Merge branch 'develop' into enable_navi_32_ci

src/targets/gpu/include/migraphx/gpu/device/argmin.hpp

@@ -25,7 +25,7 @@
 #define MIGRAPHX_GUARD_RTGLIB_DEVICE_ARGMIN_HPP
 
 #include <migraphx/argument.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/device/config.hpp>
 #include <hip/hip_runtime_api.h>
 
 namespace migraphx {
@@ -33,7 +33,10 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-void argmin(hipStream_t stream, const argument& result, const argument& arg, int64_t axis);
+void MIGRAPHX_DEVICE_EXPORT argmin(hipStream_t stream,
+                                   const argument& result,
+                                   const argument& arg,
+                                   int64_t axis);
 
 } // namespace device
 } // namespace gpu

src/targets/gpu/include/migraphx/gpu/device/config.hpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+#ifndef MIGRAPHX_GUARD_RTGLIB_DEVICE_CONFIG_HPP
+#define MIGRAPHX_GUARD_RTGLIB_DEVICE_CONFIG_HPP
+
+#include <migraphx/config.hpp>
+#include <migraphx/gpu/device/export.h>
+
+#endif

src/targets/gpu/include/migraphx/gpu/device/contiguous.hpp

@@ -25,7 +25,7 @@
 #define MIGRAPHX_GUARD_MIGRAPHLIB_KERNELS_HPP
 
 #include <migraphx/argument.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/device/config.hpp>
 #include <hip/hip_runtime_api.h>
 
 namespace migraphx {
@@ -33,7 +33,9 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-void contiguous(hipStream_t stream, const argument& result, const argument& arg);
+void MIGRAPHX_DEVICE_EXPORT contiguous(hipStream_t stream,
+                                       const argument& result,
+                                       const argument& arg);
 
 } // namespace device
 } // namespace gpu

src/targets/gpu/include/migraphx/gpu/device/fill.hpp

@@ -25,7 +25,7 @@
 #define MIGRAPHX_GUARD_RTGLIB_DEVICE_FILL_HPP
 
 #include <migraphx/argument.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/device/config.hpp>
 #include <hip/hip_runtime_api.h>
 
 namespace migraphx {
@@ -33,7 +33,7 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-void fill(hipStream_t stream, const argument& result, unsigned long val);
+void MIGRAPHX_DEVICE_EXPORT fill(hipStream_t stream, const argument& result, unsigned long val);
 
 } // namespace device
 } // namespace gpu

src/targets/gpu/include/migraphx/gpu/device/gather.hpp

@@ -25,7 +25,7 @@
 #define MIGRAPHX_GUARD_RTGLIB_DEVICE_GATHER_HPP
 
 #include <migraphx/argument.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/device/config.hpp>
 #include <hip/hip_runtime_api.h>
 
 namespace migraphx {
@@ -33,7 +33,8 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-argument gather(hipStream_t stream, argument result, argument arg1, argument arg2, int64_t axis);
+argument MIGRAPHX_DEVICE_EXPORT
+gather(hipStream_t stream, argument result, argument arg1, argument arg2, int64_t axis);
 
 } // namespace device
 } // namespace gpu

src/targets/gpu/include/migraphx/gpu/device/int8_gemm_pack.hpp

@@ -25,7 +25,7 @@
 #define MIGRAPHX_GUARD_RTGLIB_DEVICE_INT8_GEMM_PACK_HPP
 
 #include <migraphx/argument.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/device/config.hpp>
 #include <hip/hip_runtime_api.h>
 
 namespace migraphx {
@@ -33,9 +33,13 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-void int8_gemm_pack_a(hipStream_t stream, const argument& result, const argument& arg);
+void MIGRAPHX_DEVICE_EXPORT int8_gemm_pack_a(hipStream_t stream,
+                                             const argument& result,
+                                             const argument& arg);
 
-void int8_gemm_pack_b(hipStream_t stream, const argument& result, const argument& arg);
+void MIGRAPHX_DEVICE_EXPORT int8_gemm_pack_b(hipStream_t stream,
+                                             const argument& result,
+                                             const argument& arg);
 
 } // namespace device
 } // namespace gpu

src/targets/gpu/include/migraphx/gpu/device/logsoftmax.hpp

@@ -25,7 +25,7 @@
 #define MIGRAPHX_GUARD_RTGLIB_DEVICE_LOGSOFTMAX_HPP
 
 #include <migraphx/argument.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/device/config.hpp>
 #include <hip/hip_runtime_api.h>
 
 namespace migraphx {
@@ -33,7 +33,10 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-void logsoftmax(hipStream_t stream, const argument& result, const argument& arg, int64_t axis);
+void MIGRAPHX_DEVICE_EXPORT logsoftmax(hipStream_t stream,
+                                       const argument& result,
+                                       const argument& arg,
+                                       int64_t axis);
 
