Skip to content

Commit

Permalink
Merge pull request #2776 from NHZlX/mobilenet_gpu
Browse files Browse the repository at this point in the history 
Mobilenet gpu implementation
  • Loading branch information
@NHZlX
NHZlX committed Jul 21, 2017
2 parents e2880f1 + 6c528cb commit 91d2a57
Show file tree
Hide file tree
Showing 7 changed files with 993 additions and 22 deletions.
134 changes: 115 additions & 19 deletions paddle/function/ConvOpTest.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -31,13 +31,22 @@ class ConvolutionTest {
ConvolutionTest(const std::string& conv1,
const std::string& conv2,
TestType type,
bool useGroups = true,
std::string algo = "auto") {
for (size_t batchSize : {1, 32}) {
for (size_t inputSize : {7, 14, 54}) {
for (size_t filterSize : {1, 3, 5}) {
for (size_t inputChannels : {3, 64}) {
for (size_t outputChannels : {3, 64, 128}) {
if (inputChannels < outputChannels) break;
for (size_t outputChannels : {3, 64}) {
if (inputChannels > outputChannels) break;
size_t groups;
if (!useGroups) {
groups = 1;
} else {
if (outputChannels % inputChannels != 0) continue;
groups = inputChannels;
}

for (size_t stride : {1, 2}) {
for (size_t padding : {0, 1}) {
if (padding >= filterSize) break;
Expand All @@ -62,13 +71,24 @@ class ConvolutionTest {
FuncConfig()
.set("paddings", paddings)
.set("strides", strides)
.set("groups", (size_t)1)
.set("groups", groups)
.set("algo", algo));

TensorShape input{
batchSize, inputChannels, inputSize, inputSize};
TensorShape filter{
outputChannels, inputChannels, filterSize, filterSize};

TensorShape filter;
if (groups > 1)
filter = TensorShape({groups,
outputChannels / groups,
inputChannels / groups,
filterSize,
filterSize});
else
filter = TensorShape({outputChannels,
inputChannels,
filterSize,
filterSize});
TensorShape output{
batchSize, outputChannels, outputSize, outputSize};

Expand All @@ -85,7 +105,8 @@ class ConvolutionTest {
} else if (type == kBackwardFilterTest) {
test.addInputs(BufferArg(VALUE_TYPE_FLOAT, output));
test.addInputs(BufferArg(VALUE_TYPE_FLOAT, input));
test.addOutputs(BufferArg(VALUE_TYPE_FLOAT, filter));
test.addOutputs(BufferArg(VALUE_TYPE_FLOAT, filter),
ADD_TO);
test.run();
}
}
Expand All @@ -106,14 +127,23 @@ class ConvolutionTest2 {
ConvolutionTest2(const std::string& conv1,
const std::string& conv2,
TestType type,
bool useGroups = true,
std::string algo = "auto") {
for (size_t batchSize : {16}) {
for (size_t inputHeight : {7, 31}) {
for (size_t inputWidth : {10, 54}) {
for (size_t filterHeight : {1, 5}) {
for (size_t filterWidth : {3, 7}) {
for (size_t inputChannels : {7}) {
for (size_t outputChannels : {32}) {
for (size_t outputChannels : {7}) {
size_t groups;
if (!useGroups) {
groups = 1;
} else {
if (outputChannels % inputChannels != 0) continue;
groups = inputChannels;
}

size_t stride = 1;
size_t padding = 0;
size_t outputHeight =
Expand Down Expand Up @@ -141,13 +171,24 @@ class ConvolutionTest2 {
FuncConfig()
.set("paddings", paddings)
.set("strides", strides)
.set("groups", (size_t)1)
.set("groups", groups)
.set("algo", algo));

TensorShape input{
batchSize, inputChannels, inputHeight, inputWidth};
TensorShape filter{
outputChannels, inputChannels, filterHeight, filterWidth};

TensorShape filter;
if (groups > 1)
filter = TensorShape({groups,
outputChannels / groups,
inputChannels / groups,
filterHeight,
filterWidth});
else
filter = TensorShape({outputChannels,
inputChannels,
filterHeight,
filterWidth});
TensorShape output{
batchSize, outputChannels, outputHeight, outputWidth};

Expand All @@ -164,7 +205,8 @@ class ConvolutionTest2 {
} else if (type == kBackwardFilterTest) {
test.addInputs(BufferArg(VALUE_TYPE_FLOAT, output));
test.addInputs(BufferArg(VALUE_TYPE_FLOAT, input));
test.addOutputs(BufferArg(VALUE_TYPE_FLOAT, filter));
test.addOutputs(BufferArg(VALUE_TYPE_FLOAT, filter),
ADD_TO);
test.run();
}
}
Expand All @@ -177,34 +219,88 @@ class ConvolutionTest2 {
}
};

