tensor 0.1.0

checksums.yaml

@@ -0,0 +1,7 @@
+---
+SHA256:
+  metadata.gz: ab1431a30c46e725f4719be6d7af3f06f77720741f858448528f23528452baea
+  data.tar.gz: 2282380204a6d8b8a839365b265cb48443dd16e1e6206bff1d6d7c68af573421
+SHA512:
+  metadata.gz: 785a6cf18705e00c2bad28b89a1b2682b139f2423f3ba683fdfef5f2bf3204b4f1a5681c47d2734f8235585bc36e20d2c80790e0abfbf0ee5f625e2384728361
+  data.tar.gz: 7cdea95dba2a3169e10d8b1d425ada3bb06113dfca07d63287d7cc54f9014a025d31adb8a445375c5f9b00c96d5ae0532d39c39d588211f5308a36caa9743596

data/AGENTS.md

@@ -0,0 +1,38 @@
+# Repository Guidelines
+
+## Project Structure & Module Organization
+
+- Core C extension lives in `ruby_matrix.c` at the repo root.
+- Build configuration is handled by `extconf.rb`, which generates the `Makefile`.
+- Build artifacts include `ruby_matrix.bundle`, `ruby_matrix.o`, and other compiled outputs; treat these as generated files.
+
+## Build, Test, and Development Commands
+
+- `ruby extconf.rb` – generate a fresh `Makefile` for the Ruby C extension (preferred over editing `Makefile` directly).
+- `make` – compile the extension, enabling OpenMP and OpenBLAS when available.
+- `make clean` – remove compiled artifacts before rebuilding or changing toolchain settings.
+- After building, validate in a host Ruby project or with a small script that exercises matrix operations and ReLU on sample data.
+
+## Coding Style & Naming Conventions
+
+- C code uses 4-space indentation, braces on the same line, and descriptive names (`Matrix`, `matrix_multiply`, `matrix_relu_grad`, etc.).
+- Prefer clear, explicit error handling via `rb_raise` with actionable messages.
+- Keep configuration changes in `extconf.rb` (OpenMP, OpenBLAS, include/library paths); do not hand-edit the generated `Makefile`.
+- Avoid introducing one-letter variables in new code; keep naming consistent with existing functions and types.
+
+## Testing Guidelines
+
+- There is no standalone test suite in this repository; rely on the consuming Ruby project’s tests and targeted smoke scripts.
+- When adding behavior, include a minimal Ruby script (e.g., under `test/` or `examples/` if created) that constructs small matrices and verifies key operations.
+- Aim to keep new behavior covered by at least one fast, deterministic check that can run locally after `make`.
+
+## Commit & Pull Request Guidelines
+
+- Use concise, imperative commit messages (e.g., “Optimize float32 multiply path”, “Fix NEON detection on ARM”).
+- Keep changes focused: separate performance tuning, API changes, and build-system adjustments into distinct commits when possible.
+- In PR descriptions, summarize the change, note any performance impact or dependency changes (e.g., new `libomp`/OpenBLAS requirements), and describe how you validated the build.
+
+## Security & Configuration Tips
+
+- This extension depends on system libraries like OpenMP (`libomp`) and OpenBLAS; prefer configuration via `extconf.rb` and environment variables rather than hardcoded absolute paths.
+- Avoid embedding machine-specific paths in committed files; keep paths generic and documented so the extension can build on similar environments.

