Configure the cost-center of MPEG2 VME shader on haswell

This is derived from the neighbour macroblock based on MPEG2 spec.

Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
(cherry picked from commit a51860aba2d2713a9c8c817c430ffa93abd5145a)

3 files changed, 18 insertions, 11 deletions

diff --git a/src/shaders/vme/mpeg2_inter_haswell.asm b/src/shaders/vme/mpeg2_inter_haswell.asm
index bd41a90..cc9efa4 100644
--- a/src/shaders/vme/mpeg2_inter_haswell.asm
+++ b/src/shaders/vme/mpeg2_inter_haswell.asm
@@ -39,15 +39,16 @@ shl	(2) pic_ref.0<1>:uw	r4.24<2,2,1>:uw 4:uw	{align1};
 mov	(2) pic_ref.16<1>:uw	r4.20<2,2,1>:uw	{align1};
 mov  (8) mb_mvp_ref.0<1>:ud	0:ud		{align1};
 mov  (8) mb_ref_win.0<1>:ud	0:ud		{align1};
+mov  (8) mba_result.0<1>:ud	0x0:ud		{align1};
+mov  (8) mbb_result.0<1>:ud	0x0:ud		{align1};
+mov  (8) mbc_result.0<1>:ud	0x0:ud		{align1};
+
 and.z.f0.0 (1)		null:uw	mb_hwdep<0,1,0>:uw		0x04:uw   {align1};
 (f0.0) jmpi (1) __mb_hwdep_end;
 /* read back the data for MB A */
 /* the layout of MB result is: rx.0(Available). rx.4(MVa), rX.8(MVb), rX.16(Pred_L0 flag),
 *  rX.18 (Pred_L1 flag), rX.20(Forward reference ID), rX.22(Backwared reference ID)
 */
-mov  (8) mba_result.0<1>:ud	0x0:ud		{align1};
-mov  (8) mbb_result.0<1>:ud	0x0:ud		{align1};
-mov  (8) mbc_result.0<1>:ud	0x0:ud		{align1};
 mba_start:
 mov  (8) mb_msg0.0<1>:ud	0:ud		{align1};
 and.z.f0.0 (1)		null:uw	input_mb_intra_ub<0,1,0>:ub	INTRA_PRED_AVAIL_FLAG_AE:uw   {align1};
@@ -330,6 +331,8 @@ mov	(1)	mb_mvp_ref.2<1>:w		RET_ARG<0,1,0>:w	{align1};
 
 __mb_hwdep_end:
         
+mov     (2)     mv_cc_ref.0<1>:w	mba_result.4<2,2,1>:w	{align1};
+
 /* Calibrate the ref window for MPEG2 */
 mov  (1) vme_m0.0<1>:W		-16:W			{align1};
 mov  (1) vme_m0.2<1>:W		-12:W			{align1};
@@ -460,8 +463,9 @@ mov  (1) vme_m1.0<1>:UD         ADAPTIVE_SEARCH_ENABLE:ud {align1} ;
 mov  (1) vme_m1.4<1>:UB         r4.28<0,1,0>:UB {align1};          
 mov  (1) vme_m1.8<1>:UD         START_CENTER + SEARCH_PATH_LEN:UD {align1};
 /* Set the MV cost center */ 
-mov  (1) vme_m1.16<1>:ud	0:ud	{align1};
-mov  (1) vme_m1.20<1>:ud	0:ud	{align1};
+mov  (1) vme_m1.16<1>:ud	mv_cc_ref.0<0,1,0>:ud	{align1};
+mov  (1) vme_m1.20<1>:ud	mv_cc_ref.0<0,1,0>:ud	{align1};
+
 mov  (8) vme_msg_1.0<1>:UD      vme_m1.0<8,8,1>:UD {align1};
 
