You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@singa.apache.org by wa...@apache.org on 2015/11/26 12:32:03 UTC
[6/6] incubator-singa git commit: SINGA-104 Add Context Class
SINGA-104 Add Context Class
check with cpplint
Project: http://git-wip-us.apache.org/repos/asf/incubator-singa/repo
Commit: http://git-wip-us.apache.org/repos/asf/incubator-singa/commit/b2cfa17b
Tree: http://git-wip-us.apache.org/repos/asf/incubator-singa/tree/b2cfa17b
Diff: http://git-wip-us.apache.org/repos/asf/incubator-singa/diff/b2cfa17b
Branch: refs/heads/master
Commit: b2cfa17b8564dff993d195b0dd89be0bad0813a6
Parents: e3bda08
Author: WANG Sheng <wa...@gmail.com>
Authored: Thu Nov 26 19:30:42 2015 +0800
Committer: WANG Sheng <wa...@gmail.com>
Committed: Thu Nov 26 19:30:42 2015 +0800
----------------------------------------------------------------------
include/singa/utils/context.h | 74 ++--
src/test/test_context.cc | 35 +-
src/test/test_math.cc | 747 ++++++++++++++++++-------------------
src/test/test_msg.cc | 2 +-
src/test/test_paramslicer.cc | 2 +-
5 files changed, 423 insertions(+), 437 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/b2cfa17b/include/singa/utils/context.h
----------------------------------------------------------------------
diff --git a/include/singa/utils/context.h b/include/singa/utils/context.h
index 5066633..905b810 100644
--- a/include/singa/utils/context.h
+++ b/include/singa/utils/context.h
@@ -22,22 +22,19 @@
#ifndef SINGA_UTILS_CONTEXT_H_
#define SINGA_UTILS_CONTEXT_H_
-#include <vector>
-#include <random>
+#include <glog/logging.h>
#include <chrono>
+#include <random>
#include <thread>
#include <unordered_map>
-#include <glog/logging.h>
-
+#include <vector>
#ifdef USE_GPU
#include "singa/utils/cuda_utils.h"
#endif
-
namespace singa {
-
/**
* Context is used as a global singleton, which stores the mapping from CPU
* thread id to GPU device id. If a thread has no GPU, then its associated
@@ -52,30 +49,29 @@ class Context {
/**
* Destructor, release random generators and handlers.
*/
- ~Context() {
+ ~Context() {
#ifdef USE_GPU
- for (auto& entry : device_id_) {
- if (entry.second != -1) {
- cudaSetDevice(entry.second);
- if (cublas_handle_[entry.second] != nullptr) {
- cublasDestroy(cublas_handle_[entry.second]);
- cublas_handle_[entry.second] = nullptr;
- }
- if(curand_generator_[entry.second] != nullptr) {
- curandDestroyGenerator(curand_generator_[entry.second]);
- curand_generator_[entry.second] = nullptr;
- }
- }
- }
+ for (auto& entry : device_id_) {
+ if (entry.second != -1) {
+ cudaSetDevice(entry.second);
+ if (cublas_handle_[entry.second] != nullptr) {
+ cublasDestroy(cublas_handle_[entry.second]);
+ cublas_handle_[entry.second] = nullptr;
+ }
+ if (curand_generator_[entry.second] != nullptr) {
+ curandDestroyGenerator(curand_generator_[entry.second]);
+ curand_generator_[entry.second] = nullptr;
+ }
+ }
+ }
#endif
- for (auto& entry : rand_generator_) {
- if (entry.second != nullptr) {
- delete entry.second;
- entry.second = nullptr;
- }
- }
-
- }
+ for (auto& entry : rand_generator_) {
+ if (entry.second != nullptr) {
+ delete entry.second;
+ entry.second = nullptr;
+ }
+ }
+ }
/**
* Constructor, init handlers and GPU rand generators to nullptr.
*/
@@ -90,12 +86,12 @@ class Context {
* @return the ID of the device attached to a given CPU thread, or -1 if this
* thread has not been attached GPU device.
*/
- int device_id(const std::thread::id& tid) {
+ int device_id(const std::thread::id& tid) {
if (device_id_.find(tid) != device_id_.end())
return device_id_[tid];
else
return -1;
- }
+ }
/**
* Setup the CPU thread, which may be assigned a GPU device.
* If there is no GPU device, then set did to -1.
@@ -168,7 +164,7 @@ class Context {
/**
* Get the rand generator of the GPU device assigned to the given thread.
*/
- curandGenerator_t curand_generator(const std::thread::id thread_id) {
+ curandGenerator_t curand_generator(const std::thread::id thread_id) {
return curand_generator(device_id(thread_id));
}
/**
@@ -177,10 +173,10 @@ class Context {
* @return random generator. If it does not exist, then create one.
* The random seed will be set to CURAND_RNG_PSEUDO_DEFAULT if it is not set.
