aboutsummaryrefslogtreecommitdiff
path: root/src/core/CL/cl_kernels
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/CL/cl_kernels')
-rw-r--r--src/core/CL/cl_kernels/softmax_layer.cl4
-rw-r--r--src/core/CL/cl_kernels/softmax_layer_quantized.cl4
2 files changed, 2 insertions, 6 deletions
diff --git a/src/core/CL/cl_kernels/softmax_layer.cl b/src/core/CL/cl_kernels/softmax_layer.cl
index 01f5de47cf..4d2d89dd73 100644
--- a/src/core/CL/cl_kernels/softmax_layer.cl
+++ b/src/core/CL/cl_kernels/softmax_layer.cl
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -306,7 +306,6 @@ __kernel void softmax_layer_max_shift_exp_sum_parallel(
}
#ifdef NON_MULTIPLE_OF_GRID_SIZE
// How many work-items needed to complete the computation.
- //TODO: Optimize this calculation (avoid %).
int boundary_workitems = (SRC_WIDTH % (GRID_SIZE * VECTOR_SIZE)) / VECTOR_SIZE;
if(lid < boundary_workitems)
{
@@ -417,7 +416,6 @@ __kernel void softmax_layer_max_shift_exp_sum_parallel(
sum1D += data;
}
#ifdef NON_MULTIPLE_OF_GRID_SIZE
- //TODO: Optimize the calculation (avoid %).
boundary_workitems = (SRC_WIDTH % (GRID_SIZE * VECTOR_SIZE)) / VECTOR_SIZE;
if(lid < boundary_workitems)
{
diff --git a/src/core/CL/cl_kernels/softmax_layer_quantized.cl b/src/core/CL/cl_kernels/softmax_layer_quantized.cl
index b7a6e00dfa..4d5006d804 100644
--- a/src/core/CL/cl_kernels/softmax_layer_quantized.cl
+++ b/src/core/CL/cl_kernels/softmax_layer_quantized.cl
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -324,7 +324,6 @@ __kernel void softmax_layer_max_shift_exp_sum_quantized_parallel(
}
#ifdef NON_MULTIPLE_OF_GRID_SIZE
// How many work-items needed to complete the computation.
- //TODO: Optimize this calculation (avoid %).
int boundary_workitems = (SRC_WIDTH % (GRID_SIZE * VECTOR_SIZE)) / VECTOR_SIZE;
if(lid < boundary_workitems)
{
@@ -429,7 +428,6 @@ __kernel void softmax_layer_max_shift_exp_sum_quantized_parallel(
sum1D = sum1D + select(0, data_fp, data_diff >= (VEC_INT)(DIFF_MIN));
}
#ifdef NON_MULTIPLE_OF_GRID_SIZE
- //TODO: Optimize the calculation (avoid %).
boundary_workitems = (SRC_WIDTH % (GRID_SIZE * VECTOR_SIZE)) / VECTOR_SIZE;
if(lid < boundary_workitems)
{