            This mode is otherwise called 'blend'.
+ *
+ *   SRCA_OVER means we blend the source pixels to the destination
+ *             with dstFactor=srcA*(1-srcA) srcFactor=srcA (non-premul source).
+ *             This mode is otherwise called 'blend_srca'
+ *
+ *   clamp    means we fetch source pixels from a texture with u/v clamping
+ *
+ *   mod      means the source pixels are modulated (multiplied) by the
+ *            a/r/g/b of the current context's color. Typically used for
+ *            fade-in / fade-out.
+ *
+ *   dither   means we dither 32 bit values to 16 bits
+ */
 static shortcut_t shortcuts[] = {
     { { { 0x03515104, 0x00000077, { 0x00000A01, 0x00000000 } },
         { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
-        "565 fb, 8888 tx, blend", scanline_t32cb16blend, init_y_noop },
+        "565 fb, 8888 tx, blend SRC_OVER", scanline_t32cb16blend, init_y_noop },
     { { { 0x03010104, 0x00000077, { 0x00000A01, 0x00000000 } },
         { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
-        "565 fb, 8888 tx", scanline_t32cb16, init_y_noop  },  
+        "565 fb, 8888 tx, SRC", scanline_t32cb16, init_y_noop  },
+    /* same as first entry, but with dithering */
+    { { { 0x03515104, 0x00000177, { 0x00000A01, 0x00000000 } },
+        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
+        "565 fb, 8888 tx, blend SRC_OVER dither", scanline_t32cb16blend_dither, init_y_noop },
+    /* same as second entry, but with dithering */
+    { { { 0x03010104, 0x00000177, { 0x00000A01, 0x00000000 } },
+        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
+        "565 fb, 8888 tx, SRC dither", scanline_t32cb16_dither, init_y_noop  },
+    /* this is used during the boot animation - CHEAT: ignore dithering */
+    { { { 0x03545404, 0x00000077, { 0x00000A01, 0x00000000 } },
+        { 0xFFFFFFFF, 0xFFFFFEFF, { 0xFFFFFFFF, 0x0000003F } } },
+        "565 fb, 8888 tx, blend dst:ONE_MINUS_SRCA src:SRCA", scanline_t32cb16blend_srca, init_y_noop },
+    /* special case for arbitrary texture coordinates (think scaling) */
+    { { { 0x03515104, 0x00000077, { 0x00000001, 0x00000000 } },
+        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
+        "565 fb, 8888 tx, SRC_OVER clamp", scanline_t32cb16blend_clamp, init_y },
+    { { { 0x03515104, 0x00000177, { 0x00000001, 0x00000000 } },
+        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
+        "565 fb, 8888 tx, SRC_OVER clamp dither", scanline_t32cb16blend_clamp_dither, init_y },
+    /* another case used during emulation */
+    { { { 0x03515104, 0x00000077, { 0x00001001, 0x00000000 } },
+        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
+        "565 fb, 8888 tx, SRC_OVER clamp modulate", scanline_t32cb16blend_clamp_mod, init_y },
+    /* and this */
+    { { { 0x03515104, 0x00000077, { 0x00001002, 0x00000000 } },
+        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
+        "565 fb, x888 tx, SRC_OVER clamp modulate", scanline_x32cb16blend_clamp_mod, init_y },
+    { { { 0x03515104, 0x00000177, { 0x00001001, 0x00000000 } },
+        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
+        "565 fb, 8888 tx, SRC_OVER clamp modulate dither", scanline_t32cb16blend_clamp_mod_dither, init_y },
+    { { { 0x03515104, 0x00000177, { 0x00001002, 0x00000000 } },
+        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
+        "565 fb, x888 tx, SRC_OVER clamp modulate dither", scanline_x32cb16blend_clamp_mod_dither, init_y },
+    { { { 0x03010104, 0x00000077, { 0x00000001, 0x00000000 } },
+        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
+        "565 fb, 8888 tx, SRC clamp", scanline_t32cb16_clamp, init_y  },
+    { { { 0x03010104, 0x00000077, { 0x00000002, 0x00000000 } },
+        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
+        "565 fb, x888 tx, SRC clamp", scanline_t32cb16_clamp, init_y  },
+    { { { 0x03010104, 0x00000177, { 0x00000001, 0x00000000 } },
+        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
+        "565 fb, 8888 tx, SRC clamp dither", scanline_t32cb16_clamp_dither, init_y  },
+    { { { 0x03010104, 0x00000177, { 0x00000002, 0x00000000 } },
+        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
