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

add unit test for non int4 load

Browse Source
This commit is contained in:
mengchi.hmc
2021-04-23 14:33:46 +08:00
parent bb35a3ba6f
commit f4b0a33633
6 changed files with 311 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

84
test/unit/conv/device/conv2d_fprop_implicit_gemm_f16nhwc_f16nhwc_f16nhwc_tensor_op_f16_sm80.cu
View File

@ -117,5 +117,89 @@ TEST(SM80_Device_Conv2d_Fprop_Optimized_ImplicitGemm_f16nhwc_f16nhwc_f16nhwc_ten
EXPECT_TRUE(test::conv::device::TestAllConv2d<Conv2dFprop>());
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_Device_Conv2d_Fprop_Optimized_ImplicitGemm_f16nhwc_f16nhwc_f16nhwc_tensor_op_f16_align2,
128x128_64x3_64x64x64) {
/// 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 Conv2dFpropKernel = typename cutlass::conv::kernel::DefaultConv2dFprop<
ElementA, cutlass::layout::TensorNHWC,
ElementB, cutlass::layout::TensorNHWC,
ElementC, cutlass::layout::TensorNHWC,
ElementAccumulator,
cutlass::arch::OpClassTensorOp,
cutlass::arch::Sm80,
cutlass::gemm::GemmShape<128, 128, 64>,
cutlass::gemm::GemmShape<64, 64, 64>,
cutlass::gemm::GemmShape<16, 8, 16>,
cutlass::epilogue::thread::LinearCombination<
ElementC,
128 / cutlass::sizeof_bits<ElementC>::value,
ElementAccumulator,
ElementCompute
>,
cutlass::gemm::threadblock::GemmIdentityThreadblockSwizzle<>,
3,
cutlass::arch::OpMultiplyAdd,
cutlass::conv::IteratorAlgorithm::kOptimized,
2,
2
>::Kernel;
using Conv2dFprop = cutlass::conv::device::ImplicitGemmConvolution<Conv2dFpropKernel>;
/// Run all unit test sizes with device-level Conv2d instance
EXPECT_TRUE(test::conv::device::TestAllConv2d<Conv2dFprop>());
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_Device_Conv2d_Fprop_Optimized_ImplicitGemm_f16nhwc_f16nhwc_f16nhwc_tensor_op_f16_align4,
128x128_64x3_64x64x64) {
/// 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 Conv2dFpropKernel = typename cutlass::conv::kernel::DefaultConv2dFprop<
ElementA, cutlass::layout::TensorNHWC,
ElementB, cutlass::layout::TensorNHWC,
ElementC, cutlass::layout::TensorNHWC,
ElementAccumulator,
cutlass::arch::OpClassTensorOp,
cutlass::arch::Sm80,
cutlass::gemm::GemmShape<128, 128, 64>,
cutlass::gemm::GemmShape<64, 64, 64>,
cutlass::gemm::GemmShape<16, 8, 16>,
cutlass::epilogue::thread::LinearCombination<
ElementC,
128 / cutlass::sizeof_bits<ElementC>::value,
ElementAccumulator,
ElementCompute
>,
cutlass::gemm::threadblock::GemmIdentityThreadblockSwizzle<>,
3,
cutlass::arch::OpMultiplyAdd,
cutlass::conv::IteratorAlgorithm::kOptimized,
4,
4
>::Kernel;
using Conv2dFprop = cutlass::conv::device::ImplicitGemmConvolution<Conv2dFpropKernel>;
/// Run all unit test sizes with device-level Conv2d instance
EXPECT_TRUE(test::conv::device::TestAllConv2d<Conv2dFprop>());
}
////////////////////////////////////////////////////////////////////////////////
#endif // CUTLASS_ARCH_MMA_SM80_SUPPORTED

129
test/unit/conv/device/conv2d_fprop_implicit_gemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_sm75.cu
View File

