Merge branch 'develop' into bnorm_bwd_pr

example/04_gemm_add_add_fastgelu/run_gemm_add_add_fastgelu_example.inc

@@ -124,7 +124,7 @@ bool run_gemm_add_add_fastgelu(const ProblemSize& problem_size, const ExecutionC
 
     if(config.do_verification)
     {
-        Tensor<AccDataType> c_m_n(HostTensorDescriptor{M, N});
+        Tensor<AccDataType> c_m_n({M, N});
 
         auto ref_gemm    = ReferenceGemmInstance{};
         auto ref_invoker = ref_gemm.MakeInvoker();
@@ -147,9 +147,9 @@ bool run_gemm_add_add_fastgelu(const ProblemSize& problem_size, const ExecutionC
 #ifdef BUILD_INT4_EXAMPLE
         const Tensor<EDataType> e_m_n_device_result_converted(e_m_n_device_result);
 
-        return ck::utils::check_err(e_m_n_device_result_converted.mData, e_m_n_host_result.mData);
+        return ck::utils::check_err(e_m_n_device_result_converted, e_m_n_host_result);
 #else
-        return ck::utils::check_err(e_m_n_device_result.mData, e_m_n_host_result.mData);
+        return ck::utils::check_err(e_m_n_device_result, e_m_n_host_result);
 #endif
     }
 

example/09_convnd_fwd/convnd_fwd_common.hpp

@@ -10,6 +10,7 @@
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
+#include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
@@ -84,7 +85,7 @@ bool run_grouped_conv_fwd(bool do_verification,
     std::array<ck::index_t, NDimSpatial> input_left_pads{};
     std::array<ck::index_t, NDimSpatial> input_right_pads{};
 
-    auto copy = [](auto& x, auto& y) { std::copy(x.begin(), x.end(), y.begin()); };
+    auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
 
     copy(in_g_n_c_wis_desc.GetLengths(), a_g_n_c_wis_lengths);
     copy(in_g_n_c_wis_desc.GetStrides(), a_g_n_c_wis_strides);
@@ -164,7 +165,7 @@ bool run_grouped_conv_fwd(bool do_verification,
         out_device_buf.FromDevice(out_device.mData.data());
 
         return ck::utils::check_err(
-            out_device.mData, out_host.mData, "Error: incorrect results!", 1e-5f, 1e-4f);
+            out_device, out_host, "Error: incorrect results!", 1e-5f, 1e-4f);
     }
 
     return true;

example/09_convnd_fwd/convnd_fwd_dl_common.hpp

@@ -10,6 +10,7 @@
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
+#include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
@@ -88,7 +89,7 @@ bool run_grouped_conv_fwd_dl(bool do_verification,
     std::array<ck::index_t, NDimSpatial> input_left_pads{};
     std::array<ck::index_t, NDimSpatial> input_right_pads{};
 
-    auto copy = [](auto& x, auto& y) { std::copy(x.begin(), x.end(), y.begin()); };
+    auto copy = [](auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
 
     copy(in_g_n_c_wis_desc.GetLengths(), a_g_n_c_wis_lengths);
     copy(in_g_n_c_wis_desc.GetStrides(), a_g_n_c_wis_strides);

example/10_convnd_fwd_multiple_d_multiple_reduce/common.hpp

@@ -16,6 +16,7 @@
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
+#include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/convolution_parameter.hpp"
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
@@ -140,9 +141,7 @@ make_r0_host_tensor_descriptor(const ck::utils::conv::ConvParam& problem_size)
 {
     std::vector<ck::index_t> dimensions{problem_size.G_, problem_size.N_};
 
-    std::copy(begin(problem_size.output_spatial_lengths_),
-              end(problem_size.output_spatial_lengths_),
-              std::back_inserter(dimensions));
+    ck::ranges::copy(problem_size.output_spatial_lengths_, std::back_inserter(dimensions));
 
     return HostTensorDescriptor(dimensions);
 }
@@ -158,10 +157,3 @@ void unpack_host_tensor_descriptor(const HostTensorDescriptor& descriptor,
     assert(size(descriptor.GetStrides()) == size(strides));
     std::copy_n(begin(descriptor.GetStrides()), size(descriptor.GetStrides()), begin(strides));
 }
-
-template <typename Range, typename OutputIterator>
-auto copy(const Range& range, OutputIterator iter)
-    -> decltype(std::copy(std::begin(range), std::end(range), iter))
-{
-    return std::copy(std::begin(range), std::end(range), iter);
-}

