Skip to content

Commit

Permalink
+add Base implementation of class ResizerBf16Bilinear.
Browse files Browse the repository at this point in the history
  • Loading branch information
@ermig1979
ermig1979 committed Dec 26, 2024
1 parent f2af786 commit 3bdb232
Show file tree
Hide file tree
Showing 5 changed files with 116 additions and 27 deletions.
1 change: 1 addition & 0 deletions docs/2025.html
View file
Original file line number Diff line number Diff line change
Expand Up @@ -46,6 +46,7 @@ <h5>New features</h5>
<li>AMX-BF16 kernel DepthwiseConvolution_k5p2d1s1w4 for class SynetMergedConvolution16b.</li>
<li>AMX-BF16 kernel DepthwiseConvolution_k3p1d1s1w8 for class SynetMergedConvolution16b.</li>
<li>AMX-BF16 kernel DepthwiseConvolution_k3p1d1s1w6 for class SynetMergedConvolution16b.</li>
<li>Base implementation of class ResizerBf16Bilinear.</li>
</ul>
<h5>Improving</h5>
<ul>
Expand Down
2 changes: 2 additions & 0 deletions src/Simd/SimdBaseResizer.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -39,6 +39,8 @@ namespace Simd
return new ResizerShortBilinear(param);
else if (param.IsFloatBilinear())
return new ResizerFloatBilinear(param);
else if (param.IsBf16Bilinear())
return new ResizerBf16Bilinear(param);
else if (param.IsByteBicubic())
return new ResizerByteBicubic(param);
else if (param.IsByteArea2x2())
Expand Down
106 changes: 85 additions & 21 deletions src/Simd/SimdBaseResizerBilinear.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -24,6 +24,7 @@
#include "Simd/SimdMemory.h"
#include "Simd/SimdResizer.h"
#include "Simd/SimdCopy.h"
#include "Simd/SimdBFloat16.h"

namespace Simd
{
Expand Down Expand Up @@ -128,7 +129,7 @@ namespace Simd
}
}

//---------------------------------------------------------------------
//-------------------------------------------------------------------------------------------------

ResizerShortBilinear::ResizerShortBilinear(const ResParam& param)
: Resizer(param)
Expand Down Expand Up @@ -247,25 +248,11 @@ namespace Simd
}
}

//---------------------------------------------------------------------
//-------------------------------------------------------------------------------------------------

ResizerFloatBilinear::ResizerFloatBilinear(const ResParam & param)
: Resizer(param)
static void EstimateIndexAlpha(const ResParam& param, size_t srcSize, size_t dstSize, size_t channels, int32_t* indices, float* alphas)
{
_ay.Resize(_param.dstH, false, _param.align);
_iy.Resize(_param.dstH, false, _param.align);
EstimateIndexAlpha(_param.srcH, _param.dstH, 1, _iy.data, _ay.data);
size_t rs = _param.dstW * _param.channels;
_ax.Resize(rs, false, _param.align);
_ix.Resize(rs, false, _param.align);
EstimateIndexAlpha(_param.srcW, _param.dstW, _param.channels, _ix.data, _ax.data);
_bx[0].Resize(rs, false, _param.align);
_bx[1].Resize(rs, false, _param.align);
}

void ResizerFloatBilinear::EstimateIndexAlpha(size_t srcSize, size_t dstSize, size_t channels, int32_t * indices, float * alphas)
{
if (_param.method == SimdResizeMethodBilinear)
if (param.method == SimdResizeMethodBilinear)
{
float scale = (float)srcSize / dstSize;
for (size_t i = 0; i < dstSize; ++i)
Expand All @@ -290,8 +277,8 @@ namespace Simd
alphas[offset] = alpha;
}
}
}
else if (_param.method == SimdResizeMethodBilinearCaffe)
}
else if (param.method == SimdResizeMethodBilinearCaffe)
{
float scale = dstSize > 1 ? float(srcSize - 1) / float(dstSize - 1) : 0.0f;
for (size_t i = 0; i < dstSize; ++i)
Expand All @@ -312,7 +299,7 @@ namespace Simd
}
}
}
else if (_param.method == SimdResizeMethodBilinearPytorch)
else if (param.method == SimdResizeMethodBilinearPytorch)
{
float scale = (float)srcSize / dstSize;
for (size_t i = 0; i < dstSize; ++i)
Expand Down Expand Up @@ -342,6 +329,22 @@ namespace Simd
assert(0);
}

