From 10d79752d05c990575f30a9bfa6b6e225d9117f6 Mon Sep 17 00:00:00 2001
From: Maxence Naud <maxence.naud@cea.fr>
Date: Sun, 20 Oct 2024 14:31:29 +0000
Subject: [PATCH] Resolve "Missing ``dilation`` parameter in Conv[DepthWise]
forward kernel"
---
.../operator/ConvDepthWiseImpl_kernels.hpp | 42 +++++++++-------
.../backend/cpu/operator/ConvImpl_kernels.hpp | 48 +++++++++++--------
2 files changed, 55 insertions(+), 35 deletions(-)
diff --git a/include/aidge/backend/cpu/operator/ConvDepthWiseImpl_kernels.hpp b/include/aidge/backend/cpu/operator/ConvDepthWiseImpl_kernels.hpp
index ff9bb148..39a35703 100644
--- a/include/aidge/backend/cpu/operator/ConvDepthWiseImpl_kernels.hpp
+++ b/include/aidge/backend/cpu/operator/ConvDepthWiseImpl_kernels.hpp
@@ -38,7 +38,7 @@ namespace Aidge {
*/
template <class I, class W, class B, class O>
void ConvDepthWiseImpl1D_cpu_forward_kernel(const std::array<DimSize_t, 1>& strideDims,
- const std::array<DimSize_t, 1>& /*dilationDims*/,
+ const std::array<DimSize_t, 1>& dilationDims,
const std::array<DimSize_t, 1>& kernelDims,
const std::array<DimSize_t, 3>& inputDims,
const void *input_,
@@ -56,6 +56,8 @@ void ConvDepthWiseImpl1D_cpu_forward_kernel(const std::array<DimSize_t, 1>& stri
const std::size_t oxSize =
static_cast<std::size_t>(std::floor(static_cast<float>(inputDims[2] - kernelDims[0] + strideDims[0]) /
static_cast<float>(strideDims[0])));
+ const DimSize_t dilated_kernel_x = dilationDims[0]*(kernelDims[0] - 1) + 1;
+
// TODO: kernel computation
// output (batch, outCh, Xout, Yout)
@@ -71,15 +73,17 @@ void ConvDepthWiseImpl1D_cpu_forward_kernel(const std::array<DimSize_t, 1>& stri
const std::size_t iIndex = (ch + batch*inputDims[1]) * inputDims[2];
const std::size_t wIndex = ch * kernelDims[0];
for (std::size_t ox = 0; ox < oxSize; ++ox) {
- const signedsize difx = static_cast<signedsize>(- ox * strideDims[0]);
- const std::size_t sxMin = static_cast<std::size_t>(std::max(difx, signedsize(0)));
- const std::size_t sxMax = (static_cast<signedsize>(inputDims[2]) + difx) < 0 ? 0 : ((inputDims[2] + difx) > kernelDims[0] ? kernelDims[0] : inputDims[2] + difx);
+ // const signedsize difx = static_cast<signedsize>(- ox * strideDims[0]);
+ // const std::size_t sxMin = static_cast<std::size_t>(std::max(difx, signedsize(0)));
+ // const std::size_t sxMax = (static_cast<signedsize>(inputDims[2]) + difx) < 0 ? 0 : ((inputDims[2] + difx) > kernelDims[0] ? kernelDims[0] : inputDims[2] + difx);
+ const std::size_t sxMin = 0;
+ const std::size_t sxMax = dilated_kernel_x;
const std::size_t oIndexFull = oIndex + ox;
const signedsize ix = static_cast<signedsize>(ox * strideDims[0]);
- for (std::size_t sx = sxMin; sx < sxMax; ++sx) {
+ for (std::size_t sx = sxMin; sx*dilationDims[0] < sxMax; ++sx) {
output[oIndexFull] += weights[wIndex + sx] *
- input[iIndex + static_cast<std::size_t>(ix+static_cast<signedsize>(sx))];
+ input[iIndex + static_cast<std::size_t>(ix+static_cast<signedsize>(sx*dilationDims[0]))];
}
}
}
@@ -113,7 +117,7 @@ REGISTRAR(ConvDepthWiseImpl1D_cpu,
*/
template <class I, class W, class B, class O>
void ConvDepthWiseImpl2D_cpu_forward_kernel(const std::array<DimSize_t, 2>& strideDims,
- const std::array<DimSize_t, 2>& /*dilationDims*/,
+ const std::array<DimSize_t, 2>& dilationDims,
const std::array<DimSize_t, 2>& kernelDims,
const std::array<DimSize_t, 4>& inputDims,
const void *input_,
@@ -132,10 +136,12 @@ void ConvDepthWiseImpl2D_cpu_forward_kernel(const std::array<DimSize_t, 2>& stri
const std::size_t oxSize =
