Update the shader for DI on IVB
authorXiang, Haihao <haihao.xiang@intel.com>
Thu, 23 Aug 2012 02:55:00 +0000 (22:55 -0400)
committerXiang, Haihao <haihao.xiang@intel.com>
Thu, 23 Aug 2012 02:55:00 +0000 (22:55 -0400)
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
src/i965_post_processing.c
src/shaders/post_processing/gen7/Makefile.am
src/shaders/post_processing/gen7/NV12_DI_NV12.g4a [new file with mode: 0644]
src/shaders/post_processing/gen7/dndi.asm
src/shaders/post_processing/gen7/dndi.g7b

index 6e29430..c8988bd 100755 (executable)
@@ -610,7 +610,7 @@ static const uint32_t pp_nv12_avs_gen7[][4] = {
 };
 
 static const uint32_t pp_nv12_dndi_gen7[][4] = {
-// #include "shaders/post_processing/gen7/dndi.g7b"
+#include "shaders/post_processing/gen7/dndi.g7b"
 };
 
 static const uint32_t pp_nv12_dn_gen7[][4] = {
index 275efa5..b784414 100644 (file)
@@ -17,6 +17,7 @@ INTEL_PP_G4A = \
        DI_Save_NV12_16x4.g4a           \
        DI_Save_PA_16x4.g4a             \
        EOT.g4a                         \
+       NV12_DI_NV12.g4a                \
        NV12_DNDI_422CP.g4a             \
        NV12_DNDI_PA.g4a                \
        NV12_DNUV_NV12.g4a              \
diff --git a/src/shaders/post_processing/gen7/NV12_DI_NV12.g4a b/src/shaders/post_processing/gen7/NV12_DI_NV12.g4a
new file mode 100644 (file)
index 0000000..668b61f
--- /dev/null
@@ -0,0 +1,392 @@
+/*
+ *  Copyright 2000-2011 Intel Corporation All Rights Reserved
+ *
+ *  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.
+ */
+//   57    // Total instruction count
+//    1    // Total kernel count
+
+.kernel NV12_DI_NV12
+.code
+
+
+
+// FileName:   DI.asm 
+// Author:             Vivek Kumar
+// Description:        Tasks for DI only case (16x4 block)
+
+
+
+
+// Module name: common.inc
+//
+// Common header file for all Video-Processing kernels
+//
+
+.default_execution_size (16)
+.default_register_type  :ub
+
+.reg_count_total        128
+.reg_count_payload      7
+
+//========== Common constants ==========
+
+
+//========== Macros ==========
+
+
+//Fast Jump, For more details see "Set_Layer_N.asm"
+
+
+//========== Defines ====================
+
+//========== Static Parameters (Common To All) ==========
+//r1
+
+
+//r2
+
+                                    //  e.g.            byte0   byte1  byte2
+                                    // YUYV               0       1      3
+                                    // YVYU               0       3      1
+
+//Color Pipe (IECP) parameters
+
+
+//ByteCopy
+
+
+//r4
+
+                                    //  e.g.              byte0           byte1           byte2
+                                    // YUYV                 0               1               3
+                                    // YVYU                 0               3               1
+
+
+//========== Inline parameters (Common To All) ===========
+
+
+//============== Binding Index Table===========
+//Common between DNDI and DNUV
+
+
+//================= Common Message Descriptor =====
+// Message descriptor for thread spawning
+// Message Descriptors
+//                = 000 0001 (min message len 1 ) 0,0000 (resp len 0   -add later)
+//                  0000,0000,0000
+//                  0001(Spawn a root thread),0001 (Root thread spawn thread)
+//                = 0x02000011
+// Thread Spawner Message Descriptor
+
+
+// Message descriptor for atomic operation add
+// Message Descriptors
+//                = 000 0110 (min message len 6 ) 0,0000 (resp len 0   -add later)
+//                  1(header present)001,10(typed atomic operation)0(return enabled)0(slot group, low 8 bits),0111 (AOP_Add)
+//                  0000,0000 (Binding table index, added later)
+//                = 0x02000011
+
+// Atomic Operation Add Message Descriptor
+
+
+// Message descriptor for dataport media write
+        // Message Descriptors
