Force output shape to be correct for pooling operations

The avg_pool2d and max_pool2d operators trigger an error_if
statement if the output shape provided does not match the output
shape generated by the operation given the input arguments.

 * Add pseudocode for floor operation
 * Fixed type in RESCALE op

Signed-off-by: Matthew Haddon <matthew.haddon@arm.com>
Change-Id: I087faa331c30becc05719cd1261f436019dcdeb5

3 files changed, 18 insertions, 9 deletions

diff --git a/chapters/pseudocode.adoc b/chapters/pseudocode.adoc
index e6b7604..5e6af5b 100644
--- a/chapters/pseudocode.adoc
+++ b/chapters/pseudocode.adoc
@@ -128,7 +128,7 @@ out_t truncate(in_t input)
   output is the sizeof(out_t) least significant bits in input.
 ----
 
-The following definition is used to flatten a list of lists into a single list
+The following definition is used to flatten a list of lists into a single list.
 
 [source,c++]
 ----
@@ -142,10 +142,13 @@ in_t* flatten(in_t lists[]) {
 }
 ----
 
-The following helper function returns the length of a list
+Generic helper functions used to keep the pseudocode concise.
 
 [source,c++]
 ----
 int length(in_t input)
-    return number of elements in input
-----
+    return number of elements in input list
+
+int floor(in_t input)
+    return input value rounded down to nearest integer
+----
\ No newline at end of file
diff --git a/chapters/tensor_ops.adoc b/chapters/tensor_ops.adoc
index 0b78c8f..ad4d75d 100644
--- a/chapters/tensor_ops.adoc
+++ b/chapters/tensor_ops.adoc
@@ -96,8 +96,11 @@ ERROR_IF(pad_top < 0 || pad_bottom < 0 || pad_left < 0 || pad_right < 0);
 // a divide-by-zero.
 ERROR_IF(pad_right >= kernel_x || pad_left >= kernel_x);
 ERROR_IF(pad_top >= kernel_y || pad_bottom >= kernel_y);
+// Output shape must match expected shape given the input shape
+// and arguments provided
+ERROR_IF(H != floor((IH + pad_top + pad_bottom + stride_y - kernel_y) / stride_y))
+ERROR_IF(W != floor((IW + pad_left + pad_right + stride_x - kernel_x) / stride_x))
 
-pad = flatten([0,0], pad, [0,0]);
 for_each(0 <= n < N, 0 <= oy < H, 0 <= ox < W, 0 <= c < C ) {
     in_t output_val;
     acc_t acc = 0;
@@ -428,6 +431,10 @@ ERROR_IF(pad_top < 0 || pad_bottom < 0 || pad_left < 0 || pad_right < 0);
 // input values will be used.
 ERROR_IF(pad_right >= kernel_x || pad_left >= kernel_x);
 ERROR_IF(pad_top >= kernel_y || pad_bottom >= kernel_y);
+// Output shape must match expected shape given the input shape
+// and arguments provided
+ERROR_IF(H != floor((IH + pad_top + pad_bottom + stride_y - kernel_y) / stride_y))
+ERROR_IF(W != floor((IW + pad_left + pad_right + stride_x - kernel_x) / stride_x))
 
 for_each(0 <= n < N, 0 <= oy < H, 0 <= ox < W, 0 <= c < C ) {
     in_t acc = minimum_value<in_t>;
@@ -436,7 +443,6 @@ for_each(0 <= n < N, 0 <= oy < H, 0 <= ox < W, 0 <= c < C ) {
     for_each( 0 <= ky < kernel_y, 0 <= kx < kernel_x ) {
         y = iy + ky;
         x = ix + kx;
-        ERROR_IF(y >= IH + pad_bottom || x >= IW + pad_right);
         if (y >= 0 && y < IH && x >= 0 && x < IW) {
             in_t value = tensor_read<in_t>(input, [N,IH,IW,C], [n,y,x,c]);
             acc = apply_max(acc, value);
diff --git a/chapters/type_conversion.adoc b/chapters/type_conversion.adoc
index 8619fc3..aa2218f 100644
--- a/chapters/type_conversion.adoc
+++ b/chapters/type_conversion.adoc
@@ -124,9 +124,9 @@ for_each(index in shape) {
 |Any|signed 32 to signed 8|int32_t|int8_t
 |Any|signed 32 to signed 16|int32_t|int16_t
 |Any|signed 32 to signed 32|int32_t|int32_t
-|Any|signed 48 to signed 8|int48_t_t|int8_t
-|Any|signed 48 to signed 16|int48_t_t|int16_t
-|Any|signed 48 to signed 32|int48_t_t|int32_t
+|Any|signed 48 to signed 8|int48_t|int8_t
+|Any|signed 48 to signed 16|int48_t|int16_t
+|Any|signed 48 to signed 32|int48_t|int32_t
 |Any|unsigned 8 to signed 8|uint8_t|int8_t
 |Any|signed 8 to unsigned 8|int8_t|uint8_t
 |===