+        "565 fb, x888 tx, SRC clamp dither", scanline_t32cb16_clamp_dither, init_y  },
+    { { { 0x03010104, 0x00000077, { 0x00000004, 0x00000000 } },
+        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
+        "565 fb, 565 tx, SRC clamp", scanline_t16cb16_clamp, init_y  },
+    { { { 0x03515104, 0x00000077, { 0x00001004, 0x00000000 } },
+        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
+        "565 fb, 565 tx, SRC_OVER clamp", scanline_t16cb16blend_clamp_mod, init_y  },
     { { { 0x03515104, 0x00000077, { 0x00000000, 0x00000000 } },
         { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0xFFFFFFFF } } },
         "565 fb, 8888 fixed color", scanline_col32cb16blend, init_y_packed  },  
@@ -243,6 +350,12 @@ static void pick_scanline(context_t* c)
         }
     }
 
+#ifdef DEBUG_NEEDS
+    LOGI("Needs: n=0x%08x p=0x%08x t0=0x%08x t1=0x%08x",
+         c->state.needs.n, c->state.needs.p,
+         c->state.needs.t[0], c->state.needs.t[1]);
+#endif
+
 #endif // DEBUG__CODEGEN_ONLY
 
     c->init_y = init_y;
@@ -797,6 +910,678 @@ discard:
 #pragma mark Scanline
 #endif
 
+/* Used to parse a 32-bit source texture linearly. Usage is:
+ *
+ * horz_iterator32  hi(context);
+ * while (...) {
+ *    uint32_t  src_pixel = hi.get_pixel32();
+ *    ...
+ * }
+ *
+ * Use only for one-to-one texture mapping.
+ */
+struct horz_iterator32 {
+    horz_iterator32(context_t* c) {
+        const int x = c->iterators.xl;
+        const int y = c->iterators.y;
+        texture_t& tx = c->state.texture[0];
+        const int32_t u = (tx.shade.is0>>16) + x;
+        const int32_t v = (tx.shade.it0>>16) + y;
+        m_src = reinterpret_cast<uint32_t*>(tx.surface.data)+(u+(tx.surface.stride*v));
+    }
+    uint32_t  get_pixel32() {
+        return *m_src++;
+    }
+protected:
+    uint32_t* m_src;
+};
+
+/* A variant for 16-bit source textures. */
+struct horz_iterator16 {
+    horz_iterator16(context_t* c) {
+        const int x = c->iterators.xl;
+        const int y = c->iterators.y;
+        texture_t& tx = c->state.texture[0];
+        const int32_t u = (tx.shade.is0>>16) + x;
+        const int32_t v = (tx.shade.it0>>16) + y;
+        m_src = reinterpret_cast<uint16_t*>(tx.surface.data)+(u+(tx.surface.stride*v));
+    }
+    uint16_t  get_pixel16() {
+        return *m_src++;
+    }
+protected:
+    uint16_t* m_src;
+};
+
+/* A clamp iterator is used to iterate inside a texture with GGL_CLAMP.
+ * After initialization, call get_src16() or get_src32() to get the current
+ * texture pixel value.
+ */
+struct clamp_iterator {
+    clamp_iterator(context_t* c) {
+        const int xs = c->iterators.xl;
+        texture_t& tx = c->state.texture[0];
+        texture_iterators_t& ti = tx.iterators;
+        m_s = (xs * ti.dsdx) + ti.ydsdy;
+        m_t = (xs * ti.dtdx) + ti.ydtdy;
+        m_ds = ti.dsdx;
+        m_dt = ti.dtdx;
+        m_width_m1 = tx.surface.width - 1;
+        m_height_m1 = tx.surface.height - 1;
+        m_data = tx.surface.data;
+        m_stride = tx.surface.stride;
+    }
+    uint16_t get_pixel16() {
+        int  u, v;
+        get_uv(u, v);
+        uint16_t* src = reinterpret_cast<uint16_t*>(m_data) + (u + (m_stride*v));
+        return src[0];
+    }
+    uint32_t get_pixel32() {
+        int  u, v;
+        get_uv(u, v);
+        uint32_t* src = reinterpret_cast<uint32_t*>(m_data) + (u + (m_stride*v));
+        return src[0];
+    }
+private:
+    void   get_uv(int& u, int& v) {
+        int  uu = m_s >> 16;
+        int  vv = m_t >> 16;
+        if (uu < 0)
+            uu = 0;
+        if (uu > m_width_m1)
+            uu = m_width_m1;
+        if (vv < 0)
+            vv = 0;
+        if (vv > m_height_m1)
+            vv = m_height_m1;
+        u = uu;
+        v = vv;
+        m_s += m_ds;
+        m_t += m_dt;
+    }
+
+    GGLfixed  m_s, m_t;
+    GGLfixed  m_ds, m_dt;
+    int       m_width_m1, m_height_m1;
+    uint8_t*  m_data;
+    int       m_stride;
+};
+
+/*
+ * The 'horizontal clamp iterator' variant corresponds to the case where
+ * the 'v' coordinate doesn't change. This is useful to avoid one mult and
+ * extra adds / checks per pixels, if the blending/processing operation after
+ * this is very fast.
