Skip to content

L
libva
Commit 20975a94 authored by Xiang, Haihao's avatar Xiang, Haihao
Browse files
Options

i965_drv_video: add video processing kernels

parent 5f2030ba
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 7730 additions and 1 deletion
configure.ac
View file @ 20975a94
...@@ -163,6 +163,7 @@ AC_OUTPUT([ ...@@ -163,6 +163,7 @@ AC_OUTPUT([
i965_drv_video/shaders/mpeg2/Makefile i965_drv_video/shaders/mpeg2/Makefile
i965_drv_video/shaders/mpeg2/vld/Makefile i965_drv_video/shaders/mpeg2/vld/Makefile
i965_drv_video/shaders/render/Makefile i965_drv_video/shaders/render/Makefile
i965_drv_video/shaders/post_processing/Makefile
test/Makefile test/Makefile
test/basic/Makefile test/basic/Makefile
test/decode/Makefile test/decode/Makefile
......
i965_drv_video/shaders/Makefile.am
View file @ 20975a94
SUBDIRS = h264 mpeg2 render SUBDIRS = h264 mpeg2 render post_processing
i965_drv_video/shaders/post_processing/Common/AYUV_Load_16x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: AYUV_Load_16x8.asm
//----------------------------------------------------------------
#include "AYUV_Load_16x8.inc"
// In order to load 64x8 AYUV data (16x8 pixels), we need to divide the data
// into two regions and load them separately.
//
// 32 byte 32 byte
//|----------------|----------------|
//| | |
//| A | B |8
//| | |
//| | |
//|----------------|----------------|
// Load the first 32x8 data block
// Packed data block should be loaded as 32x8 pixel block
add (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w wSRC_H_ORI_OFFSET<2;2,1>:w // Source Block origin
shl (1) rMSGSRC.0<1>:d acc0:w 2:w { NoDDClr } // H. block origin need to be four times larger
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_YUV:ud { NoDDChk } // Block width and height (32x8)
mov (8) mMSGHDRY<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_YUV(0)<1> mMSGHDRY udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_YUV+nBI_CURRENT_SRC_YUV:ud
//Load the second 32x8 data block
// Offset the origin X - move to next 32 colomns
add (1) rMSGSRC.0<1>:d rMSGSRC.0<0;1,0>:d 32:w // Increase X origin by 8
// Size stays the same - 32x8
mov (8) mMSGHDRY<1>:ud rMSGSRC<8;8,1>:ud // Copy message description to message header
send (8) udSRC_YUV(8)<1> mMSGHDRY udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_YUV+nBI_CURRENT_SRC_YUV:ud
// Give AYUV region addresses to address register
mov (1) SRC_YUV_OFFSET<1>:ud 0x00400038*32:ud //Address registers contain starting addresses of two halves
//Directly move the data to destination
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (16) uwDEST_Y(%1)<1> r[SRC_YUV_OFFSET,%1*32+2]<8,4>:ub
mov (16) uwDEST_U(%1)<1> r[SRC_YUV_OFFSET,%1*32+1]<8,4>:ub
mov (16) uwDEST_V(%1)<1> r[SRC_YUV_OFFSET,%1*32+0]<8,4>:ub
}
\ No newline at end of file
i965_drv_video/shaders/post_processing/Common/AYUV_Load_16x8.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: AYUV_Load_16x8.inc
//
// AYUV data are first loaded to bottom I/O REGION_2, then unpacked to planar data
// and stored in top I/O REGION_1
#undef nY_NUM_OF_ROWS
#define nY_NUM_OF_ROWS 8 // Number of Y rows per block
#define nDPR_BLOCK_SIZE_YUV nBLOCK_WIDTH_32+nBLOCK_HEIGHT_8 // Y block size 32x8
#define nDPR_MSG_SIZE_YUV nRESLEN_8 // # of MRF's to hold Y block data (8)
//Temporary storage for unpacked AYUV data
#define rUNPACK_TEMP REG(r,nTEMP0)
.declare udUNPACK_TEMP Base=rUNPACK_TEMP ElementSize=4 SrcRegion=<8;8,1> Type=ud //1 GRF
.declare ubUNPACK_TEMP Base=rUNPACK_TEMP ElementSize=1 SrcRegion=<32;32,1> Type=ub //1 GRF
.declare ubBOT_Y_IO Base=REG(r,nBOT_Y) ElementSize=1 SrcRegion=REGION(32,1) Type=ub
#define udSRC_YUV udBOT_Y_IO
#define ubSRC_YUV ubBOT_Y_IO
#define nSRC_YUV_REG nBOT_Y
#define uwDEST_Y uwTOP_Y
#define uwDEST_U uwTOP_U
#define uwDEST_V uwTOP_V
#define SRC_YUV_OFFSET a0.0
#define nSRC_REGION nREGION_1 // REGION_1 will be the source region for first kernel
// End of AYUV_Load_16x8.inc
i965_drv_video/shaders/post_processing/Common/Expansion.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: Expansion.inc
// Number of U/V rows per block definition
#undef nUV_NUM_OF_ROWS
#ifdef EXPAND_9x5
#define nUV_NUM_OF_ROWS 6
#else
#define nUV_NUM_OF_ROWS 8
#endif
// Source/destination region definitions
#undef uwDEST_U
#undef uwDEST_V
#if (nSRC_REGION==nREGION_1)
#define uwDEST_U uwTOP_U
#define uwDEST_V uwTOP_V
#elif (nSRC_REGION==nREGION_2)
#define uwDEST_U uwBOT_U
#define uwDEST_V uwBOT_V
#endif
// End of Expansion.inc
i965_drv_video/shaders/post_processing/Common/IMC3_Load_8x4.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: IMC3_Load_8x4.asm
//
//----------------------------------------------------------------
#define IMC3_LOAD_8x4
#include "PL3_Load.inc"
// Load 16x8 planar Y ----------------------------------------------------------
add (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w wSRC_H_ORI_OFFSET<2;2,1>:w // Source Y Block origin
#if !defined(LOAD_UV_ONLY)
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_Y:ud // Block width and height (16x8)
mov (8) mMSGHDRY<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_Y(0)<1> mMSGHDRY udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_Y+nBI_CURRENT_SRC_Y:ud
#endif
// Load 8x4 planar U and V -----------------------------------------------------
asr (2) rMSGSRC.0<1>:d rMSGSRC.0<2;2,1>:d 1:w // U/V block origin should be half of Y's
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_UV:ud // U/V block width and height (8x4)
mov (8) mMSGHDRU<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_U(0)<1> mMSGHDRU udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_CURRENT_SRC_U:ud
mov (8) mMSGHDRV<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_V(0)<1> mMSGHDRU udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_CURRENT_SRC_V:ud
// Convert to word-aligned format ----------------------------------------------
#if !defined(LOAD_UV_ONLY)
$for (nY_NUM_OF_ROWS-1; >-1; -1) {
mov (16) uwDEST_Y(0,%1*16)<1> ubSRC_Y(0,%1*16)
}
#endif
$for (nUV_NUM_OF_ROWS/2-1; >-1; -1) {
mov (16) uwDEST_U(0, %1*16)<1> ubSRC_U(0, %1*16)
mov (16) uwDEST_V(0, %1*16)<1> ubSRC_V(0, %1*16)
}
// End of IMC3_Load_8x4
i965_drv_video/shaders/post_processing/Common/IMC3_Load_8x5.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: IMC3_Load_8x5.asm
//
//----------------------------------------------------------------
#define IMC3_LOAD_8x5
#include "PL3_Load.inc"
// Load 16x8 planar Y ----------------------------------------------------------
add (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w wSRC_H_ORI_OFFSET<2;2,1>:w // Source Y Block origin
#if !defined(LOAD_UV_ONLY)
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_Y:ud // Block width and height (16x8)
mov (8) mMSGHDRY<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_Y(0)<1> mMSGHDRY udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_Y+nBI_CURRENT_SRC_Y:ud
#endif
// Load 8x5 planar U and V -----------------------------------------------------
asr (2) rMSGSRC.0<1>:d rMSGSRC.0<2;2,1>:d 1:w // U/V block origin should be half of Y's
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_UV:ud // U/V block width and height (8x5)
mov (8) mMSGHDRU<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_U(0)<1> mMSGHDRU udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_CURRENT_SRC_U:ud
mov (8) mMSGHDRV<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_V(0)<1> mMSGHDRU udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_CURRENT_SRC_V:ud
// Convert to word-aligned format ----------------------------------------------
#if !defined(LOAD_UV_ONLY)
$for (nY_NUM_OF_ROWS-1; >-1; -1) {
mov (16) uwDEST_Y(0,%1*16)<1> ubSRC_Y(0,%1*16)
}
#endif
$for (nUV_NUM_OF_ROWS/2-1; >-1; -1) {
mov (16) uwDEST_U(0, %1*16)<1> ubSRC_U(0, %1*16)
mov (16) uwDEST_V(0, %1*16)<1> ubSRC_V(0, %1*16)
}
// End of IMC3_Load_8x5
i965_drv_video/shaders/post_processing/Common/IMC3_Load_9x5.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: IMC3_Load_9x5.asm
//
//----------------------------------------------------------------
// This module loads 16x8 Y, 9x5 U and 9x5 V planar data blocks for CSC module
// and stores it in byte-aligned format.
//----------------------------------------------------------------
#define IMC3_LOAD_9x5
#include "PL3_Load.inc"
// Load 16x8 planar Y ----------------------------------------------------------
add (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w wSRC_H_ORI_OFFSET<2;2,1>:w // Source Y Block origin
#if !defined(LOAD_UV_ONLY)
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_Y:ud // Block width and height (16x8)
mov (8) mMSGHDRY<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_Y(0)<1> mMSGHDRY udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_Y+nBI_CURRENT_SRC_Y:ud
#endif
// Load 9x5 planar U and V -----------------------------------------------------
asr (2) rMSGSRC.0<1>:d rMSGSRC.0<2;2,1>:d 1:w // U/V block origin should be half of Y's
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_UV:ud // U/V block width and height (12x5)
mov (8) mMSGHDRU<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_U(0)<1> mMSGHDRU udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_CURRENT_SRC_U:ud
mov (8) mMSGHDRV<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_V(0)<1> mMSGHDRU udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_CURRENT_SRC_V:ud
// Convert to word-aligned format ----------------------------------------------
#if !defined(LOAD_UV_ONLY)
$for (nY_NUM_OF_ROWS-1; >-1; -1) {
mov (16) uwDEST_Y(0,%1*16)<1> ubSRC_Y(0,%1*16)
}
#endif
$for(nUV_NUM_OF_ROWS-2; >-1; -1) {
mov (16) uwDEST_U(0, %1*16)<1> ubSRC_U(0, %1*16)
mov (16) uwDEST_V(0, %1*16)<1> ubSRC_V(0, %1*16)
}
// End of IMC3_Load_9x5
i965_drv_video/shaders/post_processing/Common/Init_All_Regs.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
#ifdef GT // to remove error messages of un-initialized GRF
.declare udGRF_space Base=r0.0 ElementSize=4 SrcRegion=REGION(8,1) Type=ud
$for (7; <80; 1) {
mov (8) udGRF_space(%1)<1> 0:ud
}
#else
#endif
\ No newline at end of file
i965_drv_video/shaders/post_processing/Common/Multiple_Loop.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
/////////////////////////////////////////////////////////////////////////////////
// Multiple_Loop.asm
// This lable is for satisfying component kernel build.
// DL will remove this label and reference the real one in Multiple_Loop_Head.asm.
#if defined(COMPONENT)
VIDEO_PROCESSING_LOOP:
#endif
//===== Possible build flags for component kernels
// 1) INC_SCALING
// 2) INC_BLENDING
// 3) INC_BLENDING and INC_SCALING
// 4) (no flags)
#define MxN_MULTIPLE_BLOCKS
//------------------------------------------------------------------------------
#if defined(MxN_MULTIPLE_BLOCKS)
// Do Multiple Block Processing ------------------------------------------------
// The 1st block has been processed before entering the loop
// Processed all blocks?
add.z.f0.0 (1) wNUM_BLKS:w wNUM_BLKS:w -1:w
// Reached multi-block width?
add (1) wORIX:w wORIX:w 16:w
cmp.l.f0.1 (1) null:w acc0.0:w wFRAME_ENDX:w // acc0.0 has wORIX
#if defined(INC_SCALING)
// Update SRC_VID_H_ORI for scaling
mul (1) REG(r,nTEMP0):f fVIDEO_STEP_X:f 16.0:f
add (1) fSRC_VID_H_ORI:f REG(r,nTEMP0):f fSRC_VID_H_ORI:f
#endif
#if defined(INC_BLENDING)
// Update SRC_ALPHA_H_ORI for blending
mul (1) REG(r,nTEMP0):f fALPHA_STEP_X:f 16.0:f
add (1) fSRC_ALPHA_H_ORI:f REG(r,nTEMP0):f fSRC_ALPHA_H_ORI:f
#endif
(f0.0)jmpi (1) END_VIDEO_PROCESSING // All blocks are done - Exit loop
(f0.1)jmpi (1) VIDEO_PROCESSING_LOOP // If not the end of row, goto the beginning of the loop
//If end of row, restart Horizontal offset and calculate Vertical offsets next row.
mov (1) wORIX:w wCOPY_ORIX:w
add (1) wORIY:w wORIY:w 8:w
#if defined(INC_SCALING)
// Update SRC_VID_H_ORI and SRC_VID_V_ORI for scaling
mov (1) fSRC_VID_H_ORI:f fFRAME_VID_ORIX:f // Reset normalised X origin to 0 for video and alpha
mul (1) REG(r,nTEMP0):f fVIDEO_STEP_Y:f 8.0:f
add (1) fSRC_VID_V_ORI:f REG(r,nTEMP0):f fSRC_VID_V_ORI:f
#endif
#if defined(INC_BLENDING)
// Update SRC_ALPHA_H_ORI and SRC_ALPHA_V_ORI for blending
mov (1) fSRC_ALPHA_H_ORI:f fFRAME_ALPHA_ORIX:f // Reset normalised X origin to 0 for video and alpha
mul (1) REG(r,nTEMP0):f fALPHA_STEP_Y:f 8.0:f
add (1) fSRC_ALPHA_V_ORI:f REG(r,nTEMP0):f fSRC_ALPHA_V_ORI:f
#endif
jmpi (1) VIDEO_PROCESSING_LOOP // Continue Loop
END_VIDEO_PROCESSING:
nop
#endif
END_THREAD // End of Thread
\ No newline at end of file
i965_drv_video/shaders/post_processing/Common/Multiple_Loop_Head.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//////////////////////////////////////////////////////////////////////////////////
// Multiple_Loop_Head.asm
// This code sets up the loop control for multiple blocks per thread
mul (1) wFRAME_ENDX:w ubBLK_CNT_X:ub 16:uw { NoDDClr } // Build multi-block loop counters
mov (1) wNUM_BLKS:w ubNUM_BLKS:ub { NoDDClr, NoDDChk } // Copy num blocks to word variable
mov (1) wCOPY_ORIX:w wORIX:w { NoDDChk } // Copy multi-block origin in pixel
mov (2) fFRAME_VID_ORIX<1>:f fSRC_VID_H_ORI<4;2,2>:f // Copy src video origin for scaling, and alpha origin for blending
add (1) wFRAME_ENDX:w wFRAME_ENDX:w wORIX:w // Continue building multi-block loop counters
VIDEO_PROCESSING_LOOP: // Loop back entry point as the biginning of the loop for multiple blocks
// Beginning of the loop
i965_drv_video/shaders/post_processing/Common/NV11_Load_4x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: NV11_Load_4x8.asm
//----------------------------------------------------------------
#define NV11_LOAD_4x8
#include "PL2_Load.inc"
// Load 16x8 NV11 Y ------------------------------------------------------------
add (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w wSRC_H_ORI_OFFSET<2;2,1>:w // Source Y Block origin
#if !defined(LOAD_UV_ONLY)
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_Y:ud // Y block width and height (16x8)
mov (8) mMSGHDRY<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_Y(0)<1> mMSGHDRY udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_Y+nBI_CURRENT_SRC_Y:ud
#endif
// Load 8x8 NV11 UV ----------------------------------------------------------
asr (1) rMSGSRC.0<1>:d rMSGSRC.0<0;1,0>:d 1:w // U/V block origin should be half of Y's
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_UV:ud // U/V block width and height (8x8)
mov (8) mMSGHDRU<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_U(0)<1> mMSGHDRU udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_CURRENT_SRC_UV:ud
// Convert to word-aligned format ----------------------------------------------
#if !defined(LOAD_UV_ONLY)
$for (nY_NUM_OF_ROWS-1; >-1; -1) {
mov (16) uwDEST_Y(0,%1*16)<1> ubSRC_Y(0,%1*16)
}
#endif
$for (nUV_NUM_OF_ROWS/4-1; >-1; -1) {
mov (16) uwDEST_U(0,%1*16)<1> ubSRC_U(0,%1*32)<32;16,2>
mov (16) uwDEST_V(0,%1*16)<1> ubSRC_U(0,%1*32+1)<32;16,2>
}
// End of NV11_Load_4x8
i965_drv_video/shaders/post_processing/Common/NV11_Load_5x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: NV11_Load_5x8.asm
//----------------------------------------------------------------
#define NV11_LOAD_5x8
#include "PL2_Load.inc"
// Load 16x8 NV11 Y ------------------------------------------------------------
add (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w wSRC_H_ORI_OFFSET<2;2,1>:w // Source Y Block origin
#if !defined(LOAD_UV_ONLY)
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_Y:ud // Y block width and height (16x8)
mov (8) mMSGHDRY<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_Y(0)<1> mMSGHDRY udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_Y+nBI_CURRENT_SRC_Y:ud
#endif
// Load 12x8 NV11 UV ---------------------------------------------------------
asr (1) rMSGSRC.0<1>:d rMSGSRC.0<0;1,0>:d 1:w // U/V block origin should be half of Y's
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_UV:ud // U/V block width and height (12x8)
mov (8) mMSGHDRU<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_U(0)<1> mMSGHDRU udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_CURRENT_SRC_UV:ud
// Convert to word-aligned format ----------------------------------------------
#if !defined(LOAD_UV_ONLY)
$for (nY_NUM_OF_ROWS-1; >-1; -1) {
mov (16) uwDEST_Y(0,%1*16)<1> ubSRC_Y(0,%1*16)
}
#endif
$for (nUV_NUM_OF_ROWS/2-1; >-1; -1) {
mov (16) uwDEST_U(0,%1*16)<1> ubSRC_U(0,%1*32)<16;8,2>
mov (16) uwDEST_V(0,%1*16)<1> ubSRC_U(0,%1*32+1)<16;8,2>
}
// End of NV11_Load_5x8
i965_drv_video/shaders/post_processing/Common/NV12_Load_8x4.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: NV12_Load_8x4.asm
//----------------------------------------------------------------
#define NV12_LOAD_8x4
#include "PL2_Load.inc"
// Load 16x8 planar Y ----------------------------------------------------------
add (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w wSRC_H_ORI_OFFSET<2;2,1>:w // Source Y Block origin
#if !defined(LOAD_UV_ONLY)
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_Y:ud // Y block width and height (16x8)
mov (8) mMSGHDRY<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_Y(0)<1> mMSGHDRY udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_Y+nBI_CURRENT_SRC_Y:ud
#endif
// Load 8x4 planar U and V -----------------------------------------------------
asr (1) rMSGSRC.1<1>:d rMSGSRC.1<0;1,0>:d 1:w // U/V block origin should be half of Y's
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_UV:ud // U/V block width and height (16x4)
mov (8) mMSGHDRU<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_U(0)<1> mMSGHDRU udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_CURRENT_SRC_UV:ud
// Convert to word-aligned format ----------------------------------------------
#if !defined(LOAD_UV_ONLY)
$for (nY_NUM_OF_ROWS-1; >-1; -1) {
mov (16) uwDEST_Y(0,%1*16)<1> ubSRC_Y(0,%1*16)
}
#endif
$for (nUV_NUM_OF_ROWS/2-1; >-1; -1) {
mov (16) uwDEST_U(0,%1*16)<1> ubSRC_U(0,%1*32)<32;16,2>
mov (16) uwDEST_V(0,%1*16)<1> ubSRC_U(0,%1*32+1)<32;16,2>
}
// End of NV12_Load_8x4
i965_drv_video/shaders/post_processing/Common/NV12_Load_8x5.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: NV12_Load_8x5.asm
//----------------------------------------------------------------
#define NV12_LOAD_8x5
#include "PL2_Load.inc"
// Load 16x8 planar Y ----------------------------------------------------------
add (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w wSRC_H_ORI_OFFSET<2;2,1>:w // Source Y Block origin
#if !defined(LOAD_UV_ONLY)
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_Y:ud // Y block width and height (16x8)
mov (8) mMSGHDRY<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_Y(0)<1> mMSGHDRY udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_Y+nBI_CURRENT_SRC_Y:ud
#endif
// Load 8x5 planar U and V -----------------------------------------------------
asr (1) rMSGSRC.1<1>:d rMSGSRC.1<0;1,0>:d 1:w // U/V block origin should be half of Y's
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_UV:ud // U/V block width and height (16x5)
mov (8) mMSGHDRU<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_U(0)<1> mMSGHDRU udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_CURRENT_SRC_UV:ud
// Convert to word-aligned format ----------------------------------------------
#if !defined(LOAD_UV_ONLY)
$for (nY_NUM_OF_ROWS-1; >-1; -1) {
mov (16) uwDEST_Y(0,%1*16)<1> ubSRC_Y(0,%1*16)
}
#endif
$for (nUV_NUM_OF_ROWS/2-1; >-1; -1) {
