My blog site. 2025-12-22T00:00:00+00:00 https://johnmaxrin.github.io/posts/feed.xml 2025-12-22T00:00:00+00:00 2025-12-22T00:00:00+00:00 https://johnmaxrin.github.io/posts/hipc-experience/ <p>Attending my first academic conference was a milestone moment for me, both professionally and personally. I had the opportunity to attend HiPC 25, which took place at the The Westin Hyderabad Mindspace from December 17 to December 20. <img src="/img/hipc_3.jpg" alt="HiPC Conference at Westin Hyderabad" /></p> <p>From the very first day, the experience felt energizing. I met people from all over the world, each bringing their own background, research interests, and perspectives. Conversations flowed easily, from deep technical discussions to casual chats about universities, advisors, and life as a researcher. Being in an environment where everyone shared at least one common thread made networking feel natural rather than forced.</p> <p>Outside the conference halls, I also got the chance to roam around Hyderabad, completely carefree. I visited several places across the city, explored the local culture, and met many people along the way. The balance between academic intensity and personal exploration made the experience even more memorable. <img src="/img/hipc_2.jpeg" alt="HiPC Conference at Westin Hyderabad" /></p> <h2 id="expectations-vs-reality">Expectations vs Reality<a class="zola-anchor" href="#expectations-vs-reality" aria-label="Anchor link for: expectations-vs-reality" style="visibility: hidden;"></a> </h2> <p>Before attending, I carried a few assumptions about conferences.</p> <p>I expected conferences to be boring. In reality, it was far from that. If you wanted, you could attend technical sessions all day. If not, you could simply talk to people. Everyone at the conference shared something in common, whether it was their research area, their alma mater, or similar academic journeys.</p> <p><img src="/img/hipc_alma.jpeg" alt="HiPC Conference at Westin Hyderabad" /></p> <p>I also assumed people would be rude, unapproachable, or pretend to know everything. The reality was the exact opposite. Almost everyone I interacted with was humble, welcoming, and genuinely open to exchanging ideas and feedback.</p> <p>Even small things surprised me. I thought the food would be bad, but it turned out to be excellent. The climate was pleasant too, which made moving around the city and the venue comfortable throughout the event.</p> <h2 id="first-presentation-experience">First Presentation Experience<a class="zola-anchor" href="#first-presentation-experience" aria-label="Anchor link for: first-presentation-experience" style="visibility: hidden;"></a> </h2> <p>One of the most defining moments of the conference was my presentation. This was the first time I presented my work to a wider audience. <img src="/img/hipc_4.jpg" alt="HiPC Conference at Westin Hyderabad" /></p> <p>Condensing months or even years of work into a 15 minute presentation is challenging. Choosing the right words at the right moment matters a lot in such time constrained situations. I relied heavily on presenter notes, along with my own explanations throughout the talk, to stay on track.</p> <p>Rupesh often says that the real way to judge a talk is by the number of hands raised afterward. By that measure, I walked away very happy. After the presentation, several people approached me to discuss the paper further. Those interactions were deeply validating.</p> <h2 id="lessons-learned">Lessons Learned<a class="zola-anchor" href="#lessons-learned" aria-label="Anchor link for: lessons-learned" style="visibility: hidden;"></a> </h2> <p>This first conference taught me several practical lessons that I will carry forward.</p> <ul> <li> <p>Pack as little as possible. Traveling light makes it much easier to move around and stay flexible.</p> </li> <li> <p>For short presentations, presenter notes are extremely helpful. They act as a safety net without breaking the flow of the talk.</p> </li> <li> <p>Most importantly, talk to as many people as you can. It costs nothing, and the conversations you have can be far more valuable than you expect.</p> </li> </ul> 2025-11-03T00:00:00+00:00 2025-11-03T00:00:00+00:00 https://johnmaxrin.github.io/posts/dhir/ <h1 id="the-challenge">The Challenge<a class="zola-anchor" href="#the-challenge" aria-label="Anchor link for: the-challenge" style="visibility: hidden;"></a> </h1> <p>Modern computing is increasingly heterogeneous. Your laptop has a CPU and maybe a GPU. Data centers have CPUs, GPUs, TPUs, and specialized accelerators. High-performance computing clusters combine all of these across multiple nodes connected by high-speed networks.</p> <p>Writing programs that efficiently utilize this heterogeneous hardware is incredibly challenging. You need to:</p> <ul> <li>Partition your computation across different devices</li> <li>Express parallelism within each device (threads, CUDA blocks, MPI ranks)</li> <li>Move data between devices efficiently</li> <li>Coordinate execution and handle dependencies</li> </ul> <p>Traditionally, this means writing tangled code mixing MPI, OpenMP, CUDA, and custom communication logic. It's error-prone, hard to maintain, and difficult to port to new hardware configurations.</p> <p>This is what DHIR is solving. DHIR (Distributed Heterogeneous Intermediate Representation) is an MLIR dialect for expressing distributed, heterogeneous computations as tasks that can execute across CPUs, GPUs, and distributed systems.</p> <h1 id="core-concepts">Core Concepts<a class="zola-anchor" href="#core-concepts" aria-label="Anchor link for: core-concepts" style="visibility: hidden;"></a> </h1> <h2 id="tasks">Tasks<a class="zola-anchor" href="#tasks" aria-label="Anchor link for: tasks" style="visibility: hidden;"></a> </h2> <p>Tasks are the fundamental unit of computation in DHIR, analogous to functions in C/C++. Each task:</p> <ul> <li>Encapsulates a specific computation</li> <li>Has explicit inputs and outputs</li> <li>Can be scheduled for execution on different devices</li> <li>Contains parallelism information</li> </ul> <h2 id="schedule">Schedule<a class="zola-anchor" href="#schedule" aria-label="Anchor link for: schedule" style="visibility: hidden;"></a> </h2> <p>Just like the main function.</p> <h2 id="system-config-json"><code>system_config.json</code><a class="zola-anchor" href="#system-config-json" aria-label="Anchor link for: system-config-json" style="visibility: hidden;"></a> </h2> <p>Helps in defining the distributed system that you have. As of now we have</p> <pre class="z-code"><code><span class="z-text z-plain">module attributes {avial.target_devices = [#dlti.target_device_spec&lt;&quot;type&quot; = &quot;node&quot;, &quot;arch&quot; = &quot;x86_64&quot;, &quot;cost&quot; = 1.000000e+00 : f32, &quot;node_id&quot; = &quot;node0&quot;, &quot;gpu_count&quot; = 0 : i32&gt;, #dlti.target_device_spec&lt;&quot;type&quot; = &quot;node&quot;, &quot;arch&quot; = &quot;x86_64&quot;, &quot;cost&quot; = 1.000000e+00 : f32, &quot;node_id&quot; = &quot;node1&quot;, &quot;gpu_count&quot; = 0 : i32&gt;, #dlti.target_device_spec&lt;&quot;type&quot; = &quot;node&quot;, &quot;arch&quot; = &quot;x86_64&quot;, &quot;cost&quot; = 5.000000e-01 : f32, &quot;node_id&quot; = &quot;node2&quot;, &quot;gpu_count&quot; = 1 : i32&gt;]} </span></code></pre> 2025-10-16T00:00:00+00:00 2025-10-16T00:00:00+00:00 https://johnmaxrin.github.io/posts/areytutorial/ <p>Debugging MLIR code can be tricky. The <strong>Arey dialect</strong> helps you instrument your IR with simple operations like printing values or checking assertions making your life much easier when hunting down bugs in your compiler passes.</p> <hr /> <h2 id="what-is-the-arey-dialect">What Is the Arey Dialect?<a class="zola-anchor" href="#what-is-the-arey-dialect" aria-label="Anchor link for: what-is-the-arey-dialect" style="visibility: hidden;"></a> </h2> <p>The Arey dialect provides utility operations such as:</p> <ul> <li><code>arey.print</code> — print runtime values</li> <li><code>arey.print_str</code> — print a constant string</li> <li><code>arey.assert</code> — assert conditions on runtime values</li> </ul> <p>Think of it as <strong>printf debugging for your IR</strong>, A lightweight way to peek into runtime behavior before full lowering.</p> <hr /> <h2 id="example-matrix-multiplication">Example: Matrix Multiplication<a class="zola-anchor" href="#example-matrix-multiplication" aria-label="Anchor link for: example-matrix-multiplication" style="visibility: hidden;"></a> </h2> <p>Suppose your MLIR code looks like this:</p> <pre data-lang="mlir" class="language-mlir z-code"><code class="language-mlir" data-lang="mlir"><span class="z-text z-plain">module { </span><span class="z-text z-plain"> func.func @matmul(%arg0: i32, %arg1: i32, %arg2: i32, </span><span class="z-text z-plain"> %arg3: memref&lt;?x128xf32&gt;, %arg4: memref&lt;?x128xf32&gt;, %arg5: memref&lt;?x128xf32&gt;) { </span><span class="z-text z-plain"> %cst = arith.constant 0.0 : f32 </span><span class="z-text z-plain"> %0 = arith.index_cast %arg0 : i32 to index </span><span class="z-text z-plain"> %1 = arith.index_cast %arg1 : i32 to index </span><span class="z-text z-plain"> %2 = arith.index_cast %arg2 : i32 to index </span><span class="z-text z-plain"> </span><span class="z-text z-plain"> affine.for %i = 0 to 128 { </span><span class="z-text z-plain"> affine.for %j = 0 to 128 { </span><span class="z-text z-plain"> affine.store %cst, %arg5[%i, %j] : memref&lt;?x128xf32&gt; </span><span class="z-text z-plain"> affine.for %k = 0 to 128 { </span><span class="z-text z-plain"> %a = affine.load %arg3[%i, %k] : memref&lt;?x128xf32&gt; </span><span class="z-text z-plain"> %b = affine.load %arg4[%k, %j] : memref&lt;?x128xf32&gt; </span><span class="z-text z-plain"> %p = arith.mulf %a, %b : f32 </span><span class="z-text z-plain"> %c = affine.load %arg5[%i, %j] : memref&lt;?x128xf32&gt; </span><span class="z-text z-plain"> %r = arith.addf %c, %p : f32 </span><span class="z-text z-plain"> affine.store %r, %arg5[%i, %j] : memref&lt;?x128xf32&gt; </span><span class="z-text z-plain"> } </span><span class="z-text z-plain"> } </span><span class="z-text z-plain"> } </span><span class="z-text z-plain"> return </span><span class="z-text z-plain"> } </span><span class="z-text z-plain">} </span></code></pre> <hr /> <h2 id="adding-arey-instrumentation">Adding Arey Instrumentation<a class="zola-anchor" href="#adding-arey-instrumentation" aria-label="Anchor link for: adding-arey-instrumentation" style="visibility: hidden;"></a> </h2> <p>Let’s say you want to check what <code>%2</code> is at runtime, or ensure <code>%arg0</code> equals 1000 or simply want to print <code>Hello World</code> You can easily do that with Arey ops:</p> <pre data-lang="mlir" class="language-mlir z-code"><code class="language-mlir" data-lang="mlir"><span class="z-text z-plain">arey.assert %arg0 : i32 eq 1000 </span><span class="z-text z-plain">arey.print %2 : index </span><span class="z-text z-plain">arey.print_str &quot;Hello World&quot; </span></code></pre> <p>Resulting instrumented IR:</p> <pre data-lang="mlir" class="language-mlir z-code"><code class="language-mlir" data-lang="mlir"><span class="z-text z-plain">module { </span><span class="z-text z-plain"> func.func @matmul(%arg0: i32, %arg1: i32, %arg2: i32, </span><span class="z-text z-plain"> %arg3: memref&lt;?x128xf32&gt;, %arg4: memref&lt;?x128xf32&gt;, %arg5: memref&lt;?x128xf32&gt;) { </span><span class="z-text z-plain"> %cst = arith.constant 0.0 : f32 </span><span class="z-text z-plain"> %0 = arith.index_cast %arg0 : i32 to index </span><span class="z-text z-plain"> %1 = arith.index_cast %arg1 : i32 to index </span><span class="z-text z-plain"> %2 = arith.index_cast %arg2 : i32 to index </span><span class="z-text z-plain"> </span><span class="z-text z-plain"> arey.assert %arg0 : i32 eq 1000 </span><span class="z-text z-plain"> arey.print %2 : index </span><span class="z-text z-plain"> </span><span class="z-text z-plain"> affine.for %i = 0 to 128 { </span><span class="z-text z-plain"> arey.print_str &quot;Hello World&quot; </span><span class="z-text z-plain"> ... </span><span class="z-text z-plain"> } </span><span class="z-text z-plain"> return </span><span class="z-text z-plain"> } </span><span class="z-text z-plain">} </span></code></pre> <hr /> <h2 id="lowering-to-llvm">Lowering to LLVM<a class="zola-anchor" href="#lowering-to-llvm" aria-label="Anchor link for: lowering-to-llvm" style="visibility: hidden;"></a> </h2> <p>Once you’re done instrumenting, simply run the conversion pass:</p> <pre data-lang="bash" class="language-bash z-code"><code class="language-bash" data-lang="bash"><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">--convert-arey-to-llvm</span></span> </span></code></pre> <hr /> <h2 id="why-use-arey">Why Use Arey?<a class="zola-anchor" href="#why-use-arey" aria-label="Anchor link for: why-use-arey" style="visibility: hidden;"></a> </h2> <ul> <li>Easy to debug MLIR at runtime</li> <li>No need to hand-write printf lowering</li> <li>Fully compatible with LLVM lowering</li> <li>Great for teaching, research, and debugging custom passes</li> </ul> <hr /> <h2 id="tl-dr">TL;DR<a class="zola-anchor" href="#tl-dr" aria-label="Anchor link for: tl-dr" style="visibility: hidden;"></a> </h2> <p>The <strong>Arey dialect</strong> is your IR-level debugging toolkit. Add print statements, assertions, or custom messages directly into MLIR and see them reflected in your lowered LLVM IR.</p> <blockquote> <p>Think of Arey as your IR’s <strong>printf debugger</strong> but cleaner and compiler-aware.</p> </blockquote> <h2 id="get-arey-dialect">Get <a rel="nofollow noreferrer" href="https://github.com/johnmaxrin/arey/releases/tag/v0.1.0">Arey Dialect</a><a class="zola-anchor" href="#get-arey-dialect" aria-label="Anchor link for: get-arey-dialect" style="visibility: hidden;"></a> </h2> 2025-09-23T00:00:00+00:00 2025-09-23T00:00:00+00:00 https://johnmaxrin.github.io/posts/fully-software-defined-cpu/ <p>I’ve always wondered: why do we build so many complex features directly into hardware, when software could handle them?</p> <p>Take <strong>register renaming</strong> as an example. Modern CPUs add special circuits to handle name conflicts between registers during execution. But what if our compilers were smart enough to generate code that avoided those conflicts in the first place? We wouldn’t need extra hardware just to rename registers.</p> <h3 id="why-hardware-became-smart">Why hardware became "Smart"<a class="zola-anchor" href="#why-hardware-became-smart" aria-label="Anchor link for: why-hardware-became-smart" style="visibility: hidden;"></a> </h3> <p>Over time, CPUs evolved to handle tricky problems automatically:</p> <ul> <li><strong>Out-of-order execution</strong> lets hardware rearrange instructions to keep pipelines full.</li> <li><strong>Branch prediction</strong> tries to guess the future to avoid stalls.</li> <li><strong>Caches</strong> hide memory latency.</li> </ul> <p>All of this makes CPUs faster, but also incredibly complex and power-hungry.</p> <h3 id="what-if-hardware-stayed-simple">What if hardware stayed simple?<a class="zola-anchor" href="#what-if-hardware-stayed-simple" aria-label="Anchor link for: what-if-hardware-stayed-simple" style="visibility: hidden;"></a> </h3> <p>Now imagine flipping this around: the CPU is <strong>dumb but fast</strong>, and the <strong>software is smart</strong>.</p> <p>This isn’t just a fantasy. There are real-world hints of it:</p> <ul> <li><strong>VLIW (Very Long Instruction Word)</strong> machines, like Intel’s Itanium, tried moving scheduling decisions to the compiler.</li> <li><strong>Interpreters and JIT compilers</strong> (like the JVM or JavaScript engines) already generate optimized code on the fly, sometimes beating hardware predictions.</li> <li><strong>Software-managed caches</strong> exist in some GPUs and embedded processors, where programmers explicitly move data in and out of scratchpad memory.