quic-sanirudh commented on PR #16642:
URL: https://github.com/apache/tvm/pull/16642#issuecomment-2033631336
> Ooh, interesting. I wasn't aware that `R.layout_transform` was implemented
in terms of `Schedule.layout_transform`. Can you take a look at #16832 and see
if it resolves your issue?
Lunderberg commented on PR #16642:
URL: https://github.com/apache/tvm/pull/16642#issuecomment-2032858841
Ooh, interesting. I wasn't aware that `R.layout_transform` was implemented
in terms of `Schedule.layout_transform`. Can you take a look at
https://github.com/apache/tvm/pull/16832 and
quic-sanirudh commented on PR #16642:
URL: https://github.com/apache/tvm/pull/16642#issuecomment-2032679685
@Lunderberg After this pass introduced support for computing struct_info for
PrimFuncs, I'm seeing a case where a primfunc whose output buffer is modified
with the
Lunderberg merged PR #16642:
URL: https://github.com/apache/tvm/pull/16642
--
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.
To unsubscribe, e-mail:
Lunderberg commented on PR #16642:
URL: https://github.com/apache/tvm/pull/16642#issuecomment-2020783444
Rebased onto main to resolve a merge conflict.
--
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
Lunderberg commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1535744877
##
src/tir/ir/function.cc:
##
@@ -21,12 +21,52 @@
* \file src/tir/ir/function.cc
* \brief The function data structure.
*/
+#include
#include
+#include
slyubomirsky commented on PR #16642:
URL: https://github.com/apache/tvm/pull/16642#issuecomment-2013697225
Thank you for your changes, I am very excited to have purity checking and
more accurate StructInfo for PrimFuncs. I had one question remaining about the
inference for argument types,
slyubomirsky commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1534643145
##
tests/python/relax/test_transform_compute_prim_value.py:
##
@@ -76,5 +76,29 @@ def compute_symbolic_expr(N: T.int64) -> T.bool:
T.ret(N % 16 == 0)
slyubomirsky commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1534634573
##
src/tir/ir/function.cc:
##
@@ -21,12 +21,52 @@
* \file src/tir/ir/function.cc
* \brief The function data structure.
*/
+#include
#include
+#include
slyubomirsky commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1534634573
##
src/tir/ir/function.cc:
##
@@ -21,12 +21,52 @@
* \file src/tir/ir/function.cc
* \brief The function data structure.
*/
+#include
#include
+#include
Lunderberg commented on PR #16642:
URL: https://github.com/apache/tvm/pull/16642#issuecomment-1995945407
The additional test case has been added, and fixed, so this PR is ready for
review/merge.
--
This is an automated message from the Apache Git Service.
To respond to the message,
Lunderberg commented on PR #16642:
URL: https://github.com/apache/tvm/pull/16642#issuecomment-1989472731
This PR needs an additional test case before merging. On the main branch,
the following IRModule is valid. However, with this PR, it fails during
parsing.
```python
Lunderberg commented on PR #16642:
URL: https://github.com/apache/tvm/pull/16642#issuecomment-1986125550
> That's a good idea. What would be the analogue to such annotations in a
PrimFunc? Do we think we need it?
(Apologies, missed the notification on this one.) For a `PrimFunc`, it
slyubomirsky commented on PR #16642:
URL: https://github.com/apache/tvm/pull/16642#issuecomment-1974030446
That's a good idea. What would be the analogue to such annotations in a
PrimFunc? Do we think we need it?
--
This is an automated message from the Apache Git Service.
To respond to
Lunderberg commented on PR #16642:
URL: https://github.com/apache/tvm/pull/16642#issuecomment-197283
> Do you think it's necessary to have an override to assert a PrimFunc is
pure?
I think I prefer having the overrides at an individual operator level,
rather than on the function
slyubomirsky commented on PR #16642:
URL: https://github.com/apache/tvm/pull/16642#issuecomment-1972363867
Excellent, I'm glad to see an analysis for PrimFunc purity. Do you think
it's necessary to have an override to assert one is pure? It's doubtful that we
would need it right now but it
Lunderberg commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1507068274
##
src/relax/transform/compute_prim_value.cc:
##
@@ -0,0 +1,94 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license
Lunderberg commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1506827430
##
tests/python/relax/test_transform_compute_prim_value.py:
##
@@ -76,5 +76,29 @@ def compute_symbolic_expr(N: T.int64) -> T.bool:
T.ret(N % 16 == 0)
slyubomirsky commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1506785821
##
tests/python/relax/test_transform_compute_prim_value.py:
##
@@ -76,5 +76,29 @@ def compute_symbolic_expr(N: T.int64) -> T.bool:
T.ret(N % 16 == 0)
slyubomirsky commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1506784633
##
python/tvm/relax/transform/transform.py:
##
@@ -463,6 +463,16 @@ def KillAfterLastUse() -> tvm.ir.transform.Pass:
return _ffi_api.KillAfterLastUse() #
Lunderberg commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1505082724
##
python/tvm/relax/transform/transform.py:
##
@@ -463,6 +463,16 @@ def KillAfterLastUse() -> tvm.ir.transform.Pass:
return _ffi_api.KillAfterLastUse() # type:
slyubomirsky commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1505049818
##
python/tvm/relax/transform/transform.py:
##
@@ -463,6 +463,16 @@ def KillAfterLastUse() -> tvm.ir.transform.Pass:
return _ffi_api.KillAfterLastUse() #
slyubomirsky commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1505049818
##
python/tvm/relax/transform/transform.py:
##
@@ -463,6 +463,16 @@ def KillAfterLastUse() -> tvm.ir.transform.Pass:
return _ffi_api.KillAfterLastUse() #
slyubomirsky commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1505049818
##
python/tvm/relax/transform/transform.py:
##
@@ -463,6 +463,16 @@ def KillAfterLastUse() -> tvm.ir.transform.Pass:
return _ffi_api.KillAfterLastUse() #
Lunderberg commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1504921950
##
python/tvm/relax/transform/transform.py:
##
@@ -463,6 +463,16 @@ def KillAfterLastUse() -> tvm.ir.transform.Pass:
return _ffi_api.KillAfterLastUse() # type:
slyubomirsky commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1504800879
##
src/relax/transform/compute_prim_value.cc:
##
@@ -0,0 +1,94 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license
slyubomirsky commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1504800473
##
python/tvm/relax/transform/transform.py:
##
@@ -463,6 +463,16 @@ def KillAfterLastUse() -> tvm.ir.transform.Pass:
return _ffi_api.KillAfterLastUse() #
Lunderberg commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1504424495
##
src/relax/transform/compute_prim_value.cc:
##
@@ -0,0 +1,94 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license
Lunderberg commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1504405971
##
python/tvm/relax/transform/transform.py:
##
@@ -463,6 +463,16 @@ def KillAfterLastUse() -> tvm.ir.transform.Pass:
return _ffi_api.KillAfterLastUse() # type:
slyubomirsky commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1503443020
##
src/relax/transform/compute_prim_value.cc:
##
@@ -0,0 +1,94 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license
slyubomirsky commented on code in PR #16642:
URL: https://github.com/apache/tvm/pull/16642#discussion_r1503442355
##
python/tvm/relax/transform/transform.py:
##
@@ -463,6 +463,16 @@ def KillAfterLastUse() -> tvm.ir.transform.Pass:
return _ffi_api.KillAfterLastUse() #
31 matches
Mail list logo