Add CSC conversion from NV12 to RGBX for VPP on Ivy
[profile/ivi/vaapi-intel-driver.git] / src / i965_post_processing.c
index ab0fbbf..a82c796 100755 (executable)
 #define SURFACE_STATE_OFFSET(index)             (SURFACE_STATE_PADDED_SIZE * index)
 #define BINDING_TABLE_OFFSET                    SURFACE_STATE_OFFSET(MAX_PP_SURFACES)
 
+#define GPU_ASM_BLOCK_WIDTH         16
+#define GPU_ASM_BLOCK_HEIGHT        8
+#define GPU_ASM_X_OFFSET_ALIGNMENT  4
+
 static const uint32_t pp_null_gen5[][4] = {
 #include "shaders/post_processing/gen5_6/null.g4b.gen5"
 };
@@ -631,6 +635,7 @@ static const uint32_t pp_pa_load_save_pl3_gen7[][4] = {
 static const uint32_t pp_rgbx_load_save_nv12_gen7[][4] = {
 };
 static const uint32_t pp_nv12_load_save_rgbx_gen7[][4] = {
+#include "shaders/post_processing/gen7/pl2_to_rgbx.g7b"
 };
 
 static VAStatus gen7_pp_plx_avs_initialize(VADriverContextP ctx, struct i965_post_processing_context *pp_context,
@@ -829,6 +834,239 @@ static struct pp_module pp_modules_gen7[] = {
             NULL,
         },
     
