Clean up exceptions

ark/api/data_type.cpp

@@ -7,7 +7,7 @@
 
 #include "bfloat16.h"
 #include "half.h"
-#include "logging.h"
+#include "logging.hpp"
 #include "model/model_data_type.hpp"
 
 namespace ark {
@@ -50,7 +50,7 @@
     }
     auto it = instances.find(type_name);
     if (it == instances.end()) {
-        ERR(InvalidUsageError, "Unknown data type: ", type_name);
+        ERR(UnsupportedError, "Unknown data type: ", type_name);
     }
     return *(it->second);
 }

ark/api/dims.cpp

@@ -5,8 +5,7 @@
 
 #include <vector>
 
-#include "error.hpp"
-#include "logging.h"
+#include "logging.hpp"
 
 namespace ark {
 

ark/api/error_test.cpp

@@ -0,0 +1,29 @@
+// Copyright (c) Microsoft Corporation.
+// Licensed under the MIT license.
+
+#include "ark/error.hpp"
+
+#include "unittest/unittest_utils.h"
+
+ark::unittest::State test_error() {
+    UNITTEST_THROW(throw ark::ModelError("test"), ark::ModelError);
+
+    try {
+        throw ark::ModelError("test");
+    } catch (const ark::ModelError &e) {
+        UNITTEST_EQ(std::string(e.what()), "test");
+    }
+
+    try {
+        throw ark::ModelError("test");
+    } catch (const ark::BaseError &e) {
+        UNITTEST_EQ(std::string(e.what()), "test");
+    }
+
+    return ark::unittest::SUCCESS;
+}
+
+int main() {
+    UNITTEST(test_error);
+    return 0;
+}

ark/api/executor.cpp

@@ -19,7 +19,7 @@
 #include "gpu/gpu_kernel.h"
 #include "gpu/gpu_logging.h"
 #include "gpu/gpu_manager.h"
-#include "logging.h"
+#include "logging.hpp"
 #include "model/model_buffer.hpp"
 #include "model/model_data_type.hpp"
 #include "model/model_tensor.hpp"

ark/api/init.cpp

@@ -9,9 +9,7 @@
 
 #include "env.h"
 #include "file_io.h"
-#include "logging.h"
-
-#define SHM_DIR "/dev/shm/"
+#include "logging.hpp"
 
 namespace ark {
 

ark/api/model.cpp

@@ -5,7 +5,7 @@
 
 #include <limits>
 
-#include "logging.h"
+#include "logging.hpp"
 
 namespace ark {
 

ark/api/model_graph.cpp

@@ -3,7 +3,7 @@
 
 #include "ark/model_graph.hpp"
 
-#include "logging.h"
+#include "logging.hpp"
 #include "model/model_graph_impl.hpp"
 #include "model/model_node.hpp"
 

ark/api/model_test.cpp

@@ -5,7 +5,7 @@
 
 #include <algorithm>
 
-#include "logging.h"
+#include "logging.hpp"
 #include "model/model_node.hpp"
 #include "model/model_op.hpp"
 #include "unittest/unittest_utils.h"

ark/api/planner.cpp

@@ -48,8 +48,7 @@
 static void check_config_field(const ModelOpRef op, const Json &config,
                                const std::string &field) {
     if (!config.contains(field)) {
-        ERR(NotFoundError, "Config field not found: ", field, " in ",
-            op->name());
+        ERR(PlanError, "Config field not found: ", field, " in ", op->name());
     }
 }
 

ark/arch.cpp

@@ -5,7 +5,7 @@
 
 #include <map>
 
-#include "logging.h"
+#include "logging.hpp"
 
 namespace ark {
 
@@ -85,7 +85,7 @@
     }
     auto it = instances.find(type_name);
     if (it == instances.end()) {
-        ERR(InvalidUsageError, "Unknown architecture type: ", type_name);
+        ERR(UnsupportedError, "Unknown architecture type: ", type_name);
     }
     return it->second;
 }

