Merge branch 'p-jb-mr0-release-hwc' into d-jb-mr1-release

Conflicts:
	hwc/hwc.c

Change-Id: Ice74d3e62a2d6ccadb2d4333f0acc39d6366e0e7
Signed-off-by: Daniel Levin <dendy@ti.com>

5 files changed, 2921 insertions, 131 deletions

diff --git a/hwc/Android.mk b/hwc/Android.mk
index c6a2eae..60b421d 100644
--- a/hwc/Android.mk
+++ b/hwc/Android.mk
@@ -1,4 +1,3 @@
-# HWC under heavy development and should not be included in builds for now
 LOCAL_PATH := $(call my-dir)
 
 # HAL module implementation, not prelinked and stored in
@@ -7,8 +6,9 @@ include $(CLEAR_VARS)
 LOCAL_PRELINK_MODULE := false
 LOCAL_ARM_MODE := arm
 LOCAL_MODULE_PATH := $(TARGET_OUT_SHARED_LIBRARIES)/../vendor/lib/hw
-LOCAL_SHARED_LIBRARIES := liblog libEGL libcutils libutils libhardware libhardware_legacy libz
-LOCAL_SRC_FILES := hwc.c
+LOCAL_SHARED_LIBRARIES := liblog libEGL libcutils libutils libhardware libhardware_legacy libz \
+                          libion
+LOCAL_SRC_FILES := hwc.c rgz_2d.c
 LOCAL_STATIC_LIBRARIES := libpng
 
 LOCAL_MODULE_TAGS := optional
@@ -16,6 +16,11 @@ LOCAL_MODULE_TAGS := optional
 LOCAL_MODULE := hwcomposer.omap4
 LOCAL_CFLAGS := -DLOG_TAG=\"ti_hwc\"
 LOCAL_C_INCLUDES += external/libpng external/zlib
+
+LOCAL_C_INCLUDES += \
+    $(LOCAL_PATH)/../edid/inc
+LOCAL_SHARED_LIBRARIES += libedid
+
 # LOG_NDEBUG=0 means verbose logging enabled
 # LOCAL_CFLAGS += -DLOG_NDEBUG=0
 include $(BUILD_SHARED_LIBRARY)
diff --git a/hwc/hal_public.h b/hwc/hal_public.h
index 3d64c4c..a7dfb08 100644
--- a/hwc/hal_public.h
+++ b/hwc/hal_public.h
@@ -164,4 +164,25 @@ typedef struct IMG_buffer_format_public_t
 }
 IMG_buffer_format_public_t;
 
+/*
+ * These are vendor specific pixel formats, by (informal) convention IMGTec
+ * formats start from the top of the range, TI formats start from the bottom
+ */
+#define HAL_PIXEL_FORMAT_BGRX_8888      0x1FF
+#define HAL_PIXEL_FORMAT_TI_NV12        0x100
+#define HAL_PIXEL_FORMAT_TI_UNUSED      0x101 /* Free for use */
+#define HAL_PIXEL_FORMAT_TI_NV12_1D     0x102
+
+#ifndef GRALLOC_USAGE_SYSTEM_HEAP
+#define GRALLOC_USAGE_SYSTEM_HEAP GRALLOC_USAGE_PRIVATE_0
+#else
+#error GRALLOC_USAGE_SYSTEM_HEAP should only be defined by hal_public.h
+#endif
+
+#ifndef GRALLOC_USAGE_PHYS_CONTIG
+#define GRALLOC_USAGE_PHYS_CONTIG GRALLOC_USAGE_PRIVATE_1
+#else
+#error GRALLOC_USAGE_PHYS_CONTIG should only be defined by hal_public.h
+#endif
 #endif /* HAL_PUBLIC_H */
+
diff --git a/hwc/hwc.c b/hwc/hwc.c
index 74012dd..af46847 100644
--- a/hwc/hwc.c
+++ b/hwc/hwc.c
@@ -25,6 +25,7 @@
 #include <linux/omapfb.h>
 #include <sys/mman.h>
 #include <sys/resource.h>
+#include <stdbool.h>
 
 #include <cutils/properties.h>
 #include <cutils/log.h>
@@ -35,8 +36,15 @@
 #include <EGL/egl.h>
 #include <hardware_legacy/uevent.h>
 #include <png.h>
+#include <utils/Timers.h>
 
 #include <system/graphics.h>
+#include <ui/S3DFormat.h>
+#include <edid_parser.h>
+
+#include <linux/bltsville.h>
+
+#define MAX_HWC_LAYERS 32
 
 #define ASPECT_RATIO_TOLERANCE 0.02f
 
@@ -50,15 +58,18 @@
 #define DIV_ROUND_UP(a, b) (((a) + (b) - 1) / (b))
 
 #include <video/dsscomp.h>
+#include <video/omap_hwc.h>
 
 #include "hal_public.h"
+#include "rgz_2d.h"
+
+#include <linux/ion.h>
+#include <linux/omap_ion.h>
+#include <ion/ion.h>
 
 #define MAX_HW_OVERLAYS 4
 #define NUM_NONSCALING_OVERLAYS 1
-#define HAL_PIXEL_FORMAT_BGRX_8888      0x1FF
-#define HAL_PIXEL_FORMAT_TI_NV12        0x100
-#define HAL_PIXEL_FORMAT_TI_NV12_PADDED 0x101
-#define MAX_TILER_SLOT (16 << 20)
+#define NUM_EXT_DISPLAY_BACK_BUFFERS 2
 
 struct ext_transform_t {
     __u8 rotation : 3;          /* 90-degree clockwise rotations */
@@ -94,6 +105,11 @@ struct omap4_hwc_ext {
     __u32 yres;
     float m[2][3];                      /* external transformation matrix */
     hwc_rect_t mirror_region;           /* region of screen to mirror */
+
+    bool s3d_enabled;
+    bool s3d_capable;
+    enum S3DLayoutType s3d_type;
+    enum S3DLayoutOrder s3d_order;
 };
 typedef struct omap4_hwc_ext omap4_hwc_ext_t;
 
@@ -103,6 +119,17 @@ enum {
     EXT_HFLIP       = (1 << 2), /* flip l-r on output (after rotation) */
 };
 
+enum bltpolicy {
+    BLTPOLICY_DISABLED = 0,
+    BLTPOLICY_DEFAULT = 1,    /* Default blit policy */
+    BLTPOLICY_ALL,            /* Test mode to attempt to blit all */
+};
+
+enum bltmode {
+    BLTMODE_PAINT = 0,    /* Attempt to blit layer by layer */
+    BLTMODE_REGION = 1,   /* Attempt to blit layers via regions */
+};
+
 /* ARGB image */
 struct omap4_hwc_img {
     int width;
@@ -119,6 +146,22 @@ struct omap4_hwc_module {
 };
 typedef struct omap4_hwc_module omap4_hwc_module_t;
 
+struct counts {
+    unsigned int possible_overlay_layers;
+    unsigned int composited_layers;
+    unsigned int scaled_layers;
+    unsigned int RGB;
+    unsigned int BGR;
+    unsigned int NV12;
+    unsigned int dockable;
+    unsigned int protected;
+
+    unsigned int max_hw_overlays;
+    unsigned int max_scaling_overlays;
+    unsigned int mem;
+    unsigned int s3d;
+};
+
 struct omap4_hwc_device {
     /* static data */
     hwc_composer_device_1_t base;
@@ -138,26 +181,47 @@ struct omap4_hwc_device {
 
     int flags_rgb_order;
     int flags_nv12_only;
+    float upscaled_nv12_limit;
 
+    int on_tv;                  /* using a tv */
     int force_sgx;
     omap4_hwc_ext_t ext;        /* external mirroring data */
     int idle;
-    int ovls_blending;
 
-    /* composition data */
-    struct dsscomp_setup_dispc_data dsscomp_data;
+    float primary_m[2][3];          /* internal transformation matrix */
+    int primary_transform;
+    int primary_rotation;
+    hwc_rect_t primary_region;
+
     buffer_handle_t *buffers;
     int use_sgx;
     int swap_rb;
-    unsigned int post2_layers;
+    unsigned int post2_layers; /* Buffers used with DSS pipes*/
+    unsigned int post2_blit_buffers; /* Buffers used with blit */
     int ext_ovls;               /* # of overlays on external display for current composition */
     int ext_ovls_wanted;        /* # of overlays that should be on external display for current composition */
     int last_ext_ovls;          /* # of overlays on external/internal display for last composition */
     int last_int_ovls;
+
+    enum S3DLayoutType s3d_input_type;
+    enum S3DLayoutOrder s3d_input_order;
+
+    enum bltmode blt_mode;
+    enum bltpolicy blt_policy;
+
+    int blit_flags;
+    int blit_num;
+    struct omap_hwc_data comp_data; /* This is a kernel data structure */
+    struct rgz_blt_entry blit_ops[RGZ_MAX_BLITS];
+    struct counts stats;
+    int    ion_fd;
+    struct ion_handle *ion_handles[2];
+
 };
 typedef struct omap4_hwc_device omap4_hwc_device_t;
 
 #define HAL_FMT(f) ((f) == HAL_PIXEL_FORMAT_TI_NV12 ? "NV12" : \
+                    (f) == HAL_PIXEL_FORMAT_TI_NV12_1D ? "NV12" : \
                     (f) == HAL_PIXEL_FORMAT_YV12 ? "YV12" : \
                     (f) == HAL_PIXEL_FORMAT_BGRX_8888 ? "xRGB32" : \
                     (f) == HAL_PIXEL_FORMAT_RGBX_8888 ? "xBGR32" : \
@@ -171,6 +235,34 @@ typedef struct omap4_hwc_device omap4_hwc_device_t;
                     (f) == OMAP_DSS_COLOR_RGB16 ? "RGB565" : "??")
 
 static int debug = 0;
+static int debugpost2 = 0;
+static int debugblt = 0;
+static rgz_t grgz;
+static struct bvsurfgeom gscrngeom;
+
+static void showfps(void)
+{
+    static int framecount = 0;
+    static int lastframecount = 0;
+    static nsecs_t lastfpstime = 0;
+    static float fps = 0;
+    char value[PROPERTY_VALUE_MAX];
+
+    property_get("debug.hwc.showfps", value, "0");
+    if (!atoi(value)) {
+        return;
+    }
+
+    framecount++;
+    if (!(framecount & 0x7)) {
+        nsecs_t now = systemTime(SYSTEM_TIME_MONOTONIC);
+        nsecs_t diff = now - lastfpstime;
+        fps = ((framecount - lastframecount) * (float)(s2ns(1))) / diff;
+        lastfpstime = now;
+        lastframecount = framecount;
+        ALOGI("%d Frames, %f FPS", framecount, fps);
+    }
+}
 
 static void dump_layer(hwc_layer_1_t const* l)
 {
@@ -241,7 +333,7 @@ static void dump_printf(struct dump_buf *buf, const char *fmt, ...)
 
 static void dump_set_info(omap4_hwc_device_t *hwc_dev, hwc_display_contents_1_t* list)
 {
-    struct dsscomp_setup_dispc_data *dsscomp = &hwc_dev->dsscomp_data;
+    struct dsscomp_setup_dispc_data *dsscomp = &hwc_dev->comp_data.dsscomp_data;
     char logbuf[1024];
     struct dump_buf log = {
         .buf = logbuf,
@@ -255,7 +347,14 @@ static void dump_set_info(omap4_hwc_device_t *hwc_dev, hwc_display_contents_1_t*
             dump_printf(&log, " ");
         hwc_layer_1_t *layer = &list->hwLayers[i];
         IMG_native_handle_t *handle = (IMG_native_handle_t *)layer->handle;
-        dump_printf(&log, "%p:%s,", handle, layer->compositionType == HWC_OVERLAY ? "DSS" : "SGX");
+        if (hwc_dev->post2_blit_buffers) {
+            if ((i + 1) < hwc_dev->post2_layers)
+                dump_printf(&log, "%p:%s,", handle, "DSS");
+            else
+                dump_printf(&log, "%p:%s,", handle, "BV2D");
+        }
+        else
+            dump_printf(&log, "%p:%s,", handle, layer->compositionType == HWC_OVERLAY ? "DSS" : "SGX");
         if ((layer->flags & HWC_SKIP_LAYER) || !handle) {
             dump_printf(&log, "SKIP");
             continue;
@@ -286,6 +385,13 @@ static void dump_set_info(omap4_hwc_device_t *hwc_dev, hwc_display_contents_1_t*
             dump_printf(&log, " ");
         dump_printf(&log, "%p", hwc_dev->buffers[i]);
     }
+    if (hwc_dev->post2_blit_buffers) {
+        dump_printf(&log, "} B{");
+        for (i = hwc_dev->post2_layers;
+             i < hwc_dev->post2_blit_buffers + hwc_dev->post2_layers; i++) {
+            dump_printf(&log, "%p ", hwc_dev->buffers[i]);
+        }
+    }
     dump_printf(&log, "}%s\n", hwc_dev->use_sgx ? " swap" : "");
 
     ALOGD("%s", log.buf);
@@ -302,7 +408,7 @@ static int omap4_hwc_is_valid_format(int format)
     case HAL_PIXEL_FORMAT_BGRA_8888:
     case HAL_PIXEL_FORMAT_BGRX_8888:
     case HAL_PIXEL_FORMAT_TI_NV12:
-    case HAL_PIXEL_FORMAT_TI_NV12_PADDED:
+    case HAL_PIXEL_FORMAT_TI_NV12_1D:
         return 1;
 
     default:
@@ -310,6 +416,17 @@ static int omap4_hwc_is_valid_format(int format)
     }
 }
 
+static __u32 get_s3d_layout_type(hwc_layer_1_t *layer)
+{
+    return (layer->flags & S3DLayoutTypeMask) >> S3DLayoutTypeShift;
+}
+
+static __u32 get_s3d_layout_order(hwc_layer_1_t *layer)
+{
+    return (layer->flags & S3DLayoutOrderMask) >> S3DLayoutOrderShift;
+}
+
+
 static int scaled(hwc_layer_1_t *layer)
 {
     int w = WIDTH(layer->sourceCrop);
@@ -318,7 +435,9 @@ static int scaled(hwc_layer_1_t *layer)
     if (layer->transform & HWC_TRANSFORM_ROT_90)
         swap(w, h);
 
-    return WIDTH(layer->displayFrame) != w || HEIGHT(layer->displayFrame) != h;
+    /* An S3D layer also needs scaling due to subsampling */
+    return WIDTH(layer->displayFrame) != w || HEIGHT(layer->displayFrame) != h
+            || get_s3d_layout_type(layer) != eMono;
 }
 
 static int is_protected(hwc_layer_1_t *layer)
@@ -342,6 +461,21 @@ static int is_RGB(IMG_native_handle_t *handle)
         return 0;
     }
 }
+static int get_rgb_bpp(IMG_native_handle_t *handle)
+{
+    switch(handle->iFormat)
+    {
+    case HAL_PIXEL_FORMAT_BGRA_8888:
+    case HAL_PIXEL_FORMAT_BGRX_8888:
+    case HAL_PIXEL_FORMAT_RGBX_8888:
+    case HAL_PIXEL_FORMAT_RGBA_8888:
+        return 32;
+    case HAL_PIXEL_FORMAT_RGB_565:
+        return 16;
+    default:
+        return 0;
+    }
+}
 
 static int is_BGR_format(int format)
 {
@@ -364,13 +498,32 @@ static int is_NV12(IMG_native_handle_t *handle)
     switch(handle->iFormat)
     {
     case HAL_PIXEL_FORMAT_TI_NV12:
-    case HAL_PIXEL_FORMAT_TI_NV12_PADDED:
+    case HAL_PIXEL_FORMAT_TI_NV12_1D:
         return 1;
     default:
         return 0;
     }
 }
 
+static int is_upscaled_NV12(omap4_hwc_device_t *hwc_dev, hwc_layer_1_t *layer)
+{
+    if (!layer)
+        return 0;
+
+    IMG_native_handle_t *handle = (IMG_native_handle_t *)layer->handle;
+    if (!is_NV12(handle))
+        return 0;
+
+    int w = WIDTH(layer->sourceCrop);
+    int h = HEIGHT(layer->sourceCrop);
+
+    if (layer->transform & HWC_TRANSFORM_ROT_90)
+        swap(w, h);
+
+    return (WIDTH(layer->displayFrame) >= w * hwc_dev->upscaled_nv12_limit ||
+            HEIGHT(layer->displayFrame) >= h * hwc_dev->upscaled_nv12_limit);
+}
+
 static int dockable(hwc_layer_1_t *layer)
 {
     IMG_native_handle_t *handle = (IMG_native_handle_t *)layer->handle;
@@ -419,7 +572,7 @@ omap4_hwc_setup_layer_base(struct dss2_ovl_cfg *oc, int index, int format, int b
         break;
 
     case HAL_PIXEL_FORMAT_TI_NV12:
-    case HAL_PIXEL_FORMAT_TI_NV12_PADDED:
+    case HAL_PIXEL_FORMAT_TI_NV12_1D:
         oc->color_mode = OMAP_DSS_COLOR_NV12;
         bits_per_pixel = 8;
         oc->cconv = ctbl_bt601_5;
@@ -661,10 +814,25 @@ crop_to_rect(struct dss2_ovl_cfg *cfg, struct hwc_rect vis_rect)
 }
 
 static void
+omap4_hwc_apply_transform(float transform[2][3],struct dss2_ovl_cfg *oc)
+{
+    float x, y, w, h;
+
+    /* display position */
+    x = transform[0][0] * oc->win.x + transform[0][1] * oc->win.y + transform[0][2];
+    y = transform[1][0] * oc->win.x + transform[1][1] * oc->win.y + transform[1][2];
+    w = transform[0][0] * oc->win.w + transform[0][1] * oc->win.h;
+    h = transform[1][0] * oc->win.w + transform[1][1] * oc->win.h;
+    oc->win.x = m_round(w > 0 ? x : x + w);
+    oc->win.y = m_round(h > 0 ? y : y + h);
+    oc->win.w = m_round(w > 0 ? w : -w);
+    oc->win.h = m_round(h > 0 ? h : -h);
+}
+
+static void
 omap4_hwc_adjust_ext_layer(omap4_hwc_ext_t *ext, struct dss2_ovl_info *ovl)
 {
     struct dss2_ovl_cfg *oc = &ovl->cfg;
-    float x, y, w, h;
 
     /* crop to clone region if mirroring */
     if (!ext->current.docking &&
@@ -673,15 +841,7 @@ omap4_hwc_adjust_ext_layer(omap4_hwc_ext_t *ext, struct dss2_ovl_info *ovl)
         return;
     }
 
-    /* display position */
-    x = ext->m[0][0] * oc->win.x + ext->m[0][1] * oc->win.y + ext->m[0][2];
-    y = ext->m[1][0] * oc->win.x + ext->m[1][1] * oc->win.y + ext->m[1][2];
-    w = ext->m[0][0] * oc->win.w + ext->m[0][1] * oc->win.h;
-    h = ext->m[1][0] * oc->win.w + ext->m[1][1] * oc->win.h;
-    oc->win.x = m_round(w > 0 ? x : x + w);
-    oc->win.y = m_round(h > 0 ? y : y + h);
-    oc->win.w = m_round(w > 0 ? w : -w);
-    oc->win.h = m_round(h > 0 ? h : -h);
+    omap4_hwc_apply_transform(ext->m, oc);
 
     /* combining transformations: F^a*R^b*F^i*R^j = F^(a+b)*R^(j+b*(-1)^i), because F*R = R^(-1)*F */
     oc->rotation += (oc->mirror ? -1 : 1) * ext->current.rotation;
@@ -690,33 +850,28 @@ omap4_hwc_adjust_ext_layer(omap4_hwc_ext_t *ext, struct dss2_ovl_info *ovl)
         oc->mirror = !oc->mirror;
 }
 