+        gen7_pp_plx_avs_initialize,
+    },
+            
+};
+
+static const uint32_t pp_null_gen75[][4] = {
+};
+
+static const uint32_t pp_nv12_load_save_nv12_gen75[][4] = {
+#include "shaders/post_processing/gen7/pl2_to_pl2.g75b"
+};
+
+static const uint32_t pp_nv12_load_save_pl3_gen75[][4] = {
+#include "shaders/post_processing/gen7/pl2_to_pl3.g75b"
+};
+
+static const uint32_t pp_pl3_load_save_nv12_gen75[][4] = {
+#include "shaders/post_processing/gen7/pl3_to_pl2.g75b"
+};
+
+static const uint32_t pp_pl3_load_save_pl3_gen75[][4] = {
+#include "shaders/post_processing/gen7/pl3_to_pl3.g75b"
+};
+
+static const uint32_t pp_nv12_scaling_gen75[][4] = {
+#include "shaders/post_processing/gen7/avs.g75b"
+};
+
+static const uint32_t pp_nv12_avs_gen75[][4] = {
+#include "shaders/post_processing/gen7/avs.g75b"
+};
+
+static const uint32_t pp_nv12_dndi_gen75[][4] = {
+// #include "shaders/post_processing/gen7/dndi.g75b"
+};
+
+static const uint32_t pp_nv12_dn_gen75[][4] = {
+// #include "shaders/post_processing/gen7/nv12_dn_nv12.g75b"
+};
+static const uint32_t pp_nv12_load_save_pa_gen75[][4] = {
+#include "shaders/post_processing/gen7/pl2_to_pa.g75b"
+};
+static const uint32_t pp_pl3_load_save_pa_gen75[][4] = {
+#include "shaders/post_processing/gen7/pl3_to_pa.g75b"
+};
+static const uint32_t pp_pa_load_save_nv12_gen75[][4] = {
+#include "shaders/post_processing/gen7/pa_to_pl2.g75b"
+};
+static const uint32_t pp_pa_load_save_pl3_gen75[][4] = {
+#include "shaders/post_processing/gen7/pa_to_pl3.g75b"
+};
+static const uint32_t pp_rgbx_load_save_nv12_gen75[][4] = {
+};
+static const uint32_t pp_nv12_load_save_rgbx_gen75[][4] = {
+};
+
+static struct pp_module pp_modules_gen75[] = {
+    {
+        {
+            "NULL module (for testing)",
+            PP_NULL,
+            pp_null_gen75,
+            sizeof(pp_null_gen75),
+            NULL,
+        },
+
+        pp_null_initialize,
+    },
+
+    {
+        {
+            "NV12_NV12",
+            PP_NV12_LOAD_SAVE_N12,
+            pp_nv12_load_save_nv12_gen75,
+            sizeof(pp_nv12_load_save_nv12_gen75),
+            NULL,
+        },
+
+        gen7_pp_plx_avs_initialize,
+    },
+
+    {
+        {
+            "NV12_PL3",
+            PP_NV12_LOAD_SAVE_PL3,
+            pp_nv12_load_save_pl3_gen75,
+            sizeof(pp_nv12_load_save_pl3_gen75),
+            NULL,
+        },
+        
+        gen7_pp_plx_avs_initialize,
+    },
+
+    {
+        {
+            "PL3_NV12",
+            PP_PL3_LOAD_SAVE_N12,
+            pp_pl3_load_save_nv12_gen75,
+            sizeof(pp_pl3_load_save_nv12_gen75),
+            NULL,
+        },
+
+        gen7_pp_plx_avs_initialize,
+    },
+
+    {
+        {
+            "PL3_PL3",
+            PP_PL3_LOAD_SAVE_N12,
+            pp_pl3_load_save_pl3_gen75,
+            sizeof(pp_pl3_load_save_pl3_gen75),
+            NULL,
+        },
+
+        gen7_pp_plx_avs_initialize,
+    },
+
+    {
+        {
+            "NV12 Scaling module",
+            PP_NV12_SCALING,
+            pp_nv12_scaling_gen75,
+            sizeof(pp_nv12_scaling_gen75),
+            NULL,
+        },
+
+        gen7_pp_plx_avs_initialize,
+    },
+
+    {
+        {
+            "NV12 AVS module",
+            PP_NV12_AVS,
+            pp_nv12_avs_gen75,
+            sizeof(pp_nv12_avs_gen75),
+            NULL,
+        },
+
+        gen7_pp_plx_avs_initialize,
+    },
+
+    {
+        {
+            "NV12 DNDI module",
+            PP_NV12_DNDI,
+            pp_nv12_dndi_gen75,
+            sizeof(pp_nv12_dndi_gen75),
+            NULL,
+        },
+
+        gen7_pp_nv12_dndi_initialize,
+    },
+
+    {
+        {
+            "NV12 DN module",
+            PP_NV12_DN,
+            pp_nv12_dn_gen75,
+            sizeof(pp_nv12_dn_gen75),
+            NULL,
+        },
+
+        gen7_pp_nv12_dn_initialize,
+    },
+    {
+        {
+            "NV12_PA module",
+            PP_NV12_LOAD_SAVE_PA,
+            pp_nv12_load_save_pa_gen75,
+            sizeof(pp_nv12_load_save_pa_gen75),
+            NULL,
+        },
+    
+        gen7_pp_plx_avs_initialize,
+    },
+
+    {
+        {
+            "PL3_PA module",
+            PP_PL3_LOAD_SAVE_PA,
+            pp_pl3_load_save_pa_gen75,
+            sizeof(pp_pl3_load_save_pa_gen75),
+            NULL,
+        },
+    
+        gen7_pp_plx_avs_initialize,
+    },
+
+    {
+        {
+            "PA_NV12 module",
+            PP_PA_LOAD_SAVE_NV12,
+            pp_pa_load_save_nv12_gen75,
+            sizeof(pp_pa_load_save_nv12_gen75),
+            NULL,
+        },
+    
+        gen7_pp_plx_avs_initialize,
+    },
+
+    {
+        {
+            "PA_PL3 module",
+            PP_PA_LOAD_SAVE_PL3,
+            pp_pa_load_save_pl3_gen75,
+            sizeof(pp_pa_load_save_pl3_gen75),
+            NULL,
+        },
+    
+        gen7_pp_plx_avs_initialize,
+    },
+    
+    {
+        {
+            "RGBX_NV12 module",
+            PP_RGBX_LOAD_SAVE_NV12,
+            pp_rgbx_load_save_nv12_gen75,
+            sizeof(pp_rgbx_load_save_nv12_gen75),
+            NULL,
+        },
+    
+        pp_plx_load_save_plx_initialize,
+    },
+
+    {
+        {
+            "NV12_RGBX module",
+            PP_NV12_LOAD_SAVE_RGBX,
+            pp_nv12_load_save_rgbx_gen75,
+            sizeof(pp_nv12_load_save_rgbx_gen75),
+            NULL,
+        },
+    
         pp_plx_load_save_plx_initialize,
     },
             