example/10_convnd_fwd_multiple_d_multiple_reduce/run_convnd_fwd_max_example.inc

@@ -77,15 +77,12 @@ bool run_convnd_fwd_max(const ck::utils::conv::ConvParam& problem_size,
     {
     case 0: break;
     case 1:
-        ck::utils::FillUniformDistributionIntegerValue<ADataType>{-8, 7}(conv_input.begin(),
-                                                                         conv_input.end());
-        ck::utils::FillUniformDistributionIntegerValue<BDataType>{-8, 7}(conv_weight.begin(),
-                                                                         conv_weight.end());
+        ck::utils::FillUniformDistributionIntegerValue<ADataType>{-8, 7}(conv_input);
+        ck::utils::FillUniformDistributionIntegerValue<BDataType>{-8, 7}(conv_weight);
         break;
     default:
-        ck::utils::FillUniformDistribution<ADataType>{-5, 5}(conv_input.begin(), conv_input.end());
-        ck::utils::FillUniformDistribution<BDataType>{-5, 5}(conv_weight.begin(),
-                                                             conv_weight.end());
+        ck::utils::FillUniformDistribution<ADataType>{-5, 5}(conv_input);
+        ck::utils::FillUniformDistribution<BDataType>{-5, 5}(conv_weight);
     }
 
     DeviceMem conv_input_device_buf(sizeof(ADataType) * conv_input.mDesc.GetElementSpaceSize());
@@ -123,10 +120,10 @@ bool run_convnd_fwd_max(const ck::utils::conv::ConvParam& problem_size,
         conv_output_g_n_k_wos_desc, conv_output_g_n_k_wos_lengths, conv_output_g_n_k_wos_strides);
     unpack_host_tensor_descriptor(r0_desc, r0_lengths, r0_strides);
 
-    copy(problem_size.conv_filter_strides_, begin(conv_filter_strides));
-    copy(problem_size.conv_filter_dilations_, begin(conv_filter_dilations));
-    copy(problem_size.input_left_pads_, begin(input_left_pads));
-    copy(problem_size.input_right_pads_, begin(input_right_pads));
+    ck::ranges::copy(problem_size.conv_filter_strides_, begin(conv_filter_strides));
+    ck::ranges::copy(problem_size.conv_filter_dilations_, begin(conv_filter_dilations));
+    ck::ranges::copy(problem_size.input_left_pads_, begin(input_left_pads));
+    ck::ranges::copy(problem_size.input_right_pads_, begin(input_right_pads));
 
     // run Conv + Reduction on device
     auto conv     = DeviceInstance<NDimSpatial>{};
@@ -276,16 +273,13 @@ bool run_convnd_fwd_max(const ck::utils::conv::ConvParam& problem_size,
         conv_output_device_buf.FromDevice(conv_output_device.mData.data());
         r0_device_buf.FromDevice(r0_device.mData.data());
 
-        return ck::utils::check_err(conv_output_device.mData,
-                                    conv_output_host.mData,
+        return ck::utils::check_err(conv_output_device,
+                                    conv_output_host,
                                     "Error: incorrect results! (Matrix E)",
                                     1e-5f,
                                     1e-4f) &&
-               ck::utils::check_err(r0_device.mData,
-                                    r0_host.mData,
-                                    "Error: incorrect results! (Matrix R0)",
-                                    1e-5f,
-                                    1e-4f);
+               ck::utils::check_err(
+                   r0_device, r0_host, "Error: incorrect results! (Matrix R0)", 1e-5f, 1e-4f);
     }
 
     return true;

example/12_reduce/reduce_blockwise.cpp

@@ -142,7 +142,7 @@ bool reduce_blockwise_test(bool do_verification,
 
         std::array<int, ShapeType::NumReduceDim_> arrReduceDims;
 
-        std::copy(reduceDims.begin(), reduceDims.end(), arrReduceDims.begin());
+        ck::ranges::copy(reduceDims, arrReduceDims.begin());
 
         result = reduce_blockwise_impl<InOutDataType,
                                        AccDataType,

example/12_reduce/reduce_blockwise_impl.hpp

@@ -10,6 +10,7 @@
 #include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp"
 
