Skip to content

fix(bindings): two silent wrong results in the vector_add examples#2575

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

Open
LeSingh1 wants to merge 1 commit into
base:main
Choose a base branch
from
+22 −5
Open

fix(bindings): two silent wrong results in the vector_add examples #2575

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
8 changes: 7 additions & 1 deletion cuda_bindings/examples/0_Introduction/vector_add_drv.py
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -101,9 +101,15 @@ def main():
# h_C contains the result in host memory
check_cuda_errors(cuda.cuMemcpyDtoH(h_c, d_c, nbytes))

# A completed C `for` loop leaves i == n, which is why the C sample can
# test `i == N` for success. A completed Python loop leaves i == n - 1, so
# a flag is needed: `i + 1 != n` is also what `break` at the last index
# produces, and a wrong value in h_c[n - 1] would be reported as a pass.
mismatch = False
for i in range(n):
sum_all = h_a[i] + h_b[i]
if math.fabs(h_c[i] - sum_all) > 1e-7:
mismatch = True
break

# Free device memory
Expand All@@ -112,7 +118,7 @@ def main():
check_cuda_errors(cuda.cuMemFree(d_c))

check_cuda_errors(cuda.cuCtxDestroy(cu_context))
if i + 1 != n:
if mismatch:
print("Result = FAIL", file=sys.stderr)
sys.exit(1)

Expand Down
19 changes: 15 additions & 4 deletions cuda_bindings/examples/0_Introduction/vector_add_mmap.py
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -157,8 +157,13 @@ def simple_malloc_multi_device_mmap(size, resident_devices, mapping_devices, ali
simple_free_multi_device_mmap(dptr, size)
return status, None, None

# Each accessDescriptor will describe the mapping requirement for a single device
access_descriptors = [cuda.CUmemAccessDesc()] * len(mapping_devices)
# Each accessDescriptor will describe the mapping requirement for a single device.
# One CUmemAccessDesc per device: `[CUmemAccessDesc()] * n` would store n
# references to a single mutable object, so the loop below would overwrite
# the same descriptor n times and cuMemSetAccess would receive the last
# device repeated n times -- succeeding, while every other device faults on
# first touch.
access_descriptors = [cuda.CUmemAccessDesc() for _ in mapping_devices]

# Prepare the access descriptor array indicating where and how the backings should be visible.
for idx in range(len(mapping_devices)):
Expand DownExpand Up@@ -290,10 +295,16 @@ def main():
# h_C contains the result in host memory
check_cuda_errors(cuda.cuMemcpyDtoH(h_c, d_c, size))

# Verify result
# Verify result.
# A completed C `for` loop leaves i == n, which is why the C sample can
# test `i == N` for success. A completed Python loop leaves i == n - 1, so
# a flag is needed: `i + 1 != n` is also what `break` at the last index
# produces, and a wrong value in h_c[n - 1] would be reported as a pass.
mismatch = False
for i in range(n):
sum_all = h_a[i] + h_b[i]
if math.fabs(h_c[i] - sum_all) > 1e-7:
mismatch = True
break

check_cuda_errors(simple_free_multi_device_mmap(d_a, allocation_size))
Expand All@@ -302,7 +313,7 @@ def main():

check_cuda_errors(cuda.cuCtxDestroy(cu_context))

if i + 1 != n:
if mismatch:
print("Result = FAIL", file=sys.stderr)
sys.exit(1)

Expand Down
Loading