SANITIZER API

v2024.1.1 | March 2024

Reference Manual
 TABLE OF CONTENTS

Chapter 1. Introduction.........................................................................................1
1.1. Overview................................................................................................... 1
Chapter 2. Usage................................................................................................. 2
2.1. Compatibility and Requirements....................................................................... 2
2.2. Callback API............................................................................................... 2
2.2.1. Driver and Runtime API Callbacks................................................................ 3
2.2.2. Resource Callbacks.................................................................................. 4
2.2.3. Synchronization Callbacks..........................................................................4
2.2.4. Launch Callbacks.................................................................................... 5
2.2.5. Memcpy Callbacks................................................................................... 5
2.2.6. Memset Callbacks................................................................................... 5
2.2.7. Batch Memory Operations Callbacks............................................................. 5
2.3. Patching API............................................................................................... 5
2.3.1. Writing a Patch...................................................................................... 6
2.3.2. Insert a Patch........................................................................................ 6
2.4. Memory API................................................................................................ 7
2.5. Special cases.............................................................................................. 8
2.5.1. Device graph launch................................................................................ 8
Chapter 3. Limitations.......................................................................................... 9

www.nvidia.com
Sanitizer API v2024.1.1 | ii
Chapter 1.
INTRODUCTION

1.1. Overview
The Compute Sanitizer API enables the creation of sanitizing and tracing tools that
target CUDA applications. Examples of such tools are memory and race condition
checkers. The Compute Sanitizer API is composed of three APIs: the callback API, the
patching API and the memory API. It is delivered as a dynamic library on supported
platforms.

www.nvidia.com
Sanitizer API v2024.1.1 | 1
Chapter 2.
USAGE

2.1. Compatibility and Requirements

The Compute Sanitizer tools require CUDA 11.0 or newer.
The Compute Sanitizer API requires CUDA 10.1 or newer. Compute Sanitizer API calls
will fail with SANITIZER_ERROR_NOT_INITIALIZED if the CUDA driver version is not
compatible with the Compute Sanitizer version.

2.2. Callback API

The Compute Sanitizer Callback API allows you to register a callback into user code. The
callback is invoked when the application calls a CUDA runtime or driver function, or
when certain events occur in the CUDA driver. The following terminology is used by the
Callback API.
‣ Callback domain: Callbacks are grouped into domains to make it easier to associate
callback functions with groups of related CUDA functions or events. The following
callback domains are defined by Sanitizer_CallbackDomain.
1. CUDA driver functions
2. CUDA runtime functions
3. CUDA resource tracking
4. CUDA synchronization notification
5. CUDA grid launches
6. CUDA memcpy operations
7. CUDA memset operations
8. CUDA batch memory operations
‣ Callback ID: Each callback is given a unique ID within the corresponding callback
domain in order to identify it within the callback function. The CUDA driver
API IDs are defined in sanitizer_driver_cbid.h and the CUDA runtime
API IDs are defined in sanitizer_runtime_cbid.h. Other callback IDs are

www.nvidia.com
Sanitizer API v2024.1.1 | 2
 Usage

defined in sanitizer_callbacks.h. All of these headers are included as part of

sanitizer.h.
‣ Callback Function: The callback function must be of the type
Sanitizer_CallbackFunc. This function type has two arguments that specify the
callback: the domain and the ID that identifies why the callback is occurring. The
type also has a cbdata argument that is used to pass data specific to the callback.
‣ Subscriber: A subscriber is used to associate each of the callback functions with one
or more CUDA API functions. There can be at most one subscriber initialized with
sanitizerSubscribe at any time. Before initializing a new subscriber, the existing
one must be finalized with sanitizerUnsubscribe .
The subscriber should be initialized prior to making any CUDA API call to ensure
correctness of the reported data.
Each callback domain is described in detail below. Unless explicitly stated, it is not
supported to call any CUDA runtime or driver API from within a callback function.
Doing so may cause the application to hang. However, it is supported to call Compute
Sanitizer Memory APIs from within callback functions.

2.2.1. Driver and Runtime API Callbacks

Using the Callback API with the SANITIZER_CB_DOMAIN_DRIVER_API or
SANITIZER_CB_DOMAIN_RUNTIME_API domains, a callback function can be associated
with one or more CUDA API functions. When those CUDA functions are called in the
application, the callback function is invoked as well. For these domains, the cbdata
argument to the callback function will be of the type Sanitizer_CallbackData.
You can call cudaDeviceSynchronize, cudaStreamSynchronize,
cuCtxSynchronize and cuStreamSynchronize from within a driver or runtime API
callback function.
The following code shows a typical sequence used to associate a callback function with
one or more CUDA API functions. For simplicity, error checking code was removed.

