[example] ttnn fused kernel with block-based compute

[brnorris03]

Description

This pull request introduces a fused elementwise operation example for TTNN, showing how to efficiently perform chained elementwise computations (such as Output = exp(A + B) + C) without storing intermediate results in memory. The implementation maximizes DST register utilization, processes tiles in blocks for better memory access patterns, and reduces memory bandwidth by eliminating intermediate reads/writes. The changes include a detailed README, a compute kernel for the fused operation, and a ternary reader kernel for block-based input loading.

• Implemented kernels/compute/fused_elementwise.cpp: a block-based compute kernel that fuses three operations (A + B, unary op (exp or relu), then + C) in DST registers, processing up to 4 tiles per DST cycle and handling remainder tiles for arbitrary input sizes.

• Added kernels/dataflow/reader_ternary.cpp: a reader kernel that loads tiles from three input tensors in blocks, reserves buffer space efficiently, and synchronizes reads with a single barrier for improved performance.

• The kernels avoid intermediate buffer storage, use block-based synchronization, and maximize hardware register usage for reduced DRAM accesses and improved throughput.

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Performance improvement
• Code refactoring
• Build/CI changes

Related Issues

Testing

Test Configuration

• OS:
• Python version:
• Compiler:

Tests Added/Modified

• Unit tests added
• Integration tests added
• Existing tests updated
• All tests pass locally

MLIR Changes (if applicable)

• Verified with TTLANG_VERBOSE_PASSES=1
• Checked initial and final MLIR outputs
• No unexpected IR transformations

Checklist

• Code follows the project's style guidelines
• Pre-commit hooks pass (pre-commit run --all-files)
• Self-review completed
• Comments added for complex code
• Documentation updated (if needed)
• No new compiler warnings
• Breaking changes documented (if any)
• CHANGELOG.md updated (if needed)

Additional Notes