mov (16) uwDEST_U(0,%1*16)<1> ubSRC_U(0,%1*32)<16;8,2>
mov (16) uwDEST_V(0,%1*16)<1> ubSRC_U(0,%1*32+1)<16;8,2>
}
// End of NV12_Load_8x5
i965_drv_video/shaders/post_processing/Common/NV12_Load_9x5.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: NV12_Load_9x5.asm
//----------------------------------------------------------------
#define NV12_LOAD_9x5
#include "PL2_Load.inc"
// Load 16x8 planar Y ----------------------------------------------------------
add (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w wSRC_H_ORI_OFFSET<2;2,1>:w // Source Y Block origin
#if !defined(LOAD_UV_ONLY)
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_Y:ud // Y block width and height (16x8)
mov (8) mMSGHDRY<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_Y(0)<1> mMSGHDRY udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_Y+nBI_CURRENT_SRC_Y:ud
#endif
// Load 9x5 planar U and V -----------------------------------------------------
asr (1) rMSGSRC.1<1>:d rMSGSRC.1<0;1,0>:d 1:w // U/V block origin should be half of Y's
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_UV:ud // U/V block width and height (20x5)
mov (8) mMSGHDRU<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_U(0)<1> mMSGHDRU udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_CURRENT_SRC_UV:ud
// Convert to word-aligned format ----------------------------------------------
#if !defined(LOAD_UV_ONLY)
$for (nY_NUM_OF_ROWS-1; >-1; -1) {
mov (16) uwDEST_Y(0,%1*16)<1> ubSRC_Y(0,%1*16)
}
#endif
$for(nUV_NUM_OF_ROWS-2; >-1; -1) {
mov (16) uwDEST_U(0,%1*16)<1> ubSRC_U(0,%1*32)<16;8,2>
mov (16) uwDEST_V(0,%1*16)<1> ubSRC_U(0,%1*32+1)<16;8,2>
}
// End of NV12_Load_9x5
i965_drv_video/shaders/post_processing/Common/P208_Load_8x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: P208_Load_8x8.asm
//----------------------------------------------------------------
#define P208_LOAD_8x8
#include "PL2_Load.inc"
add (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w wSRC_H_ORI_OFFSET<2;2,1>:w // Source Y Block origin
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_Y:ud // Y Block width and height (16x8) (U/V block size is the same)
// Load 16x8 P208 Y ------------------------------------------------------------
#if !defined(LOAD_UV_ONLY)
mov (8) mMSGHDRY<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_Y(0)<1> mMSGHDRY udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_Y+nBI_CURRENT_SRC_Y:ud
#endif
// Load 16x8 planar UV -----------------------------------------------------
mov (8) mMSGHDRU<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_U(0)<1> mMSGHDRU udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_CURRENT_SRC_UV:ud
// Convert to word-aligned format ----------------------------------------------
#if !defined(LOAD_UV_ONLY)
$for (0; <nY_NUM_OF_ROWS; 1) {
mov (16) uwDEST_Y(0,%1*16) ubSRC_Y(0,%1*16)
}
#endif
$for (0; <nUV_NUM_OF_ROWS/2; 1) {
mov (16) uwDEST_U(0,%1*16) ubSRC_U(0,%1*32)<32;16,2>
mov (16) uwDEST_V(0,%1*16) ubSRC_U(0,%1*32+1)<32;16,2>
}
// End of P208_Load_8x8.asm
i965_drv_video/shaders/post_processing/Common/P208_Load_9x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: P208_Load_9x8.asm
//----------------------------------------------------------------
#define P208_LOAD_9x8
#include "PL2_Load.inc"
add (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w wSRC_H_ORI_OFFSET<2;2,1>:w // Source Y Block origin
// Load 16x8 P208 Y ------------------------------------------------------------
#if !defined(LOAD_UV_ONLY)
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_Y:ud // Y block width and height (16x8)
mov (8) mMSGHDRY<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_Y(0)<1> mMSGHDRY udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_Y+nBI_CURRENT_SRC_Y:ud
#endif
// Load 16x8 planar UV -----------------------------------------------------
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_UV:ud // U/V block width and height (20x8)
mov (8) mMSGHDRU<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_U(0)<1> mMSGHDRU udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_CURRENT_SRC_UV:ud
// Convert to word-aligned format ----------------------------------------------
#if !defined(LOAD_UV_ONLY)
$for (0; <nY_NUM_OF_ROWS; 1) {
mov (16) uwDEST_Y(0,%1*16) ubSRC_Y(0,%1*16)
}
#endif
$for (0; <nUV_NUM_OF_ROWS; 1) {
mov (16) uwDEST_U(0,%1*16) ubSRC_U(0,%1*32)<32;16,2>
mov (16) uwDEST_V(0,%1*16) ubSRC_U(0,%1*32+1)<32;16,2>
}
// End of P208_Load_9x8.asm
i965_drv_video/shaders/post_processing/Common/PA_Load.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PA_Load.inc
//
// YUV422 data are first loaded to bottom I/O REGION_2, then unpacked to planar data
// and stored in top I/O REGION_1
#undef nY_NUM_OF_ROWS
#undef nUV_NUM_OF_ROWS
#define nY_NUM_OF_ROWS 8 // Number of Y rows per block
#define nUV_NUM_OF_ROWS 8 // Number of U/V rows per block
#if defined(PA_LOAD_8x8)
#define nDPR_BLOCK_SIZE_YUV nBLOCK_WIDTH_32+nBLOCK_HEIGHT_8 // Y block size 32x8
#define nDPR_MSG_SIZE_YUV nRESLEN_8 // # of MRF's to hold Y block data (8)
#endif
#if defined(PA_LOAD_9x8)
#define nDPR_BLOCK_SIZE_YUV_MAIN nBLOCK_WIDTH_32+nBLOCK_HEIGHT_8 // Main YUV block size 32x8
#define nDPR_MSG_SIZE_YUV_MAIN nRESLEN_8 // # of MRF's to hold Y block data (8)
#define nDPR_BLOCK_SIZE_YUV_ADDITION nBLOCK_WIDTH_4+nBLOCK_HEIGHT_8 // Additional YUV block size 4x8
#define nDPR_MSG_SIZE_YUV_ADDITION nRESLEN_1 // # of MRF's to hold Y block data (8)
#endif
#define udSRC_YUV udBOT_Y_IO
#define nSRC_YUV_REG nBOT_Y
#define uwDEST_Y uwTOP_Y
#define uwDEST_U uwTOP_U
#define uwDEST_V uwTOP_V
#define nSRC_REGION nREGION_1 // REGION_1 will be the source region for first kernel
// End of PA_Load.inc
i965_drv_video/shaders/post_processing/Common/PA_Load_8x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PA_Load_8x8.asm
//----------------------------------------------------------------
#define PA_LOAD_8x8
#include "PA_Load.inc"
// Load 16x8 packed data block
// Packed data block should be loaded as 32x8 pixel block
add (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w wSRC_H_ORI_OFFSET<2;2,1>:w // Source Block origin
shl (1) rMSGSRC.0<1>:d acc0:w 1:w // H. block origin need to be doubled
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_YUV:ud // Block width and height (32x8)
mov (8) mMSGHDRY<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_YUV(0)<1> mMSGHDRY udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_YUV+nBI_CURRENT_SRC_YUV:ud
// Unpack to "planar" YUV422 format in word-aligned bytes
add (4) pCF_Y_OFFSET<1>:uw ubSRC_CF_OFFSET<4;4,1>:ub nSRC_YUV_REG*nGRFWIB:w // Initial Y,U,V offset in YUV422 block
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (16) uwDEST_Y(0, %1*16)<1> r[pCF_Y_OFFSET, %1*nGRFWIB]REGION(16,2)
mov (8) uwDEST_U(0, %1*8)<1> r[pCF_U_OFFSET, %1*nGRFWIB]REGION(8,4)
mov (8) uwDEST_V(0, %1*8)<1> r[pCF_V_OFFSET, %1*nGRFWIB]REGION(8,4)
}
// End of PA_Load_8x8
i965_drv_video/shaders/post_processing/Common/PA_Load_9x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PA_Load_9x8.asm
//----------------------------------------------------------------
// This module loads 16x8 Y, 9x8 U and 9x8 V planar data blocks for CSC module
// and stores it in word-aligned format.
//----------------------------------------------------------------
#define PA_LOAD_9x8
#include "PA_Load.inc"
// Load 18x8 packed data block
// Packed data block should be loaded as 36x8 pixel block
add (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w wSRC_H_ORI_OFFSET<2;2,1>:w // Source Block origin
shl (1) rMSGSRC.0<1>:d acc0:w 1:w // H. block origin need to be doubled
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_YUV_MAIN:ud // Block width and height (32x8)
mov (8) mMSGHDRY<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_YUV(0)<1> mMSGHDRY udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_YUV_MAIN+nBI_CURRENT_SRC_YUV:ud
add (1) rMSGSRC.0<1>:d rMSGSRC.0:d 32:w //the last 4 pixels are read again for optimization
mov (1) rMSGSRC.2<1>:ud nDPR_BLOCK_SIZE_YUV_ADDITION:ud // Block width and height (4x8)
mov (8) mMSGHDRY<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_YUV(8)<1> mMSGHDRY udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_YUV_ADDITION+nBI_CURRENT_SRC_YUV:ud
// Unpack to "planar" YUV422 format in word-aligned bytes
add (4) pCF_Y_OFFSET<1>:uw ubSRC_CF_OFFSET<4;4,1>:ub nSRC_YUV_REG*nGRFWIB:w // Initial Y,U,V offset in YUV422 block
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (16) uwDEST_Y(0, %1*16)<1> r[pCF_Y_OFFSET, %1*nGRFWIB]REGION(16,2)
mov (8) uwDEST_U(0, %1*16)<1> r[pCF_U_OFFSET, %1*nGRFWIB]REGION(8,4)
mov (8) uwDEST_V(0, %1*16)<1> r[pCF_V_OFFSET, %1*nGRFWIB]REGION(8,4)
}
$for(0; <nUV_NUM_OF_ROWS; 1) {
mov (1) uwDEST_U(0, %1*16+8)<1> r[pCF_U_OFFSET, %1*4+256]REGION(1,0)
mov (1) uwDEST_V(0, %1*16+8)<1> r[pCF_V_OFFSET, %1*4+256]REGION(1,0)
}
//UV expansion done in PL9x8_PL16x8.asm module
// End of PA_Load_9x8
i965_drv_video/shaders/post_processing/Common/PL16x8_PL8x4.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL16x8_PL8x4.asm
//----------------------------------------------------------------
#include "common.inc"
#ifndef DEST_U //DEST_U, DEST_V not defined
#if (nSRC_REGION==nREGION_1)
#define DEST_Y uwTOP_Y
#define DEST_U uwTOP_U
#define DEST_V uwTOP_V
#elif (nSRC_REGION==nREGION_2)
#define DEST_Y uwBOT_Y
#define DEST_U uwBOT_U
#define DEST_V uwBOT_V
#endif
#endif
//Convert 444 from sampler to 422
$for (0, 0; <8; 2, 1) {
mov (8) DEST_U(0,%2*8)<1> DEST_U(%1)<16;8,2>
mov (8) DEST_V(0,%2*8)<1> DEST_V(%1)<16;8,2>
}
// Re-define new number of lines
#undef nUV_NUM_OF_ROWS
#define nUV_NUM_OF_ROWS 4
i965_drv_video/shaders/post_processing/Common/PL16x8_PL8x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL16x8_PL8x8.asm
//----------------------------------------------------------------
#include "common.inc"
#ifndef DEST_U
//DEST_U, DEST_V not defined
#if (nSRC_REGION==nREGION_1)
#define DEST_Y uwTOP_Y
#define DEST_U uwTOP_U
#define DEST_V uwTOP_V
#elif (nSRC_REGION==nREGION_2)
#define DEST_Y uwBOT_Y
#define DEST_U uwBOT_U
#define DEST_V uwBOT_V
#endif
#endif
//Convert 444 from sampler to 422
$for (0, 0; <8; 2, 1) {
mov DEST_U(%2)<1> DEST_U(%1)<16;8,2>
mov DEST_V(%2)<1> DEST_V(%1)<16;8,2>
}
i965_drv_video/shaders/post_processing/Common/PL2_Load.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL2_Load.inc
#undef nY_NUM_OF_ROWS
#undef nUV_NUM_OF_ROWS
#define nY_NUM_OF_ROWS 8 // Number of Y rows per block
#define nDPR_BLOCK_SIZE_Y nBLOCK_WIDTH_16+nBLOCK_HEIGHT_8 // Y block size 16x8
#define nDPR_MSG_SIZE_Y nRESLEN_4 // # of MRF's to hold Y block data (4)
#if defined(NV11_LOAD_4x8)
#define nUV_NUM_OF_ROWS 8 // Number of U/V rows per block
#define nDPR_BLOCK_SIZE_UV nBLOCK_WIDTH_8+nBLOCK_HEIGHT_8 // U/V block size 8x8
#define nDPR_MSG_SIZE_UV nRESLEN_2 // # of MRF's to hold U/V block data (2)
#endif
#if defined(NV11_LOAD_5x8)
#define nUV_NUM_OF_ROWS 8 // Number of U/V rows per block
#define nDPR_BLOCK_SIZE_UV nBLOCK_WIDTH_12+nBLOCK_HEIGHT_8 // U/V block size 12x8
#define nDPR_MSG_SIZE_UV nRESLEN_4 // # of MRF's to hold U/V block data (4)
#endif
#if defined(NV12_LOAD_8x4)
#define nUV_NUM_OF_ROWS 4 // Number of U/V rows per block
#define nDPR_BLOCK_SIZE_UV nBLOCK_WIDTH_16+nBLOCK_HEIGHT_4 // U/V block size 16x4
#define nDPR_MSG_SIZE_UV nRESLEN_2 // # of MRF's to hold U/V block data (2)
#endif
#if defined(NV12_LOAD_8x5)
#define nUV_NUM_OF_ROWS 6 // Number of U/V rows per block (Rounded Up to Nearest Even Number)
#define nDPR_BLOCK_SIZE_UV nBLOCK_WIDTH_16+nBLOCK_HEIGHT_5 // U/V block size 16x5
#define nDPR_MSG_SIZE_UV nRESLEN_3 // # of MRF's to hold U/V block data (3)
#endif
#if defined(NV12_LOAD_9x5)
#define nUV_NUM_OF_ROWS 6 // Number of U/V rows per block (Rounded Up to Nearest Even Number)
#define nDPR_BLOCK_SIZE_UV nBLOCK_WIDTH_20+nBLOCK_HEIGHT_5 // U/V block size 20x5
#define nDPR_MSG_SIZE_UV nRESLEN_5 // # of MRF's to hold U/V block data (5)
#endif
#if defined(P208_LOAD_8x8)
#define nUV_NUM_OF_ROWS 8 // Number of U/V rows per block
#define nDPR_BLOCK_SIZE_UV nBLOCK_WIDTH_16+nBLOCK_HEIGHT_8 // U/V block size 16x8
#define nDPR_MSG_SIZE_UV nRESLEN_4 // # of MRF's to hold U/V block data (4)
#endif
#if defined(P208_LOAD_9x8)
#define nUV_NUM_OF_ROWS 8 // Number of U/V rows per block
#define nDPR_BLOCK_SIZE_UV nBLOCK_WIDTH_20+nBLOCK_HEIGHT_8 // U/V block size 20x8
#define nDPR_MSG_SIZE_UV nRESLEN_8 // # of MRF's to hold U/V block data (8)
#endif
// Source/destination region definitions
#if !defined(udSRC_Y)
#define udSRC_Y udBOT_Y_IO // Default Y source region is top Y region
#endif
#if !defined(udSRC_U)
#define udSRC_U udBOT_U_IO // Default U source region is top U region
#endif
#define ubSRC_Y ubBOT_Y
#define nSRC_Y_REG nBOT_Y
#define ubSRC_U ubBOT_U
#define nSRC_U_REG nBOT_U
#define uwDEST_Y uwTOP_Y // However they can be transferred to word-aligned byte if desired
#define uwDEST_U uwTOP_U
#define uwDEST_V uwTOP_V
#define nSRC_REGION nREGION_1 // REGION_1 will be the source region for first kernel
// End of PL2_Load.inc
i965_drv_video/shaders/post_processing/Common/PL3_Load.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL3_Load.inc
#undef nY_NUM_OF_ROWS
#undef nUV_NUM_OF_ROWS
#define nY_NUM_OF_ROWS 8 // Number of Y rows per block
#define nDPR_BLOCK_SIZE_Y nBLOCK_WIDTH_16+nBLOCK_HEIGHT_8 // Y block size 16x8
#define nDPR_MSG_SIZE_Y nRESLEN_4 // # of MRF's to hold Y block data (4)
#if defined(IMC3_LOAD_8x4)
#define nUV_NUM_OF_ROWS 4 // Number of U/V rows per block
#define nDPR_BLOCK_SIZE_UV nBLOCK_WIDTH_8+nBLOCK_HEIGHT_4 // U/V block size 8x4
#define nDPR_MSG_SIZE_UV nRESLEN_1 // # of MRF's to hold U/V block data (1)
#endif
#if defined(IMC3_LOAD_8x5)
#define nUV_NUM_OF_ROWS 6 // Number of U/V rows per block (Rounded Up to Nearest Even Number)
#define nDPR_BLOCK_SIZE_UV nBLOCK_WIDTH_8+nBLOCK_HEIGHT_5 // U/V block size 8x5
#define nDPR_MSG_SIZE_UV nRESLEN_2 // # of MRF's to hold U/V block data (2)
#endif
#if defined(IMC3_LOAD_9x5)
#define nUV_NUM_OF_ROWS 6 // Number of U/V rows per block (Rounded Up to Nearest Even Number)
#define nDPR_BLOCK_SIZE_UV nBLOCK_WIDTH_12+nBLOCK_HEIGHT_5 // U/V block size 12x5
#define nDPR_MSG_SIZE_UV nRESLEN_3 // # of MRF's to hold U/V block data (3)
#endif
// Source/destination region definitions
#if !defined(udSRC_Y)
#define udSRC_Y udBOT_Y_IO // Default Y source region is top Y region
#endif
#if !defined(udSRC_U)
#define udSRC_U udBOT_U_IO // Default U source region is top U region
#endif
#if !defined(udSRC_V)
#define udSRC_V udBOT_V_IO // Default V source region is top V region
#endif
#define ubSRC_Y ubBOT_Y // Loading data are always in byte type
#define ubSRC_U ubBOT_U
#define ubSRC_V ubBOT_V
#define uwDEST_Y uwTOP_Y // However they can be transferred to word-aligned byte if desired
#define uwDEST_U uwTOP_U
#define uwDEST_V uwTOP_V
#define nSRC_REGION nREGION_1 // REGION_1 will be the source region for first kernel
// End of PL3_Load.inc
i965_drv_video/shaders/post_processing/Common/PL4x8_Save_NV11.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
#include "PL4x8_Save_NV11.inc"
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
#if !defined(SAVE_UV_ONLY)
// Save current planar frame Y block data (16x8) -------------------------------
mov (2) mMSGHDR.0<1>:d wORIX<2;2,1>:w // Block origin
mov (1) mMSGHDR.2<1>:ud nDPW_BLOCK_SIZE_Y:ud // Block width and height (16x8)
///* Yoni - masking is not relevant for ILK?!?
//
// //Use the mask to determine which pixels shouldn't be over-written
// cmp.ge.f0.0 (1) NULLREG BLOCK_MASK_D:ud 0x00FFFFFF:ud //Check if all pixels in the block need to be modified
// (f0.0) jmpi WritePlanarToDataPort
//
// //If mask is not all 1's, then load the entire 16x8 block
// //so that only those bytes may be modified that need to be (using the mask)
// send (8) SRC_YD(0)<1> MSGHDR MSGSRC<8;8,1>:ud DWBRMSGDSC+0x00040000+BI_DEST_Y:ud //16x8
//
// asr (2) MSGSRC.0<1>:ud ORIX<2;2,1>:w 1:w // U/V block origin should be half of Y's
// mov (1) MSGSRC.2<1>:ud 0x00030007:ud // Block width and height (8x4)
// send (8) SRC_UD(0)<1> MSGHDR MSGSRC<8;8,1>:ud DWBRMSGDSC+0x00010000+BI_DEST_U:ud
// send (8) SRC_VD(0)<1> MSGHDR MSGSRC<8;8,1>:ud DWBRMSGDSC+0x00010000+BI_DEST_V:ud
//
// //Restore the origin information
// mov (2) MSGSRC.0<1>:ud ORIX<2;2,1>:w // Block origin
// mov (1) MSGSRC.2<1>:ud 0x0007000F:ud // Block width and height (16x8)
//
// //expand U and V to be aligned on word boundary
// mov (16) SRC_UW(1)<1> SRC_U(0,16)
// mov (16) SRC_UW(0)<1> SRC_U(0, 0)
// mov (16) SRC_VW(1)<1> SRC_V(0,16)
// mov (16) SRC_VW(0)<1> SRC_V(0, 0)
//
// //Merge the data
// mov (1) f0.1:uw BLOCK_MASK_V:uw //Load the mask on flag reg
// (f0.1) mov (8) TEMP0<1>:uw BLOCK_MASK_H:uw
// (-f0.1) mov (8) TEMP0<1>:uw 0:uw
//
// // Destination is Word aligned
// $for(0; <Y_ROW_SIZE; 2) {
// mov (1) f0.1:uw TEMP(0,%1)<0;1,0>
// (-f0.1) mov (16) DEST_Y(0, %1*32)<2> SRC_Y(0, %1*16)
// (-f0.1) mov (16) DEST_U(0, %1*8)<1> SRC_U(0, %1*8) //only works for Word aligned Byte data
// (-f0.1) mov (16) DEST_V(0, %1*8)<1> SRC_V(0, %1*8) //only works for Word aligned Byte data
//
// mov (1) f0.1:uw TEMP(0,1+%1)<0;1,0>
// (-f0.1) mov (16) DEST_Y(0, 1+%1*32)<2> SRC_Y(0, 1+%1*16)
//
// }
//
//*/ Yoni - masking is not relevant for ILK?!?
WritePlanarToDataPort:
$for(0,0; <nY_NUM_OF_ROWS; 2,1) {
mov (16) mubMSGPAYLOAD(%2,0)<1> ub2DEST_Y(%1)REGION(16,2)
mov (16) mubMSGPAYLOAD(%2,16)<1> ub2DEST_Y(%1+1)REGION(16,2)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_Y+nBI_DESTINATION_Y:ud
#endif
// Save U/V data block in planar format (4x8) ----------------------------------
mov (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w // Block origin
asr (1) rMSGSRC.0<1>:d rMSGSRC.0<0;1,0>:d 1:w // U/V block origin should be half of Y's
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
$for(0,0; <nY_NUM_OF_ROWS;4,1) {
mov (16) mubMSGPAYLOAD(%2,0)<2> ub2DEST_U(%2)REGION(16,2)
mov (16) mubMSGPAYLOAD(%2,1)<2> ub2DEST_V(%2)REGION(16,2)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_UV+nBI_DESTINATION_UV:ud
// End of PL4x8_Save_NV11
i965_drv_video/shaders/post_processing/Common/PL4x8_Save_NV11.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//Module name: PL8x4_Save_NV11.inc
//
// Setup for storing planar data
//
#include "undefall.inc" //Undefine the SRC and DEST sysmbols
#define nDPW_BLOCK_SIZE_Y nBLOCK_WIDTH_16+nBLOCK_HEIGHT_8 // Y block size 16x8
#define nDPW_MSG_SIZE_Y nMSGLEN_4 // # of MRF's to hold Y block data (4)
#define nDPW_BLOCK_SIZE_UV nBLOCK_WIDTH_8+nBLOCK_HEIGHT_8 // U/V interleaved block width and height (8x8)
#define nDPW_MSG_SIZE_UV nMSGLEN_2 // # of MRF's to hold U/V block data (2)
#if (nSRC_REGION==nREGION_1)
#define udSRC_Y udBOT_Y_IO
#define udSRC_U udBOT_U_IO
#define udSRC_V udBOT_V_IO
#define ubSRC_Y ubBOT_Y
#define ubSRC_U ubBOT_U
#define ubSRC_V ubBOT_V
#define uwSRC_U uwBOT_U //For masking operation
#define uwSRC_V uwBOT_V
#define ub2DEST_Y ub2TOP_Y
#define ub2DEST_U ub2TOP_U
#define ub2DEST_V ub2TOP_V
#elif (nSRC_REGION==nREGION_2)
#define udSRC_Y udTOP_Y_IO
#define udSRC_U udTOP_U_IO
#define udSRC_V udTOP_V_IO
#define ubSRC_Y ubTOP_Y
#define ubSRC_U ubTOP_U
#define ubSRC_V ubTOP_V
#define uwSRC_U uwTOP_U //For masking operation
#define uwSRC_V uwTOP_V
#define ub2DEST_Y ub2BOT_Y
#define ub2DEST_U ub2BOT_U
#define ub2DEST_V ub2BOT_V
#endif
///* Yoni - masking is not relevant for ILK?!?