+#include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
@@ -263,10 +264,10 @@ int reduce_blockwise_impl(bool do_verification,
     std::array<index_t, NumOutDim> arrOutLengths;
     std::array<index_t, NumOutDim> arrOutStrides;
 
-    std::copy(inLengths.begin(), inLengths.end(), arrInLengths.begin());
-    std::copy(inStrides.begin(), inStrides.end(), arrInStrides.begin());
-    std::copy(outLengths.begin(), outLengths.end(), arrOutLengths.begin());
-    std::copy(outStrides.begin(), outStrides.end(), arrOutStrides.begin());
+    ck::ranges::copy(inLengths, arrInLengths.begin());
+    ck::ranges::copy(inStrides, arrInStrides.begin());
+    ck::ranges::copy(outLengths, arrOutLengths.begin());
+    ck::ranges::copy(outStrides, arrOutStrides.begin());
 
     auto reduce = DeviceReduceInstance{};
 
@@ -324,12 +325,12 @@ int reduce_blockwise_impl(bool do_verification,
 #endif
             out_dev.FromDevice(out.mData.data());
 
-        pass = pass && ck::utils::check_err(out.mData, out_ref.mData);
+        pass = pass && ck::utils::check_err(out, out_ref);
 
         if(OutputIndex)
         {
             out_index_dev.FromDevice(out_indices.mData.data());
-            pass = pass && ck::utils::check_err(out_indices.mData, out_indices_ref.mData);
+            pass = pass && ck::utils::check_err(out_indices, out_indices_ref);
         };
     };
 

example/12_reduce/reduce_blockwise_two_call.cpp

@@ -221,12 +221,12 @@ int main(int argc, char* argv[])
     std::array<index_t, 3> arrOutLengths;
     std::array<index_t, 3> arrOutStrides;
 
-    std::copy(inLengths_1.begin(), inLengths_1.end(), arrInLengths_1.begin());
-    std::copy(inStrides_1.begin(), inStrides_1.end(), arrInStrides_1.begin());
-    std::copy(inLengths_2.begin(), inLengths_2.end(), arrInLengths_2.begin());
-    std::copy(inStrides_2.begin(), inStrides_2.end(), arrInStrides_2.begin());
-    std::copy(outLengths.begin(), outLengths.end(), arrOutLengths.begin());
-    std::copy(outStrides.begin(), outStrides.end(), arrOutStrides.begin());
+    ck::ranges::copy(inLengths_1, arrInLengths_1.begin());
+    ck::ranges::copy(inStrides_1, arrInStrides_1.begin());
+    ck::ranges::copy(inLengths_2, arrInLengths_2.begin());
+    ck::ranges::copy(inStrides_2, arrInStrides_2.begin());
+    ck::ranges::copy(outLengths, arrOutLengths.begin());
+    ck::ranges::copy(outStrides, arrOutStrides.begin());
 
     auto reduce_1 = DeviceReduceInstance_1{};
 
@@ -294,7 +294,7 @@ int main(int argc, char* argv[])
     if(do_verify)
     {
         out_dev.FromDevice(out.mData.data());
-        pass = pass && ck::utils::check_err(out.mData, out_ref.mData);
+        pass = pass && ck::utils::check_err(out, out_ref);
     };
 
     return (pass ? 0 : 1);

example/12_reduce/reduce_multiblock_atomic_add.cpp

@@ -140,7 +140,7 @@ bool reduce_multiblock_atomic_add_test(bool do_verification,
 
         std::array<int, ShapeType::NumReduceDim_> a_reduceDims;
 
-        std::copy(reduceDims.begin(), reduceDims.end(), a_reduceDims.begin());
+        ck::ranges::copy(reduceDims, a_reduceDims.begin());
 
         result = reduce_multiblock_atomic_add_impl<InOutDataType,
                                                    AccDataType,

example/12_reduce/reduce_multiblock_atomic_add_impl.hpp

@@ -10,6 +10,7 @@
 #include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp"
 
+#include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
@@ -176,10 +177,10 @@ int reduce_multiblock_atomic_add_impl(bool do_verification,
     std::array<index_t, NumOutDim> arrOutLengths;
     std::array<index_t, NumOutDim> arrOutStrides;
 
