You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by GitBox <gi...@apache.org> on 2022/08/12 17:53:20 UTC
[GitHub] [tvm] vinx13 opened a new pull request, #12408: [MetaSchedule] Filter vector_load_lens based on buffer dtype
vinx13 opened a new pull request, #12408:
URL: https://github.com/apache/tvm/pull/12408
This makes the same config generic to work across workloads with different types.
cc @junrushao1994 @zxybazh @spectrometerHBH @masahi
--
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: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] junrushao1994 commented on a diff in pull request #12408: [MetaSchedule] Filter vector_load_lens based on buffer dtype
Posted by GitBox <gi...@apache.org>.
junrushao1994 commented on code in PR #12408:
URL: https://github.com/apache/tvm/pull/12408#discussion_r944843560
##########
tests/python/unittest/test_meta_schedule_schedule_rule_multi_level_tiling.py:
##########
@@ -883,13 +883,13 @@ def test_cuda_tensor_core_matmul_relu_global():
sch.compute_at(block=b70, loop=l46, preserve_unit_loops=True)
l71, l72, l73, l74, l75, l76 = sch.get_loops(block=b70)
l77 = sch.fuse(l75, l76, preserve_unit_iters=True)
-v78 = sch.sample_categorical(candidates=[1, 2, 3, 4], probs=[0.25, 0.25, 0.25, 0.25])
+v78 = sch.sample_categorical(candidates=[1, 2, 4, 8], probs=[0.25, 0.25, 0.25, 0.25])
sch.annotate(block_or_loop=b70, ann_key="meta_schedule.cooperative_fetch", ann_val=v78)
b79 = sch.cache_read(block=b19, read_buffer_index=1, storage_scope="shared")
sch.compute_at(block=b79, loop=l46, preserve_unit_loops=True)
l80, l81, l82, l83, l84, l85 = sch.get_loops(block=b79)
l86 = sch.fuse(l84, l85, preserve_unit_iters=True)
-v87 = sch.sample_categorical(candidates=[1, 2, 3, 4], probs=[0.25, 0.25, 0.25, 0.25])
+v87 = sch.sample_categorical(candidates=[1, 2, 4, 8], probs=[0.25, 0.25, 0.25, 0.25])
Review Comment:
BTW, i'm not sure if we should rule out `float3` in CUDA
##########
tests/python/unittest/test_meta_schedule_schedule_rule_multi_level_tiling.py:
##########
@@ -883,13 +883,13 @@ def test_cuda_tensor_core_matmul_relu_global():
sch.compute_at(block=b70, loop=l46, preserve_unit_loops=True)
l71, l72, l73, l74, l75, l76 = sch.get_loops(block=b70)
l77 = sch.fuse(l75, l76, preserve_unit_iters=True)
-v78 = sch.sample_categorical(candidates=[1, 2, 3, 4], probs=[0.25, 0.25, 0.25, 0.25])
+v78 = sch.sample_categorical(candidates=[1, 2, 4, 8], probs=[0.25, 0.25, 0.25, 0.25])
sch.annotate(block_or_loop=b70, ann_key="meta_schedule.cooperative_fetch", ann_val=v78)
b79 = sch.cache_read(block=b19, read_buffer_index=1, storage_scope="shared")
sch.compute_at(block=b79, loop=l46, preserve_unit_loops=True)
l80, l81, l82, l83, l84, l85 = sch.get_loops(block=b79)
l86 = sch.fuse(l84, l85, preserve_unit_iters=True)
-v87 = sch.sample_categorical(candidates=[1, 2, 3, 4], probs=[0.25, 0.25, 0.25, 0.25])
+v87 = sch.sample_categorical(candidates=[1, 2, 4, 8], probs=[0.25, 0.25, 0.25, 0.25])
Review Comment:
BTW, i'm not sure if we should rule out `float3` on CUDA
--
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: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] junrushao1994 commented on a diff in pull request #12408: [MetaSchedule] Filter vector_load_lens based on buffer dtype
Posted by GitBox <gi...@apache.org>.