</li> </ul> <p>As ISAs grew more complex, their interactions and cost models became very hard to manage. This is one big reason why attempts like VLIW struggled to gain wide adoption. <a rel="nofollow noreferrer" href="https://en.wikipedia.org/wiki/Itanium">Itanium</a>, for example, failed because writing compilers that could make VLIW efficient was likely beyond human capability. Donald Knuth himself remarked that compilers capable of making it perform even reasonably were nearly impossible to write.</p> <p>Recent research shows that this compiler barrier may no longer be insurmountable. For example, <a rel="nofollow noreferrer" href="https://publish.illinois.edu/hpvm-project/files/2025/04/Towards_Automatic_Generation_of_Efficient_Retargetable_Semantic_Driven_Optimizations-10.pdf">MISAAL</a> is a synthesis-based compiler that uses formal understanding of hardware instructions to automatically generate highly optimized code. It handles complex modern instructions quickly, something once considered impossible for humans alone. This demonstrates that with smarter, automated compilers, software could realistically take on many of the tasks that hardware currently handles. Now imagine what could be possible if we designed a CPU from the ground up with software-first thinking. Hardware could stay minimal and predictable, while the compilers, runtimes, even AI-assisted code generators handles execution, memory management, and scheduling intelligently.</p> <p>Instead of patching over decades of legacy ISA baggage, we could redefine the contract between hardware and software. The CPU would no longer need to be "smart" if the software stack is powerful enough to manage complexity efficiently. This is the vision of a fully software-defined CPU.</p> <h3 id="why-this-idea-matters">Why this idea matters<a class="zola-anchor" href="#why-this-idea-matters" aria-label="Anchor link for: why-this-idea-matters" style="visibility: hidden;"></a> </h3> <p>If we push more intelligence into compilers and runtimes, CPUs could become simpler, cheaper, and more energy-efficient. We would rely on smarter software to:</p> <ul> <li>schedule instructions,</li> <li>manage registers,</li> <li>optimize memory access,</li> <li>even decide execution order.</li> </ul> <p>It is not easy. Predicting the unpredictable is still the biggest challenge. But if we succeed, we could build a <strong>fully software-defined CPU</strong>, where hardware is just a basic engine and software does all the heavy lifting.</p> <p>What do you think, could this really be the future of computing? Let me <a href="mailto:60b36t@gmail.com">know</a>.</p> 2025-08-25T00:00:00+00:00 2025-08-25T00:00:00+00:00 https://johnmaxrin.github.io/posts/reconfig/ <p>I always wanted to create a <strong>universal compiler</strong>, Something that could take <strong>source code + a system ISA specification</strong> and directly produce the <strong>binary</strong> that runs on hardware. Imagine describing your architecture, feeding it into the compiler, and immediately being able to run programs on it.</p> <p>In this blog, I’ll share some of my learnings on this journey, starting from one of the earliest serious attempts at a retargetable compiler: <strong>Fraser’s lcc (1991)</strong>.</p> <hr /> <h2 id="fraser-s-vision-1991">Fraser’s Vision (1991)<a class="zola-anchor" href="#fraser-s-vision-1991" aria-label="Anchor link for: fraser-s-vision-1991" style="visibility: hidden;"></a> </h2> <p>The idea of a “universal compiler” isn’t new. Back in 1991, <strong>Christopher Fraser and David Hanson</strong> built <strong>lcc</strong>, a C compiler that was:</p> <ul> <li><strong>Fast</strong> (both in compilation speed and generated code quality),</li> <li><strong>Retargettable</strong> (easy to support new CPUs),</li> <li><strong>Small</strong> (compact codebase, far smaller than GCC),</li> <li><strong>High quality</strong> (conformed to ANSI C, passed validation suites).</li> </ul> <p>Their secret was <strong>separating the frontend and backend</strong>:</p> <ul> <li> <p>The <strong>frontend</strong> was target-agnostic and did a handful of simple but effective optimizations:</p> <ul> <li>Constant folding</li> <li>Local common subexpression elimination</li> <li>Simulated register declarations</li> <li>Dense jump tables for switch statements</li> </ul> </li> <li> <p>The <strong>backend</strong> was driven by a <strong>rule-based DSL</strong>.<br /> Rules defined how DAG (Directed Acyclic Graph) nodes in the intermediate representation mapped to target instructions.</p> </li> </ul> <p>Example of what a rule looked like (simplified):</p> <pre data-lang="asm" class="language-asm z-code"><code class="language-asm" data-lang="asm"><span class="z-source z-assembly"><span class="z-keyword z-control z-assembly">add</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">→</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-keyword z-control z-assembly">ADD</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-constant z-character z-decimal z-assembly">1</span> </span></code></pre> <p>These rules captured:</p> <ul> <li>The <strong>pattern of DAG operators</strong> (e.g., <code>add</code>, <code>mul</code>, <code>mem[addr]</code>)</li> <li>The <strong>target instruction form</strong></li> <li><strong>Costs/constraints</strong></li> </ul> <p>A <strong>rule compiler</strong> (written in Icon!) translated these rules into a <strong>hard-coded instruction selector</strong> (a tree-pattern matcher). The backend became a <strong>monolithic C program</strong>: fast, compact, and specific to the ISA. The same mechanism also supported <strong>peephole optimizations</strong>: rules that rewrote naïve sequences into more efficient ones.</p> <p>The frontend and backend communicated via a <strong>tight interface</strong>:</p> <ul> <li>~17 functions (<code>emit_prologue</code>, <code>emit_data</code>, …)</li> <li>A 36-operator DAG language</li> </ul> <p>This <strong>clean split</strong> made retargeting remarkably easy. To port lcc, you only needed:</p> <ol> <li>A <strong>configuration file</strong> (data sizes, alignments, calling conventions),</li> <li>A few <strong>backend support functions</strong>,</li> <li>A <strong>rule set</strong> for code generation.</li> </ol> <p>Fraser reported that each new target (VAX, MIPS, SPARC, 68020) required only <strong>377–522 lines</strong> of code. That’s a crazy low effort compared to retargeting GCC.</p> <p>👉 You can read the original paper here: <a rel="nofollow noreferrer" href="https://dl.acm.org/doi/pdf/10.1145/122616.122621">A Retargetable Compiler for ANSI C (1991)</a>.</p> <hr /> <h2 id="drawbacks-of-fraser-s-approach">Drawbacks of Fraser’s Approach<a class="zola-anchor" href="#drawbacks-of-fraser-s-approach" aria-label="Anchor link for: drawbacks-of-fraser-s-approach" style="visibility: hidden;"></a> </h2> <p>While lcc was ahead of its time, it also had <strong>serious limitations</strong>. Many of these explain why modern infrastructures like LLVM and MLIR took different paths.</p> <ol> <li> <p><strong>Limited Optimization</strong></p> <ul> <li>No global optimizations, no SSA, no dataflow analyses.</li> <li>Generated code was “good enough” but lagged behind vendor compilers and later GCC/LLVM.</li> </ul> </li> <li> <p><strong>Tiny IR (36-operator DAG)</strong></p> <ul> <li>Too low-level, closer to canonicalized trees than a flexible IR.</li> <li>Couldn’t represent advanced optimizations like loop transformations or vectorization.</li> </ul> </li> <li> <p><strong>Local Instruction Selection Only</strong></p> <ul> <li>Rules mapped one DAG at a time → no global instruction scheduling.</li> <li>Couldn’t handle complex idioms (e.g., <code>memcpy</code> → <code>rep movsb</code> on x86, delay slots on SPARC).</li> </ul> </li> <li> <p><strong>Shallow Retargetability</strong></p> <ul> <li>Worked well for clean RISC ISAs (MIPS, SPARC, VAX).</li> <li>Much harder for “irregular” ISAs like x86, DSPs, or GPUs.</li> </ul> </li> <li> <p><strong>Opaque Generated Backends</strong></p> <ul> <li>The rule compiler produced large monolithic C code.</li> <li>Fast, but hard to debug or customize compared to data-driven matchers.</li> </ul> </li> <li> <p><strong>Language Lock-In</strong></p> <ul> <li>Frontend was tightly coupled to ANSI C.</li> <li>Couldn’t serve as a general-purpose compiler infrastructure.</li> </ul> </li> <li> <p><strong>Didn’t Scale with Hardware Trends</strong></p> <ul> <li>Built before superscalar, SIMD, multicore, and GPUs became dominant.</li> <li>The tiny DAG IR couldn’t express parallelism or vector-level semantics.</li> </ul> </li> </ol> <hr /> <h2 id="why-it-still-matters">Why It Still Matters<a class="zola-anchor" href="#why-it-still-matters" aria-label="Anchor link for: why-it-still-matters" style="visibility: hidden;"></a> </h2> <p>Despite its limitations, Fraser’s approach was <strong>visionary</strong>.</p> <ul> <li>The <strong>declarative rule-based backend</strong> foreshadowed LLVM’s <strong>TableGen</strong>, MLIR’s <strong>ODS</strong>, Cranelift’s <strong>ISLE</strong>, and GNU’s <strong>CGEN</strong>.</li> <li>The <strong>tight front–backend interface</strong> showed that <strong>retargeting can be simple</strong> if the abstraction is chosen well.</li> <li>The <strong>focus on speed + smallness</strong> influenced generations of lightweight compilers used in embedded systems and education.</li> </ul> <p>In short: Fraser’s lcc was a <strong>proto-universal compiler</strong>. It wasn’t powerful enough to scale into LLVM, but it proved the idea: <em>source code + machine spec → working compiler backend</em>.</p> <hr /> <h2 id="my-takeaway">My Takeaway<a class="zola-anchor" href="#my-takeaway" aria-label="Anchor link for: my-takeaway" style="visibility: hidden;"></a> </h2> <p>As I think about building a universal compiler today, Fraser’s work reminds me:</p> <ul> <li>Keep the <strong>interfaces clean</strong>.</li> <li>Automate as much as possible (rule compilers, declarative ISAs).</li> <li>Balance <strong>simplicity vs optimization</strong> — you can always add more sophisticated passes later.</li> </ul> <p>LLVM, MLIR, and even modern DSLs for ISAs (like <a rel="nofollow noreferrer" href="https://ghidra.re/ghidra_docs/languages/html/sleigh.html">SLEIGH in Ghidra</a> or <a rel="nofollow noreferrer" href="https://github.com/riscv/sail-riscv">Sail for formal specs</a>) are essentially “Fraser’s dream, at scale”.</p> <p>The journey towards a universal compiler isn’t finished, but lcc shows we’ve been chasing this dream for decades.</p> <hr /> 2025-07-25T00:00:00+00:00 2025-07-25T00:00:00+00:00 https://johnmaxrin.github.io/posts/painmlir2/ <h1 id="understanding-memory-access-relations">Understanding Memory Access Relations<a class="zola-anchor" href="#understanding-memory-access-relations" aria-label="Anchor link for: understanding-memory-access-relations" style="visibility: hidden;"></a> </h1> <p>In <a href="/posts/painmlir/">Part 1</a> of this series, we introduced the fundamentals of the polyhedral model and how it fits into the MLIR framework. In this post, we dive into one of the most important components of that model: the memory access relation.</p> <p>A memory access relation is a mathematical mapping from loop iterations to the memory locations accessed during those iterations.</p> <p>You can think of it like this:</p> <pre class="z-code"><code><span class="z-text z-plain">AccessRel: [loop indices] → [array subscripts] </span></code></pre> <p>In the context of the polyhedral model:</p> <ul> <li>Domain: The iteration space of the loop</li> <li>Range: The memory space being accessed (array indices)</li> </ul> <p>This relation is fundamental for detecting data dependences, validating transformations, and enabling optimizations.</p> <h1 id="example-in-mlir">Example in MLIR<a class="zola-anchor" href="#example-in-mlir" aria-label="Anchor link for: example-in-mlir" style="visibility: hidden;"></a> </h1> <p>Let’s take a simple loop written in MLIR:</p> <pre data-lang="c++" class="language-c++ z-code"><code class="language-c++" data-lang="c++"><span class="z-source z-c++">affine<span class="z-punctuation z-accessor z-dot z-c++">.</span><span class="z-variable z-other z-readwrite z-member z-c++">for</span> <span class="z-keyword z-operator z-arithmetic z-c">%</span>i <span class="z-keyword z-operator z-assignment z-c">=</span> <span class="z-constant z-numeric z-integer z-decimal z-c++">0</span> to <span class="z-constant z-numeric z-integer z-decimal z-c++">7</span> <span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-begin z-c++">{</span> </span></span><span class="z-source z-c++"><span class="z-meta z-block z-c++"> <span class="z-keyword z-operator z-arithmetic z-c">%</span>a <span class="z-keyword z-operator z-assignment z-c">=</span> affine<span class="z-punctuation z-accessor z-dot z-c++">.</span><span class="z-variable z-other z-readwrite z-member z-c++">load</span> <span class="z-keyword z-operator z-arithmetic z-c">%</span>A<span class="z-meta z-brackets z-c++"><span class="z-punctuation z-section z-brackets z-begin z-c++">[</span><span class="z-keyword z-operator z-arithmetic z-c">%</span>i <span class="z-keyword z-operator z-arithmetic z-c">+</span> <span class="z-constant z-numeric z-integer z-decimal z-c++">1</span><span class="z-punctuation z-section z-brackets z-end z-c++">]</span></span> <span class="z-keyword z-operator z-ternary z-c">:</span> memref<span class="z-punctuation z-section z-generic z-begin z-c++">&lt;</span><span class="z-keyword z-operator z-ternary z-c">?</span>xf32<span class="z-punctuation z-section z-generic z-end z-c++">&gt;</span> </span></span><span class="z-source z-c++"><span class="z-meta z-block z-c++"> <span class="z-keyword z-operator z-arithmetic z-c">%</span>sum <span class="z-keyword z-operator z-assignment z-c">=</span> arith<span class="z-punctuation z-accessor z-dot z-c++">.</span><span class="z-variable z-other z-readwrite z-member z-c++">addf</span> <span class="z-keyword z-operator z-arithmetic z-c">%</span>one<span class="z-punctuation z-separator z-c++">,</span> <span class="z-keyword z-operator z-arithmetic z-c">%</span>a <span class="z-keyword z-operator z-ternary z-c">:</span> f32 </span></span><span class="z-source z-c++"><span class="z-meta z-block z-c++"> affine<span class="z-punctuation z-accessor z-dot z-c++">.</span><span class="z-variable z-other z-readwrite z-member z-c++">store</span> <span class="z-keyword z-operator z-arithmetic z-c">%</span>sum<span class="z-punctuation z-separator z-c++">,</span> <span class="z-keyword z-operator z-arithmetic z-c">%</span>A<span class="z-meta z-brackets z-c++"><span class="z-punctuation z-section z-brackets z-begin z-c++">[</span><span class="z-keyword z-operator z-arithmetic z-c">%</span>i<span class="z-punctuation z-section z-brackets z-end z-c++">]</span></span> <span class="z-keyword z-operator z-ternary z-c">:</span> memref<span class="z-punctuation z-section z-generic z-begin z-c++">&lt;</span><span class="z-keyword z-operator z-ternary z-c">?</span>xf32<span class="z-punctuation z-section z-generic z-end z-c++">&gt;</span> </span></span><span class="z-source z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-end z-c++">}</span></span> </span></code></pre> <p>Here’s what’s happening:</p> <ul> <li>We're reading from <code>A[%i + 1]</code></li> <li>We’re writing to <code>A[%i]</code> There are two memory access relations here: one for the load, and one for the store.