+                //                = 000 0001 (min message len 1 - add later) 00000 (resp len 0)
+                //                  1 (header present 1) 0 1010 (media block write) 000000
+                //                  00000000 (binding table index - set later)
+                //                = 0x020A8000
+
+
+// Message Length defines
+
+
+// Response Length defines
+
+
+// Block Width and Height Size defines
+
+
+// Extended Message Descriptors
+
+
+// Common message descriptors:
+
+
+//===================== Math Function Control ===================================
+
+
+//============ Message Registers ===============
+                             // buf4 starts from r28
+
+
+//#define mMSGHDR_EOT  r43    // Dummy Message Register for EOT
+
+
+.declare    mubMSGPAYLOAD  Base=r30 ElementSize=1 SrcRegion=<16;16,1> Type=ub
+.declare    muwMSGPAYLOAD  Base=r30 ElementSize=2 SrcRegion=<16;16,1> Type=uw
+.declare    mudMSGPAYLOAD  Base=r30 ElementSize=4 SrcRegion=<8;8,1> Type=ud
+.declare    mfMSGPAYLOAD   Base=r30 ElementSize=4 SrcRegion=<8;8,1> Type=f
+
+//=================== End of thread instruction ===========================
+
+
+//=====================Pointers Used=====================================
+
+
+//=======================================================================
+
+
+//r9-r17
+// Define temp space for any usages
+
+
+// Common Buffers
+
+
+// temp space for rotation
+
+.declare fROBUF                  Base=r9.0             ElementSize=4           SrcRegion=<8;8,1>                 DstRegion=<1>         Type=f
+
+.declare udROBUF               Base=r9.0               ElementSize=4           SrcRegion=<8;8,1>                 DstRegion=<1>         Type=ud
+
+.declare uwROBUF               Base=r9.0               ElementSize=2           SrcRegion=<16;16,1>             DstRegion=<1>           Type=uw
+
+.declare ubROBUF               Base=r9.0               ElementSize=1           SrcRegion=<16;16,1>             DstRegion=<1>           Type=ub
+
+.declare ub4ROBUF      Base=r9.0               ElementSize=1           SrcRegion=<32;8,4>              DstRegion=<4>           Type=ub
+
+
+// End of common.inc
+
+
+// FileName:    DNDI.inc
+// Author:      Vivek Kumar
+// Description: Include file for DN, DI and DNDI
+// Inputs:      DI_ENABLE, DN_ENABLE, DN_PLANAR, DN_PACKED
+
+
+
+
+// End of common.inc
+
+
+//Interface:
+//Static Parameters:
+//r1
+
+
+//====================== Binding table (Explicit To DNDI)=========================================
+
+
+.declare mudMSGHDR_DNDI     Base=r18      ElementSize=4    Type=ud
+.declare mdMSGHDR_DNDI      Base=r18      ElementSize=4    Type=d
+.declare mwMSGHDR_DNDI      Base=r18      ElementSize=2    Type=w
+
+
+.declare mudMSGHDR_STMM     Base=r20      ElementSize=4    Type=ud
+
+
+.declare mudMSGHDR_HIST     Base=r22      ElementSize=4    Type=ud
+
+
+.declare mudMSGHDR_ENC_STATS Base=r24 ElementSize=4   Type=ud
+.declare muwMSGHDR_ENC_STATS Base=r24 ElementSize=2   Type=uw
+.declare mubMSGHDR_ENC_STATS Base=r24 ElementSize=1   Type=ub
+
+
+.declare mudMSGHDR_DN_OUT   Base=r31.0      ElementSize=4  Type=ud
+.declare mdMSGHDR_DN_OUT    Base=r31.0      ElementSize=4  Type=d
+.declare mubMSGHDR_DN_OUT   Base=r31.0      ElementSize=1  Type=ub
+
+
+.declare mudMSGHDR_UVCOPY   Base=r36      ElementSize=4  Type=ud
+.declare mdMSGHDR_UVCOPY    Base=r36      ElementSize=4  Type=d
+.declare mudMSGHDR_UCOPY    Base=r36       ElementSize=4  Type=ud
+.declare mudMSGHDR_VCOPY    Base=r38       ElementSize=4  Type=ud
+
+
+.declare mudMSGHDR_DI_OUT1  Base=r18.0      ElementSize=4     Type=ud
+.declare mubMSGHDR_DI_OUT1  Base=r18.0      ElementSize=1     Type=ub
+
+
+.declare mudMSGHDR_DI_OUT2  Base=r23.0      ElementSize=4     Type=ud
+.declare mubMSGHDR_DI_OUT2  Base=r23.0      ElementSize=1     Type=ub
+
+//r45
+//Use r45 as message header, so no need to "mov" the data.
