Add print() function to fd.fusion printing to print full pre-scheduled fusion.

[kevinstephano]

I found this useful when trying to understanding the printing of the ID-Model in the test_tutorial.py file.

[github-actions]

Review updated until commit e3388d0

Description

• Add print() method to Fusion for full pre-scheduled IR output

• Include test validating FusionDefinition IR string representation

• Expose debug-level fusion IR as returnable string in Python

• Verify printed IR matches expected pre-scheduled structure

---

Changes walkthrough 📝

	Relevant files
Enhancement	runtime.cpp Add Fusion.print() to expose pre-scheduled IR in Python    python/python_direct/runtime.cpp Added print method to Fusion class in Python bindings Uses DebugStreamGuard and f.print(ss) to capture IR Returns full pre-scheduled fusion IR as a Python string Complements existing print_math method +18/-0
Tests	test_python_direct.py Test FusionDefinition.print() string representation            tests/python/direct/test_python_direct.py Added test for fd.fusion.print() output Defined expected full pre-scheduled IR string Asserted printed IR matches expected format Verified inputs, outputs, kernel, and TransformPrinter sections +28/-0

---

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🧪 PR contains tests

⚡ Recommended focus areas for review Potential Performance Overhead The use of DebugStreamGuard in the print method may introduce unnecessary overhead, as it is typically meant for debug logging rather than regular string serialization. Consider whether this guard is needed for the intended use case of printing the fusion IR. std::stringstream ss; DebugStreamGuard dsg(ss); String Equality with Formatting Sensitivity The test relies on exact string matching including whitespace and newlines, which may lead to fragile tests if formatting changes in the future. Consider using a more robust comparison method that ignores insignificant formatting differences. full_prescheduled_fusion_definition = """Inputs: T0_g_float[iS0{2}, iS1{4}, iS2{8}] T1_g_float[iS3{2}, iS4{4}, iS5{8}] Outputs: T2_g_float[iS6{2}, iS7{4}, iS8{8}] %kernel { T2_g_float[iS6{2}, iS7{4}, iS8{8}] = T0_g_float[iS0{2}, iS1{4}, iS2{8}] + T1_g_float[iS3{2}, iS4{4}, iS5{8}]; TransformPrinter : T0_g_float[iS0{2}, iS1{4}, iS2{8}] logical domain : (iS0{2}, iS1{4}, iS2{8}) contiguity: f f f loop domain : (iS0{2}, iS1{4}, iS2{8}) T1_g_float[iS3{2}, iS4{4}, iS5{8}] logical domain : (iS3{2}, iS4{4}, iS5{8}) contiguity: f f f loop domain : (iS3{2}, iS4{4}, iS5{8}) T2_g_float[iS6{2}, iS7{4}, iS8{8}] logical domain : (iS6{2}, iS7{4}, iS8{8}) contiguity: t t t loop domain : (iS6{2}, iS7{4}, iS8{8}) } // %kernel\n"""

[kevinstephano]

!test