</li> </ul> <h2 id="loop-constraints">Loop Constraints<a class="zola-anchor" href="#loop-constraints" aria-label="Anchor link for: loop-constraints" style="visibility: hidden;"></a> </h2> <p>These are the constraints on the loop iteration variable %i, extracted from the domain:</p> <pre data-lang="bash" class="language-bash z-code"><code class="language-bash" data-lang="bash"><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">Loop</span></span><span class="z-meta z-function-call z-arguments z-shell"> Constraints</span> </span><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">Domain:</span></span><span class="z-meta z-function-call z-arguments z-shell"> 0, Range: 1, Symbols: 0, Locals: 0</span> </span><span class="z-source z-shell z-bash"><span class="z-punctuation z-definition z-compound z-begin z-shell">(</span> <span class="z-punctuation z-definition z-compound z-end z-shell">)</span><span class="z-meta z-function-call z-arguments z-shell"> -<span class="z-keyword z-operator z-assignment z-redirection z-shell">&gt;</span> ( Id<span class="z-keyword z-operator z-assignment z-redirection z-shell">&lt;</span><span class="z-constant z-numeric z-integer z-decimal z-file-descriptor z-shell">0</span>x12ee37dd<span class="z-constant z-numeric z-integer z-decimal z-file-descriptor z-shell">0</span><span class="z-keyword z-operator z-assignment z-redirection z-shell">&gt;</span> </span><span class="z-meta z-function-call z-shell"></span>) <span class="z-meta z-function-call z-shell"><span class="z-support z-function z-colon z-shell">:</span></span><span class="z-meta z-function-call z-arguments z-shell"> <span class="z-keyword z-control z-regexp z-set z-begin z-shell">[</span> <span class="z-keyword z-control z-regexp z-set z-end z-shell">]</span> 2 constraints</span> </span><span class="z-source z-shell z-bash"><span class="z-punctuation z-definition z-compound z-begin z-shell">(</span><span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">Value</span></span><span class="z-meta z-function-call z-arguments z-shell"> const</span><span class="z-punctuation z-definition z-compound z-end z-shell">)</span> </span><span class="z-source z-shell z-bash"> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">1</span></span><span class="z-meta z-function-call z-arguments z-shell"> 0 <span class="z-keyword z-operator z-assignment z-redirection z-shell">&gt;</span>= 0</span> <span class="z-keyword z-operator z-logical z-continue z-shell">;</span> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">i</span></span><span class="z-meta z-function-call z-arguments z-shell"> <span class="z-keyword z-operator z-assignment z-redirection z-shell">&gt;</span>= 0</span> </span><span class="z-source z-shell z-bash"> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">-1</span></span><span class="z-meta z-function-call z-arguments z-shell"> 6 <span class="z-keyword z-operator z-assignment z-redirection z-shell">&gt;</span>= 0</span> <span class="z-keyword z-operator z-logical z-continue z-shell">;</span> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">i</span></span><span class="z-meta z-function-call z-arguments z-shell"> <span class="z-keyword z-operator z-assignment z-redirection z-shell">&lt;</span>= 6 (since loop is i = 0 to 7 exclusive</span><span class="z-meta z-function-call z-shell"></span>) </span></code></pre> <p>This defines the iteration domain: <code>0 ≤ i &lt; 7</code>.</p> <h2 id="memory-access-relations">Memory Access Relations<a class="zola-anchor" href="#memory-access-relations" aria-label="Anchor link for: memory-access-relations" style="visibility: hidden;"></a> </h2> <p>Let’s look at the access relations for both load and store. Load: <code>A[%i + 1]</code></p> <pre data-lang="bash" class="language-bash z-code"><code class="language-bash" data-lang="bash"><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">memory</span></span><span class="z-meta z-function-call z-arguments z-shell"> access rel - load</span> </span><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">Domain:</span></span><span class="z-meta z-function-call z-arguments z-shell"> 0, Range: 1, Symbols: 0, Locals: 1</span> </span><span class="z-source z-shell z-bash"><span class="z-punctuation z-definition z-compound z-begin z-shell">(</span> <span class="z-punctuation z-definition z-compound z-end z-shell">)</span><span class="z-meta z-function-call z-arguments z-shell"> -<span class="z-keyword z-operator z-assignment z-redirection z-shell">&gt;</span> ( Id<span class="z-keyword z-operator z-assignment z-redirection z-shell">&lt;</span><span class="z-constant z-numeric z-integer z-decimal z-file-descriptor z-shell">0</span>x12ee37dd<span class="z-constant z-numeric z-integer z-decimal z-file-descriptor z-shell">0</span><span class="z-keyword z-operator z-assignment z-redirection z-shell">&gt;</span> </span><span class="z-meta z-function-call z-shell"></span>) <span class="z-meta z-function-call z-shell"><span class="z-support z-function z-colon z-shell">:</span></span><span class="z-meta z-function-call z-arguments z-shell"> <span class="z-keyword z-control z-regexp z-set z-begin z-shell">[</span> <span class="z-keyword z-control z-regexp z-set z-end z-shell">]</span> 3 constraints</span> </span><span class="z-source z-shell z-bash"><span class="z-punctuation z-definition z-compound z-begin z-shell">(</span><span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">Value</span></span><span class="z-meta z-function-call z-arguments z-shell"> Local const</span><span class="z-punctuation z-definition z-compound z-end z-shell">)</span> </span><span class="z-source z-shell z-bash"> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">1</span></span><span class="z-meta z-function-call z-arguments z-shell"><span class="z-variable z-parameter z-option z-shell"><span class="z-punctuation z-definition z-parameter z-shell"> -</span>1</span> 1 = 0</span> <span class="z-keyword z-operator z-logical z-continue z-shell">;</span> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">m</span></span><span class="z-meta z-function-call z-arguments z-shell"> = i + 1</span> </span><span class="z-source z-shell z-bash"> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">1</span></span><span class="z-meta z-function-call z-arguments z-shell"> 0 0 <span class="z-keyword z-operator z-assignment z-redirection z-shell">&gt;</span>= 0</span> <span class="z-keyword z-operator z-logical z-continue z-shell">;</span> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">i</span></span><span class="z-meta z-function-call z-arguments z-shell"> <span class="z-keyword z-operator z-assignment z-redirection z-shell">&gt;</span>= 0</span> </span><span class="z-source z-shell z-bash"> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">-1</span></span><span class="z-meta z-function-call z-arguments z-shell"> 0 6 <span class="z-keyword z-operator z-assignment z-redirection z-shell">&gt;</span>= 0</span> <span class="z-keyword z-operator z-logical z-continue z-shell">;</span> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">i</span></span><span class="z-meta z-function-call z-arguments z-shell"> <span class="z-keyword z-operator z-assignment z-redirection z-shell">&lt;</span>= 6</span> </span></code></pre> <p><code>1 -1 1 = 0</code> means: Memory index <code>m</code> = loop index <code>i + 1</code> (Here, <code>m</code> is represented using a local variable)</p> <p>Store: <code>A[%i]</code></p> <pre data-lang="bash" class="language-bash z-code"><code class="language-bash" data-lang="bash"><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">memory</span></span><span class="z-meta z-function-call z-arguments z-shell"> access rel - store</span> </span><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">Domain:</span></span><span class="z-meta z-function-call z-arguments z-shell"> 0, Range: 1, Symbols: 0, Locals: 1</span> </span><span class="z-source z-shell z-bash"><span class="z-punctuation z-definition z-compound z-begin z-shell">(</span> <span class="z-punctuation z-definition z-compound z-end z-shell">)</span><span class="z-meta z-function-call z-arguments z-shell"> -<span class="z-keyword z-operator z-assignment z-redirection z-shell">&gt;</span> ( Id<span class="z-keyword z-operator z-assignment z-redirection z-shell">&lt;</span><span class="z-constant z-numeric z-integer z-decimal z-file-descriptor z-shell">0</span>x12ee37dd<span class="z-constant z-numeric z-integer z-decimal z-file-descriptor z-shell">0</span><span class="z-keyword z-operator z-assignment z-redirection z-shell">&gt;</span> </span><span class="z-meta z-function-call z-shell"></span>) <span class="z-meta z-function-call z-shell"><span class="z-support z-function z-colon z-shell">:</span></span><span class="z-meta z-function-call z-arguments z-shell"> <span class="z-keyword z-control z-regexp z-set z-begin z-shell">[</span> <span class="z-keyword z-control z-regexp z-set z-end z-shell">]</span> 3 constraints</span> </span><span class="z-source z-shell z-bash"><span class="z-punctuation z-definition z-compound z-begin z-shell">(</span><span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">Value</span></span><span class="z-meta z-function-call z-arguments z-shell"> Local const</span><span class="z-punctuation z-definition z-compound z-end z-shell">)</span> </span><span class="z-source z-shell z-bash"> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">1</span></span><span class="z-meta z-function-call z-arguments z-shell"><span class="z-variable z-parameter z-option z-shell"><span class="z-punctuation z-definition z-parameter z-shell"> -</span>1</span> 0 = 0</span> <span class="z-keyword z-operator z-logical z-continue z-shell">;</span> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">m</span></span><span class="z-meta z-function-call z-arguments z-shell"> = i</span> </span><span class="z-source z-shell z-bash"> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">1</span></span><span class="z-meta z-function-call z-arguments z-shell"> 0 0 <span class="z-keyword z-operator z-assignment z-redirection z-shell">&gt;</span>= 0</span> <span class="z-keyword z-operator z-logical z-continue z-shell">;</span> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">i</span></span><span class="z-meta z-function-call z-arguments z-shell"> <span class="z-keyword z-operator z-assignment z-redirection z-shell">&gt;</span>= 0</span> </span><span class="z-source z-shell z-bash"> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">-1</span></span><span class="z-meta z-function-call z-arguments z-shell"> 0 6 <span class="z-keyword z-operator z-assignment z-redirection z-shell">&gt;</span>= 0</span> <span class="z-keyword z-operator z-logical z-continue z-shell">;</span> <span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">i</span></span><span class="z-meta z-function-call z-arguments z-shell"> <span class="z-keyword z-operator z-assignment z-redirection z-shell">&lt;</span>= 6</span> </span></code></pre> <p>This line: <code>1 -1 0 = 0</code> means: Memory index <code>m</code> = loop index <code>i</code></p> <p>MLIR provides utilities to extract these relations automatically from your code.</p> <pre data-lang="c++" class="language-c++ z-code"><code class="language-c++" data-lang="c++"><span class="z-source z-c++">mlir<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>affine<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>MemRefAccess <span class="z-meta z-function z-c++"><span class="z-meta z-toc-list z-full-identifier z-c++"><span class="z-entity z-name z-function z-c++">src</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-variable z-parameter z-c++">loadOp</span><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span><span class="z-source z-c++">mlir<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>affine<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>MemRefAccess <span class="z-meta z-function z-c++"><span class="z-meta z-toc-list z-full-identifier z-c++"><span class="z-entity z-name z-function z-c++">dst</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-variable z-parameter z-c++">storeOp</span><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span><span class="z-source z-c++"> </span><span class="z-source z-c++">mlir<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>presburger<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>PresburgerSpace space <span class="z-keyword z-operator z-assignment z-c">=</span> </span><span class="z-source z-c++"> <span class="z-meta z-function z-c++"><span class="z-meta z-toc-list z-full-identifier z-c++">mlir<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>presburger<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>PresburgerSpace<span class="z-punctuation z-accessor z-double-colon z-c++">::</span><span class="z-entity z-name z-function z-constructor z-c++">getRelationSpace</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span><span class="z-source z-c++"> </span><span class="z-source z-c++">mlir<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>presburger<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>IntegerRelation <span class="z-meta z-function z-c++"><span class="z-meta z-toc-list z-full-identifier z-c++"><span class="z-entity z-name z-function z-c++">srcRel</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-variable z-parameter z-c++">space</span><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-separator z-c++">,</span> <span class="z-meta z-function-call z-c++"><span class="z-variable z-function z-c++">dstRel</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++">space</span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span><span class="z-source z-c++"> </span><span class="z-source z-c++"><span class="z-comment z-line z-double-slash z-c"><span class="z-punctuation z-definition z-comment z-c">//</span> Extract memory access relations into IntegerRelation </span></span><span class="z-source z-c++"><span class="z-keyword z-control z-c++">if</span> <span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span><span class="z-meta z-function-call z-c++"><span class="z-variable z-function z-c++">failed</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++">src<span class="z-punctuation z-accessor z-dot z-c++">.</span><span class="z-meta z-method-call z-c++"><span class="z-variable z-function z-member z-c++">getAccessRelation</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++">srcRel</span></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span> <span class="z-keyword z-operator z-arithmetic z-c">||</span> </span></span><span class="z-source z-c++"><span class="z-meta z-group z-c++"> <span class="z-meta z-function-call z-c++"><span class="z-variable z-function z-c++">failed</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++">dst<span class="z-punctuation z-accessor z-dot z-c++">.</span><span class="z-meta z-method-call z-c++"><span class="z-variable z-function z-member z-c++">getAccessRelation</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++">dstRel</span></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-section z-group z-end z-c++">)</span></span> <span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-begin z-c++">{</span> </span></span><span class="z-source z-c++"><span class="z-meta z-block z-c++"> <span class="z-keyword z-control z-flow z-return z-c++">return</span> mlir<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>affine<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>DependenceResult<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>Failure<span class="z-punctuation z-terminator z-c++">;</span> </span></span><span class="z-source z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-end z-c++">}</span></span> </span><span class="z-source z-c++"> </span><span class="z-source z-c++">affine<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>FlatAffineValueConstraints <span class="z-meta z-function z-c++"><span class="z-meta z-toc-list z-full-identifier z-c++"><span class="z-entity z-name z-function z-c++">srcDomain</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++">srcRel<span class="z-punctuation z-accessor z-dot z-c++">.</span><span class="z-meta z-method-call z-c++"><span class="z-variable z-function z-member z-c++">getDomainSet</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-method-call z-c++"></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span><span class="z-source z-c++">affine<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>FlatAffineValueConstraints <span class="z-meta z-function z-c++"><span class="z-meta z-toc-list z-full-identifier z-c++"><span class="z-entity z-name z-function z-c++">dstDomain</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++">dstRel<span class="z-punctuation z-accessor z-dot z-c++">.