Docs @ f2ef560

Author: MFC Action

documentation/md_gpuDebugging.html

@@ -165,59 +165,47 @@ <h2><a class="anchor" id="autotoc_md106"></a>
 Cray General Options</h2>
 <div class="fragment"><div class="line">CRAY_ACC_DEBUG: 0 (off), 1, 2, 3 (very noisy)</div>
 </div><!-- fragment --><ul>
-<li>Dumps a time-stamped log line (<code>"ACC: ...&lt;/tt&gt;) for every allocation, data transfer, kernel launch, wait, etc. Great first stop when "nothing seems to run on the GPU.</code></li>
-<li><code>Outputs on STDERR by default. Can be changed by setting <code>CRAY_ACC_DEBUG_FILE</code>.<ul>
+<li>Dumps a time-stamped log line (<code>ACC: ...</code>) for every allocation, data transfer, kernel launch, wait, etc. Great first stop when "nothing seems to run on the GPU".</li>
+<li>Outputs on STDERR by default. Can be changed by setting <code>CRAY_ACC_DEBUG_FILE</code>.<ul>
 <li>Recognizes <code>stderr</code>, <code>stdout</code>, and <code>process</code>.</li>
 <li><code>process</code> automatically generates a new file based on <code>pid</code> (each MPI process will have a different file)</li>
 </ul>
-</code></li>
-<li><code>While this environment variable specifies ACC, it can be used for both OpenACC and OpenMP</code></li>
+</li>
+<li>While this environment variable specifies ACC, it can be used for both OpenACC and OpenMP</li>
 </ul>
-<p><code></p><div class="fragment"><div class="line">CRAY_ACC_FORCE_EARLY_INIT=1</div>
+<div class="fragment"><div class="line">CRAY_ACC_FORCE_EARLY_INIT=1</div>
 </div><!-- fragment --><ul>
 <li>Force full GPU initialization at program start so you can see start-up hangs immediately</li>
 <li>Default behavior without an environment variable is to defer initialization on first use</li>
 <li>Device initialization includes initializing the GPU vendor’s low-level device runtime library (e.g., libcuda for NVIDIA GPUs) and establishing all necessary software contexts for interacting with the device</li>
 </ul>
-<p></code></p>
-<p><code></code></p>
 <h2><a class="anchor" id="autotoc_md107"></a>
-<code>Cray OpenACC Options</code></h2>
-<p><code></code></p>
-<p><code></p><div class="fragment"><div class="line">CRAY_ACC_PRESENT_DUMP_SAVE_NAMES=1</div>
+Cray OpenACC Options</h2>
+<div class="fragment"><div class="line">CRAY_ACC_PRESENT_DUMP_SAVE_NAMES=1</div>
 </div><!-- fragment --><ul>
 <li>Will cause <code>acc_present_dump()</code> to output variable names and file locations in addition to variable mappings</li>
 <li>Add <code>acc_present_dump()</code> around hotspots to help find problems with data movements<ul>
 <li>Helps more if adding <code>CRAY_ACC_DEBUG</code> environment variable</li>
 </ul>
 </li>
 </ul>
-<p></code></p>
-<p><code></code></p>
 <h1><a class="anchor" id="autotoc_md108"></a>
-<code>NVHPC Compiler Options</code></h1>
-<p><code></code></p>
-<p><code></code></p>
+NVHPC Compiler Options</h1>
 <h2><a class="anchor" id="autotoc_md109"></a>
-<code>NVHPC General Options</code></h2>
-<p><code></code></p>
-<p><code></p><div class="fragment"><div class="line">STATIC_RANDOM_SEED=1</div>
+NVHPC General Options</h2>
+<div class="fragment"><div class="line">STATIC_RANDOM_SEED=1</div>
 </div><!-- fragment --><ul>
 <li>Forces the seed returned by <code>RANDOM_SEED</code> to be constant, so it generates the same sequence of random numbers</li>
 <li>Useful for testing issues with randomized data</li>
 </ul>
-<p></code></p>
-<p><code></p><div class="fragment"><div class="line">NVCOMPILER_TERM=option[,option]</div>
+<div class="fragment"><div class="line">NVCOMPILER_TERM=option[,option]</div>
 </div><!-- fragment --><ul>
 <li><code>[no]debug</code>: Enables/disables just-in-time debugging (debugging invoked on error)</li>
 <li><code>[no]trace</code>: Enables/disables stack traceback on error</li>
 </ul>
-<p></code></p>
-<p><code></code></p>
 <h2><a class="anchor" id="autotoc_md110"></a>
-<code>NVHPC OpenACC Options</code></h2>
-<p><code></code></p>
-<p><code></p><div class="fragment"><div class="line">NVCOMPILER_ACC_NOTIFY= &lt;bitmask&gt;</div>
+NVHPC OpenACC Options</h2>
+<div class="fragment"><div class="line">NVCOMPILER_ACC_NOTIFY= &lt;bitmask&gt;</div>
 </div><!-- fragment --><ul>
 <li>Assign the environment variable to a bitmask to print out information to stderr for the following<ul>
 <li>kernel launches: 1</li>
@@ -229,15 +217,13 @@ <h2><a class="anchor" id="autotoc_md110"></a>
 </li>
 <li>1 (kernels only) is the usual first step.3 (kernels + copies) is great for "why is it so slow?"</li>
 </ul>
-<p></code></p>
-<p><code></p><div class="fragment"><div class="line">NVCOMPILER_ACC_TIME=1</div>