+ */
+static int is_context_horizontal(const context_t* c) {
+    return (c->state.texture[0].iterators.dtdx == 0);
+}
+
+struct horz_clamp_iterator {
+    uint16_t  get_pixel16() {
+        int  u = m_s >> 16;
+        m_s += m_ds;
+        if (u < 0)
+            u = 0;
+        if (u > m_width_m1)
+            u = m_width_m1;
+        const uint16_t* src = reinterpret_cast<const uint16_t*>(m_data);
+        return src[u];
+    }
+    uint32_t  get_pixel32() {
+        int  u = m_s >> 16;
+        m_s += m_ds;
+        if (u < 0)
+            u = 0;
+        if (u > m_width_m1)
+            u = m_width_m1;
+        const uint32_t* src = reinterpret_cast<const uint32_t*>(m_data);
+        return src[u];
+    }
+protected:
+    void init(const context_t* c, int shift);
+    GGLfixed       m_s;
+    GGLfixed       m_ds;
+    int            m_width_m1;
+    const uint8_t* m_data;
+};
+
+void horz_clamp_iterator::init(const context_t* c, int shift)
+{
+    const int xs = c->iterators.xl;
+    const texture_t& tx = c->state.texture[0];
+    const texture_iterators_t& ti = tx.iterators;
+    m_s = (xs * ti.dsdx) + ti.ydsdy;
+    m_ds = ti.dsdx;
+    m_width_m1 = tx.surface.width-1;
+    m_data = tx.surface.data;
+
+    GGLfixed t = (xs * ti.dtdx) + ti.ydtdy;
+    int      v = t >> 16;
+    if (v < 0)
+        v = 0;
+    else if (v >= (int)tx.surface.height)
+        v = (int)tx.surface.height-1;
+
+    m_data += (tx.surface.stride*v) << shift;
+}
+
+struct horz_clamp_iterator16 : horz_clamp_iterator {
+    horz_clamp_iterator16(const context_t* c) {
+        init(c,1);
+    };
+};
+
+struct horz_clamp_iterator32 : horz_clamp_iterator {
+    horz_clamp_iterator32(context_t* c) {
+        init(c,2);
+    };
+};
+
+/* This is used to perform dithering operations.