 } // namespace device
 } // namespace gpu

src/targets/gpu/include/migraphx/gpu/device/multinomial.hpp

@@ -25,7 +25,7 @@
 #define MIGRAPHX_GUARD_RTGLIB_DEVICE_MULTINOMIAL_HPP
 
 #include <migraphx/argument.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/device/config.hpp>
 #include <hip/hip_runtime_api.h>
 
 namespace migraphx {
@@ -33,10 +33,10 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-void multinomial(hipStream_t stream,
-                 const argument& result,
-                 const argument& arg0,
-                 const argument& arg1);
+void MIGRAPHX_DEVICE_EXPORT multinomial(hipStream_t stream,
+                                        const argument& result,
+                                        const argument& arg0,
+                                        const argument& arg1);
 
 } // namespace device
 } // namespace gpu

src/targets/gpu/include/migraphx/gpu/device/nonzero.hpp

@@ -25,7 +25,7 @@
 #define MIGRAPHX_GUARD_RTGLIB_DEVICE_NONZERO_HPP
 
 #include <migraphx/argument.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/device/config.hpp>
 #include <hip/hip_runtime_api.h>
 
 namespace migraphx {
@@ -33,7 +33,9 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-argument nonzero(hipStream_t stream, const argument& result, const argument& arg_data);
+argument MIGRAPHX_DEVICE_EXPORT nonzero(hipStream_t stream,
+                                        const argument& result,
+                                        const argument& arg_data);
 
 } // namespace device
 } // namespace gpu

src/targets/gpu/include/migraphx/gpu/device/pad.hpp

@@ -26,7 +26,7 @@
 #define MIGRAPHX_GUARD_RTGLIB_DEVICE_PAD_HPP
 
 #include <migraphx/argument.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/device/config.hpp>
 #include <hip/hip_runtime_api.h>
 
 namespace migraphx {
@@ -34,11 +34,11 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-argument pad(hipStream_t stream,
-             argument result,
-             argument arg1,
-             float value,
-             std::vector<std::int64_t> pads);
+argument MIGRAPHX_DEVICE_EXPORT pad(hipStream_t stream,
+                                    argument result,
+                                    argument arg1,
+                                    float value,
+                                    std::vector<std::int64_t> pads);
 
 } // namespace device
 } // namespace gpu

src/targets/gpu/include/migraphx/gpu/device/prefix_scan_sum.hpp

@@ -25,7 +25,7 @@
 #define MIGRAPHX_GUARD_DEVICE_PREFIX_SCAN_SUM_HPP
 
 #include <migraphx/argument.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/device/config.hpp>
 #include <hip/hip_runtime_api.h>
 
 namespace migraphx {
@@ -33,12 +33,12 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-void prefix_scan_sum(hipStream_t stream,
-                     const argument& result,
-                     const argument& arg,
-                     int32_t axis,
-                     bool exclusive,
-                     bool reverse);
+void MIGRAPHX_DEVICE_EXPORT prefix_scan_sum(hipStream_t stream,
+                                            const argument& result,
+                                            const argument& arg,
+                                            int32_t axis,
+                                            bool exclusive,
+                                            bool reverse);
 
 } // namespace device
 } // namespace gpu

src/targets/gpu/include/migraphx/gpu/device/reverse.hpp

@@ -25,7 +25,7 @@
 #define MIGRAPHX_GUARD_RTGLIB_DEVICE_REVERSE_HPP
 
 #include <migraphx/argument.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/device/config.hpp>
 #include <hip/hip_runtime_api.h>
 
 namespace migraphx {
@@ -33,8 +33,10 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-argument
-reverse(hipStream_t stream, argument result, argument arg1, const std::vector<int64_t>& axes);
+argument MIGRAPHX_DEVICE_EXPORT reverse(hipStream_t stream,
+                                        argument result,
+                                        argument arg1,
+                                        const std::vector<int64_t>& axes);
 
 } // namespace device
 } // namespace gpu

src/targets/gpu/include/migraphx/gpu/device/rnn_variable_seq_lens.hpp

@@ -25,7 +25,7 @@
 #define MIGRAPHX_GUARD_RTGLIB_DEVICE_RNN_VARIABLE_SEQ_LENS_HPP
 
 #include <migraphx/argument.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/device/config.hpp>
 #include <hip/hip_runtime_api.h>
 
 namespace migraphx {
@@ -33,22 +33,22 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-void rnn_var_sl_shift_sequence(hipStream_t stream,
-                               const argument& result,
-                               const argument& arg_hs,
-                               const argument& arg_sl);
+void MIGRAPHX_DEVICE_EXPORT rnn_var_sl_shift_sequence(hipStream_t stream,
+                                                      const argument& result,
+                                                      const argument& arg_hs,
+                                                      const argument& arg_sl);
 