ark/codegen.cpp

@@ -8,7 +8,7 @@
 #include "ark/data_type.hpp"
 #include "env.h"
 #include "file_io.h"
-#include "logging.h"
+#include "logging.hpp"
 #include "model/model_buffer.hpp"
 #include "model/model_data_type.hpp"
 #include "model/model_op.hpp"
@@ -166,7 +166,7 @@
     const std::string &template_path =
         ark_root + "/include/kernels/kernel_template.in";
     if (!is_file(template_path)) {
-        ERR(SchedulerError, "kernel template file not found: ", template_path);
+        ERR(InternalError, "kernel template file not found: ", template_path);
     }
     std::string template_code = read_file(template_path);
     std::map<std::string, std::string> replacements = {
@@ -276,7 +276,7 @@
                                 rg_json["ProcessorRange"][1]);
     if (*rg_proc_range.begin() < *proc_range.begin() ||
         *rg_proc_range.end() > *proc_range.end()) {
-        ERR(SchedulerError, "invalid processor range of resource group");
+        ERR(PlanError, "invalid processor range of resource group");
     }
     Range<size_t> rg_warp_range(rg_json["WarpRange"][0],
                                 rg_json["WarpRange"][1]);
@@ -305,7 +305,7 @@
             n_slots = total_warps / num_warps_per_task;
         }
         if (n_slots == 0) {
-            ERR(SchedulerError, "not enough resources for task group");
+            ERR(PlanError, "not enough resources for task group");
         }
 
         size_t task_b = *task_range.begin();

ark/file_io.cpp

@@ -7,7 +7,7 @@
 #include <fstream>
 #include <sstream>
 
-#include "logging.h"
+#include "logging.hpp"
 
 namespace fs = std::filesystem;
 

ark/gpu/gpu_compile.cpp

@@ -78,7 +78,7 @@
     const std::string &output_file_path) {
 #if defined(ARK_CUDA)
     if (!arch->belongs_to(ARCH_CUDA)) {
-        ERR(InvalidUsageError, "Invalid architecture: ", arch->name());
+        ERR(UnsupportedError, "Invalid architecture: ", arch->name());
     }
     const std::string &cc = arch->category()[1];
 
@@ -107,7 +107,7 @@
 #elif defined(ARK_ROCM)
     // Remove prepending "rocm_" from arch to get the compute capability
     if (!arch->belongs_to(ARCH_ROCM)) {
-        ERR(InvalidUsageError, "Invalid architecture: ", arch->name());
+        ERR(UnsupportedError, "Invalid architecture: ", arch->name());
     }
     const std::string &cc = to_lower(arch->category()[1]);
 
@@ -195,7 +195,7 @@
             }
             std::string exec_print_str = exec_print.str();
             if (exec_print_str.size() > 0) {
-                ERR(ExecutorError, "\n", compile_cmd, "\n", exec_print_str,
+                ERR(InternalError, "\n", compile_cmd, "\n", exec_print_str,
                     "\n");
             }
             LOG(INFO, "Compile succeed: ", code_file_path, " (",

ark/gpu/gpu_logging.h

@@ -5,7 +5,7 @@
 #define ARK_GPU_LOGGING_H_
 
 #include "gpu/gpu.h"
-#include "logging.h"
+#include "logging.hpp"
 
 #define GLOG(cmd)                                           \
     do {                                                    \

ark/gpu/gpu_manager.cpp

@@ -61,11 +61,13 @@ GpuManager::Impl::Impl(int gpu_id) : gpu_id_(gpu_id) {
     // E.g.: "gfx90a:sramecc+:xnack-"
     std::string gcn_arch_name = prop.gcnArchName;
     if (gcn_arch_name.substr(0, 3) != "gfx") {
-        ERR(ExecutorError, "unexpected GCN architecture name: ", gcn_arch_name);
+        ERR(UnsupportedError,
+            "unexpected GCN architecture name: ", gcn_arch_name);
     }
     size_t pos_e = gcn_arch_name.find(":");
     if (pos_e == std::string::npos) {
-        ERR(ExecutorError, "unexpected GCN architecture name: ", gcn_arch_name);
+        ERR(UnsupportedError,
+            "unexpected GCN architecture name: ", gcn_arch_name);
     }
     // E.g.: "ROCM_90A"
     auto arch_name = "ROCM_" + to_upper(gcn_arch_name.substr(3, pos_e - 3));

ark/gpu/gpu_memory.cpp

@@ -41,8 +41,7 @@
     if (align_ == 0) {
         align_ = 1;
     } else if (align_ & (align_ - 1)) {
-        ERR(InvalidUsageError, "align must be a power of 2. Given %zu.",
-            align_);
+        ERR(InternalError, "align must be a power of 2. Given %zu.", align_);
     }
     manager_.set_current();
     bytes_raw_ = bytes_ + align_ - 1;