+
+.declare mudDN_Y_OUT        Base=r45.0 ElementSize=4 SrcRegion=<8;8,1>   DstRegion=<1> Type=ud
+
+// Message response (Denoised & DI-ed pixels & statistics); Use buffer 5
+.declare udDNDI_RESP        Base=r46.0 ElementSize=4 SrcRegion=<8;8,1>   DstRegion=<1> Type=ud
+.declare uwDNDI_RESP        Base=r46.0 ElementSize=2 SrcRegion=<16;16,1> DstRegion=<1> Type=uw
+.declare ubDNDI_RESP        Base=r46.0 ElementSize=1 SrcRegion=<16;16,1> DstRegion=<1> Type=ub
+
+// Message response (UV Copy); Use buffer 5
+.declare udDNDI_UV_RESP     Base=r58.0 ElementSize=4 SrcRegion=<8;8,1>  DstRegion=<1> Type=ud
+.declare ubDNDI_UV_RESP     Base=r58.0 ElementSize=1 SrcRegion=<16;16,1>    DstRegion=<1> Type=ub
+
+//Temp GRFs: For 42X to 422 Conversion
+.declare uwDNDI_UVCOPY_TEMP Base=r10.0 ElementSize=2 SrcRegion=<16;16,1>    DstRegion=<1> Type=uw       //8 GRFs
+.declare ubDNDI_UVCOPY_TEMP Base=r10.0 ElementSize=1 SrcRegion=<16;16,1>    DstRegion=<1> Type=ub       //8 GRFs
+//---------------------------------------------------------------------------
+// Message descriptors
+//---------------------------------------------------------------------------
+// Extended message descriptor
+    // Message descriptor   for sampler read
+    //                    = 000 0010 (message len 2) 00000 (resp len - set later, 12 or 5 or 11)
+    //                      1 (header present 1) 0 11 (SIMD32/64 mode)
+    //                      1000 (message type) 0000 (DI state index)
+    //                      00000000 (binding table index - set later)
+    //                    = 0x040b8000
+
+
+// Attention: The Message Length is The Number of GRFs with Data Only, without the Header
+
+
+//---------------------------------------------------------------------------
+// VDI Return Data format
+//---------------------------------------------------------------------------
+// Defines for DI enabled
+
+
+// Defines for DI disabled
+
+
+
+// FileName:   DNDI_Command.asm 
+// Author:             Vivek Kumar
+// Description:        Sends a message to the VDI to process one DN (16x8) or DNDI (16x4) block
+
+// Prepare the DNDI send command
+mov (8)                mudMSGHDR_DNDI(0)<1>                    r0.0<8;8,1>:ud                                  // message header
+mov (1)                mwMSGHDR_DNDI(1,4)<1>                   r7.0<0;1,0>:w           { NoDDClr }             // horizontal origin    // Do we need to add offset here? -vK
+mov (1)                mwMSGHDR_DNDI(1,12)<1>                  r7.1<0;1,0>:w           { NoDDChk }             // vertical origin              // Can these 2 be combined? - vK
+
+send (8)       udDNDI_RESP(0)<1>       r18     0x2     0x4AE8003:ud
+
+// On Gen6, with VDI walker, use the XY pair returned rather than programmed above
+// VDI_RETURNED_XY is ordered XY in case of walker enables and the same as programmed in case of walker disabled
+mov (2)        r7.0<1>:w     uwDNDI_RESP(9,14)<2;2,1>  // horizontal/Vertial origin in W.14 and W.15    
+
+
+
+// FileName:   DI_STMM_Save.asm 
+// Author:             Vivek Kumar
+// Description:        Saves DI STMM Data to statistics surface in case of DI enabled (for 16x4 block)
+
+// Write STMM to memory
+mov (8)     mudMSGHDR_STMM(0)<1>       r0.0<8;8,1>:ud                                                                  // message header   
+mov (8)     mudMSGHDR_STMM(1)<1>       udDNDI_RESP(8,0)                                                // Move STMM to MRF 
+
+shr (1)     mudMSGHDR_STMM(0,0)<1>     r7.0<0;1,0>:w            1:w     { NoDDClr }            // X origin / 2
+mov (1)     mudMSGHDR_STMM(0,1)<1>     r7.1<0;1,0>:w                    { NoDDClr, NoDDChk }   // Y origin
+mov (1)     mudMSGHDR_STMM(0,2)<1>     0x30007:ud           { NoDDChk }                // block width and height (8x4)