+ */
+struct ditherer {
+    ditherer(const context_t* c) {
+        const int x = c->iterators.xl;
+        const int y = c->iterators.y;
+        m_line = &c->ditherMatrix[ ((y & GGL_DITHER_MASK)<<GGL_DITHER_ORDER_SHIFT) ];
+        m_index = x & GGL_DITHER_MASK;
+    }
+    void step(void) {
+        m_index++;
+    }
+    int  get_value(void) {
+        int ret = m_line[m_index & GGL_DITHER_MASK];
+        m_index++;
+        return ret;
+    }
+    uint16_t abgr8888ToRgb565(uint32_t s) {
+        uint32_t r = s & 0xff;
+        uint32_t g = (s >> 8) & 0xff;
+        uint32_t b = (s >> 16) & 0xff;
+        return rgb888ToRgb565(r,g,b);
+    }
+    /* The following assumes that r/g/b are in the 0..255 range each */
+    uint16_t rgb888ToRgb565(uint32_t& r, uint32_t& g, uint32_t &b) {
+        int threshold = get_value();
+        /* dither in on GGL_DITHER_BITS, and each of r, g, b is on 8 bits */
+        r += (threshold >> (GGL_DITHER_BITS-8 +5));
+        g += (threshold >> (GGL_DITHER_BITS-8 +6));
+        b += (threshold >> (GGL_DITHER_BITS-8 +5));
+        if (r > 0xff)
+            r = 0xff;
+        if (g > 0xff)
+            g = 0xff;
+        if (b > 0xff)
+            b = 0xff;
+        return uint16_t(((r & 0xf8) << 8) | ((g & 0xfc) << 3) | (b >> 3));
+    }
+protected:
+    const uint8_t* m_line;
+    int            m_index;
+};
+
+/* This structure is used to blend (SRC_OVER) 32-bit source pixels
+ * onto 16-bit destination ones. Usage is simply:
+ *
+ *   blender.blend(<32-bit-src-pixel-value>,<ptr-to-16-bit-dest-pixel>)
+ */
+struct blender_32to16 {
+    blender_32to16(context_t* c) { }
+    void write(uint32_t s, uint16_t* dst) {
+        if (s == 0)
+            return;
+        s = GGL_RGBA_TO_HOST(s);
+        int sA = (s>>24);
+        if (sA == 0xff) {
+            *dst = convertAbgr8888ToRgb565(s);
+        } else {
+            int f = 0x100 - (sA + (sA>>7));
+            int sR = (s >> (   3))&0x1F;
+            int sG = (s >> ( 8+2))&0x3F;
+            int sB = (s >> (16+3))&0x1F;
+            uint16_t d = *dst;
+            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);
+        }
+    }
+    void write(uint32_t s, uint16_t* dst, ditherer& di) {
+        if (s == 0) {
+            di.step();
+            return;
+        }
+        s = GGL_RGBA_TO_HOST(s);
+        int sA = (s>>24);
+        if (sA == 0xff) {
+            *dst = di.abgr8888ToRgb565(s);
+        } else {
+            int threshold = di.get_value() << (8 - GGL_DITHER_BITS);
+            int f = 0x100 - (sA + (sA>>7));
+            int sR = (s >> (   3))&0x1F;
+            int sG = (s >> ( 8+2))&0x3F;
+            int sB = (s >> (16+3))&0x1F;
+            uint16_t d = *dst;
+            int dR = (d>>11)&0x1f;
+            int dG = (d>>5)&0x3f;
+            int dB = (d)&0x1f;
+            sR = ((sR << 8) + f*dR + threshold)>>8;
+            sG = ((sG << 8) + f*dG + threshold)>>8;
+            sB = ((sB << 8) + f*dB + threshold)>>8;
+            if (sR > 0x1f) sR = 0x1f;
+            if (sG > 0x3f) sG = 0x3f;
+            if (sB > 0x1f) sB = 0x1f;
+            *dst = uint16_t((sR<<11)|(sG<<5)|sB);
+        }
+    }
+};
+
+/* This blender does the same for the 'blend_srca' operation.
+ * where dstFactor=srcA*(1-srcA) srcFactor=srcA
+ */
+struct blender_32to16_srcA {
+    blender_32to16_srcA(const context_t* c) { }
+    void write(uint32_t s, uint16_t* dst) {
+        if (!s) {
+            return;
+        }
+        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 f1 = (sA + (sA>>7));
+        int f2 = 0x100-f1;
+        int dR = (d>>11)&0x1f;
+        int dG = (d>>5)&0x3f;
+        int dB = (d)&0x1f;
+        sR = (f1*sR + f2*dR)>>8;
+        sG = (f1*sG + f2*dG)>>8;
+        sB = (f1*sB + f2*dB)>>8;
+        *dst = uint16_t((sR<<11)|(sG<<5)|sB);
+    }
+};
+
+/* Common init code the modulating blenders */
+struct blender_modulate {
+    void init(const context_t* c) {
+        const int r = c->iterators.ydrdy >> (GGL_COLOR_BITS-8);
+        const int g = c->iterators.ydgdy >> (GGL_COLOR_BITS-8);
+        const int b = c->iterators.ydbdy >> (GGL_COLOR_BITS-8);
+        const int a = c->iterators.ydady >> (GGL_COLOR_BITS-8);
+        m_r = r + (r >> 7);
+        m_g = g + (g >> 7);
+        m_b = b + (b >> 7);
+        m_a = a + (a >> 7);
+    }
+protected:
+    int m_r, m_g, m_b, m_a;
+};
+
+/* This blender does a normal blend after modulation.
