COMPMID-908 - Merge Activation layer with Convolution Layer (NEON. CL, GLES)

Change-Id: Iab06d0768ecf805b841e601185608aae88cf9166
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/120874
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

1 files changed, 48 insertions, 1 deletions

diff --git a/src/core/GLES_COMPUTE/cs_shaders/direct_convolution1x1.cs b/src/core/GLES_COMPUTE/cs_shaders/direct_convolution1x1.cs
index ea4e9c18e2..b42c09bbc7 100644
--- a/src/core/GLES_COMPUTE/cs_shaders/direct_convolution1x1.cs
+++ b/src/core/GLES_COMPUTE/cs_shaders/direct_convolution1x1.cs
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -26,6 +26,10 @@ layout(local_size_x = LOCAL_SIZE_X, local_size_y = LOCAL_SIZE_Y, local_size_z =
 
 #include "helpers_cs.h"
 
+#ifdef FUSED_ACTIVATION
+#include "activation_layer_helpers_cs.h"
+#endif /* FUSED_ACTIVATION */
+
 #if defined(DATA_TYPE_FP16)
 precision mediump float;
 #endif // DATA_TYPE_FP16
@@ -99,6 +103,10 @@ void main()
     pixels += LOAD(biases_ptr, VECTOR_OFFSET(biases_iter, z_index));
 #endif /* BIAS */
 
+#ifdef FUSED_ACTIVATION
+    pixels = ACT_OP(pixels);
+#endif /* FUSED_ACTIVATION */
+
     STORE_CURRENT_ITEM(dst_ptr, dst_iter, pixels);
 }
 
@@ -210,6 +218,10 @@ void main()
     pixels += b;
 #endif /* BIAS */
 
+#ifdef FUSED_ACTIVATION
+    pixels = ACT_OP(pixels);
+#endif /* FUSED_ACTIVATION */
+
     STORE_PACK4_CURRENT_ITEM_HALF(dst_ptr, dst_iter, pixels);
 }
 #elif defined(PROCESS_4X_2Y_1Z)
@@ -333,6 +345,11 @@ void main()
     pixels[1] += b;
 #endif /* BIAS */
 
+#ifdef FUSED_ACTIVATION
+    pixels[0] = ACT_OP(pixels[0]);
+    pixels[1] = ACT_OP(pixels[1]);
+#endif /* FUSED_ACTIVATION */
+
     STORE_PACK4_CURRENT_ITEM_HALF(dst_ptr, dst_iter, pixels[0]);
     STORE_PACK4_HALF(dst_ptr, TENSOR3D_OFFSET(dst_iter, 0, 1, 0), pixels[1]);
 }
@@ -470,6 +487,12 @@ void main()
     pixels[2] += b;
 #endif /* BIAS */
 
+#ifdef FUSED_ACTIVATION
+    pixels[0] = ACT_OP(pixels[0]);
+    pixels[1] = ACT_OP(pixels[1]);
+    pixels[2] = ACT_OP(pixels[2]);
+#endif /* FUSED_ACTIVATION */
+
     STORE_PACK4_CURRENT_ITEM_HALF(dst_ptr, dst_iter, pixels[0]);
     STORE_PACK4_HALF(dst_ptr, TENSOR3D_OFFSET(dst_iter, 0, 1, 0), pixels[1]);
     STORE_PACK4_HALF(dst_ptr, TENSOR3D_OFFSET(dst_iter, 0, 2, 0), pixels[2]);
@@ -609,6 +632,13 @@ void main()
     pixels1[1] += b;
 #endif /* BIAS */
 
+#ifdef FUSED_ACTIVATION
+    pixels[0]  = ACT_OP(pixels[0]);
+    pixels[1]  = ACT_OP(pixels[1]);
+    pixels1[0] = ACT_OP(pixels1[0]);
+    pixels1[1] = ACT_OP(pixels1[1]);
+#endif /* FUSED_ACTIVATION */
+
     STORE_PACK4_CURRENT_ITEM_HALF(dst_ptr, dst_iter, pixels[0]);
     STORE_PACK4_HALF(dst_ptr, TENSOR3D_OFFSET(dst_iter, 0, 1, 0), pixels[1]);
     STORE_PACK4_HALF(dst_ptr, TENSOR3D_OFFSET(dst_iter, 0, 2, 0), pixels1[0]);
@@ -745,6 +775,11 @@ void main()
         pixels[1] += b;
 #endif /* BIAS */
 
+#ifdef FUSED_ACTIVATION
+        pixels[0] = ACT_OP(pixels[0]);
+        pixels[1] = ACT_OP(pixels[1]);
+#endif /* FUSED_ACTIVATION */
+
         STORE_PACK4_CURRENT_ITEM_HALF(dst_ptr, dst_iter, pixels[0]);
         STORE_PACK4_HALF(dst_ptr, TENSOR3D_OFFSET(dst_iter, 0, 1, 0), pixels[1]);
 
@@ -868,6 +903,11 @@ void main()
     pixels[1] += b;
 #endif /* BIAS */
 
+#ifdef FUSED_ACTIVATION
+    pixels[0] = ACT_OP(pixels[0]);
+    pixels[1] = ACT_OP(pixels[1]);
+#endif /* FUSED_ACTIVATION */
+
     STORE_PACK8_CURRENT_ITEM_HALF(dst_ptr, dst_iter, pixels);
 }
 #elif defined(PROCESS_8X_2Y_1Z)
@@ -1001,6 +1041,13 @@ void main()
     pixels1[1] += b;
 #endif /* BIAS */
 
+#ifdef FUSED_ACTIVATION
+    pixels[0]  = ACT_OP(pixels[0]);
+    pixels[1]  = ACT_OP(pixels[1]);
+    pixels1[0] = ACT_OP(pixels1[0]);
+    pixels1[1] = ACT_OP(pixels1[1]);
+#endif /* FUSED_ACTIVATION */
+
     STORE_PACK8_CURRENT_ITEM_HALF(dst_ptr, dst_iter, pixels);
     STORE_PACK8_HALF(dst_ptr, TENSOR3D_OFFSET(dst_iter, 0, 1, 0), pixels1);
 }