*/
- curandGenerator_t curand_generator(const int device_id) {
+ curandGenerator_t curand_generator(const int device_id) {
CHECK_GE(device_id, 0);
if (curand_generator_.at(device_id) == nullptr) {
- // TODO handle user set seed
+ // TODO(wangwei) handle user set seed
/*
CHECK(seed_.find(tid) != seed_.end());
auto seed = seed_[tid];
@@ -189,8 +185,8 @@ class Context {
curandCreateGenerator(&curand_generator_[device_id],
CURAND_RNG_PSEUDO_DEFAULT);
}
- return curand_generator_[device_id];
- }
+ return curand_generator_[device_id];
+ }
#endif
@@ -198,19 +194,19 @@ class Context {
//!< max num of GPUs per process
const int kMaxNumGPU = 64;
//!< map from thread id to device id
- std::unordered_map<std::thread::id, int> device_id_;
+ std::unordered_map<std::thread::id, int> device_id_;
//!< map from thread id to cpu rand generator
std::unordered_map<std::thread::id, std::mt19937 *> rand_generator_;
//!< map from thread id to cpu rand generator seed
std::unordered_map<std::thread::id, int> seed_;
#ifdef USE_GPU
//!< cublas handler indexed by GPU device ID
- std::vector<cublasHandle_t> cublas_handle_;
+ std::vector<cublasHandle_t> cublas_handle_;
//!< cublas rand generator indexed by GPU device ID
- std::vector<curandGenerator_t> curand_generator_;
+ std::vector<curandGenerator_t> curand_generator_;
#endif
};
} // namespace singa
-#endif // SINGA_UTILS_MATH_ADDR_H_
+#endif // SINGA_UTILS_CONTEXT_H_
http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/b2cfa17b/src/test/test_context.cc
----------------------------------------------------------------------
diff --git a/src/test/test_context.cc b/src/test/test_context.cc
index 5e501b9..70f6d07 100644
--- a/src/test/test_context.cc
+++ b/src/test/test_context.cc
@@ -1,3 +1,24 @@
+/************************************************************
+*
+* Licensed to the Apache Software Foundation (ASF) under one
+* or more contributor license agreements. See the NOTICE file
+* distributed with this work for additional information
+* regarding copyright ownership. The ASF licenses this file
+* to you under the Apache License, Version 2.0 (the
+* "License"); you may not use this file except in compliance
+* with the License. You may obtain a copy of the License at
+*
+* http://www.apache.org/licenses/LICENSE-2.0
+*
+* Unless required by applicable law or agreed to in writing,
+* software distributed under the License is distributed on an
+* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+* KIND, either express or implied. See the License for the
+* specific language governing permissions and limitations
+* under the License.
+*
+*************************************************************/
+
#include <thread>
#include "gtest/gtest.h"
#include "singa/utils/singleton.h"
@@ -25,17 +46,17 @@ TEST(ContextTest, TestHandle) {
float A[12];
float B[12];
- for(int i = 0; i < 12; i++) {
- A[i]=i-1;
- B[i]=i+1;
+ for (int i = 0; i < 12; i++) {
+ A[i] = i - 1;
+ B[i] = i + 1;
}
float* A_gpu = NULL;
float* B_gpu = NULL;
context->SetupDevice(std::this_thread::get_id(), 0);
- cudaMalloc((void**)&A_gpu, 12 * sizeof(float));
- cudaMalloc((void**)&B_gpu, 12 * sizeof(float));
+ cudaMalloc(reinterpret_cast<void**>(&A_gpu), 12 * sizeof(float));
+ cudaMalloc(reinterpret_cast<void**>(&B_gpu), 12 * sizeof(float));
cudaMemcpy(A_gpu, A, 12 * sizeof(float), cudaMemcpyHostToDevice);
cudaMemcpy(B_gpu, B, 12 * sizeof(float), cudaMemcpyHostToDevice);
@@ -44,11 +65,11 @@ TEST(ContextTest, TestHandle) {
cublasSdot(handle, 12, A_gpu, 1, B_gpu, 1, &gpu_ret);
- for(int i = 0; i < 12;++i) {
+ for (int i = 0; i < 12; ++i) {
cpu_ret += A[i] * B[i];
}
- ASSERT_EQ(gpu_ret,cpu_ret);
+ ASSERT_EQ(gpu_ret, cpu_ret);
cudaFree(A_gpu);
cudaFree(B_gpu);
http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/b2cfa17b/src/test/test_math.cc
----------------------------------------------------------------------
diff --git a/src/test/test_math.cc b/src/test/test_math.cc
index 8f8c633..39ec2a0 100644
--- a/src/test/test_math.cc
+++ b/src/test/test_math.cc
@@ -1,3 +1,24 @@
+/************************************************************
+*
+* Licensed to the Apache Software Foundation (ASF) under one
+* or more contributor license agreements. See the NOTICE file
+* distributed with this work for additional information
+* regarding copyright ownership. The ASF licenses this file
+* to you under the Apache License, Version 2.0 (the
+* "License"); you may not use this file except in compliance
+* with the License. You may obtain a copy of the License at
+*
+* http://www.apache.org/licenses/LICENSE-2.0
+*
+* Unless required by applicable law or agreed to in writing,
+* software distributed under the License is distributed on an
+* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+* KIND, either express or implied. See the License for the
+* specific language governing permissions and limitations
+* under the License.