+ */
+struct blender_32to16_modulate : blender_modulate {
+    blender_32to16_modulate(const context_t* c) {
+        init(c);
+    }
+    void write(uint32_t s, uint16_t* dst) {
+        // blend source and destination
+        if (!s) {
+            return;
+        }
+        s = GGL_RGBA_TO_HOST(s);
+
+        /* We need to modulate s */
+        uint32_t  sA = (s >> 24);
+        uint32_t  sB = (s >> 16) & 0xff;
+        uint32_t  sG = (s >> 8) & 0xff;
+        uint32_t  sR = s & 0xff;
+
+        sA = (sA*m_a) >> 8;
+        /* Keep R/G/B scaled to 5.8 or 6.8 fixed float format */
+        sR = (sR*m_r) >> (8 - 5);
+        sG = (sG*m_g) >> (8 - 6);
+        sB = (sB*m_b) >> (8 - 5);
+
+        /* Now do a normal blend */
+        int f = 0x100 - (sA + (sA>>7));
+        uint16_t d = *dst;
+        int dR = (d>>11)&0x1f;
+        int dG = (d>>5)&0x3f;
+        int dB = (d)&0x1f;
+        sR = (sR + f*dR)>>8;
+        sG = (sG + f*dG)>>8;
+        sB = (sB + f*dB)>>8;
+        *dst = uint16_t((sR<<11)|(sG<<5)|sB);
+    }
+    void write(uint32_t s, uint16_t* dst, ditherer& di) {
+        // blend source and destination
+        if (!s) {
+            di.step();
+            return;
+        }
+        s = GGL_RGBA_TO_HOST(s);
+
+        /* We need to modulate s */
+        uint32_t  sA = (s >> 24);
+        uint32_t  sB = (s >> 16) & 0xff;
+        uint32_t  sG = (s >> 8) & 0xff;
+        uint32_t  sR = s & 0xff;
+
+        sA = (sA*m_a) >> 8;
+        /* keep R/G/B scaled to 5.8 or 6.8 fixed float format */
+        sR = (sR*m_r) >> (8 - 5);
+        sG = (sG*m_g) >> (8 - 6);
+        sB = (sB*m_b) >> (8 - 5);
+
+        /* Scale threshold to 0.8 fixed float format */
+        int threshold = di.get_value() << (8 - GGL_DITHER_BITS);
+        int f = 0x100 - (sA + (sA>>7));
+        uint16_t d = *dst;
+        int dR = (d>>11)&0x1f;
+        int dG = (d>>5)&0x3f;
+        int dB = (d)&0x1f;
+        sR = (sR + f*dR + threshold)>>8;
+        sG = (sG + f*dG + threshold)>>8;
+        sB = (sB + f*dB + threshold)>>8;
+        if (sR > 0x1f) sR = 0x1f;
+        if (sG > 0x3f) sG = 0x3f;
+        if (sB > 0x1f) sB = 0x1f;
+        *dst = uint16_t((sR<<11)|(sG<<5)|sB);
+    }
+};
+
+/* same as 32to16_modulate, except that the input is xRGB, instead of ARGB */
+struct blender_x32to16_modulate : blender_modulate {
+    blender_x32to16_modulate(const context_t* c) {
+        init(c);
+    }
+    void write(uint32_t s, uint16_t* dst) {
+        s = GGL_RGBA_TO_HOST(s);
+
+        uint32_t  sB = (s >> 16) & 0xff;
+        uint32_t  sG = (s >> 8) & 0xff;
+        uint32_t  sR = s & 0xff;
+
+        /* Keep R/G/B in 5.8 or 6.8 format */
+        sR = (sR*m_r) >> (8 - 5);
+        sG = (sG*m_g) >> (8 - 6);
+        sB = (sB*m_b) >> (8 - 5);
+
+        int f = 0x100 - m_a;
+        uint16_t d = *dst;
+        int dR = (d>>11)&0x1f;
+        int dG = (d>>5)&0x3f;
+        int dB = (d)&0x1f;
+        sR = (sR + f*dR)>>8;
+        sG = (sG + f*dG)>>8;
+        sB = (sB + f*dB)>>8;
+        *dst = uint16_t((sR<<11)|(sG<<5)|sB);
+    }
+    void write(uint32_t s, uint16_t* dst, ditherer& di) {
+        s = GGL_RGBA_TO_HOST(s);
+
+        uint32_t  sB = (s >> 16) & 0xff;
+        uint32_t  sG = (s >> 8) & 0xff;
+        uint32_t  sR = s & 0xff;
+
+        sR = (sR*m_r) >> (8 - 5);
+        sG = (sG*m_g) >> (8 - 6);
+        sB = (sB*m_b) >> (8 - 5);
+
+        /* Now do a normal blend */
