p84250761
/
InfiniTensor
forked from jiuyuan/InfiniTensor
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

support 8D tensor, add test example (#170) 

Co-authored-by: Haojie Wang <haojie0429@gmail.com>
This commit is contained in:
xgqdut2016 2023-10-31 10:47:36 +08:00 committed by GitHub
parent 23b825efc4
commit ec3adf6fa7
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 71 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
include/cuda/cuda_split_concat.h
View File

@ -3,7 +3,7 @@
#include <cstdio> #include <cstdio>
const int BATCH_SIZE = 32; // parallel tensor number. const int BATCH_SIZE = 32; // parallel tensor number.
const int DIM_MAX_SIZE = 4; const int DIM_MAX_SIZE = 8;
// Concat operator acts like element tensors composing to one big tensor,and // Concat operator acts like element tensors composing to one big tensor,and
// split operator acts like one big tensor being composed by element // split operator acts like one big tensor being composed by element

36
test/kernels/cuda/test_cuda_concat.cc
View File

@ -122,4 +122,40 @@ TEST(Concat, Cuda_dim0) {
EXPECT_TRUE(oCpu->equalData(vector<float>{0, 1, 2, 1, 1, 1, 0, 1, 2})); EXPECT_TRUE(oCpu->equalData(vector<float>{0, 1, 2, 1, 1, 1, 0, 1, 2}));
} }
TEST(Concat, CudaHigh) {
Runtime runtime = NativeCpuRuntimeObj::getInstance();
Graph gCpu = make_ref<GraphObj>(runtime);
auto t1 = gCpu->addTensor({2, 2, 3, 1, 2}, DataType::Float32);
auto t2 = gCpu->addTensor({2, 2, 1, 1, 2}, DataType::Float32);
auto t3 = gCpu->addTensor({2, 2, 2, 1, 2}, DataType::Float32);
gCpu->dataMalloc();
t1->setData(IncrementalGenerator());
t2->setData(OneGenerator());
t3->setData(OneGenerator());
auto cudaRuntime = make_ref<CudaRuntimeObj>();
Graph gCuda = make_ref<GraphObj>(cudaRuntime);
auto t1Gpu = gCuda->cloneTensor(t1);
auto t2Gpu = gCuda->cloneTensor(t2);
auto t3Gpu = gCuda->cloneTensor(t3);
auto op =
gCuda->addOp<ConcatObj>(TensorVec{t1Gpu, t2Gpu, t3Gpu}, nullptr, 2);
gCuda->dataMalloc();
t1Gpu->setData(IncrementalGenerator());
t2Gpu->setData(OneGenerator());
t3Gpu->setData(OneGenerator());
cudaRuntime->run(gCuda);
// cudaPrintTensor(op->getOutput());
// copy output from CUDA to CPU
auto oCpu = gCpu->cloneTensor(op->getOutput());
EXPECT_TRUE(oCpu->equalData(
vector<float>{0., 1., 2., 3., 4., 5., 1., 1., 1., 1., 1., 1.,
6., 7., 8., 9., 10., 11., 1., 1., 1., 1., 1., 1.,
12., 13., 14., 15., 16., 17., 1., 1., 1., 1., 1., 1.,
18., 19., 20., 21., 22., 23., 1., 1., 1., 1., 1., 1.}));
}
} // namespace infini } // namespace infini

34
test/kernels/cuda/test_cuda_split.cc
View File

@ -39,6 +39,40 @@ TEST(Split, Cuda) {
12, 13, 14, 15, 16, 17, 18, 19, 32, 33, 34, 35, 36, 37, 38, 39})); 12, 13, 14, 15, 16, 17, 18, 19, 32, 33, 34, 35, 36, 37, 38, 39}));
} }
TEST(Split, CudaHigh) {
Runtime runtime = NativeCpuRuntimeObj::getInstance();
Graph gCpu = make_ref<GraphObj>(runtime);
auto input = gCpu->addTensor({2, 6, 2, 1, 2}, DataType::Float32);
gCpu->dataMalloc();
input->setData(IncrementalGenerator());
auto cudaRuntime = make_ref<CudaRuntimeObj>();
Graph gCuda = make_ref<GraphObj>(cudaRuntime);
auto inputGpu = gCuda->cloneTensor(input);
auto op = gCuda->addOp<SplitObj>(inputGpu, std::nullopt, 1, 3);
gCuda->dataMalloc();
inputGpu->setData(IncrementalGenerator());
cudaRuntime->run(gCuda);
// copy output from CUDA to CPU
EXPECT_EQ(op->getOutputs().size(), (size_t)3);
auto o0Cpu = gCpu->cloneTensor(op->getOutput(0));
auto o1Cpu = gCpu->cloneTensor(op->getOutput(1));
auto o2Cpu = gCpu->cloneTensor(op->getOutput(2));
EXPECT_TRUE(
o0Cpu->equalData(vector<float>{0., 1., 2., 3., 4., 5., 6., 7., 24., 25.,
26., 27., 28., 29., 30., 31.}));
EXPECT_TRUE(o1Cpu->equalData(vector<float>{8., 9., 10., 11., 12., 13., 14.,
15., 32., 33., 34., 35., 36.,
37., 38., 39.}));
EXPECT_TRUE(o2Cpu->equalData(vector<float>{16., 17., 18., 19., 20., 21.,
22., 23., 40., 41., 42., 43.,
44., 45., 46., 47.}));
}
TEST(Split, Cuda_dim0) { TEST(Split, Cuda_dim0) {
Runtime runtime = NativeCpuRuntimeObj::getInstance(); Runtime runtime = NativeCpuRuntimeObj::getInstance();
Graph gCpu = make_ref<GraphObj>(runtime); Graph gCpu = make_ref<GraphObj>(runtime);