-static struct dsscomp_dispc_limitations {
-    __u8 max_xdecim_2d;
-    __u8 max_ydecim_2d;
-    __u8 max_xdecim_1d;
-    __u8 max_ydecim_1d;
-    __u32 fclk;
-    __u8 max_downscale;
-    __u8 min_width;
-    __u16 integer_scale_ratio_limit;
-    __u16 max_width;
-    __u16 max_height;
-} limits = {
-    .max_xdecim_1d = 16,
-    .max_xdecim_2d = 16,
-    .max_ydecim_1d = 16,
-    .max_ydecim_2d = 2,
-    .fclk = 170666666,
-    .max_downscale = 4,
-    .min_width = 2,
-    .integer_scale_ratio_limit = 2048,
-    .max_width = 2048,
-    .max_height = 2048,
-};
+static struct dsscomp_platform_info limits;
+
+static void
+omap4_hwc_adjust_primary_display_layer(omap4_hwc_device_t *hwc_dev, struct dss2_ovl_info *ovl)
+{
+    struct dss2_ovl_cfg *oc = &ovl->cfg;
+
+    if (crop_to_rect(&ovl->cfg, hwc_dev->primary_region) != 0) {
+        ovl->cfg.enabled = 0;
+        return;
+    }
+
+    omap4_hwc_apply_transform(hwc_dev->primary_m, oc);
+
+    /* combining transformations: F^a*R^b*F^i*R^j = F^(a+b)*R^(j+b*(-1)^i), because F*R = R^(-1)*F */
+    oc->rotation += (oc->mirror ? -1 : 1) * hwc_dev->primary_rotation;
+    oc->rotation &= 3;
+}
 
 static int omap4_hwc_can_scale(__u32 src_w, __u32 src_h, __u32 dst_w, __u32 dst_h, int is_2d,
-                               struct dsscomp_display_info *dis, struct dsscomp_dispc_limitations *limits,
-                               __u32 pclk)
+                               struct dsscomp_display_info *dis, struct dsscomp_platform_info *limits,
+                               __u32 pclk, void *handle)
 {
     __u32 fclk = limits->fclk / 1000;
     __u32 min_src_w = DIV_ROUND_UP(src_w, is_2d ? limits->max_xdecim_2d : limits->max_xdecim_1d);
@@ -739,12 +894,16 @@ static int omap4_hwc_can_scale(__u32 src_w, __u32 src_h, __u32 dst_w, __u32 dst_
 
     /* for manual panels pclk is 0, and there are no pclk based scaling limits */
     if (!pclk)
-        return !(dst_w * limits->max_downscale < min_src_w);
+        return !(dst_w < src_w / limits->max_downscale / (is_2d ? limits->max_xdecim_2d : limits->max_xdecim_1d));
 
     /* :HACK: limit horizontal downscale well below theoretical limit as we saw display artifacts */
     if (dst_w * 4 < src_w)
         return 0;
 
+    if (handle)
+        if (get_rgb_bpp(handle) == 32 && src_w > 1280 && dst_w * 3 < src_w)
+            return 0;
+
     /* max horizontal downscale is 4, or the fclk/pixclk */
     if (fclk > pclk * limits->max_downscale)
         fclk = pclk * limits->max_downscale;
@@ -771,7 +930,7 @@ static int omap4_hwc_can_scale_layer(omap4_hwc_device_t *hwc_dev, hwc_layer_1_t
     /* NOTE: layers should be able to be scaled externally since
        framebuffer is able to be scaled on selected external resolution */
     return omap4_hwc_can_scale(src_w, src_h, dst_w, dst_h, is_NV12(handle), &hwc_dev->fb_dis, &limits,
-                               hwc_dev->fb_dis.timings.pixel_clock);
+                               hwc_dev->fb_dis.timings.pixel_clock, handle);
 }
 
 static int omap4_hwc_is_valid_layer(omap4_hwc_device_t *hwc_dev,
@@ -789,7 +948,7 @@ static int omap4_hwc_is_valid_layer(omap4_hwc_device_t *hwc_dev,
     if (!is_NV12(handle)) {
         if (layer->transform)
             return 0;
-        if (mem1d(handle) > MAX_TILER_SLOT)
+        if (mem1d(handle) > limits.tiler1d_slot_size)
             return 0;
     }
 
@@ -837,10 +996,11 @@ static __u32 add_scaling_score(__u32 score,
 static int omap4_hwc_set_best_hdmi_mode(omap4_hwc_device_t *hwc_dev, __u32 xres, __u32 yres,
                                         float xpy)
 {
+    int dis_ix = hwc_dev->on_tv ? 0 : 1;
     struct _qdis {
         struct dsscomp_display_info dis;
         struct dsscomp_videomode modedb[32];
-    } d = { .dis = { .ix = 1 } };
+    } d = { .dis = { .ix = dis_ix } };
     omap4_hwc_ext_t *ext = &hwc_dev->ext;
 
     d.dis.modedb_len = sizeof(d.modedb) / sizeof(*d.modedb);
@@ -872,6 +1032,9 @@ static int omap4_hwc_set_best_hdmi_mode(omap4_hwc_device_t *hwc_dev, __u32 xres,
         __u32 ext_width = d.dis.width_in_mm;
         __u32 ext_height = d.dis.height_in_mm;
 
+        if (d.modedb[i].vmode & FB_VMODE_INTERLACED)
+            mode_yres /= 2;
+
         if (d.modedb[i].flag & FB_FLAG_RATIO_4_3) {
             ext_width = 4;
             ext_height = 3;
@@ -888,10 +1051,10 @@ static int omap4_hwc_set_best_hdmi_mode(omap4_hwc_device_t *hwc_dev, __u32 xres,
 
         /* we need to ensure that even TILER2D buffers can be scaled */
         if (!d.modedb[i].pixclock ||
-            d.modedb[i].vmode ||
+            (d.modedb[i].vmode & ~FB_VMODE_INTERLACED) ||
             !omap4_hwc_can_scale(xres, yres, ext_fb_xres, ext_fb_yres,
                                  1, &d.dis, &limits,
-                                 1000000000 / d.modedb[i].pixclock))
+                                 1000000000 / d.modedb[i].pixclock, NULL))
             continue;
 
         /* prefer CEA modes */
@@ -917,7 +1080,7 @@ static int omap4_hwc_set_best_hdmi_mode(omap4_hwc_device_t *hwc_dev, __u32 xres,
         }
     }
     if (~best) {
-        struct dsscomp_setup_display_data sdis = { .ix = 1, };
+        struct dsscomp_setup_display_data sdis = { .ix = dis_ix };
         sdis.mode = d.dis.modedb[best];
         ALOGD("picking #%d", best);
         /* only reconfigure on change */
@@ -932,10 +1095,9 @@ static int omap4_hwc_set_best_hdmi_mode(omap4_hwc_device_t *hwc_dev, __u32 xres,
         get_max_dimensions(xres, yres, xpy, d.dis.timings.x_res, d.dis.timings.y_res,
                            ext_width, ext_height, &ext_fb_xres, &ext_fb_yres);
         if (!d.dis.timings.pixel_clock ||
-            d.dis.mgr.interlaced ||
             !omap4_hwc_can_scale(xres, yres, ext_fb_xres, ext_fb_yres,
                                  1, &d.dis, &limits,
-                                 d.dis.timings.pixel_clock)) {
+                                 d.dis.timings.pixel_clock, NULL)) {
             ALOGW("DSS scaler cannot support HDMI cloning");
             return -1;
         }
@@ -948,20 +1110,6 @@ static int omap4_hwc_set_best_hdmi_mode(omap4_hwc_device_t *hwc_dev, __u32 xres,
     return 0;
 }
 
-struct counts {
-    unsigned int possible_overlay_layers;
-    unsigned int composited_layers;
-    unsigned int scaled_layers;
-    unsigned int RGB;
-    unsigned int BGR;
-    unsigned int NV12;
-    unsigned int dockable;
-    unsigned int protected;
-
-    unsigned int max_hw_overlays;
-    unsigned int max_scaling_overlays;
-    unsigned int mem;
-};
 
 static void gather_layer_statistics(omap4_hwc_device_t *hwc_dev, struct counts *num, hwc_display_contents_1_t *list)
 {
@@ -971,14 +1119,32 @@ static void gather_layer_statistics(omap4_hwc_device_t *hwc_dev, struct counts *
     for (i = 0; list && i < list->numHwLayers; i++) {
         hwc_layer_1_t *layer = &list->hwLayers[i];
         IMG_native_handle_t *handle = (IMG_native_handle_t *)layer->handle;
+        __u32 s3d_layout_type = get_s3d_layout_type(layer);
 
         layer->compositionType = HWC_FRAMEBUFFER;
 
         if (omap4_hwc_is_valid_layer(hwc_dev, layer, handle)) {
+
+            if (s3d_layout_type != eMono) {
+                /* For now we can only handle 1 S3D layer, skip any additional ones */
+                if (num->s3d > 0 || !hwc_dev->ext.dock.enabled || !hwc_dev->ext.s3d_capable) {
+                    layer->flags |= HWC_SKIP_LAYER;
+                    continue;
+                } else if (num->s3d == 0) {
+                    /* For now, S3D layer is made a dockable layer to trigger docking logic. */
+                    if (!dockable(layer)) {
+                        num->dockable++;
+                    }
+                    num->s3d++;
+                    hwc_dev->s3d_input_type = s3d_layout_type;
+                    hwc_dev->s3d_input_order = get_s3d_layout_order(layer);
+                }
+            }
+
             num->possible_overlay_layers++;
 
             /* NV12 layers can only be rendered on scaling overlays */
-            if (scaled(layer) || is_NV12(handle))
+            if (scaled(layer) || is_NV12(handle) || hwc_dev->primary_transform)
                 num->scaled_layers++;
 
             if (is_BGR(handle))
@@ -997,6 +1163,7 @@ static void gather_layer_statistics(omap4_hwc_device_t *hwc_dev, struct counts *
             num->mem += mem1d(handle);
         }
     }
+    hwc_dev->stats = *num;
 }
 
 static void decide_supported_cloning(omap4_hwc_device_t *hwc_dev, struct counts *num)
@@ -1015,6 +1182,11 @@ static void decide_supported_cloning(omap4_hwc_device_t *hwc_dev, struct counts
 
         /* reserve just a video pipeline for HDMI if docking */
         hwc_dev->ext_ovls = (num->dockable || ext->force_dock) ? 1 : 0;
+
+        if (num->s3d && (hwc_dev->ext.s3d_type != hwc_dev->s3d_input_type)) {
+            /* S3D layers are dockable, and they need two overlays */
+            hwc_dev->ext_ovls += 1;
+        }
         num->max_hw_overlays -= max(hwc_dev->ext_ovls, hwc_dev->last_ext_ovls);
 
         /* use mirroring transform if we are auto-switching to docking mode while mirroring*/
@@ -1049,16 +1221,21 @@ static void decide_supported_cloning(omap4_hwc_device_t *hwc_dev, struct counts
 
     /* if mirroring, we are limited by both internal and external overlays.  However,
        ext_ovls is always <= MAX_HW_OVERLAYS / 2 <= max_hw_overlays */
-    if (hwc_dev->ext_ovls && ext->current.enabled && !ext->current.docking)
+    if (!num->protected && hwc_dev->ext_ovls && ext->current.enabled && !ext->current.docking)
         num->max_hw_overlays = hwc_dev->ext_ovls;
 
-    num->max_scaling_overlays = num->max_hw_overlays - nonscaling_ovls;
+    /* If FB is not same resolution as LCD don't use GFX pipe line*/
+    if (hwc_dev->primary_transform) {
+        num->max_hw_overlays -= NUM_NONSCALING_OVERLAYS;
+        num->max_scaling_overlays = num->max_hw_overlays;
+    } else
+        num->max_scaling_overlays = num->max_hw_overlays - nonscaling_ovls;
 }
 
 static int can_dss_render_all(omap4_hwc_device_t *hwc_dev, struct counts *num)
 {
     omap4_hwc_ext_t *ext = &hwc_dev->ext;
-    int on_tv = ext->on_tv && ext->current.enabled;
+    int on_tv = hwc_dev->on_tv || (ext->on_tv && ext->current.enabled);
     int tform = ext->current.enabled && (ext->current.rotation || ext->current.hflip);
 
     return  !hwc_dev->force_sgx &&
@@ -1069,9 +1246,10 @@ static int can_dss_render_all(omap4_hwc_device_t *hwc_dev, struct counts *num)
             num->scaled_layers <= num->max_scaling_overlays &&
             num->NV12 <= num->max_scaling_overlays &&
             /* fits into TILER slot */
-            num->mem <= MAX_TILER_SLOT &&
+            num->mem <= limits.tiler1d_slot_size &&
             /* we cannot clone non-NV12 transformed layers */
-            (!tform || num->NV12 == num->possible_overlay_layers) &&
+            (!tform || (num->NV12 == num->possible_overlay_layers) ||
+            (num->NV12 && ext->current.docking)) &&
             /* HDMI cannot display BGR */
             (num->BGR == 0 || (num->RGB == 0 && !on_tv) || !hwc_dev->flags_rgb_order);
 }
@@ -1082,8 +1260,9 @@ static inline int can_dss_render_layer(omap4_hwc_device_t *hwc_dev,
     IMG_native_handle_t *handle = (IMG_native_handle_t *)layer->handle;
 
     omap4_hwc_ext_t *ext = &hwc_dev->ext;
-    int on_tv = ext->on_tv && ext->current.enabled;
-    int tform = ext->current.enabled && (ext->current.rotation || ext->current.hflip);
+    int cloning = ext->current.enabled && (!ext->current.docking || (handle!=NULL ? dockable(layer) : 0));
+    int on_tv = hwc_dev->on_tv || (ext->on_tv && cloning);
+    int tform = cloning && (ext->current.rotation || ext->current.hflip);
 
     return omap4_hwc_is_valid_layer(hwc_dev, layer, handle) &&
            /* cannot rotate non-NV12 layers on external display */
@@ -1103,7 +1282,7 @@ static inline int display_area(struct dss2_ovl_info *o)
 }
 
 static int clone_layer(omap4_hwc_device_t *hwc_dev, int ix) {
-    struct dsscomp_setup_dispc_data *dsscomp = &hwc_dev->dsscomp_data;
+    struct dsscomp_setup_dispc_data *dsscomp = &hwc_dev->comp_data.dsscomp_data;
     omap4_hwc_ext_t *ext = &hwc_dev->ext;
     int ext_ovl_ix = dsscomp->num_ovls - hwc_dev->post2_layers;
     struct dss2_ovl_info *o = &dsscomp->ovls[dsscomp->num_ovls];
@@ -1118,8 +1297,21 @@ static int clone_layer(omap4_hwc_device_t *hwc_dev, int ix) {
     /* reserve overlays at end for other display */
     o->cfg.ix = MAX_HW_OVERLAYS - 1 - ext_ovl_ix;
     o->cfg.mgr_ix = 1;
-    o->addressing = OMAP_DSS_BUFADDR_OVL_IX;
-    o->ba = ix;
+    /*
+    * Here the assumption is that overlay0 is the one attached to FB.
+    * Hence this clone_layer call is for FB cloning (provided use_sgx is true).
+    */
+    /* For the external displays whose transform is the same as
+    * that of primary display, ion_handles would be NULL hence
+    * the below logic doesn't execute.
+    */
+    if (ix == 0 && hwc_dev->ion_handles[sync_id%2] && hwc_dev->use_sgx) {
+        o->addressing = OMAP_DSS_BUFADDR_ION;
+        o->ba = (int)hwc_dev->ion_handles[sync_id%2];
+    } else {
+        o->addressing = OMAP_DSS_BUFADDR_OVL_IX;
+        o->ba = ix;
+    }
 
     /* use distinct z values (to simplify z-order checking) */
     o->cfg.zorder += hwc_dev->post2_layers;
@@ -1130,7 +1322,7 @@ static int clone_layer(omap4_hwc_device_t *hwc_dev, int ix) {
 }
 
 static int clone_external_layer(omap4_hwc_device_t *hwc_dev, int ix) {
-    struct dsscomp_setup_dispc_data *dsscomp = &hwc_dev->dsscomp_data;
+    struct dsscomp_setup_dispc_data *dsscomp = &hwc_dev->comp_data.dsscomp_data;
     omap4_hwc_ext_t *ext = &hwc_dev->ext;
 
     /* mirror only 1 external layer */
@@ -1170,6 +1362,152 @@ static int clone_external_layer(omap4_hwc_device_t *hwc_dev, int ix) {
     return clone_layer(hwc_dev, ix);
 }
 
+
+const char hdmiS3DTypePath[] = "/sys/devices/platform/omapdss/display1/s3d_type";
+const char hdmiS3DEnablePath[] = "/sys/devices/platform/omapdss/display1/s3d_enable";
+
+static void
+omap4_hwc_s3d_hdmi_enable(omap4_hwc_device_t *hwc_dev, bool enable)
+{
+    size_t bytesWritten;
+    char data;
+    int fd;
+
+    if (hwc_dev->ext.s3d_enabled == enable) {
+        return;
+    }
+
+    if (enable) {
+        char type[2];
+
+        switch(hwc_dev->ext.s3d_type) {
+            case eSideBySide:
+                snprintf(type, sizeof(type), "%d", HDMI_SIDE_BY_SIDE_HALF);
+                break;
+            case eTopBottom:
+                snprintf(type, sizeof(type), "%d", HDMI_TOPBOTTOM);
+                break;
+            default:
+                return;
+        }
+
+        fd = open(hdmiS3DTypePath, O_WRONLY);
+        if (fd < 0) {
+            ALOGE("Failed to open sysfs %s", hdmiS3DTypePath);
+            return;
+        }
+        bytesWritten = write(fd, type, sizeof(type));
+        close(fd);
+
+        if (bytesWritten != sizeof(type)) {
+            ALOGE("Failed to write (%s) to sysfs %s", type, hdmiS3DTypePath);
+            return;
+        }
+    }
+    data = enable ? '1' : '0';
+
+    fd = open(hdmiS3DEnablePath, O_WRONLY);
+    if (fd < 0) {
+        ALOGE("Failed to open sysfs %s", hdmiS3DEnablePath);
+        return;
+    }
+    bytesWritten = write(fd, &data, 1);
+    close(fd);
+
+    if (bytesWritten != 1) {
+        ALOGE("Failed to write(%d) to sysfs %s", enable, hdmiS3DEnablePath);
+        return;
+    }
+
+    hwc_dev->ext.s3d_enabled = enable;
+}
+
+static void
+omap4_hwc_adjust_ext_s3d_layer(omap4_hwc_device_t *hwc_dev,
+                                struct dss2_ovl_info *ovl,
+                                bool leftView)
+{
+    struct dss2_ovl_cfg *oc = &ovl->cfg;
+    float x, y, w, h;
+
+    switch (hwc_dev->s3d_input_type) {
+        case eSideBySide:
+            oc->crop.w = oc->crop.w/2;
+            if ((leftView && hwc_dev->s3d_input_order == eRightViewFirst) ||
+                (!leftView && hwc_dev->s3d_input_order == eLeftViewFirst)) {
+                oc->crop.x = oc->crop.x + oc->crop.w;
+            }
+            break;
+        case eTopBottom:
+            oc->crop.h = oc->crop.h/2;
+            if ((leftView && hwc_dev->s3d_input_order == eRightViewFirst) ||
+                (!leftView && hwc_dev->s3d_input_order == eLeftViewFirst)) {
+                oc->crop.y = oc->crop.y + oc->crop.h;
+            }
+            break;
+        default:
+            /* Should never fall here! */
+            ALOGE("Unsupported S3D layer type!");
+            break;
+    }
+
+    switch (hwc_dev->ext.s3d_type) {
+        case eSideBySide:
+            oc->win.w = oc->win.w/2;
+            if ((leftView && hwc_dev->ext.s3d_order == eRightViewFirst) ||
+                (!leftView && hwc_dev->ext.s3d_order == eLeftViewFirst)) {
+                oc->win.x = oc->win.x/2 + hwc_dev->ext.xres/2;
+            } else {
+                oc->win.x = oc->win.x/2;
+            }
+            break;
+        case eTopBottom:
+            oc->win.h = oc->win.h/2;
+            if ((leftView && hwc_dev->ext.s3d_order == eRightViewFirst) ||
+                (!leftView && hwc_dev->ext.s3d_order == eLeftViewFirst)) {
+                oc->win.y = oc->win.y/2 + hwc_dev->ext.yres/2;
+            } else {
+                oc->win.y = oc->win.y/2;
+            }
+            break;
+        default:
+            /* Currently unhandled!!! */
+            ALOGE("Unsupported S3D display type!");
+            break;
+    }
+}
+
+static int
+clone_s3d_external_layer(omap4_hwc_device_t *hwc_dev, int ix_s3d)
+{
+    struct dsscomp_setup_dispc_data *dsscomp = &hwc_dev->comp_data.dsscomp_data;
+    int r;
+
+    /* S3D layers are forced into docking layers. If the display layout and
+     * the layer layout don't match, we have to use 2 overlay pipelines */
+    r = clone_external_layer(hwc_dev, ix_s3d);
+    if (r) {
+        ALOGE("Failed to clone s3d layer (%d)", r);
+        return r;
+    }
+
+    r = clone_layer(hwc_dev, ix_s3d);
+    if (r) {
+        ALOGE("Failed to clone s3d layer (%d)", r);
+        return r;
+    }
+
+    if (dsscomp->num_ovls < 2) {
+        ALOGE("Number of overlays is inconsistent (%d)", dsscomp->num_ovls);
+        return -EINVAL;
+    }
+
+    omap4_hwc_adjust_ext_s3d_layer(hwc_dev, &dsscomp->ovls[dsscomp->num_ovls - 1], true);
+    omap4_hwc_adjust_ext_s3d_layer(hwc_dev, &dsscomp->ovls[dsscomp->num_ovls - 2], false);
+
+    return 0;
+}
+
 static int setup_mirroring(omap4_hwc_device_t *hwc_dev)
 {
     omap4_hwc_ext_t *ext = &hwc_dev->ext;
@@ -1214,6 +1552,173 @@ static void check_sync_fds(size_t numDisplays, hwc_display_contents_1_t** displa
     }
 }
 
