Merge branch 'master' into master

Author: Rohan-T144

.github/actions/spelling/allow/names.txt

@@ -1,4 +1,5 @@
 Abdulrasool
+Abdelrhman
 Abhigyan
 Alexandru
 Alja
@@ -16,6 +17,7 @@ Corlay
 Couet
 Dmitry
 Duswald
+Elrawy
 Foco
 Graenitz
 Guilherme
@@ -90,6 +92,7 @@ Yuquan
 Zarytskyi
 aaronj
 aaronjomyjoseph
+abdelrhman
 abhi
 acherjan
 acherjee
@@ -119,6 +122,7 @@ daemondzh
 davidlange
 dlange
 efremale
+elrawy
 fransham
 fsfod
 gargvaibhav

.github/actions/spelling/allow/terms.txt

@@ -31,7 +31,10 @@ cytokines
 gitlab
 gridlay
 gsoc
+gpu
 llm
+llvm
+pushforward
 linkedin
 microenvironments
 pythonized

_data/contributors.yml

@@ -338,6 +338,29 @@
       proposal: /assets/docs/Rohan_Timmaraju_Proposal_2025.pdf
       mentors: Vassil Vassilev, David Lange, Jonas Rembser, Christina Koutsou
 
+- name: Abdelrhman Elrawy
+  photo: Abdelrhman.jpg
+  info: "Google Summer of Code 2025 Contributor"
+  email: [email protected]
+  education: "Master of Applied Computing, Wilfrid Laurier University, Canada"
+  github: "https://github.com/a-elrawy"
+  active: 1
+  linkedin: "https://www.linkedin.com/in/elrawy/"
+  projects:
+    - title: "Support usage of Thrust API in Clad"
+      status: Ongoing
+      description: |
+        This project enhances Clad, a Clang-based automatic differentiation tool, 
+        by enabling it to support NVIDIA's Thrust library for GPU-parallel programming. 
+        The goal is to implement custom derivative rules for Thrust primitives like 
+        `thrust::transform` and `thrust::reduce`, making it possible to differentiate 
+        high-performance CUDA code automatically. This work bridges the gap between 
+        automatic differentiation and GPU acceleration, enabling efficient gradient 
+        computations in scientific computing and machine learning workloads.
+      proposal: /assets/docs/Abdelrhman_Elrawy_Proposal_GSoC_2025.pdf
+      mentors: Vassil Vassilev, Alexander Penev
+
+
 - name: "This could be you!"
   photo: rock.jpg
   info: See <a href="/careers">openings</a> for more info

_pages/team/abdelrhman-elrawy.md

@@ -0,0 +1,10 @@
+---
+title: "Compiler Research - Team - Abdelrhman Elrawy"
+layout: gridlay
+excerpt: "Compiler Research: Team members"
+sitemap: false
+permalink: /team/AbdelrhmanElrawy
+email: [email protected]
+---
+
+{% include team-profile.html %}

_posts/2025-05-18-support-usage-of-thrust-API-in-clad.md

@@ -0,0 +1,52 @@
+---
+title: "Supporting Thrust API in Clad"
+layout: post
+excerpt: "This summer, I am working on adding support for Thrust API in Clad, enabling automatic differentiation of GPU-accelerated code. This work bridges the gap between high-performance CUDA parallelism and source-to-source AD transformation."
+sitemap: false
+author: Abdelrhman Elrawy
+permalink: blogs/gsoc25_/
+banner_image: /images/blog/gsoc-banner.png
+date: 2025-05-18
+tags: gsoc llvm clang automatic-differentiation gpu cuda thrust
+---
+
+## About Me
+
+Hi! I’m Abdelrhman Elrawy, a graduate student in Applied Computing specializing in Machine Learning and Parallel Programming. I’ll be working on enabling **Thrust API support in Clad**, bringing GPU-accelerated parallel computing to the world of automatic differentiation.
+
+## Project Description
+
+[Clad](https://github.com/vgvassilev/clad) is a Clang-based tool for source-to-source automatic differentiation (AD). It enables gradient computations by transforming C++ code at compile time.
+
+However, many scientific and machine learning applications leverage **NVIDIA’s Thrust**, a C++ parallel algorithms library for GPUs, and currently, Clad doesn’t support differentiating through Thrust constructs. This limits the usability of Clad in high-performance CUDA code.
+
+My project addresses this gap by enabling Clad to:
+
+- Recognize and handle Thrust primitives like `thrust::transform` and `thrust::reduce`
+- Implement **custom pullback/pushforward rules** for GPU kernels
+- Ensure gradients maintain **parallel performance and correctness**
+- Benchmark and validate derivatives in real-world ML and HPC use cases
+
+## Technical Approach
+
+The project begins with a **proof-of-concept**: manually writing derivatives for common Thrust operations like `transform` and `reduce`. These are compared against finite differences to validate correctness.
+
+Following that, I’ll integrate custom differentiation logic inside Clad, building:
+- A `ThrustBuiltins.h` header for recognizing Thrust calls
+- Visitor pattern extensions in Clad’s AST traversal (e.g., `VisitCallExpr`)
+- GPU-compatible derivative utilities (e.g., CUDA-aware `thrust::fill`, `transform`)
+
+I'll also implement **unit tests**, real-world **mini-apps** (e.g., neural networks), and **benchmarks** to validate and demonstrate this feature.
+
+## Expected Outcomes
+
+By the end of GSoC 2025, Clad will be able to:
+- Differentiate through key Thrust primitives with GPU execution preserved
+- Provide documentation and tutorials for GPU-based automatic differentiation
+- Contribute a robust test suite and benchmarks to the Clad ecosystem
+
+## Related Links
+
+- [Clad GitHub](https://github.com/vgvassilev/clad)
+- [Project description](https://hepsoftwarefoundation.org/gsoc/2025/proposal_Clad-ThrustAPI.html)
+- [My GitHub](https://github.com/a-elrawy)