</span><span class="z-meta z-method-call z-c++"><span class="z-variable z-function z-member z-c++">getDomainSet</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-method-call z-c++"></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span><span class="z-source z-c++"> </span><span class="z-source z-c++">srcDomain<span class="z-punctuation z-accessor z-dot z-c++">.</span><span class="z-meta z-method-call z-c++"><span class="z-variable z-function z-member z-c++">dump</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-method-call z-c++"></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span><span class="z-source z-c++">dstDomain<span class="z-punctuation z-accessor z-dot z-c++">.</span><span class="z-meta z-method-call z-c++"><span class="z-variable z-function z-member z-c++">dump</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-method-call z-c++"></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></code></pre> <p>With this setup, you can analyze, print, or even manipulate memory access relations directly from within an MLIR pass or analysis pipeline.</p> <h1 id="coming-next">Coming Next<a class="zola-anchor" href="#coming-next" aria-label="Anchor link for: coming-next" style="visibility: hidden;"></a> </h1> <p>In the next part of this series, we’ll explore how to use these access relations to perform dependence analysis that enables transformations like loop interchange, tiling, or fusion.</p> <p>Stay tuned!</p> 2025-07-24T00:00:00+00:00 2025-07-24T00:00:00+00:00 https://johnmaxrin.github.io/posts/painmlir/ <p><a rel="nofollow noreferrer" href="http://polyhedral.info/"><strong>Polyhedral analysis</strong></a> is a mathematical framework used to analyze and transform loop nests in programs—particularly those with static control parts (where loop bounds and memory accesses are affine expressions). This model represents iteration domains, loop bounds, and memory access patterns as systems of linear inequalities. These sets can then be used for powerful optimizations such as loop fusion, tiling, interchange, and parallelization.</p> <h1 id="what-it-means-in-practice">What It Means in Practice<a class="zola-anchor" href="#what-it-means-in-practice" aria-label="Anchor link for: what-it-means-in-practice" style="visibility: hidden;"></a> </h1> <p>Suppose you have a matrix multiplication function written in MLIR using the <code>affine.for</code> dialect with dynamic sizes for matrices:</p> <pre data-lang="mlir" class="language-mlir z-code"><code class="language-mlir" data-lang="mlir"><span class="z-text z-plain">func.func @matmul(%A: memref&lt;?x?xf32&gt;, %B: memref&lt;?x?xf32&gt;, %C: memref&lt;?x?xf32&gt;) { </span><span class="z-text z-plain"> %c0 = arith.constant 0 : index </span><span class="z-text z-plain"> %c1 = arith.constant 1 : index </span><span class="z-text z-plain"> </span><span class="z-text z-plain"> %M = memref.dim %A, %c0 : memref&lt;?x?xf32&gt; </span><span class="z-text z-plain"> %N = memref.dim %B, %c1 : memref&lt;?x?xf32&gt; </span><span class="z-text z-plain"> %K = memref.dim %A, %c1 : memref&lt;?x?xf32&gt; </span><span class="z-text z-plain"> </span><span class="z-text z-plain"> affine.for %i = 0 to %M step 1 { </span><span class="z-text z-plain"> affine.for %j = 0 to %N { </span><span class="z-text z-plain"> affine.for %k = 0 to %K { </span><span class="z-text z-plain"> %a = affine.load %A[%i, %k] : memref&lt;?x?xf32&gt; </span><span class="z-text z-plain"> %b = affine.load %B[%k, %j] : memref&lt;?x?xf32&gt; </span><span class="z-text z-plain"> %c = affine.load %C[%i, %j] : memref&lt;?x?xf32&gt; </span><span class="z-text z-plain"> %prod = arith.mulf %a, %b : f32 </span><span class="z-text z-plain"> %sum = arith.addf %c, %prod : f32 </span><span class="z-text z-plain"> affine.store %sum, %C[%i, %j] : memref&lt;?x?xf32&gt; </span><span class="z-text z-plain"> } </span><span class="z-text z-plain"> } </span><span class="z-text z-plain"> } </span><span class="z-text z-plain"> return </span><span class="z-text z-plain">} </span></code></pre> <p>This code uses affine.for loops with dynamic bounds computed at runtime (%M, %N, %K) via memref.dim. The loop nests iterate over all valid (i, j, k) indices for matrix multiplication, and use affine loads and stores to access memory in a way that can be modeled with the polyhedral model.</p> <h1 id="extracting-the-polyhedral-model">Extracting the Polyhedral Model<a class="zola-anchor" href="#extracting-the-polyhedral-model" aria-label="Anchor link for: extracting-the-polyhedral-model" style="visibility: hidden;"></a> </h1> <p>MLIR provides the infrastructure to extract this iteration domain into a formal representation called a FlatAffineValueConstraints. This object stores the loop's constraints in a matrix form, representing:</p> <ul> <li>Domain dimensions (induction variables)</li> <li>Symbols (parameters/constants)</li> <li>A set of linear inequalities</li> </ul> <p>You can extract this model using the getIndexSet function:</p> <pre data-lang="mlir" class="language-mlir z-code"><code class="language-mlir" data-lang="mlir"><span class="z-text z-plain">affine::FlatAffineValueConstraints constraints; </span><span class="z-text z-plain">SmallVector&lt;Operation *, 4&gt; ops; </span><span class="z-text z-plain"> </span><span class="z-text z-plain">module-&gt;walk&lt;mlir::WalkOrder::PreOrder&gt;([&amp;](mlir::affine::AffineForOp forOp) { </span><span class="z-text z-plain"> ops.push_back(forOp); </span><span class="z-text z-plain">}); </span><span class="z-text z-plain"> </span><span class="z-text z-plain">affine::getIndexSet(ops, &amp;constraints); </span><span class="z-text z-plain">constraints.dump(); </span><span class="z-text z-plain"> </span></code></pre> <p>This generates a model like:</p> <pre data-lang="yaml" class="language-yaml z-code"><code class="language-yaml" data-lang="yaml"><span class="z-source z-yaml"><span class="z-string z-unquoted z-plain z-out z-yaml"><span class="z-entity z-name z-tag z-yaml">Domain</span></span><span class="z-punctuation z-separator z-key-value z-mapping z-yaml">:</span> <span class="z-string z-unquoted z-plain z-out z-yaml"><span class="z-entity z-name z-tag z-yaml">0, Range</span></span><span class="z-punctuation z-separator z-key-value z-mapping z-yaml">:</span> <span class="z-string z-unquoted z-plain z-out z-yaml"><span class="z-entity z-name z-tag z-yaml">3, Symbols</span></span><span class="z-punctuation z-separator z-key-value z-mapping z-yaml">:</span> <span class="z-string z-unquoted z-plain z-out z-yaml"><span class="z-entity z-name z-tag z-yaml">3, Locals</span></span><span class="z-punctuation z-separator z-key-value z-mapping z-yaml">:</span> <span class="z-constant z-numeric z-integer z-decimal z-yaml">0</span> </span><span class="z-source z-yaml"><span class="z-string z-unquoted z-plain z-out z-yaml"><span class="z-entity z-name z-tag z-yaml">( ) -&gt; ( i j k )</span></span> <span class="z-punctuation z-separator z-key-value z-mapping z-yaml">:</span> <span class="z-meta z-flow-sequence z-yaml"><span class="z-punctuation z-definition z-sequence z-begin z-yaml">[</span> <span class="z-string z-unquoted z-plain z-in z-yaml">M N K</span> <span class="z-punctuation z-definition z-sequence z-end z-yaml">]</span></span> </span><span class="z-source z-yaml"><span class="z-string z-unquoted z-plain z-out z-yaml">6 constraints</span> </span><span class="z-source z-yaml"><span class="z-string z-unquoted z-plain z-out z-yaml">(Value Value Value Value Value Value const)</span> </span><span class="z-source z-yaml"> <span class="z-string z-unquoted z-plain z-out z-yaml">1 0 0 0 0 0 0 &gt;= 0</span> </span><span class="z-source z-yaml"> <span class="z-string z-unquoted z-plain z-out z-yaml">-1 0 0 1 0 0 -1 &gt;= 0</span> </span><span class="z-source z-yaml"> <span class="z-string z-unquoted z-plain z-out z-yaml">0 1 0 0 0 0 0 &gt;= 0</span> </span><span class="z-source z-yaml"> <span class="z-string z-unquoted z-plain z-out z-yaml">0 -1 0 0 1 0 -1 &gt;= 0</span> </span><span class="z-source z-yaml"> <span class="z-string z-unquoted z-plain z-out z-yaml">0 0 1 0 0 0 0 &gt;= 0</span> </span><span class="z-source z-yaml"> <span class="z-string z-unquoted z-plain z-out z-yaml">0 0 -1 0 0 1 -1 &gt;= 0</span> </span></code></pre> <h2 id="interpreting-the-output">Interpreting the Output<a class="zola-anchor" href="#interpreting-the-output" aria-label="Anchor link for: interpreting-the-output" style="visibility: hidden;"></a> </h2> <p>This output captures the bounds of the loop nest. Here's how to interpret it:</p> <ul> <li>There are 3 loop variables (i, j, k) corresponding to %arg3, %arg4, and %arg5.</li> <li>There are 3 symbols (M, N, K) corresponding to the upper bounds 200, 300, and 400 (these can be parametric).</li> <li>The constraints describe the bounds:</li> </ul> <pre data-lang="yaml" class="language-yaml z-code"><code class="language-yaml" data-lang="yaml"><span class="z-source z-yaml"><span class="z-string z-unquoted z-plain z-out z-yaml">i &gt;= 0</span> </span><span class="z-source z-yaml"><span class="z-string z-unquoted z-plain z-out z-yaml">i &lt;= M - 1</span> </span><span class="z-source z-yaml"><span class="z-string z-unquoted z-plain z-out z-yaml">j &gt;= 0</span> </span><span class="z-source z-yaml"><span class="z-string z-unquoted z-plain z-out z-yaml">j &lt;= N - 1</span> </span><span class="z-source z-yaml"><span class="z-string z-unquoted z-plain z-out z-yaml">k &gt;= 0</span> </span><span class="z-source z-yaml"><span class="z-string z-unquoted z-plain z-out z-yaml">k &lt;= K - 1</span> </span></code></pre> <p>This forms a polyhedron in 3D space, bounded by the affine inequalities.</p> <h1 id="why-this-matters">Why This Matters<a class="zola-anchor" href="#why-this-matters" aria-label="Anchor link for: why-this-matters" style="visibility: hidden;"></a> </h1> <p>Once this model is constructed, the compiler can:</p> <ul> <li>Detect loop-carried dependencies</li> <li>Apply loop transformations legally</li> <li>Optimize data movement</li> <li>Parallelize independent loop iterations</li> </ul> <p>All of this is grounded in formal, provable correctness through the polyhedral representation.</p> <h1 id="final-thoughts">Final Thoughts<a class="zola-anchor" href="#final-thoughts" aria-label="Anchor link for: final-thoughts" style="visibility: hidden;"></a> </h1> <p>MLIR brings polyhedral analysis closer to everyday compiler IR workflows. With affine dialect, and analysis utilities like FlatAffineValueConstraints, MLIR makes it possible to perform sophisticated optimizations automatically. This is especially powerful in domains like HPC, deep learning, and scientific computing.</p> <p>If you're writing compilers or building optimizers, MLIR's polyhedral framework is not just a theory—it's a tool you can use today.</p> <h1 id="references">References<a class="zola-anchor" href="#references" aria-label="Anchor link for: references" style="visibility: hidden;"></a> </h1> <ul> <li><a rel="nofollow noreferrer" href="https://www.youtube.com/watch?v=Xg4RfgPIT-Y"> 2022 EuroLLVM Dev Mtg “Precise Polyhedral Analyses For MLIR using the FPL Presburger Library” </a></li> <li><a rel="nofollow noreferrer" href="https://www.youtube.com/watch?v=fcBkr4mk-8A"> Polyhedral Analyses using MLIR's Affine dialect - Vinayaka Bandishti </a></li> <li><a rel="nofollow noreferrer" href="https://www.youtube.com/watch?v=jBgU_EeUKYw"> Polyhedral Analyses using MLIR's Affine Dialect (contd.) </a></li> <li><a rel="nofollow noreferrer" href="https://www.youtube.com/watch?v=BKcFoP6B4Rw"> Compiler Design Module 128 : Introduction to the Polyhedral Framework </a></li> </ul> 2025-06-04T00:00:00+00:00 2025-06-04T00:00:00+00:00 https://johnmaxrin.github.io/posts/mlirgpu/ <p>This tutorial shows how to write, compile, and run a simple “Hello World” program using MLIR’s GPU dialect targeting an NVIDIA GPU (CUDA backend). We assume MLIR is built with NVPTX support and a CUDA toolkit is installed. We focus on GPU-dialect specifics: writing the kernel IR, host integration, and launch configuration, and the MLIR compilation pipeline to generate a GPU executable.</p> <h2 id="1-write-the-gpu-kernel-in-mlir">1. Write the GPU Kernel in MLIR<a class="zola-anchor" href="#1-write-the-gpu-kernel-in-mlir" aria-label="Anchor link for: 1-write-the-gpu-kernel-in-mlir" style="visibility: hidden;"></a> </h2> <p>The GPU dialect uses gpu.launch to invoke kernels on the GPU with a specified grid (blocks) and block (threads) configuration. Inside a gpu.launch region you write device code with GPU ops (like gpu.thread_id or gpu.printf) and terminate with gpu.terminator. For example, to print from the GPU, you can use a gpu.printf inside a 1×1×1 launch:</p> <pre data-lang="asm" class="language-asm z-code"><code class="language-asm" data-lang="asm"><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">f</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">f</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-source z-assembly">-</span><span class="z-entity z-name z-function z-assembly">&gt;</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">1</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">h</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">1</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">:</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span> </span><span class="z-source z-assembly"> <span class="z-support z-function z-directive z-assembly">%0</span> <span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">h</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">x</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">1</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">:</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">x</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">h</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">33</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">:</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">h</span><span class="z-entity z-name z-function z-assembly"> </span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">b</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">0</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">1</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-keyword z-control z-assembly">in</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span> <span class="z-support z-function z-directive z-assembly">%0</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">7</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span> <span class="z-support z-function z-directive z-assembly">%0</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">8</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span> <span class="z-support z-function z-directive z-assembly">%0</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly"> </span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">h</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">5</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-keyword z-control z-assembly">in</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">9</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span> <span class="z-support z-function z-directive z-assembly">%0</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">1</span><span class="z-entity z-name z-function z-assembly">0</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span> <span class="z-support z-function z-directive z-assembly">%0</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">1</span><span class="z-entity z-name z-function z-assembly">1</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span> <span class="z-support z-function z-directive z-assembly">%0</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">f</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-string z-assembly">&quot;Hi There: &quot;</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">r</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">}</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">:</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">}</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">}</span> </span></code></pre> <p>Let’s walk through this MLIR code:</p> <ul> <li> <p><code>func.func @main()</code> defines the entry point of the program.