+static void blit_reset(omap4_hwc_device_t *hwc_dev, int flags)
+{
+    hwc_dev->blit_flags = 0;
+    hwc_dev->blit_num = 0;
+    hwc_dev->post2_blit_buffers = 0;
+    hwc_dev->comp_data.blit_data.rgz_items = 0;
+
+    /* We want to maintain the rgz dirty region data if there are no geometry changes */
+    if (flags & HWC_GEOMETRY_CHANGED)
+        rgz_release(&grgz);
+}
+
+static int blit_layers(omap4_hwc_device_t *hwc_dev, hwc_display_contents_1_t *list, int bufoff)
+{
+    /* Do not blit if this frame will be composed entirely by the GPU */
+    if (!list || hwc_dev->force_sgx)
+        goto err_out;
+
+    int rgz_in_op;
+    int rgz_out_op;
+
+    switch (hwc_dev->blt_mode) {
+        case BLTMODE_PAINT:
+            rgz_in_op = RGZ_IN_HWCCHK;
+            rgz_out_op = RGZ_OUT_BVCMD_PAINT;
+            break;
+        case BLTMODE_REGION:
+        default:
+            rgz_in_op = RGZ_IN_HWC;
+            rgz_out_op = RGZ_OUT_BVCMD_REGION;
+            break;
+    }
+
+    rgz_in_params_t in = {
+        .op = rgz_in_op,
+        .data = {
+            .hwc = {
+                .dstgeom = &gscrngeom,
+                .layers = list->hwLayers,
+                .layerno = list->numHwLayers
+            }
+        }
+    };
+
+    /*
+     * This means if all the layers marked for the FRAMEBUFFER cannot be
+     * blitted, do not blit, for e.g. SKIP layers
+     */
+    if (rgz_in(&in, &grgz) != RGZ_ALL)
+        goto err_out;
+
+    unsigned int i, count = 0;
+    for (i = 0; i < list->numHwLayers; i++) {
+        if (list->hwLayers[i].compositionType != HWC_OVERLAY) {
+            count++;
+        }
+    }
+
+    rgz_out_params_t out = {
+        .op = rgz_out_op,
+        .data = {
+            .bvc = {
+                .dstgeom = &gscrngeom,
+                .noblend = 0,
+            }
+        }
+    };
+
+    if (rgz_out(&grgz, &out) != 0) {
+        ALOGE("Failed generating blits");
+        goto err_out;
+    }
+
+    /* This is a special situation where the regionizer decided no blits are
+     * needed for this frame but there are blit buffers to synchronize with. Can
+     * happen only if the regionizer is enabled otherwise it's likely a bug
+     */
+    if (rgz_out_op != RGZ_OUT_BVCMD_REGION && out.data.bvc.out_blits == 0 && out.data.bvc.out_nhndls > 0) {
+        ALOGE("Regionizer invalid output blit_num %d, post2_blit_buffers %d", out.data.bvc.out_blits, out.data.bvc.out_nhndls);
+        goto err_out;
+    }
+
+    hwc_dev->blit_flags |= HWC_BLT_FLAG_USE_FB;
+    hwc_dev->blit_num = out.data.bvc.out_blits;
+    hwc_dev->post2_blit_buffers = out.data.bvc.out_nhndls;
+    for (i = 0; i < hwc_dev->post2_blit_buffers; i++) {
+        //ALOGI("blit buffers[%d] = %p", bufoff, out.data.bvc.out_hndls[i]);
+        hwc_dev->buffers[bufoff++] = out.data.bvc.out_hndls[i];
+    }
+
+    struct rgz_blt_entry *res_blit_ops = (struct rgz_blt_entry *) out.data.bvc.cmdp;
+    memcpy(hwc_dev->comp_data.blit_data.rgz_blts, res_blit_ops, sizeof(*res_blit_ops) * out.data.bvc.cmdlen);
+    ALOGI_IF(debugblt, "blt struct sz %d", sizeof(*res_blit_ops) * out.data.bvc.cmdlen);
+    ALOGE_IF(hwc_dev->blit_num != out.data.bvc.cmdlen,"blit_num != out.data.bvc.cmdlen, %d != %d", hwc_dev->blit_num, out.data.bvc.cmdlen);
+
+    /* all layers will be rendered without SGX help either via DSS or blitter */
+    for (i = 0; i < list->numHwLayers; i++) {
+        if (list->hwLayers[i].compositionType != HWC_OVERLAY) {
+            list->hwLayers[i].compositionType = HWC_OVERLAY;
+            //ALOGI("blitting layer %d", i);
+            list->hwLayers[i].hints &= ~HWC_HINT_TRIPLE_BUFFER;
+        }
+        list->hwLayers[i].hints &= ~HWC_HINT_CLEAR_FB;
+    }
+    return 1;
+
+err_out:
+    rgz_release(&grgz);
+    return 0;
+}
+
+void debug_post2(omap4_hwc_device_t *hwc_dev, int nbufs)
+{
+    if (!debugpost2)
+        return;
+    struct dsscomp_setup_dispc_data *dsscomp = &hwc_dev->comp_data.dsscomp_data;
+    int i;
+    for (i=0; i<nbufs; i++) {
+        ALOGI("buf[%d] hndl %p", i, hwc_dev->buffers[i]);
+    }
+    for (i=0; i < dsscomp->num_ovls; i++) {
+        ALOGI("ovl[%d] ba %d", i, dsscomp->ovls[i].ba);
+    }
+}
+
+static int free_tiler2d_buffers(omap4_hwc_device_t *hwc_dev)
+{
+    int i;
+
+    for (i = 0 ; i < NUM_EXT_DISPLAY_BACK_BUFFERS; i++) {
+        ion_free(hwc_dev->ion_fd, hwc_dev->ion_handles[i]);
+        hwc_dev->ion_handles[i] = NULL;
+    }
+    return 0;
+}
+
+static int allocate_tiler2d_buffers(omap4_hwc_device_t *hwc_dev)
+{
+    int ret, i;
+    size_t stride;
+
+    if (hwc_dev->ion_fd < 0) {
+        ALOGE("No ion fd, hence can't allocate tiler2d buffers");
+        return -1;
+    }
+
+    for (i = 0; i < NUM_EXT_DISPLAY_BACK_BUFFERS; i++) {
+        if (hwc_dev->ion_handles[i])
+            return 0;
+    }
+
+    for (i = 0 ; i < NUM_EXT_DISPLAY_BACK_BUFFERS; i++) {
+        ret = ion_alloc_tiler(hwc_dev->ion_fd, hwc_dev->fb_dev->base.width, hwc_dev->fb_dev->base.height,
+                                            TILER_PIXEL_FMT_32BIT, 0, &hwc_dev->ion_handles[i], &stride);
+        if (ret)
+            goto handle_error;
+
+        ALOGI("ion handle[%d][%p]", i, hwc_dev->ion_handles[i]);
+    }
+    return 0;
+
+handle_error:
+    free_tiler2d_buffers(hwc_dev);
+    return -1;
+}
+
+
 static int omap4_hwc_prepare(struct hwc_composer_device_1 *dev, size_t numDisplays,
         hwc_display_contents_1_t** displays)
 {
@@ -1223,7 +1728,7 @@ static int omap4_hwc_prepare(struct hwc_composer_device_1 *dev, size_t numDispla
 
     hwc_display_contents_1_t* list = displays[0];  // ignore displays beyond the first
     omap4_hwc_device_t *hwc_dev = (omap4_hwc_device_t *)dev;
-    struct dsscomp_setup_dispc_data *dsscomp = &hwc_dev->dsscomp_data;
+    struct dsscomp_setup_dispc_data *dsscomp = &hwc_dev->comp_data.dsscomp_data;
     struct counts num = { .composited_layers = list ? list->numHwLayers : 0 };
     unsigned int i, ix;
 
@@ -1235,10 +1740,6 @@ static int omap4_hwc_prepare(struct hwc_composer_device_1 *dev, size_t numDispla
 
     decide_supported_cloning(hwc_dev, &num);
 
-    /* Disable the forced SGX rendering if there is only one layer */
-    if (hwc_dev->force_sgx && num.composited_layers <= 1)
-        hwc_dev->force_sgx = 0;
-
     /* phase 3 logic */
     if (can_dss_render_all(hwc_dev, &num)) {
         /* All layers can be handled by the DSS -- don't use SGX for composition */
@@ -1251,11 +1752,33 @@ static int omap4_hwc_prepare(struct hwc_composer_device_1 *dev, size_t numDispla
     }
 
     /* setup pipes */
-    dsscomp->num_ovls = hwc_dev->use_sgx;
     int z = 0;
     int fb_z = -1;
     int scaled_gfx = 0;
     int ix_docking = -1;
+    int ix_s3d = -1;
+
+    int blit_all = 0;
+    blit_reset(hwc_dev, list ? list->flags : 0);
+
+    /* If the SGX is used or we are going to blit something we need a framebuffer
+     * and a DSS pipe
+     */
+    int needs_fb = hwc_dev->use_sgx;
+
+    if (hwc_dev->blt_policy == BLTPOLICY_ALL) {
+        /* Check if we can blit everything */
+        blit_all = blit_layers(hwc_dev, list, 0);
+        if (blit_all) {
+            needs_fb = 1;
+            hwc_dev->use_sgx = 0;
+        }
+    }
+
+    /* If a framebuffer is needed, begin using VID1 for DSS overlay layers,
+     * we need GFX for FB
+     */
+    dsscomp->num_ovls = needs_fb ? 1 /*VID1*/ : 0 /*GFX*/;
 
     /* set up if DSS layers */
     unsigned int mem_used = 0;
@@ -1269,23 +1792,28 @@ static int omap4_hwc_prepare(struct hwc_composer_device_1 *dev, size_t numDispla
             (!hwc_dev->force_sgx ||
              /* render protected and dockable layers via DSS */
              is_protected(layer) ||
+             is_upscaled_NV12(hwc_dev, layer) ||
              (hwc_dev->ext.current.docking && hwc_dev->ext.current.enabled && dockable(layer))) &&
-            mem_used + mem1d(handle) < MAX_TILER_SLOT &&
+            mem_used + mem1d(handle) <= limits.tiler1d_slot_size &&
             /* can't have a transparent overlay in the middle of the framebuffer stack */
             !(is_BLENDED(layer) && fb_z >= 0)) {
 
             /* render via DSS overlay */
             mem_used += mem1d(handle);
             layer->compositionType = HWC_OVERLAY;
+            /*
+             * This hint will not be used in vanilla ICS, but maybe in
+             * JellyBean, it is useful to distinguish between blts and true
+             * overlays
+             */
+            layer->hints |= HWC_HINT_TRIPLE_BUFFER;
 
             /* clear FB above all opaque layers if rendering via SGX */
             if (hwc_dev->use_sgx && !is_BLENDED(layer))
                 layer->hints |= HWC_HINT_CLEAR_FB;
-            /* see if any of the (non-backmost) overlays are doing blending */
-            else if (is_BLENDED(layer) && i > 0)
-                hwc_dev->ovls_blending = 1;
 
             hwc_dev->buffers[dsscomp->num_ovls] = layer->handle;
+            //ALOGI("dss buffers[%d] = %p", dsscomp->num_ovls, hwc_dev->buffers[dsscomp->num_ovls]);
 
             omap4_hwc_setup_layer(hwc_dev,
                                   &dsscomp->ovls[dsscomp->num_ovls],
@@ -1295,12 +1823,12 @@ static int omap4_hwc_prepare(struct hwc_composer_device_1 *dev, size_t numDispla
                                   handle->iWidth,
                                   handle->iHeight);
 
-            dsscomp->ovls[dsscomp->num_ovls].cfg.ix = dsscomp->num_ovls;
+            dsscomp->ovls[dsscomp->num_ovls].cfg.ix = dsscomp->num_ovls + hwc_dev->primary_transform;
             dsscomp->ovls[dsscomp->num_ovls].addressing = OMAP_DSS_BUFADDR_LAYER_IX;
             dsscomp->ovls[dsscomp->num_ovls].ba = dsscomp->num_ovls;
 
             /* ensure GFX layer is never scaled */
-            if (dsscomp->num_ovls == 0) {
+            if ((dsscomp->num_ovls == 0) && (!hwc_dev->primary_transform)) {
                 scaled_gfx = scaled(layer) || is_NV12(handle);
             } else if (scaled_gfx && !scaled(layer) && !is_NV12(handle)) {
                 /* swap GFX layer with this one */
@@ -1315,6 +1843,10 @@ static int omap4_hwc_prepare(struct hwc_composer_device_1 *dev, size_t numDispla
                  display_area(&dsscomp->ovls[dsscomp->num_ovls]) > display_area(&dsscomp->ovls[ix_docking])))
                 ix_docking = dsscomp->num_ovls;
 
+            /* remember the ix for s3d layer */
+            if (get_s3d_layout_type(layer) != eMono) {
+                ix_s3d = dsscomp->num_ovls;
+            }
             dsscomp->num_ovls++;
             z++;
         } else if (hwc_dev->use_sgx) {
@@ -1334,15 +1866,31 @@ static int omap4_hwc_prepare(struct hwc_composer_device_1 *dev, size_t numDispla
     if (scaled_gfx)
         dsscomp->ovls[0].cfg.ix = dsscomp->num_ovls;
 
-    if (hwc_dev->use_sgx) {
+    if (hwc_dev->blt_policy == BLTPOLICY_DEFAULT) {
+        if (hwc_dev->use_sgx) {
+            if (blit_layers(hwc_dev, list, dsscomp->num_ovls == 1 ? 0 : dsscomp->num_ovls)) {
+                hwc_dev->use_sgx = 0;
+            }
+        }
+    }
+
+    /* If the SGX is not used and there is blit data we need a framebuffer and
+     * a DSS pipe well configured for it
+     */
+    if (needs_fb) {
         /* assign a z-layer for fb */
         if (fb_z < 0) {
-            if (num.composited_layers)
+            if (!hwc_dev->blt_policy != BLTPOLICY_DISABLED && num.composited_layers)
                 ALOGE("**** should have assigned z-layer for fb");
             fb_z = z++;
         }
-
-        hwc_dev->buffers[0] = NULL;
+        /*
+         * This is needed because if we blit all we would lose the handle of
+         * the first layer
+         */
+        if (hwc_dev->use_sgx) {
+            hwc_dev->buffers[0] = NULL;
+        }
         omap4_hwc_setup_layer_base(&dsscomp->ovls[0].cfg, fb_z,
                                    hwc_dev->fb_dev->base.format,
                                    1,   /* FB is always premultiplied */
@@ -1351,14 +1899,34 @@ static int omap4_hwc_prepare(struct hwc_composer_device_1 *dev, size_t numDispla
         dsscomp->ovls[0].cfg.pre_mult_alpha = 1;
         dsscomp->ovls[0].addressing = OMAP_DSS_BUFADDR_LAYER_IX;
         dsscomp->ovls[0].ba = 0;
+        dsscomp->ovls[0].cfg.ix = hwc_dev->primary_transform;
     }
 
     /* mirror layers */
     hwc_dev->post2_layers = dsscomp->num_ovls;
 
     omap4_hwc_ext_t *ext = &hwc_dev->ext;
-    if (ext->current.enabled && hwc_dev->ext_ovls) {
-        if (ext->current.docking && ix_docking >= 0) {
+    if (ext->current.enabled && ((!num.protected && hwc_dev->ext_ovls) ||
+              (hwc_dev->ext_ovls_wanted && hwc_dev->ext_ovls >= hwc_dev->ext_ovls_wanted))) {
+        if (ext->current.docking && ix_s3d >= 0) {
+            if (clone_s3d_external_layer(hwc_dev, ix_s3d) == 0) {
+                dsscomp->ovls[dsscomp->num_ovls - 2].cfg.zorder = z++;
+                dsscomp->ovls[dsscomp->num_ovls - 1].cfg.zorder = z++;
+                /* For now, show only the left view of an S3D layer
+                 * in the local display while we have hdmi attached */
+                switch (hwc_dev->s3d_input_type) {
+                    case eSideBySide:
+                        dsscomp->ovls[ix_s3d].cfg.crop.w = dsscomp->ovls[ix_s3d].cfg.crop.w/2;
+                        break;
+                    case eTopBottom:
+                       dsscomp->ovls[ix_s3d].cfg.crop.h = dsscomp->ovls[ix_s3d].cfg.crop.h/2;
+                        break;
+                    default:
+                        ALOGE("Unsupported S3D input type");
+                        break;
+                }
+            }
+        } else if (ext->current.docking && ix_docking >= 0) {
             if (clone_external_layer(hwc_dev, ix_docking) == 0)
                 dsscomp->ovls[dsscomp->num_ovls - 1].cfg.zorder = z++;
         } else if (ext->current.docking && ix_docking < 0 && ext->force_dock) {
@@ -1387,6 +1955,17 @@ static int omap4_hwc_prepare(struct hwc_composer_device_1 *dev, size_t numDispla
             }
         }
     }
+
+    /* Apply transform for primary display */
+    if (hwc_dev->primary_transform)
+        for (i = 0; i < dsscomp->num_ovls; i++) {
+            if(dsscomp->ovls[i].cfg.mgr_ix == 0)
+                omap4_hwc_adjust_primary_display_layer(hwc_dev, &dsscomp->ovls[i]);
+        }
+
+
+    omap4_hwc_s3d_hdmi_enable(hwc_dev, ix_s3d >= 0);
+
     ext->last = ext->current;
 
     if (z != dsscomp->num_ovls || dsscomp->num_ovls > MAX_HW_OVERLAYS)
@@ -1416,6 +1995,14 @@ static int omap4_hwc_prepare(struct hwc_composer_device_1 *dev, size_t numDispla
         hwc_dev->ext_ovls = dsscomp->num_ovls - hwc_dev->post2_layers;
     }
 
