Skip to content

MNT: Forward-port v13.4.2 to main #2938

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

Merged
leofang merged 11 commits into from
Sep 23, 2026
Merged

MNT: Forward-port v13.4.2 to main #2938

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
2b2255e
Prepare the 13.4.2 release
mdboom Sep 17, 2026
98038a8
Simplify release note
mdboom Sep 17, 2026
8181605
Fix typo
mdboom Sep 17, 2026
16d3e0b
test(pathfinder): handle cuTENSORMg removal in cuTENSOR 2.8 (#2809)
rwgk Sep 11, 2026
5c3a8e9
Add pathfinder support for cupti64_2026.3.1.dll (CTK 13.4.1) (#2896)
juenglin Sep 17, 2026
9f5f193
Prepare the 13.4.2 release (#2900)
mdboom Sep 17, 2026
f0fe2df
Record v13.4.2 release ancestry
leofang Sep 22, 2026
2dcb3f1
Fix CUDA pointer and coredump attribute types (#2929)
kkraus14 Sep 22, 2026
72116e4
fix(bindings): cover missing helper attrs and fix CU_JIT_WALL_TIME po…
juenglin Sep 22, 2026
972ee3b
[doc-only] Prepare the 13.4.3 release (#2936)
leofang Sep 23, 2026
ecf1790
Merge branch 'main' into 13.4.x
leofang Sep 23, 2026
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
157 changes: 156 additions & 1 deletion cuda_bindings/cuda/bindings/_lib/cyruntime/cyruntime.pxi
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -259,6 +259,26 @@ cdef cudaError_t getDescInfo(const cudaChannelFormatDesc* d, int *numberOfChanne
elif d[0].f in (cudaChannelFormatKind.cudaChannelFormatKindUnsignedNormalized1010102,):
if (d[0].x != 10) or (d[0].y != 10) or (d[0].z != 10) or (d[0].w != 2):
return cudaErrorInvalidChannelDescriptor
elif d[0].f in (cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Packed422,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Packed444,):
if (d[0].x != 8) or (d[0].y != 8) or (d[0].z != 8) or (d[0].w != 8):
return cudaErrorInvalidChannelDescriptor
elif d[0].f in (cudaChannelFormatKind.cudaChannelFormatKindUnsigned8SemiPlanar420,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned8SemiPlanar422,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned8SemiPlanar444,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Planar420,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Planar422,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Planar444,):
if (d[0].x != 8) or (d[0].y != 8) or (d[0].z != 8) or (d[0].w != 0):
return cudaErrorInvalidChannelDescriptor
elif d[0].f in (cudaChannelFormatKind.cudaChannelFormatKindUnsigned16SemiPlanar420,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned16SemiPlanar422,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned16SemiPlanar444,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned16Planar420,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned16Planar422,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned16Planar444,):
if (d[0].x != 16) or (d[0].y != 16) or (d[0].z != 16) or (d[0].w != 0):
return cudaErrorInvalidChannelDescriptor
else:
return cudaErrorInvalidChannelDescriptor

