Merge branch 'gridwise_2d' of github.com:ROCmSoftwarePlatform/composable_kernel into gridwise_2d

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/36_sparse_embedding/sparse_embedding3_forward_layernorm.cpp

@@ -86,12 +86,10 @@ int main()
     constexpr auto index_length   = 2048;
     constexpr AccDataType epsilon = 1e-4;
 
-    auto f_host_tensor_desc_1d = [](std::size_t len_) {
-        return HostTensorDescriptor(std::vector<std::size_t>({len_}));
-    };
+    auto f_host_tensor_desc_1d = [](std::size_t len_) { return HostTensorDescriptor({len_}); };
 
     auto f_host_tensor_desc_2d = [](std::size_t rows_, std::size_t cols_) {
-        return HostTensorDescriptor(std::vector<std::size_t>({rows_, cols_}));
+        return HostTensorDescriptor({rows_, cols_});
     };
 
     using ReferenceInstance =
@@ -203,8 +201,7 @@ int main()
             ref_invoker.Run(ref_argument);
 
             out_dev.FromDevice(out_from_dev.mData.data());
-            pass &= ck::utils::check_err(
-                out_from_dev.mData, out.mData, "Error: Incorrect results", 1e-3, 1e-3);
+            pass &= ck::utils::check_err(out_from_dev, out, "Error: Incorrect results", 1e-3, 1e-3);
         }
 
         double total_read = current_dim * index_length * 3 * sizeof(EmbType) +