// ======Start Convolution TEST======

TEST(Forward, GEMM) {
ConvolutionTest<DEVICE_TYPE_CPU, DEVICE_TYPE_CPU> test(
"NaiveConv-CPU", "GemmConv-CPU", kForwardTest);
"NaiveConv-CPU", "GemmConv-CPU", kForwardTest, false);
ConvolutionTest2<DEVICE_TYPE_CPU, DEVICE_TYPE_CPU> test2(
"NaiveConv-CPU", "GemmConv-CPU", kForwardTest);
"NaiveConv-CPU", "GemmConv-CPU", kForwardTest, false);
}

#ifndef PADDLE_ONLY_CPU
TEST(Forward, GEMM2) {
ConvolutionTest<DEVICE_TYPE_CPU, DEVICE_TYPE_GPU> test(
"GemmConv-CPU", "GemmConv-GPU", kForwardTest);
"GemmConv-CPU", "GemmConv-GPU", kForwardTest, false);
ConvolutionTest2<DEVICE_TYPE_CPU, DEVICE_TYPE_GPU> test2(
"GemmConv-CPU", "GemmConv-GPU", kForwardTest);
"GemmConv-CPU", "GemmConv-GPU", kForwardTest, false);
}

TEST(BackwardInput, GEMM) {
ConvolutionTest<DEVICE_TYPE_CPU, DEVICE_TYPE_GPU> test(
"GemmConvGradInput-CPU", "GemmConvGradInput-GPU", kBackwardInputTest);
"GemmConvGradInput-CPU",
"GemmConvGradInput-GPU",
kBackwardInputTest,
false);
ConvolutionTest2<DEVICE_TYPE_CPU, DEVICE_TYPE_GPU> test2(
"GemmConvGradInput-CPU", "GemmConvGradInput-GPU", kBackwardInputTest);
"GemmConvGradInput-CPU",
"GemmConvGradInput-GPU",
kBackwardInputTest,
false);
}

TEST(BackwardFilter, GEMM) {
ConvolutionTest<DEVICE_TYPE_CPU, DEVICE_TYPE_GPU> test(
"GemmConvGradFilter-CPU", "GemmConvGradFilter-GPU", kBackwardFilterTest);
"GemmConvGradFilter-CPU",
"GemmConvGradFilter-GPU",
kBackwardFilterTest,
false);
ConvolutionTest2<DEVICE_TYPE_CPU, DEVICE_TYPE_GPU> test2(
"GemmConvGradFilter-CPU", "GemmConvGradFilter-GPU", kBackwardFilterTest);
"GemmConvGradFilter-CPU",
"GemmConvGradFilter-GPU",
kBackwardFilterTest,
false);
}
#endif
// ======End Convolution TEST======

// ======Start DepthwiseConvolution TEST======

// TODO(zhaolong) The depthwise convolution cpu test will be added when the cpu
// version of depthwiseConv is implemented.

#ifndef PADDLE_ONLY_CPU

TEST(DepthwiseConvForward, GEMM2) {
ConvolutionTest<DEVICE_TYPE_CPU, DEVICE_TYPE_GPU> test(
"GemmConv-CPU", "DepthwiseConv-GPU", kForwardTest);
ConvolutionTest2<DEVICE_TYPE_CPU, DEVICE_TYPE_GPU> test2(
"GemmConv-CPU", "DepthwiseConv-GPU", kForwardTest);
}

TEST(DepthwiseConvBackwardInput, GEMM) {
ConvolutionTest<DEVICE_TYPE_CPU, DEVICE_TYPE_GPU> test(
"GemmConvGradInput-CPU",
"DepthwiseConvGradInput-GPU",
kBackwardInputTest);
ConvolutionTest2<DEVICE_TYPE_CPU, DEVICE_TYPE_GPU> test2(
"GemmConvGradInput-CPU",
"DepthwiseConvGradInput-GPU",
kBackwardInputTest);
}

TEST(DepthwiseConvBackwardFilter, GEMM) {
ConvolutionTest<DEVICE_TYPE_CPU, DEVICE_TYPE_GPU> test(
"GemmConvGradFilter-CPU",
"DepthwiseConvGradFilter-GPU",
kBackwardFilterTest);
ConvolutionTest2<DEVICE_TYPE_CPU, DEVICE_TYPE_GPU> test2(
"GemmConvGradFilter-CPU",
"DepthwiseConvGradFilter-GPU",
kBackwardFilterTest);
}

#endif
// ======End DepthwiseConvolution TEST======

} // namespace paddle
Loading

0 comments on commit 91d2a57

Please sign in to comment.