aboutsummaryrefslogtreecommitdiffhomepage
path: root/tensorflow/contrib/lite/kernels
diff options
context:
space:
mode:
authorGravatar Suharsh Sivakumar <suharshs@google.com>2018-09-17 20:25:23 -0700
committerGravatar TensorFlower Gardener <gardener@tensorflow.org>2018-09-17 20:28:57 -0700
commit0b7125d3c5e7128470a7a74cf8a3543eab39c2d8 (patch)
tree8a6e117399a7b7125fb0e43821452259642a5134 /tensorflow/contrib/lite/kernels
parentbb9958ab69a38cbe57d119947b635a257fa6b802 (diff)
Fix unused variable error on powerpc.
PiperOrigin-RevId: 213386145
Diffstat (limited to 'tensorflow/contrib/lite/kernels')
-rw-r--r--tensorflow/contrib/lite/kernels/internal/optimized/depthwiseconv_float.h6
-rw-r--r--tensorflow/contrib/lite/kernels/internal/optimized/depthwiseconv_uint8.h14
-rw-r--r--tensorflow/contrib/lite/kernels/internal/optimized/depthwiseconv_uint8_3x3_filter.h8
3 files changed, 13 insertions, 15 deletions
diff --git a/tensorflow/contrib/lite/kernels/internal/optimized/depthwiseconv_float.h b/tensorflow/contrib/lite/kernels/internal/optimized/depthwiseconv_float.h
index f0bea7fa1d..114575a96a 100644
--- a/tensorflow/contrib/lite/kernels/internal/optimized/depthwiseconv_float.h
+++ b/tensorflow/contrib/lite/kernels/internal/optimized/depthwiseconv_float.h
@@ -932,9 +932,6 @@ inline void DepthwiseConv(
TFLITE_DCHECK_EQ(filter_shape.DimensionsCount(), 4);
TFLITE_DCHECK_EQ(output_shape.DimensionsCount(), 4);
- const bool has_dilation = (params.dilation_width_factor != 1) ||
- (params.dilation_height_factor != 1);
-
const int batches = MatchingDim(input_shape, 0, output_shape, 0);
const int output_depth = MatchingDim(filter_shape, 3, output_shape, 3);
const int input_height = input_shape.Dims(1);
@@ -966,7 +963,8 @@ inline void DepthwiseConv(
FIXED_DEPTH_MULTIPLIER) \
if (!row_accum_func && (stride_width == 1 || ALLOW_STRIDED) && \
(input_depth == FIXED_INPUT_DEPTH || FIXED_INPUT_DEPTH == 0) && \
- depth_multiplier == FIXED_DEPTH_MULTIPLIER && !has_dilation) { \
+ depth_multiplier == FIXED_DEPTH_MULTIPLIER && \
+ dilation_height_factor == 1 && dilation_width_factor == 1) { \
row_accum_func = \
FloatDepthwiseConvAccumRow<ALLOW_STRIDED, FIXED_INPUT_DEPTH, \
FIXED_DEPTH_MULTIPLIER>; \
diff --git a/tensorflow/contrib/lite/kernels/internal/optimized/depthwiseconv_uint8.h b/tensorflow/contrib/lite/kernels/internal/optimized/depthwiseconv_uint8.h
index 494cf70504..ee3fe78a10 100644
--- a/tensorflow/contrib/lite/kernels/internal/optimized/depthwiseconv_uint8.h
+++ b/tensorflow/contrib/lite/kernels/internal/optimized/depthwiseconv_uint8.h
@@ -1718,18 +1718,15 @@ inline void DepthwiseConv(
TFLITE_DCHECK_EQ(output_depth, input_depth * depth_multiplier);
TFLITE_DCHECK_EQ(bias_shape.FlatSize(), output_depth);
- const bool has_dilation =
- (dilation_width_factor != 1) || (dilation_height_factor != 1);
-
// Enable for arm64 except for the Nvidia Linux 4 Tegra (L4T) running on
// Jetson TX-2. This compiler does not support the offsetof() macro.
#if defined(__aarch64__) && !defined(GOOGLE_L4T)
// Call kernel optimized for depthwise convolutions using 3x3 filters if
// parameters are supported.
- if (Fast3x3FilterKernelSupported(input_shape, filter_shape, stride_width,
- stride_height, has_dilation, pad_width,
- pad_height, depth_multiplier, output_shape,
- output_shift)) {
+ if (Fast3x3FilterKernelSupported(
+ input_shape, filter_shape, stride_width, stride_height,
+ dilation_width_factor, dilation_height_factor, pad_width, pad_height,
+ depth_multiplier, output_shape, output_shift)) {
DepthwiseConv3x3Filter(params, input_shape, input_data, filter_shape,
filter_data, bias_shape, bias_data, output_shape,
output_data);
@@ -1756,7 +1753,8 @@ inline void DepthwiseConv(
FIXED_DEPTH_MULTIPLIER) \
if (!row_accum_func && (stride_width == 1 || ALLOW_STRIDED) && \
(input_depth == FIXED_INPUT_DEPTH || FIXED_INPUT_DEPTH == 0) && \
- depth_multiplier == FIXED_DEPTH_MULTIPLIER && !has_dilation) { \
+ depth_multiplier == FIXED_DEPTH_MULTIPLIER && \
+ dilation_width_factor == 1 && dilation_height_factor == 1) { \
row_accum_func = \
QuantizedDepthwiseConvAccumRow<ALLOW_STRIDED, FIXED_INPUT_DEPTH, \
FIXED_DEPTH_MULTIPLIER>; \
diff --git a/tensorflow/contrib/lite/kernels/internal/optimized/depthwiseconv_uint8_3x3_filter.h b/tensorflow/contrib/lite/kernels/internal/optimized/depthwiseconv_uint8_3x3_filter.h
index 5087227182..e14d04ad02 100644
--- a/tensorflow/contrib/lite/kernels/internal/optimized/depthwiseconv_uint8_3x3_filter.h
+++ b/tensorflow/contrib/lite/kernels/internal/optimized/depthwiseconv_uint8_3x3_filter.h
@@ -3176,8 +3176,9 @@ inline void DepthwiseConvHandlePadding(const uint8* input_data,
inline bool Fast3x3FilterKernelSupported(
const RuntimeShape& input_shape, const RuntimeShape& filter_shape,
- int32 stride_width, int32 stride_height, bool has_dilation, int32 pad_width,
- int32 pad_height, int32 depth_multiplier, const RuntimeShape& output_shape,
+ int32 stride_width, int32 stride_height, int32 dilation_width_factor,
+ int32 dilation_height_factor, int32 pad_width, int32 pad_height,
+ int32 depth_multiplier, const RuntimeShape& output_shape,
int32 output_shift) {
const int32 input_height = input_shape.Dims(1);
const int32 input_width = input_shape.Dims(2);
@@ -3193,7 +3194,8 @@ inline bool Fast3x3FilterKernelSupported(
(stride_height == 1 || stride_height == 2) &&
(stride_width == stride_height) && (pad_width == 0 || pad_width == 1) &&
(pad_height == 0 || pad_height == 1) && (pad_width == pad_height) &&
- (input_depth % 8) == 0 && (output_shift > 0) && !has_dilation;
+ (input_depth % 8) == 0 && (output_shift > 0) &&
+ dilation_width_factor == 1 && dilation_height_factor == 1;
if (!supported) {
return false;