-    std::copy(inLengths.begin(), inLengths.end(), arrInLengths.begin());
-    std::copy(inStrides.begin(), inStrides.end(), arrInStrides.begin());
-    std::copy(outLengths.begin(), outLengths.end(), arrOutLengths.begin());
-    std::copy(outStrides.begin(), outStrides.end(), arrOutStrides.begin());
+    ck::ranges::copy(inLengths, arrInLengths.begin());
+    ck::ranges::copy(inStrides, arrInStrides.begin());
+    ck::ranges::copy(outLengths, arrOutLengths.begin());
+    ck::ranges::copy(outStrides, arrOutStrides.begin());
 
     auto reduce = DeviceReduceInstance{};
 
@@ -225,7 +226,7 @@ int reduce_multiblock_atomic_add_impl(bool do_verification,
     if(do_verification)
     {
         out_dev.FromDevice(out.mData.data());
-        pass = pass && ck::utils::check_err(out.mData, out_ref.mData);
+        pass = pass && ck::utils::check_err(out, out_ref);
     };
 
     return (pass ? 0 : 1);

example/13_pool2d_fwd/pool2d_fwd_common.hpp

@@ -16,6 +16,7 @@
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
 
 template <typename InDataType,
           typename OutDataType,
@@ -172,16 +173,16 @@ bool pool_test(bool do_verification,
     // tensor layout
     auto f_host_tensor_descriptor =
         [](std::size_t N_, std::size_t C_, std::size_t H, std::size_t W, auto layout) {
+            using namespace ck::literals;
+
             if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCHW>::value)
             {
-                return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}),
-                                            std::vector<std::size_t>({C_ * H * W, H * W, W, 1}));
+                return HostTensorDescriptor({N_, C_, H, W}, {C_ * H * W, H * W, W, 1_uz});
             }
             else if constexpr(ck::is_same<decltype(layout),
                                           ck::tensor_layout::convolution::NHWC>::value)
             {
-                return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}),
-                                            std::vector<std::size_t>({C_ * H * W, 1, W * C_, C_}));
+                return HostTensorDescriptor({N_, C_, H, W}, {C_ * H * W, 1_uz, W * C_, C_});
             }
         };
 
@@ -267,14 +268,14 @@ bool pool_test(bool do_verification,
 
         out_device_buf.FromDevice(out_n_c_ho_wo_device.mData.data());
 
-        pass = pass && ck::utils::check_err(out_n_c_ho_wo_device.mData, out_n_c_ho_wo_host.mData);
+        pass = pass && ck::utils::check_err(out_n_c_ho_wo_device, out_n_c_ho_wo_host);
 
         if constexpr(OutputIndex)
         {
             out_indices_device_buf.FromDevice(out_indices_n_c_ho_wo_device.mData.data());
 
-            pass = pass && ck::utils::check_err(out_indices_n_c_ho_wo_device.mData,
-                                                out_indices_n_c_ho_wo_host.mData);
+            pass = pass &&
+                   ck::utils::check_err(out_indices_n_c_ho_wo_device, out_indices_n_c_ho_wo_host);
         };
     }
 

example/14_gemm_xdl_quantization/CMakeLists.txt

+add_example_executable(example_gemm_xdl_relu_quantization_int8 gemm_xdl_relu_quantization_int8.cpp)
\ No newline at end of file

example/14_gemm_xdl_requant_relu_requant/gemm_xdl_requant_relu_requant_int8.cpp → example/14_gemm_xdl_quantization/gemm_xdl_relu_quantization_int8.cpp

@@ -15,33 +15,16 @@
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
 #include "ck/library/utility/check_err.hpp"
 
-struct RequantReluRequant
-{
-    // FIXME: We just need one scale for Relu / Leaky Relu / PRelu
-    RequantReluRequant(float scaleGemm, float scaleRelu)
-        : scaleGemm_(scaleGemm), scaleRelu_(scaleRelu)
-    {
-    }
-
-    __host__ __device__ constexpr void operator()(float& y, const float& x) const
-    {
-        float gemm_requant = scaleGemm_ * x;
-        float relu         = gemm_requant > 0 ? gemm_requant : 0;
-        float relu_requant = scaleRelu_ * relu;
-        y                  = relu_requant > 127 ? 127 : relu_requant < -128 ? -128 : relu_requant;
-    }
-
-    float scaleGemm_;
-    float scaleRelu_;
-};
-
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
 