+*
+*************************************************************/
+
#include "gtest/gtest.h"
#include "singa/utils/math_addr.h"
#include "singa/utils/math_kernel.h"
@@ -12,459 +33,407 @@ using namespace singa;
using namespace std;
TEST(MathTest, TestGemmCPU) {
- float A[3][2] = {};
- float B[3][2] = {};
- float C[2][2] = {};
- for(int i = 0; i < 3; i++)
- for(int j = 0; j < 2; j++)
- {
- A[i][j] = i+j;
- B[i][j] = i+j - i*j;
- }
- cpu_gemm(A[0], B[0], 2, 2, 3 , 1.0f, 0.0f, true, false, C[0]);
- float D[2][2] = {};
- for(int i = 0; i < 2; i++)
- for(int j = 0; j < 2; j++)
- {
- D[i][j] = 0;
- for(int k = 0; k < 3; k++)
- D[i][j] += A[k][i]*B[k][j];
- }
- for(int i = 0; i < 2; i++)
- for(int j = 0; j < 2; j++)
- {
- ASSERT_EQ(C[i][j], D[i][j]);
- }
+ float A[3][2] = {};
+ float B[3][2] = {};
+ float C[2][2] = {};
+ for (int i = 0; i < 3; i++)
+ for (int j = 0; j < 2; j++) {
+ A[i][j] = i+j;
+ B[i][j] = i+j - i*j;
+ }
+ cpu_gemm(A[0], B[0], 2, 2, 3 , 1.0f, 0.0f, true, false, C[0]);
+ float D[2][2] = {};
+ for (int i = 0; i < 2; i++)
+ for (int j = 0; j < 2; j++) {
+ D[i][j] = 0;
+ for (int k = 0; k < 3; k++)
+ D[i][j] += A[k][i]*B[k][j];
+ }
+ for (int i = 0; i < 2; i++)
+ for (int j = 0; j < 2; j++) {
+ ASSERT_EQ(C[i][j], D[i][j]);
+ }
}
TEST(MathTest, TestGemvCPU) {
- float A[4][3] = {};
- float B[4]= {};
- float C[3] = {};
- float D[3] = {};
-
- for(int i = 0; i < 3; i++)
- {
- for(int j = 0; j < 4; j++)
- {
- A[j][i] = i-j + i*j;
- }
- }
-
- for(int i = 0; i < 4; i++)B[i] = i;
- for(int i = 0; i < 3; i++)C[i] = 10;
- cpu_gemv(A[0], B, 4, 3, 1.0f, 1.0f, true, C);
-
- for(int i = 0; i < 3; i++)
- {
- for(int j = 0; j < 4; j++)
- {
- D[i] += A[j][i]*B[j];
- }
- }
- for(int i = 0; i < 3; i++)
- {
- ASSERT_EQ(C[i], D[i]+10);
- }
+ float A[4][3] = {};
+ float B[4]= {};
+ float C[3] = {};
+ float D[3] = {};
+
+ for (int i = 0; i < 3; i++) {
+ for (int j = 0; j < 4; j++) {
+ A[j][i] = i-j + i*j;
+ }
+ }
+
+ for (int i = 0; i < 4; i++)B[i] = i;
+ for (int i = 0; i < 3; i++)C[i] = 10;
+ cpu_gemv(A[0], B, 4, 3, 1.0f, 1.0f, true, C);
+
+ for (int i = 0; i < 3; i++) {
+ for (int j = 0; j < 4; j++) {
+ D[i] += A[j][i]*B[j];
+ }
+ }
+ for (int i = 0; i < 3; i++) {
+ ASSERT_EQ(C[i], D[i]+10);
+ }
}
TEST(MathTest, TestAxpyCPU) {
- float A[4][3] = {};
- float C[4][3] = {};
- float B[3][4] = {};
- float D[3][4] = {};
-
- for(int i = 0; i < 4; i++)
- {
- for(int j = 0; j < 3; j++)
- {
- A[i][j] = i-j + i*j;
- B[j][i] = i-j + i*j;
- C[i][j] = A[i][j];
- D[j][i] = B[j][i];
- }
- }
-
- cpu_axpy(A[0], 12, 2.0f, B[0]);
- for(int i = 0; i < 12; i++)
- {
- D[i / 4][i % 4] += 2*C[i / 3][i % 3];
- }
-
- for(int i = 0; i < 3; i++)
- {
- for(int j = 0; j < 4; j++)
- {
- ASSERT_EQ(B[i][j],D[i][j]);
- }
- }
+ float A[4][3] = {};
+ float C[4][3] = {};
+ float B[3][4] = {};
+ float D[3][4] = {};
+
+ for (int i = 0; i < 4; i++) {
+ for (int j = 0; j < 3; j++) {
+ A[i][j] = i-j + i*j;