+        int threshold = di.get_value() << (8 - GGL_DITHER_BITS);
+        int f = 0x100 - m_a;
+        uint16_t d = *dst;
+        int dR = (d>>11)&0x1f;
+        int dG = (d>>5)&0x3f;
+        int dB = (d)&0x1f;
+        sR = (sR + f*dR + threshold)>>8;
+        sG = (sG + f*dG + threshold)>>8;
+        sB = (sB + f*dB + threshold)>>8;
+        if (sR > 0x1f) sR = 0x1f;
+        if (sG > 0x3f) sG = 0x3f;
+        if (sB > 0x1f) sB = 0x1f;
+        *dst = uint16_t((sR<<11)|(sG<<5)|sB);
+    }
+};
+
+/* Same as above, but source is 16bit rgb565 */
+struct blender_16to16_modulate : blender_modulate {
+    blender_16to16_modulate(const context_t* c) {
+        init(c);
+    }
+    void write(uint16_t s16, uint16_t* dst) {
+        uint32_t  s = s16;
+
+        uint32_t  sR = s >> 11;
+        uint32_t  sG = (s >> 5) & 0x3f;
+        uint32_t  sB = s & 0x1f;
+
+        sR = (sR*m_r);
+        sG = (sG*m_g);
+        sB = (sB*m_b);
+
+        int f = 0x100 - m_a;
+        uint16_t d = *dst;
+        int dR = (d>>11)&0x1f;
+        int dG = (d>>5)&0x3f;
+        int dB = (d)&0x1f;
+        sR = (sR + f*dR)>>8;
+        sG = (sG + f*dG)>>8;
+        sB = (sB + f*dB)>>8;
+        *dst = uint16_t((sR<<11)|(sG<<5)|sB);
+    }
+};
+
+/* This is used to iterate over a 16-bit destination color buffer.
+ * Usage is:
+ *
+ *   dst_iterator16  di(context);
+ *   while (di.count--) {
+ *       <do stuff with dest pixel at di.dst>
+ *       di.dst++;
+ *   }
+ */
+struct dst_iterator16 {
+    dst_iterator16(const context_t* c) {
+        const int x = c->iterators.xl;
+        const int width = c->iterators.xr - x;
+        const int32_t y = c->iterators.y;
+        const surface_t* cb = &(c->state.buffers.color);
+        count = width;
+        dst = reinterpret_cast<uint16_t*>(cb->data) + (x+(cb->stride*y));
+    }
+    int        count;
+    uint16_t*  dst;
+};
+
+
+static void scanline_t32cb16_clamp(context_t* c)
+{
+    dst_iterator16  di(c);
+
+    if (is_context_horizontal(c)) {
+        /* Special case for simple horizontal scaling */
+        horz_clamp_iterator32 ci(c);
+        while (di.count--) {
+            uint32_t s = ci.get_pixel32();
+            *di.dst++ = convertAbgr8888ToRgb565(s);
+        }
+    } else {
+        /* General case */
+        clamp_iterator ci(c);
+        while (di.count--) {
+            uint32_t s = ci.get_pixel32();
+            *di.dst++ = convertAbgr8888ToRgb565(s);
+        }
+    }
+}
+
+static void scanline_t32cb16_dither(context_t* c)
+{
+    horz_iterator32 si(c);
+    dst_iterator16  di(c);
+    ditherer        dither(c);
+
+    while (di.count--) {
+        uint32_t s = si.get_pixel32();
+        *di.dst++ = dither.abgr8888ToRgb565(s);
+    }
+}
+
+static void scanline_t32cb16_clamp_dither(context_t* c)
+{
+    dst_iterator16  di(c);
+    ditherer        dither(c);
+
+    if (is_context_horizontal(c)) {
+        /* Special case for simple horizontal scaling */
+        horz_clamp_iterator32 ci(c);
+        while (di.count--) {
+            uint32_t s = ci.get_pixel32();
+            *di.dst++ = dither.abgr8888ToRgb565(s);
+        }
+    } else {
+        /* General case */
+        clamp_iterator ci(c);
+        while (di.count--) {
+            uint32_t s = ci.get_pixel32();
+            *di.dst++ = dither.abgr8888ToRgb565(s);
+        }
+    }
+}
+
+static void scanline_t32cb16blend_dither(context_t* c)
+{