-void rnn_var_sl_shift_output(hipStream_t stream,
-                             const argument& result,
-                             const argument& arg_hs,
-                             const argument& arg_sl,
-                             bool is_reverse);
+void MIGRAPHX_DEVICE_EXPORT rnn_var_sl_shift_output(hipStream_t stream,
+                                                    const argument& result,
+                                                    const argument& arg_hs,
+                                                    const argument& arg_sl,
+                                                    bool is_reverse);
 
-void rnn_var_sl_last_output(hipStream_t stream,
-                            const argument& result,
-                            const argument& arg_hs,
-                            const argument& arg_sl,
-                            bool is_reverse);
+void MIGRAPHX_DEVICE_EXPORT rnn_var_sl_last_output(hipStream_t stream,
+                                                   const argument& result,
+                                                   const argument& arg_hs,
+                                                   const argument& arg_sl,
+                                                   bool is_reverse);
 
 } // namespace device
 } // namespace gpu

src/targets/gpu/include/migraphx/gpu/device/scatter.hpp

@@ -25,7 +25,7 @@
 #define MIGRAPHX_GUARD_RTGLIB_DEVICE_SCATTER_HPP
 
 #include <migraphx/argument.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/device/config.hpp>
 #include <hip/hip_runtime_api.h>
 
 namespace migraphx {
@@ -33,7 +33,7 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-argument scatter(
+argument MIGRAPHX_DEVICE_EXPORT scatter(
     hipStream_t stream, argument result, argument arg0, argument arg1, argument arg2, int64_t axis);
 
 } // namespace device

src/targets/gpu/include/migraphx/gpu/device/topk.hpp

@@ -25,7 +25,7 @@
 #define MIGRAPHX_GUARD_RTGLIB_DEVICE_TOPK_HPP
 
 #include <migraphx/argument.hpp>
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/device/config.hpp>
 #include <hip/hip_runtime_api.h>
 
 namespace migraphx {
@@ -33,19 +33,19 @@ inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 namespace device {
 
-argument topk_smallest(hipStream_t stream,
-                       const argument& val_res,
-                       const argument& ind_res,
-                       const argument& arg,
-                       int64_t k,
-                       int64_t axis);
+argument MIGRAPHX_DEVICE_EXPORT topk_smallest(hipStream_t stream,
+                                              const argument& val_res,
+                                              const argument& ind_res,
+                                              const argument& arg,
+                                              int64_t k,
+                                              int64_t axis);
 
-argument topk_largest(hipStream_t stream,
-                      const argument& val_res,
-                      const argument& ind_res,
-                      const argument& arg,
-                      int64_t k,
-                      int64_t axis);
+argument MIGRAPHX_DEVICE_EXPORT topk_largest(hipStream_t stream,
+                                             const argument& val_res,
+                                             const argument& ind_res,
+                                             const argument& arg,
+                                             int64_t k,
+                                             int64_t axis);
 
 } // namespace device
 } // namespace gpu

src/targets/gpu/include/migraphx/gpu/device_name.hpp

@@ -24,7 +24,7 @@
 #ifndef MIGRAPHX_GUARD_GPU_DEVICE_NAME_HPP
 #define MIGRAPHX_GUARD_GPU_DEVICE_NAME_HPP
 
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/config.hpp>
 #include <string>
 
 struct hipDeviceProp_t;
@@ -33,11 +33,11 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 
-std::string get_arch_name(const hipDeviceProp_t& props);
+MIGRAPHX_GPU_EXPORT std::string get_arch_name(const hipDeviceProp_t& props);
 
-std::string get_device_name();
+MIGRAPHX_GPU_EXPORT std::string get_device_name();
 
-int get_device_id();
+MIGRAPHX_GPU_EXPORT int get_device_id();
 
 } // namespace gpu
 } // namespace MIGRAPHX_INLINE_NS

src/targets/gpu/include/migraphx/gpu/fuse_mlir.hpp

@@ -24,7 +24,6 @@
 #ifndef MIGRAPHX_GUARD_GPU_FUSE_MLIR_HPP
 #define MIGRAPHX_GUARD_GPU_FUSE_MLIR_HPP
 
-#include <migraphx/config.hpp>
 #include <migraphx/gpu/context.hpp>
 
 namespace migraphx {
@@ -34,9 +33,9 @@ struct module_pass_manager;
 
 namespace gpu {
 
-bool mlir_enabled();
+MIGRAPHX_GPU_EXPORT bool mlir_enabled();
 