@@ -1294,6 +1532,7 @@ gen7_pp_set_surface_state(VADriverContextP ctx, struct i965_post_processing_cont
                           int width, int height, int pitch, int format, 
                           int index, int is_target)
 {
+    struct i965_driver_data * const i965 = i965_driver_data(ctx);  
     struct gen7_surface_state *ss;
     dri_bo *ss_bo;
     unsigned int tiling;
@@ -1314,6 +1553,8 @@ gen7_pp_set_surface_state(VADriverContextP ctx, struct i965_post_processing_cont
     ss->ss2.height = height - 1;
     ss->ss3.pitch = pitch - 1;
     gen7_pp_set_surface_tiling(ss, tiling);
+    if (IS_HASWELL(i965->intel.device_id))
+        gen7_render_set_surface_scs(ss);
     dri_bo_emit_reloc(ss_bo,
                       I915_GEM_DOMAIN_RENDER, is_target ? I915_GEM_DOMAIN_RENDER : 0,
                       surf_bo_offset,
@@ -1494,6 +1735,10 @@ gen7_pp_set_media_rw_message_surface(VADriverContextP ctx, struct i965_post_proc
                    fourcc == VA_FOURCC('I', 'M', 'C', '1')) ? 1 : 2;
     int interleaved_uv = fourcc == VA_FOURCC('N', 'V', '1', '2');
     int packed_yuv = (fourcc == VA_FOURCC('Y', 'U', 'Y', '2') || fourcc == VA_FOURCC('U', 'Y', 'V', 'Y'));
+    int rgbx_format = (fourcc == VA_FOURCC('R', 'G', 'B', 'A') || 
+                              fourcc == VA_FOURCC('R', 'G', 'B', 'X') || 
+                              fourcc == VA_FOURCC('B', 'G', 'R', 'A') || 
+                              fourcc == VA_FOURCC('B', 'G', 'R', 'X'));
 
     if (surface->type == I965_SURFACE_TYPE_SURFACE) {
         obj_surface = SURFACE(surface->id);
@@ -1510,7 +1755,10 @@ gen7_pp_set_media_rw_message_surface(VADriverContextP ctx, struct i965_post_proc
                 width[0] = obj_surface->orig_width;     /* surface foramt is YCBCR, width is specified in units of pixels */
 
             pitch[0] = obj_surface->width * 2;
-        }
+        } else if (rgbx_format) {
+           if (is_target)
+                width[0] = obj_surface->orig_width * 4; /* surface format is R8, so quad the width */
+       }
 
         width[1] = obj_surface->cb_cr_width;
         height[1] = obj_surface->cb_cr_height;
@@ -1529,7 +1777,10 @@ gen7_pp_set_media_rw_message_surface(VADriverContextP ctx, struct i965_post_proc
         pitch[0] = obj_image->image.pitches[0];
         offset[0] = obj_image->image.offsets[0];
 
-        if (packed_yuv) {
+       if (rgbx_format) {
+           if (is_target)
+               width[0] = obj_image->image.width * 4; /* surface format is R8, so quad the width */
+       } else if (packed_yuv) {
             if (is_target)
                 width[0] = obj_image->image.width * 2;  /* surface format is R8, so double the width */
             else
@@ -1555,10 +1806,19 @@ gen7_pp_set_media_rw_message_surface(VADriverContextP ctx, struct i965_post_proc
         gen7_pp_set_surface_state(ctx, pp_context,
                                   bo, 0,
                                   width[0] / 4, height[0], pitch[0],
-                                  I965_SURFACEFORMAT_R8_SINT,
+                                  I965_SURFACEFORMAT_R8_UINT,
                                   base_index, 1);
-
-        if (!packed_yuv) {
+       if (rgbx_format) {
+               struct gen7_pp_static_parameter *pp_static_parameter = pp_context->pp_static_parameter;
+               /* the format is MSB: X-B-G-R */
+               pp_static_parameter->grf2.save_avs_rgb_swap = 0;
+               if ((fourcc == VA_FOURCC('B', 'G', 'R', 'A')) || 
+                              (fourcc == VA_FOURCC('B', 'G', 'R', 'X'))) {
+                       /* It is stored as MSB: X-R-G-B */
+                       pp_static_parameter->grf2.save_avs_rgb_swap = 1;
+               }
+       }
+        if (!packed_yuv && !rgbx_format) {
             if (interleaved_uv) {
                 gen7_pp_set_surface_state(ctx, pp_context,
                                           bo, offset[1],
@@ -1681,15 +1941,47 @@ static int
 pp_load_save_set_block_parameter(struct i965_post_processing_context *pp_context, int x, int y)
 {
     struct pp_inline_parameter *pp_inline_parameter = pp_context->pp_inline_parameter;
+    struct pp_load_save_context *pp_load_save_context = (struct pp_load_save_context *)&pp_context->private_context;
 