+    /*
+     * Whilst the mode of the display is being changed drop compositions to the
+     * display
+     */
+    if (ext->last_mode == 0 && hwc_dev->on_tv) {
+        dsscomp->num_ovls = 0;
+    }
+
     if (debug) {
         ALOGD("prepare (%d) - %s (comp=%d, poss=%d/%d scaled, RGB=%d,BGR=%d,NV12=%d) (ext=%s%s%ddeg%s %dex/%dmx (last %dex,%din)\n",
              dsscomp->sync_id,
@@ -1477,7 +2064,7 @@ static int omap4_hwc_set(struct hwc_composer_device_1 *dev,
         sur = list->sur;
     }
     omap4_hwc_device_t *hwc_dev = (omap4_hwc_device_t *)dev;
-    struct dsscomp_setup_dispc_data *dsscomp = &hwc_dev->dsscomp_data;
+    struct dsscomp_setup_dispc_data *dsscomp = &hwc_dev->comp_data.dsscomp_data;
     int err = 0;
     int invalidate;
 
@@ -1485,7 +2072,8 @@ static int omap4_hwc_set(struct hwc_composer_device_1 *dev,
 
     omap4_hwc_reset_screen(hwc_dev);
 
-    invalidate = hwc_dev->ext_ovls_wanted && !hwc_dev->ext_ovls;
+    invalidate = hwc_dev->ext_ovls_wanted && (hwc_dev->ext_ovls < hwc_dev->ext_ovls_wanted) &&
+                                              (hwc_dev->stats.protected || !hwc_dev->ext_ovls);
 
     if (debug)
         dump_set_info(hwc_dev, list);
@@ -1510,10 +2098,42 @@ static int omap4_hwc_set(struct hwc_composer_device_1 *dev,
         if (hwc_dev->force_sgx > 0)
             hwc_dev->force_sgx--;
 
+        hwc_dev->comp_data.blit_data.rgz_flags = hwc_dev->blit_flags;
+        hwc_dev->comp_data.blit_data.rgz_items = hwc_dev->blit_num;
+        int omaplfb_comp_data_sz = sizeof(hwc_dev->comp_data) +
+            (hwc_dev->comp_data.blit_data.rgz_items * sizeof(struct rgz_blt_entry));
+
+
+        unsigned int nbufs = hwc_dev->post2_layers;
+        if (hwc_dev->post2_blit_buffers) {
+            /*
+             * We don't want to pass a NULL entry in the Post2, but we need to
+             * fix up buffer handle array and overlay indexes to account for
+             * this
+             */
+            nbufs += hwc_dev->post2_blit_buffers - 1;
+
+            if (hwc_dev->post2_layers > 1) {
+                unsigned int i, j;
+                for (i = 0; i < nbufs; i++) {
+                    hwc_dev->buffers[i] = hwc_dev->buffers[i+1];
+                }
+                for (i = 1, j= 1; j < hwc_dev->post2_layers; i++, j++) {
+                    dsscomp->ovls[j].ba = i;
+                }
+            }
+        }
+        ALOGI_IF(debugblt && hwc_dev->blt_policy != BLTPOLICY_DISABLED,
+            "Post2, blits %d, ovl_buffers %d, blit_buffers %d sgx %d",
+            hwc_dev->blit_num, hwc_dev->post2_layers, hwc_dev->post2_blit_buffers,
+            hwc_dev->use_sgx);
+
+        debug_post2(hwc_dev, nbufs);
         err = hwc_dev->fb_dev->Post2((framebuffer_device_t *)hwc_dev->fb_dev,
                                  hwc_dev->buffers,
-                                 hwc_dev->post2_layers,
-                                 dsscomp, sizeof(*dsscomp));
+                                 nbufs,
+                                 dsscomp, omaplfb_comp_data_sz);
+        showfps();
     }
     hwc_dev->last_ext_ovls = hwc_dev->ext_ovls;
     hwc_dev->last_int_ovls = hwc_dev->post2_layers;
@@ -1534,7 +2154,7 @@ err_out:
 static void omap4_hwc_dump(struct hwc_composer_device_1 *dev, char *buff, int buff_len)
 {
     omap4_hwc_device_t *hwc_dev = (omap4_hwc_device_t *)dev;
-    struct dsscomp_setup_dispc_data *dsscomp = &hwc_dev->dsscomp_data;
+    struct dsscomp_setup_dispc_data *dsscomp = &hwc_dev->comp_data.dsscomp_data;
     struct dump_buf log = {
         .buf = buff,
         .buf_len = buff_len,
@@ -1548,17 +2168,22 @@ static void omap4_hwc_dump(struct hwc_composer_device_1 *dev, char *buff, int bu
         struct dss2_ovl_cfg *cfg = &dsscomp->ovls[i].cfg;
 
         dump_printf(&log, "  layer %d:\n", i);
-        dump_printf(&log, "     enabled: %s\n",
-                          cfg->enabled ? "true" : "false");
-        dump_printf(&log, "     buff: %p %dx%d stride: %d\n",
-                          hwc_dev->buffers[i], cfg->width, cfg->height, cfg->stride);
-        dump_printf(&log, "     src: (%d,%d) %dx%d\n",
-                          cfg->crop.x, cfg->crop.y, cfg->crop.w, cfg->crop.h);
-        dump_printf(&log, "     dst: (%d,%d) %dx%d\n",
-                          cfg->win.x, cfg->win.y, cfg->win.w, cfg->win.h);
-        dump_printf(&log, "     ix: %d\n", cfg->ix);
-        dump_printf(&log, "     zorder: %d\n\n", cfg->zorder);
-    }
+        dump_printf(&log, "     enabled:%s buff:%p %dx%d stride:%d\n",
+                          cfg->enabled ? "true" : "false", hwc_dev->buffers[i],
+                          cfg->width, cfg->height, cfg->stride);
+        dump_printf(&log, "     src:(%d,%d) %dx%d dst:(%d,%d) %dx%d ix:%d zorder:%d\n",
+                          cfg->crop.x, cfg->crop.y, cfg->crop.w, cfg->crop.h,
+                          cfg->win.x, cfg->win.y, cfg->win.w, cfg->win.h,
+                          cfg->ix, cfg->zorder);
+    }
+
+    if (hwc_dev->blt_policy != BLTPOLICY_DISABLED) {
+        dump_printf(&log, "  bltpolicy: %s, bltmode: %s\n",
+            hwc_dev->blt_policy == BLTPOLICY_DEFAULT ? "default" :
+                hwc_dev->blt_policy == BLTPOLICY_ALL ? "all" : "unknown",
+                    hwc_dev->blt_mode == BLTMODE_PAINT ? "paint" : "regionize");
+    }
+    dump_printf(&log, "\n");
 }
 
 static void free_png_image(omap4_hwc_device_t *hwc_dev, struct omap4_hwc_img *img)
@@ -1688,6 +2313,9 @@ static int omap4_hwc_device_close(hw_device_t* device)
             close(hwc_dev->hdmi_fb_fd);
         if (hwc_dev->fb_fd >= 0)
             close(hwc_dev->fb_fd);
+        if (hwc_dev->ion_fd >= 0)
+            ion_close(hwc_dev->ion_fd);
+
         /* pthread will get killed when parent process exits */
         pthread_mutex_destroy(&hwc_dev->lock);
         free(hwc_dev);
@@ -1722,12 +2350,103 @@ err_out:
     return err;
 }
 
+static void set_primary_display_transform_matrix(omap4_hwc_device_t *hwc_dev)
+{
+    /* create primary display translation matrix */
+    hwc_dev->fb_dis.ix = 0;/*Default display*/
+
+    int ret = ioctl(hwc_dev->dsscomp_fd, DSSCIOC_QUERY_DISPLAY, &hwc_dev->fb_dis);
+    if (ret)
+        ALOGE("failed to get display info (%d): %m", errno);
+
+    int lcd_w = hwc_dev->fb_dis.timings.x_res;
+    int lcd_h = hwc_dev->fb_dis.timings.y_res;
+    int orig_w = hwc_dev->fb_dev->base.width;
+    int orig_h = hwc_dev->fb_dev->base.height;
+    hwc_rect_t region = {.left = 0, .top = 0, .right = orig_w, .bottom = orig_h};
+    hwc_dev->primary_region = region;
+    hwc_dev->primary_rotation = ((lcd_w > lcd_h) ^ (orig_w > orig_h)) ? 1 : 0;
+    hwc_dev->primary_transform = ((lcd_w != orig_w)||(lcd_h != orig_h)) ? 1 : 0;
+
+    ALOGI("transforming FB (%dx%d) => (%dx%d) rot%d", orig_w, orig_h, lcd_w, lcd_h, hwc_dev->primary_rotation);
+
+    /* reorientation matrix is:
+       m = (center-from-target-center) * (scale-to-target) * (mirror) * (rotate) * (center-to-original-center) */
+
+    memcpy(hwc_dev->primary_m, m_unit, sizeof(m_unit));
+    m_translate(hwc_dev->primary_m, -(orig_w >> 1), -(orig_h >> 1));
+    m_rotate(hwc_dev->primary_m, hwc_dev->primary_rotation);
+    if (hwc_dev->primary_rotation & 1)
+         swap(orig_w, orig_h);
+    m_scale(hwc_dev->primary_m, orig_w, lcd_w, orig_h, lcd_h);
+    m_translate(hwc_dev->primary_m, lcd_w >> 1, lcd_h >> 1);
+}
+
+
+
+static void handle_s3d_hotplug(omap4_hwc_ext_t *ext, int state)
+{
+    struct edid_t *edid = NULL;
+    if (state) {
+        int fd = open("/sys/devices/platform/omapdss/display1/edid", O_RDONLY);
+        if (!fd)
+            return;
+        uint8_t edid_data[EDID_SIZE];
+        size_t bytes_read = read(fd, edid_data, EDID_SIZE);
+        close(fd);
+        if (bytes_read < EDID_SIZE)
+            return;
+        if (edid_parser_init(&edid, edid_data))
+            return;
+    }
+
+    ext->s3d_enabled = false;
+    ext->s3d_capable = false;
+    ext->s3d_type = eMono;
+    ext->s3d_order = eLeftViewFirst;
+
+    if (edid) {
+        ext->s3d_capable = edid_s3d_capable(edid);
+        /* For now assume Side-by-Side half support applies to all modes */
+        ext->s3d_type = eSideBySide;
+        ext->s3d_order = eLeftViewFirst;
+        edid_parser_deinit(edid);
+    }
+}
+
+
 static void handle_hotplug(omap4_hwc_device_t *hwc_dev)
 {
     omap4_hwc_ext_t *ext = &hwc_dev->ext;
     __u8 state = ext->hdmi_state;
 
+    /* Ignore external HDMI logic if the primary display is HDMI */
+    if (hwc_dev->on_tv) {
+        ALOGI("Primary display is HDMI - skip clone/dock logic");
+
+        if (state) {
+            __u32 xres = hwc_dev->fb_dev->base.width;
+            __u32 yres = hwc_dev->fb_dev->base.height;
+            if (omap4_hwc_set_best_hdmi_mode(hwc_dev, xres, yres, ext->lcd_xpy)) {
+                ALOGE("Failed to set HDMI mode");
+            }
+            set_primary_display_transform_matrix(hwc_dev);
+
+            ioctl(hwc_dev->fb_fd, FBIOBLANK, FB_BLANK_UNBLANK);
+
+            if (hwc_dev->procs && hwc_dev->procs->invalidate) {
+                hwc_dev->procs->invalidate(hwc_dev->procs);
+            }
+        } else
+            ext->last_mode = 0;
+
+        return;
+    }
+
     pthread_mutex_lock(&hwc_dev->lock);
+
+    handle_s3d_hotplug(ext, state);
+
     ext->dock.enabled = ext->mirror.enabled = 0;
     if (state) {
         /* check whether we can clone and/or dock */
@@ -1744,7 +2463,7 @@ static void handle_hotplug(omap4_hwc_device_t *hwc_dev)
         ext->dock.rotation = atoi(value) & EXT_ROTATION;
         ext->dock.hflip = (atoi(value) & EXT_HFLIP) > 0;
         ext->dock.docking = 1;
-        property_get("persist.hwc.mirroring.transform", value, hwc_dev->fb_dev->base.height > hwc_dev->fb_dev->base.width ? "3" : "0");
+        property_get("persist.hwc.mirroring.transform", value, hwc_dev->fb_dis.timings.y_res > hwc_dev->fb_dis.timings.x_res ? "3" : "0");
         ext->mirror.rotation = atoi(value) & EXT_ROTATION;
         ext->mirror.hflip = (atoi(value) & EXT_HFLIP) > 0;
         ext->mirror.docking = 0;
@@ -1771,8 +2490,19 @@ static void handle_hotplug(omap4_hwc_device_t *hwc_dev)
             } else
                 ext->mirror.enabled = 0;
         }
+        /* Allocate backup buffers for FB rotation
+        * This is required only if the FB tranform is different from that
+        * of the external display and the FB is not in TILER2D space
+        */
+        if (ext->mirror.rotation && (limits.fbmem_type != DSSCOMP_FBMEM_TILER2D))
+            allocate_tiler2d_buffers(hwc_dev);
+
     } else {
         ext->last_mode = 0;
+        if (ext->mirror.rotation && (limits.fbmem_type != DSSCOMP_FBMEM_TILER2D)) {
+            /* free tiler 2D buffer on detach */
+            free_tiler2d_buffers(hwc_dev);
+        }
     }
     ALOGI("external display changed (state=%d, mirror={%s tform=%ddeg%s}, dock={%s tform=%ddeg%s%s}, tv=%d", state,
          ext->mirror.enabled ? "enabled" : "disabled",
@@ -1866,7 +2596,7 @@ static void *omap4_hwc_hdmi_thread(void *data)
             if (hwc_dev->idle) {
                 if (hwc_dev->procs) {
                     pthread_mutex_lock(&hwc_dev->lock);
-                    invalidate = !hwc_dev->force_sgx && hwc_dev->ovls_blending;
+                    invalidate = hwc_dev->last_int_ovls > 1 && !hwc_dev->force_sgx;
                     if (invalidate) {
                         hwc_dev->force_sgx = 2;
                     }
@@ -1994,6 +2724,18 @@ static int omap4_hwc_device_open(const hw_module_t* module, const char* name,
 
     hwc_dev->base.common.tag = HARDWARE_DEVICE_TAG;
     hwc_dev->base.common.version = HWC_DEVICE_API_VERSION_1_0;
+
+    char value[PROPERTY_VALUE_MAX];
+    property_get("ro.product.board", value, "");
+    if (strncmp("blaze", value, PROPERTY_VALUE_MAX) == 0) {
+        ALOGI("Revert to legacy HWC API for fake vsync");
+        hwc_dev->base.common.version = HWC_DEVICE_API_VERSION_0_2;
+    }
+
+    if (strncmp("panda5", value, PROPERTY_VALUE_MAX) == 0) {
+        ALOGI("Revert to legacy HWC API for fake vsync");
+        hwc_dev->base.common.version = HWC_DEVICE_API_VERSION_0_2;
+    }
     hwc_dev->base.common.module = (hw_module_t *)module;
     hwc_dev->base.common.close = omap4_hwc_device_close;
     hwc_dev->base.prepare = omap4_hwc_prepare;
@@ -2013,9 +2755,9 @@ static int omap4_hwc_device_open(const hw_module_t* module, const char* name,
         goto done;
     }
 
-    hwc_dev->hdmi_fb_fd = open("/dev/graphics/fb1", O_RDWR);
-    if (hwc_dev->hdmi_fb_fd < 0) {
-        ALOGE("failed to open hdmi fb (%d)", errno);
+    int ret = ioctl(hwc_dev->dsscomp_fd, DSSCIOC_QUERY_PLATFORM, &limits);
+    if (ret) {
+        ALOGE("failed to get platform limits (%d): %m", errno);
         err = -errno;
         goto done;
     }
@@ -2042,20 +2784,51 @@ static int omap4_hwc_device_open(const hw_module_t* module, const char* name,
         goto done;
     }
 
-    hwc_dev->buffers = malloc(sizeof(buffer_handle_t) * MAX_HW_OVERLAYS);
+    /* Allocate the maximum buffers that we can receive from HWC */
+    hwc_dev->buffers = malloc(sizeof(buffer_handle_t) * MAX_HWC_LAYERS);
     if (!hwc_dev->buffers) {
         err = -ENOMEM;
         goto done;
     }
 
-    int ret = ioctl(hwc_dev->dsscomp_fd, DSSCIOC_QUERY_DISPLAY, &hwc_dev->fb_dis);
+    ret = ioctl(hwc_dev->dsscomp_fd, DSSCIOC_QUERY_DISPLAY, &hwc_dev->fb_dis);
     if (ret) {
         ALOGE("failed to get display info (%d): %m", errno);
         err = -errno;
         goto done;
     }
-    hwc_dev->ext.lcd_xpy = (float) hwc_dev->fb_dis.width_in_mm / hwc_dev->fb_dis.timings.x_res /
-                            hwc_dev->fb_dis.height_in_mm       * hwc_dev->fb_dis.timings.y_res;
+
+    hwc_dev->ion_fd = ion_open();
+    if (hwc_dev->ion_fd < 0) {
+        ALOGE("failed to open ion driver (%d)", errno);
+    }
+
+    int i;
+    for (i = 0; i < NUM_EXT_DISPLAY_BACK_BUFFERS; i++) {
+        hwc_dev->ion_handles[i] = NULL;
+    }
+
+    /* use default value in case some of requested display parameters missing */
+    hwc_dev->ext.lcd_xpy = 1.0;
+    if (hwc_dev->fb_dis.timings.x_res && hwc_dev->fb_dis.height_in_mm) {
+        hwc_dev->ext.lcd_xpy = (float)
+            hwc_dev->fb_dis.width_in_mm / hwc_dev->fb_dis.timings.x_res /
+            hwc_dev->fb_dis.height_in_mm * hwc_dev->fb_dis.timings.y_res;
+    }
+
+    if (hwc_dev->fb_dis.channel == OMAP_DSS_CHANNEL_DIGIT) {
+        ALOGI("Primary display is HDMI");
+        hwc_dev->on_tv = 1;
+    } else {
+        hwc_dev->hdmi_fb_fd = open("/dev/graphics/fb1", O_RDWR);
+        if (hwc_dev->hdmi_fb_fd < 0) {
+            ALOGE("failed to open hdmi fb (%d)", errno);
+            err = -errno;
+            goto done;
+        }
+    }
+
+    set_primary_display_transform_matrix(hwc_dev);
 
     if (pipe(hwc_dev->pipe_fds) == -1) {
             ALOGE("failed to event pipe (%d): %m", errno);
@@ -2078,7 +2851,6 @@ static int omap4_hwc_device_open(const hw_module_t* module, const char* name,
     /* get debug properties */
 
     /* see if hwc is enabled at all */
-    char value[PROPERTY_VALUE_MAX];
     property_get("debug.hwc.rgb_order", value, "1");
     hwc_dev->flags_rgb_order = atoi(value);
     property_get("debug.hwc.nv12_only", value, "0");
@@ -2121,6 +2893,32 @@ static int omap4_hwc_device_open(const hw_module_t* module, const char* name,
     ALOGI("omap4_hwc_device_open(rgb_order=%d nv12_only=%d)",
         hwc_dev->flags_rgb_order, hwc_dev->flags_nv12_only);
 
