-
Notifications
You must be signed in to change notification settings - Fork 54
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
update the scan and the copy examples
Change-Id: Idd886e2bfbd0b2df7997d6be3b2eba46c59f1941
- Loading branch information
gangche1
committed
Jan 8, 2020
1 parent
06b0d73
commit 2c05d2b
Showing
32 changed files
with
2,107 additions
and
681 deletions.
There are no files selected for viewing
265 changes: 228 additions & 37 deletions
265
test/external_contribution/CopyPipelineHigh6/CPipeline.cpp
Large diffs are not rendered by default.
Oops, something went wrong.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
234 changes: 234 additions & 0 deletions
234
test/external_contribution/CopyPipelineHigh6/ErrorDiffusion/ErrorDiffusion_genx.cpp
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,234 @@ | ||
| /* | ||
| * Copyright (c) 2017, Intel Corporation | ||
| * | ||
| * Permission is hereby granted, free of charge, to any person obtaining a | ||
| * copy of this software and associated documentation files (the "Software"), | ||
| * to deal in the Software without restriction, including without limitation | ||
| * the rights to use, copy, modify, merge, publish, distribute, sublicense, | ||
| * and/or sell copies of the Software, and to permit persons to whom the | ||
| * Software is furnished to do so, subject to the following conditions: | ||
| * | ||
| * The above copyright notice and this permission notice shall be included | ||
| * in all copies or substantial portions of the Software. | ||
| * | ||
| * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS | ||
| * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | ||
| * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | ||
| * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR | ||
| * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, | ||
| * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR | ||
| * OTHER DEALINGS IN THE SOFTWARE. | ||
| */ | ||
| #include <cm/cm.h> | ||
|
|
||
| #define BLKW 8 | ||
| #define BLKH 4 | ||
| #define ITERATION 4 | ||
|
|
||
| #define X 0 | ||
| #define Y 1 | ||
|
|
||
| /* | ||
| * Error diffusion push error to neighbour pixels, and result the algorithm need | ||
| * to run in sequence. Using following pixel processing order allow parallelism | ||
| * but not violate the dependency issue. | ||
| * Using pixelRowIndex and pixelColIndex to determine the pixel sequence to | ||
| * process the error diffusion | ||
| * | ||
| * Some pixels can process together, but some can't. To reduce the conditional | ||
| * checking, if the pixel can't parallel with other pixel, repeat to process the | ||
| * same pixel. For SIMD, it isn't any performance penalty | ||
| */ | ||
|
|
||
| short pixelLocX[32] ={ 0,1,2,3,4,5,6,7,-2,-1,0,1,2,3,4,5, | ||
| -4,-3,-2,-1,0,1,2,3,-6,-5,-4,-3,-2,-1,0,1 | ||
| }; | ||
| short pixelLocY[32] ={ 0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1, | ||
| 2,2,2,2,2,2,2,2,3,3,3,3,3,3,3,3 | ||
| }; | ||
| uchar weight[4] = {1,5,3,7}; | ||
|
|
||
| extern "C" _GENX_MAIN_ void | ||
| ErrorDiffusion_GENX( | ||
