leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 25 Packages Projects Releases Wiki Activity

CUTLASS 2.4 (Implicit GEMM convolution) (#147)

Browse Source 
CUTLASS 2.4 (Implicit GEMM Convolution)

Co-authored-by: Manish Gupta <manigupta@nvidia.com>, Haicheng Wu <haichengw@nvidia.com>, Dustyn Blasig <dblasig@nvidia.com>, Andrew Kerr <akerr@nvidia.com>
This commit is contained in:
Manish Gupta
2020-11-19 21:25:25 -08:00
committed by GitHub
co-authored by Manish Gupta <manigupta@nvidia.com>, Haicheng Wu <haichengw@nvidia.com>, Dustyn Blasig <dblasig@nvidia.com>, Andrew Kerr <akerr@nvidia.com>
parent c2b80ad4e4
commit 6615010cd0
224 changed files with 43939 additions and 1061 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/profiler/src/device_allocation.cu
+36 -1
View File
@@ -133,7 +133,18 @@ std::vector<int> DeviceAllocation::get_packed_layout(
case library::LayoutTypeID::kTensorNDHWC:
stride = get_packed_layout_stride<cutlass::layout::TensorNDHWC>(extent);
break;
case library::LayoutTypeID::kTensorNC32HW32:
stride = get_packed_layout_stride<cutlass::layout::TensorNCxHWx<32>>(extent);
break;
case library::LayoutTypeID::kTensorNC64HW64:
stride = get_packed_layout_stride<cutlass::layout::TensorNCxHWx<64>>(extent);
break;
case library::LayoutTypeID::kTensorC32RSK32:
stride = get_packed_layout_stride<cutlass::layout::TensorCxRSKx<32>>(extent);
break;
case library::LayoutTypeID::kTensorC64RSK64:
stride = get_packed_layout_stride<cutlass::layout::TensorCxRSKx<64>>(extent);
break;
default: break;
}
@@ -247,6 +258,18 @@ size_t DeviceAllocation::construct_layout(
case library::LayoutTypeID::kTensorNDHWC:
return construct_layout_<cutlass::layout::TensorNDHWC>(bytes, layout_id, extent, stride);
case library::LayoutTypeID::kTensorNC32HW32:
return construct_layout_<cutlass::layout::TensorNCxHWx<32>>(bytes, layout_id, extent, stride);
case library::LayoutTypeID::kTensorNC64HW64:
return construct_layout_<cutlass::layout::TensorNCxHWx<64>>(bytes, layout_id, extent, stride);
case library::LayoutTypeID::kTensorC32RSK32:
return construct_layout_<cutlass::layout::TensorCxRSKx<32>>(bytes, layout_id, extent, stride);
case library::LayoutTypeID::kTensorC64RSK64:
return construct_layout_<cutlass::layout::TensorCxRSKx<64>>(bytes, layout_id, extent, stride);
default: break;
}
@@ -1362,6 +1385,18 @@ static void write_tensor_csv_static_type(
case library::LayoutTypeID::kTensorNDHWC:
write_tensor_csv_static_tensor_view<T, layout::TensorNDHWC>(out, allocation);
break;
case library::LayoutTypeID::kTensorNC32HW32:
write_tensor_csv_static_tensor_view<T, layout::TensorNCxHWx<32>>(out, allocation);
break;
case library::LayoutTypeID::kTensorNC64HW64:
write_tensor_csv_static_tensor_view<T, layout::TensorNCxHWx<64>>(out, allocation);
break;
case library::LayoutTypeID::kTensorC32RSK32:
write_tensor_csv_static_tensor_view<T, layout::TensorCxRSKx<32>>(out, allocation);
break;
case library::LayoutTypeID::kTensorC64RSK64:
write_tensor_csv_static_tensor_view<T, layout::TensorCxRSKx<64>>(out, allocation);
break;
default:
throw std::runtime_error("Unhandled layout");
}