Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[SIMT] [cuda] Use correct source lane offset for warp intrinsics #4734

Merged
merged 3 commits into from
Apr 8, 2022
Merged

[SIMT] [cuda] Use correct source lane offset for warp intrinsics #4734

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
522313c
Correct lane offset
qiao-bo Apr 7, 2022
9fc39fb
Merge branch 'master' into bug
qiao-bo Apr 7, 2022
3aebc18
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Apr 7, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 15 additions & 7 deletions python/taichi/lang/simt/warp.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,38 +27,46 @@ def ballot(predicate):
def shfl_sync_i32(mask, val, offset):
return expr.Expr(
_ti_core.insert_internal_func_call(
"cuda_shfl_sync_i32", expr.make_expr_group(mask, val, offset, 32),
# lane offset is 31 for warp size 32
"cuda_shfl_sync_i32",
expr.make_expr_group(mask, val, offset, 31),
False))


def shfl_sync_f32(mask, val, offset):
return expr.Expr(
_ti_core.insert_internal_func_call(
"cuda_shfl_sync_f32", expr.make_expr_group(mask, val, offset, 32),
# lane offset is 31 for warp size 32
"cuda_shfl_sync_f32",
expr.make_expr_group(mask, val, offset, 31),
False))


def shfl_down_i32(mask, val, offset):
# Here we use 31 as the last argument since 32 (warp size) does not work
# for some reason. Using 31 leads to the desired behavior.
return expr.Expr(
_ti_core.insert_internal_func_call(
"cuda_shfl_down_sync_i32",
expr.make_expr_group(mask, val, offset, 31), False))
# lane offset is 31 for warp size 32
expr.make_expr_group(mask, val, offset, 31),
False))


def shfl_up_i32(mask, val, offset):
return expr.Expr(
_ti_core.insert_internal_func_call(
"cuda_shfl_up_sync_i32",
expr.make_expr_group(mask, val, offset, 32), False))
# lane offset is 0 for warp size 32
expr.make_expr_group(mask, val, offset, 0),
False))


def shfl_up_f32(mask, val, offset):
return expr.Expr(
_ti_core.insert_internal_func_call(
"cuda_shfl_up_sync_f32",
expr.make_expr_group(mask, val, offset, 32), False))
# lane offset is 0 for warp size 32
expr.make_expr_group(mask, val, offset, 0),
False))


def shfl_xor_i32(mask, val, offset):
Expand Down