1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
|
# VMD compatibility patch for CUDA 13
#
# CUDA 13 removed several fields from the cudaDeviceProp structure that
# VMD previously used to query device information. Specifically:
# - clockRate
# - kernelExecTimeoutEnabled
# - computeMode
# - singleToDoublePrecisionPerfRatio
#
# This patch replaces direct access to those fields with calls to
# cudaDeviceGetAttribute(), which is the recommended approach and is
# compatible with all CUDA versions.
#
# The changes are made in src/CUDAUtil.cu, in the functions:
# - vmd_cuda_device_props()
# - vmd_cuda_devpool_setdevice()
#
# Temporary variables are introduced where necessary to avoid scoping
# issues, while preserving the original logic.
#
--- a/src/CUDAUtil.cu.old 2026-10-06 11:36:05.084205696 +0200
+++ b/src/CUDAUtil.cu 2026-10-06 11:36:20.990156537 +0200
@@ -108,21 +108,21 @@
if (memb)
*memb = deviceProp.totalGlobalMem;
if (clockratekhz)
- *clockratekhz = deviceProp.clockRate;
+ cudaDeviceGetAttribute(clockratekhz, cudaDevAttrClockRate, dev);
if (smcount)
*smcount = deviceProp.multiProcessorCount;
if (asyncenginecount)
*asyncenginecount = deviceProp.asyncEngineCount;
if (kerneltimeout)
- *kerneltimeout = (deviceProp.kernelExecTimeoutEnabled != 0);
+ { int _kerneltimeout = 0; cudaDeviceGetAttribute(&_kerneltimeout, cudaDevAttrKernelExecTimeout, dev); *kerneltimeout = (_kerneltimeout != 0); }
if (integratedgpu)
*integratedgpu = (deviceProp.integrated != 0);
if (canmaphostmem)
*canmaphostmem = (deviceProp.canMapHostMemory != 0);
if (computemode)
- *computemode = deviceProp.computeMode;
+ cudaDeviceGetAttribute(computemode, cudaDevAttrComputeMode, dev);
if (spdpfpperfratio)
- *spdpfpperfratio = deviceProp.singleToDoublePrecisionPerfRatio;
+ cudaDeviceGetAttribute(spdpfpperfratio, cudaDevAttrSingleToDoublePrecisionPerfRatio, dev);
if (pageablememaccess)
*pageablememaccess = deviceProp.pageableMemoryAccess;
if (pageablememaccessuseshostpagetables)
@@ -491,7 +491,7 @@
memset(&deviceProp, 0, sizeof(cudaDeviceProp));
if (cudaGetDeviceProperties(&deviceProp, dev) == cudaSuccess) {
float smscale = ((float) deviceProp.multiProcessorCount) / 30.0f;
- double clockscale = ((double) deviceProp.clockRate) / 1295000.0;
+ int _clockratekhz = 0; cudaDeviceGetAttribute(&_clockratekhz, cudaDevAttrClockRate, dev); double clockscale = ((double) _clockratekhz) / 1295000.0;
float speedscale = smscale * ((float) clockscale);
#if 0
|