</p> </li> <li> <p><code>arith.constant</code> is used to define integer constants. In this case, <code>%c1_i32</code> is a constant value 1, and <code>%c42_i32</code> is the return value 42.</p> </li> <li> <p><code>arith.index_cast</code> converts the i32 constant to an index type, which is typically used for loop bounds or launch parameters in MLIR.</p> </li> <li> <p>The <code>gpu.launch</code> operation launches a GPU kernel. It specifies the grid and block dimensions:</p> <ul> <li> <p><code>blocks(...) in (...)</code>: Describes how many blocks will be launched.</p> </li> <li> <p><code>threads(...) in (...)</code>: Describes how many threads per block.</p> </li> </ul> <p>Here, both blocks and threads are launched with 1 in each dimension (x, y, z), so it’s a minimal configuration.</p> </li> <li> <p>Inside the GPU kernel body, we simply use <code>gpu.printf</code> to print a string "<strong>Hi There:</strong> ".</p> </li> <li> <p><code>gpu.terminator</code> marks the end of the GPU kernel region.</p> </li> <li> <p>The function returns the integer value <code>33</code>.</p> </li> </ul> <h2 id="3-the-transformation-pipeline">3 .The Transformation Pipeline<a class="zola-anchor" href="#3-the-transformation-pipeline" aria-label="Anchor link for: 3-the-transformation-pipeline" style="visibility: hidden;"></a> </h2> <p>To transform a high-level GPU MLIR program into a low-level NVVM binary representation, we can use the following mlir-opt pipeline:</p> <pre data-lang="sh" class="language-sh z-code"><code class="language-sh" data-lang="sh"><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">mlir-opt</span></span><span class="z-meta z-function-call z-arguments z-shell"> sample.mlir <span class="z-punctuation z-separator z-continuation z-line z-shell">\ </span></span></span><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-arguments z-shell"><span class="z-variable z-parameter z-option z-shell"><span class="z-punctuation z-definition z-parameter z-shell"> --</span>pass-pipeline</span><span class="z-keyword z-operator z-assignment z-option z-shell">=</span><span class="z-string z-quoted z-double z-shell"><span class="z-punctuation z-definition z-string z-begin z-shell">&quot;</span>builtin.module( </span></span></span><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-arguments z-shell"><span class="z-string z-quoted z-double z-shell"> gpu-kernel-outlining, </span></span></span><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-arguments z-shell"><span class="z-string z-quoted z-double z-shell"> gpu.module(convert-gpu-to-nvvm), </span></span></span><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-arguments z-shell"><span class="z-string z-quoted z-double z-shell"> nvvm-attach-target{chip=sm_61}, </span></span></span><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-arguments z-shell"><span class="z-string z-quoted z-double z-shell"> gpu.module(convert-gpu-to-nvvm), </span></span></span><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-arguments z-shell"><span class="z-string z-quoted z-double z-shell"> gpu-to-llvm, </span></span></span><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-arguments z-shell"><span class="z-string z-quoted z-double z-shell"> gpu-module-to-binary </span></span></span><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-arguments z-shell"><span class="z-string z-quoted z-double z-shell"> )<span class="z-punctuation z-definition z-string z-end z-shell">&quot;</span></span> <span class="z-punctuation z-separator z-continuation z-line z-shell">\ </span></span></span><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-arguments z-shell"><span class="z-variable z-parameter z-option z-shell"><span class="z-punctuation z-definition z-parameter z-shell"> -</span>o</span> example-nvvm.mlir</span> </span><span class="z-source z-shell z-bash"> </span></code></pre> <table><thead><tr><th>Pass</th><th>Description</th></tr></thead><tbody> <tr><td><code>gpu-kernel-outlining</code></td><td>Extracts GPU kernel bodies into separate <code>gpu.module</code>s</td></tr> <tr><td><code>gpu.module(convert-gpu-to-nvvm)</code></td><td>Converts high-level GPU ops (e.g. <code>gpu.launch_func</code>) into NVVM dialect</td></tr> <tr><td><code>nvvm-attach-target{chip=sm_61}</code></td><td>Specifies the NVIDIA GPU target architecture (sm_61 = Pascal, e.g., GTX 1080)</td></tr> <tr><td><code>gpu-to-llvm</code></td><td>Converts the host-side GPU ops into the LLVM dialect</td></tr> <tr><td><code>gpu-module-to-binary</code></td><td>Compiles the NVVM kernel into a binary blob and embeds it in the module as a <code>gpu.binary</code></td></tr> </tbody></table> <p>After the above transformations we get :</p> <pre data-lang="asm" class="language-asm z-code"><code class="language-asm" data-lang="asm"><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">b</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">}</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">f</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-source z-assembly">-</span><span class="z-entity z-name z-function z-assembly">&gt;</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span> </span><span class="z-source z-assembly"> <span class="z-support z-function z-directive z-assembly">%0</span> <span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-constant z-character z-decimal z-assembly">1</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">:</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">x</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">:</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-constant z-character z-decimal z-assembly">1</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-constant z-character z-decimal z-assembly">42</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">:</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">:</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">h</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">f</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">:</span><span class="z-entity z-name z-function z-assembly">:</span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">b</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-keyword z-control z-assembly">in</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-support z-function z-directive z-assembly">%0</span><span class="z-source z-assembly">,</span> <span class="z-support z-function z-directive z-assembly">%0</span><span class="z-source z-assembly">,</span> <span class="z-support z-function z-directive z-assembly">%0</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">h</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-keyword z-control z-assembly">in</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-support z-function z-directive z-assembly">%0</span><span class="z-source z-assembly">,</span> <span class="z-support z-function z-directive z-assembly">%0</span><span class="z-source z-assembly">,</span> <span class="z-support z-function z-directive z-assembly">%0</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">:</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-entity z-name z-function z-assembly"> </span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-constant z-character z-decimal z-assembly">1</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">:</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">}</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">b</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">y</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-source z-assembly">[</span><span class="z-entity z-name z-function z-assembly">#</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">b</span><span class="z-entity z-name z-function z-assembly">j</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">&lt;</span><span class="z-entity z-name z-function z-assembly">#</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">&lt;</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">h</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-string z-assembly">&quot;sm_61&quot;</span><span class="z-entity z-name z-function z-assembly">&gt;</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">I</span><span class="z-entity z-name z-function z-assembly">S</span><span class="z-entity z-name z-function z-assembly">A</span><span class="z-entity z-name z-function z-assembly">T</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">B</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">y</span><span class="z-entity z-name z-function z-assembly">T</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">I</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">M</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">44</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">:</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-source z-assembly">,</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">L</span><span class="z-entity z-name z-function z-assembly">L</span><span class="z-entity z-name z-function z-assembly">V</span><span class="z-entity z-name z-function z-assembly">M</span><span class="z-entity z-name z-function z-assembly">I</span><span class="z-entity z-name z-function z-assembly">R</span><span class="z-entity z-name z-function z-assembly">T</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">I</span><span class="z-entity z-name z-function z-assembly">S</span><span class="z-entity z-name z-function z-assembly">A</span><span class="z-entity z-name z-function z-assembly">T</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">I</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">M</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">1</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">:</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">}</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-string z-assembly">&quot;&lt;binary blob&gt;&quot;</span><span class="z-source z-assembly">]</span><span class="z-entity z-name z-function z-assembly"> </span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">}</span> </span></code></pre> <h2 id="4-mlir-to-llvm-ir">4. MLIR to LLVM IR<a class="zola-anchor" href="#4-mlir-to-llvm-ir" aria-label="Anchor link for: 4-mlir-to-llvm-ir" style="visibility: hidden;"></a> </h2> <p>We use the mlir-translate tool to convert our MLIR NVVM dialect file into LLVM IR:</p> <pre data-lang="sh" class="language-sh z-code"><code class="language-sh" data-lang="sh"><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">mlir-translate</span></span><span class="z-meta z-function-call z-arguments z-shell"> example-nvvm.mlir<span class="z-variable z-parameter z-option z-shell"><span class="z-punctuation z-definition z-parameter z-shell"> --</span>mlir-to-llvmir</span><span class="z-variable z-parameter z-option z-shell"><span class="z-punctuation z-definition z-parameter z-shell"> -</span>o</span> example.ll</span> </span></code></pre> <p>This command generates the LLVM IR representation of our GPU kernel and host interaction code, saved in example.ll.</p> <p>Here’s an excerpt of the generated LLVM IR:</p> <pre data-lang="asm" class="language-asm z-code"><code class="language-asm" data-lang="asm"><span class="z-source z-assembly"><span class="z-comment z-assembly">; ModuleID = &#39;LLVMDialectModule&#39;</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">f</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-string z-assembly">&quot;LLVMDialectModule&quot;</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-support z-function z-directive z-assembly">%0</span> <span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">y</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span><span class="z-entity z-name z-function z-assembly">}</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">b</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">y</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-source z-assembly">[</span><span class="z-constant z-character z-decimal z-assembly">2832</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">x</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">8</span><span class="z-source z-assembly">]</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-string z-assembly">&quot; ... &quot;</span><span class="z-comment z-assembly"> ; Embedded binary kernel blob</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">l</span> <span class="z-support z-function z-directive z-assembly">global</span> <span class="z-entity z-name z-function z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">b</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">g</span> <span class="z-support z-function z-directive z-assembly">global</span> <span class="z-source z-assembly">[</span><span class="z-constant z-character z-decimal z-assembly">1</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">x</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">}</span><span class="z-source z-assembly">]</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-source z-assembly">[</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">}</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">123</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">}</span> </span><span class="z-source z-assembly"><span class="z-source z-assembly">]</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">b</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">g</span> <span class="z-support z-function z-directive z-assembly">global</span> <span class="z-source z-assembly">[</span><span class="z-constant z-character z-decimal z-assembly">1</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">x</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">}</span><span class="z-source z-assembly">]</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-source z-assembly">[</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">}</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">123</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">}</span> </span><span class="z-source z-assembly"><span class="z-source z-assembly">]</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-source z-assembly">[</span><span class="z-constant z-character z-decimal z-assembly">12</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">x</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">8</span><span class="z-source z-assembly">]</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-string z-assembly">&quot;main_kernel\00&quot;</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">f</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">y</span><span class="z-entity z-name z-function z-assembly">:</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-constant z-character z-decimal z-assembly">1</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">a</span> <span class="z-support z-function z-directive z-assembly">%0</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">align</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">8</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-constant z-character z-decimal z-assembly">2</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">0</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">align</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">8</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-constant z-character z-decimal z-assembly">3</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">align</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">8</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-constant z-character z-decimal z-assembly">4</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-keyword z-control