+    int gc2d_fd = open("/dev/gcioctl", O_RDWR);
+    if (gc2d_fd < 0) {
+        ALOGI("Unable to open gc-core device (%d), blits disabled", errno);
+        hwc_dev->blt_policy = BLTPOLICY_DISABLED;
+    } else {
+        property_get("persist.hwc.bltmode", value, "1");
+        hwc_dev->blt_mode = atoi(value);
+        property_get("persist.hwc.bltpolicy", value, "1");
+        hwc_dev->blt_policy = atoi(value);
+        ALOGI("blitter present, blits mode %d, blits policy %d", hwc_dev->blt_mode, hwc_dev->blt_policy);
+        close(gc2d_fd);
+
+        if (rgz_get_screengeometry(hwc_dev->fb_fd, &gscrngeom,
+                hwc_dev->fb_dev->base.format) != 0) {
+            err = -EINVAL;
+            goto done;
+        }
+    }
+
+    property_get("persist.hwc.upscaled_nv12_limit", value, "2.");
+    sscanf(value, "%f", &hwc_dev->upscaled_nv12_limit);
+    if (hwc_dev->upscaled_nv12_limit < 0. || hwc_dev->upscaled_nv12_limit > 2048.) {
+        ALOGW("Invalid upscaled_nv12_limit (%s), setting to 2.", value);
+        hwc_dev->upscaled_nv12_limit = 2.;
+    }
+
 done:
     if (err && hwc_dev) {
         if (hwc_dev->dsscomp_fd >= 0)
diff --git a/hwc/rgz_2d.c b/hwc/rgz_2d.c
new file mode 100644
index 0000000..96c4374
--- /dev/null
+++ b/hwc/rgz_2d.c
@@ -0,0 +1,1676 @@
+/*
+ * Copyright (C) Texas Instruments - http://www.ti.com/
+ *
+ * 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.
+ */
+#include <stdio.h>
+#include <stdlib.h>
+#include <errno.h>
+#include <time.h>
+#include <assert.h>
+#include <strings.h>
+#include <dlfcn.h>
+
+#include <fcntl.h>
+#include <sys/mman.h>
+#include <linux/fb.h>
+#include <linux/bltsville.h>
+#include <video/dsscomp.h>
+#include <video/omap_hwc.h>
+
+#ifndef RGZ_TEST_INTEGRATION
+#include <cutils/log.h>
+#include <cutils/properties.h>
+#include <hardware/hwcomposer.h>
+#include "hal_public.h"
+#else
+#include "hwcomposer.h"
+#include "buffer_handle.h"
+#define ALIGN(x,a) (((x) + (a) - 1L) & ~((a) - 1L))
+#define HW_ALIGN   32
+#endif
+
+#include "rgz_2d.h"
+
+#ifdef RGZ_TEST_INTEGRATION
+extern void BVDump(const char* prefix, const char* tab, const struct bvbltparams* parms);
+#define BVDUMP(p,t,parms) BVDump(p, t, parms)
+#define HANDLE_TO_BUFFER(h) handle_to_buffer(h)
+#define HANDLE_TO_STRIDE(h) handle_to_stride(h)
+#else
+static int rgz_handle_to_stride(IMG_native_handle_t *h);
+#define BVDUMP(p,t,parms)
+#define HANDLE_TO_BUFFER(h) NULL
+/* Needs to be meaningful for TILER & GFX buffers and NV12 */
+#define HANDLE_TO_STRIDE(h) rgz_handle_to_stride(h)
+#endif
+#define DSTSTRIDE(dstgeom) dstgeom->virtstride
+
+/* Borrowed macros from hwc.c vvv - consider sharing later */
+#define min(a, b) ( { typeof(a) __a = (a), __b = (b); __a < __b ? __a : __b; } )
+#define max(a, b) ( { typeof(a) __a = (a), __b = (b); __a > __b ? __a : __b; } )
+#define swap(a, b) do { typeof(a) __a = (a); (a) = (b); (b) = __a; } while (0)
+
+#define WIDTH(rect) ((rect).right - (rect).left)
+#define HEIGHT(rect) ((rect).bottom - (rect).top)
+
+#define is_RGB(format) ((format) == HAL_PIXEL_FORMAT_BGRA_8888 || (format) == HAL_PIXEL_FORMAT_RGB_565 || (format) == HAL_PIXEL_FORMAT_BGRX_8888)
+#define is_BGR(format) ((format) == HAL_PIXEL_FORMAT_RGBX_8888 || (format) == HAL_PIXEL_FORMAT_RGBA_8888)
+#define is_NV12(format) ((format) == HAL_PIXEL_FORMAT_TI_NV12 || (format) == HAL_PIXEL_FORMAT_TI_NV12_PADDED)
+
+#define HAL_PIXEL_FORMAT_BGRX_8888 0x1FF
+#define HAL_PIXEL_FORMAT_TI_NV12 0x100
+#define HAL_PIXEL_FORMAT_TI_NV12_PADDED 0x101
+/* Borrowed macros from hwc.c ^^^ */
+#define is_OPAQUE(format) ((format) == HAL_PIXEL_FORMAT_RGB_565 || (format) == HAL_PIXEL_FORMAT_RGBX_8888 || (format) == HAL_PIXEL_FORMAT_BGRX_8888)
+
+/* OUTP the means for grabbing diagnostic data */
+#ifndef RGZ_TEST_INTEGRATION
+#define OUTP ALOGI
+#define OUTE ALOGE
+#else
+#define OUTP(...) { printf(__VA_ARGS__); printf("\n"); fflush(stdout); }
+#define OUTE OUTP
+#define ALOGD_IF(debug, ...) { if (debug) OUTP(__VA_ARGS__); }
+#endif
+
+#define IS_BVCMD(params) (params->op == RGZ_OUT_BVCMD_REGION || params->op == RGZ_OUT_BVCMD_PAINT)
+
+/* Number of framebuffers to track */
+#define RGZ_NUM_FB 2
+
+struct rgz_blts {
+    struct rgz_blt_entry bvcmds[RGZ_MAX_BLITS];
+    int idx;
+};
+
+
+static int rgz_hwc_layer_blit(rgz_out_params_t *params, rgz_layer_t *rgz_layer);
+static void rgz_blts_init(struct rgz_blts *blts);
+static void rgz_blts_free(struct rgz_blts *blts);
+static struct rgz_blt_entry* rgz_blts_get(struct rgz_blts *blts, rgz_out_params_t *params);
+static int rgz_blts_bvdirect(rgz_t* rgz, struct rgz_blts *blts, rgz_out_params_t *params);
+static void rgz_get_src_rect(hwc_layer_t* layer, blit_rect_t *subregion_rect, blit_rect_t *res_rect);
+static int hal_to_ocd(int color);
+static int rgz_get_orientation(unsigned int transform);
+static int rgz_get_flip_flags(unsigned int transform, int use_src2_flags);
+static int rgz_hwc_scaled(hwc_layer_t *layer);
+
+int debug = 0;
+struct rgz_blts blts;
+/* Represents a screen sized background layer */
+static hwc_layer_t bg_layer;
+
+static void svgout_header(int htmlw, int htmlh, int coordw, int coordh)
+{
+    OUTP("<svg xmlns=\"http://www.w3.org/2000/svg\""
+         "width=\"%d\" height=\"%d\""
+         "viewBox=\"0 0 %d %d\">",
+        htmlw, htmlh, coordw, coordh);
+}
+
+static void svgout_footer(void)
+{
+    OUTP("</svg>");
+}
+
+static void svgout_rect(blit_rect_t *r, char *color, char *text)
+{
+    OUTP("<rect x=\"%d\" y=\"%d\" width=\"%d\" height=\"%d\" fill=\"%s\" "
+         "fill-opacity=\"%f\" stroke=\"black\" stroke-width=\"1\" />",
+         r->left, r->top, r->right - r->left, r->bottom - r->top, color, 1.0f);
+
+    if (!text)
+        return;
+
+    OUTP("<text x=\"%d\" y=\"%d\" style=\"font-size:30\" fill=\"black\">%s"
+         "</text>",
+         r->left, r->top + 40, text);
+}
+
+static int empty_rect(blit_rect_t *r)
+{
+    return !r->left && !r->top && !r->right && !r->bottom;
+}
+
+static int get_top_rect(blit_hregion_t *hregion, int subregion, blit_rect_t **routp)
+{
+    int l = hregion->nlayers - 1;
+    do {
+        *routp = &hregion->blitrects[l][subregion];
+        if (!empty_rect(*routp))
+            break;
+    }
+    while (--l >= 0);
+    return l;
+}
+
+/*
+ * The idea here is that we walk the layers from front to back and count the
+ * number of layers in the hregion until the first layer which doesn't require
+ * blending.
+ */
+static int get_layer_ops(blit_hregion_t *hregion, int subregion, int *bottom)
+{
+    int l = hregion->nlayers - 1;
+    int ops = 0;
+    *bottom = -1;
+    do {
+        if (!empty_rect(&hregion->blitrects[l][subregion])) {
+            ops++;
+            *bottom = l;
+            hwc_layer_t *layer = hregion->rgz_layers[l]->hwc_layer;
+            IMG_native_handle_t *h = (IMG_native_handle_t *)layer->handle;
+            if ((layer->blending != HWC_BLENDING_PREMULT) || is_OPAQUE(h->iFormat))
+                break;
+        }
+    }
+    while (--l >= 0);
+    return ops;
+}
+
+static int get_layer_ops_next(blit_hregion_t *hregion, int subregion, int l)
+{
+    while (++l < hregion->nlayers) {
+        if (!empty_rect(&hregion->blitrects[l][subregion]))
+            return l;
+    }
+    return -1;
+}
+
+static int svgout_intersects_display(blit_rect_t *a, int dispw, int disph)
+{
+    return ((a->bottom > 0) && (a->top < disph) &&
+            (a->right > 0) && (a->left < dispw));
+}
+
+static void svgout_hregion(blit_hregion_t *hregion, int dispw, int disph)
+{
+    char *colors[] = {"red", "orange", "yellow", "green", "blue", "indigo", "violet", NULL};
+    int b;
+    for (b = 0; b < hregion->nsubregions; b++) {
+        blit_rect_t *rect;
+        (void)get_top_rect(hregion, b, &rect);
+        /* Only generate SVG for subregions intersecting the displayed area */
+        if (!svgout_intersects_display(rect, dispw, disph))
+            continue;
+        svgout_rect(rect, colors[b % 7], NULL);
+    }
+}
+
+static void rgz_out_svg(rgz_t *rgz, rgz_out_params_t *params)
+{
+    if (!rgz || !(rgz->state & RGZ_REGION_DATA)) {
+        OUTE("rgz_out_svg invoked with bad state");
+        return;
+    }
+    blit_hregion_t *hregions = rgz->hregions;
+    svgout_header(params->data.svg.htmlw, params->data.svg.htmlh,
+                  params->data.svg.dispw, params->data.svg.disph);
+    int i;
+    for (i = 0; i < rgz->nhregions; i++) {
+
+        OUTP("<!-- hregion %d (subcount %d)-->", i, hregions[i].nsubregions);
+        svgout_hregion(&hregions[i], params->data.svg.dispw,
+                       params->data.svg.disph);
+    }
+    svgout_footer();
+}
+
+/* XXX duplicate of hwc.c version */
+static void dump_layer(hwc_layer_t const* l, int iserr)
+{
+#define FMT(f) ((f) == HAL_PIXEL_FORMAT_TI_NV12 ? "NV12" : \
+                (f) == HAL_PIXEL_FORMAT_BGRX_8888 ? "xRGB32" : \
+                (f) == HAL_PIXEL_FORMAT_RGBX_8888 ? "xBGR32" : \
+                (f) == HAL_PIXEL_FORMAT_BGRA_8888 ? "ARGB32" : \
+                (f) == HAL_PIXEL_FORMAT_RGBA_8888 ? "ABGR32" : \
+                (f) == HAL_PIXEL_FORMAT_RGB_565 ? "RGB565" : "??")
+
+    OUTE("%stype=%d, flags=%08x, handle=%p, tr=%02x, blend=%04x, {%d,%d,%d,%d}, {%d,%d,%d,%d}",
+            iserr ? ">>  " : "    ",
+            l->compositionType, l->flags, l->handle, l->transform, l->blending,
+            l->sourceCrop.left,
+            l->sourceCrop.top,
+            l->sourceCrop.right,
+            l->sourceCrop.bottom,
+            l->displayFrame.left,
+            l->displayFrame.top,
+            l->displayFrame.right,
+            l->displayFrame.bottom);
+    if (l->handle) {
+        IMG_native_handle_t *h = (IMG_native_handle_t *)l->handle;
+        OUTE("%s%d*%d(%s)",
+            iserr ? ">>  " : "    ",
+            h->iWidth, h->iHeight, FMT(h->iFormat));
+        OUTE("hndl %p", l->handle);
+    }
+}
+
+static void dump_all(rgz_layer_t *rgz_layers, unsigned int layerno, unsigned int errlayer)
+{
+    unsigned int i;
+    for (i = 0; i < layerno; i++) {
+        hwc_layer_t *l = rgz_layers[i].hwc_layer;
+        OUTE("Layer %d", i);
+        dump_layer(l, errlayer == i);
+    }
+}
+
+static int rgz_out_bvdirect_paint(rgz_t *rgz, rgz_out_params_t *params)
+{
+    int rv = 0;
+    unsigned int i;
+    (void)rgz;
+
+    rgz_blts_init(&blts);
+
+    /* Begin from index 1 to remove the background layer from the output */
+    for (i = 1; i < rgz->rgz_layerno; i++) {
+        rv = rgz_hwc_layer_blit(params, &rgz->rgz_layers[i]);
+        if (rv) {
+            OUTE("bvdirect_paint: error in layer %d: %d", i, rv);
+            dump_all(rgz->rgz_layers, rgz->rgz_layerno, i);
+            rgz_blts_free(&blts);
+            return rv;
+        }
+    }
+    rgz_blts_bvdirect(rgz, &blts, params);
+    rgz_blts_free(&blts);
+    return rv;
+}
+
+static void rgz_set_async(struct rgz_blt_entry *e, int async)
+{
+    e->bp.flags = async ? e->bp.flags | BVFLAG_ASYNC : e->bp.flags & ~BVFLAG_ASYNC;
+}
+
+static void rgz_get_screen_info(rgz_out_params_t *params, struct bvsurfgeom **screen_geom)
+{
+    *screen_geom = params->data.bvc.dstgeom;
+}
+
+static int rgz_is_blending_disabled(rgz_out_params_t *params)
+{
+    return params->data.bvc.noblend;
+}
+
+static void rgz_get_displayframe_rect(hwc_layer_t *layer, blit_rect_t *res_rect)
+{
+    res_rect->left = layer->displayFrame.left;
+    res_rect->top = layer->displayFrame.top;
+    res_rect->bottom = layer->displayFrame.bottom;
+    res_rect->right = layer->displayFrame.right;
+}
+
+static void rgz_set_dst_data(rgz_out_params_t *params, blit_rect_t *subregion_rect,
+    struct rgz_blt_entry* e)
+{
+    struct bvsurfgeom *screen_geom;
+    rgz_get_screen_info(params, &screen_geom);
+
+    /* omaplfb is in charge of assigning the correct dstdesc in the kernel */
+    e->dstgeom.structsize = sizeof(struct bvsurfgeom);
+    e->dstgeom.format = screen_geom->format;
+    e->dstgeom.width = screen_geom->width;
+    e->dstgeom.height = screen_geom->height;
+    e->dstgeom.orientation = screen_geom->orientation;
+    e->dstgeom.virtstride = DSTSTRIDE(screen_geom);
+
+    e->bp.dstrect.left = subregion_rect->left;
+    e->bp.dstrect.top = subregion_rect->top;
+    e->bp.dstrect.width = WIDTH(*subregion_rect);
+    e->bp.dstrect.height = HEIGHT(*subregion_rect);
+}
+
+static void rgz_set_src_data(rgz_out_params_t *params, rgz_layer_t *rgz_layer,
+    blit_rect_t *subregion_rect, struct rgz_blt_entry* e, int is_src2)
+{
+    hwc_layer_t *hwc_layer = rgz_layer->hwc_layer;
+    struct bvbuffdesc *srcdesc = is_src2 ? &e->src2desc : &e->src1desc;
+    struct bvsurfgeom *srcgeom = is_src2 ? &e->src2geom : &e->src1geom;
+    struct bvrect *srcrect = is_src2 ? &e->bp.src2rect : &e->bp.src1rect;
+    IMG_native_handle_t *handle = (IMG_native_handle_t *)hwc_layer->handle;
+
+    srcdesc->structsize = sizeof(struct bvbuffdesc);
+    srcdesc->length = handle->iHeight * HANDLE_TO_STRIDE(handle);
+    srcdesc->auxptr = (void*)rgz_layer->buffidx;
+    srcgeom->structsize = sizeof(struct bvsurfgeom);
+    srcgeom->format = hal_to_ocd(handle->iFormat);
+    srcgeom->width = handle->iWidth;
+    srcgeom->height = handle->iHeight;
+    srcgeom->orientation = rgz_get_orientation(hwc_layer->transform);
+    srcgeom->virtstride = HANDLE_TO_STRIDE(handle);
+    if (hwc_layer->transform & HAL_TRANSFORM_ROT_90)
+        swap(srcgeom->width, srcgeom->height);
+
+    /* Find out what portion of the src we want to use for the blit */
+    blit_rect_t res_rect;
+    rgz_get_src_rect(hwc_layer, subregion_rect, &res_rect);
+    srcrect->left = res_rect.left;
+    srcrect->top = res_rect.top;
+    srcrect->width = WIDTH(res_rect);
+    srcrect->height = HEIGHT(res_rect);
+}
+
+/*
+ * Set the clipping rectangle, if part of the subregion rectangle is outside
+ * the boundaries of the destination, remove only the out-of-bounds area
+ */
+static void rgz_set_clip_rect(rgz_out_params_t *params, blit_rect_t *subregion_rect,
+    struct rgz_blt_entry* e)
+{
+    struct bvsurfgeom *screen_geom;
+    rgz_get_screen_info(params, &screen_geom);
+
+    blit_rect_t clip_rect;
+    clip_rect.left = max(0, subregion_rect->left);
+    clip_rect.top = max(0, subregion_rect->top);
+    clip_rect.bottom = min(screen_geom->height, subregion_rect->bottom);
+    clip_rect.right = min(screen_geom->width, subregion_rect->right);
+
+    e->bp.cliprect.left = clip_rect.left;
+    e->bp.cliprect.top = clip_rect.top;
+    e->bp.cliprect.width = WIDTH(clip_rect);
+    e->bp.cliprect.height = HEIGHT(clip_rect);
+}
+
+/*
+ * Configures blit entry to set src2 is the same as the destination
+ */
+static void rgz_set_src2_is_dst(rgz_out_params_t *params, struct rgz_blt_entry* e)
+{
+    /* omaplfb is in charge of assigning the correct src2desc in the kernel */
+    e->src2geom = e->dstgeom;
+    e->src2desc.structsize = sizeof(struct bvbuffdesc);
+    e->src2desc.auxptr = (void*)HWC_BLT_DESC_FB_FN(0);
+    e->bp.src2rect = e->bp.dstrect;
+}
+
+/*
+ * Configure the scaling mode according to the layer format
+ */
+static void rgz_cfg_scale_mode(struct rgz_blt_entry* e, hwc_layer_t *layer)
+{
+    /*
+     * TODO: Revisit scaling mode assignment later, output between GPU and GC320
+     * seem different
+     */
+    IMG_native_handle_t *handle = (IMG_native_handle_t *)layer->handle;
+    e->bp.scalemode = is_NV12(handle->iFormat) ? BVSCALE_9x9_TAP : BVSCALE_BILINEAR;
+}
+
+/*
+ * Copies src1 into the framebuffer
+ */
+static struct rgz_blt_entry* rgz_hwc_subregion_copy(rgz_out_params_t *params,
+    blit_rect_t *subregion_rect, rgz_layer_t *rgz_src1)
+{
+    struct rgz_blt_entry* e = rgz_blts_get(&blts, params);
+    hwc_layer_t *hwc_src1 = rgz_src1->hwc_layer;
+    e->bp.structsize = sizeof(struct bvbltparams);
+    e->bp.op.rop = 0xCCCC; /* SRCCOPY */
+    e->bp.flags = BVFLAG_CLIP | BVFLAG_ROP;
+    e->bp.flags |= rgz_get_flip_flags(hwc_src1->transform, 0);
+    rgz_set_async(e, 1);
+
+    blit_rect_t tmp_rect;
+    if (rgz_hwc_scaled(hwc_src1)) {
+        rgz_get_displayframe_rect(hwc_src1, &tmp_rect);
+        rgz_cfg_scale_mode(e, hwc_src1);
+    } else
+        tmp_rect = *subregion_rect;
+
+    rgz_set_src_data(params, rgz_src1, &tmp_rect, e, 0);
+    rgz_set_dst_data(params, &tmp_rect, e);
+    rgz_set_clip_rect(params, subregion_rect, e);
+
+    if((e->src1geom.format == OCDFMT_BGR124) ||
+       (e->src1geom.format == OCDFMT_RGB124) ||
+       (e->src1geom.format == OCDFMT_RGB16))
+        e->dstgeom.format = OCDFMT_BGR124;