//#define TEMP0 REG(r,54)
//.declare TEMP Base=TEMP0 ElementSize=2 SrcRegion=<8;8,1> Type=uw
///* Yoni - masking is not relevant for ILK?!?
i965_drv_video/shaders/post_processing/Common/PL5x8_PL16x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL5x8_PL16x8.asm
#include "Expansion.inc"
//------------------------------ Horizontal Upconversion -----------------------------
$for (nUV_NUM_OF_ROWS/2-1; >-1; -1) {
avg.sat (16) uwDEST_U(0, %1*32+16) uwDEST_U(0, %1*16+7)<1;2,0> uwDEST_U(0, %1*16+7)<1;2,1>
avg.sat (16) uwDEST_V(0, %1*32+16) uwDEST_V(0, %1*16+7)<1;2,0> uwDEST_V(0, %1*16+7)<1;2,1>
avg.sat (16) uwDEST_U(0, %1*32) uwDEST_U(0, %1*16)<1;2,0> uwDEST_U(0, %1*16)<1;2,1>
avg.sat (16) uwDEST_V(0, %1*32) uwDEST_V(0, %1*16)<1;2,0> uwDEST_V(0, %1*16)<1;2,1>
}
$for (nUV_NUM_OF_ROWS/2-1; >-1; -1) {
avg.sat (16) uwDEST_U(0, %1*32+16) uwDEST_U(0, %1*32+18)<1;2,0> uwDEST_U(0, %1*32+18)<1;2,1>
avg.sat (16) uwDEST_V(0, %1*32+16) uwDEST_V(0, %1*32+18)<1;2,0> uwDEST_V(0, %1*32+18)<1;2,1>
avg.sat (16) uwDEST_U(0, %1*32) uwDEST_U(0, %1*32)<1;2,0> uwDEST_U(0, %1*32)<1;2,1>
avg.sat (16) uwDEST_V(0, %1*32) uwDEST_V(0, %1*32)<1;2,0> uwDEST_V(0, %1*32)<1;2,1>
}
// End of PL5x8_PL16x8
i965_drv_video/shaders/post_processing/Common/PL5x8_PL8x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL5x8_PL8x8.asm
#include "Expansion.inc"
//------------------------------ Horizontal Upconversion -----------------------------
$for (0; <nUV_NUM_OF_ROWS; 1) {
avg.sat (8) uwDEST_U(0, %1*8) uwDEST_U(0, %1*8)<1;2,0> uwDEST_U(0, %1*8)<1;2,1>
avg.sat (8) uwDEST_V(0, %1*8) uwDEST_V(0, %1*8)<1;2,0> uwDEST_V(0, %1*8)<1;2,1>
}
// End of PL5x8_PL8x8
i965_drv_video/shaders/post_processing/Common/PL8x4_Save_IMC3.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL8x4_Save_IMC3.asm
//
// Save planar YUV420 frame data block of size 16x8
#include "PL8x4_Save_IMC3.inc"
//Use the mask to determine which pixels shouldn't be over-written
and (1) acc0.0<1>:ud udBLOCK_MASK<0;1,0>:ud 0x00FFFFFF:ud
cmp.ge.f0.0(1) dNULLREG acc0.0<0;1,0>:ud 0x00FFFFFF:ud //Check if all pixels in the block need to be modified
(f0.0) jmpi WritePlanarToDataPort
//If mask is not all 1's, then load the entire 16x8 block
//so that only those bytes may be modified that need to be (using the mask)
// Load 16x8 planar Y ----------------------------------------------------------
mov (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w // Block origin
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_Y:ud // Block width and height (16x8)
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_Y(0)<1> mMSGHDR udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_Y+nBI_DESTINATION_Y:ud
// Load 8x4 planar U and V -----------------------------------------------------
asr (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w 1:w // U/V block origin should be half of Y's
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_UV:ud // Block width and height (8x4)
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_U(0)<1> mMSGHDR udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_DESTINATION_U:ud
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
send (8) udSRC_V(0)<1> mMSGHDR udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_DESTINATION_V:ud
//expand U and V to be aligned on word boundary - Y remains in bytes
$for (nUV_NUM_OF_ROWS/2-1; >-1; -1) {
mov (16) uwSRC_U(0, %1*16)<1> ubSRC_U(0, %1*16)
mov (16) uwSRC_V(0, %1*16)<1> ubSRC_V(0, %1*16)
}
//Merge the data
mov (1) f0.0:uw ubBLOCK_MASK_V:ub //Load the mask on flag reg
(f0.0) mov (8) rMASK_TEMP<1>:uw uwBLOCK_MASK_H:uw
(-f0.0) mov (8) rMASK_TEMP<1>:uw 0:uw
// Destination is Word aligned
$for(0; <nY_NUM_OF_ROWS; 2) {
mov (1) f0.1:uw uwMASK_TEMP(0,%1)<0;1,0>
(-f0.1) mov (16) ub2DEST_Y(0, %1*32)<2> ubSRC_Y(0, %1*16)
(-f0.1) mov (16) ub2DEST_U(0, %1*8)<1> ubSRC_U(0, %1*8) //only works for Word aligned Byte data
(-f0.1) mov (16) ub2DEST_V(0, %1*8)<1> ubSRC_V(0, %1*8) //only works for Word aligned Byte data
mov (1) f0.1:uw uwMASK_TEMP(0,1+%1)<0;1,0>
(-f0.1) mov (16) ub2DEST_Y(0, 1+%1*32)<2> ubSRC_Y(0, 1+%1*16)
}
WritePlanarToDataPort:
#if !defined(SAVE_UV_ONLY)
// Save current planar frame Y block data (16x8) -------------------------------
mov (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w // Block origin
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_Y:ud // Block width and height (16x8)
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
$for(0,0; <nY_NUM_OF_ROWS; 2,1) {
mov(16) mubMSGPAYLOAD(%2,0)<1> ub2DEST_Y(%1)REGION(16,2)
mov(16) mubMSGPAYLOAD(%2,16)<1> ub2DEST_Y(%1+1)REGION(16,2)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_Y+nBI_DESTINATION_Y:ud
#endif
// Save U/V data block in planar format (8x4) ----------------------------------
asr (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w 1:w // U/V block origin should be half of Y's
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_UV:ud // Block width and height (8x4)
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
// Save U picture data ---------------------------------------------------------
mov (16) mubMSGPAYLOAD(0,0)<1> ub2DEST_U(0)REGION(16,2) // U rows 0,1
mov (16) mubMSGPAYLOAD(0,16)<1> ub2DEST_U(1)REGION(16,2) // U rows 2,3
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_UV+nBI_DESTINATION_U:ud
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
// Save V picture data ---------------------------------------------------------
mov (16) mubMSGPAYLOAD(0,0)<1> ub2DEST_V(0)REGION(16,2) // V rows 0,1
mov (16) mubMSGPAYLOAD(0,16)<1> ub2DEST_V(1)REGION(16,2) // V rows 2,3
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_UV+nBI_DESTINATION_V:ud
// End of PL8x4_Save_IMC3
i965_drv_video/shaders/post_processing/Common/PL8x4_Save_IMC3.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL8x4_Save_IMC3.inc
//
// Setup for storing planar data
//
#include "undefall.inc" //Undefine the SRC and DEST sysmbols
// For saving
#define nDPW_BLOCK_SIZE_Y nBLOCK_WIDTH_16+nBLOCK_HEIGHT_8 // Y block size 16x8
#define nDPW_MSG_SIZE_Y nMSGLEN_4 // # of MRF's to hold Y block data (4)
#define nDPW_BLOCK_SIZE_UV nBLOCK_WIDTH_8+nBLOCK_HEIGHT_4 // U/V block size 8x4
#define nDPW_MSG_SIZE_UV nMSGLEN_1 // # of MRF's to hold U/V block data (1)
// For masking
#undef nDPR_MSG_SIZE_Y
#define nDPR_MSG_SIZE_Y nRESLEN_4 // # of MRF's to hold Y block data (4)
#undef nDPR_MSG_SIZE_UV
#define nDPR_MSG_SIZE_UV nRESLEN_1 // # of MRF's to hold U/V block data (1)
#define rMASK_TEMP REG(r,nTEMP0)
.declare uwMASK_TEMP Base=rMASK_TEMP ElementSize=2 SrcRegion=<8;8,1> Type=uw //1 GRF
#if (nSRC_REGION==nREGION_1)
// For saving
#define ub2DEST_Y ub2TOP_Y
#define ub2DEST_U ub2TOP_U
#define ub2DEST_V ub2TOP_V
//For masking operation
#define udSRC_Y udBOT_Y_IO
#define udSRC_U udBOT_U_IO
#define udSRC_V udBOT_V_IO
#define ubSRC_Y ubBOT_Y
#define ubSRC_U ubBOT_U
#define ubSRC_V ubBOT_V
#define uwSRC_U uwBOT_U
#define uwSRC_V uwBOT_V
#elif (nSRC_REGION==nREGION_2)
// For saving
#define ub2DEST_Y ub2BOT_Y
#define ub2DEST_U ub2BOT_U
#define ub2DEST_V ub2BOT_V
//For masking operation
#define udSRC_Y udTOP_Y_IO
#define udSRC_U udTOP_U_IO
#define udSRC_V udTOP_V_IO
#define ubSRC_Y ubTOP_Y
#define ubSRC_U ubTOP_U
#define ubSRC_V ubTOP_V
#define uwSRC_U uwTOP_U
#define uwSRC_V uwTOP_V
#endif
i965_drv_video/shaders/post_processing/Common/PL8x4_Save_NV12.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL8x4_Save_NV12.asm
//
// Save entire current planar frame data block of size 16x8
//---------------------------------------------------------------
// Symbols needed to be defined before including this module
//
// DWORD_ALIGNED_DEST: only if DEST_Y, DEST_U, DEST_V data are DWord aligned
// ORIX:
//---------------------------------------------------------------
#include "PL8x4_Save_NV12.inc"
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
#if !defined(SAVE_UV_ONLY)
// Save current planar frame Y block data (16x8) -------------------------------
mov (2) mMSGHDR.0<1>:d wORIX<2;2,1>:w // Block origin
mov (1) mMSGHDR.2<1>:ud nDPW_BLOCK_SIZE_Y:ud // Block width and height (16x8)
#endif
//Use the mask to determine which pixels shouldn't be over-written
and (1) acc0<1>:ud udBLOCK_MASK<0;1,0>:ud 0x00FFFFFF:ud
cmp.ge.f0.0 (1) dNULLREG acc0<0;1,0>:ud 0x00FFFFFF:ud //Check if all pixels in the block need to be modified
(f0.0) jmpi WritePlanarToDataPort
//If mask is not all 1's, then load the entire 16x8 block
//so that only those bytes may be modified that need to be (using the mask)
send (8) udSRC_Y(0)<1> mMSGHDR udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_Y+nBI_DESTINATION_Y:ud //16x8
asr (1) rMSGSRC.1<1>:ud wORIY<0;1,0>:w 1:w { NoDDClr } // U/V block origin should be half of Y's
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_UV:ud { NoDDChk } // Block width and height (16x4)
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud //move message desrcptor to the message header
send (8) udSRC_U(0)<1> mMSGHDR udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_UV+nBI_DESTINATION_UV:ud
//Restore the origin information
mov (2) rMSGSRC.0<1>:ud wORIX<2;2,1>:w // Block origin
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_Y:ud // Block width and height (16x8)
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud //move message desrcptor to the message header
//Merge the data
mov (1) f0.1:uw ubBLOCK_MASK_V:ub //Load the mask on flag reg
(f0.1) mov (8) rMASK_TEMP<1>:uw uwBLOCK_MASK_H:uw
(-f0.1) mov (8) rMASK_TEMP<1>:uw 0:uw
//convert the mask from 16bits to 8bits by selecting every other bit
mov (1) udMASK_TEMP1(0,0)<1> 0x00040001:ud
mov (1) udMASK_TEMP1(0,1)<1> 0x00400010:ud
mov (1) udMASK_TEMP1(0,2)<1> 0x04000100:ud
mov (1) udMASK_TEMP1(0,3)<1> 0x40001000:ud
//merge the loaded block with the current block
$for(0,0; <nY_NUM_OF_ROWS; 2,1) {
mov (1) f0.1:uw uwMASK_TEMP(0, %1)<0;1,0>
(-f0.1) mov (16) ubDEST_Y(0,%1*32)<2> ubSRC_Y(0,%1*16)
and.nz.f0.1 (8) wNULLREG uwMASK_TEMP(0,%1)<0;1,0> uwMASK_TEMP1(0,0) //change the mask by selecting every other bit
(-f0.1) mov (8) ubDEST_U(0, %2*16)<2> ub2SRC_U(0, %1*8)<16;8,2>
(-f0.1) mov (8) ubDEST_V(0, %2*16)<2> ub2SRC_U(0, %1*8+1)<16;8,2>
mov (1) f0.1:uw uwMASK_TEMP(0,1+%1)<0;1,0>
(-f0.1) mov (16) ubDEST_Y(0, (1+%1)*32)<2> ubSRC_Y(0, (1+%1)*16)
}
WritePlanarToDataPort:
#if !defined(SAVE_UV_ONLY)
$for(0,0; <nY_NUM_OF_ROWS; 2,1) {
mov (16) mubMSGPAYLOAD(%2,0)<1> ub2DEST_Y(%1)REGION(16,2)
mov (16) mubMSGPAYLOAD(%2,16)<1> ub2DEST_Y(%1+1)REGION(16,2)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_Y+nBI_DESTINATION_Y:ud
#endif
//** Save 8x4 packed U and V -----------------------------------------------------
// we could write directly wORIX to mMSGHDR and then execute asr on it, that way we could
// avoid using rMSGSRC as a buffer and have one command less in code, but it is unknown whether
//it is possible to do asr on mMSGHDR so we use rMSGSRC.
mov (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w // Block origin
asr (1) rMSGSRC.1<1>:d rMSGSRC.1<0;1,0>:d 1:w // U/V block origin should be half of Y's
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_UV:ud // U/V block width and height (16x4)
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
$for(0,0; <nY_NUM_OF_ROWS;4,1) {
mov (16) mubMSGPAYLOAD(%2,0)<2> ub2DEST_U(%2)REGION(16,2)
mov (16) mubMSGPAYLOAD(%2,1)<2> ub2DEST_V(%2)REGION(16,2)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_UV+nBI_DESTINATION_UV:ud
// End of PL8x4_Save_NV12
i965_drv_video/shaders/post_processing/Common/PL8x4_Save_NV12.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//Module name: PL8x4_Save_NV12.inc
//
// Setup for storing planar data
//
#include "undefall.inc" //Undefine the SRC and DEST sysmbols
#undef nDPW_BLOCK_SIZE_Y
#undef nDPW_MSG_SIZE_Y
#undef nDPW_BLOCK_SIZE_UV
#undef nDPW_MSG_SIZE_UV
#define nDPW_BLOCK_SIZE_Y nBLOCK_WIDTH_16+nBLOCK_HEIGHT_8 // Y block size 16x8
#define nDPW_MSG_SIZE_Y nMSGLEN_4 // # of MRF's to hold Y block data (4)
#define nDPW_BLOCK_SIZE_UV nBLOCK_WIDTH_16+nBLOCK_HEIGHT_4 // U/V interleaved block width and height (16x4)
#define nDPW_MSG_SIZE_UV nMSGLEN_2 // # of MRF's to hold U/V block data (2)
// For masking
#undef nDPR_MSG_SIZE_Y
#define nDPR_MSG_SIZE_Y nRESLEN_4 // # of MRF's to hold Y block data (4)
#undef nDPR_MSG_SIZE_UV
#define nDPR_MSG_SIZE_UV nRESLEN_2
#define rMASK_TEMP REG(r,nTEMP0)
.declare uwMASK_TEMP Base=rMASK_TEMP ElementSize=2 SrcRegion=<8;8,1> Type=uw //1 GRF
#define rMASK_TEMP1 REG(r,nTEMP1)
.declare udMASK_TEMP1 Base=rMASK_TEMP1 ElementSize=4 SrcRegion=<4;4,1> Type=ud //1 GRF
.declare uwMASK_TEMP1 Base=rMASK_TEMP1 ElementSize=2 SrcRegion=<8;8,1> Type=uw //1 GRF
#if (nSRC_REGION==nREGION_1)
#define udSRC_Y udBOT_Y_IO
#define udSRC_U udBOT_U_IO
#define udSRC_V udBOT_V_IO
#define ubSRC_Y ubBOT_Y
#define ubSRC_U ubBOT_U
#define ubSRC_V ubBOT_V
#define uwSRC_U uwBOT_U //For masking operation
#define uwSRC_V uwBOT_V
#define ub2DEST_Y ub2TOP_Y
#define ub2DEST_U ub2TOP_U
#define ub2DEST_V ub2TOP_V
#define ubDEST_Y ubTOP_Y
#define ubDEST_U ubTOP_U
#define ubDEST_V ubTOP_V
#define ub2SRC_U ub2BOT_U
#elif (nSRC_REGION==nREGION_2)
#define udSRC_Y udTOP_Y_IO
#define udSRC_U udTOP_U_IO
#define udSRC_V udTOP_V_IO
#define ubSRC_Y ubTOP_Y
#define ubSRC_U ubTOP_U
#define ubSRC_V ubTOP_V
#define uwSRC_U uwTOP_U //For masking operation
#define uwSRC_V uwTOP_V
#define ub2DEST_Y ub2BOT_Y
#define ub2DEST_U ub2BOT_U
#define ub2DEST_V ub2BOT_V
#define ubDEST_Y ubBOT_Y
#define ubDEST_U ubBOT_U
#define ubDEST_V ubBOT_V
#define ub2SRC_U ub2TOP_U
#endif
///* Yoni - masking is not relevant for ILK?!?
//#define TEMP0 REG(r,54)
//.declare TEMP Base=TEMP0 ElementSize=2 SrcRegion=<8;8,1> Type=uw
///* Yoni - masking is not relevant for ILK?!?
i965_drv_video/shaders/post_processing/Common/PL8x5_PL8x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL8x5_PL8x8.asm
#include "Expansion.inc"
//------------------------------- Vertical Upconversion ------------------------------
avg.sat (8) uwDEST_U(0, 3*16+8)<1> uwDEST_U(0, 3*8) uwDEST_U(0, (1+3)*8) // Optimization
avg.sat (8) uwDEST_V(0, 3*16+8)<1> uwDEST_V(0, 3*8) uwDEST_V(0, (1+3)*8) // Optimization
$for(nUV_NUM_OF_ROWS/2-2; >-1; -1) {
mov (8) uwDEST_U(0, (1+%1)*16)<1> uwDEST_U(0, (1+%1)*8)
avg.sat (8) uwDEST_U(0, %1*16+8)<1> uwDEST_U(0, %1*8) uwDEST_U(0, (1+%1)*8)
mov (8) uwDEST_V(0, (1+%1)*16)<1> uwDEST_V(0, (1+%1)*8)
avg.sat (8) uwDEST_V(0, %1*16+8)<1> uwDEST_V(0, %1*8) uwDEST_V(0, (1+%1)*8)
}
// End of PL8x5_PL8x8
i965_drv_video/shaders/post_processing/Common/PL8x8_PL8x4.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL8x8_PL8x4.asm
//
// Convert PL 8x8 to PL8x4 in GRF
//---------------------------------------------------------------
// Symbols needed to be defined before including this module
//
// DWORD_ALIGNED_DEST: only if DEST_Y, DEST_U, DEST_V data are DWord aligned
// ORIX:
//---------------------------------------------------------------
#include "PL8x8_PL8x4.inc"
// Convert PL8x8 to PL8x4 ---------------------------------------------------------
mov (8) ubDEST_U(0,16)<2> ubDEST_U(1)<16;8,2> //selecting U every other row
mov (16) ubDEST_U(0,32)<2> ubDEST_U(2)<32;8,2> //selecting U every other row
mov (8) ubDEST_V(0,16)<2> ubDEST_V(1)<16;8,2> //selecting V every other row
mov (16) ubDEST_V(0,32)<2> ubDEST_V(2)<32;8,2> //selecting V every other row
// End of PL8x8_PL8x4.asm -------------------------------------------------------
\ No newline at end of file
i965_drv_video/shaders/post_processing/Common/PL8x8_PL8x4.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL8x8_PL8x4.inc
//
// Setup module for convert PL8x8 to PL8x4
//
//
// Source/destination region definitions
//
#include "undefall.inc" //Undefine the SRC and DEST sysmbols
#if (nSRC_REGION==nREGION_1)
//REGION_1 selected
#define ubDEST_Y ubTOP_Y
#define ubDEST_U ubTOP_U
#define ubDEST_V ubTOP_V
#elif (nSRC_REGION==nREGION_2)
//REGION_2 selected
#define ubDEST_Y ubBOT_Y
#define ubDEST_U ubBOT_U
#define ubDEST_V ubBOT_V
#endif
i965_drv_video/shaders/post_processing/Common/PL8x8_Save_P208.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL8x8_Save_P208.asm
//
// Save entire current planar frame data block of size 16x8
//---------------------------------------------------------------
// Symbols needed to be defined before including this module
//
// DWORD_ALIGNED_DEST: only if DEST_Y, DEST_U, DEST_V data are DWord aligned
// ORIX:
//---------------------------------------------------------------
#include "PL8x8_Save_P208.inc"
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
#if !defined(SAVE_UV_ONLY)
// Save current planar frame Y block data (16x8) -------------------------------
mov (2) mMSGHDR.0<1>:d wORIX<2;2,1>:w // Block origin
mov (1) mMSGHDR.2<1>:ud nDPW_BLOCK_SIZE_Y:ud // Block width and height (16x8)
WritePlanarToDataPort:
$for(0,0; <nY_NUM_OF_ROWS; 2,1) {
mov (16) mubMSGPAYLOAD(%2,0)<1> ub2DEST_Y(%1)REGION(16,2)
mov (16) mubMSGPAYLOAD(%2,16)<1> ub2DEST_Y(%1+1)REGION(16,2)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_Y+nBI_DESTINATION_Y:ud
#endif
//** Save 8x8 packed U and V -----------------------------------------------------
// we could write directly wORIX to mMSGHDR and then execute asr on it, that way we could
// avoid using rMSGSRC as a buffer and have one command less in code, but it is unknown whether
//it is possible to do asr on mMSGHDR so we use rMSGSRC.
mov (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w // Block origin
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_UV:ud // U/V block width and height (16x4)
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
$for(0,0; <nY_NUM_OF_ROWS;2,1) {
mov (16) mubMSGPAYLOAD(%2,0)<2> ub2DEST_U(%2)REGION(16,2)
mov (16) mubMSGPAYLOAD(%2,1)<2> ub2DEST_V(%2)REGION(16,2)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_UV+nBI_DESTINATION_UV:ud
//End of PL8x8_Save_P208.asm
i965_drv_video/shaders/post_processing/Common/PL8x8_Save_P208.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//Module name: PL8x8_Save_P208.inc
//
// Setup for storing planar data
//
#include "undefall.inc" //Undefine the SRC and DEST sysmbols
#define nDPW_BLOCK_SIZE_Y nBLOCK_WIDTH_16+nBLOCK_HEIGHT_8 // Y block size 16x8
#define nDPW_MSG_SIZE_Y nMSGLEN_4 // # of MRF's to hold Y block data (4)
#define nDPW_BLOCK_SIZE_UV nBLOCK_WIDTH_16+nBLOCK_HEIGHT_8 // U/V interleaved block width and height (16x8)
#define nDPW_MSG_SIZE_UV nMSGLEN_4 // # of MRF's to hold U/V block data (4)
#if (nSRC_REGION==nREGION_1)
#define udSRC_Y udBOT_Y_IO
#define udSRC_U udBOT_U_IO
#define udSRC_V udBOT_V_IO
#define ubSRC_Y ubBOT_Y
#define ubSRC_U ubBOT_U
#define ubSRC_V ubBOT_V
#define uwSRC_U uwBOT_U //For masking operation
#define uwSRC_V uwBOT_V
#define ub2DEST_Y ub2TOP_Y
#define ub2DEST_U ub2TOP_U
#define ub2DEST_V ub2TOP_V
#elif (nSRC_REGION==nREGION_2)
#define udSRC_Y udTOP_Y_IO
#define udSRC_U udTOP_U_IO
#define udSRC_V udTOP_V_IO
#define ubSRC_Y ubTOP_Y
#define ubSRC_U ubTOP_U
#define ubSRC_V ubTOP_V
#define uwSRC_U uwTOP_U //For masking operation
#define uwSRC_V uwTOP_V
#define ub2DEST_Y ub2BOT_Y
#define ub2DEST_U ub2BOT_U
#define ub2DEST_V ub2BOT_V
#endif
///* Yoni - masking is not relevant for ILK?!?