+<div class="fragment"><div class="line">NVCOMPILER_ACC_TIME=1</div>
 </div><!-- fragment --><ul>
 <li>Lightweight profiler</li>
 <li>prints a tidy end-of-run table with per-region and per-kernel times and bytes moved</li>
 <li>Do not use with CUDA profiler at the same time</li>
 </ul>
-<p></code></p>
-<p><code></p><div class="fragment"><div class="line">NVCOMPILER_ACC_DEBUG=1</div>
+<div class="fragment"><div class="line">NVCOMPILER_ACC_DEBUG=1</div>
 </div><!-- fragment --><ul>
 <li>Spews everything the runtime sees: host/device addresses, mapping events, present-table look-ups, etc.</li>
 <li>Great for "partially present" or "pointer went missing" errors.</li>
@@ -246,19 +232,15 @@ <h2><a class="anchor" id="autotoc_md110"></a>
 </ul>
 </li>
 </ul>
-<p></code></p>
-<p><code></code></p>
 <h2><a class="anchor" id="autotoc_md111"></a>
-<code>NVHPC OpenMP Options</code></h2>
-<p><code></code></p>
-<p><code></p><div class="fragment"><div class="line">LIBOMPTARGET_PROFILE=run.json</div>
+NVHPC OpenMP Options</h2>
+<div class="fragment"><div class="line">LIBOMPTARGET_PROFILE=run.json</div>
 </div><!-- fragment --><ul>
 <li>Emits a Chrome-trace (JSON) timeline you can open in chrome://tracing or Speedscope</li>
 <li>Great lightweight profiler when Nsight is overkill.</li>
 <li>Granularity in µs via <code>LIBOMPTARGET_PROFILE_GRANULARITY</code> (default 500).</li>
 </ul>
-<p></code></p>
-<p><code></p><div class="fragment"><div class="line">LIBOMPTARGET_INFO=&lt;bitmask&gt;</div>
+<div class="fragment"><div class="line">LIBOMPTARGET_INFO=&lt;bitmask&gt;</div>
 </div><!-- fragment --><ul>
 <li>Prints out different types of runtime information</li>
 <li>Human-readable log of data-mapping inserts/updates, kernel launches, copies, waits.</li>
@@ -273,44 +255,37 @@ <h2><a class="anchor" id="autotoc_md111"></a>
 </ul>
 </li>
 </ul>
-<p></code></p>
-<p><code></p><div class="fragment"><div class="line">LIBOMPTARGET_DEBUG=1</div>
+<div class="fragment"><div class="line">LIBOMPTARGET_DEBUG=1</div>
 </div><!-- fragment --><ul>
 <li>Developer-level trace (host-side)</li>
 <li>Much noisier than <code>INFO</code></li>
 <li>Only works if the runtime was built with <code>-DOMPTARGET_DEBUG</code>.</li>
 </ul>
-<p></code></p>
-<p><code></p><div class="fragment"><div class="line">LIBOMPTARGET_JIT_OPT_LEVEL=-O{0,1,2,3}</div>
+<div class="fragment"><div class="line">LIBOMPTARGET_JIT_OPT_LEVEL=-O{0,1,2,3}</div>
 </div><!-- fragment --><ul>
 <li>This environment variable can be used to change the optimization pipeline used to optimize the embedded device code as part of the device JIT.</li>
 <li>The value corresponds to the <code>-O{0,1,2,3}</code> command line argument passed to clang.</li>
 </ul>
-<p></code></p>
-<p><code></p><div class="fragment"><div class="line">LIBOMPTARGET_JIT_SKIP_OPT=1</div>
+<div class="fragment"><div class="line">LIBOMPTARGET_JIT_SKIP_OPT=1</div>
 </div><!-- fragment --><ul>
 <li>This environment variable can be used to skip the optimization pipeline during JIT compilation.</li>
 <li>If set, the image will only be passed through the backend.</li>
 <li>The backend is invoked with the <code>LIBOMPTARGET_JIT_OPT_LEVEL</code> flag.</li>
 </ul>
-<p></code></p>
-<p><code></code></p>
 <h1><a class="anchor" id="autotoc_md112"></a>
-<code>Compiler Documentation</code></h1>
-<p><code></code></p>
-<p><code></p><ul>
+Compiler Documentation</h1>
+<ul>
 <li><a href="https://cpe.ext.hpe.com/docs/24.11/cce/man7/intro_openmp.7.html#environment-variables">Cray &amp; OpenMP Docs</a></li>
 <li><a href="https://cpe.ext.hpe.com/docs/24.11/cce/man7/intro_openacc.7.html#environment-variables">Cray &amp; OpenACC Docs</a></li>
 <li><a href="https://docs.nvidia.com/hpc-sdk/compilers/hpc-compilers-user-guide/index.html?highlight=NVCOMPILER_#environment-variables">NVHPC &amp; OpenACC Docs</a></li>
 <li><a href="https://docs.nvidia.com/hpc-sdk/compilers/hpc-compilers-user-guide/index.html?highlight=NVCOMPILER_#id2">NVHPC &amp; OpenMP Docs</a></li>
-<li>[LLVM &amp; OpenMP Docs] (<a href="https://openmp.llvm.org/design/Runtimes.html">https://openmp.llvm.org/design/Runtimes.html</a>)<ul>
+<li><a href="https://openmp.llvm.org/design/Runtimes.html">LLVM &amp; OpenMP Docs</a><ul>
 <li>NVHPC is built on top of LLVM</li>
 </ul>
 </li>
 <li><a href="https://www.openmp.org/spec-html/5.1/openmp.html">OpenMP Docs</a></li>
 <li><a href="https://www.openacc.org/sites/default/files/inline-files/OpenACC.2.7.pdf">OpenACC Docs</a> </li>
 </ul>
-<p></code></p>
 </div></div><!-- contents -->
 </div><!-- PageDoc -->
 </div><!-- doc-content -->