ark/include/ark.hpp

@@ -10,6 +10,7 @@
 
 #include <ark/data_type.hpp>
 #include <ark/dims.hpp>
+#include <ark/error.hpp>
 #include <ark/executor.hpp>
 #include <ark/init.hpp>
 #include <ark/model_graph.hpp>

ark/include/ark/error.hpp

@@ -0,0 +1,47 @@
+// Copyright (c) Microsoft Corporation.
+// Licensed under the MIT license.
+
+#ifndef ARK_ERROR_HPP
+#define ARK_ERROR_HPP
+
+#include <stdexcept>
+#include <string>
+
+namespace ark {
+
+/// Base class for all ARK errors.
+class BaseError : public std::exception {
+   private:
+    std::string msg_;
+
+   public:
+    BaseError(const std::string &msg) : msg_(msg) {}
+    const char *what() const noexcept override { return msg_.c_str(); }
+};
+
+#define REGISTER_ERROR_TYPE(_name)                        \
+    class _name : public BaseError {                      \
+       public:                                            \
+        _name(const std::string &msg) : BaseError(msg) {} \
+    };
+
+/// Internal error in ARK, likely a bug.
+REGISTER_ERROR_TYPE(InternalError)
+/// Invalid usage of ARK API.
+REGISTER_ERROR_TYPE(InvalidUsageError)
+/// Invalid ARK model definition or usage.
+REGISTER_ERROR_TYPE(ModelError)
+/// Invalid ARK plan definition or usage.
+REGISTER_ERROR_TYPE(PlanError)
+/// Unsupported feature triggered.
+REGISTER_ERROR_TYPE(UnsupportedError)
+/// Error from invalid system state such as a system call failure.
+REGISTER_ERROR_TYPE(SystemError)
+/// Error from a CUDA/HIP API call.
+REGISTER_ERROR_TYPE(GpuError)
+/// Error from a unit test.
+REGISTER_ERROR_TYPE(UnitTestError)
+
+}  // namespace ark
+
+#endif  // ARK_ERROR_HPP

ark/logging.cpp

@@ -1,7 +1,7 @@
 // Copyright (c) Microsoft Corporation.
 // Licensed under the MIT license.
 
-#include "logging.h"
+#include "logging.hpp"
 
 #include <unistd.h>
 

ark/logging.h → ark/logging.hpp

@@ -1,14 +1,14 @@
 // Copyright (c) Microsoft Corporation.
 // Licensed under the MIT license.
 
-#ifndef ARK_LOGGING_H_
-#define ARK_LOGGING_H_
+#ifndef ARK_LOGGING_HPP_
+#define ARK_LOGGING_HPP_
 
 #include <iostream>
 #include <sstream>
 #include <string>
 
-#include "error.hpp"
+#include "ark/error.hpp"
 
 namespace ark {
 
@@ -94,4 +94,4 @@ inline void _err(const std::string &file, int line, T value, Args... args) {
 
 }  // namespace ark
 
-#endif  // ARK_LOGGING_H_
+#endif  // ARK_LOGGING_HPP_

ark/model/model_buffer.cpp

@@ -3,7 +3,7 @@
 
 #include "model_buffer.hpp"
 
-#include "logging.h"
+#include "logging.hpp"
 
 namespace ark {
 
@@ -51,17 +51,13 @@
 
 std::shared_ptr<ModelBuffer> ModelBuffer::deserialize(const Json &serialized) {
     if (!serialized.contains("Id")) {
-        ERR(InvalidUsageError,
-            "ModelBuffer deserialization failed: missing Id");
+        ERR(ModelError, "ModelBuffer deserialization failed: missing Id");
     } else if (!serialized.contains("Rank")) {
-        ERR(InvalidUsageError,
-            "ModelBuffer deserialization failed: missing Rank");
+        ERR(ModelError, "ModelBuffer deserialization failed: missing Rank");
     } else if (!serialized.contains("SendTags")) {
-        ERR(InvalidUsageError,
-            "ModelBuffer deserialization failed: missing SendTags");
+        ERR(ModelError, "ModelBuffer deserialization failed: missing SendTags");
     } else if (!serialized.contains("RecvTags")) {
-        ERR(InvalidUsageError,
-            "ModelBuffer deserialization failed: missing RecvTags");
+        ERR(ModelError, "ModelBuffer deserialization failed: missing RecvTags");
     }
     return std::make_shared<ModelBuffer>(serialized["Id"], serialized["Rank"],
                                          serialized["SendTags"],