Expand Down Expand Up @@ -414,12 +434,104 @@ cdef cudaError_t getDescInfo(const cudaChannelFormatDesc* d, int *numberOfChanne
elif case_desc(d, 10, 10, 10, 2, cudaChannelFormatKind.cudaChannelFormatKindUnsignedNormalized1010102):
numberOfChannels[0] = 4
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UNORM_INT_101010_2
# 2a: normalized 8/16-bit kinds — validated above but previously had no case_desc arm
elif case_desc(d, 8, 0, 0, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsignedNormalized8X1):
numberOfChannels[0] = 1
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UNORM_INT8X1
elif case_desc(d, 8, 8, 0, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsignedNormalized8X2):
numberOfChannels[0] = 2
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UNORM_INT8X2
elif case_desc(d, 8, 8, 8, 8, cudaChannelFormatKind.cudaChannelFormatKindUnsignedNormalized8X4):
numberOfChannels[0] = 4
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UNORM_INT8X4
elif case_desc(d, 8, 0, 0, 0, cudaChannelFormatKind.cudaChannelFormatKindSignedNormalized8X1):
numberOfChannels[0] = 1
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_SNORM_INT8X1
elif case_desc(d, 8, 8, 0, 0, cudaChannelFormatKind.cudaChannelFormatKindSignedNormalized8X2):
numberOfChannels[0] = 2
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_SNORM_INT8X2
elif case_desc(d, 8, 8, 8, 8, cudaChannelFormatKind.cudaChannelFormatKindSignedNormalized8X4):
numberOfChannels[0] = 4
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_SNORM_INT8X4
elif case_desc(d, 16, 0, 0, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsignedNormalized16X1):
numberOfChannels[0] = 1
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UNORM_INT16X1
elif case_desc(d, 16, 16, 0, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsignedNormalized16X2):
numberOfChannels[0] = 2
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UNORM_INT16X2
elif case_desc(d, 16, 16, 16, 16, cudaChannelFormatKind.cudaChannelFormatKindUnsignedNormalized16X4):
numberOfChannels[0] = 4
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UNORM_INT16X4
elif case_desc(d, 16, 0, 0, 0, cudaChannelFormatKind.cudaChannelFormatKindSignedNormalized16X1):
numberOfChannels[0] = 1
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_SNORM_INT16X1
elif case_desc(d, 16, 16, 0, 0, cudaChannelFormatKind.cudaChannelFormatKindSignedNormalized16X2):
numberOfChannels[0] = 2
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_SNORM_INT16X2
elif case_desc(d, 16, 16, 16, 16, cudaChannelFormatKind.cudaChannelFormatKindSignedNormalized16X4):
numberOfChannels[0] = 4
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_SNORM_INT16X4
# 2b: CUDA 13.3/13.4 packed and multi-planar YUV kinds
elif case_desc(d, 8, 8, 8, 8, cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Packed422):
numberOfChannels[0] = 4
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UINT8_PACKED_422
elif case_desc(d, 8, 8, 8, 8, cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Packed444):
numberOfChannels[0] = 4
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UINT8_PACKED_444
elif case_desc(d, 8, 8, 8, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsigned8SemiPlanar420):
numberOfChannels[0] = 3
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UINT8_SEMIPLANAR_420
elif case_desc(d, 16, 16, 16, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsigned16SemiPlanar420):
numberOfChannels[0] = 3
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UINT16_SEMIPLANAR_420
elif case_desc(d, 8, 8, 8, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsigned8SemiPlanar422):
numberOfChannels[0] = 3
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UINT8_SEMIPLANAR_422
elif case_desc(d, 16, 16, 16, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsigned16SemiPlanar422):
numberOfChannels[0] = 3
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UINT16_SEMIPLANAR_422
elif case_desc(d, 8, 8, 8, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsigned8SemiPlanar444):
numberOfChannels[0] = 3
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UINT8_SEMIPLANAR_444
elif case_desc(d, 16, 16, 16, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsigned16SemiPlanar444):
numberOfChannels[0] = 3
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UINT16_SEMIPLANAR_444
elif case_desc(d, 8, 8, 8, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Planar420):
numberOfChannels[0] = 3
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UINT8_PLANAR_420
elif case_desc(d, 16, 16, 16, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsigned16Planar420):
numberOfChannels[0] = 3
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UINT16_PLANAR_420
elif case_desc(d, 8, 8, 8, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Planar422):
numberOfChannels[0] = 3
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UINT8_PLANAR_422
elif case_desc(d, 16, 16, 16, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsigned16Planar422):
numberOfChannels[0] = 3
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UINT16_PLANAR_422
elif case_desc(d, 8, 8, 8, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Planar444):
numberOfChannels[0] = 3
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UINT8_PLANAR_444
elif case_desc(d, 16, 16, 16, 0, cudaChannelFormatKind.cudaChannelFormatKindUnsigned16Planar444):
numberOfChannels[0] = 3
format[0] = cydriver.CUarray_format_enum.CU_AD_FORMAT_UINT16_PLANAR_444
else:
return cudaErrorInvalidChannelDescriptor