+    dst_iterator16 di(c);
+    ditherer       dither(c);
+    blender_32to16 bl(c);
+    horz_iterator32  hi(c);
+    while (di.count--) {
+        uint32_t s = hi.get_pixel32();
+        bl.write(s, di.dst, dither);
+        di.dst++;
+    }
+}
+
+static void scanline_t32cb16blend_clamp(context_t* c)
+{
+    dst_iterator16  di(c);
+    blender_32to16  bl(c);
+
+    if (is_context_horizontal(c)) {
+        horz_clamp_iterator32 ci(c);
+        while (di.count--) {
+            uint32_t s = ci.get_pixel32();
+            bl.write(s, di.dst);
+            di.dst++;
+        }
+    } else {
+        clamp_iterator ci(c);
+        while (di.count--) {
+            uint32_t s = ci.get_pixel32();
+            bl.write(s, di.dst);
+            di.dst++;
+        }
+    }
+}
+
+static void scanline_t32cb16blend_clamp_dither(context_t* c)
+{
+    dst_iterator16 di(c);
+    ditherer       dither(c);
+    blender_32to16 bl(c);
+
+    clamp_iterator ci(c);
+    while (di.count--) {
+        uint32_t s = ci.get_pixel32();
+        bl.write(s, di.dst, dither);
+        di.dst++;
+    }
+}
+
+void scanline_t32cb16blend_clamp_mod(context_t* c)
+{
+    dst_iterator16 di(c);
+    blender_32to16_modulate bl(c);
+
+    clamp_iterator ci(c);
+    while (di.count--) {
+        uint32_t s = ci.get_pixel32();
+        bl.write(s, di.dst);
+        di.dst++;
+    }
+}
+
+void scanline_t32cb16blend_clamp_mod_dither(context_t* c)
+{
+    dst_iterator16 di(c);
+    blender_32to16_modulate bl(c);
+    ditherer dither(c);
+
+    clamp_iterator ci(c);
+    while (di.count--) {
+        uint32_t s = ci.get_pixel32();
+        bl.write(s, di.dst, dither);
+        di.dst++;
+    }
+}
+
+/* Variant of scanline_t32cb16blend_clamp_mod with a xRGB texture */
+void scanline_x32cb16blend_clamp_mod(context_t* c)
+{
+    dst_iterator16 di(c);
+    blender_x32to16_modulate  bl(c);
+
+    clamp_iterator ci(c);
+    while (di.count--) {
+        uint32_t s = ci.get_pixel32();
+        bl.write(s, di.dst);
+        di.dst++;
+    }
+}
+
+void scanline_x32cb16blend_clamp_mod_dither(context_t* c)
+{
+    dst_iterator16 di(c);
+    blender_x32to16_modulate  bl(c);
+    ditherer dither(c);
+
+    clamp_iterator ci(c);
+    while (di.count--) {
+        uint32_t s = ci.get_pixel32();
+        bl.write(s, di.dst, dither);
+        di.dst++;
+    }
+}
+
+void scanline_t16cb16_clamp(context_t* c)
+{
+    dst_iterator16  di(c);
+
+    /* Special case for simple horizontal scaling */
+    if (is_context_horizontal(c)) {
+        horz_clamp_iterator16 ci(c);
+        while (di.count--) {
+            *di.dst++ = ci.get_pixel16();
+        }
+    } else {
+        clamp_iterator ci(c);
+        while (di.count--) {
+            *di.dst++ = ci.get_pixel16();
+        }
+    }
+}
+
+
+
 template <typename T, typename U>
 static inline __attribute__((const))
 T interpolate(int y, T v0, U dvdx, U dvdy) {
@@ -1322,30 +2107,24 @@ void scanline_t32cb16(context_t* c)
     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);
+        *dst++ = convertAbgr8888ToRgb565(s);
         ct--;
     }
 
     while (ct >= 2) {
+#if BYTE_ORDER == BIG_ENDIAN
         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;
-        
+        d = convertAbgr8888ToRgb565_hi16(s);
+
+        s = GGL_RGBA_TO_HOST( *src++ );
+        d |= convertAbgr8888ToRgb565(s);
+#else
         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;
