APPBROWSER-391: Fix GLES COMPUTE alignment issues

APPBROWSER-402: Performance optimization for squeezenet/xray model

Change-Id: If31b186b99a6d6087164019fe94d3ac9279e3204
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/119526
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

3 files changed, 35 insertions, 35 deletions

diff --git a/src/core/GLES_COMPUTE/cs_shaders/arithmetic_add.cs b/src/core/GLES_COMPUTE/cs_shaders/arithmetic_add.cs
index 0ff43605ba..2ab6d5eac5 100755
--- a/src/core/GLES_COMPUTE/cs_shaders/arithmetic_add.cs
+++ b/src/core/GLES_COMPUTE/cs_shaders/arithmetic_add.cs
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016, 2017 ARM Limited.
+ * Copyright (c) 2016-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -29,20 +29,20 @@ layout(local_size_x = LOCAL_SIZE_X, local_size_y = LOCAL_SIZE_Y, local_size_z =
 precision mediump float;
 #define ADD(x, y) (x) + (y)
 
-/** This function add two images.
+/** This function add two tensors.
  *
- * @param[in]  src1_ptr   Pointer to the first source image. Supported data types: F16
- * @param[in]  src1_attrs The attributes of the first source image
- * @param[in]  src2_ptr   Pointer to the second source image. Supported data types: Same as @p src1_ptr
- * @param[in]  src2_attrs The attributes of the second source image
- * @param[out] dst_ptr    Pointer to the destination image. Supported data types: Same as @p src1_ptr
- * @param[in]  dst_attrs  The attributes of the destination image
+ * @param[in]  src1_ptr   Pointer to the first source tensor. Supported data types: F16
+ * @param[in]  src1_attrs The attributes of the first source tensor
+ * @param[in]  src2_ptr   Pointer to the second source tensor. Supported data types: Same as @p src1_ptr
+ * @param[in]  src2_attrs The attributes of the second source tensor
+ * @param[out] dst_ptr    Pointer to the destination tensor. Supported data types: Same as @p src1_ptr
+ * @param[in]  dst_attrs  The attributes of the destination tensor
  */
 SHADER_PARAMS_DECLARATION
 {
-    ImageAttributes src1_attrs;
-    ImageAttributes src2_attrs;
-    ImageAttributes dst_attrs;
+    Tensor3DAttributes src1_attrs;
+    Tensor3DAttributes src2_attrs;
+    Tensor3DAttributes dst_attrs;
 };
 
 TENSOR_DECLARATION(1, src1Buffer, uvec4, src1_ptr, src1_shift, 4, readonly);
@@ -51,9 +51,9 @@ TENSOR_DECLARATION(3, dstBuffer, uvec4, dst_ptr, dst_shift, 4, writeonly);
 
 void main(void)
 {
-    ImageIterator src1_iter = CONVERT_TO_IMAGE_ITERATOR(src1_attrs, src1_shift);
-    ImageIterator src2_iter = CONVERT_TO_IMAGE_ITERATOR(src2_attrs, src2_shift);
-    ImageIterator dst_iter  = CONVERT_TO_IMAGE_ITERATOR(dst_attrs, dst_shift);
+    Tensor3DIterator src1_iter = CONVERT_TO_TENSOR3D_ITERATOR(src1_attrs, src1_shift);
+    Tensor3DIterator src2_iter = CONVERT_TO_TENSOR3D_ITERATOR(src2_attrs, src2_shift);
+    Tensor3DIterator dst_iter  = CONVERT_TO_TENSOR3D_ITERATOR(dst_attrs, dst_shift);
 
     vec4 tmp1[2] = LOAD_UNPACK8_CURRENT_ITEM_HALF(src1_ptr, src1_iter);
     vec4 tmp2[2] = LOAD_UNPACK8_CURRENT_ITEM_HALF(src2_ptr, src2_iter);
@@ -62,4 +62,4 @@ void main(void)
     addition[1] = ADD(tmp1[1], tmp2[1]);
 
     STORE_PACK8_CURRENT_ITEM_HALF(dst_ptr, dst_iter, addition);
-}
\ No newline at end of file
+}
diff --git a/src/core/GLES_COMPUTE/cs_shaders/concatenate.cs b/src/core/GLES_COMPUTE/cs_shaders/concatenate.cs
index 0c8b5bf0bb..69ac50b4d0 100644
--- a/src/core/GLES_COMPUTE/cs_shaders/concatenate.cs
+++ b/src/core/GLES_COMPUTE/cs_shaders/concatenate.cs
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -53,8 +53,8 @@ void main(void)
     Tensor3DIterator src_iter = CONVERT_TO_TENSOR3D_ITERATOR(src_attrs, src_shift);
     Tensor3DIterator dst_iter = CONVERT_TO_TENSOR3D_ITERATOR(dst_attrs, dst_shift);
 
-    float tmp = LOAD(src_ptr, TENSOR3D_OFFSET(src_iter, -OFFSETS_X, -OFFSETS_Y, 0));
-    STORE(dst_ptr, TENSOR_OFFSET_ADVANCE_IN_BYTES(dst_iter, OFFSETS_Z), tmp);
+    float tmp = LOAD(src_ptr, TENSOR3D_OFFSET(src_iter, -OFFSET_X, -OFFSET_Y, 0));
+    STORE_CURRENT_ITEM(dst_ptr, dst_iter, tmp);
 }
 
 #elif defined(DATA_TYPE_FP16)
@@ -66,7 +66,7 @@ void main(void)
     Tensor3DIterator src_iter = CONVERT_TO_TENSOR3D_ITERATOR(src_attrs, src_shift);
     Tensor3DIterator dst_iter = CONVERT_TO_TENSOR3D_ITERATOR(dst_attrs, dst_shift);
 