-    pp_inline_parameter->grf5.block_vertical_mask = 0xff;
-    pp_inline_parameter->grf5.block_horizontal_mask = 0xffff;
-    pp_inline_parameter->grf5.destination_block_horizontal_origin = x * 16;
-    pp_inline_parameter->grf5.destination_block_vertical_origin = y * 8;
+    pp_inline_parameter->grf5.destination_block_horizontal_origin = x * 16 + pp_load_save_context->dest_x;
+    pp_inline_parameter->grf5.destination_block_vertical_origin = y * 8 + pp_load_save_context->dest_y;
 
     return 0;
 }
 
+static void calculate_boundary_block_mask(struct i965_post_processing_context *pp_context, const VARectangle *dst_rect)
+{
+    int i;
+    /* x offset of dest surface must be dword aligned.
+     * so we have to extend dst surface on left edge, and mask out pixels not interested
+     */
+    if (dst_rect->x%GPU_ASM_X_OFFSET_ALIGNMENT) {
+        pp_context->block_horizontal_mask_left = 0;
+        for (i=dst_rect->x%GPU_ASM_X_OFFSET_ALIGNMENT; i<GPU_ASM_BLOCK_WIDTH; i++)
+        {
+            pp_context->block_horizontal_mask_left |= 1<<i;
+        }
+    }
+    else {
+        pp_context->block_horizontal_mask_left = 0xffff;
+    }
+    
+    int dst_width_adjust = dst_rect->width + dst_rect->x%GPU_ASM_X_OFFSET_ALIGNMENT; 
+    if (dst_width_adjust%GPU_ASM_BLOCK_WIDTH){
+        pp_context->block_horizontal_mask_right = (1 << (dst_width_adjust%GPU_ASM_BLOCK_WIDTH)) - 1;
+    }
+    else {
+        pp_context->block_horizontal_mask_right = 0xffff;
+    }
+    
+    if (dst_rect->height%GPU_ASM_BLOCK_HEIGHT){
+        pp_context->block_vertical_mask_bottom = (1 << (dst_rect->height%GPU_ASM_BLOCK_HEIGHT)) - 1;
+    }
+    else {
+        pp_context->block_vertical_mask_bottom = 0xff;
+    }
+
+}
 static VAStatus
 pp_plx_load_save_plx_initialize(VADriverContextP ctx, struct i965_post_processing_context *pp_context,
                                 const struct i965_surface *src_surface,
@@ -1716,11 +2008,15 @@ pp_plx_load_save_plx_initialize(VADriverContextP ctx, struct i965_post_processin
     pp_context->pp_x_steps = pp_load_save_x_steps;
     pp_context->pp_y_steps = pp_load_save_y_steps;
     pp_context->pp_set_block_parameter = pp_load_save_set_block_parameter;
-    pp_load_save_context->dest_h = ALIGN(height[Y], 16);
-    pp_load_save_context->dest_w = ALIGN(width[Y], 16);
 
-    pp_inline_parameter->grf5.block_count_x = ALIGN(width[Y], 16) / 16;   /* 1 x N */
-    pp_inline_parameter->grf5.number_blocks = ALIGN(width[Y], 16) / 16;
+    int dst_left_edge_extend = dst_rect->x%GPU_ASM_X_OFFSET_ALIGNMENT;;
+    pp_load_save_context->dest_x = dst_rect->x - dst_left_edge_extend;
+    pp_load_save_context->dest_y = dst_rect->y;
+    pp_load_save_context->dest_h = ALIGN(dst_rect->height, 8);
+    pp_load_save_context->dest_w = ALIGN(dst_rect->width+dst_left_edge_extend, 16);
+
+    pp_inline_parameter->grf5.block_count_x = pp_load_save_context->dest_w / 16;   /* 1 x N */
+    pp_inline_parameter->grf5.number_blocks = pp_load_save_context->dest_w / 16;
 
     pp_static_parameter->grf3.horizontal_origin_offset = src_rect->x;
     pp_static_parameter->grf3.vertical_origin_offset = src_rect->y;
@@ -1846,20 +2142,20 @@ pp_nv12_scaling_initialize(VADriverContextP ctx, struct i965_post_processing_con
     pp_context->pp_y_steps = pp_scaling_y_steps;
     pp_context->pp_set_block_parameter = pp_scaling_set_block_parameter;
 
