commit 8bcb65fbf74f35444bc137bd22f9957eb0ec8cad
author Giorgio Arena <giorgio.arena@arm.com> Tue Jan 12 11:29:18 2021 +0000
committer Giorgio Arena <giorgio.arena@arm.com> Wed Jan 13 10:47:20 2021 +0000
tree d594c88de68b910e9c140543df86c0ce219b617d
parent e78ef6fda8f9dafec3a08af917a946ebe18944c5

[Nightly Failure] Fix CLDepthwiseConvolutionLayer 3x3 QASYMM8 on Midgard

- Add checks for pad top/bottom bigger than (kernel size / 2)

Resolves: COMPMID-4088

Signed-off-by: Giorgio Arena <giorgio.arena@arm.com>
Change-Id: Ifc5ea2154847d447bc5643d7607e7256aeddfcbf
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4840
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

src/core/CL/cl_kernels/depthwise_convolution_quantized.cl

diff --git a/src/core/CL/cl_kernels/depthwise_convolution_quantized.cl b/src/core/CL/cl_kernels/depthwise_convolution_quantized.cl
index 95cd44e..d39089b 100644
--- a/src/core/CL/cl_kernels/depthwise_convolution_quantized.cl
+++ b/src/core/CL/cl_kernels/depthwise_convolution_quantized.cl
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -1165,10 +1165,10 @@
     values3 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s3));
 
     // z == 1
-    // z_coord can be only negative for z = 0 so we do not need to clamp it
-    // Moreover z_coord cannot be out-of-bound for z = 1 so we do not need to clamp the offset
     z_coord = z * (int)NUM_PLANES_PROCESSED - (int)CONV_PAD_TOP + 1;
+    z_coord = min((uint)z_coord, (uint)SRC_DIM_2);
     offset  = y_offset + (int4)(z_coord * src_stride_z);
+    offset  = min(offset, (int4)max_offset);
     VEC_TYPE(VEC_SIZE)
     values4 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s0));
     VEC_TYPE(VEC_SIZE)
@@ -1179,10 +1179,10 @@
     values7 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s3));
 
     // z == 2
-    // After z = 1 we can simply add src_stride_z to offset without updating z_coord
-    // However offset can be out-of-bound so we need to check if it is greater than max_offset
-    offset += (int4)src_stride_z;
-    offset = min(offset, (int4)max_offset);
+    z_coord = z * (int)NUM_PLANES_PROCESSED - (int)CONV_PAD_TOP + 2;
+    z_coord = min((uint)z_coord, (uint)SRC_DIM_2);
+    offset  = y_offset + (int4)(z_coord * src_stride_z);
+    offset  = min(offset, (int4)max_offset);
     VEC_TYPE(VEC_SIZE)
     values8 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s0));
     VEC_TYPE(VEC_SIZE)
@@ -1193,10 +1193,10 @@
     values11 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s3));
 
     // z == 3
-    // After z = 1 we can simply add src_stride_z to offset without updating z_coord
-    // However offset can be out-of-bound so we need to check if it is greater than max_offset
-    offset += (int4)(src_stride_z);
-    offset = min(offset, (int4)max_offset);
+    z_coord = z * (int)NUM_PLANES_PROCESSED - (int)CONV_PAD_TOP + 3;
+    z_coord = min((uint)z_coord, (uint)SRC_DIM_2);
+    offset  = y_offset + (int4)(z_coord * src_stride_z);
+    offset  = min(offset, (int4)max_offset);
     VEC_TYPE(VEC_SIZE)
     values12 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(src_addr + offset.s0));
     VEC_TYPE(VEC_SIZE)