+ B[j][i] = i-j + i*j;
+ C[i][j] = A[i][j];
+ D[j][i] = B[j][i];
+ }
+ }
+
+ cpu_axpy(A[0], 12, 2.0f, B[0]);
+ for (int i = 0; i < 12; i++) {
+ D[i / 4][i % 4] += 2*C[i / 3][i % 3];
+ }
+
+ for (int i = 0; i < 3; i++) {
+ for (int j = 0; j < 4; j++) {
+ ASSERT_EQ(B[i][j], D[i][j]);
+ }
+ }
}
/*
TEST(MathTest, TestEopCPU) {
- float A[10] = {};
- float B[10] = {};
- float C[10] = {};
- float O[10] = {};
-
- for(int i = 0; i < 10; i++)
- {
- A[i] = i;
- B[i] = -i;
- C[i] = i;
-
- }
- cpu_e_f<singa::op::Set>(5, 15.0f, O, O);
- for(int i = 0; i < 5; i++)
- {
- ASSERT_EQ(O[i]-15,0);
- }
- for(int i = 5; i < 10; i++)
- {
- ASSERT_EQ(O[i],0);
- }
+ float A[10] = {};
+ float B[10] = {};
+ float C[10] = {};
+ float O[10] = {};
+
+ for (int i = 0; i < 10; i++) {
+ A[i] = i;
+ B[i] = -i;
+ C[i] = i;
+ }
+ cpu_e_f<singa::op::Set>(5, 15.0f, O, O);
+ for (int i = 0; i < 5; i++) {
+ ASSERT_EQ(O[i]-15,0);
+ }
+ for (int i = 5; i < 10; i++) {
+ ASSERT_EQ(O[i],0);
+ }
}
*/
#ifdef USE_GPU
TEST(MathTest, TestGemmGPU) {
- float A[3][2] = {};
- float B[3][2] = {};
- float C[2][2] = {};
- for(int i = 0; i < 3; i++)
- {
- for(int j = 0; j < 2; j++)
- {
- A[i][j] = i+j;
- B[i][j] = i+j - i*j;
- }
- }
-
- float* A_gpu=NULL;
- float* B_gpu=NULL;
- float* C_gpu=NULL;
-
- cudaMalloc((void**)&A_gpu, 3*2*sizeof(float));
- cudaMalloc((void**)&B_gpu, 3*2*sizeof(float));
- cudaMalloc((void**)&C_gpu, 2*2*sizeof(float));
-
- cudaMemcpy(A_gpu,A,3*2*sizeof(float),cudaMemcpyHostToDevice);
- cudaMemcpy(B_gpu,B,3*2*sizeof(float),cudaMemcpyHostToDevice);
-
- gpu_gemm<float>(A_gpu, B_gpu, 2, 2, 3 , 1, 0, true, false, C_gpu);
-
- cudaMemcpy(C,C_gpu,2*2*sizeof(float),cudaMemcpyDeviceToHost);
-
- float D[2][2] = {};
- for(int i = 0; i < 2; i++)
- {
- for(int j = 0; j < 2; j++)
- {
- D[i][j] = 0;
- for(int k = 0; k < 3; k++)
- {
- D[i][j] += A[k][i]*B[k][j];
- }
- }
- }
-
- for(int i = 0; i < 2; i++)
- {
- for(int j = 0; j < 2; j++)
- {
- ASSERT_EQ(C[i][j],D[i][j]);
- }
- }
-
- cudaFree(A_gpu);
- cudaFree(B_gpu);
- cudaFree(C_gpu);
+ float A[3][2] = {};
+ float B[3][2] = {};
+ float C[2][2] = {};
+ for (int i = 0; i < 3; i++) {
+ for (int j = 0; j < 2; j++) {
+ A[i][j] = i+j;
+ B[i][j] = i+j - i*j;
+ }
+ }
+
+ float* A_gpu = NULL;
+ float* B_gpu = NULL;
+ float* C_gpu = NULL;
+
+ cudaMalloc(reinterpret_cast<void**>(&A_gpu), 3*2*sizeof(float));
+ cudaMalloc(reinterpret_cast<void**>(&B_gpu), 3*2*sizeof(float));
+ cudaMalloc(reinterpret_cast<void**>(&C_gpu), 2*2*sizeof(float));
+
+ cudaMemcpy(A_gpu, A, 3*2*sizeof(float), cudaMemcpyHostToDevice);
+ cudaMemcpy(B_gpu, B, 3*2*sizeof(float), cudaMemcpyHostToDevice);
+
+ gpu_gemm<float>(A_gpu, B_gpu, 2, 2, 3 , 1, 0, true, false, C_gpu);
+
+ cudaMemcpy(C, C_gpu, 2*2*sizeof(float), cudaMemcpyDeviceToHost);
+
+ float D[2][2] = {};
+ for (int i = 0; i < 2; i++) {
+ for (int j = 0; j < 2; j++) {
+ D[i][j] = 0;
+ for (int k = 0; k < 3; k++) {
+ D[i][j] += A[k][i]*B[k][j];