//-------------------------------------------------------------------------------------------------

ResizerFloatBilinear::ResizerFloatBilinear(const ResParam & param)
: Resizer(param)
{
_ay.Resize(_param.dstH, false, _param.align);
_iy.Resize(_param.dstH, false, _param.align);
EstimateIndexAlpha(_param, _param.srcH, _param.dstH, 1, _iy.data, _ay.data);
size_t rs = _param.dstW * _param.channels;
_ax.Resize(rs, false, _param.align);
_ix.Resize(rs, false, _param.align);
EstimateIndexAlpha(_param, _param.srcW, _param.dstW, _param.channels, _ix.data, _ax.data);
_bx[0].Resize(rs, false, _param.align);
_bx[1].Resize(rs, false, _param.align);
}

void ResizerFloatBilinear::Run(const uint8_t * src, size_t srcStride, uint8_t * dst, size_t dstStride)
{
Run((const float*)src, srcStride / sizeof(float), (float*)dst, dstStride / sizeof(float));
Expand Down Expand Up @@ -386,6 +389,67 @@ namespace Simd
dst[dx] = pbx[0][dx]*fy0 + pbx[1][dx]*fy1;
}
}

//-------------------------------------------------------------------------------------------------

ResizerBf16Bilinear::ResizerBf16Bilinear(const ResParam& param)
: Resizer(param)
{
_ay.Resize(_param.dstH, false, _param.align);
_iy.Resize(_param.dstH, false, _param.align);
EstimateIndexAlpha(_param, _param.srcH, _param.dstH, 1, _iy.data, _ay.data);
size_t rs = _param.dstW * _param.channels;
_ax.Resize(rs, false, _param.align);
_ix.Resize(rs, false, _param.align);
EstimateIndexAlpha(_param, _param.srcW, _param.dstW, _param.channels, _ix.data, _ax.data);
_bx[0].Resize(rs, false, _param.align);
_bx[1].Resize(rs, false, _param.align);
}

void ResizerBf16Bilinear::Run(const uint8_t* src, size_t srcStride, uint8_t* dst, size_t dstStride)
{
Run((const uint16_t*)src, srcStride / sizeof(uint16_t), (uint16_t*)dst, dstStride / sizeof(uint16_t));
}

void ResizerBf16Bilinear::Run(const uint16_t* src, size_t srcStride, uint16_t* dst, size_t dstStride)
{
size_t cn = _param.channels;
size_t rs = _param.dstW * cn;
float* pbx[2] = { _bx[0].data, _bx[1].data };
int32_t prev = -2;
for (size_t dy = 0; dy < _param.dstH; dy++, dst += dstStride)
{
float fy1 = _ay[dy];
float fy0 = 1.0f - fy1;
int32_t sy = _iy[dy];
int32_t k = 0;

if (sy == prev)
k = 2;
else if (sy == prev + 1)
{
Swap(pbx[0], pbx[1]);
k = 1;
}

prev = sy;

for (; k < 2; k++)
{
float* pb = pbx[k];
const uint16_t* ps = src + (sy + k) * srcStride;
for (size_t dx = 0; dx < rs; dx++)
{
int32_t sx = _ix[dx];
float fx = _ax[dx];
pb[dx] = BFloat16ToFloat32(ps[sx]) * (1.0f - fx) + BFloat16ToFloat32(ps[sx + cn]) * fx;
}
}

for (size_t dx = 0; dx < rs; dx++)
dst[dx] = Float32ToBFloat16(pbx[0][dx] * fy0 + pbx[1][dx] * fy1);
}
}
}
}