//#define TEMP0 REG(r,54)
//.declare TEMP Base=TEMP0 ElementSize=2 SrcRegion=<8;8,1> Type=uw
///* Yoni - masking is not relevant for ILK?!?
i965_drv_video/shaders/post_processing/Common/PL8x8_Save_PA.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL8x8_Save_PA.asm
//
// Save planar YUV422 to packed YUV422 format data
//
// Note: SRC_* must reference to regions with data type "BYTE"
// in order to save to byte-aligned byte location
#include "PL8x8_Save_PA.inc"
add (4) pCF_Y_OFFSET<1>:uw ubDEST_CF_OFFSET<4;4,1>:ub nDEST_YUV_REG*nGRFWIB:w // Initial Y,U,V offset in YUV422 block
// Pack Y
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (16) r[pCF_Y_OFFSET, %1*nGRFWIB]<2> ubSRC_Y(0,%1*32)
}
// Pack U/V
$for(0; <nUV_NUM_OF_ROWS; 1) {
mov (8) r[pCF_U_OFFSET, %1*nGRFWIB]<4> ubSRC_U(0, %1*16)
mov (8) r[pCF_V_OFFSET, %1*nGRFWIB]<4> ubSRC_V(0, %1*16)
}
shl (1) rMSGSRC.0<1>:d wORIX<0;1,0>:w 1:w { NoDDClr } // H. block origin need to be doubled
mov (1) rMSGSRC.1<1>:d wORIY<0;1,0>:w { NoDDClr, NoDDChk } // Block origin
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_YUV:ud { NoDDChk } // Block width and height (32x8)
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
//Use the mask to determine which pixels shouldn't be over-written
and (1) acc0.0<1>:ud udBLOCK_MASK<0;1,0>:ud 0x00FFFFFF:ud
cmp.ge.f0.0(1) dNULLREG acc0.0<0;1,0>:ud 0x00FFFFFF:ud //Check if all pixels in the block need to be modified
(f0.0) jmpi WritePackedToDataPort
//If mask is not all 1's, then load the entire 32x8 block
//so that only those bytes may be modified that need to be (using the mask)
// Load 32x8 packed YUV 422 ----------------------------------------------------
send (8) udSRC_YUV(0)<1> mMSGHDR udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_YUV+nBI_DESTINATION_YUV:ud
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
//Merge the data
mov (1) f0.0:uw ubBLOCK_MASK_V:ub //Load the mask on flag reg
(f0.0) mov (8) rMASK_TEMP<1>:uw uwBLOCK_MASK_H:uw
(-f0.0) mov (8) rMASK_TEMP<1>:uw 0:uw
// Destination is Byte aligned
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (1) f0.1:uw uwMASK_TEMP(0,%1)<0;1,0>
(-f0.1) mov (16) uwDEST_YUV(%1)<1> uwSRC_YUV(%1) //check the UV merge - vK
}
WritePackedToDataPort:
// Packed YUV data are stored in one of the I/O regions before moving to MRF
// Note: This is necessary since indirect addressing is not supported for MRF.
// Packed data block should be saved as 32x8 pixel block
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (8) mudMSGPAYLOAD(%1)<1> udDEST_YUV(%1)REGION(8,1)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_YUV+nBI_DESTINATION_YUV:ud
// End of PL8x8_Save_PA
i965_drv_video/shaders/post_processing/Common/PL8x8_Save_PA.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL8x8_Save_PA.inc
//
// Setup for storing packed data
//
#include "undefall.inc" //Undefine the SRC and DEST sysmbols
// For saving
#define nDPW_BLOCK_SIZE_YUV nBLOCK_WIDTH_32+nBLOCK_HEIGHT_8 // YUV block size 32x8
#define nDPW_MSG_SIZE_YUV nMSGLEN_8 // # of MRF's to hold YUV block data (8)
// For masking
#undef nDPR_MSG_SIZE_YUV
#define nDPR_MSG_SIZE_YUV nRESLEN_8 // # of MRF's to hold YUV block data (8)
#define rMASK_TEMP REG(r,nTEMP0)
.declare uwMASK_TEMP Base=rMASK_TEMP ElementSize=2 SrcRegion=<8;8,1> Type=uw //1 GRF
#if (nSRC_REGION==nREGION_1)
// For saving
#define udSRC_YUV udTOP_Y_IO
#define udDEST_YUV udBOT_Y_IO
#define nDEST_YUV_REG nBOT_Y
//For masking operation
#define ubSRC_Y ub2TOP_Y
#define ubSRC_U ub2TOP_U
#define ubSRC_V ub2TOP_V
#define uwSRC_YUV uwTOP_Y
#define uwDEST_YUV uwBOT_Y
#elif (nSRC_REGION==nREGION_2)
// For saving
#define udSRC_YUV udBOT_Y_IO
#define udDEST_YUV udTOP_Y_IO
#define nDEST_YUV_REG nTOP_Y
//For masking operation
#define ubSRC_Y ub2BOT_Y
#define ubSRC_U ub2BOT_U
#define ubSRC_V ub2BOT_V
#define uwSRC_YUV uwBOT_Y
#define uwDEST_YUV uwTOP_Y
#endif
i965_drv_video/shaders/post_processing/Common/PL9x5_PL16x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL9x5_PL16x8.asm
#define EXPAND_9x5
#include "Expansion.inc"
//------------------------------ Horizontal Upconversion -----------------------------
$for (nUV_NUM_OF_ROWS-2; >-1; -1) {
avg.sat (16) uwDEST_U(0, %1*16)<1> uwDEST_U(0, %1*16)<1;2,0> uwDEST_U(0, %1*16)<1;2,1>
avg.sat (16) uwDEST_V(0, %1*16)<1> uwDEST_V(0, %1*16)<1;2,0> uwDEST_V(0, %1*16)<1;2,1>
}
#undef nUV_NUM_OF_ROWS
#define nUV_NUM_OF_ROWS 8 //use packed version of all post-processing kernels
//------------------------------- Vertical Upconversion ------------------------------
avg.sat (16) uwDEST_U(0, 3*32+16)<1> uwDEST_U(0, 3*16) uwDEST_U(0, (1+3)*16)
avg.sat (16) uwDEST_V(0, 3*32+16)<1> uwDEST_V(0, 3*16) uwDEST_V(0, (1+3)*16)
$for(nUV_NUM_OF_ROWS/2-2; >-1; -1) {
mov (16) uwDEST_U(0, (1+%1)*32)<1> uwDEST_U(0, (1+%1)*16)
avg.sat (16) uwDEST_U(0, %1*32+16)<1> uwDEST_U(0, %1*16) uwDEST_U(0, (1+%1)*16)
mov (16) uwDEST_V(0, (1+%1)*32)<1> uwDEST_V(0, (1+%1)*16)
avg.sat (16) uwDEST_V(0, %1*32+16)<1> uwDEST_V(0, %1*16) uwDEST_V(0, (1+%1)*16)
}
// End of PL9x5_PL16x8
i965_drv_video/shaders/post_processing/Common/PL9x8_PL16x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: PL9x5_PL16x8.asm
#include "Expansion.inc"
//------------------------------ Horizontal Upconversion -----------------------------
$for (0; <nUV_NUM_OF_ROWS; 1) {
avg.sat (16) uwDEST_U(0, %1*16)<1> uwDEST_U(0, %1*16)<1;2,0> uwDEST_U(0, %1*16)<1;2,1>
avg.sat (16) uwDEST_V(0, %1*16)<1> uwDEST_V(0, %1*16)<1;2,0> uwDEST_V(0, %1*16)<1;2,1>
}
// End of PL9x5_PL16x8
\ No newline at end of file
i965_drv_video/shaders/post_processing/Common/RGB16x8_Save_RGB.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: RGB16x8_Save_RGB.asm
//
// Save packed ARGB 444 frame data block of size 16x8
//
// To save 16x8 block (64x8 byte layout for ARGB8888) we need 2 send instructions
// ---------
// | 1 | 2 |
// ---------
#include "RGB16x8_Save_RGB.inc"
shl (1) rMSGSRC.0<1>:d wORIX<0;1,0>:w 2:w { NoDDClr } // H. block origin need to be quadrupled
mov (1) rMSGSRC.1<1>:d wORIY<0;1,0>:w { NoDDClr, NoDDChk } // Block origin (1st quadrant)
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_ARGB:ud { NoDDChk } // Block width and height (32x8)
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
//Use the mask to determine which pixels shouldn't be over-written
and (1) acc0.0<1>:ud udBLOCK_MASK<0;1,0>:ud 0x00FFFFFF:ud
cmp.ge.f0.0(1) dNULLREG acc0.0<0;1,0>:ud 0x00FFFFFF:ud //Check if all pixels in the block need to be modified
(f0.0) jmpi WriteARGBToDataPort
//If mask is not all 1's, then load the entire 64x8 block
//so that only those bytes may be modified that need to be (using the mask)
// Load first block 16x8 packed ARGB 444 ---------------------------------------
or (1) acc0.0<1>:ud udBLOCK_MASK<0;1,0>:ud 0xFF00FF00:ud //Check first block
cmp.e.f0.0 (1) dNULLREG acc0.0<0;1,0>:ud 0xFFFFFFFF:ud
(f0.0) jmpi SkipFirstBlockMerge //If full mask then skip this block
send (8) udSRC_ARGB(0)<1> mMSGHDR udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_ARGB+nBI_DESTINATION_RGB:ud
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
//Merge the data
mov (1) f0.0:uw ubBLOCK_MASK_V:ub //Load the mask on flag reg
(f0.0) mov (8) rMASK_TEMP<1>:uw uwBLOCK_MASK_H:uw //use sel instruction - vK
(-f0.0) mov (8) rMASK_TEMP<1>:uw 0:uw
$for(0, 0; <nY_NUM_OF_ROWS; 1, 2) { //take care of the lines in the block, they are different in the src and dest
mov (1) f0.1:uw uwMASK_TEMP(0,%1)<0;1,0>
(-f0.1) mov (8) udDEST_ARGB(%2)<1> udSRC_ARGB(%1)
}
SkipFirstBlockMerge:
// Load second block 16x8 packed ARGB 444 ---------------------------------------
or (1) acc0.0<1>:ud udBLOCK_MASK<0;1,0>:ud 0xFF0000FF:ud //Check second block
cmp.e.f0.0 (1) dNULLREG acc0.0<0;1,0>:ud 0xFFFFFFFF:ud
(f0.0) jmpi WriteARGBToDataPort //If full mask then skip this block
add (1) mMSGHDR.0<1>:d rMSGSRC.0<0;1,0>:d 32:d // Point to 2nd part
send (8) udSRC_ARGB(0)<1> mMSGHDR udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_ARGB+nBI_DESTINATION_RGB:ud
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud // Point to 1st part again
//Merge the data
mov (1) f0.0:uw ubBLOCK_MASK_V:ub //Load the mask on flag reg
(f0.0) shr (8) rMASK_TEMP<1>:uw uwBLOCK_MASK_H:uw 8:uw //load the mask for second block
(-f0.0) mov (8) rMASK_TEMP<1>:uw 0:uw
$for(0, 1; <nY_NUM_OF_ROWS; 1, 2) { //take care of the lines in the block, they are different in the src and dest
mov (1) f0.1:uw uwMASK_TEMP(0,%1)<0;1,0>
(-f0.1) mov (8) udDEST_ARGB(%2)<1> udSRC_ARGB(%1)
}
WriteARGBToDataPort:
// Move packed data to MRF and output
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (8) mudMSGPAYLOAD(%1)<1> udDEST_ARGB(%1*2)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_ARGB+nBI_DESTINATION_RGB:ud
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
add (1) mMSGHDR.0<1>:d rMSGSRC.0<0;1,0>:d 32:d // Point to 2nd part
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (8) mudMSGPAYLOAD(%1)<1> udDEST_ARGB(%1*2+1)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_ARGB+nBI_DESTINATION_RGB:ud
// End of RGB16x8_Save_RGB
i965_drv_video/shaders/post_processing/Common/RGB16x8_Save_RGB.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: RGB16x8_Save_RGB.inc
//
#include "undefall.inc" //Undefine the SRC and DEST sysmbols
// For saving
#define nDPW_BLOCK_SIZE_ARGB nBLOCK_WIDTH_32+nBLOCK_HEIGHT_8 // ARGB block size 32x8
#define nDPW_MSG_SIZE_ARGB nMSGLEN_8 // # of MRF's to hold ARGB block data (8)
// For masking
#undef nDPR_MSG_SIZE_ARGB
#define nDPR_MSG_SIZE_ARGB nRESLEN_8 // # of MRF's to hold ARGB block data (8)
#define rMASK_TEMP REG(r,nTEMP0)
.declare uwMASK_TEMP Base=rMASK_TEMP ElementSize=2 SrcRegion=<8;8,1> Type=uw //1 GRF
#if (nSRC_REGION==nREGION_1)
// For saving
#define udDEST_ARGB udTOP_Y_IO //The output of previous stage is stored here; This is modified and is written to render cache
//For masking operation
#define udSRC_ARGB udBOT_Y_IO //To hold the destination data that shouldn't be modified
#elif (nSRC_REGION==nREGION_2)
// For saving
#define udDEST_ARGB udBOT_Y_IO //The output of previous stage is stored here; This is modified and is written to render cache
//For masking operation
#define udSRC_ARGB udTOP_Y_IO //To hold the destination data that shouldn't be modified
#endif
i965_drv_video/shaders/post_processing/Common/RGB16x8_Save_RGB16.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: RGB16x8_Save_RGB16.asm
//
// Save packed RGB565 frame data block of size 16x8
//
// To save 16x8 block (32x8 byte layout for RGB565) we need 1 send instruction
// -----
// | 1 |
// -----
#include "RGB16x8_Save_RGB16.inc"
//convert 32 bit RGB to 16 bit RGB
// Truncate A8R8G8B8 to A6R5G6B5 within byte.
// That is keeping 5 MSB of R and B, and 6 MSB of G.
