Compile for gfx908 and gfx90a (#130)

* adding compilation for multiple targets

* fix build

* clean

* update Jekinsfile

* update readme

* update Jenkins

* use ck::half_t instead of ushort for bf16

* rename enum classes

* clean

* rename

* clean

library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp

@@ -9,7 +9,7 @@ namespace tensor_operation {
 namespace device {
 namespace device_conv2d_bwd_data_instance {
 
-using BF16 = ushort;
+using BF16 = bhalf_t;
 using F32  = float;
 
 template <ck::index_t... Is>
@@ -17,10 +17,10 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 static constexpr auto ConvBwdDataDefault =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
 
 static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
 
 // Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
 using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances =

library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp

@@ -17,10 +17,10 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 static constexpr auto ConvBwdDataDefault =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
 
 static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
 
 // Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
 using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances =

library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp

@@ -16,10 +16,10 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 static constexpr auto ConvBwdDataDefault =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
 
 static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
 
 // Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
 using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances =

library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp

@@ -17,10 +17,10 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 static constexpr auto ConvBwdDataDefault =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
 
 static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
 
 // Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
 using device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances =

library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp

@@ -9,7 +9,7 @@ namespace tensor_operation {
 namespace device {
 namespace device_conv2d_bwd_data_instance {
 
-using BF16 = ushort;
+using BF16 = bhalf_t;
 using F32  = float;
 
 template <ck::index_t... Is>
@@ -17,10 +17,10 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 static constexpr auto ConvBwdDataDefault =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
 
 static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
 
 // Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
 using device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances =

library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp

@@ -17,10 +17,10 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 static constexpr auto ConvBwdDataDefault =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
 
 static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
 
 // Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
 using device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances =

library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp

@@ -16,10 +16,10 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 static constexpr auto ConvBwdDataDefault =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
 
 static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
 
 // Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
 using device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances =

library/src/tensor_operation_instance/gpu/convnd_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp

@@ -17,10 +17,10 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 static constexpr auto ConvBwdDataDefault =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
 
 static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Filter1x1Stride1Pad0;
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
 
 // Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
 using device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances =

library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp

@@ -20,7 +20,7 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization_t::Default;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
 // Compilation parameters for a[k, m] * b[k, n] = c[m, n]
 using device_gemm_xdl_f16_f16_f16_km_kn_mn_instances =

library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp

@@ -20,7 +20,7 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization_t::Default;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
 // Compilation parameters for a[k, m] * b[n, k] = c[m, n]
 using device_gemm_xdl_f16_f16_f16_km_nk_mn_instances =

library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp

@@ -20,7 +20,7 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization_t::Default;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances =

library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp

@@ -20,8 +20,8 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault   = ck::tensor_operation::device::GemmSpecialization_t::Default;
-static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization_t::MNPadding;
+static constexpr auto GemmDefault   = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
 
 // Compilation parameters for a[m, k] * b[n, k] = c[m, n]
 using device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances =

library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp

@@ -20,7 +20,7 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization_t::Default;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
 // Compilation parameters for a[k, m] * b[k, n] = c[m, n]
 using device_gemm_xdl_f32_f32_f32_km_kn_mn_instances =

library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp

@@ -20,7 +20,7 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization_t::Default;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
 // Compilation parameters for a[k, m] * b[n, k] = c[m, n]
 using device_gemm_xdl_f32_f32_f32_km_nk_mn_instances =

library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp

@@ -20,7 +20,7 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization_t::Default;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_gemm_xdl_f32_f32_f32_mk_kn_mn_instances =

library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp

@@ -20,7 +20,7 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization_t::Default;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
 // Compilation parameters for a[m, k] * b[n, k] = c[m, n]
 using device_gemm_xdl_f32_f32_f32_mk_nk_mn_instances =

library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp

@@ -20,7 +20,7 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization_t::Default;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instances = std::tuple<

library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp

@@ -20,7 +20,7 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization_t::Default;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instances = std::tuple<

library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp

@@ -20,7 +20,7 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization_t::Default;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances = std::tuple<

library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp

@@ -20,7 +20,7 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization_t::Default;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances = std::tuple<