Merge branch 'amd-develop' into amd-master

library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp

@@ -345,7 +345,11 @@ void add_device_gemm_xdl_c_shuffle_f8_f8_f8_km_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Col, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances);
 
-void add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances(
+void add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_default_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances);
+
+void add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_padded_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Row, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances);
 
@@ -575,7 +579,8 @@ struct DeviceOperationInstanceFactory<
             if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
                          is_same_v<CLayout, Row>)
             {
-                add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances(op_ptrs);
+                add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_padded_instances(op_ptrs);
+                add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_default_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)

library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_scaleadd_scaleadd_relu.hpp

@@ -27,7 +27,7 @@ void add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhw
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
                                                                 NDHWGC,
                                                                 GKZYXC,
-                                                                ck::Tuple<NDHWGK, NDHWGK>,
+                                                                ck::Tuple<NDHWGK, G_K>,
                                                                 NDHWGK,
                                                                 BF16,
                                                                 BF16,
@@ -43,7 +43,7 @@ void add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhw
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
                                                                 NDHWGC,
                                                                 GKZYXC,
-                                                                ck::Tuple<NDHWGK, NDHWGK>,
+                                                                ck::Tuple<NDHWGK, G_K>,
                                                                 NDHWGK,
                                                                 F16,
                                                                 F16,
@@ -59,7 +59,7 @@ void add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhw
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
                                                                 NDHWGC,
                                                                 GKZYXC,
-                                                                ck::Tuple<NDHWGK, NDHWGK>,
+                                                                ck::Tuple<NDHWGK, G_K>,
                                                                 NDHWGK,
                                                                 F32,
                                                                 F32,
@@ -75,7 +75,7 @@ void add_device_grouped_conv3d_fwd_xdl_scaleadd_scaleadd_relu_ndhwgc_gkzyxc_ndhw
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
                                                                 NDHWGC,
                                                                 GKZYXC,
-                                                                ck::Tuple<NDHWGK, NDHWGK>,
+                                                                ck::Tuple<NDHWGK, G_K>,
                                                                 NDHWGK,
                                                                 int8_t,
                                                                 int8_t,
@@ -130,7 +130,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
         if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWGC> &&
-                     is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, NDHWGK>)
+                     is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, NDHWGK> &&
+                     DLayouts::Size() == 2 && is_same_v<tuple_element_t<0, DLayouts>, NDHWGK> &&
+                     is_same_v<tuple_element_t<1, DLayouts>, G_K>)
         {
 #ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&

library/include/ck/library/tensor_operation_instance/gpu/normalization_fwd_swish.hpp

@@ -19,7 +19,7 @@ namespace instance {
 
 // FP16
 void add_device_normalization_fwd_rank_5_3_swish_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalizationFwd<F16, F16, F16, F16, F32, Swish, 5, 3>>>&);
+    std::vector<std::unique_ptr<DeviceNormalizationFwd<F16, F16, F16, F16, F16, Swish, 5, 3>>>&);
 
 // FP32
 void add_device_normalization_fwd_rank_5_3_swish_f32_instances(
@@ -61,7 +61,7 @@ struct DeviceOperationInstanceFactory<
 
         if constexpr(is_same_v<XDataType, F16> && is_same_v<GammaDataType, F16> &&
                      is_same_v<BetaDataType, F16> && is_same_v<YDataType, F16> &&
-                     is_same_v<SaveMeanInvStdDataType, F32>)
+                     is_same_v<SaveMeanInvStdDataType, F16>)
         {
             if constexpr(Rank == 5 && NumReduceDim == 3)
             {