aboutsummaryrefslogtreecommitdiff
path: root/chapters/ewise_unary.adoc
diff options
context:
space:
mode:
Diffstat (limited to 'chapters/ewise_unary.adoc')
-rw-r--r--chapters/ewise_unary.adoc103
1 files changed, 52 insertions, 51 deletions
diff --git a/chapters/ewise_unary.adoc b/chapters/ewise_unary.adoc
index 34b64a9..7df32ba 100644
--- a/chapters/ewise_unary.adoc
+++ b/chapters/ewise_unary.adoc
@@ -19,7 +19,7 @@ Elementwise absolute value operation
|Argument|Type|Name|Shape|Description
|Input|in_t*|input1|shape|Input tensor
-|Output|out_t*|output|shape|Output tensor of same type, size as the input tensor
+|Output|in_t*|output|shape|Output tensor of same type, size as the input tensor
|===
*Operation Function:*
@@ -27,10 +27,10 @@ Elementwise absolute value operation
[source,c]
----
for_each (index in shape) {
- int32_t value1 = tensor_read<in_t>(input1, shape, index)
+ in_t value1 = tensor_read<in_t>(input1, shape, index)
if (value1 < 0)
- value1 = apply_sub<out_t>(0, value1)
- tensor_write<out_t>(output, shape, index, value1)
+ value1 = apply_sub<in_t>(0, value1)
+ tensor_write<in_t>(output, shape, index, value1)
}
----
@@ -39,8 +39,8 @@ for_each (index in shape) {
|===
|Profile|Mode|in_t|out_t
-|Any|signed 32|int32|int32
-|MI, MT|float|float|float
+|Any|signed 32|int32
+|MI, MT|float|float
|===
==== BITWISE_NOT
@@ -53,7 +53,7 @@ Elementwise bitwise NOT of input tensor.
|Argument|Type|Name|Shape|Description
|Input|in_t*|input1|shape|Input tensor
-|Output|out_t*|output|shape|Output tensor of same type, size as the input tensor
+|Output|in_t*|output|shape|Output tensor of same type, size as the input tensor
|===
*Quantization Parameters:*
@@ -65,20 +65,20 @@ None
[source,c]
----
for_each (index in shape) {
- int32_t value1 = tensor_read<in_t>(input1, shape, index)
- int32_t acc = ~value1
- tensor_write<out_t>(output, shape, index, acc)
+ in_t value1 = tensor_read<in_t>(input1, shape, index)
+ in_t acc = ~value1
+ tensor_write<in_t>(output, shape, index, acc)
}
----
*Supported Data Types:*
|===
-|Profile|Mode|in_t|out_t
+|Profile|Mode|in_t
-|Any|signed 8|aint8|aint8
-|Any|signed 16|int16|int16
-|Any|signed 32|int32|int32
+|Any|signed 8|aint8
+|Any|signed 16|int16
+|Any|signed 32|int32
|===
==== CEIL
@@ -91,15 +91,15 @@ Elementwise ceiling operation
|Argument|Type|Name|Shape|Description
|Input|in_t*|input1|shape|Input tensor
-|Output|out_t*|output|shape|Output tensor of same type, size as the input tensor
+|Output|in_t*|output|shape|Output tensor of same type, size as the input tensor
|===
*Supported Data Types:*
|===
-|Profile|Mode|in_t|out_t
+|Profile|Mode|in_t
-|MI, MT|float|float|float
+|MI, MT|float|float
|===
==== CLZ
@@ -112,7 +112,7 @@ Elementwise count leading zeros operation
|Argument|Type|Name|Shape|Description
|Input|in_t*|input1|shape|Input tensor
-|Output|out_t*|output|shape|Output tensor of same type, size as the input tensor
+|Output|in_t*|output|shape|Output tensor of same type, size as the input tensor
|===
*Operation Function:*
@@ -120,7 +120,8 @@ Elementwise count leading zeros operation
[source,c]
----
for_each (index in shape) {
- int32_t value1 = tensor_read<in_t>(input1, shape, index)
+ in_t acc = 0
+ in_t value1 = tensor_read<in_t>(input1, shape, index)
if (value1 == 0) {
acc = 32 // input1_width
}
@@ -132,15 +133,15 @@ for_each (index in shape) {
acc = acc + 1
}
}
- tensor_write<out_t>(output, shape, index, acc)
+ tensor_write<in_t>(output, shape, index, acc)
}
----
*Supported Data Types:*
|===
-|Profile|Mode|in_t|out_t
+|Profile|Mode|in_t
-|Any|signed 32|int32|int32
+|Any|signed 32|int32
|===
==== EXP
@@ -153,15 +154,15 @@ Elementwise e to the x operation
|Argument|Type|Name|Shape|Description
|Input|in_t*|input1|shape|Input tensor
-|Output|out_t*|output|shape|Output tensor of same type, size as the input tensor
+|Output|in_t*|output|shape|Output tensor of same type, size as the input tensor
|===
*Supported Data Types:*
|===
-|Profile|Mode|Input 0|Output
+|Profile|Mode|in_t
-|Any|float|float|float
+|Any|float|float
|===
==== FLOOR
@@ -174,15 +175,15 @@ Elementwise floor operation
|Argument|Type|Name|Shape|Description
|Input|in_t*|input1|shape|Input tensor
-|Output|out_t*|output|shape|Output tensor of same type, size as the input tensor
+|Output|in_t*|output|shape|Output tensor of same type, size as the input tensor
|===
*Supported Data Types:*
|===
-|Profile|Mode|in_t|out_t
+|Profile|Mode|in_t
-|MI, MT|float|float|float
+|MI, MT|float|float
|===
==== LOG
@@ -195,15 +196,15 @@ Elementwise natural logarithm operation
|Argument|Type|Name|Shape|Description
|Input|in_t*|input1|shape|Input tensor
-|Output|out_t*|output|shape|Output tensor of same type, size as the input tensor
+|Output|in_t*|output|shape|Output tensor of same type, size as the input tensor
|===
*Supported Data Types:*
|===
-|Profile|Mode|in_t|out_t
+|Profile|Mode|in_t
-|MI, MT|float|float|float
+|MI, MT|float|float
|===
==== LOGICAL_NOT
@@ -216,7 +217,7 @@ Elementwise logical NOT of input.
