默认修改为混精的gemm计算

setup.py

@@ -64,7 +64,7 @@ def get_version_add(sha: Optional[str] = None) -> str:
     lines=[]
     with open(add_version_path, 'r',encoding='utf-8') as file:
         lines = file.readlines()
-    lines[2] = "__dcu_version__ = '0.1.0+{}'\n".format(version)
+    lines[2] = "__dcu_version__ = '0.2.6+{}'\n".format(version)
     with open(add_version_path, encoding="utf-8",mode="w") as file:
         file.writelines(lines)
     file.close()

src/turbomind/triton_backend/llama/LlamaTritonModel.cc

@@ -272,10 +272,20 @@ std::unique_ptr<LlamaTritonSharedModelInstance<T>> LlamaTritonModel<T>::createSh
     std::unique_ptr<cudaDeviceProp> cuda_device_prop_ptr(new cudaDeviceProp);
     ft::check_cuda_error(cudaGetDeviceProperties(cuda_device_prop_ptr.get(), device_id));
 
+    int hgemm_switch = 0;
+    const char* env_var_value_str = std::getenv("LMDEPLOY_HGEMM_SWITCH");
+    if (env_var_value_str != nullptr) {
+        hgemm_switch = std::stoi(env_var_value_str);
+    }
+
     if (std::is_same<T, half>::value) {
-        // cublas_wrapper->setGemmConfig(CUDA_R_16F, CUDA_R_16F, CUDA_R_16F, CUDA_R_32F);
+        if(hgemm_switch == 2){
             cublas_wrapper->setGemmConfig(CUDA_R_16F, CUDA_R_16F, CUDA_R_16F, CUDA_R_16F);
         }
+        else{
+            cublas_wrapper->setGemmConfig(CUDA_R_16F, CUDA_R_16F, CUDA_R_16F, CUDA_R_32F);
+        }
+    }
     else if (std::is_same<T, float>::value) {
         cublas_wrapper->setFP32GemmConfig();
     }

src/turbomind/utils/cublasMMWrapper.cc

@@ -40,6 +40,18 @@ cublasMMWrapper::cublasMMWrapper(cublasHandle_t   cublas_handle,
     if (allocator_ != nullptr) {
         cublas_workspace_ = allocator_->reMalloc(cublas_workspace_, CUBLAS_WORKSPACE_SIZE, false);
     }
+	# hgemm-switch 0:fp32r,1:fp16r-fp32r,2:fp16r		----xzhou 20240427
+    m_ihgemm_switch = 0;
+    const char* env_var_value_str = std::getenv("LMDEPLOY_HGEMM_SWITCH");
+    if (env_var_value_str != nullptr) {
+        m_ihgemm_switch = std::stoi(env_var_value_str);
+    }
+    m_ihgemm_switch_n = 16;
+    const char* env_n_value_str = std::getenv("LMDEPLOY_HGEMM_SWITCH_N");
+    if (env_n_value_str != nullptr) {
+        m_ihgemm_switch_n = std::stoi(env_n_value_str);
+    }
+    if(m_ihgemm_switch == 0) printf("hgemm_switch=%d, hgemm_switch_n_limit=%d\n", m_ihgemm_switch, m_ihgemm_switch_n);
 }
 
 #ifdef SPARSITY_ENABLED
@@ -113,6 +125,10 @@ void cublasMMWrapper::Gemm(cublasOperation_t transa,
 {
     TM_LOG_DEBUG(__PRETTY_FUNCTION__);
     mu_->lock();
+	# hgemm-switch xzhou 20240427
+    if(m_ihgemm_switch == 1 && (m == 5120 || m == 4096 || m == 12288 || m == 11008) && n <= m_ihgemm_switch_n && Atype == CUDA_R_16F){
+        computeType = CUDA_R_16F;
+    }
     check_cuda_error(cublasGemmEx(cublas_handle_,
                                   transa,
                                   transb,
@@ -172,7 +188,12 @@ void cublasMMWrapper::Gemm(cublasOperation_t transa,
 
     mu_->lock();
     // TODO: default cublas libs
-    int  is_fp16_computeType = computeType_ == CUDA_R_16F ? 1 : 0;
+    cudaDataType_t computeType = computeType_;
+	# hgemm-switch xzhou 20240427
+    if(m_ihgemm_switch == 1 && (m == 5120 || m == 4096 || m == 12288 || m == 11008) && n <= m_ihgemm_switch_n && Atype_ == CUDA_R_16F){
+        computeType = CUDA_R_16F;
+    }
+    int  is_fp16_computeType = computeType == CUDA_R_16F ? 1 : 0;
     bool using_cublasLt      = (Atype_ == CUDA_R_16F) ? true : false;
     int  batch_count         = 1;
     // fp32 use cublas as default
@@ -323,7 +344,7 @@ void cublasMMWrapper::Gemm(cublasOperation_t transa,
                                       C,
                                       Ctype_,
                                       ldc,
-                                      computeType_,
+                                      computeType,
                                       static_cast<cublasGemmAlgo_t>(cublasAlgo)));
         sync_check_cuda_error();
     // }
@@ -343,7 +364,7 @@ void cublasMMWrapper::setFP16GemmConfig()
     Atype_       = CUDA_R_16F;
     Btype_       = CUDA_R_16F;
     Ctype_       = CUDA_R_16F;
-    computeType_ = CUDA_R_16F;
+    computeType_ = CUDA_R_32F;
 }
 
 #ifdef ENABLE_BF16

src/turbomind/utils/cublasMMWrapper.h

@@ -48,6 +48,9 @@ protected:
     cublasAlgoMap* cublas_algo_map_;
     std::mutex*    mu_;
 
+    int m_ihgemm_switch;
+    int m_ihgemm_switch_n;
+
     IAllocator* allocator_        = nullptr;
     void*       cublas_workspace_ = nullptr;