-    pp_scaling_context->dest_x = dst_rect->x;
+    int dst_left_edge_extend = dst_rect->x%GPU_ASM_X_OFFSET_ALIGNMENT;
+    float src_left_edge_extend = (float)dst_left_edge_extend*src_rect->width/dst_rect->width;
+    pp_scaling_context->dest_x = dst_rect->x - dst_left_edge_extend;
     pp_scaling_context->dest_y = dst_rect->y;
-    pp_scaling_context->dest_w = ALIGN(dst_rect->width, 16);
-    pp_scaling_context->dest_h = ALIGN(dst_rect->height, 16);
-    pp_scaling_context->src_normalized_x = (float)src_rect->x / in_w;
+    pp_scaling_context->dest_w = ALIGN(dst_rect->width + dst_left_edge_extend, 16);
+    pp_scaling_context->dest_h = ALIGN(dst_rect->height, 8);
+    pp_scaling_context->src_normalized_x = (float)(src_rect->x - src_left_edge_extend)/ in_w;
     pp_scaling_context->src_normalized_y = (float)src_rect->y / in_h;
 
     pp_static_parameter->grf1.r1_6.normalized_video_y_scaling_step = (float) src_rect->height / in_h / dst_rect->height;
 
-    pp_inline_parameter->grf5.normalized_video_x_scaling_step = (float) src_rect->width / in_w / dst_rect->width;
+    pp_inline_parameter->grf5.normalized_video_x_scaling_step = (float) (src_rect->width + src_left_edge_extend)/ in_w / (dst_rect->width + dst_left_edge_extend);
     pp_inline_parameter->grf5.block_count_x = pp_scaling_context->dest_w / 16;   /* 1 x N */
     pp_inline_parameter->grf5.number_blocks = pp_scaling_context->dest_w / 16;
-    pp_inline_parameter->grf5.block_vertical_mask = 0xff;
-    pp_inline_parameter->grf5.block_horizontal_mask = 0xffff;
 
     dst_surface->flags = src_surface->flags;
 
@@ -2236,23 +2532,23 @@ pp_nv12_avs_initialize(VADriverContextP ctx, struct i965_post_processing_context
     pp_context->pp_y_steps = pp_avs_y_steps;
     pp_context->pp_set_block_parameter = pp_avs_set_block_parameter;
 
-    pp_avs_context->dest_x = dst_rect->x;
+    int dst_left_edge_extend = dst_rect->x%GPU_ASM_X_OFFSET_ALIGNMENT;
+    float src_left_edge_extend = (float)dst_left_edge_extend*src_rect->width/dst_rect->width;
+    pp_avs_context->dest_x = dst_rect->x - dst_left_edge_extend;
     pp_avs_context->dest_y = dst_rect->y;
-    pp_avs_context->dest_w = ALIGN(dst_rect->width, 16);
-    pp_avs_context->dest_h = ALIGN(dst_rect->height, 16);
-    pp_avs_context->src_normalized_x = (float)src_rect->x / in_w;
+    pp_avs_context->dest_w = ALIGN(dst_rect->width + dst_left_edge_extend, 16);
+    pp_avs_context->dest_h = ALIGN(dst_rect->height, 8);
+    pp_avs_context->src_normalized_x = (float)(src_rect->x - src_left_edge_extend)/ in_w;
     pp_avs_context->src_normalized_y = (float)src_rect->y / in_h;
-    pp_avs_context->src_w = src_rect->width;
+    pp_avs_context->src_w = src_rect->width + src_left_edge_extend;
     pp_avs_context->src_h = src_rect->height;
 
     pp_static_parameter->grf4.r4_2.avs.nlas = nlas;
     pp_static_parameter->grf1.r1_6.normalized_video_y_scaling_step = (float) src_rect->height / in_h / dst_rect->height;
 
-    pp_inline_parameter->grf5.normalized_video_x_scaling_step = (float) src_rect->width / in_w / dst_rect->width;
+    pp_inline_parameter->grf5.normalized_video_x_scaling_step = (float) (src_rect->width + src_left_edge_extend)/ in_w / (dst_rect->width + dst_left_edge_extend);
     pp_inline_parameter->grf5.block_count_x = 1;        /* M x 1 */
     pp_inline_parameter->grf5.number_blocks = pp_avs_context->dest_h / 8;
-    pp_inline_parameter->grf5.block_vertical_mask = 0xff;
-    pp_inline_parameter->grf5.block_horizontal_mask = 0xffff;
     pp_inline_parameter->grf6.video_step_delta = 0.0;
 
     dst_surface->flags = src_surface->flags;
@@ -2356,10 +2652,13 @@ gen7_pp_plx_avs_initialize(VADriverContextP ctx, struct i965_post_processing_con
     struct i965_sampler_8x8_state *sampler_8x8_state;
     int index, i;
     int width[3], height[3], pitch[3], offset[3];
+    int src_width, src_height;
 