+ }
+ }
+ }
+
+ for (int i = 0; i < 2; i++) {
+ for (int j = 0; j < 2; j++) {
+ ASSERT_EQ(C[i][j], D[i][j]);
+ }
+ }
+
+ cudaFree(A_gpu);
+ cudaFree(B_gpu);
+ cudaFree(C_gpu);
}
TEST(MathTest, TestGemvGPU) {
- float A[4][3] = {};
- float B[4]= {};
- float C[3] = {};
- float D[3] = {};
-
- for(int i = 0; i < 4; i++)
- {
- for(int j = 0; j < 3; j++)
- {
- A[i][j] = i-j + i*j;
- }
- }
-
- for(int i = 0; i < 4; i++)B[i] = i;
- for(int i = 0; i < 3; i++)C[i] = 10;
-
- float* A_gpu=NULL;
- float* B_gpu=NULL;
- float* C_gpu=NULL;
-
- cudaMalloc((void**)&A_gpu, 4*3*sizeof(float));
- cudaMalloc((void**)&B_gpu, 4*sizeof(float));
- cudaMalloc((void**)&C_gpu, 3*sizeof(float));
-
- cudaMemcpy(A_gpu,A,4*3*sizeof(float),cudaMemcpyHostToDevice);
- cudaMemcpy(B_gpu,B,4*sizeof(float),cudaMemcpyHostToDevice);
- cudaMemcpy(C_gpu,C,3*sizeof(float),cudaMemcpyHostToDevice);
-
- gpu_gemv<float>(A_gpu, B_gpu, 4, 3, 1, 1, true, C_gpu);
-
- cudaMemcpy(C,C_gpu,3*sizeof(float),cudaMemcpyDeviceToHost);
-
- for(int i = 0; i < 3; i++)
- {
- for(int j = 0; j < 4; j++)
- {
- D[i] += A[j][i]*B[j];
- }
- }
-
- for(int i = 0; i < 3; i++)
- {
- ASSERT_EQ(C[i],D[i]+10);
- }
-
- cudaFree(A_gpu);
- cudaFree(B_gpu);
- cudaFree(C_gpu);
+ float A[4][3] = {};
+ float B[4]= {};
+ float C[3] = {};
+ float D[3] = {};
+
+ for (int i = 0; i < 4; i++) {
+ for (int j = 0; j < 3; j++) {
+ A[i][j] = i-j + i*j;
+ }
+ }
+
+ for (int i = 0; i < 4; i++) B[i] = i;
+ for (int i = 0; i < 3; i++) C[i] = 10;
+
+ float* A_gpu = NULL;
+ float* B_gpu = NULL;
+ float* C_gpu = NULL;
+
+ cudaMalloc(reinterpret_cast<void**>(&A_gpu), 4*3*sizeof(float));
+ cudaMalloc(reinterpret_cast<void**>(&B_gpu), 4*sizeof(float));
+ cudaMalloc(reinterpret_cast<void**>(&C_gpu), 3*sizeof(float));
+
+ cudaMemcpy(A_gpu, A, 4*3*sizeof(float), cudaMemcpyHostToDevice);
+ cudaMemcpy(B_gpu, B, 4*sizeof(float), cudaMemcpyHostToDevice);
+ cudaMemcpy(C_gpu, C, 3*sizeof(float), cudaMemcpyHostToDevice);
+
+ gpu_gemv<float>(A_gpu, B_gpu, 4, 3, 1, 1, true, C_gpu);
+
+ cudaMemcpy(C, C_gpu, 3*sizeof(float), cudaMemcpyDeviceToHost);
+
+ for (int i = 0; i < 3; i++) {
+ for (int j = 0; j < 4; j++) {
+ D[i] += A[j][i]*B[j];
+ }
+ }
+
+ for (int i = 0; i < 3; i++) {
+ ASSERT_EQ(C[i], D[i]+10);
+ }
+
+ cudaFree(A_gpu);
+ cudaFree(B_gpu);
+ cudaFree(C_gpu);
}
/*
TEST(MathTest, TestAxpyGPU) {
- float A[4][3] = {};
- float C[4][3] = {};
- float B[3][4] = {};
- float D[3][4] = {};
-
- for(int i = 0; i < 4; i++)
- {
- for(int j = 0; j < 3; j++)
- {
- A[i][j] = i-j + i*j;
- B[j][i] = i-j + i*j;
- C[i][j] = A[i][j];
- D[j][i] = B[j][i];
- }
- }
-
- float* A_gpu=NULL;
- float* B_gpu=NULL;
-
- cudaMalloc((void**)&A_gpu, 4*3*sizeof(float));
- cudaMalloc((void**)&B_gpu, 3*4*sizeof(float));
-
- cudaMemcpy(A_gpu,A,4*3*sizeof(float),cudaMemcpyHostToDevice);
- cudaMemcpy(B_gpu,B,3*4*sizeof(float),cudaMemcpyHostToDevice);
-
- gpu_axpy<float>(A_gpu, 12, 2, B_gpu);
-