Sanitizer_SubscriberHandle handle;
MyDataStruct *my_data = ...;
...
sanitizerSubscribe(&handle, my_callback, my_data);
sanitizerEnableDomain(1, handle, SANITIZER_CB_DOMAIN_RUNTIME_API);

First, sanitizerSubscribe is used to initialize a subscriber with the my_callback

callback function. Next, sanitizerEnableDomain is used to associate that callback
with all the CUDA runtime functions. Using this code sequence will cause my_callback
to be called twice each time any of the CUDA runtime API functions are invoked,
once on entry to the CUDA function and once just before the CUDA function
exits. Compute Sanitizer callback API functions sanitizerEnableCallback and
sanitizerEnableAllDomains can also be used to associate CUDA API functions with
a callback.

www.nvidia.com
Sanitizer API v2024.1.1 | 3
 Usage

The following code shows a typical callback function.

void SANITIZERAPI
my_callback(void *userdata,
Sanitizer_CallbackDomain domain,
Sanitizer_CallbackId cbid,
const void *cbdata)
{
const Sanitizer_CallbackData *cbInfo = (Sanitizer_CallbackData *)cbdata;
MyDataStruct *my_data = (MyDataStruct *)userdata;

if ((domain == SANITIZER_CB_DOMAIN_RUNTIME_API) &&

(cbid == SANITIZER_RUNTIME_TRACE_CBID_cudaMemcpy_v3020) &&
(cbInfo->callbackSite == SANITIZER_API_ENTER))
{
cudaMemcpy_v3020_params *funcParams = (cudaMemcpy_v3020_params *)
(cbInfo->functionParams);
size_t count = funcParams->count;
enum cudaMemcpyKind kind = funcParams->kind
...
}
...
}

In the callback function, Sanitizer_CallbackDomain and Sanitizer_CallbackId

parameters can be used to determine which CUDA API function invocation is triggering
this callback. In the example above, we are checking for the CUDA runtime cudaMemcpy
function. The cbdata parameter holds a structure of useful information that can
be used within the callback. In this case, we use the callbackSite member of
the structure to detect that the callback is occurring on entry to cudaMemcpy, and
we use the functionParams member to access the parameters to cudaMemcpy.
To access the parameters, we first cast functionParams to a structure type
corresponding to the cudaMemcpy function. These parameter structures are contained in
generated_cuda_runtime_api_meta.h, generated_cuda_meta.h and a number of
other files.

2.2.2. Resource Callbacks

Using the Callback API with the SANITIZER_CB_DOMAIN_RESOURCE domain, a callback
function can be associated with some CUDA resource creation and destruction events.
For example, when a CUDA context is created, the callback function is invoked with a
callback ID equal to SANITIZER_CBID_RESOURCE_CONTEXT_CREATED. For this domain,
the cbdata argument is one of the following types:
‣ Sanitizer_ResourceContextData for CUDA context creation and destruction
‣ Sanitizer_ResourceStreamData for CUDA stream creation and destruction
‣ Sanitizer_ResourceModuleData for CUDA module load and unload
‣ Sanitizer_ResourceMemoryData for CUDA memory allocation and de-allocation

2.2.3. Synchronization Callbacks

Using the Callback API with the SANITIZER_CB_DOMAIN_SYNCHRONIZE domain, a
callback function can be associated with CUDA context and stream synchronizations.
For example, when a CUDA context is synchronized, the callback function is invoked

www.nvidia.com
Sanitizer API v2024.1.1 | 4
 Usage

with a callback ID equal to SANITIZER_CBID_SYNCHRONIZE_CONTEXT_SYNCHRONIZED.

For this domain, the cbdata argument is of the type Sanitizer_SynchronizeData.

2.2.4. Launch Callbacks

Using the Callback API with the SANITIZER_CB_DOMAIN_LAUNCH domain, a callback
function can be associated with CUDA kernel launches. For example, when a CUDA
kernel launch has started, the callback function is invoked with a callback ID equal to
SANITIZER_CBID_LAUNCH_BEGIN. For this domain, the cbdata argument is of the type
Sanitizer_LaunchData.