$for (0, 0; <nY_NUM_OF_ROWS; 1, 2) {
shr uwCSC_TEMP(%1,0)<1> ubDEST_ARGB(%2,0)<32;8,4> 3:w // B >> 3
shl (16) uwTEMP_RGB16(0)<1> uwDEST_ARGB(%2,1)<16;8,2> 8:w // R << 8
and (16) uwTEMP_RGB16(0)<1> uwTEMP_RGB16(0) 0xF800:uw
or (16) uwCSC_TEMP(%1,0)<1> uwCSC_TEMP(%1,0)<16;16,1> uwTEMP_RGB16(0)
shr (16) uwTEMP_RGB16(0)<1> uwDEST_ARGB(%2,0)<16;8,2> 5:w // G >> 5
and (16) uwTEMP_RGB16(0)<1> uwTEMP_RGB16(0) 0x07E0:uw
or (16) uwCSC_TEMP(%1,0)<1> uwCSC_TEMP(%1,0)<16;16,1> uwTEMP_RGB16(0)
}
mov (2) rMSGSRC.0<1>:d wORIX<2;2,1>:w // Block origin (1st quadrant)
shl (1) rMSGSRC.0<1>:d wORIX<0;1,0>:w 1:w // H. block origin need to be doubled for byte offset
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_RGB16:ud // Block width and height (32x8)
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
//Use the mask to determine which pixels shouldn't be over-written
and (1) acc0.0<1>:ud udBLOCK_MASK<0;1,0>:ud 0x00FFFFFF:ud
cmp.ge.f0.0(1) dNULLREG acc0.0<0;1,0>:ud 0x00FFFFFF:ud //Check if all pixels in the block need to be modified
(f0.0) jmpi WriteRGB16ToDataPort
//If mask is not all 1's, then load the entire 32x8 block
//so that only those bytes may be modified that need to be (using the mask)
// Load 32x8 packed RGB565 -----------------------------------------------------
send (8) udSRC_RGB16(0)<1> mMSGHDR udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_RGB16+nBI_DESTINATION_RGB:ud
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
//Merge the data
mov (1) f0.0:uw ubBLOCK_MASK_V:ub //Load the mask on flag reg
(f0.0) mov (8) rMASK_TEMP<1>:uw uwBLOCK_MASK_H:uw //use sel instruction - vK
(-f0.0) mov (8) rMASK_TEMP<1>:uw 0:uw
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (1) f0.1:uw uwMASK_TEMP(0,%1)<0;1,0>
(-f0.1) mov (16) uwCSC_TEMP(%1)<1> uwSRC_RGB16(%1)
}
WriteRGB16ToDataPort:
// Move packed data to MRF and output
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (8) mudMSGPAYLOAD(%1)<1> udCSC_TEMP(%1)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_RGB16+nBI_DESTINATION_RGB:ud
// End of RGB16x8_Save_RGB16
i965_drv_video/shaders/post_processing/Common/RGB16x8_Save_RGB16.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: RGB16x8_Save_RGB16.inc
//
#include "undefall.inc" //Undefine the SRC and DEST sysmbols
// For saving
#define nDPW_BLOCK_SIZE_RGB16 nBLOCK_WIDTH_32+nBLOCK_HEIGHT_8 // RGB16 block size 32x8
#define nDPW_MSG_SIZE_RGB16 nMSGLEN_8 // # of MRF's to hold RGB16 block data (8)
// For conversion to 16bit
.declare uwTEMP_RGB16 Base=REG(r,nTEMP1) ElementSize=2 SrcRegion=<16;16,1> Type=uw //1 GRF
// For masking
#undef nDPR_MSG_SIZE_RGB16
#define nDPR_MSG_SIZE_RGB16 nRESLEN_8 // # of MRF's to hold ARGB block data (8)
#define rMASK_TEMP REG(r,nTEMP0)
.declare uwMASK_TEMP Base=rMASK_TEMP ElementSize=2 SrcRegion=<8;8,1> Type=uw //1 GRF
#if (nSRC_REGION==nREGION_1)
// For saving
#define ubDEST_ARGB ubTOP_Y //Data from previous module
#define uwDEST_ARGB uwTOP_Y //Data from previous module
#define udCSC_TEMP udBOT_Y_IO //Data Converted to 16 bits
#define uwCSC_TEMP uwBOT_Y
//For masking operation
#define udSRC_RGB16 udTOP_Y_IO //To hold the destination data that shouldn't be modified
#define uwSRC_RGB16 uwTOP_Y
#elif (nSRC_REGION==nREGION_2)
// For saving
#define ubDEST_ARGB ubBOT_Y //Data from previous module
#define uwDEST_ARGB uwBOT_Y //Data from previous module
#define udCSC_TEMP udTOP_Y_IO //Data Converted to 16 bits
#define uwCSC_TEMP uwTOP_Y
//For masking operation
#define udSRC_RGB16 udBOT_Y_IO //To hold the destination data that shouldn't be modified
#define uwSRC_RGB16 uwBOT_Y
#endif
i965_drv_video/shaders/post_processing/Common/RGB16x8_Save_Y416.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: RGB16x8_Save_Y416.asm
//
// Save packed ARGB 444 frame data block of size 16x8
//
// To save 16x8 block (128x8 byte layout for ARGB 16bit per component) we need 4 send instructions
// -----------------
// | 1 | 2 | 3 | 4 |
// -----------------
#include "RGB16x8_Save_RGB.inc"
shl (1) rMSGSRC.0<1>:d wORIX<0;1,0>:w 3:w { NoDDClr } // H. block origin need to become 8 times
mov (1) rMSGSRC.1<1>:d wORIY<0;1,0>:w { NoDDClr, NoDDChk } // Block origin (1st quadrant)
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_ARGB:ud { NoDDChk } // Block width and height (32x8)
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
/* Not needed for validation kernels for now -vK
//Use the mask to determine which pixels shouldn't be over-written
and (1) acc0.0<1>:ud udBLOCK_MASK<0;1,0>:ud 0x00FFFFFF:ud
cmp.ge.f0.0(1) dNULLREG acc0.0<0;1,0>:ud 0x00FFFFFF:ud //Check if all pixels in the block need to be modified
(f0.0) jmpi WriteARGBToDataPort
//If mask is not all 1's, then load the entire 64x8 block
//so that only those bytes may be modified that need to be (using the mask)
// Load first block 16x8 packed ARGB 444 ---------------------------------------
or (1) acc0.0<1>:ud udBLOCK_MASK<0;1,0>:ud 0xFF00FF00:ud //Check first block
cmp.e.f0.0 (1) dNULLREG acc0.0<0;1,0>:ud 0xFFFFFFFF:ud
(f0.0) jmpi SkipFirstBlockMerge //If full mask then skip this block
send (8) udSRC_ARGB(0)<1> mMSGHDR udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_ARGB+nBI_DESTINATION_RGB:ud
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
//Merge the data
mov (1) f0.0:uw ubBLOCK_MASK_V:ub //Load the mask on flag reg
(f0.0) mov (8) rMASK_TEMP<1>:uw uwBLOCK_MASK_H:uw //use sel instruction - vK
(-f0.0) mov (8) rMASK_TEMP<1>:uw 0:uw
$for(0, 0; <nY_NUM_OF_ROWS; 1, 2) { //take care of the lines in the block, they are different in the src and dest
mov (1) f0.1:uw uwMASK_TEMP(0,%1)<0;1,0>
(-f0.1) mov (8) udDEST_ARGB(%2)<1> udSRC_ARGB(%1)
}
SkipFirstBlockMerge:
// Load second block 16x8 packed ARGB 444 ---------------------------------------
or (1) acc0.0<1>:ud udBLOCK_MASK<0;1,0>:ud 0xFF0000FF:ud //Check second block
cmp.e.f0.0 (1) dNULLREG acc0.0<0;1,0>:ud 0xFFFFFFFF:ud
(f0.0) jmpi WriteARGBToDataPort //If full mask then skip this block
add (1) mMSGHDR.0<1>:d rMSGSRC.0<0;1,0>:d 32:d // Point to 2nd part
send (8) udSRC_ARGB(0)<1> mMSGHDR udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_ARGB+nBI_DESTINATION_RGB:ud
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud // Point to 1st part again
//Merge the data
mov (1) f0.0:uw ubBLOCK_MASK_V:ub //Load the mask on flag reg
(f0.0) shr (8) rMASK_TEMP<1>:uw uwBLOCK_MASK_H:uw 8:uw //load the mask for second block
(-f0.0) mov (8) rMASK_TEMP<1>:uw 0:uw
$for(0, 1; <nY_NUM_OF_ROWS; 1, 2) { //take care of the lines in the block, they are different in the src and dest
mov (1) f0.1:uw uwMASK_TEMP(0,%1)<0;1,0>
(-f0.1) mov (8) udDEST_ARGB(%2)<1> udSRC_ARGB(%1)
}
*/
WriteARGBToDataPort:
// Move packed data to MRF and output
//Write 1st 4X8 pixels
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (8) mudMSGPAYLOAD(%1)<1> udDEST_ARGB(%1*4)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_ARGB+nBI_DESTINATION_RGB:ud
//Write 2nd 4X8 pixels
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
add (1) mMSGHDR.0<1>:d rMSGSRC.0<0;1,0>:d 32:d // Point to 2nd part
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (8) mudMSGPAYLOAD(%1)<1> udDEST_ARGB(%1*4+1)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_ARGB+nBI_DESTINATION_RGB:ud
//Write 3rd 4X8 pixels
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
add (1) mMSGHDR.0<1>:d rMSGSRC.0<0;1,0>:d 64:d // Point to 2nd part
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (8) mudMSGPAYLOAD(%1)<1> udDEST_ARGB(%1*4+2)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_ARGB+nBI_DESTINATION_RGB:ud
//Write 4th 4X8 pixels
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
add (1) mMSGHDR.0<1>:d rMSGSRC.0<0;1,0>:d 96:d // Point to 2nd part
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (8) mudMSGPAYLOAD(%1)<1> udDEST_ARGB(%1*4+3)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_ARGB+nBI_DESTINATION_RGB:ud
// End of RGB16x8_Save_Y416
i965_drv_video/shaders/post_processing/Common/RGB16x8_Save_Y416.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: RGB16x8_Save_Y416.inc
//
#include "undefall.inc" //Undefine the SRC and DEST sysmbols
// For saving
#define nDPW_BLOCK_SIZE_ARGB nBLOCK_WIDTH_32+nBLOCK_HEIGHT_8 // ARGB block size 32x8
#define nDPW_MSG_SIZE_ARGB nMSGLEN_8 // # of MRF's to hold ARGB block data (8)
// For masking
#undef nDPR_MSG_SIZE_ARGB
#define nDPR_MSG_SIZE_ARGB nRESLEN_8 // # of MRF's to hold ARGB block data (8)
#define rMASK_TEMP REG(r,nTEMP0)
.declare uwMASK_TEMP Base=rMASK_TEMP ElementSize=2 SrcRegion=<8;8,1> Type=uw //1 GRF
#if (nSRC_REGION==nREGION_1)
// For saving
#define udDEST_ARGB udTOP_Y_IO //The output of previous stage is stored here; This is modified and is written to render cache
//For masking operation
#define udSRC_ARGB udBOT_Y_IO //To hold the destination data that shouldn't be modified
#elif (nSRC_REGION==nREGION_2)
// For saving
#define udDEST_ARGB udBOT_Y_IO //The output of previous stage is stored here; This is modified and is written to render cache
//For masking operation
#define udSRC_ARGB udTOP_Y_IO //To hold the destination data that shouldn't be modified
#endif
i965_drv_video/shaders/post_processing/Common/RGB_Pack.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
.declare SRC_B Base=REG(r,10) ElementSize=2 SrcRegion=REGION(8,1) DstRegion=<1> Type=uw
.declare SRC_G Base=REG(r,18) ElementSize=2 SrcRegion=REGION(8,1) DstRegion=<1> Type=uw
.declare SRC_R Base=REG(r,26) ElementSize=2 SrcRegion=REGION(8,1) DstRegion=<1> Type=uw
.declare SRC_A Base=REG(r,34) ElementSize=2 SrcRegion=REGION(8,1) DstRegion=<1> Type=uw
#define DEST_ARGB ubBOT_ARGB
#undef nSRC_REGION
#define nSRC_REGION nREGION_2
//Pack directly to mrf as optimization - vK
$for(0, 0; <8; 1, 2) {
// mov (16) DEST_ARGB(%2,0)<4> SRC_B(%1) { Compr, NoDDClr } // 16 B
// mov (16) DEST_ARGB(%2,1)<4> SRC_G(%1) { Compr, NoDDClr, NoDDChk } // 16 G
// mov (16) DEST_ARGB(%2,2)<4> SRC_R(%1) { Compr, NoDDClr, NoDDChk } // 16 R //these 2 inst can be merged - vK
// mov (16) DEST_ARGB(%2,3)<4> SRC_A(%1) { Compr, NoDDChk } //DEST_RGB_FORMAT<0;1,0>:ub { Compr, NoDDChk } // 16 A
mov (8) DEST_ARGB(%2, 0)<4> SRC_B(%1) { NoDDClr } // 8 B
mov (8) DEST_ARGB(%2, 1)<4> SRC_G(%1) { NoDDClr, NoDDChk } // 8 G
mov (8) DEST_ARGB(%2, 2)<4> SRC_R(%1) { NoDDClr, NoDDChk } // 8 R
mov (8) DEST_ARGB(%2, 3)<4> SRC_A(%1) { NoDDChk } // 8 A
mov (8) DEST_ARGB(%2+1,0)<4> SRC_B(%1,8) { NoDDClr } // 8 B
mov (8) DEST_ARGB(%2+1,1)<4> SRC_G(%1,8) { NoDDClr, NoDDChk } // 8 G
mov (8) DEST_ARGB(%2+1,2)<4> SRC_R(%1,8) { NoDDClr, NoDDChk } // 8 R
mov (8) DEST_ARGB(%2+1,3)<4> SRC_A(%1,8) { NoDDChk } // 8 A
}
i965_drv_video/shaders/post_processing/Common/SetupVPKernel.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Modual name: SetupVPKernel.asm
//
// Initial setup for running video-processing kernels
//
#include "common.inc"
//
// Now, begin source code....
//
.code
#include "Init_All_Regs.asm"
mov (8) rMSGSRC.0<1>:ud r0.0<8;8,1>:ud // Initialize message payload header with R0
#if defined (INC_BLENDING)
mul (1) fALPHA_STEP_X:f fSCALING_STEP_RATIO:f fVIDEO_STEP_X:f //StepX_ratio = AlphaStepX / VideoStepX
#endif
// End of SetupVPKernel
i965_drv_video/shaders/post_processing/Common/common.inc 0 → 100644
View file @ 20975a94
This diff is collapsed.
i965_drv_video/shaders/post_processing/Common/readSampler16x1.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: readSampler16x1.asm
//
// Read one row of pix through sampler
//
//#define SAMPLER_MSG_DSC 0x166A0000 // ILK Sampler Message Descriptor
// Send Message [DevILK] Message Descriptor
// MBZ MsgL=5 MsgR=8 H MBZ SIMD MsgType SmplrIndx BindTab
// 000 0 101 0 1000 1 0 10 0000 0000 00000000
// 0 A 8 A 0 0 0 0
// MsgL=1+2*2(u,v)=5 MsgR=8
#define SAMPLER_MSG_DSC 0x0A8A0000 // ILK Sampler Message Descriptor
// Assume MSGSRC is set already in the caller
//mov (8) rMSGSRC.0<1>:ud 0:ud // Unused fileds
// Read 16 sampled pixels and stored them in float32 in 8 GRFs
// 422 data is expanded to 444, return 8 GRF in the order of RGB- (UYV-).
// 420 data has three surfaces, return 8 GRF. Valid is always in the 1st GRF when in R8. Make sure no overwrite the following 3 GRFs.
// alpha data is expanded to 4444, return 8 GRF in the order of RGBA (UYVA).
mov(16) mMSGHDR<1>:uw rMSGSRC<16;16,1>:uw
send (16) DATABUF(0)<1> mMSGHDR udDUMMY_NULL 0x2 SAMPLER_MSG_DSC+SAMPLER_IDX+BINDING_IDX:ud
i965_drv_video/shaders/post_processing/Common/undefall.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Modual name: undefall.inc
//
// undefine all global symbol for new process
//
//Source definitions
#undef ubSRC_Y
#undef ubSRC_U
#undef ubSRC_V
#undef ub2SRC_Y
#undef ub2SRC_U
#undef ub2SRC_V
#undef ub4SRC_Y
#undef ub4SRC_U
#undef ub4SRC_V
#undef uwSRC_Y
#undef uwSRC_U
#undef uwSRC_V
#undef udSRC_Y
#undef udSRC_U
#undef udSRC_V
#undef udSRC_YUV
#undef nSRC_YUV_REG
//Destination definitions
#undef ubDEST_Y
#undef ubDEST_U
#undef ubDEST_V
#undef ub2DEST_Y
#undef ub2DEST_U
#undef ub2DEST_V
#undef ub4DEST_Y
#undef ub4DEST_U
#undef ub4DEST_V
#undef uwDEST_Y
#undef uwDEST_U
#undef uwDEST_V
#undef udDEST_Y
#undef udDEST_U
#undef udDEST_V
#undef udDEST_YUV
#undef nDEST_YUV_REG
#undef ubDEST_ARGB
// End of undefall.inc
i965_drv_video/shaders/post_processing/Core_Kernels/AVS_IEF.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: AVS_IEF.inc
#ifndef _AVS_INF_INC_
#define _AVS_INF_INC_
#include "undefall.inc" //Undefine the SRC and DEST sysmbols
// Message Header
// m0.7 31:0 Debug
// m0.6 31:0 Debug
// m0.5 31:0 Ignored
// m0.4 31:0 Ignored
// m0.3 31:0 Ignored
// m0.2 31:16 Ignored
// 15 Alpha Write Channel Mask enable=0, disable=1
// 14 Blue Write Channel Mask (V)
// 13 Green Write Channel Mask (Y)
// 12 Red Write Channel Mask (U)
// 11:0 Ignored
// m0.1 Ignored
// m0.0 Ignored
#define mAVS_8x8_HDR m0 // Message Header
#define mAVS_PAYLOAD m1 // Message Payload Header
#define mAVS_8x8_HDR_2 m2 // Message Header
#define mAVS_PAYLOAD_2 m3 // Message Payload Header
#define mAVS_8x8_HDR_UV m2 // Message Header
#define mAVS_PAYLOAD_UV m3 // Message Payload Header
#define rAVS_8x8_HDR rMSGSRC // Mirror of Message Header
#define rAVS_PAYLOAD r9 // Mirror of Message Payload Header
// AVS payload
// m1.7 Ignored
// m1.6 Pixel 0 V Address ---> ORIY (Y0)
// m1.5 Delta V ---> Step Y
// m1.4 Ignored
// m1.3 Ignored
// m1.2 Pixel 0 U Address ---> ORIX (X0)
// m1.1 U 2nd Derivative ---> NLAS dx
// m1.0 Delta U ---> Step X
// Sampler Message Descriptor
// 31:29 Reserved 000
// 28:25 Message length 0010
// 24:20 Response length xxxxx ---> 4GRFs for each enabled channel
// 19 Header Present 1
// 18 MBZ 0
// 17:16 SIMD Mode 11 ---> SIMD64
// 15:12 Message Type 0011 ---> sample_8x8
// 11:8 Sampler Index xxxx
// 7:0 Binding Table Index xxxxxxxx
#define nAVS_MSG_DSC_1CH 0x044BB000
#define nAVS_MSG_DSC_2CH 0x048BB000
#define nAVS_MSG_DSC_3CH 0x04CBB000
#define nAVS_MSG_DSC_4CH 0x050BB000
#define nAVS_RED_CHANNEL_ONLY 0x0000E000 // Enable Red channel only
#define nAVS_GREEN_CHANNEL_ONLY 0x0000D000 // Enable Green channel only
#define nAVS_RED_BLUE_CHANNELS 0x0000A000 // Enable Red and Blue channels
#define nAVS_RGB_CHANNELS 0x00008000 // Enable RGB(YUV) channels
#define nAVS_ALL_CHANNELS 0x00000000 // Enable all channels (ARGB\AYUV)
.declare ubAVS_RESPONSE Base=REG(r,nTEMP8) ElementSize=1 SrcRegion=REGION(16,1) Type=ub
.declare uwAVS_RESPONSE Base=REG(r,nTEMP8) ElementSize=2 SrcRegion=REGION(16,1) Type=uw
.declare ubAVS_RESPONSE_2 Base=REG(r,nTEMP24) ElementSize=1 SrcRegion=REGION(16,1) Type=ub
.declare uwAVS_RESPONSE_2 Base=REG(r,nTEMP24) ElementSize=2 SrcRegion=REGION(16,1) Type=uw
#if (nSRC_REGION==nREGION_2)
#define uwDEST_Y uwBOT_Y
#define uwDEST_U uwBOT_U
#define uwDEST_V uwBOT_V
#define ubDEST_Y ubBOT_Y
#undef nSRC_REGION
#define nSRC_REGION nREGION_2
#else //(nSRC_REGION==nREGION_1)
#define uwDEST_Y uwTOP_Y
#define uwDEST_U uwTOP_U
#define uwDEST_V uwTOP_V
#define ubDEST_Y ubTOP_Y
#undef nSRC_REGION
#define nSRC_REGION nREGION_1
#endif
#endif //_AVS_INF_INC_
i965_drv_video/shaders/post_processing/Core_Kernels/AVS_SetupFirstBlock.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//------------------------------------------------------------------------------
// AVS_SetupFirstBlock.asm
//------------------------------------------------------------------------------
// Setup Message Header
// mov (8) mAVS_8x8_HDR<1>:ud rMSGSRC<8;8,1>:ud
// Check NLAS Enable bit
and.z.f0.0 (1) wNULLREG uwNLAS_ENABLE:uw BIT15:uw
(f0.0)mov (1) fVIDEO_STEP_DELTA:f 0.0:f
// Setup Message Payload Header for 1st block of Media Sampler 8x8
mov (1) rAVS_PAYLOAD.0:f fVIDEO_STEP_DELTA:f //NLAS dx
mov (1) rAVS_PAYLOAD.1:f fVIDEO_STEP_X:f //Step X
mov (1) rAVS_PAYLOAD.5:f fVIDEO_STEP_Y:f //Step Y
mov (2) rAVS_PAYLOAD.2<4>:f fSRC_VID_H_ORI<2;2,1>:f //Orig X and Y
i965_drv_video/shaders/post_processing/Core_Kernels/AVS_SetupSecondBlock.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//------------------------------------------------------------------------------
// AVS_SetupSecondBlock.asm
//------------------------------------------------------------------------------
//NLAS calculations for 2nd block of Media Sampler 8x8:
// X(i) = X0 + dx*i + ddx*i*(i-1)/2 ==> X(8) = X0 + dx*8 +ddx*28
// dx(i)= dx(0) + ddx*i ==> dx(8)= dx + ddx*8
// Calculating X(8)
mov (1) acc0.2<1>:f fSRC_VID_H_ORI:f
mac (1) acc0.2<1>:f fVIDEO_STEP_X:f 8.0:f
mac (1) rAVS_PAYLOAD.2:f fVIDEO_STEP_DELTA:f 28.0:f
// Calculating dx(8)
mov (1) acc0.1<1>:f fVIDEO_STEP_X:f
mac (1) rAVS_PAYLOAD.1:f fVIDEO_STEP_DELTA:f 8.0:f
i965_drv_video/shaders/post_processing/Core_Kernels/DI.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: DI.inc
#ifdef GT
// GT DI Kernel
#else // ILK
// ILK DI Kernel
#endif
//---------------------------------------------------------------------------
// Binding table indices
//---------------------------------------------------------------------------
