Add the support of color BT709/SMPTE240M for color-space conversion on BDW

This is picked up from that on Haswell/Ivybridge.

Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>

5 files changed, 33 insertions, 103 deletions

diff --git a/src/i965_render.c b/src/i965_render.c
index 3c99614..8e14d87 100644
--- a/src/i965_render.c
+++ b/src/i965_render.c
@@ -2862,7 +2862,8 @@ gen8_render_color_calc_state(VADriverContextP ctx)
 
 static void
 gen8_render_upload_constants(VADriverContextP ctx,
-                             struct object_surface *obj_surface)
+                             struct object_surface *obj_surface,
+                             unsigned int flags)
 {
     struct i965_driver_data *i965 = i965_driver_data(ctx);
     struct i965_render_state *render_state = &i965->render_state;
@@ -2873,6 +2874,8 @@ gen8_render_upload_constants(VADriverContextP ctx,
     float brightness = (float)i965->brightness_attrib->value / 255; /* YUV is float in the shader */
     float hue = (float)i965->hue_attrib->value / 180 * PI;
     float saturation = (float)i965->saturation_attrib->value / DEFAULT_SATURATION;
+    float *yuv_to_rgb;
+    unsigned int color_flag;
 
     dri_bo_map(render_state->dynamic_state.bo, 1);
     assert(render_state->dynamic_state.bo->virtual);
@@ -2907,6 +2910,15 @@ gen8_render_upload_constants(VADriverContextP ctx,
     *color_balance_base++ = cos(hue) * contrast * saturation;
     *color_balance_base++ = sin(hue) * contrast * saturation;
 
+    color_flag = flags & VA_SRC_COLOR_MASK;
+    yuv_to_rgb = (float *)constant_buffer + 8;
+    if (color_flag == VA_SRC_BT709)
+        memcpy(yuv_to_rgb, yuv_to_rgb_bt709, sizeof(yuv_to_rgb_bt709));
+    else if (color_flag == VA_SRC_SMPTE_240)
+        memcpy(yuv_to_rgb, yuv_to_rgb_smpte_240, sizeof(yuv_to_rgb_smpte_240));
+    else
+        memcpy(yuv_to_rgb, yuv_to_rgb_bt601, sizeof(yuv_to_rgb_bt601));
+
     dri_bo_unmap(render_state->dynamic_state.bo);
 }
 
@@ -2925,7 +2937,7 @@ gen8_render_setup_states(
     gen8_render_cc_viewport(ctx);
     gen8_render_color_calc_state(ctx);
     gen8_render_blend_state(ctx);
-    gen8_render_upload_constants(ctx, obj_surface);
+    gen8_render_upload_constants(ctx, obj_surface, flags);
     i965_render_upload_vertex(ctx, obj_surface, src_rect, dst_rect);
 }
 
@@ -4024,7 +4036,7 @@ gen8_emit_wm_state(VADriverContextP ctx, int kernel)
 
     BEGIN_BATCH(batch, 11);
     OUT_BATCH(batch, GEN6_3DSTATE_CONSTANT_PS | (11 - 2));
-    OUT_BATCH(batch, 1);
+    OUT_BATCH(batch, URB_CS_ENTRY_SIZE);
     OUT_BATCH(batch, 0);
     /*DW3-4. Constant buffer 0 */
     OUT_BATCH(batch, render_state->curbe_offset);
diff --git a/src/shaders/render/exa_wm_src_affine.g8a b/src/shaders/render/exa_wm_src_affine.g8a
index 1d4efcc..7927c3b 100644
--- a/src/shaders/render/exa_wm_src_affine.g8a
+++ b/src/shaders/render/exa_wm_src_affine.g8a
@@ -35,8 +35,6 @@ define(`vh',    `g69')
 define(`bl',    `g2.0<8,8,1>F')
 define(`bh',    `g4.0<8,8,1>F')
 
-define(`a0_a_x',`g7.0<0,1,0>F')
-define(`a0_a_y',`g7.16<0,1,0>F')
 