     /* source surface */
     gen7_pp_set_media_rw_message_surface(ctx, pp_context, src_surface, 0, 0,
                                          width, height, pitch, offset);
+    src_width = width[0];
+    src_height = height[0];
 
     /* destination surface */
     gen7_pp_set_media_rw_message_surface(ctx, pp_context, dst_surface, 24, 1,
@@ -2528,7 +2827,7 @@ gen7_pp_plx_avs_initialize(VADriverContextP ctx, struct i965_post_processing_con
     pp_static_parameter->grf2.avs_wa_five_div_256_width = (float) 5.0 / (256 * dw);
 
     pp_static_parameter->grf3.sampler_load_horizontal_scaling_step_ratio = (float) pp_avs_context->src_w / dw;
-    pp_static_parameter->grf4.sampler_load_vertical_scaling_step = (float) 1.0 / pp_avs_context->dest_h;
+    pp_static_parameter->grf4.sampler_load_vertical_scaling_step = (float) src_rect->height / src_height / pp_avs_context->dest_h;
     pp_static_parameter->grf5.sampler_load_vertical_frame_origin = -(float)pp_avs_context->dest_y / pp_avs_context->dest_h;
     pp_static_parameter->grf6.sampler_load_horizontal_frame_origin = -(float)pp_avs_context->dest_x / dw;
 
@@ -3624,6 +3923,8 @@ gen6_pp_initialize(
     else
         va_status = VA_STATUS_ERROR_UNIMPLEMENTED;
 
+    calculate_boundary_block_mask(pp_context, dst_rect);
+    
     return va_status;
 }
 
@@ -3795,6 +4096,46 @@ gen6_interface_descriptor_load(VADriverContextP ctx,
     ADVANCE_BATCH(batch);
 }
 
+static void update_block_mask_parameter(struct i965_post_processing_context *pp_context, int x, int y, int x_steps, int y_steps) 
+{
+    struct pp_inline_parameter *pp_inline_parameter = pp_context->pp_inline_parameter;
+
+    pp_inline_parameter->grf5.block_vertical_mask = 0xff;
+    pp_inline_parameter->grf6.block_vertical_mask_bottom = pp_context->block_vertical_mask_bottom;
+    // for the first block, it always on the left edge. the second block will reload horizontal_mask from grf6.block_horizontal_mask_middle
+    pp_inline_parameter->grf5.block_horizontal_mask = pp_context->block_horizontal_mask_left;
+    pp_inline_parameter->grf6.block_horizontal_mask_middle = 0xffff;
+    pp_inline_parameter->grf6.block_horizontal_mask_right = pp_context->block_horizontal_mask_right;
+
+    /* 1 x N */
+    if (x_steps == 1) {
+        if (y == y_steps-1) {
+            pp_inline_parameter->grf5.block_vertical_mask = pp_context->block_vertical_mask_bottom;
+        }
+        else {
+            pp_inline_parameter->grf6.block_vertical_mask_bottom = 0xff;
+        }
+    }
+
+    /* M x 1 */
+    if (y_steps == 1) {
+        if (x == 0) { // all blocks in this group are on the left edge
+            pp_inline_parameter->grf6.block_horizontal_mask_middle = pp_context->block_horizontal_mask_left;
+            pp_inline_parameter->grf6.block_horizontal_mask_right = pp_context->block_horizontal_mask_left; 
+        }
+        else if (x == x_steps-1) {
+            pp_inline_parameter->grf5.block_horizontal_mask = pp_context->block_horizontal_mask_right;
+            pp_inline_parameter->grf6.block_horizontal_mask_middle = pp_context->block_horizontal_mask_right;
+        }
+        else {
+            pp_inline_parameter->grf5.block_horizontal_mask = 0xffff;
+            pp_inline_parameter->grf6.block_horizontal_mask_middle = 0xffff;
+            pp_inline_parameter->grf6.block_horizontal_mask_right = 0xffff;
+        }
+    }
+
+}
+
 static void
 gen6_pp_object_walker(VADriverContextP ctx,
                       struct i965_post_processing_context *pp_context)
@@ -3825,6 +4166,10 @@ gen6_pp_object_walker(VADriverContextP ctx,
     for (y = 0; y < y_steps; y++) {
         for (x = 0; x < x_steps; x++) {
             if (!pp_context->pp_set_block_parameter(pp_context, x, y)) {
+                // some common block parameter update goes here, apply to all pp functions
+                if (IS_GEN6(i965->intel.device_id))
+                    update_block_mask_parameter (pp_context, x, y, x_steps, y_steps);
+                
                 *command_ptr++ = (CMD_MEDIA_OBJECT | (command_length_in_dws - 2));
                 *command_ptr++ = 0;
                 *command_ptr++ = 0;
@@ -4040,6 +4385,50 @@ i965_vpp_clear_surface(VADriverContextP ctx,
     intel_batchbuffer_end_atomic(batch);
 }
 
