[Refactor] Phaseout LLVM Dependency by Making it Optional (#247)

* remove llvm build

* [Refactor] Update kernel compilation and profiling in examples

- Replaced `tilelang.lower` with `tilelang.compile` in multiple example scripts to streamline kernel compilation.
- Updated profiling calls to utilize the new `get_profiler` method, enhancing performance measurement consistency.
- Adjusted assertions and benchmarking methods to align with the new profiling structure across various examples, ensuring correctness and clarity in performance evaluations.

* lint fix

* License Update

* [Refactor] Improve code formatting and documentation in CUDA header and HIP runtime files

- Adjusted formatting in `cuda.h` for better readability, including alignment of comments and struct fields.
- Cleaned up whitespace and improved comment clarity in `rt_mod_hip.cc` to enhance code maintainability.

* [Refactor] Enhance formatting and clarity in CUDA header and HIP runtime files

- Improved comment alignment and readabilit...

src/target/rt_mod_hip.cc

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
-
 #if defined(__linux__)
 #include <sys/stat.h>
 #endif
@@ -14,108 +11,6 @@
 namespace tvm {
 namespace codegen {
 
-#define HIPRTC_CALL(x)                                                         \
-  \  
-  {                                                                            \
-    \  
-    hiprtcResult result = x;                                                   \
-    \  
-    if (result != HIPRTC_SUCCESS) {                                            \
-      \  
-      LOG(FATAL)                                                               \
-          << "HiprtcError: " #x " failed with error: "                         \
-          << hiprtcGetErrorString(result);                                     \
-      \  
-                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                   \
-    }                                                                          \
-    \  
-                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                   \
-  }
-
-static std::string FindHIPIncludePath() {
-#if defined(_WIN32)
-  const std::string delimiter = "\\";
-#else
-  const std::string delimiter = "/";
-#endif
-  std::string hip_include_path;
-  const char *hip_path_env = std::getenv("HIP_PATH");
-  if (hip_path_env != nullptr) {
-    hip_include_path += hip_path_env;
-    hip_include_path += delimiter + "include";
-    return hip_include_path;
-  }
-
-#if defined(__linux__)
-  struct stat st;
-  hip_include_path = "/opt/rocm/hip/include";
-  if (stat(hip_include_path.c_str(), &st) == 0) {
-    return hip_include_path;
-  }
-
-  if (stat("/usr/include/hip/hip_runtime.h", &st) == 0) {
-    return "/usr/include/hip";
-  }
-#endif
-  LOG(FATAL) << "Cannot find HIP include path."
-             << "HIP_PATH is not set or ROCm is not installed in the default "
-                "installation path."
-             << "In other than linux, it is necessary to set HIP_PATH.";
-  return hip_include_path;
-}
-
-static std::string HIPRTCCompile(const std::string &code,
-                                 bool include_path = false) {
-  std::vector<std::string> compile_params;
-  std::vector<const char *> param_cstrings{};
-  hiprtcProgram prog;
-  std::string cc =
-      "gfx900"; // Default target architecture (can be changed as needed)
-  int major, minor;
-  hipError_t e1 = hipDeviceGetAttribute(
-      &major, hipDeviceAttributeComputeCapabilityMajor, 0);
-  hipError_t e2 = hipDeviceGetAttribute(
-      &minor, hipDeviceAttributeComputeCapabilityMinor, 0);
-
-  if (e1 == hipSuccess && e2 == hipSuccess) {
-    cc = "gfx" + std::to_string(major * 100 + minor * 10);
-  } else {
-    LOG(WARNING) << "cannot detect compute capability from your device, "
-                 << "fall back to gfx900.";
-  }
-
-  compile_params.push_back("--gpu-architecture=" + cc);
-
-  if (include_path) {
-    std::string include_option = "--include-path=" + FindHIPIncludePath();
-    compile_params.push_back(include_option);
-  }
-
-  for (const auto &string : compile_params) {
-    param_cstrings.push_back(string.c_str());
-  }
-  HIPRTC_CALL(
-      hiprtcCreateProgram(&prog, code.c_str(), nullptr, 0, nullptr, nullptr));
-  hiprtcResult compile_res =
-      hiprtcCompileProgram(prog, param_cstrings.size(), param_cstrings.data());
-
-  size_t log_size;
-  HIPRTC_CALL(hiprtcGetProgramLogSize(prog, &log_size));
-  std::string log;
-  log.resize(log_size);
-  HIPRTC_CALL(hiprtcGetProgramLog(prog, &log[0]));
-  ICHECK_EQ(compile_res, HIPRTC_SUCCESS) << log;
-  size_t code_size;
-  HIPRTC_CALL(hiprtcGetCodeSize(prog, &code_size));
-
-  std::string code_out;
-  code_out.resize(code_size);
-  HIPRTC_CALL(hiprtcGetCode(prog, &code_out[0]));
-  HIPRTC_CALL(hiprtcDestroyProgram(&prog));
-
-  return code_out;
-}
-
 static std::unordered_map<std::string, runtime::FunctionInfo>
 ExtractFuncInfo(const IRModule &mod) {
   std::unordered_map<std::string, runtime::FunctionInfo> fmap;
@@ -173,13 +68,36 @@ runtime::Module BuildTileLangHIP(IRModule mod, Target target) {
     if (ptx[0] != '/')
       fmt = "hsaco";
   } else {
-    ptx = HIPRTCCompile(code, false);
+    ICHECK(false) << "tilelang_callback_hip_compile is not set";
   }
   return ROCMModuleCreate(ptx, fmt, ExtractFuncInfo(mod), code, std::string());
 }
 