+        d = convertAbgr8888ToRgb565(s);
 
-#if BYTE_ORDER == BIG_ENDIAN
-        d = (d>>16) | (d<<16);
+        s = GGL_RGBA_TO_HOST( *src++ );
+        d |= convertAbgr8888ToRgb565(s) << 16;
 #endif
-
         *dst32++ = d;
         ct -= 2;
     }
@@ -1357,6 +2136,7 @@ last_one:
 
 void scanline_t32cb16blend(context_t* c)
 {
+#if ((ANDROID_CODEGEN >= ANDROID_CODEGEN_ASM) && defined(__arm__))
     int32_t x = c->iterators.xl;
     size_t ct = c->iterators.xr - x;
     int32_t y = c->iterators.y;
@@ -1368,33 +2148,55 @@ void scanline_t32cb16blend(context_t* c)
     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);
+    dst_iterator16  di(c);
+    horz_iterator32  hi(c);
+    blender_32to16  bl(c);
+    while (di.count--) {
+        uint32_t s = hi.get_pixel32();
+        bl.write(s, di.dst);
+        di.dst++;
     }
 #endif
 }
 
+void scanline_t32cb16blend_srca(context_t* c)
+{
+    dst_iterator16  di(c);
+    horz_iterator32  hi(c);
+    blender_32to16_srcA  blender(c);
+
+    while (di.count--) {
+        uint32_t s = hi.get_pixel32();
+        blender.write(s,di.dst);
+        di.dst++;
+    }
+}
+
+void scanline_t16cb16blend_clamp_mod(context_t* c)
+{
+    const int a = c->iterators.ydady >> (GGL_COLOR_BITS-8);
+    if (a == 0) {
+        return;
+    }
+
+    if (a == 255) {
+        scanline_t16cb16_clamp(c);
+        return;
+    }
+
+    dst_iterator16  di(c);
+    blender_16to16_modulate  blender(c);
+    clamp_iterator  ci(c);
+
+    while (di.count--) {
+        uint16_t s = ci.get_pixel16();
+        blender.write(s, di.dst);
+        di.dst++;
+    }
+}
+
 void scanline_memcpy(context_t* c)
 {
     int32_t x = c->iterators.xl;
@@ -1518,26 +2320,3 @@ void rect_memcpy(context_t* c, size_t 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, ASSEMBLY_SCRATCH_SIZE));
-    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
-}
-
diff --git a/libpixelflinger/tests/codegen/Android.mk b/libpixelflinger/tests/codegen/Android.mk
index 1bc4214..aa320fc 100644
--- a/libpixelflinger/tests/codegen/Android.mk
+++ b/libpixelflinger/tests/codegen/Android.mk
@@ -2,12 +2,15 @@ LOCAL_PATH:= $(call my-dir)
 include $(CLEAR_VARS)
 
 LOCAL_SRC_FILES:= \
-	codegen.cpp
+	codegen.cpp.arm
 
 LOCAL_SHARED_LIBRARIES := \
 	libcutils \
     libpixelflinger
 
+LOCAL_C_INCLUDES := \
+	system/core/libpixelflinger
+
 LOCAL_MODULE:= test-opengl-codegen
 
 LOCAL_MODULE_TAGS := tests
diff --git a/libpixelflinger/tests/codegen/codegen.cpp b/libpixelflinger/tests/codegen/codegen.cpp
index 1865888..94e2481 100644
--- a/libpixelflinger/tests/codegen/codegen.cpp
+++ b/libpixelflinger/tests/codegen/codegen.cpp
@@ -1,9 +1,54 @@
 #include <stdio.h>
 #include <stdint.h>
 
-extern "C" void ggl_test_codegen(
-        uint32_t n, uint32_t p, uint32_t t0, uint32_t t1);
+#include "private/pixelflinger/ggl_context.h"
 
+#include "buffer.h"
+#include "scanline.h"
+
+#include "codeflinger/CodeCache.h"
+#include "codeflinger/GGLAssembler.h"
+#include "codeflinger/ARMAssembler.h"
+
+#if defined(__arm__)
+#   define ANDROID_ARM_CODEGEN  1
+#else
+#   define ANDROID_ARM_CODEGEN  0
+#endif
+
+#define ASSEMBLY_SCRATCH_SIZE   2048
+
+using namespace android;
+
+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; }
+};
+
+static 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, ASSEMBLY_SCRATCH_SIZE));
+    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
+}
 
 int main(int argc, char** argv)
 {