Merge branch 'develop' into hip_tensor_permute

test/conv_tensor_rearrange/test_conv_tensor_rearrange_interface.cpp

@@ -53,7 +53,7 @@ class TestConvTensorRearrangeInterface : public ::testing::Test
     template <typename ConvTensorRearrangeOp>
     bool Run()
     {
-
+        const auto G = conv_param.G_;
         const auto N = conv_param.N_;
         const auto C = conv_param.C_;
         const auto FakeC =
@@ -71,13 +71,13 @@ class TestConvTensorRearrangeInterface : public ::testing::Test
         const auto image_desc =
             ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<ImLayout>(
                 conv_param);
-        const auto gemm_desc = HostTensorDescriptor({NDoHoWo, CZYX});
+        const auto gemm_desc = HostTensorDescriptor({G, NDoHoWo, CZYX});
 
         std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
         std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
         std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
         std::array<ck::index_t, NDimSpatial + 3> input_g_n_c_wis_strides{};
-        std::array<ck::index_t, 2> output_m_k_strides{};
+        std::array<ck::index_t, 3> output_g_m_k_strides{};
         std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
         std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
         std::array<ck::index_t, NDimSpatial> input_left_pads{};
@@ -89,7 +89,7 @@ class TestConvTensorRearrangeInterface : public ::testing::Test
         copy(conv_param.filter_spatial_lengths_, filter_spatial_lengths);
         copy(conv_param.output_spatial_lengths_, output_spatial_lengths);
         copy(image_desc.GetStrides(), input_g_n_c_wis_strides);
-        copy(gemm_desc.GetStrides(), output_m_k_strides);
+        copy(gemm_desc.GetStrides(), output_g_m_k_strides);
         copy(conv_param.conv_filter_strides_, conv_filter_strides);
         copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
         copy(conv_param.input_left_pads_, input_left_pads);
@@ -100,13 +100,14 @@ class TestConvTensorRearrangeInterface : public ::testing::Test
             auto img2col  = DeviceImgToColInstance{};
             auto argument = img2col.MakeArgument(nullptr,
                                                  nullptr,
+                                                 G,
                                                  N,
                                                  IsCPacked ? C : FakeC,
                                                  input_spatial_lengths,
                                                  filter_spatial_lengths,
                                                  output_spatial_lengths,
                                                  input_g_n_c_wis_strides,
-                                                 output_m_k_strides,
+                                                 output_g_m_k_strides,
                                                  conv_filter_strides,
                                                  conv_filter_dilations,
                                                  input_left_pads,
@@ -119,13 +120,14 @@ class TestConvTensorRearrangeInterface : public ::testing::Test
             auto col2img  = DeviceColToimgInstance{};
             auto argument = col2img.MakeArgument(nullptr,
                                                  nullptr,
+                                                 G,
                                                  N,
                                                  IsCPacked ? C : FakeC,
                                                  input_spatial_lengths,
                                                  filter_spatial_lengths,
                                                  output_spatial_lengths,
                                                  input_g_n_c_wis_strides,
-                                                 output_m_k_strides,
+                                                 output_g_m_k_strides,
                                                  conv_filter_strides,
                                                  conv_filter_dilations,
                                                  input_left_pads,

test/grouped_gemm/test_grouped_gemm_interface.cpp

@@ -108,6 +108,10 @@ TEST_F(TestGGemmSplitKInterface_MKNKMN, KLoops)
 
     // kloops % 2
     Ks = std::vector<int>{256, 512, 320, 768};
+    EXPECT_FALSE(
+        DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, kbatch));
+
+    Ks = std::vector<int>{256, 512, 384, 768};
     EXPECT_TRUE(
         DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, kbatch));