Revert "add cusparseLt 0.4 to speed up ffn1,ffn2,qkvo multiplication,…

…speed up…" (#109)
qingshui · Jan 16, 2024 · 8888367 · 8888367
1 parent ae56e86
commit 8888367
Show file tree

Hide file tree

Showing 30 changed files with 461 additions and 2,255 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -61,7 +61,6 @@ option(WITH_ASCEND_CL "Compile PaddlePaddle with ASCEND CL" ${WITH_ASCEND})
 option(WITH_ASCEND_CXX11 "Compile PaddlePaddle with ASCEND and CXX11 ABI" OFF)
 option(WITH_ONNXRUNTIME "Compile PaddlePaddle with ONNXRUNTIME" OFF)
 option(WITH_CUSPARSELT "Compile PaddlePaddle with CUSPARSELT" OFF)
-option(WITH_CUSPARSELT4 "Compile PaddlePaddle with CUSPARSELT version 0.4" ON)
 # Note(zhouwei): It use option above, so put here
 include(init)
 include(generic) # simplify cmake module

diff --git a/cmake/external/cusparselt4.cmake b/cmake/external/cusparselt4.cmake
diff --git a/cmake/inference_lib.cmake b/cmake/inference_lib.cmake
@@ -109,7 +109,7 @@ function(copy_part_of_thrid_party TARGET DST)
         DSTS ${dst_dir} ${dst_dir})
     endif()
 
-    if(WITH_SPARSELT OR WITH_SPARSELT4)
+    if(WITH_SPARSELT)
       set(dst_dir "${DST}/third_party/install/cusparselt")
       copy(
         ${TARGET}

diff --git a/cmake/third_party.cmake b/cmake/third_party.cmake
@@ -491,11 +491,6 @@ if(WITH_CUSPARSELT)
   list(APPEND third_party_deps extern_cusparselt)
 endif()
 
-if(WITH_CUSPARSELT4)
-  include(external/cusparselt4) # download, build, install cusparselt 0.4
-  list(APPEND third_party_deps extern_cusparselt)
-endif()
-
 if(WITH_GPU
    AND NOT WITH_ARM
    AND NOT WIN32

diff --git a/paddle/fluid/inference/api/analysis_predictor.cc b/paddle/fluid/inference/api/analysis_predictor.cc
@@ -49,7 +49,6 @@
 #include "paddle/fluid/inference/utils/model_utils.h"
 #include "paddle/fluid/inference/utils/singleton.h"
 #include "paddle/fluid/memory/memcpy.h"
-#include "paddle/fluid/operators/fused/cusparseLt.h"
 #include "paddle/fluid/platform/cpu_helper.h"
 #include "paddle/fluid/platform/device/gpu/gpu_info.h"
 #include "paddle/fluid/platform/device_context.h"
@@ -436,8 +435,6 @@ void AnalysisPredictor::InitDeviceContexts() {
           gpu_context->SetSolverHandle(
               gpu_resource->GetSolverDnHandleCreator());
           gpu_context->SetSparseHandle(gpu_resource->GetSparseHandleCreator());
-          gpu_context->SetSparseLtHandle(
-                 gpu_resource->GetSparseLtHandleCreator());
           gpu_context->SetEigenDevice(gpu_resource->GetGpuEigenDeviceCreator());
           gpu_context->SetComputeCapability(
               gpu_resource->GetGpuComputeCapability());
@@ -450,10 +447,6 @@ void AnalysisPredictor::InitDeviceContexts() {
               gpu_resource->GetGPUMultiProcessors());
           gpu_context->SetDriverVersion(gpu_resource->GetGpuDriverVersion());
           gpu_context->SetRuntimeVersion(gpu_resource->GetGpuRuntimeVersion());
-          paddle::operators::WeightCache::Instance().init(
-              place_,
-              phi::Stream(
-                  reinterpret_cast<phi::StreamId>(gpu_resource->GetStream())));
           VLOG(1) << "thread id is " << std::this_thread::get_id()
                   << ", stream id is "
                   << reinterpret_cast<void *>(gpu_resource->GetStream())

