Narrowing the scope of PR to OCP FP8 enablement only

library/include/ck/library/utility/host_tensor_generator.hpp

@@ -304,107 +304,3 @@ struct GeneratorTensor_Diagonal
         return pred ? value : T{0};
     }
 };
-
-/**
- * @brief Used to generate tensor entries from coefficients of Leibniz formula for Pi.
- *
- * @tparam T The type of the tensor values.
- *
- * Usage: For verification of GEMM
- *    a_m_k.GenerateTensorValue(GeneratorTensor_PI<ADataType>{});
- *    b_k_n.GenerateTensorValue(GeneratorTensor_1<BDataType>{1});
- *
- *    c = a * b;
- *
- *    We expect that |c[i][j]-M_PI| <= truncation_error(K)
- */
-template <typename T>
-struct GeneratorTensor_PI
-{
-    template <typename... Ts>
-    T operator()(Ts... Xs) const
-    {
-        static constexpr double pi = 3.14159265358979323846;
-
-        std::array<ck::index_t, sizeof...(Ts)> dims = {{static_cast<ck::index_t>(Xs)...}};
-
-        if constexpr(dims.size() > 0)
-        {
-            constexpr auto last_dim = dims.size() - 1;
-            size_t i                = dims[last_dim];
-            float fi                = i;
-            float tmp               = (i % 2 == 0) ? 4.0 : -4.0;
-            tmp /= (2.0 * fi + 1.0);
-            return ck::type_convert<T>(tmp);
-        }
-        else
-        {
-            return ck::type_convert<T>(pi);
-        }
-    }
-
-    static double truncation_error(size_t N) { return 4.0 / (2.0 * N + 1.0); }
-};
-
-/**
- * @brief Used to generate tensor entries from coefficients of non-alternating version of Leibniz
- * formula for Pi.
- *
- * @tparam T The type of the tensor values.
- *
- * Usage: For verification of GEMM
- *    a_m_k.GenerateTensorValue(GeneratorTensor_PI_A<ADataType>{});
- *    b_k_n.GenerateTensorValue(GeneratorTensor_PI_B<BDataType>{});
- *
- *    c = a * b;
- *
- *    We expect that |c[i][j]-M_PI| <= 0.00013 for K >= 4096 and a,b,c are float.
- */
-template <typename T>
-struct GeneratorTensor_PI_A
-{
-    static constexpr double pi = 3.14159265358979323846;
-
-    template <typename... Ts>
-    T operator()(Ts... Xs) const
-    {
-        std::array<ck::index_t, sizeof...(Ts)> dims = {{static_cast<ck::index_t>(Xs)...}};
-
-        if constexpr(dims.size() > 0)
-        {
-            constexpr auto last_dim = dims.size() - 1;
-            size_t i                = dims[last_dim];
-            float fi                = i;
-            float tmp               = 2.0 / (4.0 * fi + 1.0);
-            return ck::type_convert<T>(tmp);
-        }
-        else
-        {
-            return ck::type_convert<T>(pi / 2.0);
-        }
-    }
-};
-
-template <typename T>
-struct GeneratorTensor_PI_B
-{
-    static constexpr double pi = 3.14159265358979323846;
-
-    template <typename... Ts>
-    T operator()(Ts... Xs) const
-    {
-        std::array<ck::index_t, sizeof...(Ts)> dims = {{static_cast<ck::index_t>(Xs)...}};
-
-        if constexpr(dims.size() > 0)
-        {
-            size_t i  = dims[0];
-            float fi  = i;
-            float tmp = 4.0 / (4.0 * fi + 3.0);
-            return ck::type_convert<T>(tmp);
-        }
-        else
-        {
-            return ck::type_convert<T>(2.0);
-        }
-    }
-};

test/grouped_convnd_bwd_weight/CMakeLists.txt

-add_custom_target(test_grouped_convnd_bwd_weight_target)
-
 if(GPU_TARGETS MATCHES "gfx9" OR DL_KERNELS)
       add_gtest_executable(test_grouped_convnd_bwd_weight test_grouped_convnd_bwd_weight.cpp)
       target_link_libraries(test_grouped_convnd_bwd_weight PRIVATE utility device_grouped_conv1d_bwd_weight_instance device_grouped_conv2d_bwd_weight_instance device_grouped_conv3d_bwd_weight_instance)
-      add_dependencies(test_grouped_convnd_bwd_weight_target test_grouped_convnd_bwd_weight)
    elseif(GPU_TARGETS MATCHES "gfx11")
       add_gtest_executable(test_grouped_convnd_bwd_weight test_grouped_convnd_bwd_weight.cpp)
       target_link_libraries(test_grouped_convnd_bwd_weight PRIVATE utility device_grouped_conv3d_bwd_weight_instance)
-      add_dependencies(test_grouped_convnd_bwd_weight_target test_grouped_convnd_bwd_weight)
 endif()
 add_gtest_executable(test_grouped_convnd_bwd_weight_interface_xdl test_grouped_convnd_bwd_weight_interface_xdl.cpp)
 if(result EQUAL 0)
    target_link_libraries(test_grouped_convnd_bwd_weight_interface_xdl PRIVATE utility)
-   add_dependencies(test_grouped_convnd_bwd_weight_target test_grouped_convnd_bwd_weight_interface_xdl)
 endif()
 add_gtest_executable(test_grouped_convnd_bwd_weight_interface_wmma test_grouped_convnd_bwd_weight_interface_wmma.cpp)
 if(result EQUAL 0)
    target_link_libraries(test_grouped_convnd_bwd_weight_interface_wmma PRIVATE utility)
-   add_dependencies(test_grouped_convnd_bwd_weight_target test_grouped_convnd_bwd_weight_interface_wmma)
 endif()
 add_gtest_executable(test_grouped_conv_bwd_weight_xdl_bilinear test_grouped_conv_bwd_weight_xdl_bilinear.cpp)
 if(result EQUAL 0)
    target_link_libraries(test_grouped_conv_bwd_weight_xdl_bilinear PRIVATE utility device_grouped_conv3d_bwd_weight_bilinear_instance)
-   add_dependencies(test_grouped_convnd_bwd_weight_target test_grouped_conv_bwd_weight_xdl_bilinear)
 endif()

test/pool/test_pool_fwd_common.hpp

@@ -10,6 +10,8 @@ using F16  = ck::half_t;
 using BF16 = ck::bhalf_t;
 using F32  = float;
 using I32  = int32_t;
+using I8   = int8_t;
+using F8   = ck::f8_t;
 using ck::index_t;
 using NDHWC = ck::tensor_layout::convolution::NDHWC;
 using NHWC  = ck::tensor_layout::convolution::NHWC;