experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

TFLite GPU: Rename SOFT_MAX and SoftMax to SOFTMAX and Softmax, respectively.

Browse Source 
PiperOrigin-RevId: 260970382
This commit is contained in:
Juhyun Lee 2019-07-31 11:51:21 -07:00 committed by TensorFlower Gardener
parent 49002f2e95
commit 39db34c731
8 changed files with 57 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
tensorflow/lite/delegates/gpu/common/model_builder.cc
View File

@ -1737,7 +1737,7 @@ class SoftmaxOperationParser : public TFLiteOperationParser {
const TfLiteRegistration* registration, GraphFloat32* graph,
ObjectReader* reader) final {
Node* node = graph->NewNode();
node->operation.type = ToString(OperationType::SOFT_MAX);
node->operation.type = ToString(OperationType::SOFTMAX);
RETURN_IF_ERROR(reader->AddInput(node, 0));
RETURN_IF_ERROR(reader->AddOutputs(node));
@ -1753,8 +1753,7 @@ class SoftmaxOperationParser : public TFLiteOperationParser {
// auto mul_node = reader->NewPassthroughNode(node);
// mul_node->operation.type = ToString(OperationType::MUL);
}
// TODO(impjdi): Rename to SoftmaxAttributes.
SoftMaxAttributes attr;
SoftmaxAttributes attr;
attr.axis = Axis::CHANNELS; // always by channels
node->operation.attributes = attr;
return OkStatus();

6
tensorflow/lite/delegates/gpu/common/operations.cc
View File

@ -106,8 +106,8 @@ std::string ToString(enum OperationType op) {
return "sin";
case OperationType::SLICE:
return "slice";
case OperationType::SOFT_MAX:
return "soft_max";
case OperationType::SOFTMAX:
return "softmax";
case OperationType::SPACE_TO_BATCH:
return "space_to_batch";
case OperationType::SQRT:
@ -158,7 +158,7 @@ OperationType OperationTypeFromString(const std::string& name) {
{"sigmoid", OperationType::SIGMOID},
{"sin", OperationType::SIN},
{"slice", OperationType::SLICE},
{"soft_max", OperationType::SOFT_MAX},
{"softmax", OperationType::SOFTMAX},
{"sqrt", OperationType::SQRT},
{"square", OperationType::SQUARE},
{"subtract", OperationType::SUB},

4
tensorflow/lite/delegates/gpu/common/operations.h
View File

@ -63,7 +63,7 @@ enum class OperationType {
SIGMOID,
SIN,
SLICE,
SOFT_MAX,
SOFTMAX,
SPACE_TO_BATCH,
SQRT,
SQUARE,
@ -239,7 +239,7 @@ struct PReLUAttributes {
alpha;
};
struct SoftMaxAttributes {
struct SoftmaxAttributes {
Axis axis = Axis::UNKNOWN;
};

2
tensorflow/lite/delegates/gpu/gl/kernels/registry.cc
View File

@ -86,7 +86,7 @@ class Registry : public NodeShader {
insert_op(Type::RELU, NewReLUNodeShader);
insert_op(Type::RESHAPE, NewReshapeNodeShader);
insert_op(Type::SLICE, NewSliceNodeShader);
insert_op(Type::SOFT_MAX, NewSoftMaxNodeShader);
insert_op(Type::SOFTMAX, NewSoftmaxNodeShader);
insert_op(Type::UPSAMPLE_2D, NewUpsamplingNodeShader);
insert_elementwise_op(Type::ABS);

14
tensorflow/lite/delegates/gpu/gl/kernels/softmax.cc
View File

@ -33,14 +33,14 @@ namespace gpu {
namespace gl {
namespace {
class SoftMax : public NodeShader {
class Softmax : public NodeShader {
public:
Status GenerateCode(const GenerationContext& ctx,
GeneratedCode* generated_code) const final {
auto input = ctx.graph->FindInputs(ctx.node->id)[0];
auto output = ctx.graph->FindOutputs(ctx.node->id)[0];
auto attr =
absl::any_cast<SoftMaxAttributes>(ctx.node->operation.attributes);
const auto* input = ctx.graph->FindInputs(ctx.node->id)[0];
const auto* output = ctx.graph->FindOutputs(ctx.node->id)[0];
const auto& attr = absl::any_cast<const SoftmaxAttributes&>(
ctx.node->operation.attributes);
if (input->tensor.shape != output->tensor.shape) {
return InvalidArgumentError("Input and output shape does not match");
}
@ -89,8 +89,8 @@ class SoftMax : public NodeShader {
} // namespace
std::unique_ptr<NodeShader> NewSoftMaxNodeShader() {
return absl::make_unique<SoftMax>();
std::unique_ptr<NodeShader> NewSoftmaxNodeShader() {
return absl::make_unique<Softmax>();
}
} // namespace gl

2
tensorflow/lite/delegates/gpu/gl/kernels/softmax.h
View File

@ -25,7 +25,7 @@ namespace tflite {
namespace gpu {
namespace gl {
std::unique_ptr<NodeShader> NewSoftMaxNodeShader();
std::unique_ptr<NodeShader> NewSoftmaxNodeShader();
} // namespace gl
} // namespace gpu