diff --git a/paddle/fluid/inference/api/infer_context.h b/paddle/fluid/inference/api/infer_context.h
@@ -34,7 +34,6 @@ class InferGPUContext : public phi::GPUContext {
   using phi::GPUContext::SetEigenDevice;
   using phi::GPUContext::SetSolverHandle;
   using phi::GPUContext::SetSparseHandle;
-  using phi::GPUContext::SetSparseLtHandle;
   using phi::GPUContext::SetStream;
   // using phi::GPUContext::SetDnnWorkspaceHandle;
   using phi::GPUContext::SetComputeCapability;

diff --git a/paddle/fluid/inference/api/resource_manager.cc b/paddle/fluid/inference/api/resource_manager.cc
@@ -168,7 +168,6 @@ void GPUContextResource::DestroyGPUResource() {
   DestroyBlasLtHandle();
   DestroySolverHandle();
   DestroySparseHandle();
-  DestroySparseLtHandle();
 }
 
 void GPUContextResource::InitGpuProperties() {
@@ -229,20 +228,6 @@ void GPUContextResource::DestroySparseHandle() {
   phi::DestroySparseHandle(sparse_handle_);
 }
 
-void GPUContextResource::InitSparseLtHandle() {
-  if (sparselt_handle_ == nullptr) {
-    sparselt_handle_ = &sparselt_ori_handle_;
-    phi::InitSparseLtHandle(sparselt_handle_);
-  }
-}
-
-void GPUContextResource::DestroySparseLtHandle() {
-  if (sparselt_handle_ != nullptr) {
-    phi::DestroySparseLtHandle(sparselt_handle_);
-    sparselt_handle_ = nullptr;
-  }
-}
-
 phi::Place GPUContextResource::Place() const { return place_; }
 
 gpuStream_t GPUContextResource::GetStream() const { return stream_; }
@@ -337,14 +322,6 @@ GPUContextResource::GetSparseHandleCreator() {
   };
 }
 
-std::function<cusparseLtHandle_t*()>
-GPUContextResource::GetSparseLtHandleCreator() {
-  return [&]() {
-    InitSparseLtHandle();
-    return sparselt_handle_;
-  };
-}
-
 Eigen::GpuDevice* GPUContextResource::GetGpuEigenDevice() const {
   return gpu_eigen_device_.get();
 }

diff --git a/paddle/fluid/inference/api/resource_manager.h b/paddle/fluid/inference/api/resource_manager.h
@@ -63,7 +63,6 @@ class GPUContextResource {
   std::function<phi::blasLtHandle_t()> GetBlasLtHandleCreator();
   std::function<phi::solverHandle_t()> GetSolverDnHandleCreator();
   std::function<phi::sparseHandle_t()> GetSparseHandleCreator();
-  std::function<cusparseLtHandle_t*()> GetSparseLtHandleCreator();
   std::function<Eigen::GpuDevice*()> GetGpuEigenDeviceCreator();
 
   gpuStream_t GetStream() const;
@@ -107,8 +106,6 @@ class GPUContextResource {
   void DestroySolverHandle();
   void InitSparseHandle();
   void DestroySparseHandle();
-  void InitSparseLtHandle();
-  void DestroySparseLtHandle();
 
  private:
   phi::Place place_;
@@ -133,8 +130,6 @@ class GPUContextResource {
   dnnHandle_t dnn_handle_{nullptr};
   phi::solverHandle_t solver_handle_{nullptr};
   phi::sparseHandle_t sparse_handle_{nullptr};
-  cusparseLtHandle_t* sparselt_handle_{nullptr};
-  cusparseLtHandle_t sparselt_ori_handle_;
   // DnnWorkspaceHandle
 };
 #endif