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

CUTLASS 2.10 bug fixes and minor updates. (#626)

Browse Source
This commit is contained in:
Andrew Kerr
2022-09-15 16:20:33 -04:00
committed by GitHub
parent 2cc2c7ba1f
commit fc9ebc645b
5 changed files with 63 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

55
test/unit/conv/device/conv2d_dgrad_implicit_gemm_f16nhwc_f16nhwc_f16nhwc_tensor_op_f16_sm80.cu
View File

@@ -240,4 +240,59 @@ TEST(SM80_Device_Conv2d_Dgrad_Optimized_ImplicitGemm_f16nhwc_f16nhwc_f16nhwc_ten
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_Device_Conv2d_Dgrad_Optimized_ImplicitGemm_f16nhwc_f16nhwc_f16nhwc_tensor_op_f16_align2,
128x64_64x3_64x32x64) {
/// Conv operation element types for the Gemm equivalent (ImplicitGemm)
using ElementA = cutlass::half_t;
using ElementB = cutlass::half_t;
using ElementC = cutlass::half_t;
using ElementAccumulator = cutlass::half_t;
using ElementCompute = cutlass::half_t;
/// Device-level Conv2d instance
using Conv2dDgradKernel = typename cutlass::conv::kernel::DefaultConv2dDgrad<
ElementA, cutlass::layout::TensorNHWC,
ElementB, cutlass::layout::TensorNHWC,
ElementC, cutlass::layout::TensorNHWC,
ElementAccumulator,
cutlass::arch::OpClassTensorOp,
cutlass::arch::Sm80,
cutlass::gemm::GemmShape<128, 64, 32>,
cutlass::gemm::GemmShape<64, 32, 32>,
cutlass::gemm::GemmShape<16, 8, 16>,
cutlass::epilogue::thread::LinearCombination<
ElementC,
2,
ElementAccumulator,
ElementCompute
>,
cutlass::gemm::threadblock::GemmIdentityThreadblockSwizzle<>,
3,
cutlass::arch::OpMultiplyAdd,
cutlass::conv::IteratorAlgorithm::kOptimized,
cutlass::conv::StrideSupport::kUnity,
2,
2
>::Kernel;
using Conv2dDgrad = cutlass::conv::device::ImplicitGemmConvolution<Conv2dDgradKernel>;
test::conv::device::Conv2dProblemVector problem_size_list;
// run specific problem size in the unit test first
problem_size_list.push_back(cutlass::conv::Conv2dProblemSize(
{35, 100, 50, 64}, // input size (NHWC)
{22, 1, 1, 64}, // filter size (KRSC)
{0, 0, 0, 0}, // padding (pad_h, _, pad_w, _)
{1, 1}, // stride (stride_h, stride_w)
{1, 1} // dilation (dilation_h, dilation_w)
));
/// Run all unit test sizes with device-level Conv2d instance
EXPECT_TRUE(test::conv::device::TestAllConv2d<Conv2dDgrad>(problem_size_list));
}
////////////////////////////////////////////////////////////////////////////////
#endif // CUTLASS_ARCH_MMA_SM80_SUPPORTED