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::Relu;
+using CElementOp   = ck::tensor_operation::element_wise::Activation_Mul_Clamp<ActivationOp>;
 
 using ADataType        = int8_t;
 using BDataType        = int8_t;
@@ -67,7 +50,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
      CShuffleDataType,           // typename CShuffleDataType,
      PassThrough,                // typename AElementwiseOperation,
      PassThrough,                // typename BElementwiseOperation,
-     RequantReluRequant,         // typename CElementwiseOperation,
+     CElementOp,         // typename CElementwiseOperation,
      GemmDefault,                // GemmSpecialization GemmSpec,
      1,                          // index_t NumGemmKPrefetchStage,
      256,                        // index_t BlockSize,
@@ -100,13 +83,8 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
      16>;                        // index_t CShuffleBlockTransferScalarPerVector_NPerBlock>
 // clang-format on
 
-using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
-                                                                        BDataType,
-                                                                        CDataType,
-                                                                        float,
-                                                                        PassThrough,
-                                                                        PassThrough,
-                                                                        RequantReluRequant>;
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, float, PassThrough, PassThrough, CElementOp>;
 
 int main(int argc, char* argv[])
 {
@@ -123,8 +101,7 @@ int main(int argc, char* argv[])
     ck::index_t StrideB = 4096;
     ck::index_t StrideC = 4096;
 
-    float scale_gemm = 0.03;
-    float scale_relu = 1;
+    float quant_multiplier = 0.03;
 
     if(argc == 4)
     {
@@ -157,15 +134,15 @@ int main(int argc, char* argv[])
 
     auto f_host_tensor_descriptor =
         [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
             if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
             {
-                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
-                                            std::vector<std::size_t>({stride, 1}));
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
             }
             else
             {
-                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
-                                            std::vector<std::size_t>({1, stride}));
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
             }
         };
 
@@ -199,7 +176,7 @@ int main(int argc, char* argv[])
 
     auto a_element_op = PassThrough{};
     auto b_element_op = PassThrough{};
-    auto c_element_op = RequantReluRequant{scale_gemm, scale_relu};
+    auto c_element_op = CElementOp{quant_multiplier, ActivationOp{}};
 
     // do GEMM
     auto gemm     = DeviceGemmInstance{};
@@ -249,7 +226,7 @@ int main(int argc, char* argv[])
 
         ref_invoker.Run(ref_argument);
 
-        return ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData) ? 0 : 1;
+        return ck::utils::check_err(c_m_n_device_result, c_m_n_host_result) ? 0 : 1;
     }
 
     return 0;

example/14_gemm_xdl_requant_relu_requant/CMakeLists.txt

-add_example_executable(example_gemm_xdl_requant_relu_requant_int8 gemm_xdl_requant_relu_requant_int8.cpp)
\ No newline at end of file

example/15_grouped_gemm/grouped_gemm_xdl_bfp16.cpp

@@ -16,6 +16,7 @@
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
 
 template <ck::index_t... Is>

example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp

@@ -16,6 +16,7 @@
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
 
 template <ck::index_t... Is>

example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp

@@ -16,6 +16,7 @@
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
 
 template <ck::index_t... Is>

example/15_grouped_gemm/grouped_gemm_xdl_int4.cpp

@@ -16,6 +16,7 @@
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
 
 template <ck::index_t... Is>

example/15_grouped_gemm/grouped_gemm_xdl_int8.cpp

@@ -16,6 +16,7 @@
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
 
 template <ck::index_t... Is>

example/15_grouped_gemm/run_grouped_gemm_example.inc

@@ -52,15 +52,15 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co
 
     auto f_host_tensor_descriptor =
         [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
             if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
             {
-                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
-                                            std::vector<std::size_t>({stride, 1}));
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
             }
             else
             {
-                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
-                                            std::vector<std::size_t>({1, stride}));
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
             }
         };
 
@@ -208,10 +208,10 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co
 
 #ifdef BUILD_INT4_EXAMPLE
             const Tensor<EDataType> c_device_result_converted(c_device_tensors[i]);
-            pass &= ck::utils::check_err(c_device_result_converted.mData, c_host_tensors[i].mData);
+            pass &= ck::utils::check_err(c_device_result_converted, c_host_tensors[i]);
 
 #else
-            pass &= ck::utils::check_err(c_device_tensors[i].mData, c_host_tensors[i].mData);
+            pass &= ck::utils::check_err(c_device_tensors[i], c_host_tensors[i]);
 #endif
         }
     }