data/extconf.rb

@@ -0,0 +1,148 @@
+require 'mkmf'
+require 'rbconfig'
+
+# Logging level
+$VERBOSE = true
+
+host_os   = RbConfig::CONFIG['host_os']
+on_macos  = host_os =~ /darwin/
+
+# Debug environment variables
+puts "PKG_CONFIG_PATH: #{ENV['PKG_CONFIG_PATH']}"
+puts "PATH: #{ENV['PATH']}"
+puts "pkg-config location: #{`which pkg-config`.chomp}"
+
+# Set absolute path to pkg-config if needed (primarily for Homebrew on macOS)
+if on_macos && ENV['PKG_CONFIG'].to_s.empty?
+  ENV['PKG_CONFIG'] = '/opt/homebrew/bin/pkg-config'
+end
+puts "Using pkg-config at: #{ENV['PKG_CONFIG']}"
+
+# Check for required headers
+puts "Checking for ruby.h..."
+unless have_header('ruby.h')
+  abort "ruby.h not found. Please ensure Ruby development headers are installed."
+end
+
+# Detect architecture
+is_arm = RbConfig::CONFIG['host_cpu'] =~ /arm|aarch64/
+is_x86 = RbConfig::CONFIG['host_cpu'] =~ /x86|x64|i386/
+
+# Detect if we're using Apple's clang
+is_apple_clang = RbConfig::CONFIG['CC'] =~ /^clang/ && RUBY_PLATFORM =~ /darwin/
+
+# OpenMP support for macOS
+if is_apple_clang
+  puts "Detected Apple's clang, using LLVM OpenMP..."
+  # Use LLVM's OpenMP
+  libomp_prefix = "/opt/homebrew/opt/libomp"
+  if Dir.exist?(libomp_prefix)
+    $INCFLAGS << " -I#{libomp_prefix}/include"
+    $LDFLAGS << " -L#{libomp_prefix}/lib"
+    $CFLAGS << " -Xpreprocessor -fopenmp"
+    $LDFLAGS << " -lomp"
+    have_library('omp')
+    puts "LLVM OpenMP support enabled"
+  else
+    puts "Warning: OpenMP not found. Some optimizations will be disabled."
+    puts "To enable OpenMP: brew install libomp"
+  end
+else
+  # Standard OpenMP support for non-Apple compilers
+  if have_library('omp') || have_library('gomp')
+    $CFLAGS << " -fopenmp"
+    $LDFLAGS << " -fopenmp"
+    puts "OpenMP support enabled"
+  else
+    puts "Warning: OpenMP not found. Some optimizations will be disabled."
+  end
+end
+
+# Optional OpenBLAS support
+openblas_include_path = nil
+openblas_lib_path     = nil
+
+if on_macos
+  # Allow overriding OpenBLAS location, otherwise try Homebrew cellars
+  explicit_openblas = ENV['OPENBLAS_DIR']
+  if explicit_openblas && Dir.exist?(explicit_openblas)
+    openblas_root = explicit_openblas
+  else
+    openblas_versions = Dir.glob("/opt/homebrew/Cellar/openblas/*").sort
+    openblas_root = openblas_versions.last
+  end
+
+  if openblas_root && Dir.exist?(openblas_root)
+    openblas_include_path = File.join(openblas_root, "include")
+    openblas_lib_path     = File.join(openblas_root, "lib")
+    $INCFLAGS << " -I#{openblas_include_path}"
+    $LDFLAGS << " -L#{openblas_lib_path}"
+    $LIBPATH << openblas_lib_path
+  else
+    puts "OpenBLAS not found under /opt/homebrew/Cellar; BLAS optimizations may be disabled."
+  end
+end
+
+# Check for BLAS support using pkg-config (cross-platform)
+puts "Checking for BLAS support..."
+if pkg_config("openblas")
+  puts "BLAS support found via pkg-config! Enabling BLAS optimizations."
+elsif openblas_lib_path && Dir.exist?(openblas_lib_path)
+  puts "pkg-config failed to find OpenBLAS. Using manual configuration from #{openblas_lib_path}."
+  $LDFLAGS << " -lopenblas"
+else
+  puts "OpenBLAS not found. Continuing without BLAS-specific optimizations."
+end
+
+# Clean up incompatible OpenMP flags from pkg-config when using Apple's clang
+if is_apple_clang
+  $CFLAGS = $CFLAGS.to_s.gsub(/\s-fopenmp\b/, "")
+end
+
+# Check for math library
+have_library('m')
+
+# Check for SIMD headers based on architecture
+puts "Checking for SIMD headers..."
+if is_x86 && have_header('immintrin.h')
+  $CFLAGS << " -mavx2"
+  $defs << "-DHAVE_IMMINTRIN_H"
+  puts "AVX2 support found! Enabling AVX2 optimizations."
+elsif is_arm && have_header('arm_neon.h')
+  if RbConfig::CONFIG['host_os'] =~ /darwin/
+    $CFLAGS << " -DHAVE_ARM_NEON_H"
+    puts "ARM NEON support found on macOS! Enabling NEON optimizations."
+  else
+    $CFLAGS << " -DHAVE_ARM_NEON_H -mfpu=neon"
+    puts "ARM NEON support found! Enabling NEON optimizations."
+  end
+  $defs << "-DHAVE_ARM_NEON_H"
+else
+  puts "No SIMD headers found. SIMD optimizations will not be available."
+end
+
+# Add optimization flags
+$CFLAGS << " -O3 -march=native"
+$CFLAGS << " -fno-common -pipe"
+$CFLAGS << " -D_XOPEN_SOURCE -D_DARWIN_C_SOURCE -D_DARWIN_UNLIMITED_SELECT -D_REENTRANT"
+
+# Additional include and library paths
+$INCFLAGS << " -I/opt/homebrew/include"
+$INCFLAGS << " -I/usr/local/include"
+$LDFLAGS << " -L/opt/homebrew/lib"
+$LDFLAGS << " -L/usr/local/lib"
+
+# Ensure proper linking for neural network operations
+$defs << "-DUSE_NEURAL_OPS"
+$defs << "-DUSE_PARALLEL_OPS"
+
+# Print final configuration
+puts "\nFinal Configuration:"
+puts "Compiler: #{RbConfig::CONFIG['CC']}"
+puts "CFLAGS: #{$CFLAGS}"
+puts "LDFLAGS: #{$LDFLAGS}"
+puts "INCFLAGS: #{$INCFLAGS}"
+puts "Definitions: #{$defs.join(' ')}"
+
+# Create the makefile for the native extension (tensor_ext)
+create_makefile('tensor_ext')

data/lib/tensor.rb

@@ -0,0 +1,6 @@
+begin
+  require "tensor_ext"
+rescue LoadError
+  # Fallback for local development where the bundle lives next to the repo root
+  require File.expand_path("../tensor_ext", __dir__)
+end