static_cast<std::size_t>(std::floor(static_cast<float>(inputDims[2] - kernelDims[0] + strideDims[0]) /
static_cast<float>(strideDims[0])));
+ const DimSize_t dilated_kernel_x = dilationDims[0]*(kernelDims[0] - 1) + 1;
// output W size
const std::size_t oySize =
static_cast<std::size_t>(std::floor(static_cast<float>(inputDims[3] - kernelDims[1] + strideDims[1]) /
static_cast<float>(strideDims[1])));
+ const DimSize_t dilated_kernel_y = dilationDims[1]*(kernelDims[1] - 1) + 1;
// TODO: kernel computation
// output (batch, outCh, Xout, Yout)
@@ -151,13 +157,17 @@ void ConvDepthWiseImpl2D_cpu_forward_kernel(const std::array<DimSize_t, 2>& stri
const std::size_t iIndex = (ch + batch*inputDims[1]) * inputDims[2] * inputDims[3];
const std::size_t wIndex = ch * kernelDims[0] * kernelDims[1];
for (std::size_t ox = 0; ox < oxSize; ++ox) {
- const signedsize difx = static_cast<signedsize>(- ox * strideDims[0]);
- const std::size_t sxMin = static_cast<std::size_t>(std::max(difx, signedsize(0)));
- const std::size_t sxMax = (static_cast<signedsize>(inputDims[2]) + difx) < 0 ? 0 : ((inputDims[2] + difx) > kernelDims[0] ? kernelDims[0] : inputDims[2] + difx);
+ // const signedsize difx = static_cast<signedsize>(- ox * strideDims[0]);
+ // const std::size_t sxMin = static_cast<std::size_t>(std::max(difx, signedsize(0)));
+ // const std::size_t sxMax = (static_cast<signedsize>(inputDims[2]) + difx) < 0 ? 0 : ((inputDims[2] + difx) > kernelDims[0] ? kernelDims[0] : inputDims[2] + difx);
+ const std::size_t sxMin = 0;
+ const std::size_t sxMax = dilated_kernel_x;
for (std::size_t oy = 0; oy < oySize; ++oy) {
- const signedsize dify = static_cast<signedsize>(- oy * strideDims[1]);
- const std::size_t syMin = static_cast<std::size_t>(std::max(dify, signedsize(0)));
- const std::size_t syMax = (static_cast<signedsize>(inputDims[3]) + dify) < 0 ? 0 : ((inputDims[3] + dify) > kernelDims[1] ? kernelDims[1] : inputDims[3] + dify);
+ // const signedsize dify = static_cast<signedsize>(- oy * strideDims[1]);
+ // const std::size_t syMin = static_cast<std::size_t>(std::max(dify, signedsize(0)));
+ // const std::size_t syMax = (static_cast<signedsize>(inputDims[3]) + dify) < 0 ? 0 : ((inputDims[3] + dify) > kernelDims[1] ? kernelDims[1] : inputDims[3] + dify);
+ const std::size_t syMin = 0;
+ const std::size_t syMax = dilated_kernel_y;
const std::size_t oIndexFull = oIndex + ox*oySize + oy;
const signedsize ix = static_cast<signedsize>(ox * strideDims[0]);
const signedsize iy = static_cast<signedsize>(oy * strideDims[1]);
@@ -173,10 +183,10 @@ void ConvDepthWiseImpl2D_cpu_forward_kernel(const std::array<DimSize_t, 2>& stri
weights[wIndex + 2*kernelDims[1] + 1] * input[iIndex + static_cast<std::size_t>(ix+2)*inputDims[3] + static_cast<std::size_t>(iy+1)] +
weights[wIndex + 2*kernelDims[1] + 2] * input[iIndex + static_cast<std::size_t>(ix+2)*inputDims[3] + static_cast<std::size_t>(iy+2)]);
} else {
- for (std::size_t sx = sxMin; sx < sxMax; ++sx) {
- for (std::size_t sy = syMin; sy < syMax; ++sy) {
+ for (std::size_t sx = sxMin; sx*dilationDims[0] < sxMax; ++sx) {
+ for (std::size_t sy = syMin; sy*dilationDims[1] < syMax; ++sy) {
output[oIndexFull] += weights[wIndex + sx*kernelDims[1] + sy] *
- input[iIndex + static_cast<std::size_t>(ix+static_cast<signedsize>(sx))*inputDims[3] + static_cast<std::size_t>(iy+static_cast<signedsize>(sy))];
+ input[iIndex + static_cast<std::size_t>(ix+static_cast<signedsize>(sx*dilationDims[0]))*inputDims[3] + static_cast<std::size_t>(iy+static_cast<signedsize>(sy*dilationDims[1]))];
}
}