2.2.5. Memcpy Callbacks

Using the Callback API with the SANITIZER_CB_DOMAIN_MEMCPY domain, a callback
function can be associated with CUDA memcpy operations. For example, when a
cudaMemcpy API function is called, the callback function is invoked with a callback ID
equal to SANITIZER_CBID_MEMCPY_STARTING. For this domain, the cbdata argument
is of the type Sanitizer_MemcpyData.

2.2.6. Memset Callbacks

Using the Callback API with the SANITIZER_CB_DOMAIN_MEMSET domain, a callback
function can be associated with CUDA memset operations. For example, when a
cudaMemset API function is called, the callback function is invoked with a callback ID
equal to SANITIZER_CBID_MEMSET_STARTING. For this domain, the cbdata argument
is of the type Sanitizer_MemsetData.

2.2.7. Batch Memory Operations Callbacks

Using the Callback API with the SANITIZER_CB_DOMAIN_BATCH_MEMOP domain, a
callback function can be associated with CUDA batch memory operations. For example,
when a cuStreamWriteValue API function is called, the callback function is invoked
with a callback ID equal to SANITIZER_CBID_BATCH_MEMOP_WRITE. For this domain,
the cbdata argument is of the type Sanitizer_BatchMemopData.

2.3. Patching API

The Compute Sanitizer Patching API allows you to load patch functions and insert them
into user code. Patch functions will be invoked when the application's CUDA code
executes certain instructions or calls certain CUDA device functions. The following
terminology is used by the Patching API:
‣ Instruction ID: Each patchable event is given a unique ID than can be passed to
patching API functions to specify that these events should be patched. Instruction
IDs are defined by Sanitizer_InstructionId.
‣ Instrumentation point: A location in the original CUDA code that is being
instrumented by the Compute Sanitizer API. Upon execution, the user code path is

www.nvidia.com
Sanitizer API v2024.1.1 | 5
 Usage

modified so that a patch gets executed either before or after the patched event. All
patches are executed prior to the event, with the exception of device-side malloc.
‣ Patch: A CUDA __device__ function that the Compute Sanitizer will insert
into another existing CUDA code. Patch function signatures must match the one
expected by the API (see below for the expected signature types).

2.3.1. Writing a Patch

The patch must follow the function signature required by the Compute Sanitizer
API for a given instruction ID. The mapping of instruction ID to function
signature is documented in the comments of Sanitizer_InstructionId in
sanitizer_patching.h. For instance, if we wish to patch a memory access using
the instruction ID SANITIZER_INSTRUCTION_MEMORY_ACCESS, we need to use the
SanitizerCallbackMemoryAccess type.

extern "C" __device__

SanitizerPatchResult SANITIZERAPI my_memory_access_callback(
void* userdata,
uint64_t pc,
void* ptr,
uint32_t accessSize,
uint32_t flags)
{
MyDeviceDataStruct *my_data = (MyDeviceDataStruct *)userdata

if ((flags & SANITIZER_MEMORY_DEVICE_FLAG_WRITE) != 0)

// log write
else
// log read

return SANITIZER_PATCH_SUCCESS;
}

In this patch, we log write and read accesses to a structure we allocated previously.
extern "C" ensures that the patch name will not be mangled, allowing us to use its
name as a string directly in calls to sanitizerPatchInstructions (see below).
There can be multiple patches defined in a single CUDA file. This file must then be
compiled using the following nvcc options:

$ nvcc --cubin --compile-as-tools-patch MySanitizerPatches.cu -o

MySanitizerPatches.cubin

The --cubin option can be replaced by --fatbin if a fatbin is preferred over a cubin as
the output file.

2.3.2. Insert a Patch

Once the patch has been generated, it can be inserted into user code by using the
following procedure:
1. Load the patch. There are two APIs used to load the patch:
sanitizerAddPatchesFromFile and sanitizerAddPatches. They use the same
input format as cuModuleLoad and cuModuleLoadData, respectively.

www.nvidia.com
Sanitizer API v2024.1.1 | 6
 Usage

2. Select which instructions to patch by using the sanitizerPatchInstructions

API.
3. Patch user code by using the sanitizerPatchModule API.
4. Optionally, set the callback data for patches by using the
sanitizerSetCallbackData API.
The following code shows a typical sequence using these APIs. For simplicity, error
checking was removed.