z-assembly">call</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">M</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">G</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">F</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-constant z-character z-decimal z-assembly">3</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">)</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-constant z-character z-decimal z-assembly">5</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-keyword z-control z-assembly">call</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">S</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">C</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-entity z-name z-function z-assembly">)</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-keyword z-control z-assembly">call</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">L</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">h</span><span class="z-entity z-name z-function z-assembly">K</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-constant z-character z-decimal z-assembly">4</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">1</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">1</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">1</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">1</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">1</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">1</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">0</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-constant z-character z-decimal z-assembly">5</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-constant z-character z-decimal z-assembly">2</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">0</span><span class="z-entity z-name z-function z-assembly">)</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-keyword z-control z-assembly">call</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">S</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">S</span><span class="z-entity z-name z-function z-assembly">y</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">h</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">z</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-constant z-character z-decimal z-assembly">5</span><span class="z-entity z-name z-function z-assembly">)</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-keyword z-control z-assembly">call</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">S</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">D</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">y</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">%</span><span class="z-constant z-character z-decimal z-assembly">5</span><span class="z-entity z-name z-function z-assembly">)</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-keyword z-control z-assembly">ret</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">42</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">}</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">f</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">section</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-string z-assembly">&quot;.text.startup&quot;</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">y</span><span class="z-entity z-name z-function z-assembly">:</span> </span><span class="z-source z-assembly"> <span class="z-support z-function z-directive z-assembly">%0</span> <span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-keyword z-control z-assembly">call</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">M</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">L</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">b</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">y</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">2832</span><span class="z-entity z-name z-function z-assembly">)</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span> <span class="z-support z-function z-directive z-assembly">%0</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">align</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">8</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-keyword z-control z-assembly">ret</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">d</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">}</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">M</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">L</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-entity z-name z-function z-assembly">)</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">f</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">section</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-string z-assembly">&quot;.text.startup&quot;</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">y</span><span class="z-entity z-name z-function z-assembly">:</span> </span><span class="z-source z-assembly"> <span class="z-support z-function z-directive z-assembly">%0</span> <span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">k</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">_</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">align</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">8</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-keyword z-control z-assembly">call</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">M</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">U</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-support z-function z-directive z-assembly">ptr</span> <span class="z-support z-function z-directive z-assembly">%0</span><span class="z-entity z-name z-function z-assembly">)</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-keyword z-control z-assembly">ret</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">d</span> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">}</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">M</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">U</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly">)</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">M</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">G</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">F</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly">)</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">S</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">C</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-entity z-name z-function z-assembly">)</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">L</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">h</span><span class="z-entity z-name z-function z-assembly">K</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">6</span><span class="z-entity z-name z-function z-assembly">4</span><span class="z-entity z-name z-function z-assembly">)</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">S</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">S</span><span class="z-entity z-name z-function z-assembly">y</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">h</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">n</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">z</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly">)</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">@</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">g</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">S</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">D</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">y</span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-support z-function z-directive z-assembly">ptr</span><span class="z-entity z-name z-function z-assembly">)</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">!</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">d</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-variable z-parameter z-register z-assembly">flags</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">!</span><span class="z-entity z-name z-function z-assembly">{</span><span class="z-entity z-name z-function z-assembly">!</span><span class="z-constant z-character z-decimal z-assembly">0</span><span class="z-entity z-name z-function z-assembly">}</span> </span><span class="z-source z-assembly"> </span><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">!</span><span class="z-constant z-character z-decimal z-assembly">0</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">!</span><span class="z-entity z-name z-function z-assembly">{</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">2</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">!</span><span class="z-string z-assembly">&quot;Debug Info Version&quot;</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-constant z-character z-decimal z-assembly">3</span><span class="z-entity z-name z-function z-assembly">}</span> </span></code></pre> <ul> <li> <p>The kernel binary is embedded as a large internal constant <code>(@main_kernel_binary)</code>, representing the compiled GPU code.</p> </li> <li> <p>The <code>@main_kernel_module</code> global pointer stores the loaded module handle.</p> </li> <li> <p>Global constructors and destructors (<code>@llvm.global_ctors</code> and <code>@llvm.global_dtors</code>) ensure the kernel module is loaded and unloaded properly at program startup and shutdown.</p> </li> <li> <p>The <code>@main</code> function performs the following:</p> <ul> <li> <p>Loads the kernel module handle.</p> </li> <li> <p>Retrieves the kernel function by name (main_kernel).</p> </li> <li> <p>Creates a GPU stream.</p> </li> <li> <p>Launches the kernel with specified grid and block dimensions.</p> </li> <li> <p>Synchronizes and destroys the stream.</p> </li> <li> <p>Returns an arbitrary integer (42).</p> </li> </ul> </li> <li> <p>The loader and unloader functions handle module lifecycle via calls to external runtime functions (mgpuModuleLoad, mgpuModuleUnload, etc.).</p> </li> </ul> <p>This LLVM IR forms the basis for generating the final PTX assembly in the next compilation step.</p> <h2 id="5-define-the-mgpuruntime-interface-for-mlir-gpu-execution">5. Define the mgpuruntime Interface for MLIR GPU Execution<a class="zola-anchor" href="#5-define-the-mgpuruntime-interface-for-mlir-gpu-execution" aria-label="Anchor link for: 5-define-the-mgpuruntime-interface-for-mlir-gpu-execution" style="visibility: hidden;"></a> </h2> <p>MLIR’s GPU dialect allows the lowering of GPU kernels, but to actually execute those kernels, you need a runtime implementation that can load modules, launch kernels, and manage CUDA streams. The following C++ code implements a minimal runtime using CUDA’s <strong>Driver API</strong>.</p> <pre data-lang="cpp" class="language-cpp z-code"><code class="language-cpp" data-lang="cpp"><span class="z-source z-c++"><span class="z-meta z-preprocessor z-include z-c++"><span class="z-keyword z-control z-import z-include z-c++">#include</span> <span class="z-string z-quoted z-other z-lt-gt z-include z-c++"><span class="z-punctuation z-definition z-string z-begin z-c++">&lt;</span>cuda.h<span class="z-punctuation z-definition z-string z-end z-c++">&gt;</span></span> </span></span><span class="z-source z-c++"><span class="z-meta z-preprocessor z-include z-c++"><span class="z-keyword z-control z-import z-include z-c++">#include</span> <span class="z-string z-quoted z-other z-lt-gt z-include z-c++"><span class="z-punctuation z-definition z-string z-begin z-c++">&lt;</span>iostream<span class="z-punctuation z-definition z-string z-end z-c++">&gt;</span></span> </span></span><span class="z-source z-c++"><span class="z-meta z-preprocessor z-include z-c++"><span class="z-keyword z-control z-import z-include z-c++">#include</span> <span class="z-string z-quoted z-other z-lt-gt z-include z-c++"><span class="z-punctuation z-definition z-string z-begin z-c++">&lt;</span>iomanip<span class="z-punctuation z-definition z-string z-end z-c++">&gt;</span></span> </span></span><span class="z-source z-c++"> </span><span class="z-source z-c++"><span class="z-storage z-modifier z-c++">extern</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span>C<span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span> <span class="z-storage z-type z-c">void</span><span class="z-keyword z-operator z-c++">*</span> <span class="z-meta z-function z-c++"><span class="z-meta z-toc-list z-full-identifier z-c++"><span class="z-entity z-name z-function z-c++">mgpuModuleLoad</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-storage z-type z-c++">void</span><span class="z-keyword z-operator z-c">*</span> <span class="z-variable z-parameter z-c++">data</span><span class="z-punctuation z-separator z-c++">,</span> <span class="z-storage z-type z-c">long</span> <span class="z-storage z-type z-c">long</span> <span class="z-variable z-parameter z-c++">size</span><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-meta z-function z-c++"> </span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-begin z-c++">{</span></span></span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> CUmodule mod<span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> CUdevice device<span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> CUcontext ctx<span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> CUresult res <span class="z-keyword z-operator z-assignment z-c">=</span> <span class="z-meta z-function-call z-c++"><span class="z-variable z-function z-c++">cuInit</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-constant z-numeric z-integer z-decimal z-c++">0</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-keyword z-control z-c++">if</span> <span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span>res <span class="z-keyword z-operator z-comparison z-c">!=</span> CUDA_SUCCESS<span class="z-punctuation z-section z-group z-end z-c++">)</span></span> <span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-begin z-c++">{</span> </span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-meta z-block z-c++"> std<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>cerr <span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span>cuInit failed with error code <span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span> <span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span> res <span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span><span class="z-constant z-character z-escape z-c">\n</span><span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-meta z-block z-c++"> <span class="z-keyword z-control z-flow z-return z-c++">return</span> <span class="z-constant z-language z-c++">nullptr</span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-meta z-block z-c++"> <span class="z-punctuation z-section z-block z-end z-c++">}</span></span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-comment z-line z-double-slash z-c"><span class="z-punctuation z-definition z-comment z-c">//</span> Get CUDA device </span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> res <span class="z-keyword z-operator z-assignment z-c">=</span> <span class="z-meta z-function-call z-c++"><span class="z-variable z-function z-c++">cuDeviceGet</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-keyword z-operator z-c">&amp;</span>device<span class="z-punctuation z-separator z-c++">,</span> <span class="z-constant z-numeric z-integer z-decimal z-c++">0</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> <span class="z-comment z-line z-double-slash z-c"><span class="z-punctuation z-definition z-comment z-c">//</span> Use device 0 </span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-keyword z-control z-c++">if</span> <span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span>res <span class="z-keyword z-operator z-comparison z-c">!