- cudaMemcpy(A,A_gpu,4*3*sizeof(float),cudaMemcpyDeviceToHost);
- cudaMemcpy(B,B_gpu,3*4*sizeof(float),cudaMemcpyDeviceToHost);
-
- //for(int i = 0; i < 12; i++)D[0][i] += 2*C[0][i];
-
- for(int i = 0; i < 4; i++)
- {
- for(int j = 0; j < 3; j++)
- {
- D[i][j] += C[i][j];
- ASSERT_EQ(B[i][j],D[i][j]);
- }
- }
-
- cudaFree(A_gpu);
- cudaFree(B_gpu);
+ float A[4][3] = {};
+ float C[4][3] = {};
+ float B[3][4] = {};
+ float D[3][4] = {};
+
+ for (int i = 0; i < 4; i++)
+ {
+ for (int j = 0; j < 3; j++)
+ {
+ A[i][j] = i-j + i*j;
+ B[j][i] = i-j + i*j;
+ C[i][j] = A[i][j];
+ D[j][i] = B[j][i];
+ }
+ }
+
+ float* A_gpu=NULL;
+ float* B_gpu=NULL;
+
+ cudaMalloc((void**)&A_gpu, 4*3*sizeof(float));
+ cudaMalloc((void**)&B_gpu, 3*4*sizeof(float));
+
+ cudaMemcpy(A_gpu,A,4*3*sizeof(float),cudaMemcpyHostToDevice);
+ cudaMemcpy(B_gpu,B,3*4*sizeof(float),cudaMemcpyHostToDevice);
+
+ gpu_axpy<float>(A_gpu, 12, 2, B_gpu);
+
+ cudaMemcpy(A,A_gpu,4*3*sizeof(float),cudaMemcpyDeviceToHost);
+ cudaMemcpy(B,B_gpu,3*4*sizeof(float),cudaMemcpyDeviceToHost);
+
+ //for (int i = 0; i < 12; i++)D[0][i] += 2*C[0][i];
+
+ for (int i = 0; i < 4; i++)
+ {
+ for (int j = 0; j < 3; j++)
+ {
+ D[i][j] += C[i][j];
+ ASSERT_EQ(B[i][j],D[i][j]);
+ }
+ }
+
+ cudaFree(A_gpu);
+ cudaFree(B_gpu);
}
*/
TEST(MathTest, TestDotGPU) {
- float A[12];
- float B[12];
-
- for(int i = 0; i < 12; i++)
- {
- A[i]=i-1;
- B[i]=i+1;
- }
+ float A[12];
+ float B[12];
- float* A_gpu=NULL;
- float* B_gpu=NULL;
+ for (int i = 0; i < 12; i++) {
+ A[i] = i - 1;
+ B[i] = i + 1;
+ }
- cudaMalloc((void**)&A_gpu, 12*sizeof(float));
- cudaMalloc((void**)&B_gpu, 12*sizeof(float));
+ float* A_gpu = NULL;
+ float* B_gpu = NULL;
- cudaMemcpy(A_gpu,A,12*sizeof(float),cudaMemcpyHostToDevice);
- cudaMemcpy(B_gpu,B,12*sizeof(float),cudaMemcpyHostToDevice);
- float gpu_ret=gpu_dot<float>(A_gpu,B_gpu,12);
+ cudaMalloc(reinterpret_cast<void**>(&A_gpu), 12*sizeof(float));
+ cudaMalloc(reinterpret_cast<void**>(&B_gpu), 12*sizeof(float));
- float cpu_ret=0.0f;
- for(int i = 0; i < 12; i++)
- {
- cpu_ret+=A[i]*B[i];
- }
+ cudaMemcpy(A_gpu, A, 12*sizeof(float), cudaMemcpyHostToDevice);
+ cudaMemcpy(B_gpu, B, 12*sizeof(float), cudaMemcpyHostToDevice);
+ float gpu_ret = gpu_dot<float>(A_gpu, B_gpu, 12);
- ASSERT_EQ(gpu_ret,cpu_ret);
+ float cpu_ret = 0.0f;
+ for (int i = 0; i < 12; i++) {
+ cpu_ret += A[i] * B[i];
+ }
- cudaFree(A_gpu);
- cudaFree(B_gpu);
+ ASSERT_EQ(gpu_ret, cpu_ret);
+ cudaFree(A_gpu);
+ cudaFree(B_gpu);
}
TEST(MathTest, TestSingaSumColGPU) {
+ float A[3][4];
+ float B[4];
+ float C[4];
- float A[3][4];
- float B[4];
- float C[4];
-
- for(int i = 0; i < 3; i++)
- {
- for(int j = 0; j < 4; j++)
- {
- A[i][j]=i+j;
- }
- }
-
- for(int i = 0; i < 4; i++)
- {
- B[i]=0.0f;
- C[i]=0.0f;
- }
-
- float* A_gpu=NULL;
- float* B_gpu=NULL;
-
- cudaMalloc((void**)&A_gpu, 12*sizeof(float));
- cudaMalloc((void**)&B_gpu, 4*sizeof(float));