| SurfaceIndex SrcSI, | ||
| SurfaceIndex ErrBufRowSI, | ||
| SurfaceIndex ErrBufColSI, | ||
| SurfaceIndex DstSI, | ||
| int widthBlk | ||
| ) | ||
| { | ||
| matrix<uint, BLKH, 8> input; | ||
| matrix<uint, BLKH, 8> output; | ||
| matrix<short, 5, 64> errBuf = 0; | ||
| vector<short, 64> errRowTemp; | ||
| vector<short, 2> pos, pos4; | ||
| vector<uint, 32> inputVectorX(pixelLocX); | ||
| vector<uint, 32> inputVectorY(pixelLocY); | ||
| matrix<uint, 4, 16> tmpInput; | ||
|
|
||
| cm_wait(); | ||
|
|
||
| pos(X) = get_thread_origin_x(); | ||
| pos(Y) = get_thread_origin_y(); | ||
|
|
||
| pos4(X) = pos(X) * BLKW; | ||
| pos4(Y) = pos(Y) * BLKH; | ||
|
|
||
| inputVectorX += pos4(X); | ||
| inputVectorY += pos4(Y); | ||
|
|
||
| read_typed(SrcSI, CM_R_ENABLE, tmpInput.select<1,1,16,1>(0,0), | ||
| inputVectorX.select<16,1>(0), inputVectorY.select<16,1>(0)); | ||
| input.select<1,1,8,1>(0,0) = tmpInput.select<1,1,8,1>(0,0); | ||
| input.select<1,1,8,1>(1,0) = tmpInput.select<1,1,8,1>(0,8); | ||
|
|
||
| read_typed(SrcSI, CM_R_ENABLE, tmpInput.select<1,1,16,1>(0,0), | ||
| inputVectorX.select<16,1>(16), inputVectorY.select<16,1>(16)); | ||
| input.select<1,1,8,1>(2,0) = tmpInput.select<1,1,8,1>(0,0); | ||
| input.select<1,1,8,1>(3,0) = tmpInput.select<1,1,8,1>(0,8); | ||
|
|
||
| // Need to update, try to use single column for column error buffer and | ||
| // single row for row buffer (single row for row buffer may not possible) | ||
| int colErrBufOffset = (pos(Y))*128; | ||
| int rowErrBufOffset = (pos(X)+1)*64-24; | ||
|
|
||
|
|
||
| if (pos(Y) != 0) | ||
| read(DWALIGNED(ErrBufRowSI), rowErrBufOffset, errRowTemp.select<40,1>(0)); | ||
| else | ||
| errRowTemp.select<40,1>(0) = 0; | ||
|
|
||
| errBuf.select<1,1,40,1>(0,24) = errRowTemp.select<40,1>(0); | ||
|
|
||
| if (pos(X) != 0) | ||
| read(ErrBufColSI, colErrBufOffset , errRowTemp.select<40,1>(0)); | ||
| else | ||
| errRowTemp.select<40,1>(0) = 0; | ||
|
|
||
| errBuf.select<1,1,12,1>(1,16) = errRowTemp.select<12,1>(0); | ||
| errBuf.select<1,1,12,1>(2,8) = errRowTemp.select<12,1>(12); | ||
| errBuf.select<1,1,12,1>(3,0) = errRowTemp.select<12,1>(24); | ||
| errBuf.select<1,1,4,1>(4,0) = errRowTemp.select<4,1>(36); | ||
|
|
||
| // The read extra one column to left and right, so, it will directly map into | ||
| // error buffer | ||
| // | ||
| // The input buffer is a diaganol box, but we read a square to wrap this box | ||
|
|
||
| // The read extra one column to left and right, so, it will directly map into | ||
| // error buffer | ||
| // The input buffer is a diaganol box, but we read a square to wrap this box | ||
| // ________ | ||
| // _| _| | ||
| // _| _| | ||
| // _| _| | ||
| //|_______ | | ||
| // | ||
| vector<uint, 8> oldpixel, newpixel; | ||
|
|
||
| vector_ref<uchar, 32> oldpixelc = oldpixel.format<uchar>(); | ||
| vector_ref<uchar, 32> newpixelc = newpixel.format<uchar>(); | ||
| vector<float, 32> fxs; | ||
|
|
||
| vector<float, 4> weightf(weight); | ||
|
|
||
| weightf =weightf/16.0f; | ||
|
|
||
| // this is error buffer size matrix<short, 5, 64> errBuf = 0; | ||
| int colIndex, rowIndex; | ||