=</span> CUDA_SUCCESS<span class="z-punctuation z-section z-group z-end z-c++">)</span></span> <span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-begin z-c++">{</span> </span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-meta z-block z-c++"> std<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>cerr <span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span>cuDeviceGet failed with error code <span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span> <span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span> res <span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span><span class="z-constant z-character z-escape z-c">\n</span><span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-meta z-block z-c++"> <span class="z-keyword z-control z-flow z-return z-c++">return</span> <span class="z-constant z-language z-c++">nullptr</span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-meta z-block z-c++"> <span class="z-punctuation z-section z-block z-end z-c++">}</span></span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-comment z-line z-double-slash z-c"><span class="z-punctuation z-definition z-comment z-c">//</span> Create context </span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> res <span class="z-keyword z-operator z-assignment z-c">=</span> <span class="z-meta z-function-call z-c++"><span class="z-variable z-function z-c++">cuCtxCreate</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-keyword z-operator z-c">&amp;</span>ctx<span class="z-punctuation z-separator z-c++">,</span> <span class="z-constant z-numeric z-integer z-decimal z-c++">0</span><span class="z-punctuation z-separator z-c++">,</span> device</span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-keyword z-control z-c++">if</span> <span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span>res <span class="z-keyword z-operator z-comparison z-c">!=</span> CUDA_SUCCESS<span class="z-punctuation z-section z-group z-end z-c++">)</span></span> <span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-begin z-c++">{</span> </span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-meta z-block z-c++"> std<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>cerr <span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span>cuCtxCreate failed with error code <span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span> <span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span> res <span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span><span class="z-constant z-character z-escape z-c">\n</span><span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-meta z-block z-c++"> <span class="z-keyword z-control z-flow z-return z-c++">return</span> <span class="z-constant z-language z-c++">nullptr</span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-meta z-block z-c++"> <span class="z-punctuation z-section z-block z-end z-c++">}</span></span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-comment z-line z-double-slash z-c"><span class="z-punctuation z-definition z-comment z-c">//</span>CUresult initRes = cuInit(0); </span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> res <span class="z-keyword z-operator z-assignment z-c">=</span> <span class="z-meta z-function-call z-c++"><span class="z-variable z-function z-c++">cuModuleLoadData</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-keyword z-operator z-c">&amp;</span>mod<span class="z-punctuation z-separator z-c++">,</span> data</span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> std<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>cout<span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span><span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span>RES: <span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span><span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span>res<span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span><span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span><span class="z-constant z-character z-escape z-c">\n</span><span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-keyword z-control z-c++">if</span> <span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span>res <span class="z-keyword z-operator z-comparison z-c">!=</span> CUDA_SUCCESS<span class="z-punctuation z-section z-group z-end z-c++">)</span></span> std<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>cerr <span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span>cuModuleLoadData failed<span class="z-constant z-character z-escape z-c">\n</span><span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-keyword z-control z-flow z-return z-c++">return</span> mod<span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"></span></span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-end z-c++">}</span></span></span> </span><span class="z-source z-c++"> </span><span class="z-source z-c++"><span class="z-storage z-modifier z-c++">extern</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span>C<span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span> <span class="z-storage z-type z-c">void</span> <span class="z-meta z-function z-c++"><span class="z-meta z-toc-list z-full-identifier z-c++"><span class="z-entity z-name z-function z-c++">mgpuModuleUnload</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-storage z-type z-c++">void</span><span class="z-keyword z-operator z-c">*</span> <span class="z-variable z-parameter z-c++">mod</span><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-meta z-function z-c++"> </span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-begin z-c++">{</span></span></span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-meta z-function-call z-c++"><span class="z-variable z-function z-c++">cuModuleUnload</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span>CUmodule<span class="z-punctuation z-section z-group z-end z-c++">)</span></span>mod</span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"></span></span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-end z-c++">}</span></span></span> </span><span class="z-source z-c++"> </span><span class="z-source z-c++"><span class="z-storage z-modifier z-c++">extern</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span>C<span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span> <span class="z-storage z-type z-c">void</span><span class="z-keyword z-operator z-c++">*</span> <span class="z-meta z-function z-c++"><span class="z-meta z-toc-list z-full-identifier z-c++"><span class="z-entity z-name z-function z-c++">mgpuModuleGetFunction</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-storage z-type z-c++">void</span><span class="z-keyword z-operator z-c">*</span> <span class="z-variable z-parameter z-c++">mod</span><span class="z-punctuation z-separator z-c++">,</span> <span class="z-storage z-modifier z-c++">const</span> <span class="z-storage z-type z-c">char</span><span class="z-keyword z-operator z-c">*</span> <span class="z-variable z-parameter z-c++">name</span><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-meta z-function z-c++"> </span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-begin z-c++">{</span></span></span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> CUfunction func<span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> CUresult res <span class="z-keyword z-operator z-assignment z-c">=</span> <span class="z-meta z-function-call z-c++"><span class="z-variable z-function z-c++">cuModuleGetFunction</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-keyword z-operator z-c">&amp;</span>func<span class="z-punctuation z-separator z-c++">,</span> <span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span>CUmodule<span class="z-punctuation z-section z-group z-end z-c++">)</span></span>mod<span class="z-punctuation z-separator z-c++">,</span> name</span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> std<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>cout<span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span><span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span>Res <span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span><span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span>res <span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span><span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span><span class="z-constant z-character z-escape z-c">\n</span><span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-keyword z-control z-c++">if</span> <span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span>res <span class="z-keyword z-operator z-comparison z-c">!=</span> CUDA_SUCCESS<span class="z-punctuation z-section z-group z-end z-c++">)</span></span> std<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>cerr <span class="z-keyword z-operator z-arithmetic z-c">&lt;&lt;</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span>cuModuleGetFunction failed<span class="z-constant z-character z-escape z-c">\n</span><span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-keyword z-control z-flow z-return z-c++">return</span> func<span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"></span></span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-end z-c++">}</span></span></span> </span><span class="z-source z-c++"> </span><span class="z-source z-c++"><span class="z-storage z-modifier z-c++">extern</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span>C<span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span> <span class="z-storage z-type z-c">void</span><span class="z-keyword z-operator z-c++">*</span> <span class="z-meta z-function z-c++"><span class="z-meta z-toc-list z-full-identifier z-c++"><span class="z-entity z-name z-function z-c++">mgpuStreamCreate</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-meta z-function z-c++"> </span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-begin z-c++">{</span></span></span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> CUstream stream<span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-meta z-function-call z-c++"><span class="z-variable z-function z-c++">cuStreamCreate</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-keyword z-operator z-c">&amp;</span>stream<span class="z-punctuation z-separator z-c++">,</span> <span class="z-constant z-numeric z-integer z-decimal z-c++">0</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-keyword z-control z-flow z-return z-c++">return</span> stream<span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"></span></span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-end z-c++">}</span></span></span> </span><span class="z-source z-c++"> </span><span class="z-source z-c++"><span class="z-storage z-modifier z-c++">extern</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span>C<span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span> <span class="z-storage z-type z-c">void</span> <span class="z-meta z-function z-c++"><span class="z-meta z-toc-list z-full-identifier z-c++"><span class="z-entity z-name z-function z-c++">mgpuLaunchKernel</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-storage z-type z-c++">void</span><span class="z-keyword z-operator z-c">*</span> <span class="z-variable z-parameter z-c++">func</span><span class="z-punctuation z-separator z-c++">,</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"> <span class="z-storage z-type z-c">long</span> <span class="z-storage z-type z-c">long</span> <span class="z-variable z-parameter z-c++">gx</span><span class="z-punctuation z-separator z-c++">,</span> <span class="z-storage z-type z-c">long</span> <span class="z-storage z-type z-c">long</span> <span class="z-variable z-parameter z-c++">gy</span><span class="z-punctuation z-separator z-c++">,</span> <span class="z-storage z-type z-c">long</span> <span class="z-storage z-type z-c">long</span> <span class="z-variable z-parameter z-c++">gz</span><span class="z-punctuation z-separator z-c++">,</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"> <span class="z-storage z-type z-c">long</span> <span class="z-storage z-type z-c">long</span> <span class="z-variable z-parameter z-c++">bx</span><span class="z-punctuation z-separator z-c++">,</span> <span class="z-storage z-type z-c">long</span> <span class="z-storage z-type z-c">long</span> <span class="z-variable z-parameter z-c++">by</span><span class="z-punctuation z-separator z-c++">,</span> <span class="z-storage z-type z-c">long</span> <span class="z-storage z-type z-c">long</span> <span class="z-variable z-parameter z-c++">bz</span><span class="z-punctuation z-separator z-c++">,</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"> <span class="z-storage z-type z-c">int</span> <span class="z-variable z-parameter z-c++">shared</span><span class="z-punctuation z-separator z-c++">,</span> <span class="z-storage z-type z-c++">void</span><span class="z-keyword z-operator z-c">*</span> <span class="z-variable z-parameter z-c++">stream</span><span class="z-punctuation z-separator z-c++">,</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"> <span class="z-storage z-type z-c++">void</span><span class="z-keyword z-operator z-c">*</span> <span class="z-variable z-parameter z-c++">params</span><span class="z-punctuation z-separator z-c++">,</span> <span class="z-storage z-type z-c++">void</span><span class="z-keyword z-operator z-c">*</span><span class="z-punctuation z-separator z-c++">,</span> <span class="z-storage z-type z-c">long</span> <span class="z-variable z-parameter z-c++">long</span><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-meta z-function z-c++"> </span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-begin z-c++">{</span></span></span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-storage z-type z-c">void</span><span class="z-keyword z-operator z-c">*</span><span class="z-keyword z-operator z-c">*</span> param_array <span class="z-keyword z-operator z-assignment z-c">=</span> <span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span><span class="z-storage z-type z-c">void</span><span class="z-keyword z-operator z-c">*</span><span class="z-keyword z-operator z-c">*</span><span class="z-punctuation z-section z-group z-end z-c++">)</span></span>params<span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-meta z-function-call z-c++"><span class="z-variable z-function z-c++">cuLaunchKernel</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span>CUfunction<span class="z-punctuation z-section z-group z-end z-c++">)</span></span>func<span class="z-punctuation z-separator z-c++">,</span> </span></span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"> gx<span class="z-punctuation z-separator z-c++">,</span> gy<span class="z-punctuation z-separator z-c++">,</span> gz<span class="z-punctuation z-separator z-c++">,</span> </span></span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"> bx<span class="z-punctuation z-separator z-c++">,</span> by<span class="z-punctuation z-separator z-c++">,</span> bz<span class="z-punctuation z-separator z-c++">,</span> </span></span></span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"> shared<span class="z-punctuation z-separator z-c++">,</span> <span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span>CUstream<span class="z-punctuation z-section z-group z-end z-c++">)</span></span>stream<span class="z-punctuation z-separator z-c++">,</span> param_array<span class="z-punctuation z-separator z-c++">,</span> <span class="z-constant z-language z-c++">nullptr</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"></span></span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-end z-c++">}</span></span></span> </span><span class="z-source z-c++"> </span><span class="z-source z-c++"><span class="z-storage z-modifier z-c++">extern</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span>C<span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span> <span class="z-storage z-type z-c">void</span> <span class="z-meta z-function z-c++"><span class="z-meta z-toc-list z-full-identifier z-c++"><span class="z-entity z-name z-function z-c++">mgpuStreamSynchronize</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-storage z-type z-c++">void</span><span class="z-keyword