+String BuildTileLangHIPWithoutCompile(IRModule mod, Target target) {
+  using tvm::runtime::Registry;
+  bool output_ssa = false;
+  CodeGenTileLangHIP cg;
+  cg.Init(output_ssa);
+
+  for (auto kv : mod->functions) {
+    ICHECK(kv.second->IsInstance<PrimFuncNode>())
+        << "CodeGenTileLangHIP: Can only take PrimFunc";
+    auto f = Downcast<PrimFunc>(kv.second);
+    auto calling_conv = f->GetAttr<Integer>(tvm::attr::kCallingConv);
+    ICHECK(calling_conv == CallingConv::kDeviceKernelLaunch);
+    cg.AddFunction(f);
+  }
+
+  std::string code = cg.Finish();
+  if (const auto *f = Registry::Get("tilelang_callback_hip_postproc")) {
+    code = (*f)(code, target).operator std::string();
+  }
+  return String(code);
+}
 TVM_REGISTER_GLOBAL("target.build.tilelang_hip")
     .set_body_typed(BuildTileLangHIP);
+TVM_REGISTER_GLOBAL("target.build.tilelang_hip_without_compile")
+    .set_body_typed(BuildTileLangHIPWithoutCompile);
 
 } // namespace codegen
 } // namespace tvm

src/target/utils.cc

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
-
 /*!
  * \file tl/target/utils.cc
  * \brief helper functions for target attributes.

src/target/utils.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
-
 /*!
  * \file tl/target/utils.h
  * \brief helper functions for target attributes.

src/tl_templates/cuda/common.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include <cuda_runtime.h>

src/tl_templates/cuda/copy.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include "common.h"

src/tl_templates/cuda/copy_sm90.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include <cuda.h>

src/tl_templates/cuda/cuda_fp8.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include <cute/numeric/numeric_types.hpp>

src/tl_templates/cuda/gemm.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 #if (defined(__CUDA_ARCH_LIST__) && (__CUDA_ARCH_LIST__ >= 900))
 #include "gemm_sm90.h"

src/tl_templates/cuda/gemm_sm70.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include <cutlass/cutlass.h>

src/tl_templates/cuda/gemm_sm80.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include <cute/algorithm/clear.hpp>

src/tl_templates/cuda/gemm_sm90.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include <cute/arch/mma_sm80.hpp>

src/tl_templates/cuda/ldsm.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include "common.h"

src/tl_templates/cuda/reduce.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include "common.h"

src/tl_templates/cuda/threadblock_swizzle.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include "common.h"

src/tl_templates/hip/common.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include <ck_tile/core.hpp>

src/tl_templates/hip/copy.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include "common.h"

src/tl_templates/hip/gemm.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include "common.h"

src/tl_templates/hip/ldsm.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include "common.h"
\ No newline at end of file

src/tl_templates/hip/reduce.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include "common.h"

src/tl_templates/hip/threadblock_swizzle.h

-// Copyright (c) Microsoft Corporation.
-// Licensed under the MIT License.
 #pragma once
 
 #include "common.h"