+
+    return e;
+}
+
+/*
+ * Blends two layers and write the result in the framebuffer, src1 must be the
+ * top most layer while src2 is the one behind. If src2 is NULL means src1 will
+ * be blended with the current content of the framebuffer.
+ */
+static struct rgz_blt_entry* rgz_hwc_subregion_blend(rgz_out_params_t *params,
+    blit_rect_t *subregion_rect, rgz_layer_t *rgz_src1, rgz_layer_t *rgz_src2)
+{
+    struct rgz_blt_entry* e = rgz_blts_get(&blts, params);
+    hwc_layer_t *hwc_src1 = rgz_src1->hwc_layer;
+    e->bp.structsize = sizeof(struct bvbltparams);
+    e->bp.op.blend = BVBLEND_SRC1OVER;
+    e->bp.flags = BVFLAG_CLIP | BVFLAG_BLEND;
+    e->bp.flags |= rgz_get_flip_flags(hwc_src1->transform, 0);
+    rgz_set_async(e, 1);
+
+    blit_rect_t tmp_rect;
+    if (rgz_hwc_scaled(hwc_src1)) {
+        rgz_get_displayframe_rect(hwc_src1, &tmp_rect);
+        rgz_cfg_scale_mode(e, hwc_src1);
+    } else
+        tmp_rect = *subregion_rect;
+
+    rgz_set_src_data(params, rgz_src1, &tmp_rect, e, 0);
+    rgz_set_dst_data(params, &tmp_rect, e);
+    rgz_set_clip_rect(params, subregion_rect, e);
+
+    if (rgz_src2) {
+        /*
+         * NOTE: Due to an API limitation it's not possible to blend src1 and
+         * src2 if both have scaling, hence only src1 is used for now
+         */
+        hwc_layer_t *hwc_src2 = rgz_src2->hwc_layer;
+        if (rgz_hwc_scaled(hwc_src2))
+            OUTE("src2 layer %p has scaling, this is not supported", hwc_src2);
+        e->bp.flags |= rgz_get_flip_flags(hwc_src2->transform, 1);
+        rgz_set_src_data(params, rgz_src2, subregion_rect, e, 1);
+    } else
+        rgz_set_src2_is_dst(params, e);
+
+    return e;
+}
+
+/*
+ * Clear the destination buffer, if rect is NULL means the whole screen, rect
+ * cannot be outside the boundaries of the screen
+ */
+static void rgz_out_clrdst(rgz_out_params_t *params, blit_rect_t *rect)
+{
+    struct rgz_blt_entry* e = rgz_blts_get(&blts, params);
+    e->bp.structsize = sizeof(struct bvbltparams);
+    e->bp.op.rop = 0xCCCC; /* SRCCOPY */
+    e->bp.flags = BVFLAG_CLIP | BVFLAG_ROP;
+    rgz_set_async(e, 1);
+
+    struct bvsurfgeom *screen_geom;
+    rgz_get_screen_info(params, &screen_geom);
+
+    e->src1desc.structsize = sizeof(struct bvbuffdesc);
+    e->src1desc.length = 4; /* 1 pixel, 32bpp */
+    /*
+     * With the HWC we don't bother having a buffer for the fill we'll get the
+     * OMAPLFB to fixup the src1desc and stride if the auxiliary pointer is -1
+     */
+    e->src1desc.auxptr = (void*)-1;
+    e->src1geom.structsize = sizeof(struct bvsurfgeom);
+    e->src1geom.format = OCDFMT_RGBA24;
+    e->bp.src1rect.left = e->bp.src1rect.top = e->src1geom.orientation = 0;
+    e->src1geom.height = e->src1geom.width = e->bp.src1rect.height = e->bp.src1rect.width = 1;
+
+    blit_rect_t clear_rect;
+    if (rect) {
+        clear_rect.left = rect->left;
+        clear_rect.top = rect->top;
+        clear_rect.right = rect->right;
+        clear_rect.bottom = rect->bottom;
+    } else {
+        clear_rect.left = clear_rect.top = 0;
+        clear_rect.right = screen_geom->width;
+        clear_rect.bottom = screen_geom->height;
+    }
+
+    rgz_set_dst_data(params, &clear_rect, e);
+    rgz_set_clip_rect(params, &clear_rect, e);
+}
+
+static int rgz_out_bvcmd_paint(rgz_t *rgz, rgz_out_params_t *params)
+{
+    int rv = 0;
+    params->data.bvc.out_blits = 0;
+    params->data.bvc.out_nhndls = 0;
+    rgz_blts_init(&blts);
+    rgz_out_clrdst(params, NULL);
+
+    unsigned int i, j;
+
+    /* Begin from index 1 to remove the background layer from the output */
+    for (i = 1, j = 0; i < rgz->rgz_layerno; i++) {
+        rgz_layer_t *rgz_layer = &rgz->rgz_layers[i];
+        hwc_layer_t *l = rgz_layer->hwc_layer;
+
+        //OUTP("blitting meminfo %d", rgz->rgz_layers[i].buffidx);
+
+        /*
+         * See if it is needed to put transparent pixels where this layer
+         * is located in the screen
+         */
+        if (rgz_layer->buffidx == -1) {
+            struct bvsurfgeom *scrgeom = params->data.bvc.dstgeom;
+            blit_rect_t srcregion;
+            srcregion.left = max(0, l->displayFrame.left);
+            srcregion.top = max(0, l->displayFrame.top);
+            srcregion.bottom = min(scrgeom->height, l->displayFrame.bottom);
+            srcregion.right = min(scrgeom->width, l->displayFrame.right);
+            rgz_out_clrdst(params, &srcregion);
+            continue;
+        }
+
+        rv = rgz_hwc_layer_blit(params, rgz_layer);
+        if (rv) {
+            OUTE("bvcmd_paint: error in layer %d: %d", i, rv);
+            dump_all(rgz->rgz_layers, rgz->rgz_layerno, i);
+            rgz_blts_free(&blts);
+            return rv;
+        }
+        params->data.bvc.out_hndls[j++] = l->handle;
+        params->data.bvc.out_nhndls++;
+    }
+
+    /* Last blit is made sync to act like a fence for the previous async blits */
+    struct rgz_blt_entry* e = &blts.bvcmds[blts.idx-1];
+    rgz_set_async(e, 0);
+
+    /* FIXME: we want to be able to call rgz_blts_free and populate the actual
+     * composition data structure ourselves */
+    params->data.bvc.cmdp = blts.bvcmds;
+    params->data.bvc.cmdlen = blts.idx;
+
+    if (params->data.bvc.out_blits >= RGZ_MAX_BLITS) {
+        rv = -1;
+    // rgz_blts_free(&blts); // FIXME
+    }
+    return rv;
+}
+
+static float getscalew(hwc_layer_t *layer)
+{
+    int w = WIDTH(layer->sourceCrop);
+    int h = HEIGHT(layer->sourceCrop);
+
+    if (layer->transform & HWC_TRANSFORM_ROT_90)
+        swap(w, h);
+
+    return ((float)WIDTH(layer->displayFrame)) / (float)w;
+}
+
+static float getscaleh(hwc_layer_t *layer)
+{
+    int w = WIDTH(layer->sourceCrop);
+    int h = HEIGHT(layer->sourceCrop);
+
+    if (layer->transform & HWC_TRANSFORM_ROT_90)
+        swap(w, h);
+
+    return ((float)HEIGHT(layer->displayFrame)) / (float)h;
+}
+
+static int rgz_bswap(int *a, int *b)
+{
+    if (*a > *b) {
+        int tmp = *b;
+        *b = *a;
+        *a = tmp;
+        return 1;
+    }
+    return 0;
+}
+
+/*
+ * Simple bubble sort on an array
+ */
+static void rgz_bsort(int *a, int len)
+{
+    int i, s;
+
+    do {
+        s=0;
+        for (i=0; i+1<len; i++) {
+            if (rgz_bswap(&a[i], &a[i+1]))
+                s = 1;
+        }
+    } while (s);
+}
+
+/*
+ * Leave only unique numbers in a sorted array
+ */
+static int rgz_bunique(int *a, int len)
+{
+    int unique = 1;
+    int base = 0;
+    while (base + 1 < len) {
+        if (a[base] == a[base + 1]) {
+            int skip = 1;
+            while (base + skip < len && a[base] == a[base + skip])
+                skip++;
+            if (base + skip == len)
+                break;
+            int i;
+            for (i = 0; i < skip - 1; i++)
+                a[base + 1 + i] = a[base + skip];
+        }
+        unique++;
+        base++;
+    }
+    return unique;
+}
+
+static int rgz_hwc_layer_sortbyy(rgz_layer_t *ra, int rsz, int *out, int *width, int screen_height)
+{
+    int outsz = 0;
+    int i;
+    *width = 0;
+    for (i = 0; i < rsz; i++) {
+        hwc_layer_t *layer = ra[i].hwc_layer;
+        /* Maintain regions inside display boundaries */
+        int top = layer->displayFrame.top;
+        int bottom = layer->displayFrame.bottom;
+        out[outsz++] = max(0, top);
+        out[outsz++] = min(bottom, screen_height);
+        int right = layer->displayFrame.right;
+        *width = *width > right ? *width : right;
+    }
+    rgz_bsort(out, outsz);
+    return outsz;
+}
+
+static int rgz_hwc_intersects(blit_rect_t *a, hwc_rect_t *b)
+{
+    return ((a->bottom > b->top) && (a->top < b->bottom) &&
+            (a->right > b->left) && (a->left < b->right));
+}
+
+static void rgz_gen_blitregions(blit_hregion_t *hregion, int screen_width)
+{
+/*
+ * 1. Get the offsets (left/right positions) of each layer within the
+ *    hregion. Assume that layers describe the bounds of the hregion.
+ * 2. We should then be able to generate an array of rects
+ * 3. Each layer will have a different z-order, for each z-order
+ *    find the intersection. Some intersections will be empty.
+ */
+
+    int offsets[RGZ_SUBREGIONMAX];
+    int noffsets=0;
+    int l, r;
+    for (l = 0; l < hregion->nlayers; l++) {
+        hwc_layer_t *layer = hregion->rgz_layers[l]->hwc_layer;
+        /* Make sure the subregion is not outside the boundaries of the screen */
+        int left = layer->displayFrame.left;
+        int right = layer->displayFrame.right;
+        offsets[noffsets++] = max(0, left);
+        offsets[noffsets++] = min(right, screen_width);
+    }
+    rgz_bsort(offsets, noffsets);
+    noffsets = rgz_bunique(offsets, noffsets);
+    hregion->nsubregions = noffsets - 1;
+    bzero(hregion->blitrects, sizeof(hregion->blitrects));
+    for (r = 0; r + 1 < noffsets; r++) {
+        blit_rect_t subregion;
+        subregion.top = hregion->rect.top;
+        subregion.bottom = hregion->rect.bottom;
+        subregion.left = offsets[r];
+        subregion.right = offsets[r+1];
+
+        ALOGD_IF(debug, "                sub l %d r %d",
+            subregion.left, subregion.right);
+        for (l = 0; l < hregion->nlayers; l++) {
+            hwc_layer_t *layer = hregion->rgz_layers[l]->hwc_layer;
+            if (rgz_hwc_intersects(&subregion, &layer->displayFrame)) {
+
+                hregion->blitrects[l][r] = subregion;
+
+                ALOGD_IF(debug, "hregion->blitrects[%d][%d] (%d %d %d %d)", l, r,
+                        hregion->blitrects[l][r].left,
+                        hregion->blitrects[l][r].top,
+                        hregion->blitrects[l][r].right,
+                        hregion->blitrects[l][r].bottom);
+            }
+        }
+    }
+}
+
+static int rgz_hwc_scaled(hwc_layer_t *layer)
+{
+    int w = WIDTH(layer->sourceCrop);
+    int h = HEIGHT(layer->sourceCrop);
+
+    if (layer->transform & HWC_TRANSFORM_ROT_90)
+        swap(w, h);
+
+    return WIDTH(layer->displayFrame) != w || HEIGHT(layer->displayFrame) != h;
+}
+
+static int rgz_in_valid_hwc_layer(hwc_layer_t *layer)
+{
+    IMG_native_handle_t *handle = (IMG_native_handle_t *)layer->handle;
+    if ((layer->flags & HWC_SKIP_LAYER) || !handle)
+        return 0;
+
+    if (is_NV12(handle->iFormat))
+        return handle->iFormat == HAL_PIXEL_FORMAT_TI_NV12;
+
+    /* FIXME: The following must be removed when GC supports vertical/horizontal
+     * buffer flips, please note having a FLIP_H and FLIP_V means 180 rotation
+     * which is supported indeed
+     */
+    if (layer->transform) {
+        int is_flipped = !!(layer->transform & HWC_TRANSFORM_FLIP_H) ^ !!(layer->transform & HWC_TRANSFORM_FLIP_V);
+        if (is_flipped) {
+            ALOGE("Layer %p is flipped %d", layer, layer->transform);
+            return 0;
+        }
+    }
+
+    switch(handle->iFormat) {
+    case HAL_PIXEL_FORMAT_BGRX_8888:
+    case HAL_PIXEL_FORMAT_RGBX_8888:
+    case HAL_PIXEL_FORMAT_RGB_565:
+    case HAL_PIXEL_FORMAT_RGBA_8888:
+    case HAL_PIXEL_FORMAT_BGRA_8888:
+        break;
+    default:
+        return 0;
+    }
+    return 1;
+}
+
+/* Reset dirty region data and state */
+static void rgz_delete_region_data(rgz_t *rgz){
+    if (!rgz)
+        return;
+    if (rgz->hregions)
+        free(rgz->hregions);
+    rgz->hregions = NULL;
+    rgz->nhregions = 0;
+    rgz->state &= ~RGZ_REGION_DATA;
+}
+
+static void rgz_handle_dirty_region(rgz_t *rgz, int reset_counters)
+{
+    unsigned int i;
+    for (i = 0; i < rgz->rgz_layerno; i++) {
+        rgz_layer_t *rgz_layer = &rgz->rgz_layers[i];
+        void *new_handle;
+
+        /*
+         * We don't care about the handle for background and layers with the
+         * clear fb hint, but we want to maintain a layer state for dirty
+         * region handling.
+         */
+        if (i == 0 || rgz_layer->buffidx == -1)
+            new_handle = (void*)0x1;
+        else
+            new_handle = (void*)rgz_layer->hwc_layer->handle;
+
+        if (reset_counters || new_handle != rgz_layer->dirty_hndl) {
+            rgz_layer->dirty_count = RGZ_NUM_FB;
+            rgz_layer->dirty_hndl = new_handle;
+        } else
+            rgz_layer->dirty_count -= rgz_layer->dirty_count ? 1 : 0;
+
+    }
+}
+
+static int rgz_in_hwccheck(rgz_in_params_t *p, rgz_t *rgz)
+{
+    hwc_layer_t *layers = p->data.hwc.layers;
+    int layerno = p->data.hwc.layerno;
+
+    rgz->state &= ~RGZ_STATE_INIT;
+
+    if (!layers)
+        return -1;
+
+    /* For debugging */
+    //dump_all(layers, layerno, 0);
+
+    /*
+     * Store buffer index to be sent in the HWC Post2 list. Any overlay
+     * meminfos must come first
+     */
+    int l, memidx = 0;
+    for (l = 0; l < layerno; l++) {
+        /*
+         * Workaround: If a NV12 layer is present in the list, don't even try
+         * to blit. There is a performance degradation while playing video and
+         * using GC at the same time.
+         */
+        IMG_native_handle_t *handle = (IMG_native_handle_t *)layers[l].handle;
+        if (!(layers[l].flags & HWC_SKIP_LAYER) && handle && is_NV12(handle->iFormat))
+            return -1;
+
+        if (layers[l].compositionType == HWC_OVERLAY)
+            memidx++;
+    }
+
+    int possible_blit = 0, candidates = 0;
+
+    /*
+     * Insert the background layer at the beginning of the list, maintain a
+     * state for dirty region handling
+     */
+    rgz_layer_t *rgz_layer = &rgz->rgz_layers[0];
+    rgz_layer->hwc_layer = &bg_layer;
+
+    for (l = 0; l < layerno; l++) {
+        if (layers[l].compositionType == HWC_FRAMEBUFFER) {
+            candidates++;
+            if (rgz_in_valid_hwc_layer(&layers[l]) &&
+                    possible_blit < RGZ_INPUT_MAXLAYERS) {
+                rgz_layer_t *rgz_layer = &rgz->rgz_layers[possible_blit+1];
+                rgz_layer->hwc_layer = &layers[l];
+                rgz_layer->buffidx = memidx++;
+                possible_blit++;
+            }
+            continue;
+        }
+
+        if (layers[l].hints & HWC_HINT_CLEAR_FB) {
+            candidates++;
+            if (possible_blit < RGZ_INPUT_MAXLAYERS) {
+                /*
+                 * Use only the layer rectangle as an input to regionize when the clear
+                 * fb hint is present, mark this layer to identify it.
+                 */
+                rgz_layer_t *rgz_layer = &rgz->rgz_layers[possible_blit+1];
+                rgz_layer->buffidx = -1;
+                rgz_layer->hwc_layer = &layers[l];
+                possible_blit++;
+            }
+        }
+    }
+
+    if (!possible_blit || possible_blit != candidates) {
+        return -1;
+    }
+
+    unsigned int blit_layers = possible_blit + 1; /* Account for background layer */
+    int reset_dirty_counters = rgz->rgz_layerno != blit_layers ? 1 : 0;
+    /*
+     * The layers we are going to blit differ in number from the previous frame,
+     * we can't trust anymore the region data, calculate it again
+     */
+    if (reset_dirty_counters)
+        rgz_delete_region_data(rgz);
+
+    rgz->state |= RGZ_STATE_INIT;
+    rgz->rgz_layerno = blit_layers;
+
+    rgz_handle_dirty_region(rgz, reset_dirty_counters);
+
+    return RGZ_ALL;
+}
+
+static int rgz_in_hwc(rgz_in_params_t *p, rgz_t *rgz)
+{
+    int yentries[RGZ_SUBREGIONMAX];
+    int dispw;  /* widest layer */
+    int screen_width = p->data.hwc.dstgeom->width;
+    int screen_height = p->data.hwc.dstgeom->height;
+
+    if (!(rgz->state & RGZ_STATE_INIT)) {
+        OUTE("rgz_process started with bad state");
+        return -1;
+    }
+
+    /* If there is already region data avoid parsing it again */
+    if (rgz->state & RGZ_REGION_DATA) {
+        return 0;
+    }
+
+    int layerno = rgz->rgz_layerno;
+
+    /* Find the horizontal regions */
+    rgz_layer_t *rgz_layers = rgz->rgz_layers;
+    int ylen = rgz_hwc_layer_sortbyy(rgz_layers, layerno, yentries, &dispw, screen_height);
+
+    ylen = rgz_bunique(yentries, ylen);
+
+    /* at this point we have an array of horizontal regions */
+    rgz->nhregions = ylen - 1;
+
+    blit_hregion_t *hregions = calloc(rgz->nhregions, sizeof(blit_hregion_t));
+    if (!hregions) {
+        OUTE("Unable to allocate memory for hregions");
+        return -1;
+    }
+    rgz->hregions = hregions;
+
+    ALOGD_IF(debug, "Allocated %d regions (sz = %d), layerno = %d", rgz->nhregions, rgz->nhregions * sizeof(blit_hregion_t), layerno);
