Bug #1809
closed
3.1RC1 No more CUDA Support
Added by Moody Mob over 8 years ago.
Updated almost 7 years ago.
Description
Compilation : ./configure --enable-geoip --enable-lua --enable-python --enable-prelude --enable-cuda --prefix=/opt/suricata
/opt/suricata/bin/suricata -c /opt/suricata/etc/suricata/suricata.yaml --af-packet=wlan0
14/6/2016 -- 09:36:55 - <Info> - GPU Device 1: GM20B, 2 Multiprocessors, 998MHz, CUDA Compute Capability 5.3
14/6/2016 -- 09:36:55 - <Notice> - This is Suricata version 3.1RC1 RELEASE
14/6/2016 -- 09:36:56 - <Error> - [ERRCODE: SC_ERR_CUDA_ERROR(134)] - cuMemcpyHtoD failed. Returned errocode - CUDA_ERROR_INVALID_VALUE
14/6/2016 -- 09:36:56 - <Error> - [ERRCODE: SC_ERR_AC_CUDA_ERROR(138)] - SCCudaMemcpyHtoD failure.
Same options under 3.0.1 run perfectly.
- Description updated (diff)
- Target version set to 3.1.1
I'll see if I can get it patched up.
One issue is simple:
diff --git a/src/app-layer-detect-proto.c b/src/app-layer-detect-proto.c
index fb25b5b..541fa24 100644
--- a/src/app-layer-detect-proto.c
+++ b/src/app-layer-detect-proto.c
@@ -1530,6 +1530,12 @@ int AppLayerProtoDetectSetup(void)
uint16_t spm_matcher = SinglePatternMatchDefaultMatcher();
uint16_t mpm_matcher = PatternMatchDefaultMatcher();
+#ifdef __SC_CUDA_SUPPORT__
+ /* CUDA won't work here, so fall back to AC */
+ if (mpm_matcher == MPM_AC_CUDA)
+ mpm_matcher = DEFAULT_MPM;
+#endif
+
alpd_ctx.spm_global_thread_ctx = SpmInitGlobalThreadCtx(spm_matcher);
if (alpd_ctx.spm_global_thread_ctx == NULL) {
SCLogError(SC_ERR_FATAL, "Unable to alloc SpmGlobalThreadCtx.");
But it seems it requires some more work to actually work correctly.
- Assignee set to Anonymous
- Target version changed from 3.1.1 to TBD
3.1.1 will contain the above fix, but other fixes may be needed. If anyone has interest in helping to address those, please step forward.
- Status changed from New to Closed
- Assignee deleted (
Anonymous)
- Target version deleted (
TBD)
Also available in: Atom
PDF