#define nBIDX_DI_PRV 10 // Previous DI-ed frame
#define nBIDX_DI_CUR 13 // Current DI-ed frame
#define nBIDX_DN 7 // Denoised frame
#define nBIDX_STAT 20 // Statistics
#define nBIDX_DI_Source 4 // Source Surface
//---------------------------------------------------------------------------
// Message descriptors
//---------------------------------------------------------------------------
// Extended message descriptor
#define nSMPL_ENGINE 0x2
#define nDATAPORT_WRITE 0x5
#define nTS_EOT 0x27 // with End-Of-Thread bit ON
// Message descriptor for end-of-thread
// = 000 0001 (message len) 00000 (resp len)
// 0 (header present 0) 00000000000000 0 (URB dereferenced) 0000
#define nEOT_MSGDSC 0x02000000
// 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
// comment begin
// The following is commented out because of walker feature
// It corresponds to the #ifdef GT #else and #endif
//#define nSMPL_MSGDSC 0x040b8000
//#define nSMPL_RESP_LEN_DI 0x00c00000 // 12
//#define nSMPL_RESP_LEN_NODI_PL 0x00500000 // 5
//#define nSMPL_RESP_LEN_NODI_PA 0x00900000 // 9
//#define nSMPL_RESP_LEN_NODN 0x00900000 // 9
//#define nSMPL_RESP_LEN_PDI 0x00b00000 // 11
// comment end
#ifdef GT
#define nSMPL_MSGDSC 0x040b8000
#define nSMPL_RESP_LEN_DI 0x00c00000 // 12
#define nSMPL_RESP_LEN_NODI_PL 0x00500000 // 5 //DI disable, the XY stored in 5th GRF, no impact to return length
#define nSMPL_RESP_LEN_NODI_PA 0x00900000 // 9 //DI disable, the XY stored in 5th GRF, no impact to return length
#define nSMPL_RESP_LEN_NODN 0x00a00000 // 10 //NO DN, originally use 9, now we need use 10 to store the XY with walker
#define nSMPL_RESP_LEN_PDI 0x00b00000 // 11
#else
#define nSMPL_MSGDSC 0x040b8000
#define nSMPL_RESP_LEN_DI 0x00c00000 // 12
#define nSMPL_RESP_LEN_NODI_PL 0x00500000 // 5
#define nSMPL_RESP_LEN_NODI_PA 0x00900000 // 9
#define nSMPL_RESP_LEN_NODN 0x00900000 // 9
#define nSMPL_RESP_LEN_PDI 0x00b00000 // 11
#endif
// Message descriptor for dataport media write
#ifdef GT
// = 000 0000 (message len - set later) 00000 (resp len 0)
// 1 (header present 1) 0 0 1010 (media block write) 00000
// 00000000 (binding table index - set later)
// = 0x00094000
#define nDPMW_MSGDSC 0x00094000
#else // ILK
// = 000 0000 (message len - set later) 00000 (resp len 0)
// 1 (header present 1) 000 0 010 (media block write) 0000
// 00000000 (binding table index - set later)
// = 0x00082000
#define nDPMW_MSGDSC 0x00082000
#endif
#define nDPMW_MSG_LEN_STMM 0x04000000 // 2 - STMM
#define nDPMW_MSG_LEN_DH 0x04000000 // 2 - Denoise history
#define nDPMW_MSG_LEN_PA_DN 0x0a000000 // 5 - Denoised output
#define nDPMW_MSG_LEN_PA_NODI 0x12000000 // 9 - Denoised output - denoise only - DI disabled
#define nDPMW_MSG_LEN_PL_DN 0x06000000 // 3 - Denoised output
#define nDPMW_MSG_LEN_PL_NODI 0x0a000000 // 5 - Denoised output - denoise only - DI disabled
#define nDPMW_MSG_LEN_DI 0x0a000000 // 5 - DI output
//---------------------------------------------------------------------------
// Static and inline parameters
//---------------------------------------------------------------------------
// Static parameters
.declare ubTFLD_FIRST Base=r1.27 ElementSize=1 Type=ub // top field first
.declare ubSRCYUVOFFSET Base=r1.4 ElementSize=1 Type=ub // source packed format
.declare ubDSTYUVOFFSET Base=r1.8 ElementSize=1 Type=ub // destination packed format
.declare uwSPITCH_DIV2 Base=r1.10 ElementSize=2 Type=uw // statistics surface pitch divided by 2
// Inline parameters
.declare uwXORIGIN Base=r5.0 ElementSize=2 Type=uw // X and Y origin
.declare uwYORIGIN Base=r5.1 ElementSize=2 Type=uw
//---------------------------------------------------------------------------
// Kernel GRF variables
//---------------------------------------------------------------------------
// Message response (Denoised & DI-ed pixels & statistics)
.declare dRESP Base=r8 ElementSize=4 Type=d // Response message (12 or 5 or 11)
.declare ubRESP Base=r8 ElementSize=1 Type=ub
.declare dSTMM Base=r16 ElementSize=4 Type=d // STMM
.declare ubDN_HIST_NODI Base=r12 ElementSize=1 Type=ub // Denoise history data (DI disabled)
.declare ubDN_HIST_DI Base=r17 ElementSize=1 Type=ub // Denoise history data (DI enabled)
.declare uwRETURNED_POSITION_DI Base=r17 ElementSize=2 Type=uw // XY_Return_Data (DI enabled)
.declare uwRETURNED_POSITION_DN Base=r12 ElementSize=2 Type=uw // XY_Return_Data (DI disabled)
.declare ub1ST_FLD_DN Base=r12 ElementSize=1 Type=ub // 1st field Denoised data (DI enabled)
.declare d1ST_FLD_DN Base=r12 ElementSize=4 Type=d
.declare ub2ND_FLD_DN Base=r18 ElementSize=1 Type=ub // 2nd field Denoised data (DI enabled)
.declare d2ND_FLD_DN Base=r18 ElementSize=4 Type=d
.declare ubPRV_DI Base=r8 ElementSize=1 Type=ub // Previous frame DI (DI enabled)
.declare ubCUR_DI Base=r12 ElementSize=1 Type=ub // Previous frame DI (DI enabled)
// Packed denoised output
.declare ubDN_YUV Base=r22 ElementSize=1 Type=ub // Denoised YUV422
.declare dDN_YUV Base=r22 ElementSize=4 Type=d
#define npDN_YUV 704 // = 22*32 = 0x280
// Packed DI output
.declare dDI_YUV_PRV Base=r32 ElementSize=4 Type=d // Previous frame DI output
.declare dDI_YUV_CUR Base=r36 ElementSize=4 Type=d // Current frame DI output
#define npDI_YUV 1024 // = 32*32 = 0x
// For packed output
#define p422_YOFFSET a0.2
#define p422_UOFFSET a0.3
#define p422_VOFFSET a0.4
#define pDN_TFLDSRC a0.6
#define pDN_BFLDSRC a0.7
#define npRESP 192 // = 6*32
// Message source
.declare udMSGSRC Base=r70 ElementSize=4 Type=ud
.declare uwMSGSRC Base=r70 ElementSize=2 Type=uw
.declare dMSGSRC Base=r70 ElementSize=4 Type=d
//---------------------------------------------------------------------------
// Kernel MRF variables
//---------------------------------------------------------------------------
#define mMSGHDR_SMPL m1 // Sampler response: m1~m2
.declare mudMSGHDR_SMPL Base=m1 ElementSize=4 Type=ud
.declare muwMSGHDR_SMPL Base=m1 ElementSize=2 Type=uw
#define mMSGHDR_DN m3 // Denoise output: m3~m7 for PA, m3~m5 for PL
.declare mdMSGHDR_DN Base=m3 ElementSize=4 Type=d
#define mMSGHDR_STAT m8 // Statistics output: m8~m9
.declare mdMSGHDR_STAT Base=m8 ElementSize=4 Type=d
.declare mubMSGHDR_STAT Base=m8 ElementSize=1 Type=ub
#define mMSGHDR_DI m10 // DI output: m10~m14
.declare mdMSGHDR_DI Base=m10 ElementSize=4 Type=d
#define mMSGHDR_EOT m15 // EOT
#ifdef GT
#define MSGSRC
#else
#define MSGSRC null:ud
#endif
//---------------------------------------------------------------------------
// End of thread instruction
//---------------------------------------------------------------------------
#ifdef GT
#define END_THREAD send (8) null<1>:d mMSGHDR_EOT nTS_EOT nEOT_MSGDSC
#else // ILK
#define END_THREAD send (8) null<1>:d mMSGHDR_EOT null:ud nTS_EOT nEOT_MSGDSC
#endif
// end of DI.inc
i965_drv_video/shaders/post_processing/Core_Kernels/DI_Hist_Save.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Write denoise history to memory
shr (2) rMSGSRC.0<1>:ud wORIX<2;2,1>:w 2:w NODDCLR // X,Y origin / 4
add (1) rMSGSRC.0<1>:ud rMSGSRC.0<0;1,0>:ud uwSPITCH_DIV2<0;1,0>:uw NODDCLR_NODDCHK // Add pitch to X origin
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_HIST:ud NODDCHK // block width and height (4x2)
mov (8) mMSGHDR_HIST<1>:ud rMSGSRC.0<8;8,1>:ud // message header
mov (1) mudMSGHDR_HIST(1)<1> udRESP(nDI_HIST_OFFSET,0)<0;1,0> // Move denoise history to MRF
send (8) dNULLREG mMSGHDR_HIST udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPMW_MSG_LEN_HIST+nBI_STMM_HISTORY_OUTPUT:ud
i965_drv_video/shaders/post_processing/Core_Kernels/DI_SAVE_PA.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
shl (1) rMSGSRC.0<1>:ud wORIX<0;1,0>:w 1:w NODDCLR // H. block origin need to be doubled
mov (1) rMSGSRC.1<1>:ud wORIY<0;1,0>:w NODDCLR_NODDCHK // Block origin
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_DI:ud NODDCHK // Block width and height (32x8)
add (4) pCF_Y_OFFSET<1>:uw ubDEST_CF_OFFSET<4;4,1>:ub nDEST_YUV_REG*nGRFWIB:w // Initial Y,U,V offset in YUV422 block
// Pack 2nd field Y
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (16) r[pCF_Y_OFFSET, %1*nGRFWIB]<2> ubRESP(nDI_PREV_FRAME_LUMA_OFFSET,%1*16)
}
// Pack 1st field Y
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (16) r[pCF_Y_OFFSET, %1+4*nGRFWIB]<2> ubRESP(nDI_CURR_FRAME_LUMA_OFFSET,%1*16)
}
// Pack 2nd field U
$for(0; <nUV_NUM_OF_ROWS; 1) {
mov (8) r[pCF_U_OFFSET, %1*nGRFWIB]<4> ubRESP(nDI_PREV_FRAME_CHROMA_OFFSET,%1*16+1)<16;8,2> //U pixels
}
// Pack 1st field U
$for(0; <nUV_NUM_OF_ROWS; 1) {
mov (8) r[pCF_U_OFFSET, %1+4*nGRFWIB]<4> ubRESP(nDI_CURR_FRAME_CHROMA_OFFSET,%1*16+1)<16;8,2> //U pixels
}
// Pack 2nd field V
$for(0; <nUV_NUM_OF_ROWS; 1) {
mov (8) r[pCF_V_OFFSET, %1*nGRFWIB]<4> ubRESP(nDI_PREV_FRAME_CHROMA_OFFSET,%1*16)<16;8,2> //Vpixels
}
// Packs1st field V
$for(0; <nUV_NUM_OF_ROWS; 1) {
mov (8) r[pCF_V_OFFSET, %1+4*nGRFWIB]<4> ubRESP(nDI_CURR_FRAME_CHROMA_OFFSET,%1*16)<16;8,2> //Vpixels
}
//save the previous frame
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
$for(0; <4; 1) {
mov (8) mudMSGPAYLOAD(%1)<1> udDEST_YUV(%1)REGION(8,1)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_DI+nBI_DESTINATION_1_YUV:ud
//save the current frame
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
$for(0; <4; 1) {
mov (8) mudMSGPAYLOAD(%1)<1> udDEST_YUV(%1+4)REGION(8,1)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_DI+nBI_DESTINATION_2_YUV:ud
i965_drv_video/shaders/post_processing/Core_Kernels/DNDI.inc 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Module name: DI.inc
#ifdef GT
// GT DI Kernel
#else // ILK
// ILK DI Kernel
#endif
#include "undefall.inc"
//---------------------------------------------------------------------------
// 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
#define nSMPL_DI_MSGDSC 0x040b8000
#define nSMPL_RESP_LEN_DNDI nRESLEN_12 // 12 - for DN + DI Alg
#define nSMPL_RESP_LEN_DN_PL nRESLEN_5 // 5 - for DN Planar Alg
#define nSMPL_RESP_LEN_DN_PA nRESLEN_9 // 9 - for DN Packed Alg
#define nSMPL_RESP_LEN_DI nRESLEN_9 // 9 - for DI Only Alg
#define nSMPL_RESP_LEN_PDI nRESLEN_11 // 11 - for Partial DI Alg
// Attention: The Message Length is The Number of GRFs with Data Only, without the Header
#define nDPMW_MSG_LEN_STMM nMSGLEN_1 // 1 - For STMM Save
#define nDPMW_MSG_LEN_HIST nMSGLEN_1 // 1 - For Denoise History Save
#define nDPMW_MSG_LEN_PA_DN_DI nMSGLEN_4 // 4 - For DN Curr Save
#define nDPMW_MSG_LEN_PA_DN_NODI nMSGLEN_8 // 8 - For DN Curr Save (denoise only - DI disabled)
#define nDPMW_MSG_LEN_PL_DN_DI nMSGLEN_2 // 2 - For DN Curr Save
#define nDPMW_MSG_LEN_PL_DN_NODI nMSGLEN_4 // 4 - For DN Curr Save (denoise only - DI disabled)
#define nDPW_BLOCK_SIZE_STMM nBLOCK_WIDTH_8+nBLOCK_HEIGHT_4 // Y block size 8x4
#undef nDPW_BLOCK_SIZE_DI
#undef nDPW_MSG_SIZE_DI
#define nDPW_BLOCK_SIZE_DI nBLOCK_WIDTH_32+nBLOCK_HEIGHT_4
#define nDPW_MSG_SIZE_DI nMSGLEN_4
//---------------------------------------------------------------------------
// Kernel GRF variables
//---------------------------------------------------------------------------
// Defines for DI enabled
#define nDI_PREV_FRAME_LUMA_OFFSET 0
#define nDI_PREV_FRAME_CHROMA_OFFSET 2
#define nDI_CURR_FRAME_LUMA_OFFSET 4
#define nDI_CURR_FRAME_CHROMA_OFFSET 6
#define nDI_STMM_OFFSET 8
#define nDI_HIST_OFFSET 9
#define nDI_CURR_2ND_FIELD_LUMA_OFFSET 10
#define nDI_CURR_2ND_FIELD_CHROMA_OFFSET 11
// Defines for DI disabled
#define nNODI_LUMA_OFFSET 0
#define nNODI_HIST_OFFSET 4
#define nNODI_CHROMA_OFFSET 5
#ifdef DI_ENABLE
#define nHIST_OFFSET nDI_HIST_OFFSET
#undef nY_NUM_OF_ROWS
#define nY_NUM_OF_ROWS 8 // Number of Y rows per block (4 rows for each frame)
#undef nUV_NUM_OF_ROWS
#define nUV_NUM_OF_ROWS 8 // Number of U/V rows per block
#endif
#ifdef DI_DISABLE
#define nHIST_OFFSET nNODI_HIST_OFFSET
#endif
#if (nSRC_REGION==nREGION_2)
#define ub2SRC_Y ub2BOT_Y
#define ub2SRC_U ub2BOT_U
#define ub2SRC_V ub2BOT_V
#define uwDEST_Y uwBOT_Y
#define uwDEST_U uwBOT_U
#define uwDEST_V uwBOT_V
#define nDEST_YUV_REG nTOP_Y
#define udDEST_YUV udTOP_Y_IO
#define nRESP nTEMP0 // DI return message requires 12 GRFs
#define nDN_YUV nTOP_Y // Space for Packing DN for next run requires 8 GRFs
#undef nSRC_REGION
#define nSRC_REGION nREGION_2
#else
#define ub2SRC_Y ub2TOP_Y
#define ub2SRC_U ub2TOP_U
#define ub2SRC_V ub2TOP_V
#define uwDEST_Y uwTOP_Y
#define uwDEST_U uwTOP_U
#define uwDEST_V uwTOP_V
#define nDEST_YUV_REG nBOT_Y
#define udDEST_YUV udBOT_Y_IO
#define nRESP nTEMP0 // DI return message requires 12 GRFs
#define nDN_YUV nBOT_Y // Space for Packing DN for next run requires 8 GRFs
#undef nSRC_REGION
#define nSRC_REGION nREGION_1 // REGION_1 will be the source region for first kernel
#endif
// Message response (Denoised & DI-ed pixels & statistics)
.declare udRESP Base=REG(r,nRESP) ElementSize=4 SrcRegion=REGION(8,1) DstRegion=<1> Type=ud
.declare ubRESP Base=REG(r,nRESP) ElementSize=1 SrcRegion=REGION(16,1) DstRegion=<1> Type=ub
// For Denoised Curr Output (Used as Priv in Next Run)
.declare ubDN_YUV Base=REG(r,nDN_YUV) ElementSize=1 Type=ub
.declare udDN_YUV Base=REG(r,nDN_YUV) ElementSize=4 Type=ud
#define npDN_YUV nDN_YUV*nGRFWIB
// For DI Process Output (1st and 2nd Frames Output)
//.declare udDI_YUV_PRIV Base=REG(r,nTEMP0) ElementSize=4 Type=ud // Previous frame DI output
//.declare udDI_YUV_CURR Base=REG(r,nTEMP0) ElementSize=4 Type=ud // Current frame DI output
//#define npDI_YUV nTEMP0*nGRFWIB
//---------------------------------------------------------------------------
// Kernel MRF variables
//---------------------------------------------------------------------------
#define mMSG_SMPL m1 // Sampler Command is in: m1~m2
.declare mudMSG_SMPL Base=mMSG_SMPL ElementSize=4 Type=ud
.declare muwMSG_SMPL Base=mMSG_SMPL ElementSize=2 Type=uw
#define mMSGHDR_DN m1 // Denoise Output: m1~m9 for PA, m3~m5 for PL
.declare mudMSGHDR_DN Base=mMSGHDR_DN ElementSize=4 Type=ud
.declare mubMSGHDR_DN Base=mMSGHDR_DN ElementSize=1 Type=ub
#define mMSGHDR_STMM m11 // STMM Output: m11~m12
.declare mudMSGHDR_STMM Base=mMSGHDR_STMM ElementSize=4 Type=ud
#define mMSGHDR_HIST m13 // HIST Output: m13~m14
.declare mudMSGHDR_HIST Base=mMSGHDR_HIST ElementSize=1 Type=ud
#define mMSGHDR_DI_1ST m1 // DI output: m1~m5
.declare mudMSGHDR_DI_1ST Base=mMSGHDR_DI_1ST ElementSize=4 Type=ud
#define mMSGHDR_DI_2ND m6 // DI output: m6~m10
.declare mudMSGHDR_DI_2ND Base=mMSGHDR_DI_2ND ElementSize=4 Type=ud
// end of DNDI.inc
i965_drv_video/shaders/post_processing/Core_Kernels/DNDI_COMMAND.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Activate the DNDI send command
mov (8) mudMSG_SMPL(0)<1> rMSGSRC.0<8;8,1>:ud NODDCLR // message header
mov (1) muwMSG_SMPL(1,4)<1> wORIX<0;1,0>:w NODDCLR_NODDCHK// horizontal origin
mov (1) muwMSG_SMPL(1,12)<1> wORIY<0;1,0>:w NODDCLR_NODDCHK // vertical origin
//mov (2) muwMSG_SMPL(1,4)<2> wORIX<2;2,1>:w NODDCHK// problem during compile !! when using this line
send (8) udRESP(0)<1> mMSG_SMPL udDUMMY_NULL nSMPL_ENGINE nSMPL_DI_MSGDSC+nSMPL_RESP_LEN+nBI_CURRENT_SRC_YUV_HW_DI:ud
i965_drv_video/shaders/post_processing/Core_Kernels/DNDI_Hist_Save.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
// Write denoise history to memory
shr (2) rMSGSRC.0<1>:ud wORIX<2;2,1>:w 2:w NODDCLR // X,Y origin / 4
add (1) rMSGSRC.0<1>:ud rMSGSRC.0<0;1,0>:ud uwSPITCH_DIV2<0;1,0>:uw NODDCLR_NODDCHK// Add pitch to X origin
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_HIST:ud NODDCHK // block width and height (4x2)
mov (8) mMSGHDR_HIST<1>:ud rMSGSRC.0<8;8,1>:ud // message header
mov (2) mudMSGHDR_HIST(1)<1> udRESP(nNODI_HIST_OFFSET,0)<2;2,1> // Move denoise history to MRF
send (8) dNULLREG mMSGHDR_HIST udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPMW_MSG_LEN_HIST+nBI_STMM_HISTORY_OUTPUT:ud
i965_drv_video/shaders/post_processing/Core_Kernels/PA_AVS_IEF_16x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//---------- PA_AVS_IEF_16x8.asm ----------
#include "AVS_IEF.inc"
//------------------------------------------------------------------------------
// 2 sampler reads for 8x8 YUV packed
//------------------------------------------------------------------------------
#include "PA_AVS_IEF_Sample.asm"
//------------------------------------------------------------------------------
// Unpacking sampler reads to 4:4:4 internal planar
//------------------------------------------------------------------------------
#include "PA_AVS_IEF_Unpack_16x8.asm"
//------------------------------------------------------------------------------
i965_drv_video/shaders/post_processing/Core_Kernels/PA_AVS_IEF_8x4.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//---------- PA_AVS_IEF_8x4.asm ----------
#include "AVS_IEF.inc"
//------------------------------------------------------------------------------
// 2 sampler reads for 8x8 YUV packed
//------------------------------------------------------------------------------
#include "PA_AVS_IEF_Sample.asm"
//------------------------------------------------------------------------------
// Unpacking sampler data to 4:2:0 internal planar
//------------------------------------------------------------------------------
#include "PA_AVS_IEF_Unpack_8x4.asm"
//------------------------------------------------------------------------------
i965_drv_video/shaders/post_processing/Core_Kernels/PA_AVS_IEF_8x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//---------- PA_AVS_IEF_8x8.asm ----------
#include "AVS_IEF.inc"
//------------------------------------------------------------------------------
// 2 sampler reads for 8x8 YUV packed
//------------------------------------------------------------------------------
#include "PA_AVS_IEF_Sample.asm"
//------------------------------------------------------------------------------
// Unpacking sampler data to 4:2:2 internal planar
//------------------------------------------------------------------------------