 mov (8) vme_msg_2<1>:UD		vme_m2.0<8,8,1>:UD {align1};
diff --git a/src/shaders/vme/mpeg2_inter_haswell.g75b b/src/shaders/vme/mpeg2_inter_haswell.g75b
index cdaeb4c..277f116 100644
--- a/src/shaders/vme/mpeg2_inter_haswell.g75b
+++ b/src/shaders/vme/mpeg2_inter_haswell.g75b
@@ -12,11 +12,11 @@
    { 0x00200001, 0x2a500129, 0x00450094, 0x00000000 },
    { 0x00600001, 0x2ac00061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2a800061, 0x00000000, 0x00000000 },
-   { 0x01000005, 0x20002d28, 0x000000a6, 0x00040004 },
-   { 0x00010020, 0x34001c00, 0x00001400, 0x00000740 },
    { 0x00600001, 0x2ae00061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2b000061, 0x00000000, 0x00000000 },
    { 0x00600001, 0x2b200061, 0x00000000, 0x00000000 },
+   { 0x01000005, 0x20002d28, 0x000000a6, 0x00040004 },
+   { 0x00010020, 0x34001c00, 0x00001400, 0x00000710 },
    { 0x00600001, 0x2b400061, 0x00000000, 0x00000000 },
    { 0x01000005, 0x20002e28, 0x000000a5, 0x00600060 },
    { 0x00210001, 0x2af401ed, 0x00000000, 0xffffffff },
@@ -122,14 +122,15 @@
    { 0x00000001, 0x2fa401ad, 0x00000b04, 0x00000000 },
    { 0x00000001, 0x2fa801ad, 0x00000b24, 0x00000000 },
    { 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
-   { 0x00000020, 0x34001c00, 0x00001400, 0x00000720 },
+   { 0x00000020, 0x34001c00, 0x00001400, 0x00000730 },
    { 0x00000001, 0x2ac001ad, 0x00000fe4, 0x00000000 },
    { 0x00000001, 0x2fa001ad, 0x00000ae6, 0x00000000 },
    { 0x00000001, 0x2fa401ad, 0x00000b06, 0x00000000 },
    { 0x00000001, 0x2fa801ad, 0x00000b26, 0x00000000 },
    { 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
-   { 0x00000020, 0x34001c00, 0x00001400, 0x000006c0 },
+   { 0x00000020, 0x34001c00, 0x00001400, 0x000006d0 },
    { 0x00000001, 0x2ac201ad, 0x00000fe4, 0x00000000 },
+   { 0x00200001, 0x2a2001ad, 0x00450ae4, 0x00000000 },
    { 0x00000001, 0x244001ed, 0x00000000, 0xfff0fff0 },
    { 0x00000001, 0x244201ed, 0x00000000, 0xfff4fff4 },
    { 0x00000001, 0x2fa00021, 0x00000440, 0x00000000 },
@@ -180,8 +181,8 @@
    { 0x00000001, 0x24600061, 0x00000000, 0x00000002 },
    { 0x00000001, 0x24640231, 0x0000009c, 0x00000000 },
    { 0x00000001, 0x24680061, 0x00000000, 0x30003030 },
-   { 0x00000001, 0x24700061, 0x00000000, 0x00000000 },
-   { 0x00000001, 0x24740061, 0x00000000, 0x00000000 },
+   { 0x00000001, 0x24700021, 0x00000a20, 0x00000000 },
+   { 0x00000001, 0x24740021, 0x00000a20, 0x00000000 },
    { 0x00600001, 0x28200021, 0x008d0460, 0x00000000 },
    { 0x00600001, 0x28400021, 0x008d0560, 0x00000000 },
    { 0x00000001, 0x28600061, 0x00000000, 0x01010101 },
diff --git a/src/shaders/vme/vme75_mpeg2.inc b/src/shaders/vme/vme75_mpeg2.inc
index c8e42cc..b297a49 100644
--- a/src/shaders/vme/vme75_mpeg2.inc
+++ b/src/shaders/vme/vme75_mpeg2.inc
@@ -19,3 +19,5 @@ define(`INTER_PART_MASK',               `0x7e000000')
 define(`mpeg2_ref',		`r83')
 define(`pic_ref',		`r82')
 define(`INTRA16_DC_PRED',      `0xBB')
+/* Cost center ref */
+define(`mv_cc_ref',		`r81')