+VAStatus
+i965_scaling_processing(
+    VADriverContextP   ctx,
+    VASurfaceID        src_surface_id,
+    const VARectangle *src_rect,
+    VASurfaceID        dst_surface_id,
+    const VARectangle *dst_rect,
+    unsigned int       flags)
+{
+    VAStatus va_status = VA_STATUS_SUCCESS;
+    struct i965_driver_data *i965 = i965_driver_data(ctx);
+    struct object_surface *src_surface_obj = SURFACE(src_surface_id);
+    struct object_surface *dst_surface_obj = SURFACE(dst_surface_id);
+    assert(src_surface_obj->fourcc == VA_FOURCC('N', 'V', '1', '2'));
+    assert(dst_surface_obj->fourcc == VA_FOURCC('N', 'V', '1', '2'));
+
+    if (HAS_PP(i965) && (flags & I965_PP_FLAG_AVS)) {
+        struct i965_surface src_surface;
+        struct i965_surface dst_surface;
+
+         _i965LockMutex(&i965->pp_mutex);
+
+         src_surface.id = src_surface_id;
+         src_surface.type = I965_SURFACE_TYPE_SURFACE;
+         src_surface.flags = I965_SURFACE_FLAG_FRAME;
+         dst_surface.id = dst_surface_id;
+         dst_surface.type = I965_SURFACE_TYPE_SURFACE;
+         dst_surface.flags = I965_SURFACE_FLAG_FRAME;
+
+         va_status = i965_post_processing_internal(ctx, i965->pp_context,
+                                                   &src_surface,
+                                                   src_rect,
+                                                   &dst_surface,
+                                                   dst_rect,
+                                                   PP_NV12_AVS,
+                                                   NULL);
+
+         _i965UnlockMutex(&i965->pp_mutex);
+    }
+
+    return va_status;
+}
+
 VASurfaceID
 i965_post_processing(
     VADriverContextP   ctx,
@@ -4470,8 +4859,11 @@ i965_post_processing_context_init(VADriverContextP ctx,
     assert(NUM_PP_MODULES == ARRAY_ELEMS(pp_modules_gen5));
     assert(NUM_PP_MODULES == ARRAY_ELEMS(pp_modules_gen6));
     assert(NUM_PP_MODULES == ARRAY_ELEMS(pp_modules_gen7));
+    assert(NUM_PP_MODULES == ARRAY_ELEMS(pp_modules_gen75));
 
-    if (IS_GEN7(i965->intel.device_id))
+    if (IS_HASWELL(i965->intel.device_id))
+        memcpy(pp_context->pp_modules, pp_modules_gen75, sizeof(pp_context->pp_modules));
+    else if (IS_GEN7(i965->intel.device_id))
         memcpy(pp_context->pp_modules, pp_modules_gen7, sizeof(pp_context->pp_modules));
     else if (IS_GEN6(i965->intel.device_id))
         memcpy(pp_context->pp_modules, pp_modules_gen6, sizeof(pp_context->pp_modules));
@@ -4539,7 +4931,7 @@ static const int proc_frame_to_pp_frame[3] = {
     I965_SURFACE_FLAG_BOTTOME_FIELD_FIRST
 };
 
-static void 
+void 
 i965_proc_picture(VADriverContextP ctx, 
                   VAProfile profile, 
                   union codec_state *codec_state,
@@ -4571,9 +4963,8 @@ i965_proc_picture(VADriverContextP ctx,
     src_surface.type = I965_SURFACE_TYPE_SURFACE;
     src_surface.flags = proc_frame_to_pp_frame[pipeline_param->filter_flags & 0x3];
 
+    VASurfaceID out_surface_id = VA_INVALID_ID;
     if (obj_surface->fourcc != VA_FOURCC('N', 'V', '1', '2')) {
-        VASurfaceID out_surface_id = VA_INVALID_ID;
-
         src_surface.id = pipeline_param->surface;
         src_surface.type = I965_SURFACE_TYPE_SURFACE;
         src_surface.flags = I965_SURFACE_FLAG_FRAME;
@@ -4637,15 +5028,11 @@ i965_proc_picture(VADriverContextP ctx,
         dst_rect.height = in_height;
     }
 