| #pragma unroll | ||
| for (int i=0; i < ITERATION; i++) | ||
| { | ||
| fxs = 0; | ||
| // oldpixel should be from input one row | ||
| oldpixel = input.select<1,1,8,1>(i,0); | ||
|
|
||
| colIndex = 24 - i*8; | ||
| rowIndex = i; | ||
| if (pos(X) == widthBlk) | ||
| errBuf.select<5,1,36,1>(0, 28) = 0; | ||
|
|
||
| fxs += errBuf.select<1,1,32,1>(rowIndex,colIndex) * weightf[0]; | ||
|
|
||
| fxs += errBuf.select<1,1,32,1>(rowIndex,colIndex+4) * weightf[1]; | ||
|
|
||
| fxs += errBuf.select<1,1,32,1>(rowIndex,colIndex+8) * weightf[2]; | ||
|
|
||
| if (pos(X) == 0) | ||
| { | ||
| if (i == 1) | ||
| fxs.select<8,1>(0) = 0; | ||
| else if (i == 2) | ||
| fxs.select<16,1>(0) = 0; | ||
| else if (i == 3) | ||
| fxs.select<24,1>(0) = 0; | ||
| } | ||
|
|
||
| rowIndex += 1; | ||
|
|
||
| vector<float, 4> pixelleft (errBuf.select<1,1,4,1>(rowIndex, colIndex)); | ||
| vector<float, 4> singlefxs; | ||
| int rowi; | ||
|
|
||
| #pragma unroll | ||
| for (int ii=0; ii < 8; ii++) | ||
| { | ||
| rowi = ii*4; | ||
|
|
||
| singlefxs = pixelleft * weightf[3]; | ||
|
|
||
| singlefxs = singlefxs + fxs.select<4,1>(rowi) + | ||
| oldpixelc.select<4,1>(rowi); | ||
|
|
||
| newpixelc.select<4,1>(rowi) = (singlefxs > 127.0f)* 0xff; | ||
| fxs.select<4,1>(rowi) = cm_rndd<float>(singlefxs - newpixelc.select<4,1>(rowi)); | ||
| pixelleft = fxs.select<4,1>(rowi); | ||
|
|
||
| } | ||
|
|
||
| // Temporary workaround the right border incorrect data | ||
| if (pos(X) == widthBlk) | ||
| { | ||
| if (i == 1) | ||
| fxs.select<24,1>(8) = 0; | ||
| else if (i == 2) | ||
| fxs.select<16,1>(16) = 0; | ||
| else if (i == 3) | ||
| fxs.select<8,1>(24) = 0; | ||
| } | ||
| errBuf.select<1,1,32,1>(rowIndex,colIndex+4) = fxs; | ||
|
|
||
| output.select<1,1,8,1>(rowIndex-1, 0) = newpixel; | ||
|
|
||
| } | ||
|
|
||
| //output write need to use scatter write as it index is | ||
| matrix<uint, 4, 16> tmpOut; | ||
|
|
||
| tmpOut.select<1,1,8,1>(0,0) = output.select<1,1,8,1>(0,0); | ||
| tmpOut.select<1,1,8,1>(0,8) = output.select<1,1,8,1>(1,0); | ||
|
|
||
| write_typed(DstSI, CM_R_ENABLE, tmpOut, | ||
| inputVectorX.select<16,1>(0), inputVectorY.select<16,1>(0)); | ||
|
|
||
| tmpOut.select<1,1,8,1>(0,0) = output.select<1,1,8,1>(2,0); | ||
| tmpOut.select<1,1,8,1>(0,8) = output.select<1,1,8,1>(3,0); | ||
| write_typed(DstSI, CM_R_ENABLE, tmpOut, | ||
| inputVectorX.select<16,1>(16), inputVectorY.select<16,1>(16)); | ||
|
|
||
| vector<short, 64> errBufColOut; | ||
|
|
||
| errBufColOut.select<12,1>(0) = errBuf.select<1,1,12,1>(1,48); | ||
| errBufColOut.select<12,1>(12) = errBuf.select<1,1,12,1>(2,40); | ||
| errBufColOut.select<12,1>(24) = errBuf.select<1,1,12,1>(3,32); | ||
| errBufColOut.select<4,1>(36) = errBuf.select<1,1,4,1>(4,32); | ||
|
|
||
| colErrBufOffset = (pos(Y))*128; | ||
| rowErrBufOffset = (pos(X))*64; | ||
|
|
||
| write(ErrBufRowSI, rowErrBufOffset, errBuf.select<1,1,32,1>(4,4).format<short>()); | ||
| write(ErrBufColSI, colErrBufOffset, errBufColOut.select<64,1>(0)); | ||
|
|
||
| cm_fence(); | ||
| cm_signal(); | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.