- cudaMemcpy(A_gpu,A,12*sizeof(float),cudaMemcpyHostToDevice);
-
- singa_gpu_sum_col(A_gpu,B_gpu,3,4,4);
-
- cudaMemcpy(B,B_gpu,4*sizeof(float),cudaMemcpyDeviceToHost);
-
- for(int i = 0; i < 4; i++)
- {
- for(int j = 0; j < 3; j++)
- {
- C[i]+=A[j][i];
- }
- }
-
- for(int i = 0; i <4; i++)
- {
- ASSERT_EQ(B[i],C[i]);
- }
-
- cudaFree(A_gpu);
- cudaFree(B_gpu);
+ for (int i = 0; i < 3; i++) {
+ for (int j = 0; j < 4; j++) {
+ A[i][j] = i + j;
+ }
+ }
+
+ for (int i = 0; i < 4; i++) {
+ B[i] = 0.0f;
+ C[i] = 0.0f;
+ }
+
+ float* A_gpu = NULL;
+ float* B_gpu = NULL;
+
+ cudaMalloc(reinterpret_cast<void**>(&A_gpu), 12*sizeof(float));
+ cudaMalloc(reinterpret_cast<void**>(&B_gpu), 4*sizeof(float));
+ cudaMemcpy(A_gpu, A, 12*sizeof(float), cudaMemcpyHostToDevice);
+
+ singa_gpu_sum_col(A_gpu, B_gpu, 3, 4, 4);
+
+ cudaMemcpy(B, B_gpu, 4*sizeof(float), cudaMemcpyDeviceToHost);
+
+ for (int i = 0; i < 4; i++) {
+ for (int j = 0; j < 3; j++) {
+ C[i] += A[j][i];
+ }
+ }
+
+ for (int i = 0; i < 4; i++) {
+ ASSERT_EQ(B[i], C[i]);
+ }
+
+ cudaFree(A_gpu);
+ cudaFree(B_gpu);
}
TEST(MathTest, TestSingaAddVecRowGPU) {
-
- float A[3][4];
- float B[4];
- float C[3][4];
- float D[3][4];
-
- for(int i = 0; i < 4; i++)
- {
- B[i]=i;
- }
-
- for(int i = 0; i < 3; i++)
- {
- for(int j = 0; j < 4; j++)
- {
- A[i][j]=i+j;
- D[i][j]=A[i][j]+B[j];
- }
- }
-
-
- float* A_gpu=NULL;
- float* B_gpu=NULL;
- float* C_gpu=NULL;
-
- cudaMalloc((void**)&A_gpu, 3*4*sizeof(float));
- cudaMalloc((void**)&B_gpu, 4*sizeof(float));
- cudaMalloc((void**)&C_gpu, 3*4*sizeof(float));
- cudaMemcpy(A_gpu,A,3*4*sizeof(float),cudaMemcpyHostToDevice);
- cudaMemcpy(B_gpu,B,4*sizeof(float),cudaMemcpyHostToDevice);
-
- singa_gpu_add_vec_row(B_gpu,A_gpu,C_gpu,3,4,4);
-
- cudaMemcpy(C,C_gpu,3*4*sizeof(float),cudaMemcpyDeviceToHost);
-
- for(int i = 0; i < 3; i++)
- {
- for(int j = 0; j < 4; j++)
- {
- ASSERT_EQ(C[i][j],D[i][j]);
- }
- }
-
- cudaFree(A_gpu);
- cudaFree(B_gpu);
- cudaFree(C_gpu);
+ float A[3][4];
+ float B[4];
+ float C[3][4];
+ float D[3][4];
+
+ for (int i = 0; i < 4; i++) {
+ B[i] = i;
+ }
+
+ for (int i = 0; i < 3; i++) {
+ for (int j = 0; j < 4; j++) {
+ A[i][j] = i + j;
+ D[i][j] = A[i][j] + B[j];
+ }
+ }
+
+ float* A_gpu = NULL;
+ float* B_gpu = NULL;
+ float* C_gpu = NULL;
+
+ cudaMalloc(reinterpret_cast<void**>(&A_gpu), 3*4*sizeof(float));
+ cudaMalloc(reinterpret_cast<void**>(&B_gpu), 4*sizeof(float));
+ cudaMalloc(reinterpret_cast<void**>(&C_gpu), 3*4*sizeof(float));
+ cudaMemcpy(A_gpu, A, 3*4*sizeof(float), cudaMemcpyHostToDevice);
+ cudaMemcpy(B_gpu, B, 4*sizeof(float), cudaMemcpyHostToDevice);
+
+ singa_gpu_add_vec_row(B_gpu, A_gpu, C_gpu, 3, 4, 4);
+
+ cudaMemcpy(C, C_gpu, 3*4*sizeof(float), cudaMemcpyDeviceToHost);
+
+ for (int i = 0; i < 3; i++) {
+ for (int j = 0; j < 4; j++) {
+ ASSERT_EQ(C[i][j], D[i][j]);
+ }
+ }
+
+ cudaFree(A_gpu);
+ cudaFree(B_gpu);
+ cudaFree(C_gpu);
}