junrushao1994 commented on code in PR #12408:
URL: https://github.com/apache/tvm/pull/12408#discussion_r945110854
##########
tests/python/unittest/test_meta_schedule_schedule_rule_multi_level_tiling.py:
##########
@@ -883,13 +883,13 @@ def test_cuda_tensor_core_matmul_relu_global():
sch.compute_at(block=b70, loop=l46, preserve_unit_loops=True)
l71, l72, l73, l74, l75, l76 = sch.get_loops(block=b70)
l77 = sch.fuse(l75, l76, preserve_unit_iters=True)
-v78 = sch.sample_categorical(candidates=[1, 2, 3, 4], probs=[0.25, 0.25, 0.25, 0.25])
+v78 = sch.sample_categorical(candidates=[1, 2, 4, 8], probs=[0.25, 0.25, 0.25, 0.25])
sch.annotate(block_or_loop=b70, ann_key="meta_schedule.cooperative_fetch", ann_val=v78)
b79 = sch.cache_read(block=b19, read_buffer_index=1, storage_scope="shared")
sch.compute_at(block=b79, loop=l46, preserve_unit_loops=True)
l80, l81, l82, l83, l84, l85 = sch.get_loops(block=b79)
l86 = sch.fuse(l84, l85, preserve_unit_iters=True)
-v87 = sch.sample_categorical(candidates=[1, 2, 3, 4], probs=[0.25, 0.25, 0.25, 0.25])
+v87 = sch.sample_categorical(candidates=[1, 2, 4, 8], probs=[0.25, 0.25, 0.25, 0.25])
Review Comment:
That makes sense!
--
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: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] junrushao1994 merged pull request #12408: [MetaSchedule] Filter vector_load_lens based on buffer dtype
Posted by GitBox <gi...@apache.org>.
junrushao1994 merged PR #12408:
URL: https://github.com/apache/tvm/pull/12408
--
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: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] vinx13 commented on a diff in pull request #12408: [MetaSchedule] Filter vector_load_lens based on buffer dtype
Posted by GitBox <gi...@apache.org>.
vinx13 commented on code in PR #12408:
URL: https://github.com/apache/tvm/pull/12408#discussion_r944857397
##########
tests/python/unittest/test_meta_schedule_schedule_rule_multi_level_tiling.py:
##########
@@ -883,13 +883,13 @@ def test_cuda_tensor_core_matmul_relu_global():
sch.compute_at(block=b70, loop=l46, preserve_unit_loops=True)
l71, l72, l73, l74, l75, l76 = sch.get_loops(block=b70)
l77 = sch.fuse(l75, l76, preserve_unit_iters=True)
-v78 = sch.sample_categorical(candidates=[1, 2, 3, 4], probs=[0.25, 0.25, 0.25, 0.25])
+v78 = sch.sample_categorical(candidates=[1, 2, 4, 8], probs=[0.25, 0.25, 0.25, 0.25])
sch.annotate(block_or_loop=b70, ann_key="meta_schedule.cooperative_fetch", ann_val=v78)
b79 = sch.cache_read(block=b19, read_buffer_index=1, storage_scope="shared")
sch.compute_at(block=b79, loop=l46, preserve_unit_loops=True)
l80, l81, l82, l83, l84, l85 = sch.get_loops(block=b79)
l86 = sch.fuse(l84, l85, preserve_unit_iters=True)
-v87 = sch.sample_categorical(candidates=[1, 2, 3, 4], probs=[0.25, 0.25, 0.25, 0.25])
+v87 = sch.sample_categorical(candidates=[1, 2, 4, 8], probs=[0.25, 0.25, 0.25, 0.25])
Review Comment:
This is fp16
--
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: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [tvm] vinx13 commented on pull request #12408: [MetaSchedule] Filter vector_load_lens based on buffer dtype
Posted by GitBox <gi...@apache.org>.
vinx13 commented on PR #12408:
URL: https://github.com/apache/tvm/pull/12408#issuecomment-1213545622
@tvm-bot rerun
--
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: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org