}
diff --git a/include/aidge/backend/cpu/operator/ConvImpl_kernels.hpp b/include/aidge/backend/cpu/operator/ConvImpl_kernels.hpp
index cc3bd57c..e800c252 100644
--- a/include/aidge/backend/cpu/operator/ConvImpl_kernels.hpp
+++ b/include/aidge/backend/cpu/operator/ConvImpl_kernels.hpp
@@ -40,7 +40,7 @@ namespace Aidge {
*/
template <class I, class W, class B, class O>
void ConvImpl1D_cpu_forward_kernel(const std::array<DimSize_t, 1>& strideDims,
- const std::array<DimSize_t, 1>& /*dilationDims*/,
+ const std::array<DimSize_t, 1>& dilationDims,
const std::array<DimSize_t, 1>& kernelDims,
const std::array<DimSize_t, 3>& inputDims,
DimSize_t outChannels,
@@ -57,8 +57,9 @@ void ConvImpl1D_cpu_forward_kernel(const std::array<DimSize_t, 1>& strideDims,
// output H size
const std::size_t oxSize =
- static_cast<std::size_t>(std::floor(static_cast<float>(inputDims[2] - kernelDims[0] + strideDims[0]) /
+ static_cast<std::size_t>(std::floor(static_cast<float>(inputDims[2] - dilationDims[0]*(kernelDims[0] - 1) - 1 + strideDims[0]) /
static_cast<float>(strideDims[0])));
+ const DimSize_t dilated_kernel_x = dilationDims[0]*(kernelDims[0] - 1) + 1;
// TODO: kernel computation
// output (batch, outCh, Xout, Yout)
@@ -76,15 +77,17 @@ void ConvImpl1D_cpu_forward_kernel(const std::array<DimSize_t, 1>& strideDims,
const std::size_t iIndex = (inCh + batch*inputDims[1]) * inputDims[2];
const std::size_t wIndex = (inCh + outCh*inputDims[1]) * kernelDims[0];
for (std::size_t ox = 0; ox < oxSize; ++ox) {
- const signedsize difx = static_cast<signedsize>(- ox * strideDims[0]);
- const std::size_t sxMin = static_cast<std::size_t>(std::max(difx, signedsize(0)));
- const std::size_t sxMax = (static_cast<signedsize>(inputDims[2]) + difx) < 0 ? 0 : ((inputDims[2] + difx) > kernelDims[0] ? kernelDims[0] : inputDims[2] + difx);
+ // const signedsize difx = static_cast<signedsize>(- ox * strideDims[0]);
+ // const std::size_t sxMin = static_cast<std::size_t>(std::max(difx, signedsize(0)));
+ // const std::size_t sxMax = (static_cast<signedsize>(inputDims[2]) + difx) < 0 ? 0 : ((inputDims[2] + difx) > kernelDims[0] ? kernelDims[0] : inputDims[2] + difx);
+ const std::size_t sxMin = 0;
+ const std::size_t sxMax = dilated_kernel_x;
const std::size_t oIndexFull = oIndex + ox;
const signedsize ix = static_cast<signedsize>(ox * strideDims[0]);
- for (std::size_t sx = sxMin; sx < sxMax; ++sx) {
+ for (std::size_t sx = sxMin; sx*dilationDims[0] < sxMax; ++sx) {
output[oIndexFull] += weights[wIndex + sx] *
- input[iIndex + static_cast<std::size_t>(ix+static_cast<signedsize>(sx))];
+ input[iIndex + static_cast<std::size_t>(ix+static_cast<signedsize>(sx*dilationDims[0]))];
}
}
}
@@ -122,7 +125,7 @@ REGISTRAR(ConvImpl1D_cpu,
*/
template <class I, class W, class B, class O>
void ConvImpl2D_cpu_forward_kernel(const std::array<DimSize_t, 2>& strideDims,
- const std::array<DimSize_t, 2>& /*dilationDims*/,
+ const std::array<DimSize_t, 2>& dilationDims,
const std::array<DimSize_t, 2>& kernelDims,
const std::array<DimSize_t, 4> &inputDims,
DimSize_t outChannels,
@@ -139,12 +142,15 @@ void ConvImpl2D_cpu_forward_kernel(const std::array<DimSize_t, 2>& strideDims,
// output H size
const std::size_t oxSize =
- static_cast<std::size_t>(std::floor(static_cast<float>(inputDims[2] - kernelDims[0] + strideDims[0]) /
+ static_cast<std::size_t>(std::floor(static_cast<float>(inputDims[2] - dilationDims[0]*(kernelDims[0] - 1) - 1 + strideDims[0]) /
static_cast<float>(strideDims[0])));
+ const DimSize_t dilated_kernel_x = dilationDims[0]*(kernelDims[0] - 1) + 1;
