deploy: 508fa70

Author: FranciscoThiesen

cppcon2025/cppcon_2025_slides.html

@@ -811,12 +811,12 @@ <h1 id="how-fast-are-we">How fast are we? ...</h1>
 <p><strong>3.4 GB/s</strong> - 14x faster than nlohmann, 2.5x faster than Serde!</p>
 </section>
 </foreignObject></svg><svg data-marpit-svg="" viewBox="0 0 1280 720"><foreignObject width="1280" height="720"><section id="69" data-paginate="true" data-theme="default" lang="C" data-marpit-pagination="69" style="--paginate:true;--theme:default;" data-marpit-pagination-total="81">
-<h1 id="ablation-study-how-we-achieved-34-gbs">Ablation Study: How We Achieved 3.4 GB/s</h1>
+<h1 id="ablation-study-how-we-achieved-32-gbs">Ablation Study: How We Achieved 3.2 GB/s</h1>
 <p><strong>What is Ablation?</strong><br />
 From neuroscience: systematically remove parts to understand function</p>
-<p><strong>Our Approach:</strong></p>
+<p><strong>Our Approach (Apple Silicon M2):</strong></p>
 <ol>
-<li><strong>Baseline</strong>: All optimizations enabled (3,435 MB/s)</li>
+<li><strong>Baseline</strong>: All optimizations enabled (3,211 MB/s)</li>
 <li><strong>Disable one optimization</strong> at a time</li>
 <li><strong>Measure performance impact</strong></li>
 <li><strong>Calculate contribution</strong>: <code>(Baseline - Disabled) / Disabled</code></li>
@@ -847,7 +847,7 @@ <h2 id="the-power-of-compile-time">The Power of Compile-Time</h2>
 </code></pre>
 </section>
 </foreignObject></svg><svg data-marpit-svg="" viewBox="0 0 1280 720"><foreignObject width="1280" height="720"><section id="72" data-paginate="true" data-theme="default" lang="C" data-marpit-pagination="72" style="--paginate:true;--theme:default;" data-marpit-pagination-total="81">
-<h1 id="consteval-performance-impact">Consteval Performance Impact</h1>
+<h1 id="consteval-performance-impact-apple-silicon">Consteval Performance Impact (Apple Silicon)</h1>
 <table>
 <thead>
 <tr>
@@ -861,17 +861,17 @@ <h1 id="consteval-performance-impact">Consteval Performance Impact</h1>
 <tbody>
 <tr>
 <td>Twitter</td>
-<td>3,231 MB/s</td>
-<td>1,624 MB/s</td>
+<td>3,211 MB/s</td>
+<td>1,607 MB/s</td>
 <td>-50%</td>
-<td><strong>1.99x</strong></td>
+<td><strong>2.00x</strong></td>
 </tr>
 <tr>
 <td>CITM</td>
-<td>2,341 MB/s</td>
-<td>883 MB/s</td>
-<td>-62%</td>
-<td><strong>2.65x</strong></td>
+<td>2,360 MB/s</td>
+<td>978 MB/s</td>
+<td>-59%</td>
+<td><strong>2.41x</strong></td>
 </tr>
 </tbody>
 </table>
@@ -900,7 +900,7 @@ <h1 id="optimization-2-simd-string-escaping">Optimization #2: SIMD String Escapi
 </code></pre>
 </section>
 </foreignObject></svg><svg data-marpit-svg="" viewBox="0 0 1280 720"><foreignObject width="1280" height="720"><section id="74" data-paginate="true" data-theme="default" lang="C" data-marpit-pagination="74" style="--paginate:true;--theme:default;" data-marpit-pagination-total="81">
-<h1 id="simd-escaping-performance-impact">SIMD Escaping Performance Impact</h1>