#include "PA_AVS_IEF_Unpack_8x8.asm"
//------------------------------------------------------------------------------
i965_drv_video/shaders/post_processing/Core_Kernels/PA_AVS_IEF_Sample.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//---------- PA_AVS_IEF_Sample.asm ----------
//------------------------------------------------------------------------------
// 2 sampler reads for 8x8 YUV packed
//------------------------------------------------------------------------------
// 1st 8x8 setup
#include "AVS_SetupFirstBlock.asm"
// Enable RGB(YUV) channels
mov (1) rAVS_8x8_HDR.2:ud nAVS_RGB_CHANNELS:ud
mov (16) mAVS_8x8_HDR.0:ud rAVS_8x8_HDR.0<8;8,1>:ud // Copy msg header and payload mirrors to MRFs
send (1) uwAVS_RESPONSE(0)<1> mAVS_8x8_HDR udDUMMY_NULL nSMPL_ENGINE nAVS_MSG_DSC_3CH+nSI_SRC_YUV+nBI_CURRENT_SRC_YUV
// Return YUV in 12 GRFs
// 2nd 8x8 setup
#include "AVS_SetupSecondBlock.asm"
mov (16) mAVS_8x8_HDR_2.0:ud rAVS_8x8_HDR.0<8;8,1>:ud // Copy msg header and payload mirrors to MRFs
send (1) uwAVS_RESPONSE_2(0)<1> mAVS_8x8_HDR_2 udDUMMY_NULL nSMPL_ENGINE nAVS_MSG_DSC_3CH+nSI_SRC_YUV+nBI_CURRENT_SRC_YUV
// Return YUV in 12 GRFs
i965_drv_video/shaders/post_processing/Core_Kernels/PA_AVS_IEF_Unpack_8x4.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//---------- PA_AVS_IEF_Unpack_8x8.asm ----------
// Yoni: In order to optimize unpacking, 3 methods are being checked:
// 1. AVS_ORIGINAL
// 2. AVS_ROUND_TO_8_BITS
// 3. AVS_INDIRECT_ACCESS
//
// Only 1 method should stay in the code
//#define AVS_ROUND_TO_8_BITS
//#define AVS_INDIRECT_ACCESS
// Move first 8x8 words of Y to dest GRF
mov (8) uwDEST_Y(0)<1> ubAVS_RESPONSE(2,1)<16;4,2>
mov (8) uwDEST_Y(1)<1> ubAVS_RESPONSE(2,8+1)<16;4,2>
mov (8) uwDEST_Y(2)<1> ubAVS_RESPONSE(3,1)<16;4,2>
mov (8) uwDEST_Y(3)<1> ubAVS_RESPONSE(3,8+1)<16;4,2>
mov (8) uwDEST_Y(4)<1> ubAVS_RESPONSE(8,1)<16;4,2>
mov (8) uwDEST_Y(5)<1> ubAVS_RESPONSE(8,8+1)<16;4,2>
mov (8) uwDEST_Y(6)<1> ubAVS_RESPONSE(9,1)<16;4,2>
mov (8) uwDEST_Y(7)<1> ubAVS_RESPONSE(9,8+1)<16;4,2>
// Move first 4x8 words of V to dest GRF
mov (4) uwDEST_V(0)<1> ubAVS_RESPONSE(0,1)<16;2,4>
mov (4) uwDEST_V(0,8)<1> ubAVS_RESPONSE(1,1)<16;2,4>
mov (4) uwDEST_V(1)<1> ubAVS_RESPONSE(6,1)<16;2,4>
mov (4) uwDEST_V(1,8)<1> ubAVS_RESPONSE(7,1)<16;2,4>
// Move first 4x8 words of U to dest GRF
mov (4) uwDEST_U(0)<1> ubAVS_RESPONSE(4,1)<16;2,4>
mov (4) uwDEST_U(0,8)<1> ubAVS_RESPONSE(5,1)<16;2,4>
mov (4) uwDEST_U(1)<1> ubAVS_RESPONSE(10,1)<16;2,4>
mov (4) uwDEST_U(1,8)<1> ubAVS_RESPONSE(11,1)<16;2,4>
// Move second 8x8 words of Y to dest GRF
mov (8) uwDEST_Y(0,8)<1> ubAVS_RESPONSE_2(2,1)<16;4,2>
mov (8) uwDEST_Y(1,8)<1> ubAVS_RESPONSE_2(2,8+1)<16;4,2>
mov (8) uwDEST_Y(2,8)<1> ubAVS_RESPONSE_2(3,1)<16;4,2>
mov (8) uwDEST_Y(3,8)<1> ubAVS_RESPONSE_2(3,8+1)<16;4,2>
mov (8) uwDEST_Y(4,8)<1> ubAVS_RESPONSE_2(8,1)<16;4,2>
mov (8) uwDEST_Y(5,8)<1> ubAVS_RESPONSE_2(8,8+1)<16;4,2>
mov (8) uwDEST_Y(6,8)<1> ubAVS_RESPONSE_2(9,1)<16;4,2>
mov (8) uwDEST_Y(7,8)<1> ubAVS_RESPONSE_2(9,8+1)<16;4,2>
// Move second 4x8 words of V to dest GRF
mov (4) uwDEST_V(0,4)<1> ubAVS_RESPONSE_2(0,1)<16;2,4>
mov (4) uwDEST_V(0,12)<1> ubAVS_RESPONSE_2(1,1)<16;2,4>
mov (4) uwDEST_V(1,4)<1> ubAVS_RESPONSE_2(6,1)<16;2,4>
mov (4) uwDEST_V(1,12)<1> ubAVS_RESPONSE_2(7,1)<16;2,4>
// Move second 4x8 words of U to dest GRF
mov (4) uwDEST_U(0,4)<1> ubAVS_RESPONSE_2(4,1)<16;2,4>
mov (4) uwDEST_U(0,12)<1> ubAVS_RESPONSE_2(5,1)<16;2,4>
mov (4) uwDEST_U(1,4)<1> ubAVS_RESPONSE_2(10,1)<16;2,4>
mov (4) uwDEST_U(1,12)<1> ubAVS_RESPONSE_2(11,1)<16;2,4>
//------------------------------------------------------------------------------
// Re-define new number of lines
#undef nUV_NUM_OF_ROWS
#undef nY_NUM_OF_ROWS
#define nY_NUM_OF_ROWS 8
#define nUV_NUM_OF_ROWS 8
i965_drv_video/shaders/post_processing/Core_Kernels/PA_AVS_IEF_Unpack_8x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//---------- PA_AVS_IEF_Unpack_8x8.asm ----------
// Yoni: In order to optimize unpacking, 3 methods are being checked:
// 1. AVS_ORIGINAL
// 2. AVS_ROUND_TO_8_BITS
// 3. AVS_INDIRECT_ACCESS
//
// Only 1 method should stay in the code
//#define AVS_ROUND_TO_8_BITS
//#define AVS_INDIRECT_ACCESS
// Move first 8x8 words of Y to dest GRF
mov (8) uwDEST_Y(0)<1> ubAVS_RESPONSE(2,1)<16;4,2>
mov (8) uwDEST_Y(1)<1> ubAVS_RESPONSE(2,8+1)<16;4,2>
mov (8) uwDEST_Y(2)<1> ubAVS_RESPONSE(3,1)<16;4,2>
mov (8) uwDEST_Y(3)<1> ubAVS_RESPONSE(3,8+1)<16;4,2>
mov (8) uwDEST_Y(4)<1> ubAVS_RESPONSE(8,1)<16;4,2>
mov (8) uwDEST_Y(5)<1> ubAVS_RESPONSE(8,8+1)<16;4,2>
mov (8) uwDEST_Y(6)<1> ubAVS_RESPONSE(9,1)<16;4,2>
mov (8) uwDEST_Y(7)<1> ubAVS_RESPONSE(9,8+1)<16;4,2>
// Move first 4x8 words of V to dest GRF
mov (4) uwDEST_V(0)<1> ubAVS_RESPONSE(0,1)<16;2,4>
mov (4) uwDEST_V(0,8)<1> ubAVS_RESPONSE(0,8+1)<16;2,4>
mov (4) uwDEST_V(1)<1> ubAVS_RESPONSE(1,1)<16;2,4>
mov (4) uwDEST_V(1,8)<1> ubAVS_RESPONSE(1,8+1)<16;2,4>
mov (4) uwDEST_V(2)<1> ubAVS_RESPONSE(6,1)<16;2,4>
mov (4) uwDEST_V(2,8)<1> ubAVS_RESPONSE(6,8+1)<16;2,4>
mov (4) uwDEST_V(3)<1> ubAVS_RESPONSE(7,1)<16;2,4>
mov (4) uwDEST_V(3,8)<1> ubAVS_RESPONSE(7,8+1)<16;2,4>
// Move first 4x8 words of U to dest GRF
mov (4) uwDEST_U(0)<1> ubAVS_RESPONSE(4,1)<16;2,4>
mov (4) uwDEST_U(0,8)<1> ubAVS_RESPONSE(4,8+1)<16;2,4>
mov (4) uwDEST_U(1)<1> ubAVS_RESPONSE(5,1)<16;2,4>
mov (4) uwDEST_U(1,8)<1> ubAVS_RESPONSE(5,8+1)<16;2,4>
mov (4) uwDEST_U(2)<1> ubAVS_RESPONSE(10,1)<16;2,4>
mov (4) uwDEST_U(2,8)<1> ubAVS_RESPONSE(10,8+1)<16;2,4>
mov (4) uwDEST_U(3)<1> ubAVS_RESPONSE(11,1)<16;2,4>
mov (4) uwDEST_U(3,8)<1> ubAVS_RESPONSE(11,8+1)<16;2,4>
// Move second 8x8 words of Y to dest GRF
mov (8) uwDEST_Y(0,8)<1> ubAVS_RESPONSE_2(2,1)<16;4,2>
mov (8) uwDEST_Y(1,8)<1> ubAVS_RESPONSE_2(2,8+1)<16;4,2>
mov (8) uwDEST_Y(2,8)<1> ubAVS_RESPONSE_2(3,1)<16;4,2>
mov (8) uwDEST_Y(3,8)<1> ubAVS_RESPONSE_2(3,8+1)<16;4,2>
mov (8) uwDEST_Y(4,8)<1> ubAVS_RESPONSE_2(8,1)<16;4,2>
mov (8) uwDEST_Y(5,8)<1> ubAVS_RESPONSE_2(8,8+1)<16;4,2>
mov (8) uwDEST_Y(6,8)<1> ubAVS_RESPONSE_2(9,1)<16;4,2>
mov (8) uwDEST_Y(7,8)<1> ubAVS_RESPONSE_2(9,8+1)<16;4,2>
// Move second 4x8 words of V to dest GRF
mov (4) uwDEST_V(0,4)<1> ubAVS_RESPONSE_2(0,1)<16;2,4>
mov (4) uwDEST_V(0,12)<1> ubAVS_RESPONSE_2(0,8+1)<16;2,4>
mov (4) uwDEST_V(1,4)<1> ubAVS_RESPONSE_2(1,1)<16;2,4>
mov (4) uwDEST_V(1,12)<1> ubAVS_RESPONSE_2(1,8+1)<16;2,4>
mov (4) uwDEST_V(2,4)<1> ubAVS_RESPONSE_2(6,1)<16;2,4>
mov (4) uwDEST_V(2,12)<1> ubAVS_RESPONSE_2(6,8+1)<16;2,4>
mov (4) uwDEST_V(3,4)<1> ubAVS_RESPONSE_2(7,1)<16;2,4>
mov (4) uwDEST_V(3,12)<1> ubAVS_RESPONSE_2(7,8+1)<16;2,4>
// Move second 4x8 words of U to dest GRF
mov (4) uwDEST_U(0,4)<1> ubAVS_RESPONSE_2(4,1)<16;2,4>
mov (4) uwDEST_U(0,12)<1> ubAVS_RESPONSE_2(4,8+1)<16;2,4>
mov (4) uwDEST_U(1,4)<1> ubAVS_RESPONSE_2(5,1)<16;2,4>
mov (4) uwDEST_U(1,12)<1> ubAVS_RESPONSE_2(5,8+1)<16;2,4>
mov (4) uwDEST_U(2,4)<1> ubAVS_RESPONSE_2(10,1)<16;2,4>
mov (4) uwDEST_U(2,12)<1> ubAVS_RESPONSE_2(10,8+1)<16;2,4>
mov (4) uwDEST_U(3,4)<1> ubAVS_RESPONSE_2(11,1)<16;2,4>
mov (4) uwDEST_U(3,12)<1> ubAVS_RESPONSE_2(11,8+1)<16;2,4>
//------------------------------------------------------------------------------
// Re-define new number of lines
#undef nUV_NUM_OF_ROWS
#undef nY_NUM_OF_ROWS
#define nY_NUM_OF_ROWS 8
#define nUV_NUM_OF_ROWS 8
i965_drv_video/shaders/post_processing/Core_Kernels/PA_DNDI_ALG.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
#define DI_ENABLE
#include "DNDI.inc"
#ifdef DI_ONLY
#undef nSMPL_RESP_LEN
#define nSMPL_RESP_LEN nSMPL_RESP_LEN_DI // set the number of GRF
#else
#undef nSMPL_RESP_LEN
#define nSMPL_RESP_LEN nSMPL_RESP_LEN_DNDI // set the number of GRF
#endif
#undef nDPW_BLOCK_SIZE_HIST
#define nDPW_BLOCK_SIZE_HIST nBLOCK_WIDTH_4+nBLOCK_HEIGHT_1 // HIST Block Size for Write is 4x2
#undef nDPW_BLOCK_SIZE_DN
#define nDPW_BLOCK_SIZE_DN nBLOCK_WIDTH_32+nBLOCK_HEIGHT_4 // DN Block Size for Write is 32x4
////////////////////////////////////// Run the DN Algorithm ///////////////////////////////////////
#include "DNDI_Command.asm"
////////////////////////////////////// Rearrange for Internal Planar //////////////////////////////
//// move the previous frame Y component to internal planar format
//$for (0; <nY_NUM_OF_ROWS/2; 1) {
// mov (16) uwDEST_Y(%1,0)<1> ubRESP(nDI_PREV_FRAME_LUMA_OFFSET,%1*16)
//}
//// move the previous frame U,V components to internal planar format
//$for (0; <nUV_NUM_OF_ROWS/2; 1) {
// mov (8) uwDEST_U(0,%1*8)<1> ubRESP(nDI_PREV_FRAME_CHROMA_OFFSET,%1*16+1)<16;8,2> //U pixels
// mov (8) uwDEST_V(0,%1*8)<1> ubRESP(nDI_PREV_FRAME_CHROMA_OFFSET,%1*16)<16;8,2> //V pixels
//}
//// move the current frame Y component to internal planar format
//$for (0; <nY_NUM_OF_ROWS/2; 1) {
// mov (16) uwDEST_Y(%1+4,0)<1> ubRESP(nDI_CURR_FRAME_LUMA_OFFSET,%1*16)
//}
//// move the current frame U,V components to internal planar format
//$for (0; <nUV_NUM_OF_ROWS/2; 1) {
// mov (8) uwDEST_U(2,%1*8)<1> ubRESP(nDI_CURR_FRAME_CHROMA_OFFSET,%1*16+1)<16;8,2> //U pixels
// mov (8) uwDEST_V(2,%1*8)<1> ubRESP(nDI_CURR_FRAME_CHROMA_OFFSET,%1*16)<16;8,2> //V pixels
//}
////////////////////////////////////// Save the STMM Data for Next Run /////////////////////////
// Write STMM to memory
shr (1) rMSGSRC.0<1>:ud wORIX<0;1,0>:w 1:w NODDCLR // X origin / 2
mov (1) rMSGSRC.1<1>:ud wORIY<0;1,0>:w NODDCLR_NODDCHK // Y origin
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_STMM:ud NODDCHK // block width and height (8x4)
mov (8) mudMSGHDR_STMM(0)<1> rMSGSRC.0<8;8,1>:ud // message header
mov (8) mudMSGHDR_STMM(1)<1> udRESP(nDI_STMM_OFFSET,0) // Move STMM to MRF
send (8) dNULLREG mMSGHDR_STMM udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPMW_MSG_LEN_STMM+nBI_STMM_HISTORY_OUTPUT:ud
////////////////////////////////////// Save the History Data for Next Run /////////////////////////
#ifdef DI_ONLY
#else
#include "DI_Hist_Save.asm"
////////////////////////////////////// Pack and Save the DN Curr Frame for Next Run ///////////////
// check top/bottom field first
cmp.e.f0.0 (1) null<1>:w ubTFLD_FIRST<0;1,0>:ub 1:w
add (4) pCF_Y_OFFSET<1>:uw ubSRC_CF_OFFSET<4;4,1>:ub npDN_YUV:uw
//set the save DN position
shl (1) rMSGSRC.0<1>:ud wORIX<0;1,0>:w 1:w NODDCLR // X origin * 2
mov (1) rMSGSRC.1<1>:ud wORIY<0;1,0>:w NODDCLR_NODDCHK // Y origin
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_DN:ud NODDCHK // block width and height (8x4)
mov (8) mudMSGHDR_DN(0)<1> rMSGSRC.0<8;8,1>:ud
(f0.0) jmpi (1) TOP_FIELD_FIRST
BOTTOM_FIELD_FIRST:
//$for (0,0; <nY_NUM_OF_ROWS/2; 2,1) {
// mov (16) r[pCF_Y_OFFSET, %1*32]<2>:ub ubRESP(nDI_CURR_2ND_FIELD_LUMA_OFFSET,%2*16) // 2nd field luma from current frame (line 0,2)
// mov (16) r[pCF_Y_OFFSET, %1+1*32]<2>:ub ubRESP(nDI_CURR_FRAME_LUMA_OFFSET+%2,16) // 1st field luma from current frame (line 1,3)
// mov (8) r[pCF_U_OFFSET, %1*32]<4>:ub ubRESP(nDI_CURR_2ND_FIELD_CHROMA_OFFSET,%2*16+1)<16;8,2> // 2nd field U from current frame (line 0,2)
// mov (8) r[pCF_V_OFFSET, %1*32]<4>:ub ubRESP(nDI_CURR_2ND_FIELD_CHROMA_OFFSET,%2*16)<16;8,2> // 2nd field V from current frame (line 0,2)
// mov (8) r[pCF_U_OFFSET, %1+1*32]<4>:ub ubRESP(nDI_CURR_FRAME_CHROMA_OFFSET+%2,16+1)<16;8,2> // 1st field U from current frame (line 1,3)
// mov (8) r[pCF_V_OFFSET, %1+1*32]<4>:ub ubRESP(nDI_CURR_FRAME_CHROMA_OFFSET+%2,16)<16;8,2> // 1st field U from current frame (line 1,3)
//}
$for (0,0; <nY_NUM_OF_ROWS/2; 2,1) {
mov (16) r[pCF_Y_OFFSET, %1*32]<2>:ub ubRESP(nDI_CURR_2ND_FIELD_LUMA_OFFSET,%2*16) // 2nd field luma from current frame (line 0,2)
mov (16) r[pCF_Y_OFFSET, %1+1*32]<2>:ub ubRESP(nDI_CURR_FRAME_LUMA_OFFSET+%2,16) // 1st field luma from current frame (line 1,3)
}
$for (0,0; <nY_NUM_OF_ROWS/2; 2,1) {
mov (8) r[pCF_U_OFFSET, %1*32]<4>:ub ubRESP(nDI_CURR_2ND_FIELD_CHROMA_OFFSET,%2*16+1)<16;8,2> // 2nd field U from current frame (line 0,2)
mov (8) r[pCF_U_OFFSET, %1+1*32]<4>:ub ubRESP(nDI_CURR_FRAME_CHROMA_OFFSET+%2,16+1)<16;8,2> // 1st field U from current frame (line 1,3)
}
$for (0,0; <nY_NUM_OF_ROWS/2; 2,1) {
mov (8) r[pCF_V_OFFSET, %1*32]<4>:ub ubRESP(nDI_CURR_2ND_FIELD_CHROMA_OFFSET,%2*16)<16;8,2> // 2nd field V from current frame (line 0,2)
mov (8) r[pCF_V_OFFSET, %1+1*32]<4>:ub ubRESP(nDI_CURR_FRAME_CHROMA_OFFSET+%2,16)<16;8,2> // 1st field U from current frame (line 1,3)
}
jmpi (1) SAVE_DN_CURR
TOP_FIELD_FIRST:
//$for (0,0; <nY_NUM_OF_ROWS/2; 2,1) {
// mov (16) r[pCF_Y_OFFSET, %1*32]<2>:ub ubRESP(nDI_CURR_FRAME_LUMA_OFFSET+%2,0) // 1st field luma from current frame (line 0,2)
// mov (16) r[pCF_Y_OFFSET, %1+1*32]<2>:ub ubRESP(nDI_CURR_2ND_FIELD_LUMA_OFFSET,%2*16) // 2nd field luma from current frame (line 1,3)
// mov (8) r[pCF_U_OFFSET, %1*32]<4>:ub ubRESP(nDI_CURR_FRAME_CHROMA_OFFSET+%2,1)<16;8,2> // 1st field U from current frame (line 0,2)
// mov (8) r[pCF_V_OFFSET, %1*32]<4>:ub ubRESP(nDI_CURR_FRAME_CHROMA_OFFSET+%2,0)<16;8,2> // 1st field V from current frame (line 0,2)
// mov (8) r[pCF_U_OFFSET, %1+1*32]<4>:ub ubRESP(nDI_CURR_2ND_FIELD_CHROMA_OFFSET,%2*16+1)<16;8,2> // 2nd field U from current frame (line 1,3)
// mov (8) r[pCF_V_OFFSET, %1+1*32]<4>:ub ubRESP(nDI_CURR_2ND_FIELD_CHROMA_OFFSET,%2*16)<16;8,2> // 2nd field V from current frame (line 1,3)
//}
$for (0,0; <nY_NUM_OF_ROWS/2; 2,1) {
mov (16) r[pCF_Y_OFFSET, %1*32]<2>:ub ubRESP(nDI_CURR_FRAME_LUMA_OFFSET+%2,0) // 1st field luma from current frame (line 0,2)
mov (16) r[pCF_Y_OFFSET, %1+1*32]<2>:ub ubRESP(nDI_CURR_2ND_FIELD_LUMA_OFFSET,%2*16) // 2nd field luma from current frame (line 1,3)
}
$for (0,0; <nY_NUM_OF_ROWS/2; 2,1) {
mov (8) r[pCF_U_OFFSET, %1*32]<4>:ub ubRESP(nDI_CURR_FRAME_CHROMA_OFFSET+%2,1)<16;8,2> // 1st field U from current frame (line 0,2)
mov (8) r[pCF_U_OFFSET, %1+1*32]<4>:ub ubRESP(nDI_CURR_2ND_FIELD_CHROMA_OFFSET,%2*16+1)<16;8,2> // 2nd field U from current frame (line 1,3)
}
$for (0,0; <nY_NUM_OF_ROWS/2; 2,1) {
mov (8) r[pCF_V_OFFSET, %1*32]<4>:ub ubRESP(nDI_CURR_FRAME_CHROMA_OFFSET+%2,0)<16;8,2> // 1st field V from current frame (line 0,2)
mov (8) r[pCF_V_OFFSET, %1+1*32]<4>:ub ubRESP(nDI_CURR_2ND_FIELD_CHROMA_OFFSET,%2*16)<16;8,2> // 2nd field V from current frame (line 1,3)
}
SAVE_DN_CURR:
$for(0; <nY_NUM_OF_ROWS/2; 1) {
mov (8) mudMSGHDR_DN(%1+1)<1> udDN_YUV(%1)REGION(8,1)
}
send (8) dNULLREG mMSGHDR_DN udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPMW_MSG_LEN_PA_DN_DI+nBI_DESTINATION_YUV:ud
#endif
// Save Processed frames
#include "DI_Save_PA.asm"
i965_drv_video/shaders/post_processing/Core_Kernels/PA_DN_ALG.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
#define DI_DISABLE
#include "DNDI.inc"
#undef nY_NUM_OF_ROWS
#define nY_NUM_OF_ROWS 8 // Number of Y rows per block
#undef nUV_NUM_OF_ROWS
#define nUV_NUM_OF_ROWS 8 // Number of U/V rows per block
#undef nSMPL_RESP_LEN
#define nSMPL_RESP_LEN nSMPL_RESP_LEN_DN_PA // Set the Number of GRFs in DNDI response
#undef nDPW_BLOCK_SIZE_DN
#define nDPW_BLOCK_SIZE_DN nBLOCK_WIDTH_32+nBLOCK_HEIGHT_8 // DN Curr Block Size for Write is 32x8
#undef nDPW_BLOCK_SIZE_HIST
#define nDPW_BLOCK_SIZE_HIST nBLOCK_WIDTH_4+nBLOCK_HEIGHT_2 // HIST Block Size for Write is 4x2
////////////////////////////////////// Run the DN Algorithm ///////////////////////////////////////
#include "DNDI_COMMAND.asm"
////////////////////////////////////// Save the History Data for Next Run /////////////////////////
#include "DNDI_Hist_Save.asm"
////////////////////////////////////// Pack and Save the DN Curr Frame for Next Run ///////////////
add (4) pCF_Y_OFFSET<1>:uw ubDEST_CF_OFFSET<4;4,1>:ub npDN_YUV:w
$for (0; <nY_NUM_OF_ROWS; 1) {
mov (16) r[pCF_Y_OFFSET, %1*32]<2>:ub ubRESP(nNODI_LUMA_OFFSET,%1*16)<16;16,1> // copy line of Y
}
$for (0; <nUV_NUM_OF_ROWS; 1) {
mov (8) r[pCF_U_OFFSET, %1*32]<4>:ub ubRESP(nNODI_CHROMA_OFFSET,%1*16+1)<16;8,2> // copy line of U
mov (8) r[pCF_V_OFFSET, %1*32]<4>:ub ubRESP(nNODI_CHROMA_OFFSET,%1*16)<16;8,2> // copy line of V
}
shl (1) rMSGSRC.0<1>:ud wORIX<0;1,0>:w 1:w // X origin * 2 (422 output)
mov (1) rMSGSRC.1<1>:ud wORIY<0;1,0>:w // Y origin
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_DN:ud // block width and height (32x8)
mov (8) mMSGHDR_DN<1>:ud rMSGSRC<8;8,1>:ud // message header
$for(0; <nY_NUM_OF_ROWS; 2) {
mov (16) mudMSGHDR_DN(1+%1)<1> udDN_YUV(%1)REGION(8,1) // Move DN Curr to MRF
}
send (8) dNULLREG mMSGHDR_DN udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPMW_MSG_LEN_PA_DN_NODI+nBI_DESTINATION_YUV:ud
i965_drv_video/shaders/post_processing/Core_Kernels/PA_Scaling.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//---------- PA_Scaling.asm ----------
#include "Scaling.inc"
// Build 16 elements ramp in float32 and normalized it
// mov (8) SAMPLER_RAMP(0)<1> 0x76543210:v
// add (8) SAMPLER_RAMP(1)<1> SAMPLER_RAMP(0) 8.0:f
mov (4) SAMPLER_RAMP(0)<1> 0x48403000:vf //3, 2, 1, 0 in float vector
mov (4) SAMPLER_RAMP(0,4)<1> 0x5C585450:vf //7, 6, 5, 4 in float vector
add (8) SAMPLER_RAMP(1)<1> SAMPLER_RAMP(0) 8.0:f
//Module: PrepareScaleCoord.asm
// Setup for sampler msg hdr
mov (2) rMSGSRC.0<1>:ud 0:ud { NoDDClr } // Unused fields
mov (1) rMSGSRC.2<1>:ud 0:ud { NoDDChk } // Write and offset
// Calculate 16 v based on the step Y and vertical origin
mov (16) mfMSGPAYLOAD(2)<1> fSRC_VID_V_ORI<0;1,0>:f
mov (16) SCALE_COORD_Y<1>:f fSRC_VID_V_ORI<0;1,0>:f
// Calculate 16 u based on the step X and hori origin
// line (16) mfMSGPAYLOAD(0)<1> SCALE_STEP_X<0;1,0>:f SAMPLER_RAMP(0) // Assign to mrf directly
mov (16) acc0:f fSRC_VID_H_ORI<0;1,0>:f { Compr }
mac (16) mfMSGPAYLOAD(0)<1> fVIDEO_STEP_X<0;1,0>:f SAMPLER_RAMP(0) { Compr }
//Setup the constants for line instruction
mov (1) SCALE_LINE_P255<1>:f 255.0:f { NoDDClr } //{ NoDDClr, NoDDChk }
mov (1) SCALE_LINE_P0_5<1>:f 0.5:f { NoDDChk }
//------------------------------------------------------------------------------
$for (0; <nY_NUM_OF_ROWS; 1) {
// Read 16 sampled pixels and store them in float32 in 8 GRFs in the order of BGRA (VYUA).