+    int i, j;
+    for (i = 0; i < rgz->nhregions; i++) {
+        hregions[i].rect.top = yentries[i];
+        hregions[i].rect.bottom = yentries[i+1];
+        /* Avoid hregions outside the display boundaries */
+        hregions[i].rect.left = 0;
+        hregions[i].rect.right = dispw > screen_width ? screen_width : dispw;
+        hregions[i].nlayers = 0;
+        for (j = 0; j < layerno; j++) {
+            hwc_layer_t *layer = rgz_layers[j].hwc_layer;
+            if (rgz_hwc_intersects(&hregions[i].rect, &layer->displayFrame)) {
+                int l = hregions[i].nlayers++;
+                hregions[i].rgz_layers[l] = &rgz_layers[j];
+            }
+        }
+    }
+
+    /* Calculate blit regions */
+    for (i = 0; i < rgz->nhregions; i++) {
+        rgz_gen_blitregions(&hregions[i], screen_width);
+        ALOGD_IF(debug, "hregion %3d: nsubregions %d", i, hregions[i].nsubregions);
+        ALOGD_IF(debug, "           : %d to %d: ",
+            hregions[i].rect.top, hregions[i].rect.bottom);
+        for (j = 0; j < hregions[i].nlayers; j++)
+            ALOGD_IF(debug, "              %p ", hregions[i].rgz_layers[j]->hwc_layer);
+    }
+    rgz->state |= RGZ_REGION_DATA;
+    return 0;
+}
+
+/*
+ * generate a human readable description of the layer
+ *
+ * idx, flags, fmt, type, sleft, stop, sright, sbot, dleft, dtop, \
+ * dright, dbot, rot, flip, blending, scalew, scaleh, visrects
+ *
+ */
+static void rgz_print_layer(hwc_layer_t *l, int idx, int csv)
+{
+    char big_log[1024];
+    int e = sizeof(big_log);
+    char *end = big_log + e;
+    e -= snprintf(end - e, e, "<!-- LAYER-DAT: %d", idx);
+
+
+    e -= snprintf(end - e, e, "%s %p", csv ? "," : " hndl:",
+            l->handle ? l->handle : NULL);
+
+    e -= snprintf(end - e, e, "%s %s", csv ? "," : " flags:",
+        l->flags & HWC_SKIP_LAYER ? "skip" : "none");
+
+    IMG_native_handle_t *handle = (IMG_native_handle_t *)l->handle;
+    if (handle) {
+        e -= snprintf(end - e, e, "%s", csv ? ", " : " fmt: ");
+        switch(handle->iFormat) {
+        case HAL_PIXEL_FORMAT_BGRA_8888:
+            e -= snprintf(end - e, e, "bgra"); break;
+        case HAL_PIXEL_FORMAT_RGB_565:
+            e -= snprintf(end - e, e, "rgb565"); break;
+        case HAL_PIXEL_FORMAT_BGRX_8888:
+            e -= snprintf(end - e, e, "bgrx"); break;
+        case HAL_PIXEL_FORMAT_RGBX_8888:
+            e -= snprintf(end - e, e, "rgbx"); break;
+        case HAL_PIXEL_FORMAT_RGBA_8888:
+            e -= snprintf(end - e, e, "rgba"); break;
+        case HAL_PIXEL_FORMAT_TI_NV12:
+        case HAL_PIXEL_FORMAT_TI_NV12_PADDED:
+            e -= snprintf(end - e, e, "nv12"); break;
+        default:
+            e -= snprintf(end - e, e, "unknown");
+        }
+        e -= snprintf(end - e, e, "%s", csv ? ", " : " type: ");
+        if (handle->usage & GRALLOC_USAGE_HW_RENDER)
+            e -= snprintf(end - e, e, "hw");
+        else if (handle->usage & GRALLOC_USAGE_SW_READ_MASK ||
+                 handle->usage & GRALLOC_USAGE_SW_WRITE_MASK)
+            e -= snprintf(end - e, e, "sw");
+        else
+            e -= snprintf(end - e, e, "unknown");
+    } else {
+        e -= snprintf(end - e, e, csv ? ", unknown" : " fmt: unknown");
+        e -= snprintf(end - e, e, csv ? ", na" : " type: na");
+    }
+    e -= snprintf(end - e, e, csv ? ", %d, %d, %d, %d" : " src: %d %d %d %d",
+        l->sourceCrop.left, l->sourceCrop.top, l->sourceCrop.right,
+        l->sourceCrop.bottom);
+    e -= snprintf(end - e, e, csv ? ", %d, %d, %d, %d" : " disp: %d %d %d %d",
+        l->displayFrame.left, l->displayFrame.top,
+        l->displayFrame.right, l->displayFrame.bottom);
+
+    e -= snprintf(end - e, e, "%s %s", csv ? "," : " rot:",
+        l->transform & HWC_TRANSFORM_ROT_90 ? "90" :
+            l->transform & HWC_TRANSFORM_ROT_180 ? "180" :
+            l->transform & HWC_TRANSFORM_ROT_270 ? "270" : "none");
+
+    char flip[5] = "";
+    strcat(flip, l->transform & HWC_TRANSFORM_FLIP_H ? "H" : "");
+    strcat(flip, l->transform & HWC_TRANSFORM_FLIP_V ? "V" : "");
+    if (!(l->transform & (HWC_TRANSFORM_FLIP_V|HWC_TRANSFORM_FLIP_H)))
+        strcpy(flip, "none");
+    e -= snprintf(end - e, e, "%s %s", csv ? "," : " flip:", flip);
+
+    e -= snprintf(end - e, e, "%s %s", csv ? "," : " blending:",
+        l->blending == HWC_BLENDING_NONE ? "none" :
+        l->blending == HWC_BLENDING_PREMULT ? "premult" :
+        l->blending == HWC_BLENDING_COVERAGE ? "coverage" : "invalid");
+
+    e -= snprintf(end - e, e, "%s %1.3f", csv ? "," : " scalew:", getscalew(l));
+    e -= snprintf(end - e, e, "%s %1.3f", csv ? "," : " scaleh:", getscaleh(l));
+
+    e -= snprintf(end - e, e, "%s %d", csv ? "," : " visrect:",
+        l->visibleRegionScreen.numRects);
+
+    if (!csv) {
+        e -= snprintf(end - e, e, " -->");
+        OUTP("%s", big_log);
+
+        size_t i = 0;
+        for (; i < l->visibleRegionScreen.numRects; i++) {
+            hwc_rect_t const *r = &l->visibleRegionScreen.rects[i];
+            OUTP("<!-- LAYER-VIS: %d: rect: %d %d %d %d -->",
+                    i, r->left, r->top, r->right, r->bottom);
+        }
+    } else {
+        size_t i = 0;
+        for (; i < l->visibleRegionScreen.numRects; i++) {
+            hwc_rect_t const *r = &l->visibleRegionScreen.rects[i];
+            e -= snprintf(end - e, e, ", %d, %d, %d, %d",
+                    r->left, r->top, r->right, r->bottom);
+        }
+        e -= snprintf(end - e, e, " -->");
+        OUTP("%s", big_log);
+    }
+}
+
+static void rgz_print_layers(hwc_layer_list_t* list, int csv)
+{
+    size_t i;
+    for (i = 0; i < list->numHwLayers; i++) {
+        hwc_layer_t *l = &list->hwLayers[i];
+        rgz_print_layer(l, i, csv);
+    }
+}
+
+static int hal_to_ocd(int color)
+{
+    switch(color) {
+    case HAL_PIXEL_FORMAT_BGRA_8888:
+        return OCDFMT_BGRA24;
+    case HAL_PIXEL_FORMAT_BGRX_8888:
+        return OCDFMT_BGR124;
+    case HAL_PIXEL_FORMAT_RGB_565:
+        return OCDFMT_RGB16;
+    case HAL_PIXEL_FORMAT_RGBA_8888:
+        return OCDFMT_RGBA24;
+    case HAL_PIXEL_FORMAT_RGBX_8888:
+        return OCDFMT_RGB124;
+    case HAL_PIXEL_FORMAT_TI_NV12:
+        return OCDFMT_NV12;
+    case HAL_PIXEL_FORMAT_YV12:
+        return OCDFMT_YV12;
+    default:
+        return OCDFMT_UNKNOWN;
+    }
+}
+
+/*
+ * The loadbltsville fn is only needed for testing, the bltsville shared
+ * libraries aren't planned to be used directly in production code here
+ */
+static BVFN_MAP bv_map;
+static BVFN_BLT bv_blt;
+static BVFN_UNMAP bv_unmap;
+#ifndef RGZ_TEST_INTEGRATION
+gralloc_module_t const *gralloc;
+#endif
+#define BLTSVILLELIB "libbltsville_cpu.so"
+
+#ifdef RGZ_TEST_INTEGRATION
+static int loadbltsville(void)
+{
+    void *hndl = dlopen(BLTSVILLELIB, RTLD_LOCAL | RTLD_LAZY);
+    if (!hndl) {
+        OUTE("Loading bltsville failed");
+        return -1;
+    }
+    bv_map = (BVFN_MAP)dlsym(hndl, "bv_map");
+    bv_blt = (BVFN_BLT)dlsym(hndl, "bv_blt");
+    bv_unmap = (BVFN_UNMAP)dlsym(hndl, "bv_unmap");
+    if(!bv_blt || !bv_map || !bv_unmap) {
+        OUTE("Missing bltsville fn %p %p %p", bv_map, bv_blt, bv_unmap);
+        return -1;
+    }
+    OUTP("Loaded %s", BLTSVILLELIB);
+
+#ifndef RGZ_TEST_INTEGRATION
+    hw_module_t const* module;
+    int err = hw_get_module(GRALLOC_HARDWARE_MODULE_ID, &module);
+    if (err != 0) {
+        OUTE("Loading gralloc failed");
+        return -1;
+    }
+    gralloc = (gralloc_module_t const *)module;
+#endif
+    return 0;
+}
+#else
+static int loadbltsville(void) {
+    return 0;
+}
+#endif
+
+#ifndef RGZ_TEST_INTEGRATION
+static int rgz_handle_to_stride(IMG_native_handle_t *h)
+{
+    int bpp = is_NV12(h->iFormat) ? 0 : (h->iFormat == HAL_PIXEL_FORMAT_RGB_565 ? 2 : 4);
+    int stride = ALIGN(h->iWidth, HW_ALIGN) * bpp;
+    return stride;
+}
+
+#endif
+
+extern void BVDump(const char* prefix, const char* tab, const struct bvbltparams* parms);
+
+static int rgz_get_orientation(unsigned int transform)
+{
+    int orientation = 0;
+    if ((transform & HWC_TRANSFORM_FLIP_H) && (transform & HWC_TRANSFORM_FLIP_V))
+        orientation += 180;
+    if (transform & HWC_TRANSFORM_ROT_90)
+        orientation += 90;
+
+    return orientation;
+}
+
+static int rgz_get_flip_flags(unsigned int transform, int use_src2_flags)
+{
+    /*
+     * If vertical and horizontal flip flags are set it means a 180 rotation
+     * (with no flip) is intended for the layer, so we return 0 in that case.
+     */
+    int flip_flags = 0;
+    if (transform & HWC_TRANSFORM_FLIP_H)
+        flip_flags |= (use_src2_flags ? BVFLAG_HORZ_FLIP_SRC2 : BVFLAG_HORZ_FLIP_SRC1);
+    if (transform & HWC_TRANSFORM_FLIP_V)
+        flip_flags = flip_flags ? 0 : flip_flags | (use_src2_flags ? BVFLAG_VERT_FLIP_SRC2 : BVFLAG_VERT_FLIP_SRC1);
+    return flip_flags;
+}
+
+static int rgz_hwc_layer_blit(rgz_out_params_t *params, rgz_layer_t *rgz_layer)
+{
+    static int loaded = 0;
+    if (!loaded)
+        loaded = loadbltsville() ? : 1; /* attempt load once */
+
+    hwc_layer_t* layer = rgz_layer->hwc_layer;
+    blit_rect_t srcregion;
+    rgz_get_displayframe_rect(layer, &srcregion);
+
+    int noblend = rgz_is_blending_disabled(params);
+    if (!noblend && layer->blending == HWC_BLENDING_PREMULT)
+        rgz_hwc_subregion_blend(params, &srcregion, rgz_layer, NULL);
+    else
+        rgz_hwc_subregion_copy(params, &srcregion, rgz_layer);
+
+    return 0;
+}
+
+/*
+ * Calculate the src rectangle on the basis of the layer display, source crop
+ * and subregion rectangles. Additionally any rotation will be taken in
+ * account. The resulting rectangle is written in res_rect.
+ */
+static void rgz_get_src_rect(hwc_layer_t* layer, blit_rect_t *subregion_rect, blit_rect_t *res_rect)
+{
+    IMG_native_handle_t *handle = (IMG_native_handle_t *)layer->handle;
+    int res_left = 0;
+    int res_top = 0;
+    int delta_left;
+    int delta_top;
+    int res_width;
+    int res_height;
+
+    /*
+     * If the layer is scaled we use the whole cropping rectangle from the
+     * source and just move the clipping rectangle for the region we want to
+     * blit, this is done to prevent any artifacts when blitting subregions of
+     * a scaled layer. If there is a transform, adjust the width and height
+     * accordingly to match the rotated buffer geometry.
+     */
+    if (rgz_hwc_scaled(layer)) {
+        delta_top = 0;
+        delta_left = 0;
+        res_width = WIDTH(layer->sourceCrop);
+        res_height = HEIGHT(layer->sourceCrop);
+        if (layer->transform & HAL_TRANSFORM_ROT_90)
+            swap(res_width , res_height);
+    } else {
+        delta_top = subregion_rect->top - layer->displayFrame.top;
+        delta_left = subregion_rect->left - layer->displayFrame.left;
+        res_width = WIDTH(*subregion_rect);
+        res_height = HEIGHT(*subregion_rect);
+    }
+
+    /*
+     * Calculate the top, left offset from the source cropping rectangle
+     * depending on the rotation
+     */
+    switch(layer->transform) {
+        case 0:
+            res_left = layer->sourceCrop.left + delta_left;
+            res_top = layer->sourceCrop.top + delta_top;
+            break;
+        case HAL_TRANSFORM_ROT_90:
+            res_left = handle->iHeight - layer->sourceCrop.bottom + delta_left;
+            res_top = layer->sourceCrop.left + delta_top;
+            break;
+        case HAL_TRANSFORM_ROT_180:
+            res_left = handle->iWidth - layer->sourceCrop.right + delta_left;
+            res_top = handle->iHeight - layer->sourceCrop.bottom + delta_top;
+            break;
+        case HAL_TRANSFORM_ROT_270:
+            res_left = layer->sourceCrop.top + delta_left;
+            res_top = handle->iWidth - layer->sourceCrop.right + delta_top;
+            break;
+        default:
+            OUTE("Invalid transform value %d", layer->transform);
+    }
+
+    /* Resulting rectangle has the subregion dimensions */
+    res_rect->left = res_left;
+    res_rect->top = res_top;
+    res_rect->right = res_left + res_width;
+    res_rect->bottom = res_top + res_height;
+}
+
+static void rgz_batch_entry(struct rgz_blt_entry* e, unsigned int flag, unsigned int set)
+{
+    e->bp.flags &= ~BVFLAG_BATCH_MASK;
+    e->bp.flags |= flag;
+    e->bp.batchflags |= set;
+}
+
+static int rgz_hwc_subregion_blit(blit_hregion_t *hregion, int sidx, rgz_out_params_t *params)
+{
+    static int loaded = 0;
+    if (!loaded)
+        loaded = loadbltsville() ? : 1; /* attempt load once */
+
+    int lix;
+    int ldepth = get_layer_ops(hregion, sidx, &lix);
+    if (ldepth == 0) {
+        /* Impossible, there are no layers in this region even if the
+         * background is covering the whole screen
+         */
+        OUTE("hregion %p subregion %d doesn't have any ops", hregion, sidx);
+        return -1;
+    }
+
+    /* Determine if this region is dirty */
+    int dirty = 0, dirtylix = lix;
+    while (dirtylix != -1) {
+        rgz_layer_t *rgz_layer = hregion->rgz_layers[dirtylix];
+        if (rgz_layer->dirty_count){
+            /* One of the layers is dirty, we need to generate blits for this subregion */
+            dirty = 1;
+            break;
+        }
+        dirtylix = get_layer_ops_next(hregion, sidx, dirtylix);
+    }
+
+    if (!dirty)
+        return 0;
+
+    /* Check if the bottom layer is the background */
+    if (hregion->rgz_layers[lix]->hwc_layer == &bg_layer) {
+        if (ldepth == 1) {
+            /* Background layer is the only operation, clear subregion */
+            rgz_out_clrdst(params, &hregion->blitrects[lix][sidx]);
+            return 0;
+        } else {
+            /* No need to generate blits with background layer if there is
+             * another layer on top of it, discard it
+             */
+            ldepth--;
+            lix = get_layer_ops_next(hregion, sidx, lix);
+        }
+    }
+
+    /*
+     * See if the depth most layer needs to be ignored. If this layer is the
+     * only operation, we need to clear this subregion.
+     */
+    if (hregion->rgz_layers[lix]->buffidx == -1) {
+        ldepth--;
+        if (!ldepth) {
+            rgz_out_clrdst(params, &hregion->blitrects[lix][sidx]);
+            return 0;
+        }
+        lix = get_layer_ops_next(hregion, sidx, lix);
+    }
+
+    int noblend = rgz_is_blending_disabled(params);
+
+    if (!noblend && ldepth > 1) { /* BLEND */
+        blit_rect_t *rect = &hregion->blitrects[lix][sidx];
+        struct rgz_blt_entry* e;
+
+        int s2lix = lix;
+        lix = get_layer_ops_next(hregion, sidx, lix);
+
+        /*
+         * We save a read and a write from the FB if we blend the bottom
+         * two layers, we can do this only if both layers are not scaled
+         */
+        int first_batchflags = 0;
+        if (!rgz_hwc_scaled(hregion->rgz_layers[lix]->hwc_layer) &&
+            !rgz_hwc_scaled(hregion->rgz_layers[s2lix]->hwc_layer)) {
+            e = rgz_hwc_subregion_blend(params, rect, hregion->rgz_layers[lix],
+                hregion->rgz_layers[s2lix]);
+            first_batchflags |= BVBATCH_SRC2;
+        } else {
+            /* Return index to the first operation and make a copy of the first layer */
+            lix = s2lix;
+            e = rgz_hwc_subregion_copy(params, rect, hregion->rgz_layers[lix]);
+            first_batchflags |= BVBATCH_OP | BVBATCH_SRC2;
+        }
+        rgz_batch_entry(e, BVFLAG_BATCH_BEGIN, 0);
+
+        /* Rest of layers blended with FB */
+        int first = 1;
+        while((lix = get_layer_ops_next(hregion, sidx, lix)) != -1) {
+            int batchflags = 0;
+            e = rgz_hwc_subregion_blend(params, rect, hregion->rgz_layers[lix], NULL);
+            if (first) {
+                first = 0;
+                batchflags |= first_batchflags;
+            }
+            /*
+             * TODO: This will work when scaling is introduced, however we need
+             * to think on a better way to optimize this.
+             */
+            batchflags |= BVBATCH_SRC1 | BVBATCH_SRC1RECT_ORIGIN| BVBATCH_SRC1RECT_SIZE |
+                BVBATCH_DSTRECT_ORIGIN | BVBATCH_DSTRECT_SIZE | BVBATCH_SRC2RECT_ORIGIN |
+                BVBATCH_SRC2RECT_SIZE | BVBATCH_SCALE;
+            rgz_batch_entry(e, BVFLAG_BATCH_CONTINUE, batchflags);
+        }
+
+        if (e->bp.flags & BVFLAG_BATCH_BEGIN)