-struct fuse_mlir
+struct MIGRAPHX_GPU_EXPORT fuse_mlir
 {
     context* ctx = nullptr;
     std::string name() const { return "gpu::fuse_mlir"; }

src/targets/gpu/include/migraphx/gpu/hip.hpp

@@ -24,7 +24,7 @@
 #ifndef MIGRAPHX_GUARD_MIGRAPHLIB_HIP_HPP
 #define MIGRAPHX_GUARD_MIGRAPHLIB_HIP_HPP
 
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/config.hpp>
 #include <migraphx/argument.hpp>
 #include <migraphx/literal.hpp>
 #include <migraphx/check_shapes.hpp>
@@ -38,26 +38,26 @@ namespace gpu {
 
 struct context;
 
-std::string hip_error(int error);
+MIGRAPHX_GPU_EXPORT std::string hip_error(int error);
 
-argument allocate_gpu(const shape& s, bool host = false);
+MIGRAPHX_GPU_EXPORT argument allocate_gpu(const shape& s, bool host = false);
 
-argument register_on_gpu(const argument& arg);
+MIGRAPHX_GPU_EXPORT argument register_on_gpu(const argument& arg);
 
-argument to_gpu(const argument& arg, bool host = false);
+MIGRAPHX_GPU_EXPORT argument to_gpu(const argument& arg, bool host = false);
 
-argument from_gpu(const argument& arg);
+MIGRAPHX_GPU_EXPORT argument from_gpu(const argument& arg);
 
-void set_device(std::size_t id);
+MIGRAPHX_GPU_EXPORT void set_device(std::size_t id);
 
-void gpu_sync();
-void gpu_sync(const context& ctx);
+MIGRAPHX_GPU_EXPORT void gpu_sync();
+MIGRAPHX_GPU_EXPORT void gpu_sync(const context& ctx);
 
-void gpu_copy(context& ctx, const argument& src, const argument& dst);
-void copy_to_gpu(context& ctx, const argument& src, const argument& dst);
-void copy_from_gpu(context& ctx, const argument& src, const argument& dst);
+MIGRAPHX_GPU_EXPORT void gpu_copy(context& ctx, const argument& src, const argument& dst);
+MIGRAPHX_GPU_EXPORT void copy_to_gpu(context& ctx, const argument& src, const argument& dst);
+MIGRAPHX_GPU_EXPORT void copy_from_gpu(context& ctx, const argument& src, const argument& dst);
 
-argument get_preallocation(context& ctx, const std::string& id);
+MIGRAPHX_GPU_EXPORT argument get_preallocation(context& ctx, const std::string& id);
 
 struct hip_allocate
 {
@@ -187,7 +187,8 @@ struct hip_copy
     std::ptrdiff_t output_alias(const std::vector<shape>&) const { return 1; }
 };
 
-void store_preallocated_param(context& ctx, const std::string& id, const argument& a);
+MIGRAPHX_GPU_EXPORT void
+store_preallocated_param(context& ctx, const std::string& id, const argument& a);
 
 struct hip_allocate_memory
 {

src/targets/gpu/include/migraphx/gpu/kernel.hpp

@@ -24,7 +24,7 @@
 #ifndef MIGRAPHX_GUARD_RTGLIB_KERNEL_HPP
 #define MIGRAPHX_GUARD_RTGLIB_KERNEL_HPP
 
-#include <migraphx/config.hpp>
+#include <migraphx/gpu/config.hpp>
 #include <migraphx/gpu/pack_args.hpp>
 #include <hip/hip_runtime_api.h>
 #include <memory>
@@ -37,7 +37,7 @@ namespace gpu {
 
 struct kernel_impl;
 
-struct kernel
+struct MIGRAPHX_GPU_EXPORT kernel
 {
     kernel() = default;
     kernel(const char* image, const std::string& name);

src/targets/gpu/include/migraphx/gpu/lowering.hpp

@@ -24,7 +24,6 @@
 #ifndef MIGRAPHX_GUARD_RTGLIB_MIOPEN_LOWERING_HPP
 #define MIGRAPHX_GUARD_RTGLIB_MIOPEN_LOWERING_HPP
 
-#include <migraphx/config.hpp>
 #include <migraphx/gpu/context.hpp>
 
 namespace migraphx {
@@ -40,7 +39,7 @@ namespace gpu {
  * * Maps instructions to their GPU-specific counterparts.
  * * Inserts `allocate` instructions before GPU operators.
  */
-struct lowering
+struct MIGRAPHX_GPU_EXPORT lowering
 {
     context* ctx;
     bool offload_copy;