+
+send (8)       null<1>:d       r20     0x5             0x40A8021:ud      
+
+
+
+// FileName:   DNDI_Enc_Stats_Save.asm
+// Author:             Vivek Kumar
+// Description:        Saves Encoder Statistics data to statistics surface in case of DI enabled (for 16x4 block)
+
+// Write encoder statistics to memory
+//Currently enable this only on Gen6 validation
+mov (8)                mudMSGHDR_ENC_STATS(1)<1>       0x0:ud                                          // Init payload MRF
+mov (8)                mudMSGHDR_ENC_STATS(0)<1>       r0.0<8;8,1>:ud                          // message header
+
+shr (1)                mudMSGHDR_ENC_STATS(0,0)<1>             r7.0<0;1,0>:w                   1:w     { NoDDClr }                     //enable the flag after testing on si                                   { NoDDClr }     // X origin / 2
+mul (1)                acc0.1<1>:ud                                    r7.1<0;1,0>:w                           3:w                                                                                                                                                                                     // Y origin * 3
+shr (1)                mudMSGHDR_ENC_STATS(0,1)<1>             acc0.1<0;1,0>:ud                        2:w             { NoDDClr, NoDDChk }            //enable the flag after testing on si                      { NoDDClr, NoDDChk } // Y origin * 3/4
+mov (1)                mudMSGHDR_ENC_STATS(0,2)<1>             0x20007:ud                              { NoDDChk }                     //enable the flag after testing on si                                           { NoDDChk } // block width and height (8x3)
+add (2)                mudMSGHDR_ENC_STATS(0,0)<1>             mudMSGHDR_ENC_STATS(0,0)<2;2,1>         r1.12<2;2,1>:uw                                 // Add pitch to X,Y origin
+
+
+       //Data block for Encoder Statistics
+       //----------------------------------------------------
+       //|  0  |   1  |   2   |  3  |  4  |  5  |  6  |  7  | Bytes
+       //----------------------------------------------------
+       //| BNE | MCNT | FCNT | TCNT |  X  |  X  |  X  |  X  |
+       //----------------------------------------------------
+       //|   DcTpT    |     SVCM    |   DcBpT   |   DcTpB   |
+       //----------------------------------------------------
+       //|   SHCM     |     STAD    |   DcTcB   |   DcBpB   |
+       //----------------------------------------------------
+       mov (1)         mudMSGHDR_ENC_STATS(1,0)<1>             udDNDI_RESP(9,1)<0;1,0>                 { NoDDClr }                     // Move encoder statistics to MRF
+       mov (2)         mudMSGHDR_ENC_STATS(1,3)<2>             udDNDI_RESP(9,3)<2;2,1>                 { NoDDClr, NoDDChk }            // Move encoder statistics to MRF
+       mov (2)         mudMSGHDR_ENC_STATS(1,2)<2>             udDNDI_RESP(9,5)<2;2,1>                 { NoDDChk }                     // Move encoder statistics to MRF
+
+
+send (8)   null<1>:d    r24    0x5    0x40A8021:ud
+
+
+
+// FileName:    DI_Save_NV12_16x4.asm
+// Author:      Vivek Kumar
+// Description: Save two 16x4 blocks of DI output in Packed format
+
+
+// add (4) a0.4<1>:uw   r2.28<4;4,1>:ub   608:w               // Initial Y,U,V offset in YUV422 block; it starts at m20
+
+mov (8) r28.0<1>:ud     r0.0<8;8,1>:ud
+mov (1) r28.0<1>:d      r7.0<0;1,0>:w            { NoDDClr }          // H. block origin need to be doubled
+mov (1) r28.1<1>:d      r7.1<0;1,0>:w                 { NoDDClr, NoDDChk }       // Block origin
+mov (1) r28.2<1>:ud     0x3000F:ud          { NoDDChk }          // Block width and height (32x8)
+
+//prepare the message headers
+mov (8) r18.0<1>:ud       r28<8;8,1>:ud
+mov (8) r23.0<1>:ud       r28<8;8,1>:ud
+
+//Bottom field Y