if d[0].f in (cudaChannelFormatKind.cudaChannelFormatKindNV12,
cudaChannelFormatKind.cudaChannelFormatKindUnsignedBlockCompressed6H,
cudaChannelFormatKind.cudaChannelFormatKindSignedBlockCompressed6H,):
cudaChannelFormatKind.cudaChannelFormatKindSignedBlockCompressed6H,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned8SemiPlanar420,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned16SemiPlanar420,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned8SemiPlanar422,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned16SemiPlanar422,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned8SemiPlanar444,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned16SemiPlanar444,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Planar420,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned16Planar420,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Planar422,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned16Planar422,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Planar444,
cudaChannelFormatKind.cudaChannelFormatKindUnsigned16Planar444,):
if numberOfChannels[0] != 3:
return cudaErrorInvalidChannelDescriptor
else:
Expand Down Expand Up @@ -536,6 +648,49 @@ cdef cudaError_t getChannelFormatDescFromDriverDesc(cudaChannelFormatDesc* pRunt
channel_size = 8
elif pDriverDesc[0].Format == cydriver.CU_AD_FORMAT_UNORM_INT_101010_2:
pRuntimeDesc[0].f = cudaChannelFormatKind.cudaChannelFormatKindUnsignedNormalized1010102
# 2b: CUDA 13.3/13.4 packed and multi-planar YUV formats
elif pDriverDesc[0].Format == cydriver.CU_AD_FORMAT_UINT8_PACKED_422:
pRuntimeDesc[0].f = cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Packed422
channel_size = 8
elif pDriverDesc[0].Format == cydriver.CU_AD_FORMAT_UINT8_PACKED_444:
pRuntimeDesc[0].f = cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Packed444
channel_size = 8
elif pDriverDesc[0].Format == cydriver.CU_AD_FORMAT_UINT8_SEMIPLANAR_420:
pRuntimeDesc[0].f = cudaChannelFormatKind.cudaChannelFormatKindUnsigned8SemiPlanar420
channel_size = 8
elif pDriverDesc[0].Format == cydriver.CU_AD_FORMAT_UINT16_SEMIPLANAR_420:
pRuntimeDesc[0].f = cudaChannelFormatKind.cudaChannelFormatKindUnsigned16SemiPlanar420
channel_size = 16
elif pDriverDesc[0].Format == cydriver.CU_AD_FORMAT_UINT8_SEMIPLANAR_422:
pRuntimeDesc[0].f = cudaChannelFormatKind.cudaChannelFormatKindUnsigned8SemiPlanar422
channel_size = 8
elif pDriverDesc[0].Format == cydriver.CU_AD_FORMAT_UINT16_SEMIPLANAR_422:
pRuntimeDesc[0].f = cudaChannelFormatKind.cudaChannelFormatKindUnsigned16SemiPlanar422
channel_size = 16
elif pDriverDesc[0].Format == cydriver.CU_AD_FORMAT_UINT8_SEMIPLANAR_444:
pRuntimeDesc[0].f = cudaChannelFormatKind.cudaChannelFormatKindUnsigned8SemiPlanar444
channel_size = 8
elif pDriverDesc[0].Format == cydriver.CU_AD_FORMAT_UINT16_SEMIPLANAR_444:
pRuntimeDesc[0].f = cudaChannelFormatKind.cudaChannelFormatKindUnsigned16SemiPlanar444
channel_size = 16
elif pDriverDesc[0].Format == cydriver.CU_AD_FORMAT_UINT8_PLANAR_420:
pRuntimeDesc[0].f = cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Planar420
channel_size = 8
elif pDriverDesc[0].Format == cydriver.CU_AD_FORMAT_UINT16_PLANAR_420:
pRuntimeDesc[0].f = cudaChannelFormatKind.cudaChannelFormatKindUnsigned16Planar420
channel_size = 16
elif pDriverDesc[0].Format == cydriver.CU_AD_FORMAT_UINT8_PLANAR_422:
pRuntimeDesc[0].f = cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Planar422
channel_size = 8
elif pDriverDesc[0].Format == cydriver.CU_AD_FORMAT_UINT16_PLANAR_422:
pRuntimeDesc[0].f = cudaChannelFormatKind.cudaChannelFormatKindUnsigned16Planar422
channel_size = 16
elif pDriverDesc[0].Format == cydriver.CU_AD_FORMAT_UINT8_PLANAR_444:
pRuntimeDesc[0].f = cudaChannelFormatKind.cudaChannelFormatKindUnsigned8Planar444
channel_size = 8
elif pDriverDesc[0].Format == cydriver.CU_AD_FORMAT_UINT16_PLANAR_444:
pRuntimeDesc[0].f = cudaChannelFormatKind.cudaChannelFormatKindUnsigned16Planar444
channel_size = 16
else:
return cudaErrorInvalidChannelDescriptor

Expand Down
7 changes: 5 additions & 2 deletions cuda_bindings/cuda/bindings/_lib/utils.pxd
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@ cimport cuda.bindings.driver as driver
cimport cuda.bindings.cydriver as cydriver
cimport cuda.bindings.cyruntime as cyruntime
from libcpp.vector cimport vector
from libcpp cimport bool as cpp_bool
from cpython.buffer cimport PyBuffer_Release, Py_buffer

cdef class _HelperKernelParams:
Expand Down Expand Up @@ -35,6 +36,7 @@ cdef class _HelperCUmemPool_attribute:

# Return values
cdef int _int_val
cdef unsigned int _uint_val
cdef driver.cuuint64_t _cuuint64_t_val


Expand All @@ -60,7 +62,7 @@ cdef class _HelperCUpointer_attribute:
cdef unsigned int _uint
cdef int _int
cdef driver.CUdeviceptr _devptr
cdef void** _void
cdef void* _void
cdef driver.CUDA_POINTER_ATTRIBUTE_P2P_TOKENS _token
cdef bint _bool
cdef unsigned long long _ull
Expand Down Expand Up @@ -163,5 +165,6 @@ cdef class _HelperCUcoredumpSettings:
cdef object _references # keeps caller bytes alive so _charstar stays valid

# Return values
cdef bint _bool
cdef cpp_bool _bool
cdef unsigned int _uint
cdef char* _charstar
Loading
Loading