-    uvec2 tmp = LOAD(src_ptr, TENSOR3D_OFFSET(src_iter, -OFFSETS_X, -OFFSETS_Y, 0));
-    STORE(dst_ptr, TENSOR_OFFSET_ADVANCE_IN_BYTES(dst_iter, OFFSETS_Z), tmp);
+    uvec2 tmp = LOAD(src_ptr, TENSOR3D_OFFSET(src_iter, -OFFSET_X, -OFFSET_Y, 0));
+    STORE_CURRENT_ITEM(dst_ptr, dst_iter, tmp);
 }
 #endif /*DATA_TYPE_FP16*/
diff --git a/src/core/GLES_COMPUTE/cs_shaders/scale.cs b/src/core/GLES_COMPUTE/cs_shaders/scale.cs
index b2689a257d..b72c3392aa 100644
--- a/src/core/GLES_COMPUTE/cs_shaders/scale.cs
+++ b/src/core/GLES_COMPUTE/cs_shaders/scale.cs
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016, 2018 ARM Limited.
+ * Copyright (c) 2016-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -29,23 +29,23 @@ layout(local_size_x = LOCAL_SIZE_X, local_size_y = LOCAL_SIZE_Y, local_size_z =
 // We DO have to use highp for DATA_TYPE_FP16 float here to calculate the coordinates of source tensor. float is highp by default, but we still write it down here to make it more clearly, and mediump is only used for src/dst tensor in shader body.
 precision highp float;
 
-/** Performs an affine transformation on an image interpolating with the NEAREAST NEIGHBOUR method. Input and output are single channel FP16.
+/** Performs an affine transformation on an tensor interpolating with the NEAREAST NEIGHBOUR method. Input and output are single channel FP16.
  *
  * @param[in]  src_ptr      Pointer to the source tensor. Supported data types: FP16.
  * @param[in]  src_attrs    The attributes of the source tensor
  * @param[out] dst_ptr      Pointer to the destination tensor. Supported data types: FP16. (Must be the same as the input)
  * @param[in]  dst_attrs    The attributes of the destination tensor
- * @param[in]  input_width  Input image width
- * @param[in]  input_height Input image height
+ * @param[in]  input_width  Input tensor width
+ * @param[in]  input_height Input tensor height
  * @param[in]  scale        The scale factor along x/y dimension
  */
 SHADER_PARAMS_DECLARATION
 {
-    ImageAttributes src_attrs;
-    ImageAttributes dst_attrs;
-    float           input_width;
-    float           input_height;
-    vec2            scale;
+    Tensor3DAttributes src_attrs;
+    Tensor3DAttributes dst_attrs;
+    float              input_width;
+    float              input_height;
+    vec2               scale;
 };
 
 #if defined(DATA_TYPE_FP16)
@@ -75,8 +75,8 @@ vec4[2] clamp_to_border_with_size(vec4[2] coords, float width, float height, flo
 
 void main()
 {
-    ImageIterator src_iter = CONVERT_TO_IMAGE_ITERATOR_NO_STEP(src_attrs, src_shift);
-    ImageIterator dst_iter = CONVERT_TO_IMAGE_ITERATOR(dst_attrs, dst_shift);
+    Tensor3DIterator src_iter = CONVERT_TO_TENSOR3D_ITERATOR_NO_STEP(src_attrs, src_shift);
+    Tensor3DIterator dst_iter = CONVERT_TO_TENSOR3D_ITERATOR(dst_attrs, dst_shift);
 
     vec4[2] tc = clamp_to_border_with_size(transform_nearest(vec2(gl_GlobalInvocationID.x << uint(2), gl_GlobalInvocationID.y), scale), input_width, input_height, float(BORDER_SIZE));
 
@@ -85,7 +85,7 @@ void main()
 
     for(int i = 0; i < 4; i++)
     {
-        uint offset_in_bytes = image_offset_in_bytes(src_iter, int(tc[0][i]), int(tc[1][i]));
+        uint offset_in_bytes = tensor3D_offset_in_bytes(src_iter, int(tc[0][i]), int(tc[1][i]), int(gl_GlobalInvocationID.z));
 
         s = LOAD_UNPACK2_HALF(src_ptr, uint(offset_in_bytes >> src_shift));
 
@@ -107,15 +107,15 @@ TENSOR_DECLARATION(2, dstBuffer, uvec4, dst_ptr, dst_shift, 4, writeonly);
 
 void main()
 {
-    ImageIterator src_iter = CONVERT_TO_IMAGE_ITERATOR_NO_STEP(src_attrs, src_shift);
-    ImageIterator dst_iter = CONVERT_TO_IMAGE_ITERATOR(dst_attrs, dst_shift);
+    Tensor3DIterator src_iter = CONVERT_TO_TENSOR3D_ITERATOR_NO_STEP(src_attrs, src_shift);
+    Tensor3DIterator dst_iter = CONVERT_TO_TENSOR3D_ITERATOR(dst_attrs, dst_shift);
 
     uvec2 tc = uvec2(gl_GlobalInvocationID.x << uint(2), gl_GlobalInvocationID.y >> uint(1));
 
     mediump vec4 s = vec4(0.0f);
     mediump      vec4[2] d;
 
-    s = LOAD_UNPACK4_HALF(src_ptr, IMAGE_OFFSET(src_iter, int(tc[0]), int(tc[1])));
+    s = LOAD_UNPACK4_HALF(src_ptr, TENSOR3D_OFFSET(src_iter, int(tc[0]), int(tc[1]), int(gl_GlobalInvocationID.z)));
 
     d[0] = vec4(s.x, s.x, s.y, s.y);
     d[1] = vec4(s.z, s.z, s.w, s.w);