 /* U */
 pln (8) ul<1>F a0_a_x bl { align1 }; /* pixel 0-7 */
diff --git a/src/shaders/render/exa_wm_src_affine.g8b b/src/shaders/render/exa_wm_src_affine.g8b
index 0273257..f5f9eca 100644
--- a/src/shaders/render/exa_wm_src_affine.g8b
+++ b/src/shaders/render/exa_wm_src_affine.g8b
@@ -1,4 +1,4 @@
-   { 0x0060005a, 0x28403ae8, 0x3a0000e0, 0x008d0040 },
-   { 0x0060005a, 0x28603ae8, 0x3a0000e0, 0x008d0080 },
-   { 0x0060005a, 0x28803ae8, 0x3a0000f0, 0x008d0040 },
-   { 0x0060005a, 0x28a03ae8, 0x3a0000f0, 0x008d0080 },
+   { 0x0060005a, 0x28403ae8, 0x3a000140, 0x008d0040 },
+   { 0x0060005a, 0x28603ae8, 0x3a000140, 0x008d0080 },
+   { 0x0060005a, 0x28803ae8, 0x3a000150, 0x008d0040 },
+   { 0x0060005a, 0x28a03ae8, 0x3a000150, 0x008d0080 },
diff --git a/src/shaders/render/exa_wm_yuv_rgb.g8a b/src/shaders/render/exa_wm_yuv_rgb.g8a
index 62669c8..9da53c8 100644
--- a/src/shaders/render/exa_wm_yuv_rgb.g8a
+++ b/src/shaders/render/exa_wm_yuv_rgb.g8a
@@ -28,79 +28,5 @@
  */
 
