[CINN][new hardware] SYCL third PR: complete the SYCL logic (#71204)

feixi21 · web-flow · commit 5879c629cba0 · 2025-03-06T09:55:43.000+08:00
* complete sycl logic

* fix_bugs

* fix_bugs
diff --git a/paddle/cinn/backends/codegen_device_util.h b/paddle/cinn/backends/codegen_device_util.h
@@ -25,6 +25,9 @@
 #ifdef CINN_WITH_HIP
 #include "paddle/cinn/backends/hip/codegen_hip_dev.h"
 #endif
+#ifdef CINN_WITH_SYCL
+#include "paddle/cinn/backends/sycl/codegen_sycl_dev.h"
+#endif
 #include "paddle/cinn/cinn.h"
 #include "paddle/cinn/ir/ir.h"
 #include "paddle/cinn/ir/ir_mutator.h"
@@ -140,7 +143,14 @@ struct CollectHostFunctionVisitor : public ir::IRMutator<> {
           shared_mem_bytes = codegen_dev.GetDynSharedMemOffset();
 #endif
         },
-        [&](common::HygonDCUArchSYCL) { CINN_NOT_IMPLEMENTED });
+        [&](common::HygonDCUArchSYCL) {
+#ifdef CINN_WITH_SYCL
+          sycl::CodeGenSyclDevice codegen_dev(
+              cinn::common::DefaultHygonDcuSyclTarget());
+          codegen_dev.Compile(ir::LoweredFunc(func));
+          shared_mem_bytes = codegen_dev.GetDynSharedMemOffset();
+#endif
+        });
 
     VLOG(6) << "Add a call node for func->name " << func->name << "\n"
             << "grid_dim: (" << func->cuda_axis_info.grid_dim(0) << ", "
diff --git a/paddle/cinn/backends/compiler.cc b/paddle/cinn/backends/compiler.cc
@@ -35,6 +35,11 @@
 #include "paddle/cinn/backends/hip/compiler_hip.h"
 #include "paddle/cinn/runtime/hip/hip_module.h"
 #endif
+#ifdef CINN_WITH_SYCL
+#include "paddle/cinn/backends/sycl/codegen_sycl_dev.h"
+#include "paddle/cinn/backends/sycl/compiler_sycl.h"
+#include "paddle/cinn/runtime/sycl/sycl_module.h"
+#endif
 #include "paddle/cinn/adt/adt.h"
 
 PD_DECLARE_string(cinn_source_code_save_path);
@@ -288,7 +293,19 @@ std::string Compiler::GetSourceCode(const ir::Module& module) {
         CINN_NOT_IMPLEMENTED
 #endif
       },
-      [&](common::HygonDCUArchSYCL) -> std::string { CINN_NOT_IMPLEMENTED });
+      [&](common::HygonDCUArchSYCL) -> std::string {
+#ifdef CINN_WITH_SYCL
+        auto _host_module_device_module_ =
+            SplitDeviceAndHostModule(module);  // NOLINT
+        auto& host_module = std::get<0>(_host_module_device_module_);
+        auto& device_module = std::get<1>(_host_module_device_module_);
+        sycl::CodeGenSyclDevice codegen(target_);
+        auto source_code = codegen.Compile(device_module);
+        return source_code;
+#else
+        CINN_NOT_IMPLEMENTED
+#endif
+      });
 }
 
 void Compiler::BuildDefault(const Module& module) {
@@ -393,7 +410,38 @@ void Compiler::RegisterHipModuleSymbol() {
 #endif
 }
 