-    obj_surface = SURFACE(proc_state->current_render_target);
-    i965_check_alloc_surface_bo(ctx, obj_surface, !!tiling, VA_FOURCC('N','V','1','2'), SUBSAMPLE_YUV420);
-    i965_vpp_clear_surface(ctx, &proc_context->pp_context, proc_state->current_render_target, pipeline_param->output_background_color); 
-    
     for (i = 0; i < pipeline_param->num_filters; i++) {
         struct object_buffer *obj_buffer = BUFFER(pipeline_param->filters[i]);
         VAProcFilterParameterBufferBase *filter_param = (VAProcFilterParameterBufferBase *)obj_buffer->buffer_store->buffer;
         VAProcFilterType filter_type = filter_param->type;
-        VASurfaceID out_surface_id = VA_INVALID_ID;
+        out_surface_id = VA_INVALID_ID;
         int kernel_index = procfilter_to_pp_flag[filter_type];
 
         if (kernel_index != PP_NULL &&
@@ -4678,11 +5065,35 @@ i965_proc_picture(VADriverContextP ctx,
         }
     }
 
-    dst_surface.id = proc_state->current_render_target;
+    obj_surface = SURFACE(proc_state->current_render_target);
+    int csc_needed = 0;
+    if (obj_surface->fourcc && obj_surface->fourcc !=  VA_FOURCC('N','V','1','2')){
+        csc_needed = 1;
+        out_surface_id = VA_INVALID_ID;
+        status = i965_CreateSurfaces(ctx,
+                                     obj_surface->orig_width,
+                                     obj_surface->orig_height,
+                                     VA_RT_FORMAT_YUV420, 
+                                     1,
+                                     &out_surface_id);
+        assert(status == VA_STATUS_SUCCESS);
+        tmp_surfaces[num_tmp_surfaces++] = out_surface_id;
+        struct object_surface *csc_surface = SURFACE(out_surface_id);
+        i965_check_alloc_surface_bo(ctx, csc_surface, !!tiling, VA_FOURCC('N','V','1','2'), SUBSAMPLE_YUV420);
+        dst_surface.id = out_surface_id;
+    } else {
+        i965_check_alloc_surface_bo(ctx, obj_surface, !!tiling, VA_FOURCC('N','V','1','2'), SUBSAMPLE_YUV420);
+        dst_surface.id = proc_state->current_render_target;
+    }
+
     dst_surface.type = I965_SURFACE_TYPE_SURFACE;
+    i965_vpp_clear_surface(ctx, &proc_context->pp_context, proc_state->current_render_target, pipeline_param->output_background_color); 
 
+    // load/save doesn't support different origin offset for src and dst surface
     if (src_rect.width == dst_rect.width &&
-        src_rect.height == dst_rect.height) {
+        src_rect.height == dst_rect.height &&
+        src_rect.x == dst_rect.x &&
+        src_rect.y == dst_rect.y) {
         i965_post_processing_internal(ctx, &proc_context->pp_context,
                                       &src_surface,
                                       &src_rect,
@@ -4702,6 +5113,15 @@ i965_proc_picture(VADriverContextP ctx,
                                       NULL);
     }
 
+    if (csc_needed) {
+        src_surface.id = dst_surface.id;
+        src_surface.type = dst_surface.type;
+        src_surface.flags = dst_surface.flags;
+        dst_surface.id = proc_state->current_render_target;
+        dst_surface.type = I965_SURFACE_TYPE_SURFACE;
+        i965_image_processing(ctx, &src_surface, &dst_rect, &dst_surface, &dst_rect);
+    }
+    
     if (num_tmp_surfaces)
         i965_DestroySurfaces(ctx,
                              tmp_surfaces,
@@ -4728,7 +5148,7 @@ i965_proc_context_init(VADriverContextP ctx, struct object_config *obj_config)
 
     proc_context->base.destroy = i965_proc_context_destroy;
     proc_context->base.run = i965_proc_picture;
-    proc_context->base.batch = intel_batchbuffer_new(intel, I915_EXEC_RENDER);
+    proc_context->base.batch = intel_batchbuffer_new(intel, I915_EXEC_RENDER, 0);
     i965_post_processing_context_init(ctx, &proc_context->pp_context, proc_context->base.batch);
 
     return (struct hw_context *)proc_context;