+<h1 id="simd-escaping-performance-impact-apple-silicon">SIMD Escaping Performance Impact (Apple Silicon)</h1>
 <table>
 <thead>
 <tr>
@@ -914,17 +914,17 @@ <h1 id="simd-escaping-performance-impact">SIMD Escaping Performance Impact</h1>
 <tbody>
 <tr>
 <td>Twitter</td>
-<td>3,231 MB/s</td>
-<td>2,245 MB/s</td>
-<td>-31%</td>
-<td><strong>1.44x</strong></td>
+<td>3,211 MB/s</td>
+<td>2,269 MB/s</td>
+<td>-29%</td>
+<td><strong>1.42x</strong></td>
 </tr>
 <tr>
 <td>CITM</td>
-<td>2,341 MB/s</td>
-<td>2,273 MB/s</td>
-<td>-3%</td>
-<td><strong>1.03x</strong></td>
+<td>2,360 MB/s</td>
+<td>2,259 MB/s</td>
+<td>-4%</td>
+<td><strong>1.04x</strong></td>
 </tr>
 </tbody>
 </table>
@@ -954,15 +954,15 @@ <h1 id="optimization-3-fast-digit-counting">Optimization #3: Fast Digit Counting
 <tbody>
 <tr>
 <td>Twitter</td>
-<td>3,231 MB/s</td>
-<td>3,041 MB/s</td>
+<td>3,211 MB/s</td>
+<td>3,035 MB/s</td>
 <td><strong>1.06x</strong></td>
 </tr>
 <tr>
 <td>CITM</td>
-<td>2,341 MB/s</td>
-<td>1,841 MB/s</td>
-<td><strong>1.27x</strong></td>
+<td>2,360 MB/s</td>
+<td>1,767 MB/s</td>
+<td><strong>1.34x</strong></td>
 </tr>
 </tbody>
 </table>
@@ -992,8 +992,8 @@ <h1 id="optimizations-4--5-branch-hints--buffer-growth">Optimizations #4 &amp; #
 <tbody>
 <tr>
 <td>Twitter &amp; CITM</td>
-<td>~2%</td>
-<td>1.02x</td>
+<td>~1%</td>
+<td>1.01x</td>
 </tr>
 </tbody>
 </table>
@@ -1013,40 +1013,40 @@ <h1 id="combined-performance-impact">Combined Performance Impact</h1>
 <tbody>
 <tr>
 <td><strong>Consteval</strong></td>
-<td>+99% (1.99x)</td>
-<td>+165% (2.65x)</td>
+<td>+100% (2.00x)</td>
+<td>+141% (2.41x)</td>
 </tr>
 <tr>
 <td><strong>SIMD Escaping</strong></td>
-<td>+44% (1.44x)</td>
-<td>+3% (1.03x)</td>
+<td>+42% (1.42x)</td>
+<td>+4% (1.04x)</td>
 </tr>
 <tr>
 <td><strong>Fast Digits</strong></td>
 <td>+6% (1.06x)</td>
-<td>+27% (1.27x)</td>
+<td>+34% (1.34x)</td>
 </tr>
 <tr>
 <td><strong>Branch Hints</strong></td>
-<td>+1.5%</td>
-<td>+1.5%</td>
+<td>+1%</td>
+<td>+5%</td>
 </tr>
 <tr>
 <td><strong>Buffer Growth</strong></td>
-<td>+0.8%</td>
-<td>+0.8%</td>
+<td>-0.4%</td>
+<td>+2%</td>
 </tr>
 <tr>
 <td><strong>TOTAL</strong></td>
-<td><strong>~2.95x faster</strong></td>
-<td><strong>~3.5x faster</strong></td>
+<td><strong>~2.9x faster</strong></td>
+<td><strong>~3.4x faster</strong></td>
 </tr>
 </tbody>
 </table>
 <p><strong>From Baseline to Optimized:</strong></p>
 <ul>
-<li>Twitter: ~1,100 MB/s → 3,231 MB/s</li>
-<li>CITM: ~670 MB/s → 2,341 MB/s</li>
+<li>Twitter: ~1,100 MB/s → 3,211 MB/s</li>
+<li>CITM: ~700 MB/s → 2,360 MB/s</li>
 </ul>
 </section>
 </foreignObject></svg><svg data-marpit-svg="" viewBox="0 0 1280 720"><foreignObject width="1280" height="720"><section id="78" data-paginate="true" data-theme="default" lang="C" data-marpit-pagination="78" style="--paginate:true;--theme:default;" data-marpit-pagination-total="81">