-void Compiler::RegisterSyclModuleSymbol() { CINN_NOT_IMPLEMENTED }
+void Compiler::RegisterSyclModuleSymbol() {
+#ifdef CINN_WITH_SYCL
+  syclrtc::Compiler compiler;
+  std::string source_code =
+      sycl::CodeGenSyclDevice::GetSourceHeader() + device_fn_code_;
+  std::string hsaco = compiler(source_code);
+  PADDLE_ENFORCE_EQ(
+      !hsaco.empty(),
+      true,
+      ::common::errors::Fatal("Compile hsaco failed from source code:\n%s",
+                              source_code));
+  using runtime::sycl::SYCLModule;
+  sycl_module_.reset(new SYCLModule(source_code, hsaco, SYCLModule::Kind::so));
+  // get device id
+  using cinn::runtime::BackendAPI;
+  int device_id = BackendAPI::get_backend(target_)->get_device();
+  // register kernel
+  RuntimeSymbols symbols;
+  for (const auto& kernel_fn_name : device_fn_name_) {
+    auto fn_kernel = sycl_module_->GetFunction(kernel_fn_name);
+    PADDLE_ENFORCE_NOT_NULL(
+        fn_kernel,
+        ::common::errors::Fatal("HIP GetFunction Error: get valid kernel."));
+    fn_ptr_.push_back(reinterpret_cast<void*>(fn_kernel));
+    symbols.RegisterVar(kernel_fn_name + "_ptr_",
+                        reinterpret_cast<void*>(fn_kernel));
+  }
+  engine_->RegisterModuleRuntimeSymbols(std::move(symbols));
+#else
+  CINN_NOT_IMPLEMENTED
+#endif
+}
 
 void Compiler::CompileCudaModule(const Module& module,
                                  const std::string& code) {
@@ -473,7 +521,39 @@ void Compiler::CompileHipModule(const Module& module, const std::string& code) {
 
 void Compiler::CompileSyclModule(const Module& module,
                                  const std::string& code) {
+#ifdef CINN_WITH_SYCL
+  auto _host_module_device_module_ =
+      SplitDeviceAndHostModule(module);  // NOLINT
+  auto& host_module = std::get<0>(_host_module_device_module_);
+  auto& device_module = std::get<1>(_host_module_device_module_);
+  VLOG(3) << "[SYCL] host module:\n" << host_module;
+  VLOG(3) << "[SYCL] device module:\n" << device_module;
+  std::string source_code;
+  if (!FLAGS_cinn_debug_custom_code_path.empty()) {
+    std::string file_path = FLAGS_cinn_debug_custom_code_path;
+    source_code = GetFileContent(file_path);
+  } else if (code.empty()) {
+    sycl::CodeGenSyclDevice codegen(target_);
+    source_code = codegen.Compile(device_module);
+  } else {
+    source_code = code;
+  }
+  PADDLE_ENFORCE_EQ(
+      !source_code.empty(),
+      true,
+      ::common::errors::Fatal(
+          "Compile SYCL code failed from device module:\n%s", device_module));
+  VLOG(3) << "[SYCL]:\n" << source_code;
+  SourceCodePrint::GetInstance()->write(source_code);
+  device_fn_code_ += source_code;
+  for (auto& fn : device_module.functions()) {
+    std::string kernel_fn_name = fn->name;
+    device_fn_name_.emplace_back(kernel_fn_name);
+  }
+  engine_->Link<CodeGenGpuHost>(host_module);
+#else
   CINN_NOT_IMPLEMENTED
+#endif
 }
 
 void Compiler::CompileX86Module(const Module& module) {
diff --git a/paddle/cinn/backends/compiler.h b/paddle/cinn/backends/compiler.h
@@ -32,6 +32,9 @@
 #ifdef CINN_WITH_HIP
 #include "paddle/cinn/runtime/hip/hip_module.h"
 #endif
+#ifdef CINN_WITH_SYCL
+#include "paddle/cinn/runtime/sycl/sycl_module.h"
+#endif
 
 namespace cinn {
 namespace backends {
@@ -167,6 +170,9 @@ class Compiler final {
 #ifdef CINN_WITH_HIP
   std::unique_ptr<runtime::hip::HIPModule> hip_module_;
 #endif
+#ifdef CINN_WITH_SYCL
+  std::unique_ptr<runtime::sycl::SYCLModule> sycl_module_;
+#endif
 };
 
 }  // namespace backends
diff --git a/paddle/cinn/common/target.cc b/paddle/cinn/common/target.cc
@@ -359,10 +359,10 @@ const Target &DefaultHygonDcuSyclTarget() {
 const Target &DefaultDeviceTarget() {
 #ifdef CINN_WITH_CUDA
   return DefaultNVGPUTarget();
-#elif defined(CINN_WITH_HIP)
-  return DefaultHygonDcuHipTarget();
 #elif defined(CINN_WITH_SYCL)
   return DefaultHygonDcuSyclTarget();
+#elif defined(CINN_WITH_HIP)
+  return DefaultHygonDcuHipTarget();
 #endif
 }
 