+            rgz_batch_entry(e, 0, 0);
+        else
+            rgz_batch_entry(e, BVFLAG_BATCH_END, 0);
+
+    } else { /* COPY */
+        blit_rect_t *rect = &hregion->blitrects[lix][sidx];
+        if (noblend)    /* get_layer_ops() doesn't understand this so get the top */
+            lix = get_top_rect(hregion, sidx, &rect);
+        rgz_hwc_subregion_copy(params, rect, hregion->rgz_layers[lix]);
+    }
+    return 0;
+}
+
+struct bvbuffdesc gscrndesc = {
+    .structsize = sizeof(struct bvbuffdesc), .length = 0,
+    .auxptr = MAP_FAILED
+};
+struct bvsurfgeom gscrngeom = {
+    .structsize = sizeof(struct bvsurfgeom), .format = OCDFMT_UNKNOWN
+};
+
+static void rgz_blts_init(struct rgz_blts *blts)
+{
+    bzero(blts, sizeof(*blts));
+}
+
+static void rgz_blts_free(struct rgz_blts *blts)
+{
+    /* TODO ??? maybe we should dynamically allocate this */
+    rgz_blts_init(blts);
+}
+
+static struct rgz_blt_entry* rgz_blts_get(struct rgz_blts *blts, rgz_out_params_t *params)
+{
+    struct rgz_blt_entry *ne;
+    if (blts->idx < RGZ_MAX_BLITS) {
+        ne = &blts->bvcmds[blts->idx++];
+        if (IS_BVCMD(params))
+            params->data.bvc.out_blits++;
+    } else {
+        OUTE("!!! BIG PROBLEM !!! run out of blit entries");
+        ne = &blts->bvcmds[blts->idx - 1]; /* Return last slot */
+    }
+    return ne;
+}
+
+static int rgz_blts_bvdirect(rgz_t *rgz, struct rgz_blts *blts, rgz_out_params_t *params)
+{
+    struct bvbatch *batch = NULL;
+    int rv = -1;
+    int idx = 0;
+
+    while (idx < blts->idx) {
+        struct rgz_blt_entry *e = &blts->bvcmds[idx];
+        if (e->bp.flags & BVFLAG_BATCH_MASK)
+            e->bp.batch = batch;
+        rv = bv_blt(&e->bp);
+        if (rv) {
+            OUTE("BV_BLT failed: %d", rv);
+            BVDUMP("bv_blt:", "  ", &e->bp);
+            return -1;
+        }
+        if (e->bp.flags & BVFLAG_BATCH_BEGIN)
+            batch = e->bp.batch;
+        idx++;
+    }
+    return rv;
+}
+
+static int rgz_out_region(rgz_t *rgz, rgz_out_params_t *params)
+{
+    if (!(rgz->state & RGZ_REGION_DATA)) {
+        OUTE("rgz_out_region invoked with bad state");
+        return -1;
+    }
+
+    rgz_blts_init(&blts);
+    ALOGD_IF(debug, "rgz_out_region:");
+
+    if (IS_BVCMD(params))
+        params->data.bvc.out_blits = 0;
+
+    int i;
+    for (i = 0; i < rgz->nhregions; i++) {
+        blit_hregion_t *hregion = &rgz->hregions[i];
+        int s;
+        ALOGD_IF(debug, "h[%d] nsubregions = %d", i, hregion->nsubregions);
+        if (hregion->nlayers == 0) {
+            /* Impossible, there are no layers in this region even if the
+             * background is covering the whole screen
+             */
+            OUTE("hregion %p doesn't have any ops", hregion);
+            return -1;
+        }
+        for (s = 0; s < hregion->nsubregions; s++) {
+            ALOGD_IF(debug, "h[%d] -> [%d]", i, s);
+            if (rgz_hwc_subregion_blit(hregion, s, params))
+                return -1;
+        }
+    }
+
+    int rv = 0;
+
+    if (IS_BVCMD(params)) {
+        unsigned int j;
+        params->data.bvc.out_nhndls = 0;
+        /* Begin from index 1 to remove the background layer from the output */
+        for (j = 1, i = 0; j < rgz->rgz_layerno; j++) {
+            rgz_layer_t *rgz_layer = &rgz->rgz_layers[j];
+            /* We don't need the handles for layers marked as -1 */
+            if (rgz_layer->buffidx == -1)
+                continue;
+            hwc_layer_t *layer = rgz_layer->hwc_layer;
+            params->data.bvc.out_hndls[i++] = layer->handle;
+            params->data.bvc.out_nhndls++;
+        }
+
+        if (blts.idx > 0) {
+            /* Last blit is made sync to act like a fence for the previous async blits */
+            struct rgz_blt_entry* e = &blts.bvcmds[blts.idx-1];
+            rgz_set_async(e, 0);
+        }
+
+        /* FIXME: we want to be able to call rgz_blts_free and populate the actual
+         * composition data structure ourselves */
+        params->data.bvc.cmdp = blts.bvcmds;
+        params->data.bvc.cmdlen = blts.idx;
+        if (params->data.bvc.out_blits >= RGZ_MAX_BLITS)
+            rv = -1;
+        //rgz_blts_free(&blts);
+    } else {
+        rv = rgz_blts_bvdirect(rgz, &blts, params);
+        rgz_blts_free(&blts);
+    }
+
+    return rv;
+}
+
+void rgz_profile_hwc(hwc_layer_list_t* list, int dispw, int disph)
+{
+    if (!list)  /* A NULL composition list can occur */
+        return;
+
+#ifndef RGZ_TEST_INTEGRATION
+    static char regiondump2[PROPERTY_VALUE_MAX] = "";
+    char regiondump[PROPERTY_VALUE_MAX];
+    property_get("debug.2dhwc.region", regiondump, "0");
+    int dumpregions = strncmp(regiondump, regiondump2, PROPERTY_VALUE_MAX);
+    if (dumpregions)
+        strncpy(regiondump2, regiondump, PROPERTY_VALUE_MAX);
+    else {
+        dumpregions = !strncmp(regiondump, "all", PROPERTY_VALUE_MAX) &&
+                      (list->flags & HWC_GEOMETRY_CHANGED);
+        static int iteration = 0;
+        if (dumpregions)
+            sprintf(regiondump, "iteration %d", iteration++);
+    }
+
+    char dumplayerdata[PROPERTY_VALUE_MAX];
+    /* 0 - off, 1 - human readable, 2 - CSV */
+    property_get("debug.2dhwc.dumplayers", dumplayerdata, "0");
+    int dumplayers = atoi(dumplayerdata);
+#else
+    char regiondump[] = "";
+    int dumplayers = 1;
+    int dumpregions = 0;
+#endif
+    if (dumplayers && (list->flags & HWC_GEOMETRY_CHANGED)) {
+        OUTP("<!-- BEGUN-LAYER-DUMP: %d -->", list->numHwLayers);
+        rgz_print_layers(list, dumplayers == 1 ? 0 : 1);
+        OUTP("<!-- ENDED-LAYER-DUMP -->");
+    }
+
+    if(!dumpregions)
+        return;
+
+    rgz_t rgz;
+    rgz_in_params_t ip = { .data = { .hwc = {
+                           .layers = list->hwLayers,
+                           .layerno = list->numHwLayers } } };
+    ip.op = RGZ_IN_HWCCHK;
+    if (rgz_in(&ip, &rgz) == RGZ_ALL) {
+        ip.op = RGZ_IN_HWC;
+        if (rgz_in(&ip, &rgz) == RGZ_ALL) {
+            OUTP("<!-- BEGUN-SVG-DUMP: %s -->", regiondump);
+            OUTP("<b>%s</b>", regiondump);
+            rgz_out_params_t op = {
+                .op = RGZ_OUT_SVG,
+                .data = {
+                    .svg = {
+                        .dispw = dispw, .disph = disph,
+                        .htmlw = 450, .htmlh = 800
+                    }
+                },
+            };
+            rgz_out(&rgz, &op);
+            OUTP("<!-- ENDED-SVG-DUMP -->");
+        }
+    }
+    rgz_release(&rgz);
+}
+
+int rgz_get_screengeometry(int fd, struct bvsurfgeom *geom, int fmt)
+{
+    /* Populate Bltsville destination buffer information with framebuffer data */
+    struct fb_fix_screeninfo fb_fixinfo;
+    struct fb_var_screeninfo fb_varinfo;
+
+    ALOGI("Attempting to get framebuffer device info.");
+    if(ioctl(fd, FBIOGET_FSCREENINFO, &fb_fixinfo)) {
+        OUTE("Error getting fb_fixinfo");
+        return -EINVAL;
+    }
+
+    if(ioctl(fd, FBIOGET_VSCREENINFO, &fb_varinfo)) {
+        ALOGE("Error gettting fb_varinfo");
+        return -EINVAL;
+    }
+
+    bzero(&bg_layer, sizeof(bg_layer));
+    bg_layer.displayFrame.left = bg_layer.displayFrame.top = 0;
+    bg_layer.displayFrame.right = fb_varinfo.xres;
+    bg_layer.displayFrame.bottom = fb_varinfo.yres;
+
+    bzero(geom, sizeof(*geom));
+    geom->structsize = sizeof(*geom);
+    geom->width = fb_varinfo.xres;
+    geom->height = fb_varinfo.yres;
+    geom->virtstride = fb_fixinfo.line_length;
+    geom->format = hal_to_ocd(fmt);
+    /* Always set to 0, src buffers will contain rotation values as needed */
+    geom->orientation = 0;
+    return 0;
+}
+
+int rgz_in(rgz_in_params_t *p, rgz_t *rgz)
+{
+    int rv = -1;
+    switch (p->op) {
+    case RGZ_IN_HWC:
+        rv = rgz_in_hwccheck(p, rgz);
+        if (rv == RGZ_ALL)
+            rv = rgz_in_hwc(p, rgz) ? 0 : RGZ_ALL;
+        break;
+    case RGZ_IN_HWCCHK:
+        bzero(rgz, sizeof(rgz_t));
+        rv = rgz_in_hwccheck(p, rgz);
+        break;
+    default:
+        return -1;
+    }
+    return rv;
+}
+
+void rgz_release(rgz_t *rgz)
+{
+    if (!rgz)
+        return;
+    if (rgz->hregions)
+        free(rgz->hregions);
+    bzero(rgz, sizeof(*rgz));
+}
+
+int rgz_out(rgz_t *rgz, rgz_out_params_t *params)
+{
+    switch (params->op) {
+    case RGZ_OUT_SVG:
+        rgz_out_svg(rgz, params);
+        return 0;
+    case RGZ_OUT_BVDIRECT_PAINT:
+        return rgz_out_bvdirect_paint(rgz, params);
+    case RGZ_OUT_BVCMD_PAINT:
+        return rgz_out_bvcmd_paint(rgz, params);
+    case RGZ_OUT_BVDIRECT_REGION:
+    case RGZ_OUT_BVCMD_REGION:
+        return rgz_out_region(rgz, params);
+    default:
+        return -1;
+    }
+}
+
diff --git a/hwc/rgz_2d.h b/hwc/rgz_2d.h
new file mode 100644
index 0000000..44867b8
--- /dev/null
+++ b/hwc/rgz_2d.h
@@ -0,0 +1,290 @@
+/*
+ * Copyright (C) Texas Instruments - http://www.ti.com/
+ *
+ * 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.
+ */
+#ifndef __RGZ_2D__
+#define __RGZ_2D__
+
+#include <linux/bltsville.h>
+
+/*
+ * Maximum number of layers used to generate subregion rectangles in a
+ * horizontal region.
+ */
+#define RGZ_MAXLAYERS 13
+
+/*
+ * Maximum number of layers the regionizer will accept as input. Account for an
+ * additional 'background layer' to generate empty subregion rectangles.
+ */
+#define RGZ_INPUT_MAXLAYERS (RGZ_MAXLAYERS - 1)
+
+/*
+ * Regionizer data
+ *
+ * This is an oqaque structure passed in by the client
+ */
+struct rgz;
+typedef struct rgz rgz_t;
+
+/*
+ * With an open framebuffer file descriptor get the geometry of
+ * the device
+ */
+int rgz_get_screengeometry(int fd, struct bvsurfgeom *geom, int fmt);
+
+/*
+ * Regionizer input parameters
+ */
+struct rgz_in_hwc {
+    int flags;
+    int layerno;
+    hwc_layer_t *layers;
+    struct bvsurfgeom *dstgeom;
+};
+
+typedef struct rgz_in_params {
+    int op; /* See RGZ_IN_* */
+    union {
+        struct rgz_in_hwc hwc;
+    } data;
+} rgz_in_params_t;
+
+/*
+ * Validate whether the HWC layers can be rendered
+ *
+ * Arguments (rgz_in_params_t):
+ * op                RGZ_IN_HWCCHK
+ * data.hwc.layers   HWC layer array
+ * data.hwc.layerno  HWC layer array size
+ *
+ * Returns:
+ * rv = RGZ_ALL, -1 failure
+ */
+#define RGZ_IN_HWCCHK 1
+
+/*
+ * Regionize the HWC layers
+ *
+ * This generates region data which can be used with regionizer
+ * output function. This call will validate whether all or some of the
+ * layers can be rendered.
+ *
+ * The caller must use rgz_release when done with the region data
+ *
+ * Arguments (rgz_in_params_t):
+ * op                RGZ_IN_HWC
+ * data.hwc.layers   HWC layer array
+ * data.hwc.layerno  HWC layer array size
+ *
+ * Returns:
+ * rv = RGZ_ALL, -1 failure
+ */
+#define RGZ_IN_HWC 2
+
+int rgz_in(rgz_in_params_t *param, rgz_t *rgz);
+
+/* This means all layers can be blitted */
+#define RGZ_ALL 1
+
+/*
+ * Free regionizer resources
+ */
+void rgz_release(rgz_t *rgz);
+
+/*
+ * Regionizer output operations
+ */
+struct rgz_out_bvcmd {
+    void *cmdp;
+    int cmdlen;
+    struct bvsurfgeom *dstgeom;
+    int noblend;
+    buffer_handle_t out_hndls[RGZ_INPUT_MAXLAYERS]; /* OUTPUT */
+    int out_nhndls; /* OUTPUT */
+    int out_blits; /* OUTPUT */
+};
+
+struct rgz_out_svg {
+    int dispw;
+    int disph;
+    int htmlw;
+    int htmlh;
+};
+
+struct rgz_out_bvdirect {
+    struct bvbuffdesc *dstdesc;
+    struct bvsurfgeom *dstgeom;
+    int noblend;
+};
+
+typedef struct rgz_out_params {
+    int op; /* See RGZ_OUT_* */
+    union {
+        struct rgz_out_bvcmd bvc;
+        struct rgz_out_bvdirect bv;
+        struct rgz_out_svg svg;
+    } data;
+} rgz_out_params_t;
+
+/*
+ * Regionizer output commands
+ */
+
+/*
+ * Output SVG from regionizer
+ *
+ * rgz_out_params_t:
+ *
+ * op              RGZ_OUT_SVG
+ * data.svg.dispw
+ * data.svg.disph  Display width and height these values will be the
+ *                 viewport dimensions i.e. the logical coordinate space
+ *                 rather than the physical size
+ * data.svg.htmlw
+ * data.svg.htmlh  HTML output dimensions
+ */
+#define RGZ_OUT_SVG 0
+
+/*
+ * This commands generates bltsville command data structures for HWC which will
+ * paint layer by layer
+ *
+ * rgz_out_params_t:
+ *
+ * op                   RGZ_OUT_BVCMD_PAINT
+ * data.bvc.cmdp        Pointer to buffer with cmd data
+ * data.bvc.cmdlen      length of cmdp
+ * data.bvc.dstgeom     bltsville struct describing the destination geometry
+ * data.bvc.noblend     Test option to disable blending
+ * data.bvc.out_hndls   Array of buffer handles (OUTPUT)
+ * data.bvc.out_nhndls  Number of buffer handles (OUTPUT)
+ * data.bvc.out_blits   Number of blits (OUTPUT)
+ */
+#define RGZ_OUT_BVCMD_PAINT 1
+
+/*
+ * This commands generates bltsville command data structures for HWC which will
+ * render via regions. This will involve a complete redraw of the screen.
+ *
+ * See RGZ_OUT_BVCMD_PAINT
+ */
+#define RGZ_OUT_BVCMD_REGION 2
+
+/*
+ * Perform actual blits painting each layer from back to front - this is a test
+ * command
+ *
+ * rgz_out_params_t:
+ *
+ * op                  RGZ_OUT_BVDIRECT_PAINT
+ * data.bv.dstdesc     bltsville struct describing the destination buffer
+ * data.bv.dstgeom     bltsville struct describing the destination geometry
+ * data.bv.list        List of HWC layers to blit, only HWC_OVERLAY layers
+ *                     will be rendered
+ * data.bv.noblend     Test option to disable blending
+ */
+#define RGZ_OUT_BVDIRECT_PAINT 3
+/*
+ * Perform actual blits where each blit is a subregion - this is a test mode
+ */
+#define RGZ_OUT_BVDIRECT_REGION 5
+
+int rgz_out(rgz_t *rgz, rgz_out_params_t* params);
+
+/*
+ * Produce instrumented logging of layer data
+ */
+void rgz_profile_hwc(hwc_layer_list_t* list, int dispw, int disph);
+
+/*
+ * ----------------------------------
+ * IMPLEMENTATION DETAILS FOLLOW HERE
+ * ----------------------------------
+ */
+
+/*
+ * Regionizer blit data structures
+ */
+typedef struct blit_rect {
+    int left, top, right, bottom;
+} blit_rect_t;
+
+/*
+ * A hregion is a horizontal area generated from the intersection of layers
+ * for a given composition.
+ *
+ * ----------------------------------------
+ * | layer 0                              |
+ * |           xxxxxxxxxxxxxxxxxx         |
+ * |           x layer 1        x         |
+ * |           x                x         |
+ * |           x        xxxxxxxxxxxxxxxxxxx
+ * |           x        x layer 2         x
+ * |           x        x                 x
+ * |           xxxxxxxxxx                 x
+ * |                    x                 x
+ * |                    x                 x
+ * ---------------------xxxxxxxxxxxxxxxxxxx
+ *
+ * This can be broken up into a number of horizontal regions:
+ *
+ * ----------------------------------------
+ * | H1                                l0 |
+ * |-----------xxxxxxxxxxxxxxxxxx---------|
+ * | H2        x                x         |
+ * |        l0 x            l01 x      l0 |
+ * |-----------x--------xxxxxxxxxxxxxxxxxxx
+ * | H3        x        x       x         x
+ * |        l0 x    l01 x  l012 x     l02 x
+ * |-----------xxxxxxxxxxxxxxxxxx---------x
+ * | H4                 x                 x
+ * |                 l0 x             l02 x
+ * ---------------------xxxxxxxxxxxxxxxxxxx
+ *
+ * Each hregion is just an array of rectangles. By accounting for the layers
+ * at different z-order, and hregion becomes a multi-dimensional array e.g. in
+ * the diagram above H4 has 2 sub-regions, layer 0 intersects with the first
+ * region and layers 0 and 2 intersect with the second region.
+ */
+#define RGZ_SUBREGIONMAX ((RGZ_MAXLAYERS << 1) - 1)
+#define RGZ_MAX_BLITS (RGZ_SUBREGIONMAX * RGZ_SUBREGIONMAX)
+
+typedef struct rgz_layer {
+    hwc_layer_t *hwc_layer;
+    int buffidx;
+    int dirty_count;
+    void* dirty_hndl;
+} rgz_layer_t;
+
+typedef struct blit_hregion {
+    blit_rect_t rect;
+    rgz_layer_t *rgz_layers[RGZ_MAXLAYERS];
+    int nlayers;
+    int nsubregions;
+    blit_rect_t blitrects[RGZ_MAXLAYERS][RGZ_SUBREGIONMAX]; /* z-order | rectangle */
+} blit_hregion_t;
+
+enum { RGZ_STATE_INIT = 1, RGZ_REGION_DATA = 2} ;
+
+struct rgz {
+    /* All fields here are opaque to the caller */
+    blit_hregion_t *hregions;
+    int nhregions;
+    int state;
+    unsigned int rgz_layerno;
+    rgz_layer_t rgz_layers[RGZ_MAXLAYERS];
+};
+
+#endif /* __RGZ_2D__ */