|Argument|Type|Name|Shape|Description
|Input|in_t*|input1|shape|Input tensor
-|Output|out_t*|output|shape|Output tensor of same type, size as the input tensor
+|Output|in_t*|output|shape|Output tensor of same type, size as the input tensor
|===
*Quantization Parameters:*
@@ -252,7 +253,7 @@ Elementwise negation operation
|Argument|Type|Name|Shape|Description
|Input|in_t*|input1|shape|Input tensor
-|Output|out_t*|output|shape|Output tensor of same type, size as the input tensor
+|Output|in_t*|output|shape|Output tensor of same type, size as the input tensor
|===
*Quantization Parameters:*
@@ -261,7 +262,7 @@ Elementwise negation operation
|Argument|Type|Name|Shape|Description
|Attribute|in_t|input1_zp|-|Input 1 zero point
-|Attribute|out_t|output_zp|-|Output zero point
+|Attribute|in_t|output_zp|-|Output zero point
|===
*Operation Function:*
@@ -269,24 +270,24 @@ Elementwise negation operation
[source,c]
----
assert(in_t == aint8_t || input_zp == 0) // Zero point only for asymmetric int8
-assert(out_t == aint8_t || output_zp == 0) // Zero point only for asymmetric int8
+assert(in_t == aint8_t || output_zp == 0) // Zero point only for asymmetric int8
for_each (index in shape) {
- int32_t value1 = tensor_read<in_t>(input1, shape, index)
- int32_t acc = appl_sub<int32_t>(0, value1 - input1_zp)
- acc = apply_clip(acc, minimum<out_t>, maximum<out_t>)
- tensor_write<out_t>(output + output_zp, shape, index, acc)
+ in_t value1 = tensor_read<in_t>(input1, shape, index)
+ in_t acc = appl_sub<in_t>(0, value1 - input1_zp)
+ acc = apply_clip(acc, minimum<in_t>, maximum<in_t>)
+ tensor_write<in_t>(output + output_zp, shape, index, acc)
}
----
*Supported Data Types:*
|===
-|Profile|Mode|Input 0|Output
+|Profile|Mode|in_t
-|Any|signed 8|aint8|aint8
-|Any|signed 16|int16|int16
-|Any|signed 32|int32|int32
-|MI, MT|float|float|float
+|Any|signed 8|aint8
+|Any|signed 16|int16
+|Any|signed 32|int32
+|MI, MT|float|float
|===
==== RECIPROCAL
@@ -299,15 +300,15 @@ Elementwise reciprocal operation. For integer operation, a TABLE should be used
|Argument|Type|Name|Shape|Description
|Input|in_t*|input1|shape|Input tensor
-|Output|out_t*|output|shape|Output tensor of same type, size as the input tensor
+|Output|in_t*|output|shape|Output tensor of same type, size as the input tensor
|===
*Supported Data Types:*
|===
-|Profile|Mode|Input 0|Output
+|Profile|Mode|in_t
-|MI, MT|float|float|float
+|MI, MT|float|float
|===
==== RSQRT
@@ -320,13 +321,13 @@ Elementwise reciprocal square root operation. For integer operation, a TABLE sho
|Argument|Type|Name|Shape|Description
|Input|in_t*|input1|shape|Input tensor
-|Output|out_t*|output|shape|Output tensor of same type, size as the input tensor
+|Output|in_t*|output|shape|Output tensor of same type, size as the input tensor
|===
*Supported Data Types:*
|===
-|Profile|Mode|Input 0|Output
+|Profile|Mode|in_t
-|MI, MT|float|float|float
-|=== \ No newline at end of file
+|MI, MT|float|float
+|===
generated by cgit v1.2.1 (git 2.23.0) at 2024-05-17 06:21:00 +0000