Hzfengsy commented on code in PR #15142:
URL: https://github.com/apache/tvm/pull/15142#discussion_r1240639829
##########
tests/python/unittest/test_meta_schedule_schedule_rule_mlt_tc.py:
##########
@@ -211,7 +212,7 @@ def matmul_relu_0(A: T.Buffer((128, 128), "float16"), B:
T.Buffer((128, 128), "f
actual = generate_design_space(
kind="cuda",
mod=mod,
- target=tvm.target.Target("cuda"),
+ target=tvm.target.Target("cuda --arch=sm_50"),
Review Comment:
It's so confusing that we mark arch `sm_50` for tensor core unit tests. I
don't think these changes are related to the `compute_inline` enhancement
##########
tests/python/unittest/test_tir_schedule_compute_inline.py:
##########
@@ -622,8 +621,8 @@ def elementwise_overcomputed_producer_reverse_inlined(
for i, j in T.grid(128, 128):
with T.block("B"):
vi, vj = T.axis.remap("SS", [i, j])
- T.where(i < 127 and j < 127)
- C[vi, vj] = A[vi, vj] * 2.0 + 1.0
+ if vi < 127 and vj < 127:
Review Comment:
What's the reason to change `T.where` to `IF` branches? There are two major
concerns:
1. with `If` branch we may lose changes for further schedule opportunities
after `compute_inline`
2. It's breaking changes for the existing codebase
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