55
tensorflow/lite/delegates/gpu/gl/kernels/softmax_test.cc
View File

@ -31,7 +31,7 @@ namespace gpu {
namespace gl {
namespace {
TEST(SoftmaxTest, WorksForChannelsAxis) {
TEST(SoftmaxTest, Softmax) {
TensorRef<BHWC> input;
input.type = DataType::FLOAT32;
input.ref = 0;
@ -42,13 +42,13 @@ TEST(SoftmaxTest, WorksForChannelsAxis) {
output.ref = 1;
output.shape = BHWC(1, 2, 2, 1);
SoftMaxAttributes attr;
SoftmaxAttributes attr;
attr.axis = Axis::CHANNELS;
SingleOpModel model({ToString(OperationType::SOFT_MAX), attr}, {input},
SingleOpModel model({ToString(OperationType::SOFTMAX), attr}, {input},
{output});
ASSERT_TRUE(model.PopulateTensor(0, {0.1, 0.2, 0.1, 0.2}));
ASSERT_OK(model.Invoke(*NewSoftMaxNodeShader()));
ASSERT_OK(model.Invoke(*NewSoftmaxNodeShader()));
EXPECT_THAT(model.GetOutput(0), Pointwise(FloatNear(1e-6), {1, 1, 1, 1}));
}
@ -63,15 +63,13 @@ TEST(SoftmaxTest, DoesNotWorkForHeightAxis) {
output.ref = 1;
output.shape = BHWC(1, 2, 2, 1);
SoftMaxAttributes attr;
SoftmaxAttributes attr;
attr.axis = Axis::HEIGHT;
SingleOpModel model({ToString(OperationType::SOFT_MAX), attr}, {input},
SingleOpModel model({ToString(OperationType::SOFTMAX), attr}, {input},
{output});
ASSERT_TRUE(model.PopulateTensor(0, {1, 2, 3, 4}));
ASSERT_THAT(
model.Invoke(*NewSoftMaxNodeShader()).message(),
testing::HasSubstr("Softmax is only supported for channels axis."));
ASSERT_TRUE(model.PopulateTensor(0, {0.1, 0.2, 0.3, 0.4}));
EXPECT_FALSE(model.Invoke(*NewSoftmaxNodeShader()).ok());
}
TEST(SoftmaxTest, DoesNotWorkForWidthAxis) {
@ -85,15 +83,40 @@ TEST(SoftmaxTest, DoesNotWorkForWidthAxis) {
output.ref = 1;
output.shape = BHWC(1, 2, 2, 1);
SoftMaxAttributes attr;
SoftmaxAttributes attr;
attr.axis = Axis::WIDTH;
SingleOpModel model({ToString(OperationType::SOFT_MAX), attr}, {input},
SingleOpModel model({ToString(OperationType::SOFTMAX), attr}, {input},
{output});
ASSERT_TRUE(model.PopulateTensor(0, {1, 2, 3, 4}));
ASSERT_THAT(
model.Invoke(*NewSoftMaxNodeShader()).message(),
testing::HasSubstr("Softmax is only supported for channels axis."));
ASSERT_TRUE(model.PopulateTensor(0, {0.1, 0.2, 0.3, 0.4}));
EXPECT_FALSE(model.Invoke(*NewSoftmaxNodeShader()).ok());
}
TEST(SoftmaxTest, Softmax1x1) {
TensorRef<BHWC> input;
input.type = DataType::FLOAT32;
input.ref = 0;
input.shape = BHWC(1, 1, 1, 4);
TensorRef<BHWC> output;
output.type = DataType::FLOAT32;
output.ref = 1;
output.shape = BHWC(1, 1, 1, 4);
SoftmaxAttributes attr;
attr.axis = Axis::CHANNELS;
const double sum =
std::exp(0.1) + std::exp(0.2) + std::exp(0.3) + std::exp(0.4);
SingleOpModel model({ToString(OperationType::SOFTMAX), attr}, {input},
{output});
ASSERT_TRUE(model.PopulateTensor(0, {0.1, 0.2, 0.3, 0.4}));
ASSERT_OK(model.Invoke(*NewSoftmaxNodeShader()));
EXPECT_THAT(
model.GetOutput(0),
Pointwise(FloatNear(1e-6), {std::exp(0.1) / sum, std::exp(0.2) / sum,
std::exp(0.3) / sum, std::exp(0.4) / sum}));
}
} // namespace

4
tensorflow/lite/delegates/gpu/metal/api.cc
View File

@ -220,9 +220,9 @@ Status Compile(const GraphFloat32& graph, const RuntimeOptions& options,
Slice(node_id, inputs[0], outputs[0],
absl::any_cast<SliceAttributes>(node->operation.attributes));
break;
case OperationType::SOFT_MAX: {
case OperationType::SOFTMAX: {
auto attr =
absl::any_cast<SoftMaxAttributes>(node->operation.attributes);
absl::any_cast<SoftmaxAttributes>(node->operation.attributes);
if (attr.axis != Axis::CHANNELS) {
return UnimplementedError("Softmax supports only CHANNELS dimension");
}