adding summary and faqs

Author: anupras-mohapatra-arm

content/learning-paths/servers-and-cloud-computing/performix-instruction-mix/_index.md

@@ -19,11 +19,57 @@ prerequisites:
     - Basic understanding of matrix multiplication
     - Basic understanding of writing SIMD code with Neon or SVE
 
+# START generated_summary_faq
+generated_summary_faq:
+  template_version: summary-faq-v3
+  generated_at: '2026-07-15T19:50:40Z'
+  generator: ai
+  ai_assisted: true
+  ai_review_required: true
+  model: gpt-5
+  prompt_template: summary-faq-v3
+  source_hash: 22b8b268d361bfa326a92845a7a0878e26719ef182c3e907fd6da5dcd5503388
+  summary_generated_at: '2026-07-15T19:50:40Z'
+  summary_source_hash: 22b8b268d361bfa326a92845a7a0878e26719ef182c3e907fd6da5dcd5503388
+  faq_generated_at: '2026-07-15T19:50:40Z'
+  faq_source_hash: 22b8b268d361bfa326a92845a7a0878e26719ef182c3e907fd6da5dcd5503388
+  summary: >-
+    You'll build a GPT-2 inference example on an Arm Linux target and use Arm Performix to analyze
+    execution behavior. After running the Code Hotspots recipe to locate the dominant `matmul`
+    kernel, you'll use the Instruction Mix recipe to compare scalar and vector execution patterns
+    on Arm Neoverse. You'll then compare the scalar `gpt2` baseline with `gpt2_neon`, `gpt2_sve`,
+    optional `gpt2_user`, and the KleidiAI-backed `gpt2_kai_sve` path, validating vectorization
+    and throughput changes with Performix data and tokens-per-second results.
+  faqs:
+  - question: Which Performix recipe should I run first to find the bottleneck?
+    answer: >-
+      Start with the Code Hotspots recipe. Set the launch command to the baseline binary and use
+      `-n 150` to bias the profile toward inference time, then review the top functions.
+  - question: How do I know that matrix multiplication is the hotspot before I change any code?
+    answer: >-
+      The Code Hotspots report ranks functions by sampled time, so `kernels::matmul_ref()` should appear
+      near the top. Use that evidence to decide where to focus subsequent Instruction Mix profiling.
+  - question: Where do I add my Neon or SVE intrinsics, and what part of the loop should I change?
+    answer: >-
+      Edit `src/kernels/matmul_user.cpp`. Focus on the accumulation loop, `acc += row[j] * x[j];`,
+      and consider lane utilization, loop unrolling, and how you handle the tail.
+  - question: What binaries should I expect after building, and where are the reference implementations?
+    answer: >-
+      The build produces the scalar baseline `gpt2`, reference variants `gpt2_neon` and `gpt2_sve`,
+      and `gpt2_user` when the user kernel is enabled. The reference kernels are implemented in
+      `matmul_neon.cpp` and `matmul_sve.cpp`.
+  - question: How do I verify that my vectorization worked using Performix?
+    answer: >-
+      Profile your rebuilt binary with the Instruction Mix recipe and compare the proportion of
+      vector instructions with the baseline. You can also run the reference `gpt2_neon` and `gpt2_sve`
+      and compare instruction mix and the program's tokens-per-second summary.
+# END generated_summary_faq
+
 author:
     - Kieran Hejmadi
     - Oliver Grainge
 
-generate_summary_faq: true
+generate_summary_faq: false
 rerun_summary: false
 rerun_faqs: false