@@ -400,10 +400,10 @@ int GetMaxBlocks() {
 const Target &DefaultTarget() {
 #ifdef CINN_WITH_CUDA
   return DefaultNVGPUTarget();
-#elif defined(CINN_WITH_HIP)
-  return DefaultHygonDcuHipTarget();
 #elif defined(CINN_WITH_SYCL)
   return DefaultHygonDcuSyclTarget();
+#elif defined(CINN_WITH_HIP)
+  return DefaultHygonDcuHipTarget();
 #else
   return DefaultHostTarget();
 #endif
diff --git a/paddle/cinn/optim/optimize.cc b/paddle/cinn/optim/optimize.cc
@@ -107,7 +107,7 @@ ir::LoweredFunc Optimize(ir::LoweredFunc fn,
         VLOG(10) << "After Optimize TransBufferWithDynamicShape:" << copied;
 #endif
       },
-      [&](common::HygonDCUArchHIP) {
+      [&](std::variant<common::HygonDCUArchHIP, common::HygonDCUArchSYCL>) {
 #ifdef CINN_WITH_HIP
         ir::SetCudaAxisInfo(copied);
         if (remove_gpu_for_loops) {
@@ -124,7 +124,6 @@ ir::LoweredFunc Optimize(ir::LoweredFunc fn,
         VLOG(10) << "After Optimize CudaSyncThreadsDropIfThenElse:" << copied;
 #endif
       },
-      [&](common::HygonDCUArchSYCL) { CINN_NOT_IMPLEMENTED },
       [](auto) {});
 
   SimplifyUnitBlock(&copied->body);
diff --git a/paddle/cinn/optim/trans_buffer_with_dynamic_shape.cc b/paddle/cinn/optim/trans_buffer_with_dynamic_shape.cc
@@ -200,7 +200,19 @@ LogicalResult TransBufferWithDynamicShapePass::Run(ir::LoweredFunc func) {
                 "The shared memory size used by current kernel is greater "
                 "than the max shared memory per block"));
       },
-      [&](common::HygonDCUArchSYCL) { CINN_NOT_IMPLEMENTED });
+      [&](common::HygonDCUArchSYCL) {
+        using cinn::runtime::BackendAPI;
+        size_t max_shm_per_block =
+            BackendAPI::get_backend(common::HygonDCUArchSYCL{})
+                ->get_device_property(
+                    BackendAPI::DeviceProperty::MaxSharedMemoryPerBlock);
+        PADDLE_ENFORCE_LE(
+            mutator.shared_mem_size_used(),
+            max_shm_per_block,
+            ::common::errors::InvalidArgument(
+                "The shared memory size used by current kernel is greater "
+                "than the max shared memory per block"));
+      });
   return LogicalResult::success();
 }
 
diff --git a/paddle/cinn/runtime/sycl/sycl_backend_api.cc b/paddle/cinn/runtime/sycl/sycl_backend_api.cc
@@ -45,9 +45,11 @@ void SYCLBackendAPI::Init(Arch arch) {
         backend = ::sycl::backend::ext_oneapi_hip;
       });
   // look for matched devices
-  for (auto device : devices) {
-    if (device.get_backend() == backend) {
-      this->devices.push_back(device);
+  if (this->devices.size() < 8) {
+    for (auto device : devices) {
+      if (device.get_backend() == backend) {
+        this->devices.push_back(device);
+      }
     }
   }
   if (this->devices.size() == 0) {
diff --git a/paddle/cinn/runtime/use_extern_funcs.h b/paddle/cinn/runtime/use_extern_funcs.h
@@ -18,6 +18,11 @@
 #ifdef CINN_WITH_CUDA
 #include "paddle/cinn/runtime/cuda/use_extern_funcs.h"
 #endif
+
 #ifdef CINN_WITH_HIP
 #include "paddle/cinn/runtime/hip/use_extern_funcs.h"
 #endif
+
+#ifdef CINN_WITH_SYCL
+#include "paddle/cinn/runtime/sycl/use_extern_funcs.h"
+#endif
