masahi commented on a change in pull request #7233:
URL: https://github.com/apache/tvm/pull/7233#discussion_r560015060
##########
File path: python/tvm/topi/cuda/scatter.py
##########
@@ -312,19 +313,18 @@ def gen_ir_4d(data, indices, updates, axis, out,
update_func):
out_ptr = ib.buffer_ptr(out)
data_ptr = ib.buffer_ptr(data)
with ib.new_scope():
- i = te.thread_axis("blockIdx.x")
- ib.scope_attr(i, "thread_extent", n)
- j = te.thread_axis("blockIdx.y")
- ib.scope_attr(j, "thread_extent", c)
- k = te.thread_axis("blockIdx.z")
- ib.scope_attr(k, "thread_extent", h)
+ fused = n * c * h * w
+ num_thread =
int(tvm.target.Target.current(allow_none=False).max_num_threads)
+ num_blocks = ceil_div(fused, num_thread)
Review comment:
Now all memcpy are done by the same IR.
----------------------------------------------------------------
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:
[email protected]