TEST(MathTest, TestSingaSetValueGPU) {
+ float A[3][4];
+ float* A_gpu = NULL;
- float A[3][4];
+ cudaMalloc(reinterpret_cast<void**>(&A_gpu), 3*4*sizeof(float));
- float* A_gpu=NULL;
+ cudaMemcpy(A_gpu, A, 3*4*sizeof(float), cudaMemcpyHostToDevice);
- cudaMalloc((void**)&A_gpu, 3*4*sizeof(float));
+ singa_gpu_set_value(A_gpu, 4.0, 3*4);
- cudaMemcpy(A_gpu,A,3*4*sizeof(float),cudaMemcpyHostToDevice);
+ cudaMemcpy(A, A_gpu, 3*4*sizeof(float), cudaMemcpyDeviceToHost);
- singa_gpu_set_value(A_gpu,4.0,3*4);
+ for (int i = 0; i < 3; i++) {
+ for (int j = 0; j < 4; j++) {
+ ASSERT_EQ(A[i][j], 4.0f);
+ }
+ }
- cudaMemcpy(A,A_gpu,3*4*sizeof(float),cudaMemcpyDeviceToHost);
-
- for(int i = 0; i < 3; i++)
- {
- for(int j = 0; j < 4; j++)
- {
- ASSERT_EQ(A[i][j],4.0f);
- }
- }
-
- cudaFree(A_gpu);
+ cudaFree(A_gpu);
}
TEST(MathTest, TestEopGPU) {
+ float A[10] = {};
+ float B[10] = {};
- float A[10] = {};
- float B[10] = {};
-
- for(int i = 0; i < 10; i++)
- {
- A[i] = i;
- B[i] = -i;
- }
+ for (int i = 0; i < 10; i++) {
+ A[i] = i;
+ B[i] = -i;
+ }
- float* A_gpu=NULL;
- float* B_gpu=NULL;
+ float* A_gpu = NULL;
+ float* B_gpu = NULL;
- cudaMalloc((void**)&A_gpu, 10*sizeof(float));
- cudaMalloc((void**)&B_gpu, 10*sizeof(float));
+ cudaMalloc(reinterpret_cast<void**>(&A_gpu), 10*sizeof(float));
+ cudaMalloc(reinterpret_cast<void**>(&B_gpu), 10*sizeof(float));
- cudaMemcpy(A_gpu,A,10*sizeof(float),cudaMemcpyHostToDevice);
- cudaMemcpy(B_gpu,B,10*sizeof(float),cudaMemcpyHostToDevice);
+ cudaMemcpy(A_gpu, A, 10*sizeof(float), cudaMemcpyHostToDevice);
+ cudaMemcpy(B_gpu, B, 10*sizeof(float), cudaMemcpyHostToDevice);
- gpu_e_f<singa::op::Sigmoid<float>, float>(10, A_gpu, B_gpu);
+ gpu_e_f<singa::op::Sigmoid<float>, float>(10, A_gpu, B_gpu);
- cudaFree(A_gpu);
- cudaFree(B_gpu);
+ cudaFree(A_gpu);
+ cudaFree(B_gpu);
}
#endif // USE_GPU
http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/b2cfa17b/src/test/test_msg.cc
----------------------------------------------------------------------
diff --git a/src/test/test_msg.cc b/src/test/test_msg.cc
index d5d9f20..db83b1c 100644
--- a/src/test/test_msg.cc
+++ b/src/test/test_msg.cc
@@ -42,7 +42,7 @@ TEST(MsgTest, AddrTest) {
}
TEST(MsgTest, AddFrameTest) {
- int buf[5]={1,2,3,4,5};
+ int buf[5] = {1, 2, 3, 4, 5};
Msg msg;
msg.AddFrame("abcdefg", 7);
msg.AddFrame(buf, sizeof(int) * 5);
http://git-wip-us.apache.org/repos/asf/incubator-singa/blob/b2cfa17b/src/test/test_paramslicer.cc
----------------------------------------------------------------------
diff --git a/src/test/test_paramslicer.cc b/src/test/test_paramslicer.cc
index c693da1..bc7dedd 100644
--- a/src/test/test_paramslicer.cc
+++ b/src/test/test_paramslicer.cc
@@ -25,7 +25,7 @@
using namespace singa;
-const int param_size[]={2400,32,25600,32, 51200,64,57600,10};
+const int param_size[] = {2400, 32, 25600, 32, 51200, 64, 57600, 10};
/*
class ParamSlicerTest : public ::testing::Test {