kazum commented on a change in pull request #5502:
URL: https://github.com/apache/incubator-tvm/pull/5502#discussion_r421965191
##########
File path: python/tvm/relay/op/_tensor_grad.py
##########
@@ -116,12 +117,60 @@ def sinh_grad(orig, grad):
x = orig.args[0]
return [grad * cosh(x)]
+
+@register_gradient("acos")
+def acos_grad(orig, grad):
+ """Returns [grad * -1/((1 - (x ^ 2)) ^ 1/2)]"""
+ x = orig.args[0]
+ a = const(2.0)
+ ones = ones_like(x)
+ return [grad * (-ones / sqrt(ones_like(x) - power(x, a)))]
Review comment:
ones_like(x) => ones
##########
File path: python/tvm/relay/op/_tensor_grad.py
##########
@@ -116,12 +117,60 @@ def sinh_grad(orig, grad):
x = orig.args[0]
return [grad * cosh(x)]
+
+@register_gradient("acos")
+def acos_grad(orig, grad):
+ """Returns [grad * -1/((1 - (x ^ 2)) ^ 1/2)]"""
+ x = orig.args[0]
+ a = const(2.0)
+ ones = ones_like(x)
+ return [grad * (-ones / sqrt(ones_like(x) - power(x, a)))]
+
+
+@register_gradient("acosh")
+def acosh_grad(orig, grad):
+ """Returns [grad * 1/((x - 1) ^ 1/2 * (x + 1) ^ 1/2)]"""
+ x = orig.args[0]
+ a = const(2.0)
+ ones = ones_like(x)
+ return [grad * ones / sqrt(power(x, a) - ones)]
+
+
+@register_gradient("asin")
+def asin_grad(orig, grad):
+ """Returns [grad * 1/((1 - (x ^ 2)) ^ (1/2))]"""
+ x = orig.args[0]
+ a = const(2.0)
+ ones = ones_like(x)
+ return [grad * ones / sqrt(ones - power(x, a))]
+
+
+@register_gradient("asinh")
+def asinh_grad(orig, grad):
+ """Returns [grad * 1/((1 + (x ^ 2)) ^ (1/2))]"""
+ x = orig.args[0]
+ a = const(2.0)
+ ones = ones_like(x)
+ return [grad * ones / sqrt(ones + power(x, a))]
+
+
@register_gradient("atan")
def atan_grad(orig, grad):
"""Returns [grad * 1 / (1 + x ^ 2)]"""
x = orig.args[0]
a = const(2.0)
- return [grad * ones_like(x) / (ones_like(x) + power(x, a))]
+ ones = ones_like(x)
+ return [grad * ones / (ones + power(x, a))]
+
+
+@register_gradient("atanh")
+def atanh_grad(orig, grad):
+ """Returns xx[grad * 1 / (1 - x ^ 2)]"""
Review comment:
Remove `xx`.
##########
File path: python/tvm/relay/op/_tensor_grad.py
##########
@@ -116,12 +117,60 @@ def sinh_grad(orig, grad):
x = orig.args[0]
return [grad * cosh(x)]
+
+@register_gradient("acos")
+def acos_grad(orig, grad):
+ """Returns [grad * -1/((1 - (x ^ 2)) ^ 1/2)]"""
+ x = orig.args[0]
+ a = const(2.0)
+ ones = ones_like(x)
+ return [grad * (-ones / sqrt(ones_like(x) - power(x, a)))]
+
+
+@register_gradient("acosh")
+def acosh_grad(orig, grad):
+ """Returns [grad * 1/((x - 1) ^ 1/2 * (x + 1) ^ 1/2)]"""
+ x = orig.args[0]
+ a = const(2.0)
+ ones = ones_like(x)
+ return [grad * ones / sqrt(power(x, a) - ones)]
Review comment:
I think all the `power(x, const(2.0))` can be replaced with `x * x`.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
