From 0dc72089dccc10d3b55096ade5fc4d72de6cb96f Mon Sep 17 00:00:00 2001
From: Thomas Schwinge <tschwinge@baylibre.com>
Date: Thu, 7 Mar 2024 14:42:07 +0100
Subject: [PATCH] GCN, nvptx: Errors during device probing are fatal
Currently, we silently disable libgomp GCN and nvptx plugins/devices in
presence of certain error conditions during device probing, thus typically
silently resorting to host-fallback execution. Make such errors fatal, similar
as for any other device access later on, so that we early and reliably notice
when things go wrong. (Keep just two cases non-fatal: (a) libgomp GCN or nvptx
plugins are available but 'libhsa-runtime64.so.1' or 'libcuda.so.1' are not,
and (b) those are available, but the corresponding devices are not.)
This resolves the issue that we've got execution test cases unexpectedly
PASSing, despite:
libgomp: GCN fatal error: Run-time could not be initialized
Runtime message: HSA_STATUS_ERROR_OUT_OF_RESOURCES: The runtime failed to allocate the necessary resources. This error may also occur when the core runtime library needs to spawn threads or create internal OS-specific events.
..., and therefore they were not offloaded to the GCN device, but ran in
host-fallback execution mode. What happend in that scenario is that in
'init_hsa_context' during the initial 'GOMP_OFFLOAD_get_num_devices' we ran
into 'HSA_STATUS_ERROR_OUT_OF_RESOURCES', but it wasn't fatal, but just
silently disabled the libgomp plugin/device.
Especially "entertaining" were cases where such unintended host-fallback
execution happened during effective-target checks like
'offload_device_available' (host-fallback execution there meaning: no offload
device available), but actual test cases then were running with an offload
device available, and therefore mis-configured.
include/
* cuda/cuda.h (CUresult): Add 'CUDA_ERROR_NO_DEVICE'.
libgomp/
* plugin/plugin-gcn.c (init_hsa_context): Add and handle
'bool probe' parameter. Adjust all users; errors during device
probing are fatal.
* plugin/plugin-nvptx.c (nvptx_get_num_devices): Aside from
'CUDA_ERROR_NO_DEVICE', errors during device probing are fatal.
---
include/cuda/cuda.h | 1 +
libgomp/plugin/plugin-gcn.c | 14 ++++++++------
libgomp/plugin/plugin-nvptx.c | 4 +++-
3 files changed, 12 insertions(+), 7 deletions(-)
@@ -57,6 +57,7 @@ typedef enum {
CUDA_ERROR_OUT_OF_MEMORY = 2,
CUDA_ERROR_NOT_INITIALIZED = 3,
CUDA_ERROR_DEINITIALIZED = 4,
+ CUDA_ERROR_NO_DEVICE = 100,
CUDA_ERROR_INVALID_CONTEXT = 201,
CUDA_ERROR_INVALID_HANDLE = 400,
CUDA_ERROR_NOT_FOUND = 500,
@@ -1511,10 +1511,12 @@ assign_agent_ids (hsa_agent_t agent, void *data)
}
/* Initialize hsa_context if it has not already been done.
- Return TRUE on success. */
+ If !PROBE: returns TRUE on success.
+ If PROBE: returns TRUE on success or if the plugin/device shall be silently
+ ignored, and otherwise emits an error and returns FALSE. */
static bool
-init_hsa_context (void)
+init_hsa_context (bool probe)
{
hsa_status_t status;
int agent_index = 0;
@@ -1529,7 +1531,7 @@ init_hsa_context (void)
GOMP_PLUGIN_fatal ("%s\n", msg);
else
GCN_WARNING ("%s\n", msg);
- return false;
+ return probe ? true : false;
}
status = hsa_fns.hsa_init_fn ();
if (status != HSA_STATUS_SUCCESS)
@@ -3321,8 +3323,8 @@ GOMP_OFFLOAD_version (void)
int
GOMP_OFFLOAD_get_num_devices (unsigned int omp_requires_mask)
{
- if (!init_hsa_context ())
- return 0;
+ if (!init_hsa_context (true))
+ exit (EXIT_FAILURE);
/* Return -1 if no omp_requires_mask cannot be fulfilled but
devices were present. */
if (hsa_context.agent_count > 0
@@ -3339,7 +3341,7 @@ GOMP_OFFLOAD_get_num_devices (unsigned int omp_requires_mask)
bool
GOMP_OFFLOAD_init_device (int n)
{
- if (!init_hsa_context ())
+ if (!init_hsa_context (false))
return false;
if (n >= hsa_context.agent_count)
{
@@ -622,12 +622,14 @@ nvptx_get_num_devices (void)
CUresult r = CUDA_CALL_NOCHECK (cuInit, 0);
/* This is not an error: e.g. we may have CUDA libraries installed but
no devices available. */
- if (r != CUDA_SUCCESS)
+ if (r == CUDA_ERROR_NO_DEVICE)
{
GOMP_PLUGIN_debug (0, "Disabling nvptx offloading; cuInit: %s\n",
cuda_error (r));
return 0;
}
+ else if (r != CUDA_SUCCESS)
+ GOMP_PLUGIN_fatal ("cuInit error: %s", cuda_error (r));
}
CUDA_CALL_ASSERT (cuDeviceGetCount, &n);
--
2.34.1