+mov (8)   mudMSGHDR_DI_OUT1(1)<1>              udDNDI_RESP(0,0)
+mov (8)   mudMSGHDR_DI_OUT1(2)<1>              udDNDI_RESP(0,8)
+// Top field Y
+mov (8)   mudMSGHDR_DI_OUT2(1)<1>      udDNDI_RESP(4,0)
+mov (8)   mudMSGHDR_DI_OUT2(2)<1>      udDNDI_RESP(4,8)
+
+//Change origin to U/V block
+asr (1) r28.1<1>:d             r28.1<0;1,0>:d    1:w                   { NoDDClr }             // U/V block origin should be half of Y's
+mov (1) r28.2<1>:ud     0x1000F:ud             { NoDDChk }             // Block width and height (16x2)
+
+// Bottom field U/V
+mov (16)   r22.0<2>:ub       ubDNDI_RESP(2, 1)<32;8,2> { NoDDClr }
+mov (16)   r22.1<2>:ub       ubDNDI_RESP(2, 0)<32;8,2> { NoDDChk }
+
+// Top field U/V
+mov (16)   r27.0<2>:ub       ubDNDI_RESP(6, 1)<32;8,2> { NoDDClr }
+mov (16)   r27.1<2>:ub       ubDNDI_RESP(6, 0)<32;8,2> { NoDDChk }
+
+//copy message desrcptor to the message header
+mov (8) r21<1>:ud                      r28<8;8,1>:ud
+mov (8) r26<1>:ud                      r28<8;8,1>:ud
+
+//Send out Y component on previous frame to surface
+send (8)    null<1>:d    r18   0x5    0x60A801B:ud
+//Send out Y component on current frame to surface
+send (8)    null<1>:d    r23   0x5    0x60A801E:ud
+//Send out U/V component on previous frame to surface
+send (8)    null<1>:d    r21   0x5    0x40A801C:ud
+//Send out U/V component on current frame to surface
+send (8)    null<1>:d    r26   0x5    0x40A801F:ud
+
+
+//End of Thread message
+
+mov (8) r127<1>:ud r0.0<8;8,1>:ud 
+ send (1) null<1>:d r127 0x27 0x02000010
+
+
+.end_code 
+.end_kernel
index 65962a5..b820fdb 100644 (file)
@@ -2,9 +2,7 @@
 .kernel DNDI
 .code
 
-#include "DI_Core.g4a"
-#include "DI_Save_NV12_16x4.g4a"
-#include "EOT.g4a"
+#include "NV12_DI_NV12.g4a"
 
 .end_code  
 
index ad26525..ffa9d49 100644 (file)
    { 0x00200c01, 0x432c0021, 0x004506ec, 0x00000000 },
    { 0x00200801, 0x43280021, 0x004506f4, 0x00000000 },
    { 0x05600031, 0x20000e24, 0x00000300, 0x040a8021 },
-   { 0x00200401, 0x236001a5, 0x004500e0, 0x00000000 },
-   { 0x00000801, 0x23680061, 0x00000000, 0x0003000f },
+   { 0x00600001, 0x23800021, 0x008d0000, 0x00000000 },
+   { 0x00000401, 0x238001a5, 0x000000e0, 0x00000000 },
+   { 0x00000c01, 0x238401a5, 0x000000e2, 0x00000000 },
+   { 0x00000801, 0x23880061, 0x00000000, 0x0003000f },
+   { 0x00600001, 0x22400021, 0x008d0380, 0x00000000 },
+   { 0x00600001, 0x22e00021, 0x008d0380, 0x00000000 },
    { 0x00600001, 0x22600021, 0x008d05c0, 0x00000000 },
    { 0x00600001, 0x22800021, 0x008d05e0, 0x00000000 },
    { 0x00600001, 0x23000021, 0x008d0640, 0x00000000 },
    { 0x00600001, 0x23200021, 0x008d0660, 0x00000000 },
-   { 0x00600001, 0x22400021, 0x008d0360, 0x00000000 },
-   { 0x00600001, 0x22e00021, 0x008d0360, 0x00000000 },
-   { 0x0000040c, 0x23643ca5, 0x00000364, 0x00010001 },
-   { 0x00000801, 0x23680061, 0x00000000, 0x0001000f },
-   { 0x00800401, 0x42a00231, 0x00ce0601, 0x00000000 },
-   { 0x00800801, 0x42a10231, 0x00ce0600, 0x00000000 },
-   { 0x00800401, 0x43400231, 0x00ce0681, 0x00000000 },
-   { 0x00800801, 0x43410231, 0x00ce0680, 0x00000000 },
-   { 0x00600001, 0x22a00021, 0x008d0360, 0x00000000 },
-   { 0x00600001, 0x23400021, 0x008d0360, 0x00000000 },
+   { 0x0000040c, 0x23843ca5, 0x00000384, 0x00010001 },
+   { 0x00000801, 0x23880061, 0x00000000, 0x0001000f },
+   { 0x00800401, 0x42c00231, 0x00ce0601, 0x00000000 },
+   { 0x00800801, 0x42c10231, 0x00ce0600, 0x00000000 },
+   { 0x00800401, 0x43600231, 0x00ce0681, 0x00000000 },
+   { 0x00800801, 0x43610231, 0x00ce0680, 0x00000000 },
+   { 0x00600001, 0x22a00021, 0x008d0380, 0x00000000 },
+   { 0x00600001, 0x23400021, 0x008d0380, 0x00000000 },
    { 0x05600031, 0x20000e24, 0x00000240, 0x060a801b },
    { 0x05600031, 0x20000e24, 0x000002e0, 0x060a801e },
    { 0x05600031, 0x20000e24, 0x000002a0, 0x040a801c },