intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-09 06:23:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

Optimize copying buffers with misaligned pointers

Browse Source 
Resolves: NEO-5476
Signed-off-by: Maciej Dziuban <maciej.dziuban@intel.com>
This commit is contained in:
Maciej Dziuban
2021-01-26 10:38:20 +00:00
committed by Compute-Runtime-Automation
parent 9290637a8e
commit 566a761aaa
9 changed files with 116 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
shared/source/built_ins/kernels/copy_buffer_to_buffer.builtin_kernel
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2017-2020 Intel Corporation
* Copyright (C) 2017-2021 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -41,6 +41,27 @@ __kernel void CopyBufferToBufferMiddle(
vstore4(loaded, gid, pDst);
}
__kernel void CopyBufferToBufferMiddleMisaligned(
__global const uint* pSrc,
__global uint* pDst,
uint srcOffsetInBytes,
uint dstOffsetInBytes,
uint misalignmentInBits)
{
const size_t gid = get_global_id(0);
pDst += dstOffsetInBytes >> 2;
pSrc += srcOffsetInBytes >> 2;
const uint4 src0 = vload4(gid, pSrc);
const uint4 src1 = vload4(gid + 1, pSrc);
uint4 result;
result.x = (src0.x >> misalignmentInBits) | (src0.y << (32 - misalignmentInBits));
result.y = (src0.y >> misalignmentInBits) | (src0.z << (32 - misalignmentInBits));
result.z = (src0.z >> misalignmentInBits) | (src0.w << (32 - misalignmentInBits));
result.w = (src0.w >> misalignmentInBits) | (src1.x << (32 - misalignmentInBits));
vstore4(result, gid, pDst);
}
__kernel void CopyBufferToBufferRightLeftover(
const __global uchar* pSrc,
__global uchar* pDst,

23
shared/source/built_ins/kernels/copy_buffer_to_buffer_stateless.builtin_kernel
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2019-2020 Intel Corporation
* Copyright (C) 2019-2021 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -41,6 +41,27 @@ __kernel void CopyBufferToBufferMiddle(
vstore4(loaded, gid, pDst);
}
__kernel void CopyBufferToBufferMiddleMisaligned(
__global const uint* pSrc,
__global uint* pDst,
ulong srcOffsetInBytes,
ulong dstOffsetInBytes,
uint misalignmentInBits)
{
const size_t gid = get_global_id(0);
pDst += dstOffsetInBytes >> 2;
pSrc += srcOffsetInBytes >> 2;
const uint4 src0 = vload4(gid, pSrc);
const uint4 src1 = vload4(gid + 1, pSrc);
uint4 result;
result.x = (src0.x >> misalignmentInBits) | (src0.y << (32 - misalignmentInBits));
result.y = (src0.y >> misalignmentInBits) | (src0.z << (32 - misalignmentInBits));
result.z = (src0.z >> misalignmentInBits) | (src0.w << (32 - misalignmentInBits));
result.w = (src0.w >> misalignmentInBits) | (src1.x << (32 - misalignmentInBits));
vstore4(result, gid, pDst);
}
__kernel void CopyBufferToBufferRightLeftover(
const __global uchar* pSrc,
__global uchar* pDst,

23
shared/test/unit_test/test_files/5842966610900731025_images.cl → shared/test/unit_test/test_files/17473353783988150586_images.cl
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2020 Intel Corporation
* Copyright (C) 2020-2021 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -46,6 +46,27 @@ __kernel void CopyBufferToBufferMiddle(
vstore4(loaded, gid, pDst);
}
__kernel void CopyBufferToBufferMiddleMisaligned(
__global const uint* pSrc,
__global uint* pDst,
uint srcOffsetInBytes,
uint dstOffsetInBytes,
uint misalignmentInBits)
{
const size_t gid = get_global_id(0);
pDst += dstOffsetInBytes >> 2;
pSrc += srcOffsetInBytes >> 2;
const uint4 src0 = vload4(gid, pSrc);
const uint4 src1 = vload4(gid + 1, pSrc);
uint4 result;
result.x = (src0.x >> misalignmentInBits) | (src0.y << (32 - misalignmentInBits));
result.y = (src0.y >> misalignmentInBits) | (src0.z << (32 - misalignmentInBits));
result.z = (src0.z >> misalignmentInBits) | (src0.w << (32 - misalignmentInBits));
result.w = (src0.w >> misalignmentInBits) | (src1.x << (32 - misalignmentInBits));
vstore4(result, gid, pDst);
}
__kernel void CopyBufferToBufferRightLeftover(
const __global uchar* pSrc,
__global uchar* pDst,

2
shared/test/unit_test/test_files/17034883254829439342_options.txt → shared/test/unit_test/test_files/17473353783988150586_images_options.txt
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2020 Intel Corporation
* Copyright (C) 2020-2021 Intel Corporation
*
* SPDX-License-Identifier: MIT
*

23
shared/test/unit_test/test_files/17034883254829439342.cl → shared/test/unit_test/test_files/2175211656417857798.cl
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2020 Intel Corporation
* Copyright (C) 2020-2021 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -46,6 +46,27 @@ __kernel void CopyBufferToBufferMiddle(
vstore4(loaded, gid, pDst);
}
__kernel void CopyBufferToBufferMiddleMisaligned(
__global const uint* pSrc,
__global uint* pDst,
uint srcOffsetInBytes,
uint dstOffsetInBytes,
uint misalignmentInBits)
{
const size_t gid = get_global_id(0);
pDst += dstOffsetInBytes >> 2;
pSrc += srcOffsetInBytes >> 2;
const uint4 src0 = vload4(gid, pSrc);
const uint4 src1 = vload4(gid + 1, pSrc);
uint4 result;
result.x = (src0.x >> misalignmentInBits) | (src0.y << (32 - misalignmentInBits));
result.y = (src0.y >> misalignmentInBits) | (src0.z << (32 - misalignmentInBits));
result.z = (src0.z >> misalignmentInBits) | (src0.w << (32 - misalignmentInBits));
result.w = (src0.w >> misalignmentInBits) | (src1.x << (32 - misalignmentInBits));
vstore4(result, gid, pDst);
}
__kernel void CopyBufferToBufferRightLeftover(
const __global uchar* pSrc,
__global uchar* pDst,

2
shared/test/unit_test/test_files/5842966610900731025_images_options.txt → shared/test/unit_test/test_files/2175211656417857798_options.txt
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2020 Intel Corporation
* Copyright (C) 2020-2021 Intel Corporation
*
* SPDX-License-Identifier: MIT
*