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

profiler/src/profiler.cpp

@@ -85,26 +85,24 @@ int main(int argc, char* argv[])
     {
         return profile_reduce(argc, argv);
     }
-    else
-    {
+
     // clang-format off
     printf("arg1: tensor operation (gemm: GEMM\n"
            "                        gemm_bias_2d: GEMM+Bias(2D)\n"
            "                        gemm_bias_relu: GEMM+Bias+ReLU\n"
            "                        gemm_bias_relu_add: GEMM+Bias+ReLU+Add\n"
            "                        gemm_reduce: GEMM+Reduce\n"
-               "                        grouped_gemm: Grouped Gemm\n"
+           "                        grouped_gemm: Grouped GEMM\n"
            "                        conv_fwd: ForwardConvolution\n"
            "                        conv_fwd_bias_relu: ForwardConvolution+Bias+ReLU\n"
            "                        conv_fwd_bias_relu_add: ForwardConvolution+Bias+ReLU+Add\n"
            "                        conv_fwd_bias_relu_atomic_add: ForwardConvolution+Bias+ReLU+AtomicAdd\n"
-               "                        conv1d_bwd_data: BackwardConvolution data 1 dim\n"
-               "                        conv2d_bwd_data: BackwardConvolution data 2 dim\n"
-               "                        conv3d_bwd_data: BackwardConvolution data 3 dim\n"
-               "                        grouped_gemm: Grouped Gemm\n"
-               "                        reduce: REDUCE\n");
+           "                        conv1d_bwd_data: BackwardConvolution data 1d\n"
+           "                        conv2d_bwd_data: BackwardConvolution data 2d\n"
+           "                        conv3d_bwd_data: BackwardConvolution data 3d\n"
+           "                        grouped_gemm: Grouped GEMM\n"
+           "                        reduce: Reduce\n");
     // clang-format on
 
     return 0;
-    }
 }

script/cmake-rocm.sh

@@ -10,9 +10,11 @@ cmake
 -D CMAKE_INSTALL_PREFIX=${MY_PROJECT_INSTALL}                                                                                                  \
 -D BUILD_DEV=OFF                                                                                                                               \
 -D CMAKE_BUILD_TYPE=Release                                                                                                                    \
--D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 -ftemplate-backtrace-limit=0 -mllvm --amdgpu-spill-vgpr-to-agpr=0 -gline-tables-only "   \
+-D CMAKE_CXX_FLAGS=" --offload-arch=gfx908 --offload-arch=gfx90a -O3 -ftemplate-backtrace-limit=0 -gline-tables-only -save-temps=$PWD"         \
 -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc                                                                                                      \
 -D CMAKE_PREFIX_PATH=/opt/rocm                                                                                                                 \
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                                                                              \
 ${MY_PROJECT_SOURCE}
 
+#-D CMAKE_CXX_FLAGS=" --offload-arch=gfx908 --offload-arch=gfx90a -O3 -ftemplate-backtrace-limit=0 -mllvm --amdgpu-spill-vgpr-to-agpr=0 -gline-tables-only -save-temps=$PWD"   \
+#-D CMAKE_CXX_FLAGS=" --offload-arch=gfx908 --offload-arch=gfx90a -O3 -ftemplate-backtrace-limit=0 -gline-tables-only -save-temps=$PWD"         \

test/include/conv_test_util.hpp

@@ -31,7 +31,7 @@ using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
 using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
 
 static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default;
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
 
 template <ck::index_t SpatialDims, typename InDataType, typename WeiDataType, typename OutDataType>
 using DeviceConvNDFwdInstance = ck::tensor_operation::device::

test/magic_number_division/magic_number_division.cpp

@@ -5,7 +5,7 @@
 #include <stdlib.h>
 #include <half.hpp>
 #include "config.hpp"
-#include "print.hpp"
+#include "magic_division.hpp"
 #include "device.hpp"
 #include "host_tensor.hpp"
 #include "host_tensor_generator.hpp"

test/reduce/reduce_no_index.cpp

@@ -51,10 +51,10 @@ struct type_mapping<ck::half_t>
 
 constexpr int Rank = 4;
 
-constexpr ReduceTensorOp_t ReduceOpId      = ReduceTensorOp_t::AVG;
-constexpr NanPropagation_t NanOpt          = NanPropagation_t::PROPAGATE_NAN;
+constexpr ReduceTensorOp ReduceOpId      = ReduceTensorOp::AVG;
+constexpr NanPropagation NanOpt          = NanPropagation::PROPAGATE_NAN;
 constexpr bool PropagateNan              = false;
-constexpr ReduceTensorIndices_t IndicesOpt = ReduceTensorIndices_t::NO_INDICES;
+constexpr ReduceTensorIndices IndicesOpt = ReduceTensorIndices::NO_INDICES;
 constexpr bool NeedIndices               = false;
 
 template <typename InDataType,

test/reduce/reduce_with_index.cpp

@@ -51,10 +51,10 @@ struct type_mapping<ck::half_t>
 
 constexpr int Rank = 4;
 
-constexpr ReduceTensorOp_t ReduceOpId      = ReduceTensorOp_t::AMAX;
-constexpr NanPropagation_t NanOpt          = NanPropagation_t::PROPAGATE_NAN;
+constexpr ReduceTensorOp ReduceOpId      = ReduceTensorOp::AMAX;
+constexpr NanPropagation NanOpt          = NanPropagation::PROPAGATE_NAN;
 constexpr bool PropagateNan              = false;
-constexpr ReduceTensorIndices_t IndicesOpt = ReduceTensorIndices_t::FLATTENED_INDICES;
+constexpr ReduceTensorIndices IndicesOpt = ReduceTensorIndices::FLATTENED_INDICES;
 constexpr bool NeedIndices               = true;
 
 template <typename InDataType,