mov (8) MSGHDR_SCALE.0:ud rMSGSRC.0<8;8,1>:ud // Copy msg header and payload mirrors to MRFs
send (16) SCALE_RESPONSE_YW(0)<1> MSGHDR_SCALE udDUMMY_NULL nSMPL_ENGINE SMPLR_MSG_DSC+nSI_SRC_SIMD16_YUV+nBI_CURRENT_SRC_YUV
// Calculate 16 v for next line
add (16) mfMSGPAYLOAD(2)<1> SCALE_COORD_Y<8;8,1>:f fVIDEO_STEP_Y<0;1,0>:f // Assign to mrf directly
add (16) SCALE_COORD_Y<1>:f SCALE_COORD_Y<8;8,1>:f fVIDEO_STEP_Y<0;1,0>:f // Assign to mrf directly
// Scale back to [0, 255], convert f to ud
line (16) acc0:f SCALE_LINE_P255<0;1,0>:f SCALE_RESPONSE_YF(0) { Compr } // Process B, V
mov (16) SCALE_RESPONSE_YD(0)<1> acc0:f { Compr }
line (16) acc0:f SCALE_LINE_P255<0;1,0>:f SCALE_RESPONSE_YF(2) { Compr } // Process B, V
mov (16) SCALE_RESPONSE_YD(2)<1> acc0:f { Compr }
line (16) acc0:f SCALE_LINE_P255<0;1,0>:f SCALE_RESPONSE_YF(4) { Compr } // Process B, V
mov (16) SCALE_RESPONSE_YD(4)<1> acc0:f { Compr }
mov (16) DEST_V(%1)<1> SCALE_RESPONSE_YB(0) //possible error due to truncation - vK
mov (16) DEST_Y(%1)<1> SCALE_RESPONSE_YB(2) //possible error due to truncation - vK
mov (16) DEST_U(%1)<1> SCALE_RESPONSE_YB(4) //possible error due to truncation - vK
}
#define nSRC_REGION nREGION_1
//------------------------------------------------------------------------------
i965_drv_video/shaders/post_processing/Core_Kernels/PL2_AVS_IEF_16x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//---------- PL2_AVS_IEF_16x8.asm ----------
#include "AVS_IEF.inc"
//------------------------------------------------------------------------------
// 2 sampler reads for 8x8 Y each
// 2 sampler read for 8x8 U and 8x8 V (NV11\P208 input surface)
//------------------------------------------------------------------------------
// 1st 8x8 setup
#include "AVS_SetupFirstBlock.asm"
// Enable green channel only
mov (1) rAVS_8x8_HDR.2:ud nAVS_GREEN_CHANNEL_ONLY:ud
mov (16) mAVS_8x8_HDR.0:ud rAVS_8x8_HDR.0<8;8,1>:ud // Copy msg header and payload mirrors to MRFs
send (1) uwAVS_RESPONSE(0)<1> mAVS_8x8_HDR udDUMMY_NULL nSMPL_ENGINE nAVS_MSG_DSC_1CH+nSI_SRC_Y+nBI_CURRENT_SRC_Y
// Return Y in 4 GRFs
// 8x8 U and V sampling
// Enable red and blue channels
mov (1) rAVS_8x8_HDR.2:ud nAVS_RED_BLUE_CHANNELS:ud
mov (16) mAVS_8x8_HDR_UV.0:ud rAVS_8x8_HDR.0<8;8,1>:ud // Copy msg header and payload mirrors to MRFs
send (1) uwAVS_RESPONSE(4)<1> mAVS_8x8_HDR_UV udDUMMY_NULL nSMPL_ENGINE nAVS_MSG_DSC_2CH+nSI_SRC_UV+nBI_CURRENT_SRC_UV
// Return U and V in 8 GRFs
// 2nd 8x8 setup
#include "AVS_SetupSecondBlock.asm"
// 2nd 8x8 Y sampling
// Enable green channel only
mov (1) rAVS_8x8_HDR.2:ud nAVS_GREEN_CHANNEL_ONLY:ud
mov (16) mAVS_8x8_HDR.0:ud rAVS_8x8_HDR.0<8;8,1>:ud // Copy msg header and payload mirrors to MRFs
send (1) uwAVS_RESPONSE_2(0)<1> mAVS_8x8_HDR udDUMMY_NULL nSMPL_ENGINE nAVS_MSG_DSC_1CH+nSI_SRC_Y+nBI_CURRENT_SRC_Y
// 2nd 8x8 U and V sampling
// Enable red and blue channels
mov (1) rAVS_8x8_HDR.2:ud nAVS_RED_BLUE_CHANNELS:ud
mov (16) mAVS_8x8_HDR_UV.0:ud rAVS_8x8_HDR.0<8;8,1>:ud // Copy msg header and payload mirrors to MRFs
send (1) uwAVS_RESPONSE_2(4)<1> mAVS_8x8_HDR_UV udDUMMY_NULL nSMPL_ENGINE nAVS_MSG_DSC_2CH+nSI_SRC_UV+nBI_CURRENT_SRC_UV
// Return U and V in 8 GRFs
//------------------------------------------------------------------------------
// Unpacking sampler reads to 4:4:4 internal planar
//------------------------------------------------------------------------------
#include "PL2_AVS_IEF_Unpack_16x8.asm"
i965_drv_video/shaders/post_processing/Core_Kernels/PL2_AVS_IEF_8x4.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//---------- PL2_AVS_IEF_8x4.asm ----------
#include "AVS_IEF.inc"
//------------------------------------------------------------------------------
// 2 sampler reads for 8x8 Y each
// 1 sampler read for 8x8 U and 8x8 V (NV11\NV12 input surface)
//------------------------------------------------------------------------------
// 1st 8x8 setup
#include "AVS_SetupFirstBlock.asm"
// Enable green channel only
mov (1) rAVS_8x8_HDR.2:ud nAVS_GREEN_CHANNEL_ONLY:ud
mov (16) mAVS_8x8_HDR.0:ud rAVS_8x8_HDR.0<8;8,1>:ud // Copy msg header and payload mirrors to MRFs
send (1) uwAVS_RESPONSE(0)<1> mAVS_8x8_HDR udDUMMY_NULL nSMPL_ENGINE nAVS_MSG_DSC_1CH+nSI_SRC_Y+nBI_CURRENT_SRC_Y
// Return Y in 4 GRFs
// 8x8 U and V sampling
// Enable red and blue channels
//Only 8x4 wil be used
mov (1) rAVS_8x8_HDR.2:ud nAVS_RED_BLUE_CHANNELS:ud
// Calculate Chroma Step Size:
// for H direction: 16 Luma samples are covered by 8 Chroma samples. Thus Chroma_Step_X = 2 * Luma_Step_X
// for V direction: 8 Luma samples are covered by 8 Chroma samples. Thus Chroma_Step_Y = Luma_Step_Y
mul (1) rAVS_PAYLOAD.1:f fVIDEO_STEP_X:f 2.0:f // Step X for chroma
mov (16) mAVS_8x8_HDR_UV.0:ud rAVS_8x8_HDR.0<8;8,1>:ud // Copy msg header and payload mirrors to MRFs
send (1) uwAVS_RESPONSE(4)<1> mAVS_8x8_HDR_UV udDUMMY_NULL nSMPL_ENGINE nAVS_MSG_DSC_2CH+nSI_SRC_UV+nBI_CURRENT_SRC_UV
// Return U and V in 8 GRFs
// 2nd 8x8 setup
#include "AVS_SetupSecondBlock.asm"
// 2nd 8x8 Y sampling
// Enable green channel only
mov (1) rAVS_8x8_HDR.2:ud nAVS_GREEN_CHANNEL_ONLY:ud
mov (16) mAVS_8x8_HDR.0:ud rAVS_8x8_HDR.0<8;8,1>:ud // Copy msg header and payload mirrors to MRFs
send (1) uwAVS_RESPONSE_2(0)<1> mAVS_8x8_HDR udDUMMY_NULL nSMPL_ENGINE nAVS_MSG_DSC_1CH+nSI_SRC_Y+nBI_CURRENT_SRC_Y
//------------------------------------------------------------------------------
// Unpacking sampler reads to 4:2:0 internal planar
//------------------------------------------------------------------------------
#include "PL2_AVS_IEF_Unpack_8x4.asm"
i965_drv_video/shaders/post_processing/Core_Kernels/PL2_AVS_IEF_8x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//---------- PL2_AVS_IEF_8x8.asm ----------
#include "AVS_IEF.inc"
//------------------------------------------------------------------------------
// 2 sampler reads for 8x8 Y each
// 1 sampler read for 8x8 U and 8x8 V (NV11\NV12 input surface)
//------------------------------------------------------------------------------
// 1st 8x8 setup
#include "AVS_SetupFirstBlock.asm"
// Enable green channel only
mov (1) rAVS_8x8_HDR.2:ud nAVS_GREEN_CHANNEL_ONLY:ud
mov (16) mAVS_8x8_HDR.0:ud rAVS_8x8_HDR.0<8;8,1>:ud // Copy msg header and payload mirrors to MRFs
send (1) uwAVS_RESPONSE(0)<1> mAVS_8x8_HDR udDUMMY_NULL nSMPL_ENGINE nAVS_MSG_DSC_1CH+nSI_SRC_Y+nBI_CURRENT_SRC_Y
// Return Y in 4 GRFs
// 8x8 U and V sampling
// Enable red and blue channels
mov (1) rAVS_8x8_HDR.2:ud nAVS_RED_BLUE_CHANNELS:ud
// Calculate Chroma Step Size:
// for H direction: 16 Luma samples are covered by 8 Chroma samples. Thus Chroma_Step_X = 2 * Luma_Step_X
// for V direction: 8 Luma samples are covered by 8 Chroma samples. Thus Chroma_Step_Y = Luma_Step_Y
mul (1) rAVS_PAYLOAD.1:f fVIDEO_STEP_X:f 2.0:f // Step X for chroma
mov (16) mAVS_8x8_HDR_UV.0:ud rAVS_8x8_HDR.0<8;8,1>:ud // Copy msg header and payload mirrors to MRFs
send (1) uwAVS_RESPONSE(4)<1> mAVS_8x8_HDR_UV udDUMMY_NULL nSMPL_ENGINE nAVS_MSG_DSC_2CH+nSI_SRC_UV+nBI_CURRENT_SRC_UV
// Return U and V in 8 GRFs
// 2nd 8x8 setup
#include "AVS_SetupSecondBlock.asm"
// 2nd 8x8 Y sampling
// Enable green channel only
mov (1) rAVS_8x8_HDR.2:ud nAVS_GREEN_CHANNEL_ONLY:ud
mov (16) mAVS_8x8_HDR.0:ud rAVS_8x8_HDR.0<8;8,1>:ud // Copy msg header and payload mirrors to MRFs
send (1) uwAVS_RESPONSE_2(0)<1> mAVS_8x8_HDR udDUMMY_NULL nSMPL_ENGINE nAVS_MSG_DSC_1CH+nSI_SRC_Y+nBI_CURRENT_SRC_Y
//------------------------------------------------------------------------------
// Unpacking sampler reads to 4:2:2 internal planar
//------------------------------------------------------------------------------
#include "PL2_AVS_IEF_Unpack_8x8.asm"
i965_drv_video/shaders/post_processing/Core_Kernels/PL2_AVS_IEF_Unpack_8x4.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//---------- PL2_AVS_IEF_8x4.asm ----------
// Move first 8x8 words of Y to dest GRF at lower 8 words of each RGF.
$for(0; <8/2; 1) {
mov (8) uwDEST_Y(%1*2)<1> ubAVS_RESPONSE(%1,1)<16;4,2> // Copy high byte in a word
mov (8) uwDEST_Y(%1*2+1)<1> ubAVS_RESPONSE(%1,8+1)<16;4,2> // Copy high byte in a word
}
// Move 8x4 words of U to dest GRF (Copy high byte in a word)
mov (8) uwDEST_U(0)<1> ubAVS_RESPONSE(4,1)<16;4,2>
mov (8) uwDEST_U(0,8)<1> ubAVS_RESPONSE(5,1)<16;4,2>
mov (8) uwDEST_U(1)<1> ubAVS_RESPONSE(8,1)<16;4,2>
mov (8) uwDEST_U(1,8)<1> ubAVS_RESPONSE(9,1)<16;4,2>
// Move 8x4 words of V to dest GRF
mov (8) uwDEST_V(0)<1> ubAVS_RESPONSE(6,1)<16;4,2>
mov (8) uwDEST_V(0,8)<1> ubAVS_RESPONSE(7,1)<16;4,2>
mov (8) uwDEST_V(1)<1> ubAVS_RESPONSE(10,1)<16;4,2>
mov (8) uwDEST_V(1,8)<1> ubAVS_RESPONSE(11,1)<16;4,2>
// Move 2nd 8x8 words of Y to dest GRF at higher 8 words of each GRF.
$for(0; <8/2; 1) {
mov (8) uwDEST_Y(%1*2,8)<1> ubAVS_RESPONSE_2(%1,1)<16;4,2> // Copy high byte in a word
mov (8) uwDEST_Y(%1*2+1,8)<1> ubAVS_RESPONSE_2(%1,8+1)<16;4,2> // Copy high byte in a word
}
//------------------------------------------------------------------------------
// Re-define new # of lines
#undef nUV_NUM_OF_ROWS
#undef nY_NUM_OF_ROWS
#define nY_NUM_OF_ROWS 8
#define nUV_NUM_OF_ROWS 4
i965_drv_video/shaders/post_processing/Core_Kernels/PL2_AVS_IEF_Unpack_8x8.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//---------- PL2_AVS_IEF_8x8.asm ----------
// Move first 8x8 words of Y to dest GRF at lower 8 words of each RGF.
$for(0; <8/2; 1) {
mov (8) uwDEST_Y(%1*2)<1> ubAVS_RESPONSE(%1,1)<16;4,2> // Copy high byte in a word
mov (8) uwDEST_Y(%1*2+1)<1> ubAVS_RESPONSE(%1,8+1)<16;4,2> // Copy high byte in a word
}
// Move 8x8 words of U to dest GRF (Copy high byte in a word)
mov (8) uwDEST_U(0)<1> ubAVS_RESPONSE(4,1)<16;4,2>
mov (8) uwDEST_U(0,8)<1> ubAVS_RESPONSE(4,8+1)<16;4,2>
mov (8) uwDEST_U(1)<1> ubAVS_RESPONSE(5,1)<16;4,2>
mov (8) uwDEST_U(1,8)<1> ubAVS_RESPONSE(5,8+1)<16;4,2>
mov (8) uwDEST_U(2)<1> ubAVS_RESPONSE(8,1)<16;4,2>
mov (8) uwDEST_U(2,8)<1> ubAVS_RESPONSE(8,8+1)<16;4,2>
mov (8) uwDEST_U(3)<1> ubAVS_RESPONSE(9,1)<16;4,2>
mov (8) uwDEST_U(3,8)<1> ubAVS_RESPONSE(9,8+1)<16;4,2>
// Move 8x8 words of V to dest GRF
mov (8) uwDEST_V(0)<1> ubAVS_RESPONSE(6,1)<16;4,2>
mov (8) uwDEST_V(0,8)<1> ubAVS_RESPONSE(6,8+1)<16;4,2>
mov (8) uwDEST_V(1)<1> ubAVS_RESPONSE(7,1)<16;4,2>
mov (8) uwDEST_V(1,8)<1> ubAVS_RESPONSE(7,8+1)<16;4,2>
mov (8) uwDEST_V(2)<1> ubAVS_RESPONSE(10,1)<16;4,2>
mov (8) uwDEST_V(2,8)<1> ubAVS_RESPONSE(10,8+1)<16;4,2>
mov (8) uwDEST_V(3)<1> ubAVS_RESPONSE(11,1)<16;4,2>
mov (8) uwDEST_V(3,8)<1> ubAVS_RESPONSE(11,8+1)<16;4,2>
// Move 2nd 8x8 words of Y to dest GRF at higher 8 words of each GRF.
$for(0; <8/2; 1) {
mov (8) uwDEST_Y(%1*2,8)<1> ubAVS_RESPONSE_2(%1,1)<16;4,2> // Copy high byte in a word
mov (8) uwDEST_Y(%1*2+1,8)<1> ubAVS_RESPONSE_2(%1,8+1)<16;4,2> // Copy high byte in a word
}
//------------------------------------------------------------------------------
// Re-define new # of lines
#undef nUV_NUM_OF_ROWS
#undef nY_NUM_OF_ROWS
#define nY_NUM_OF_ROWS 8
#define nUV_NUM_OF_ROWS 8
i965_drv_video/shaders/post_processing/Core_Kernels/PL2_Scaling.asm 0 → 100644
View file @ 20975a94
/*
* All Video Processing kernels
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
//---------- PL2_Scaling.asm ----------
#include "Scaling.inc"
// Build 16 elements ramp in float32 and normalized it
// mov (8) SAMPLER_RAMP(0)<1> 0x76543210:v
// add (8) SAMPLER_RAMP(1)<1> SAMPLER_RAMP(0) 8.0:f
mov (4) SAMPLER_RAMP(0)<1> 0x48403000:vf //3, 2, 1, 0 in float vector
mov (4) SAMPLER_RAMP(0,4)<1> 0x5C585450:vf //7, 6, 5, 4 in float vector
add (8) SAMPLER_RAMP(1)<1> SAMPLER_RAMP(0) 8.0:f
//Module: PrepareScaleCoord.asm
// Setup for sampler msg hdr
mov (2) rMSGSRC.0<1>:ud 0:ud { NoDDClr } // Unused fields
mov (1) rMSGSRC.2<1>:ud 0:ud { NoDDChk } // Write and offset
// Calculate 16 v based on the step Y and vertical origin
mov (16) mfMSGPAYLOAD(2)<1> fSRC_VID_V_ORI<0;1,0>:f
mov (16) SCALE_COORD_Y<1>:f fSRC_VID_V_ORI<0;1,0>:f
// Calculate 16 u based on the step X and hori origin
// line (16) mfMSGPAYLOAD(0)<1> SCALE_STEP_X<0;1,0>:f SAMPLER_RAMP(0) // Assign to mrf directly
mov (16) acc0:f fSRC_VID_H_ORI<0;1,0>:f { Compr }
mac (16) mfMSGPAYLOAD(0)<1> fVIDEO_STEP_X<0;1,0>:f SAMPLER_RAMP(0) { Compr }
//Setup the constants for line instruction
mov (1) SCALE_LINE_P255<1>:f 255.0:f { NoDDClr } //{ NoDDClr, NoDDChk }
mov (1) SCALE_LINE_P0_5<1>:f 0.5:f { NoDDChk }
//------------------------------------------------------------------------------
$for (0; <nY_NUM_OF_ROWS; 1) {
// Read 16 sampled pixels and store them in float32 in 8 GRFs in the order of BGRA (VYUA).
mov (8) MSGHDR_SCALE.0:ud rMSGSRC.0<8;8,1>:ud // Copy msg header and payload mirrors to MRFs
send (16) SCALE_RESPONSE_YW(0)<1> MSGHDR_SCALE udDUMMY_NULL nSMPL_ENGINE SMPLR_MSG_DSC+nSI_SRC_SIMD16_Y+nBI_CURRENT_SRC_Y
send (16) SCALE_RESPONSE_UW(0)<1> MSGHDR_SCALE udDUMMY_NULL nSMPL_ENGINE SMPLR_MSG_DSC+nSI_SRC_SIMD16_UV+nBI_CURRENT_SRC_UV
// Calculate 16 v for next line
add (16) mfMSGPAYLOAD(2)<1> SCALE_COORD_Y<8;8,1>:f fVIDEO_STEP_Y<0;1,0>:f // Assign to mrf directly
add (16) SCALE_COORD_Y<1>:f SCALE_COORD_Y<8;8,1>:f fVIDEO_STEP_Y<0;1,0>:f // Assign to mrf directly
// Scale back to [0, 255], convert f to ud
line (16) acc0:f SCALE_LINE_P255<0;1,0>:f SCALE_RESPONSE_YF(0) { Compr } // Process B, V
mov (16) SCALE_RESPONSE_YD(0)<1> acc0:f { Compr }
line (16) acc0:f SCALE_LINE_P255<0;1,0>:f SCALE_RESPONSE_UF(0) { Compr } // Process B, V
mov (16) SCALE_RESPONSE_UD(0)<1> acc0:f { Compr }
line (16) acc0:f SCALE_LINE_P255<0;1,0>:f SCALE_RESPONSE_UF(2) { Compr } // Process B, V
mov (16) SCALE_RESPONSE_UD(2)<1> acc0:f { Compr }
mov (16) DEST_Y(%1)<1> SCALE_RESPONSE_YB(0) //possible error due to truncation - vK
mov (16) DEST_U(%1)<1> SCALE_RESPONSE_UB(0) //possible error due to truncation - vK
mov (16) DEST_V(%1)<1> SCALE_RESPONSE_UB(2) //possible error due to truncation - vK
}
#define nSRC_REGION nREGION_1
//------------------------------------------------------------------------------
i965_drv_video/shaders/post_processing/Makefile.am 0 → 100644
View file @ 20975a94
This diff is collapsed.
i965_drv_video/shaders/post_processing/null.g4a 0 → 100644
View file @ 20975a94
/* Just for test */
send(16) 0 acc0<1>UW g0<8,8,1>UW thread_spawner(0, 0, 0) mlen 1 rlen 0 {align1 EOT};
i965_drv_video/shaders/post_processing/null.g4b 0 → 100644
View file @ 20975a94
{ 0x00800031, 0x24001d28, 0x008d0000, 0x87100000 },
i965_drv_video/shaders/post_processing/null.g4b.gen5 0 → 100644
View file @ 20975a94
This diff is collapsed.
i965_drv_video/shaders/post_processing/nv12_avs_nv12.asm 0 → 100644
View file @ 20975a94
This diff is collapsed.
i965_drv_video/shaders/post_processing/nv12_avs_nv12.g4b.gen5 0 → 100644
View file @ 20975a94
This diff is collapsed.
i965_drv_video/shaders/post_processing/nv12_dndi_nv12.asm 0 → 100644
View file @ 20975a94
This diff is collapsed.
i965_drv_video/shaders/post_processing/nv12_dndi_nv12.g4b.gen5 0 → 100644
View file @ 20975a94
This diff is collapsed.
i965_drv_video/shaders/post_processing/nv12_load_save_nv12.asm 0 → 100644
View file @ 20975a94
This diff is collapsed.
i965_drv_video/shaders/post_processing/nv12_scaling_nv12.asm 0 → 100644
View file @ 20975a94
This diff is collapsed.
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment