Merge branch 'develop' into wmma_op

example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_bf16.cpp

@@ -23,6 +23,7 @@ Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
 #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_batched_gemm.hpp"
 
 template <ck::index_t... Is>

example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp16.cpp

@@ -23,6 +23,7 @@ Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
 #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_batched_gemm.hpp"
 
 template <ck::index_t... Is>

example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp32.cpp

@@ -23,6 +23,7 @@ Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
 #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_batched_gemm.hpp"
 
 template <ck::index_t... Is>

example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int4.cpp

@@ -27,6 +27,7 @@ Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
 #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_batched_gemm.hpp"
 
 template <ck::index_t... Is>

example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int8.cpp

@@ -23,6 +23,7 @@ Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
 #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_batched_gemm.hpp"
 
 template <ck::index_t... Is>

example/31_batched_gemm_gemm/run_batched_gemm_gemm_example.inc

@@ -106,15 +106,15 @@ bool run_batched_gemm_gemm_example(int argc, char* argv[])
                                        std::size_t stride,
                                        std::size_t batch_stride,
                                        auto layout) {
+        using namespace ck::literals;
+
         if(std::is_same<decltype(layout), Row>::value)
         {
-            return HostTensorDescriptor(std::vector<std::size_t>({batch_count, row, col}),
-                                        std::vector<std::size_t>({batch_stride, stride, 1}));
+            return HostTensorDescriptor({batch_count, row, col}, {batch_stride, stride, 1_uz});
         }
         else
         {
-            return HostTensorDescriptor(std::vector<std::size_t>({batch_count, row, col}),
-                                        std::vector<std::size_t>({batch_stride, 1, stride}));
+            return HostTensorDescriptor({batch_count, row, col}, {batch_stride, 1_uz, stride});
         }
     };
 
@@ -270,7 +270,7 @@ bool run_batched_gemm_gemm_example(int argc, char* argv[])
         c_g_m_o_device_buf.FromDevice(c_g_m_o_device_result.mData.data());
 #endif
 
-        return ck::utils::check_err(c_g_m_o_device_result.mData, c_g_m_o_host_result.mData);
+        return ck::utils::check_err(c_g_m_o_device_result, c_g_m_o_host_result);
     }
 
     return true;

example/32_batched_gemm_scale_softmax_gemm/batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp

@@ -24,6 +24,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g
 #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_batched_gemm.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_softmax.hpp"
 

example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp

@@ -24,6 +24,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g
 #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_batched_gemm.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_softmax.hpp"
 

example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_fp16.cpp

@@ -23,6 +23,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g
 #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_batched_gemm.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_softmax.hpp"
 
@@ -245,15 +246,15 @@ int main(int argc, char* argv[])
                                        std::size_t stride,
                                        std::size_t batch_stride,
                                        auto layout) {
+        using namespace ck::literals;
+
         if(std::is_same<decltype(layout), Row>::value)
         {
-            return HostTensorDescriptor(std::vector<std::size_t>({batch_count, row, col}),
-                                        std::vector<std::size_t>({batch_stride, stride, 1}));
+            return HostTensorDescriptor({batch_count, row, col}, {batch_stride, stride, 1_uz});
         }
         else
         {
-            return HostTensorDescriptor(std::vector<std::size_t>({batch_count, row, col}),
-                                        std::vector<std::size_t>({batch_stride, 1, stride}));
+            return HostTensorDescriptor({batch_count, row, col}, {batch_stride, 1_uz, stride});
         }
     };
 
@@ -391,7 +392,7 @@ int main(int argc, char* argv[])
 
         ref_gemm1_invoker.Run(ref_gemm1_argument);
 
-        return ck::utils::check_err(c_g_m_o_device_result.mData, c_g_m_o_host_result.mData) ? 0 : 1;
+        return ck::utils::check_err(c_g_m_o_device_result, c_g_m_o_host_result) ? 0 : 1;
     }
 
     return 0;

example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp

@@ -24,6 +24,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g
 #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_batched_gemm.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_softmax.hpp"
 

example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm_permute.inc

@@ -22,7 +22,7 @@ int run(int argc, char* argv[])
 
     float alpha = 1;
 
-    bool input_permute = false;
+    bool input_permute  = false;
     bool output_permute = true;
 
     if(argc == 1)
@@ -50,7 +50,7 @@ int run(int argc, char* argv[])
 
         alpha = std::stof(argv[10]);
 
-        input_permute = std::stoi(argv[11]);
+        input_permute  = std::stoi(argv[11]);
         output_permute = std::stoi(argv[12]);
     }
     else

example/32_batched_gemm_scale_softmax_gemm/run_grouped_gemm_scale_softmax_gemm_permute.inc

@@ -7,7 +7,7 @@ int run(int argc, char* argv[])
     int init_method      = 1;
     bool time_kernel     = false;
 
-    bool input_permute = false;
+    bool input_permute  = false;
     bool output_permute = true;
 
     if(argc == 1)
@@ -26,7 +26,7 @@ int run(int argc, char* argv[])
         init_method     = std::stoi(argv[2]);
         time_kernel     = std::stoi(argv[3]);
 
-        input_permute = std::stoi(argv[4]);
+        input_permute  = std::stoi(argv[4]);
         output_permute = std::stoi(argv[5]);
     }
     else
@@ -66,10 +66,10 @@ int run(int argc, char* argv[])
     std::cout << "group count " << group_count << ". printing first 4 groups\n";
     for(std::size_t i = 0; i < group_count; i++)
     {
-        int M     = 128 * (rand() % 8 + 1);
-        int N     = 128 * (rand() % 8 + 1);
-        int K     = 40;
-        int O     = 40 * (rand() % 2 + 1);
+        int M  = 128 * (rand() % 8 + 1);
+        int N  = 128 * (rand() % 8 + 1);
+        int K  = 40;
+        int O  = 40 * (rand() % 2 + 1);
         int G0 = rand() % 3 + 1;
         int G1 = rand() % 5 + 1;
 
@@ -228,12 +228,12 @@ int run(int argc, char* argv[])
     {
         for(std::size_t i = 0; i < group_count; i++)
         {
-            const int& G0                  = g0_g1_m_n_k_o[i][0];
-            const int& G1                  = g0_g1_m_n_k_o[i][1];
-            const int& M                   = g0_g1_m_n_k_o[i][2];
-            const int& N                   = g0_g1_m_n_k_o[i][3];
-            const int& K                   = g0_g1_m_n_k_o[i][4];
-            const int& O                   = g0_g1_m_n_k_o[i][5];
+            const int& G0 = g0_g1_m_n_k_o[i][0];
+            const int& G1 = g0_g1_m_n_k_o[i][1];
+            const int& M  = g0_g1_m_n_k_o[i][2];
+            const int& N  = g0_g1_m_n_k_o[i][3];
+            const int& K  = g0_g1_m_n_k_o[i][4];
+            const int& O  = g0_g1_m_n_k_o[i][5];
 
             const auto& c_gs_ms_os_lengths = problem_descs[i].c_gs_ms_os_lengths;
             const auto& c_gs_ms_os_strides = problem_descs[i].c_gs_ms_os_strides;

example/33_multiple_reduce/dual_reduce_common.hpp

@@ -12,6 +12,7 @@
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/utility/data_type.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"
@@ -253,10 +254,10 @@ int mean_meansquare_dual_reduce_test(size_t n,
     std::array<ck::index_t, NumOutputDim> i_outLengths;
     std::array<ck::index_t, NumOutputDim> i_outStrides;
 
-    std::copy(inLengths.begin(), inLengths.end(), i_inLengths.begin());
-    std::copy(inStrides.begin(), inStrides.end(), i_inStrides.begin());
-    std::copy(outLengths.begin(), outLengths.end(), i_outLengths.begin());
-    std::copy(outStrides.begin(), outStrides.end(), i_outStrides.begin());
+    ck::ranges::copy(inLengths, i_inLengths.begin());
+    ck::ranges::copy(inStrides, i_inStrides.begin());
+    ck::ranges::copy(outLengths, i_outLengths.begin());
+    ck::ranges::copy(outStrides, i_outStrides.begin());
 
     auto dual_reduce_op = DeviceDualReduce{};
 
@@ -305,8 +306,8 @@ int mean_meansquare_dual_reduce_test(size_t n,
     {
         mean_dev.FromDevice(mean.mData.data());
         meansquare_dev.FromDevice(meansquare.mData.data());
-        pass = pass && ck::utils::check_err(mean.mData, mean_ref.mData);
-        pass = pass && ck::utils::check_err(meansquare.mData, meansquare_ref.mData);
+        pass = pass && ck::utils::check_err(mean, mean_ref);
+        pass = pass && ck::utils::check_err(meansquare, meansquare_ref);
     };
 
     return (pass ? 0 : 1);

example/34_batchnorm/batchnorm_infer_nhwc.cpp

@@ -9,6 +9,7 @@
 #include <getopt.h>
 
 #include "ck/ck.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"
@@ -220,14 +221,10 @@ bool bnorm_infer_nhwc_test(bool do_verification,
     std::array<index_t, Rank - NumReduceDim> i_scaleBiasMeanVarLengths;
     std::array<index_t, Rank - NumReduceDim> i_scaleBiasMeanVarStrides;
 
-    std::copy(inOutLengths.begin(), inOutLengths.end(), i_inOutLengths.begin());
-    std::copy(inOutStrides.begin(), inOutStrides.end(), i_inOutStrides.begin());
-    std::copy(scaleBiasMeanVarLengths.begin(),
-              scaleBiasMeanVarLengths.end(),
-              i_scaleBiasMeanVarLengths.begin());
-    std::copy(scaleBiasMeanVarStrides.begin(),
-              scaleBiasMeanVarStrides.end(),
-              i_scaleBiasMeanVarStrides.begin());
+    ck::ranges::copy(inOutLengths, i_inOutLengths.begin());
+    ck::ranges::copy(inOutStrides, i_inOutStrides.begin());
+    ck::ranges::copy(scaleBiasMeanVarLengths, i_scaleBiasMeanVarLengths.begin());
+    ck::ranges::copy(scaleBiasMeanVarStrides, i_scaleBiasMeanVarStrides.begin());
 
     int result = 0;
 
@@ -302,7 +299,7 @@ bool bnorm_infer_nhwc_test(bool do_verification,
         (void)invoker_ptr_ref->Run(argument_ptr_ref.get());
 
         y_dev.FromDevice(y.mData.data());
-        pass = pass && ck::utils::check_err(y.mData, y_ref.mData);
+        pass = pass && ck::utils::check_err(y, y_ref);
     };
 
     return (pass);

example/38_grouped_conv_bwd_data_multiple_d/common.hpp

@@ -15,6 +15,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp"
+#include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
 #include "ck/library/utility/convolution_parameter.hpp"