@ -117,5 +117,134 @@ TEST(SM75_Device_Conv2d_Fprop_Optimized_ImplicitGemm_f16nhwc_f16nhwc_f32nhwc_ten
EXPECT_TRUE(test::conv::device::TestAllConv2d<Conv2dFprop>());
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM75_Device_Conv2d_Fprop_Optimized_ImplicitGemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_align1,
128x128_32x2_64x64x32) {
/// Conv operation element types for the Gemm equivalent (ImplicitGemm)
using ElementA = cutlass::half_t;
using ElementB = cutlass::half_t;
using ElementC = float;
using ElementAccumulator = float;
using ElementCompute = float;
/// Device-level Conv2d instance
using Conv2dFpropKernel = typename cutlass::conv::kernel::DefaultConv2dFprop<
ElementA, cutlass::layout::TensorNHWC,
ElementB, cutlass::layout::TensorNHWC,
ElementC, cutlass::layout::TensorNHWC,
ElementAccumulator,
cutlass::arch::OpClassTensorOp,
cutlass::arch::Sm75,
cutlass::gemm::GemmShape<128, 128, 32>,
cutlass::gemm::GemmShape<64, 64, 32>,
cutlass::gemm::GemmShape<16, 8, 8>,
cutlass::epilogue::thread::LinearCombination<
ElementC,
128 / cutlass::sizeof_bits<ElementC>::value,
ElementAccumulator,
ElementCompute
>,
cutlass::gemm::threadblock::GemmIdentityThreadblockSwizzle<>,
2,
cutlass::arch::OpMultiplyAdd,
cutlass::conv::IteratorAlgorithm::kOptimized,
1,
1
>::Kernel;
using Conv2dFprop = cutlass::conv::device::ImplicitGemmConvolution<Conv2dFpropKernel>;
/// Run all unit test sizes with device-level Conv2d instance
EXPECT_TRUE(test::conv::device::TestAllConv2d<Conv2dFprop>());
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM75_Device_Conv2d_Fprop_Optimized_ImplicitGemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_align2,
128x128_32x2_64x64x32) {
/// Conv operation element types for the Gemm equivalent (ImplicitGemm)
using ElementA = cutlass::half_t;
using ElementB = cutlass::half_t;
using ElementC = float;
using ElementAccumulator = float;
using ElementCompute = float;
/// Device-level Conv2d instance
using Conv2dFpropKernel = typename cutlass::conv::kernel::DefaultConv2dFprop<
ElementA, cutlass::layout::TensorNHWC,
ElementB, cutlass::layout::TensorNHWC,
ElementC, cutlass::layout::TensorNHWC,
ElementAccumulator,
cutlass::arch::OpClassTensorOp,
cutlass::arch::Sm75,
cutlass::gemm::GemmShape<128, 128, 32>,
cutlass::gemm::GemmShape<64, 64, 32>,
cutlass::gemm::GemmShape<16, 8, 8>,
cutlass::epilogue::thread::LinearCombination<
ElementC,
128 / cutlass::sizeof_bits<ElementC>::value,
ElementAccumulator,
ElementCompute
>,
cutlass::gemm::threadblock::GemmIdentityThreadblockSwizzle<>,
2,
cutlass::arch::OpMultiplyAdd,
cutlass::conv::IteratorAlgorithm::kOptimized,
2,
2
>::Kernel;
using Conv2dFprop = cutlass::conv::device::ImplicitGemmConvolution<Conv2dFpropKernel>;
/// Run all unit test sizes with device-level Conv2d instance
EXPECT_TRUE(test::conv::device::TestAllConv2d<Conv2dFprop>());
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM75_Device_Conv2d_Fprop_Optimized_ImplicitGemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_align4,
128x128_32x2_64x64x32) {
/// Conv operation element types for the Gemm equivalent (ImplicitGemm)
using ElementA = cutlass::half_t;
using ElementB = cutlass::half_t;
using ElementC = float;
using ElementAccumulator = float;
using ElementCompute = float;
/// Device-level Conv2d instance
using Conv2dFpropKernel = typename cutlass::conv::kernel::DefaultConv2dFprop<
ElementA, cutlass::layout::TensorNHWC,
ElementB, cutlass::layout::TensorNHWC,
ElementC, cutlass::layout::TensorNHWC,
ElementAccumulator,
cutlass::arch::OpClassTensorOp,
cutlass::arch::Sm75,
cutlass::gemm::GemmShape<128, 128, 32>,
cutlass::gemm::GemmShape<64, 64, 32>,
cutlass::gemm::GemmShape<16, 8, 8>,
cutlass::epilogue::thread::LinearCombination<
ElementC,
128 / cutlass::sizeof_bits<ElementC>::value,
ElementAccumulator,
ElementCompute
>,
cutlass::gemm::threadblock::GemmIdentityThreadblockSwizzle<>,
2,
cutlass::arch::OpMultiplyAdd,
cutlass::conv::IteratorAlgorithm::kOptimized,
4,
4
>::Kernel;
using Conv2dFprop = cutlass::conv::device::ImplicitGemmConvolution<Conv2dFpropKernel>;
/// Run all unit test sizes with device-level Conv2d instance
EXPECT_TRUE(test::conv::device::TestAllConv2d<Conv2dFprop>());
}
////////////////////////////////////////////////////////////////////////////////
#endif // CUTLASS_ARCH_MMA_SM75_SUPPORTED