cppcon2025/cppcon_2025_slides.md

@@ -958,13 +958,13 @@ _mm512_cmple_epu8_mask(word, _mm512_set1_epi8(31));
 
 ---
 
-# Ablation Study: How We Achieved 3.4 GB/s
+# Ablation Study: How We Achieved 3.2 GB/s
 
 **What is Ablation?**
 From neuroscience: systematically remove parts to understand function
 
-**Our Approach:**
-1. **Baseline**: All optimizations enabled (3,435 MB/s)
+**Our Approach (Apple Silicon M2):**
+1. **Baseline**: All optimizations enabled (3,211 MB/s)
 2. **Disable one optimization** at a time
 3. **Measure performance impact**
 4. **Calculate contribution**: `(Baseline - Disabled) / Disabled`
@@ -1001,12 +1001,12 @@ b.append_literal(username_key);  // Just memcpy!
 
 ---
 
-# Consteval Performance Impact
+# Consteval Performance Impact (Apple Silicon)
 
 | Dataset | Baseline | No Consteval | Impact | **Speedup** |
 |---------|----------|--------------|--------|-------------|
-| Twitter | 3,231 MB/s | 1,624 MB/s | -50% | **1.99x** |
-| CITM | 2,341 MB/s | 883 MB/s | -62% | **2.65x** |
+| Twitter | 3,211 MB/s | 1,607 MB/s | -50% | **2.00x** |
+| CITM | 2,360 MB/s | 978 MB/s | -59% | **2.41x** |
 
 **Twitter Example (100 tweets):**
 - 100 tweets × 15 fields = **1,500 field names**
@@ -1039,12 +1039,12 @@ if (!needs_escape)
 
 ---
 
-# SIMD Escaping Performance Impact
+# SIMD Escaping Performance Impact (Apple Silicon)
 
 | Dataset | Baseline | No SIMD | Impact | **Speedup** |
 |---------|----------|---------|--------|-------------|
-| Twitter | 3,231 MB/s | 2,245 MB/s | -31% | **1.44x** |
-| CITM | 2,341 MB/s | 2,273 MB/s | -3% | **1.03x** |
+| Twitter | 3,211 MB/s | 2,269 MB/s | -29% | **1.42x** |
+| CITM | 2,360 MB/s | 2,259 MB/s | -4% | **1.04x** |
 
 **Why Different Impact?**
 - **Twitter**: Long text fields (tweets, descriptions) → Big win
@@ -1066,8 +1066,8 @@ fast_digit_count(value);  // Bit operations + lookup table
 
 | Dataset | Baseline | No Fast Digits | **Speedup** |
 |---------|----------|----------------|-------------|
-| Twitter | 3,231 MB/s | 3,041 MB/s | **1.06x** |
-| CITM | 2,341 MB/s | 1,841 MB/s | **1.27x** |
+| Twitter | 3,211 MB/s | 3,035 MB/s | **1.06x** |
+| CITM | 2,360 MB/s | 1,767 MB/s | **1.34x** |
 
 **CITM has ~10,000+ integers!**
 
@@ -1089,7 +1089,7 @@ if (UNLIKELY(buffer_full)) {  // CPU knows this is rare
 
 | Both Optimizations | Impact | Speedup |
 |-------------------|--------|---------|
-| Twitter & CITM | ~2% | 1.02x |
+| Twitter & CITM | ~1% | 1.01x |
 
 **Small but free!**
 
@@ -1101,16 +1101,16 @@ if (UNLIKELY(buffer_full)) {  // CPU knows this is rare
 
 | Optimization | Twitter Contribution | CITM Contribution |
 |--------------|---------------------|-------------------|
-| **Consteval** | +99% (1.99x) | +165% (2.65x) |
-| **SIMD Escaping** | +44% (1.44x) | +3% (1.03x) |
-| **Fast Digits** | +6% (1.06x) | +27% (1.27x) |
-| **Branch Hints** | +1.5% | +1.5% |
-| **Buffer Growth** | +0.8% | +0.8% |
-| **TOTAL** | **~2.95x faster** | **~3.5x faster** |
+| **Consteval** | +100% (2.00x) | +141% (2.41x) |
+| **SIMD Escaping** | +42% (1.42x) | +4% (1.04x) |
+| **Fast Digits** | +6% (1.06x) | +34% (1.34x) |
+| **Branch Hints** | +1% | +5% |
+| **Buffer Growth** | -0.4% | +2% |
+| **TOTAL** | **~2.9x faster** | **~3.4x faster** |
 
 **From Baseline to Optimized:**
-- Twitter: ~1,100 MB/s → 3,231 MB/s
-- CITM: ~670 MB/s → 2,341 MB/s
+- Twitter: ~1,100 MB/s → 3,211 MB/s
+- CITM: ~700 MB/s → 2,360 MB/s
 
 ---