CUcontext ctx = ... // current CUDA context

sanitizerAddPatchesFromFile("MySanitizerPatches.cubin", ctx);

CUmodule module = ... // module containing the user code

sanitizerPatchInstructions(SANITIZER_INSTRUCTION_MEMORY_ACCESS, module,
"my_memory_access_callback");

sanitizerPatchModule(module);

MyDeviceDataTracker *deviceDataTracker;
cudaMalloc(&deviceDataTracker, sizeof(*deviceDataTracker));

CUfunction function = ... // kernel to be launched for which we want to set the
callbackdata for the patches
sanitizerSetCallbackData(function, deviceDataTracker);

All subsequent launches using code from this CUDA module will be instrumented and
my_memory_access_callback will be invoked before every memory access. However,
the callback data is only set for all subsequent launches of the given kernel. An easy way
to have a kernel CUfunction, is through the Sanitizer launch callbacks. Instrumentation
can be removed by using the sanitizerUnpatchModule API.

2.4. Memory API

The Compute Sanitizer Memory API provides replacement functions for the CUDA
Memory API that can be safely called from within Compute Sanitizer callbacks.
‣ sanitizerAlloc is a replacement for cudaMalloc.
‣ sanitizerFree is a replacement for cudaFree.
‣ sanitizerMemcpyHostToDeviceAsync is a replacement for cudaMemcpyAsync
for host-to-device copies.
‣ sanitizerMemcpyDeviceToHost is a replacement for cudaMemcpy for device-to-
host copies.
‣ sanitizerMemset is a replacement for cudaMemset.
These functions can also be called in normal user code, where they can be mixed with
the CUDA API. For instance, memory allocated with sanitizerAlloc can be freed
with cudaFree. However, since only CUDA API calls will cause callbacks to be invoked,
this can lead to an incoherent tracking state and should be avoided.

www.nvidia.com
Sanitizer API v2024.1.1 | 7
 Usage

2.5. Special cases

In some specific scenarios, the Compute Sanitizer API may behave differently from the
general cases. These scenarios are listed in this section.

2.5.1. Device graph launch

When the application instrumented by the Compute Sanitizer API makes use of device
launched CUDA graphs, the following behaviors are to be taken into account:
‣ No callbacks will be called from the host when a CUDA graph is launched from
the device. However, graph launch-related callbacks are still called when a device
launchable graph is launched from the host, or uploaded.
‣ Patching API issued callbacks on the device when inside a device launched CUDA
graph may have a userdata from a different node if different userdata pointers were
set per node.
To compensate these limitations, it is possible to set a device launched graph specific
data using sanitizerSetDeviceGraphData. It can be called during the launch on the
host of a device launchable graph or of a graph containing device graph launches. The
data set with sanitizerSetDeviceGraphData will then be retrievable from any device
graph launched from this host launched graph. To retrieve this data from a patching API
device callback, the following code can be used:

void* userdata = **((void***)cudaGetCurrentGraphExec());

It is however recommended to perform NULL checks at each dereferencing.

www.nvidia.com
Sanitizer API v2024.1.1 | 8
Chapter 3.
LIMITATIONS

No known issues at this time.

www.nvidia.com
Sanitizer API v2024.1.1 | 9
Notice
ALL NVIDIA DESIGN SPECIFICATIONS, REFERENCE BOARDS, FILES, DRAWINGS,
DIAGNOSTICS, LISTS, AND OTHER DOCUMENTS (TOGETHER AND SEPARATELY,
"MATERIALS") ARE BEING PROVIDED "AS IS." NVIDIA MAKES NO WARRANTIES,
EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO THE
MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF
NONINFRINGEMENT, MERCHANTABILITY, AND FITNESS FOR A PARTICULAR
PURPOSE.
Information furnished is believed to be accurate and reliable. However, NVIDIA
Corporation assumes no responsibility for the consequences of use of such
information or for any infringement of patents or other rights of third parties
that may result from its use. No license is granted by implication of otherwise
under any patent rights of NVIDIA Corporation. Specifications mentioned in this
publication are subject to change without notice. This publication supersedes and
replaces all other information previously supplied. NVIDIA Corporation products
are not authorized as critical components in life support devices or systems
without express written approval of NVIDIA Corporation.

Trademarks
NVIDIA and the NVIDIA logo are trademarks or registered trademarks of NVIDIA
Corporation in the U.S. and other countries. Other company and product names
may be trademarks of the respective companies with which they are associated.

Copyright
© 2019-2024 NVIDIA Corporation and affiliates. All rights reserved.

This product includes software developed by the Syncro Soft SRL (http://
www.sync.ro/).

www.nvidia.com