 include(`exa_wm.g4i')
-
-define(`YCbCr_base',	`src_sample_base')
-
-define(`Cr',		`src_sample_b')
-define(`Cr_01',		`src_sample_b_01')
-define(`Cr_23',		`src_sample_b_23')
-
-define(`Y',		`src_sample_r')
-define(`Y_01',		`src_sample_r_01')
-define(`Y_23',		`src_sample_r_23')
-
-define(`Cb',		`src_sample_g')
-define(`Cb_01',		`src_sample_g_01')
-define(`Cb_23',		`src_sample_g_23')
-
-define(`Crn',		`mask_sample_g')
-define(`Crn_01',	`mask_sample_g_01')
-define(`Crn_23',	`mask_sample_g_23')
-
-define(`Yn',		`mask_sample_r')
-define(`Yn_01',		`mask_sample_r_01')
-define(`Yn_23',		`mask_sample_r_23')
-
-define(`Cbn',		`mask_sample_b')
-define(`Cbn_01',	`mask_sample_b_01')
-define(`Cbn_23',	`mask_sample_b_23')
-
-    /* color space conversion function:
-     * R = Clamp ( 1.164(Y-16/255) + 1.596(Cr-128/255), 0, 1)
-     * G = Clamp ( 1.164(Y-16/255) - 0.813(Cr-128/255) - 0.392(Cb-128/255), 0, 1)
-     * B = Clamp ( 1.164(Y-16/255) + 2.017(Cb-128/255), 0, 1)
-     */
-
-    /* Normalize Y, Cb and Cr:
-     *
-     * Yn = (Y - 16/255) * 1.164
-     * Crn = Cr - 128 / 255
-     * Cbn = Cb - 128 / 255
-     */
-add (16)    Yn<1>F		Y<8;8,1>F	-0.0627451F { compr align1 };
-mul (16)    Yn<1>F		Yn<8;8,1>F	1.164F	    { compr align1 };
-
-add (16)    Crn<1>F		Cr<8;8,1>F	-0.501961F  { compr align1 };
-
-add (16)    Cbn<1>F		Cb<8;8,1>F	-0.501961F  { compr align1 };
-
-    /* 
-     * R = Y + Cr * 1.596
-     */
-mov (8)    acc0<1>F		Yn_01.0<8;8,1>F		    { align1 };
-mac.sat(8) src_sample_r_01<1>F	Crn_01<8;8,1>F	1.596F	    { align1 };
-mov (8)    acc0<1>F		Yn_23.0<8;8,1>F		    { align1 };
-mac.sat(8) src_sample_r_23<1>F	Crn_23<8;8,1>F	1.596F	    { align1 };
-
-    /*
-     * G = Crn * -0.813 + Cbn * -0.392 + Y
-     */
-mov (8)    acc0<1>F		Yn_01.0<8;8,1>F		    { align1 };
-mac (8)    acc0<1>F		Crn_01.0<8;8,1>F    	-0.813F	    { align1 };
-mac.sat(8) src_sample_g_01<1>F	Cbn_01.0<8;8,1>F    	-0.392F	    { align1 };
-mov (8)    acc0<1>F		Yn_23.0<8;8,1>F		    { align1 };
-mac (8)    acc0<1>F		Crn_23.0<8;8,1>F    	-0.813F	    { align1 };
-mac.sat(8) src_sample_g_23<1>F	Cbn_23.0<8;8,1>F    	-0.392F	    { align1 };
-
-    /*
-     * B = Cbn * 2.017 + Y
-     */
-mov (8)    acc0<1>F	        Yn_01.0<8;8,1>F		    { align1 };
-mac.sat(8) src_sample_b_01<1>F	Cbn_01.0<8;8,1>F     2.017F	    { align1 };
-
-mov (8)    acc0<1>F	        Yn_23.0<8;8,1>F		    { align1 };
-mac.sat(8) src_sample_b_23<1>F	Cbn_23.0<8;8,1>F     2.017F	    { align1 };
-    /*
-     * A = 1.0
-     */
-mov (16)    src_sample_a<1>F	1.0F        { compr align1 };
+include(`exa_yuv_gen6.g4i')
+include(`exa_yuv_rgb.gxa')
diff --git a/src/shaders/render/exa_wm_yuv_rgb.g8b b/src/shaders/render/exa_wm_yuv_rgb.g8b
index 8898a39..6b6b4d1 100644
--- a/src/shaders/render/exa_wm_yuv_rgb.g8b
+++ b/src/shaders/render/exa_wm_yuv_rgb.g8b
@@ -1,19 +1,13 @@
-   { 0x00800040, 0x22c03ae8, 0x3e8d01c0, 0xbd808081 },
-   { 0x00800041, 0x22c03ae8, 0x3e8d02c0, 0x3f94fdf4 },
-   { 0x00800040, 0x23003ae8, 0x3e8d0240, 0xbf008084 },
-   { 0x00800040, 0x23403ae8, 0x3e8d0200, 0xbf008084 },
-   { 0x00600001, 0x24003ae0, 0x008d02c0, 0x00000000 },
-   { 0x80600048, 0x21c03ae8, 0x3e8d0300, 0x3fcc49ba },
-   { 0x00600001, 0x24003ae0, 0x008d02e0, 0x00000000 },
-   { 0x80600048, 0x21e03ae8, 0x3e8d0320, 0x3fcc49ba },
-   { 0x00600001, 0x24003ae0, 0x008d02c0, 0x00000000 },
-   { 0x00600048, 0x24003ae0, 0x3e8d0300, 0xbf5020c5 },
-   { 0x80600048, 0x22003ae8, 0x3e8d0340, 0xbec8b439 },
-   { 0x00600001, 0x24003ae0, 0x008d02e0, 0x00000000 },
-   { 0x00600048, 0x24003ae0, 0x3e8d0320, 0xbf5020c5 },
-   { 0x80600048, 0x22203ae8, 0x3e8d0360, 0xbec8b439 },
-   { 0x00600001, 0x24003ae0, 0x008d02c0, 0x00000000 },
-   { 0x80600048, 0x22403ae8, 0x3e8d0340, 0x40011687 },
-   { 0x00600001, 0x24003ae0, 0x008d02e0, 0x00000000 },
-   { 0x80600048, 0x22603ae8, 0x3e8d0360, 0x40011687 },
+   { 0x00800040, 0x22c03ae8, 0x3a8d01c0, 0x000000ec },
+   { 0x00800040, 0x23003ae8, 0x3a8d0200, 0x000000fc },
+   { 0x00800040, 0x23403ae8, 0x3a8d0240, 0x0000010c },
+   { 0x00800041, 0x24003ae0, 0x3a8d02c0, 0x000000e0 },
+   { 0x00800048, 0x24003ae0, 0x3a8d0300, 0x000000e4 },
+   { 0x80800048, 0x21c03ae8, 0x3a8d0340, 0x000000e8 },
+   { 0x00800041, 0x24003ae0, 0x3a8d02c0, 0x000000f0 },
+   { 0x00800048, 0x24003ae0, 0x3a8d0300, 0x000000f4 },
+   { 0x80800048, 0x22003ae8, 0x3a8d0340, 0x000000f8 },
+   { 0x00800041, 0x24003ae0, 0x3a8d02c0, 0x00000100 },
+   { 0x00800048, 0x24003ae0, 0x3a8d0300, 0x00000104 },
+   { 0x80800048, 0x22403ae8, 0x3a8d0340, 0x00000108 },
    { 0x00800001, 0x22803ee8, 0x38000000, 0x3f800000 },