// output W size
const std::size_t oySize =
- static_cast<std::size_t>(std::floor(static_cast<float>(inputDims[3] - kernelDims[1] + strideDims[1]) /
+ static_cast<std::size_t>(std::floor(static_cast<float>(inputDims[3] - dilationDims[1]*(kernelDims[1] - 1) - 1 + strideDims[1]) /
static_cast<float>(strideDims[1])));
+ const DimSize_t dilated_kernel_y = dilationDims[1]*(kernelDims[1] - 1) + 1;
+
// TODO: kernel computation
// output (batch, outCh, Xout, Yout)
@@ -162,13 +168,17 @@ void ConvImpl2D_cpu_forward_kernel(const std::array<DimSize_t, 2>& strideDims,
const std::size_t iIndex = (inCh + batch*inputDims[1]) * inputDims[2] * inputDims[3];
const std::size_t wIndex = (inCh + outCh*inputDims[1]) * kernelDims[0] * kernelDims[1];
for (std::size_t ox = 0; ox < oxSize; ++ox) {
- const signedsize difx = static_cast<signedsize>(- ox * strideDims[0]);
- const std::size_t sxMin = static_cast<std::size_t>(std::max(difx, signedsize(0)));
- const std::size_t sxMax = (static_cast<signedsize>(inputDims[2]) + difx) < 0 ? 0 : ((inputDims[2] + difx) > kernelDims[0] ? kernelDims[0] : inputDims[2] + difx);
+ // const signedsize difx = static_cast<signedsize>(- ox * strideDims[0]);
+ // const std::size_t sxMin = static_cast<std::size_t>(std::max(difx, signedsize(0)));
+ // const std::size_t sxMax = (static_cast<signedsize>(inputDims[2]) + difx) < 0 ? 0 : ((inputDims[2] + difx) > kernelDims[0] ? kernelDims[0] : inputDims[2] + difx);
+ const std::size_t sxMin = 0;
+ const std::size_t sxMax = dilated_kernel_x;
for (std::size_t oy = 0; oy < oySize; ++oy) {
- const signedsize dify = static_cast<signedsize>(- oy * strideDims[1]);
- const std::size_t syMin = static_cast<std::size_t>(std::max(dify, signedsize(0)));
- const std::size_t syMax = (static_cast<signedsize>(inputDims[3]) + dify) < 0 ? 0 : ((inputDims[3] + dify) > kernelDims[1] ? kernelDims[1] : inputDims[3] + dify);
+ // const signedsize dify = static_cast<signedsize>(- oy * strideDims[1]);
+ // const std::size_t syMin = static_cast<std::size_t>(std::max(dify, signedsize(0)));
+ // const std::size_t syMax = (static_cast<signedsize>(inputDims[3]) + dify) < 0 ? 0 : ((inputDims[3] + dify) > kernelDims[1] ? kernelDims[1] : inputDims[3] + dify);
+ const std::size_t syMin = 0;
+ const std::size_t syMax = dilated_kernel_y;
const std::size_t oIndexFull = oIndex + ox*oySize + oy;
const signedsize ix = static_cast<signedsize>(ox * strideDims[0]);
const signedsize iy = static_cast<signedsize>(oy * strideDims[1]);
@@ -184,10 +194,10 @@ void ConvImpl2D_cpu_forward_kernel(const std::array<DimSize_t, 2>& strideDims,
weights[wIndex + 2*kernelDims[1] + 1] * input[iIndex + static_cast<std::size_t>(ix+2)*inputDims[3] + static_cast<std::size_t>(iy+1)] +
weights[wIndex + 2*kernelDims[1] + 2] * input[iIndex + static_cast<std::size_t>(ix+2)*inputDims[3] + static_cast<std::size_t>(iy+2)]);
} else {
- for (std::size_t sx = sxMin; sx < sxMax; ++sx) {
- for (std::size_t sy = syMin; sy < syMax; ++sy) {
+ for (std::size_t sx = sxMin; sx*dilationDims[0] < sxMax; ++sx) {
+ for (std::size_t sy = syMin; sy*dilationDims[1] < syMax; ++sy) {
output[oIndexFull] += weights[wIndex + sx*kernelDims[1] + sy] *
- input[iIndex + static_cast<std::size_t>(ix+static_cast<signedsize>(sx))*inputDims[3] + static_cast<std::size_t>(iy+static_cast<signedsize>(sy))];
+ input[iIndex + static_cast<std::size_t>(ix+static_cast<signedsize>(sx*dilationDims[0]))*inputDims[3] + static_cast<std::size_t>(iy+static_cast<signedsize>(sy*dilationDims[1]))];
}
}
}
--
GitLab