88
test/unit/conv/device/conv2d_fprop_implicit_gemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f32_sm80.cu
View File

@ -77,5 +77,93 @@ TEST(SM80_Device_Conv2d_Fprop_Analytic_ImplicitGemm_tf32nhwc_tf32nhwc_f32nhwc_te
EXPECT_TRUE(test::conv::device::TestAllConv2d<Conv2dFprop>());
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_Device_Conv2d_Fprop_Optimized_ImplicitGemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f32_align1,
128x128_32x3_64x64x32) {
/// Conv operation element types for the Gemm equivalent (ImplicitGemm)
using ElementA = cutlass::tfloat32_t;
using ElementB = cutlass::tfloat32_t;
using ElementC = float;
using ElementAccumulator = float;
using ElementCompute = float;
/// Device-level Conv2d instance
using Conv2dFpropKernel = typename cutlass::conv::kernel::DefaultConv2dFprop<
ElementA, cutlass::layout::TensorNHWC,
ElementB, cutlass::layout::TensorNHWC,
ElementC, cutlass::layout::TensorNHWC,
ElementAccumulator,
cutlass::arch::OpClassTensorOp,
cutlass::arch::Sm80,
cutlass::gemm::GemmShape<128, 128, 16>,
cutlass::gemm::GemmShape<64, 64, 16>,
cutlass::gemm::GemmShape<16, 8, 8>,
cutlass::epilogue::thread::LinearCombination<
ElementC,
128 / cutlass::sizeof_bits<ElementC>::value,
ElementAccumulator,
ElementCompute
>,
cutlass::gemm::threadblock::GemmIdentityThreadblockSwizzle<>,
3,
cutlass::arch::OpMultiplyAdd,
cutlass::conv::IteratorAlgorithm::kOptimized,
1,
1
>::Kernel;
using Conv2dFprop = cutlass::conv::device::ImplicitGemmConvolution<Conv2dFpropKernel>;
/// Run all unit test sizes with device-level Conv2d instance
EXPECT_TRUE(test::conv::device::TestAllConv2d<Conv2dFprop>());
}
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_Device_Conv2d_Fprop_Optimized_ImplicitGemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f32_align2,
128x128_32x3_64x64x32) {
/// Conv operation element types for the Gemm equivalent (ImplicitGemm)
using ElementA = cutlass::tfloat32_t;
using ElementB = cutlass::tfloat32_t;
using ElementC = float;
using ElementAccumulator = float;
using ElementCompute = float;
/// Device-level Conv2d instance
using Conv2dFpropKernel = typename cutlass::conv::kernel::DefaultConv2dFprop<
ElementA, cutlass::layout::TensorNHWC,
ElementB, cutlass::layout::TensorNHWC,
ElementC, cutlass::layout::TensorNHWC,
ElementAccumulator,
cutlass::arch::OpClassTensorOp,
cutlass::arch::Sm80,
cutlass::gemm::GemmShape<128, 128, 16>,
cutlass::gemm::GemmShape<64, 64, 16>,
cutlass::gemm::GemmShape<16, 8, 8>,
cutlass::epilogue::thread::LinearCombination<
ElementC,
128 / cutlass::sizeof_bits<ElementC>::value,
ElementAccumulator,
ElementCompute
>,
cutlass::gemm::threadblock::GemmIdentityThreadblockSwizzle<>,
3,
cutlass::arch::OpMultiplyAdd,
cutlass::conv::IteratorAlgorithm::kOptimized,
2,
2
>::Kernel;
using Conv2dFprop = cutlass::conv::device::ImplicitGemmConvolution<Conv2dFpropKernel>;
/// Run all unit test sizes with device-level Conv2d instance
EXPECT_TRUE(test::conv::device::TestAllConv2d<Conv2dFprop>());
}
////////////////////////////////////////////////////////////////////////////////
#endif // CUTLASS_ARCH_MMA_SM80_SUPPORTED