COMPMID-3887: NEGEMMConvolutionLayer hangs up on num_threads>18
When store the Tensor shape information, the size was clamped to 32bit, which cause the memory allocated is not big enough.
Signed-off-by: Sheri Zhang <sheri.zhang@arm.com>
Change-Id: I9f7dfcd5595a143b0ed4f6973e20bcd9d776b673
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4331
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
diff --git a/arm_compute/core/TensorShape.h b/arm_compute/core/TensorShape.h
index 2187743..b455a07 100644
--- a/arm_compute/core/TensorShape.h
+++ b/arm_compute/core/TensorShape.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2016-2019 Arm Limited.
+ * Copyright (c) 2016-2020 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -36,7 +36,7 @@
namespace arm_compute
{
/** Shape of a tensor */
-class TensorShape : public Dimensions<uint32_t>
+class TensorShape : public Dimensions<size_t>
{
public:
/** Constructor to initialize the tensor shape.
diff --git a/arm_compute/core/Window.inl b/arm_compute/core/Window.inl
index 14a432a..6100d09 100644
--- a/arm_compute/core/Window.inl
+++ b/arm_compute/core/Window.inl
@@ -197,15 +197,15 @@
{
if(d == dimension)
{
- int start = _dims[d].start();
- int end = _dims[d].end();
- const int step = _dims[d].step();
+ int start = _dims[d].start();
+ int end = _dims[d].end();
+ const int step = _dims[d].step();
const int num_it = num_iterations(d);
const int rem = num_it % total;
- int work = num_it / total;
+ int work = num_it / total;
- int it_start = work * id;
+ int it_start = work * id;
if(int(id) < rem)
{
@@ -277,7 +277,7 @@
{
for(unsigned int n = first_dimension; n < shape.num_dimensions(); ++n)
{
- set(n, Window::Dimension(0, std::max(shape[n], static_cast<uint32_t>(1))));
+ set(n, Window::Dimension(0, std::max(shape[n], static_cast<size_t>(1))));
}
}
diff --git a/utils/GraphUtils.cpp b/utils/GraphUtils.cpp
index 84f0416..e543cab 100644
--- a/utils/GraphUtils.cpp
+++ b/utils/GraphUtils.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017-2019 Arm Limited.
+ * Copyright (c) 2017-2020 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -272,9 +272,16 @@
{
std::tie(permuted_shape, perm) = compute_permutation_parameters(tensor.info()->tensor_shape(), tensor.info()->data_layout());
}
+
+#ifdef __arm__
ARM_COMPUTE_EXIT_ON_MSG_VAR(image_loader->width() != permuted_shape.x() || image_loader->height() != permuted_shape.y(),
"Failed to load image file: dimensions [%d,%d] not correct, expected [%" PRIu32 ",%" PRIu32 "].",
image_loader->width(), image_loader->height(), permuted_shape.x(), permuted_shape.y());
+#else // __arm__
+ ARM_COMPUTE_EXIT_ON_MSG_VAR(image_loader->width() != permuted_shape.x() || image_loader->height() != permuted_shape.y(),
+ "Failed to load image file: dimensions [%d,%d] not correct, expected [%" PRIu64 ",%" PRIu64 "].",
+ image_loader->width(), image_loader->height(), permuted_shape.x(), permuted_shape.y());
+#endif // __arm__
// Fill the tensor with the PPM content (BGR)
image_loader->fill_planar_tensor(tensor, _bgr);
@@ -348,9 +355,16 @@
std::tie(permuted_shape, perm) = compute_permutation_parameters(tensor.info()->tensor_shape(),
tensor.info()->data_layout());
}
+
+#ifdef __arm__
ARM_COMPUTE_EXIT_ON_MSG_VAR(jpeg.width() != permuted_shape.x() || jpeg.height() != permuted_shape.y(),
"Failed to load image file: dimensions [%d,%d] not correct, expected [%" PRIu32 ",%" PRIu32 "].",
jpeg.width(), jpeg.height(), permuted_shape.x(), permuted_shape.y());
+#else // __arm__
+ ARM_COMPUTE_EXIT_ON_MSG_VAR(jpeg.width() != permuted_shape.x() || jpeg.height() != permuted_shape.y(),
+ "Failed to load image file: dimensions [%d,%d] not correct, expected [%" PRIu64 ",%" PRIu64 "].",
+ jpeg.width(), jpeg.height(), permuted_shape.x(), permuted_shape.y());
+#endif // __arm__
// Fill the tensor with the JPEG content (BGR)
jpeg.fill_planar_tensor(tensor, _bgr);