24 changes: 22 additions & 2 deletions src/Simd/SimdResizer.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -71,6 +71,12 @@ namespace Simd
(method == SimdResizeMethodBilinear || method == SimdResizeMethodBilinearCaffe || method == SimdResizeMethodBilinearPytorch);
}

bool IsBf16Bilinear() const
{
return type == SimdResizeChannelBf16 &&
(method == SimdResizeMethodBilinear || method == SimdResizeMethodBilinearCaffe || method == SimdResizeMethodBilinearPytorch);
}

bool IsByteBicubic() const
{
return type == SimdResizeChannelByte && method == SimdResizeMethodBicubic;
Expand Down Expand Up @@ -180,8 +186,6 @@ namespace Simd
Array32i _ix, _iy;
Array32f _ax, _ay, _bx[2];

void EstimateIndexAlpha(size_t srcSize, size_t dstSize, size_t channels, int32_t * indices, float * alphas);

virtual void Run(const float * src, size_t srcStride, float * dst, size_t dstStride);

public:
Expand All @@ -192,6 +196,22 @@ namespace Simd

//-------------------------------------------------------------------------------------------------

class ResizerBf16Bilinear : public Resizer
{
protected:
Array32i _ix, _iy;
Array32f _ax, _ay, _bx[2];

virtual void Run(const uint16_t* src, size_t srcStride, uint16_t* dst, size_t dstStride);

public:
ResizerBf16Bilinear(const ResParam& param);

virtual void Run(const uint8_t* src, size_t srcStride, uint8_t* dst, size_t dstStride);
};

//-------------------------------------------------------------------------------------------------

const int32_t BICUBIC_RANGE = 1 << SIMD_RESIZER_BICUBIC_BITS;
const int32_t BICUBIC_SHIFT = SIMD_RESIZER_BICUBIC_BITS * 2;
const int32_t BICUBIC_ROUND = 1 << (BICUBIC_SHIFT - 1);
Expand Down
10 changes: 6 additions & 4 deletions src/Test/TestResize.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -144,7 +144,7 @@ namespace Test
assert(0);

View src(srcW, srcH, format, NULL, TEST_ALIGN(srcW));
if (format == View::Float)
if (type == SimdResizeChannelFloat)
FillRandom32f(src);
else if (type == SimdResizeChannelShort)
FillRandom16u(src);
Expand Down Expand Up @@ -184,9 +184,9 @@ namespace Test

TEST_EXECUTE_AT_LEAST_MIN_TIME(f2.Call(src, dst2, channels, type, method));

if (format == View::Float)
if (type == SimdResizeChannelFloat)
result = result && Compare(dst1, dst2, EPS, true, 64, DifferenceAbsolute);
else if (format == View::Float)
else if (type == SimdResizeChannelBf16)
{
View dst32f1(dstW, dstH, View::Float), dst32f2(dstW, dstH, View::Float);
for (size_t row = 0; row < dstH; row++)
Expand All @@ -196,7 +196,7 @@ namespace Test
}
result = result && Compare(dst32f1, dst32f2, EPS, true, 64, DifferenceAbsolute);
}
else if(format == View::Int16)
else if(type == SimdResizeChannelShort)
result = result && Compare(dst1, dst2, 1, true, 64);
else
result = result && Compare(dst1, dst2, 0, true, 64);
Expand Down Expand Up @@ -251,6 +251,8 @@ namespace Test
{
bool result = true;

result = result && ResizerAutoTest(SimdResizeMethodBilinear, SimdResizeChannelBf16, 16, f1, f2);

result = result && ResizerAutoTest(SimdResizeMethodNearest, SimdResizeChannelBf16, 1, f1, f2);
result = result && ResizerAutoTest(SimdResizeMethodNearest, SimdResizeChannelBf16, 3, f1, f2);
result = result && ResizerAutoTest(SimdResizeMethodNearest, SimdResizeChannelBf16, 8, f1, f2);
Expand Down

0 comments on commit 3bdb232

Please sign in to comment.