z-operator z-c">*</span> <span class="z-variable z-parameter z-c++">stream</span><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-meta z-function z-c++"> </span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-begin z-c++">{</span></span></span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-meta z-function-call z-c++"><span class="z-variable z-function z-c++">cuStreamSynchronize</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span>CUstream<span class="z-punctuation z-section z-group z-end z-c++">)</span></span>stream</span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"></span></span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-end z-c++">}</span></span></span> </span><span class="z-source z-c++"> </span><span class="z-source z-c++"><span class="z-storage z-modifier z-c++">extern</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span>C<span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span> <span class="z-storage z-type z-c">void</span> <span class="z-meta z-function z-c++"><span class="z-meta z-toc-list z-full-identifier z-c++"><span class="z-entity z-name z-function z-c++">mgpuStreamDestroy</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function z-parameters z-c++"><span class="z-meta z-group z-c++"><span class="z-storage z-type z-c++">void</span><span class="z-keyword z-operator z-c">*</span> <span class="z-variable z-parameter z-c++">stream</span><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-meta z-function z-c++"> </span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-begin z-c++">{</span></span></span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"> <span class="z-meta z-function-call z-c++"><span class="z-variable z-function z-c++">cuStreamDestroy</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span>CUstream<span class="z-punctuation z-section z-group z-end z-c++">)</span></span>stream</span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></span></span><span class="z-source z-c++"><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"></span></span><span class="z-meta z-function z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-end z-c++">}</span></span></span> </span></code></pre> <p>This runtime layer allows the MLIR-compiled GPU code to be executed without needing external CUDA host code. You can compile and execute GPU kernels generated from MLIR entirely within a runtime like this, which makes it highly customizable and useful for research, prototyping, or building new compiler backends.</p> <h2 id="6-compiling-llvm-ir-to-executable">6. Compiling LLVM IR to Executable<a class="zola-anchor" href="#6-compiling-llvm-ir-to-executable" aria-label="Anchor link for: 6-compiling-llvm-ir-to-executable" style="visibility: hidden;"></a> </h2> <p>After generating the LLVM IR (example.ll), we compile it together with the GPU runtime source (mgpuruntime.cc) using clang++. This produces a host executable that will load and launch our GPU kernel.</p> <pre data-lang="sh" class="language-sh z-code"><code class="language-sh" data-lang="sh"><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">clang++</span></span><span class="z-meta z-function-call z-arguments z-shell"> mgpuruntime.cc example.ll<span class="z-variable z-parameter z-option z-shell"><span class="z-punctuation z-definition z-parameter z-shell"> -</span>O3</span><span class="z-variable z-parameter z-option z-shell"><span class="z-punctuation z-definition z-parameter z-shell"> -</span>lcuda</span><span class="z-variable z-parameter z-option z-shell"><span class="z-punctuation z-definition z-parameter z-shell"> -</span>o</span> run_kernel</span> </span></code></pre> <h2 id="7-running-the-executable">7. Running the Executable<a class="zola-anchor" href="#7-running-the-executable" aria-label="Anchor link for: 7-running-the-executable" style="visibility: hidden;"></a> </h2> <p>Finally, we run the compiled program:</p> <pre data-lang="sh" class="language-sh z-code"><code class="language-sh" data-lang="sh"><span class="z-source z-shell z-bash"><span class="z-meta z-function-call z-shell"><span class="z-variable z-function z-shell">./run_kernel</span></span> </span></code></pre> <p>The output should be:</p> <pre class="z-code"><code><span class="z-text z-plain">Hi There: </span></code></pre> 2025-04-21T00:00:00+00:00 2025-04-21T00:00:00+00:00 https://johnmaxrin.github.io/posts/musings-1/ <p>Today, the world pauses. We lost a shepherd, a father, a quiet strength — our beloved Pope has gone home.</p> <p>And in this moment, I'm reminded how some stories… don’t end. They simply change form. Like a soft candle flickering out, not into darkness, but into something eternal.</p> <p>We often seek closure — in people, in relationships, in moments. But maybe beauty lies in the unfinished. In the half-spoken goodbyes, in silent prayers, in the gentle presence that lingers even after someone is gone.</p> <p>His story may feel like it ended today. But the truth is, it lives on — in every soul he inspired, in every heart he helped lift. Just as some people leave our lives quietly, not with full stops, but with ellipses… Their story lives within us, woven into the very fabric of who we are.</p> <p>Some stories remain unfinished. And maybe… that’s what makes them eternal.</p> 2025-03-28T00:00:00+00:00 2025-03-28T00:00:00+00:00 https://johnmaxrin.github.io/posts/mlirccllvmptrs/ <p>I was working on lowering our DSL’s IR directly to LLVM IR! Yeah, a lot of high-level information had to be stripped away during lowering to LLVM, but I was convinced the adventure was worth it. When it came to representing our Graph Type in LLVM, the only option I could think of was using a struct and that’s where the real story begins.</p> <h1 id="starplat-and-graph-type">Starplat and Graph Type<a class="zola-anchor" href="#starplat-and-graph-type" aria-label="Anchor link for: starplat-and-graph-type" style="visibility: hidden;"></a> </h1> <p>Starplat is a DSL built for graph analytics, and having a type named Graph is an absolute must. Our Graph type isn’t just a simple container it does a ton of things: storing graphs in different representations, retrieving the total number of nodes and edges, and much more.</p> <p>By default, we represent the graph in Compressed Sparse Row (CSR) format. We also have a set of associated functions, like attachNodeProperty. Think of Single Source Shortest Path, You need a "DISTANCE" value attached to every node. That’s exactly what attachNodeProperty does: it assigns node properties dynamically, making the graph more than just an abstract structure.</p> <pre data-lang="cpp" class="language-cpp z-code"><code class="language-cpp" data-lang="cpp"><span class="z-source z-c++">g<span class="z-punctuation z-accessor z-dot z-c++">.</span><span class="z-meta z-method-call z-c++"><span class="z-variable z-function z-member z-c++">attachNodeProperty</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++">dist<span class="z-keyword z-operator z-assignment z-c">=</span>INF</span></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></code></pre> <p>Now, the real challenge? Representing all of this in LLVM. <br> <strong>The Plan </strong> <br> First, we need to understand how to deal with arrays in LLVM. We'll start small, Creating a simple structure with a single int32 member. If we can successfully load and store values into this member, we’ll take the next step: replacing it with an array. Then, we scale up, Adding multiple members to the struct, manipulating values inside it, and ensuring everything works as expected.</p> <p>Sounds simple enough, right? Well, let’s find out.</p> <h2 id="struct-with-single-int32">Struct with Single int32<a class="zola-anchor" href="#struct-with-single-int32" aria-label="Anchor link for: struct-with-single-int32" style="visibility: hidden;"></a> </h2> <p>In C, We can define a simple struct like this:</p> <pre data-lang="c" class="language-c z-code"><code class="language-c" data-lang="c"><span class="z-source z-c"><span class="z-meta z-struct z-c"><span class="z-storage z-type z-c">struct</span> </span></span><span class="z-source z-c"><span class="z-meta z-struct z-c"><span class="z-meta z-block z-c"><span class="z-punctuation z-section z-block z-begin z-c">{</span></span></span><span class="z-meta z-struct z-c"><span class="z-meta z-block z-c"> </span></span></span><span class="z-source z-c"><span class="z-meta z-struct z-c"><span class="z-meta z-block z-c"> <span class="z-storage z-type z-c">int</span> a<span class="z-punctuation z-terminator z-c">;</span> </span></span></span><span class="z-source z-c"><span class="z-meta z-struct z-c"><span class="z-meta z-block z-c"></span></span><span class="z-meta z-struct z-c"><span class="z-meta z-block z-c"><span class="z-punctuation z-section z-block z-end z-c">}</span></span></span><span class="z-punctuation z-terminator z-c">;</span> </span></code></pre> <p>In LLVM, this translates to:</p> <pre data-lang="asm" class="language-asm z-code"><code class="language-asm" data-lang="asm"><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">%</span><span class="z-entity z-name z-function z-assembly">G</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">h</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">y</span><span class="z-entity z-name z-function z-assembly">p</span><span class="z-entity z-name z-function z-assembly">e</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">{</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly">}</span> </span></code></pre> <p>Now that we have the mapping, the next step is to <strong> emit a graph struct using MLIR.</strong></p> <pre data-lang="cpp" class="language-cpp z-code"><code class="language-cpp" data-lang="cpp"><span class="z-source z-c++"><span class="z-storage z-type z-c">auto</span> int32ty <span class="z-keyword z-operator z-assignment z-c">=</span> builder<span class="z-punctuation z-accessor z-dot z-c++">.</span><span class="z-meta z-method-call z-c++"><span class="z-variable z-function z-member z-c++">getI32Type</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-method-call z-c++"></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span><span class="z-source z-c++"><span class="z-storage z-type z-c">auto</span> structType <span class="z-keyword z-operator z-assignment z-c">=</span> <span class="z-meta z-function-call z-c++">LLVM<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>LLVMStructType<span class="z-punctuation z-accessor z-double-colon z-c++">::</span><span class="z-variable z-function z-c++">getIdentified</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-keyword z-operator z-c">&amp;</span>context<span class="z-punctuation z-separator z-c++">,</span> <span class="z-string z-quoted z-double z-c"><span class="z-punctuation z-definition z-string z-begin z-c">&quot;</span>Graph<span class="z-punctuation z-definition z-string z-end z-c">&quot;</span></span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span><span class="z-source z-c++">structType<span class="z-punctuation z-accessor z-dot z-c++">.</span><span class="z-meta z-method-call z-c++"><span class="z-variable z-function z-member z-c++">setBody</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++"><span class="z-meta z-block z-c++"><span class="z-punctuation z-section z-block z-begin z-c++">{</span><span class="z-meta z-function-call z-c++">LLVM<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>LLVMPointerType<span class="z-punctuation z-accessor z-double-colon z-c++">::</span><span class="z-variable z-function z-c++">get</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-keyword z-operator z-c">&amp;</span>context</span></span><span class="z-meta z-function-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-separator z-c++">,</span> builder<span class="z-punctuation z-accessor z-dot z-c++">.</span><span class="z-meta z-method-call z-c++"><span class="z-variable z-function z-member z-c++">getI32Type</span><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span></span><span class="z-meta z-method-call z-c++"></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-section z-block z-end z-c++">}</span></span><span class="z-punctuation z-separator z-c++">,</span> <span class="z-constant z-language z-c">false</span></span></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span><span class="z-source z-c++"><span class="z-storage z-type z-c">auto</span> graphObj <span class="z-keyword z-operator z-assignment z-c">=</span> buidler<span class="z-punctuation z-accessor z-dot z-c++">.</span><span class="z-meta z-method-call z-c++"><span class="z-variable z-function z-member z-c++">create</span><span class="z-punctuation z-section z-generic z-begin z-c++">&lt;</span></span><span class="z-meta z-method-call z-c++">LLVM<span class="z-punctuation z-accessor z-double-colon z-c++">::</span>AllocaOp<span class="z-punctuation z-section z-generic z-end z-c++">&gt;</span></span><span class="z-meta z-method-call z-c++"><span class="z-punctuation z-section z-group z-begin z-c++">(</span></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++">loc<span class="z-punctuation z-separator z-c++">,</span> ptrType<span class="z-punctuation z-separator z-c++">,</span> structType<span class="z-punctuation z-separator z-c++">,</span> one</span></span><span class="z-meta z-method-call z-c++"><span class="z-meta z-group z-c++"><span class="z-punctuation z-section z-group z-end z-c++">)</span></span></span><span class="z-punctuation z-terminator z-c++">;</span> </span></code></pre> <p>This will generate LLVM IR similar to the following.</p> <pre data-lang="asm" class="language-asm z-code"><code class="language-asm" data-lang="asm"><span class="z-source z-assembly"><span class="z-entity z-name z-function z-assembly">%</span><span class="z-constant z-character z-decimal z-assembly">1</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">=</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">o</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">a</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">e</span> <span class="z-support z-function z-directive z-assembly">%0</span> <span class="z-entity z-name z-function z-assembly">x</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">!</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-entity z-name z-function z-assembly">s</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">r</span><span class="z-entity z-name z-function z-assembly">u</span><span class="z-entity z-name z-function z-assembly">c</span><span class="z-entity z-name z-function z-assembly">t</span><span class="z-entity z-name z-function z-assembly">&lt;</span><span class="z-string z-assembly">&quot;Graph&quot;</span><span class="z-source z-assembly">,</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly">&gt;</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">:</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">(</span><span class="z-entity z-name z-function z-assembly">i</span><span class="z-entity z-name z-function z-assembly">3</span><span class="z-entity z-name z-function z-assembly">2</span><span class="z-entity z-name z-function z-assembly">)</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-source z-assembly">-</span><span class="z-entity z-name z-function z-assembly">&gt;</span><span class="z-entity z-name z-function z-assembly"> </span><span class="z-entity z-name z-function z-assembly">!</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">l</span><span class="z-entity z-name z-function z-assembly">v</span><span class="z-entity z-name z-function z-assembly">m</span><span class="z-entity z-name z-function z-assembly">.</span><span class="z-support z-function z-directive z-assembly">ptr</span> </span></code></pre> <h1 id="struct-and-pointer">Struct and Pointer<a class="zola-anchor" href="#struct-and-pointer" aria-label="Anchor link for: struct-and-pointer" style="visibility: hidden;"></a> </h1> 2025-03-23T00:00:00+00:00 2025-03-23T00:00:00+00:00 https://johnmaxrin.github.io/posts/arith2llvm/ <p>Working on it.</p>