print shared memory limited blocks per SM

[liqiangxl]

No description provided.

[github-actions]

Auto-merge Status

✅ Internal CI is finished
✅ No failed checks
✅ PR is mergeable
ℹ️ PR mergeable_state: unstable

Description

• Added debug output for shared memory limited blocks per SM

• Prints total requested vs available shared memory bytes

• Calculates and displays blocks per SM ratio for memory planning

• Handles unlimited case when no shared memory requested

Changes walkthrough

	Relevant files
Enhancement	allocations.cpp Add shared memory blocks per SM debug logging                        csrc/runtime/allocations.cpp Added debug output section after shared memory computation Prints total requested and available shared memory bytes Calculates and logs shared memory limited blocks per SM Handles division by zero with unlimited message when no memory requested +14/-0

PR Reviewer Guide

Here are some key observations to aid the review process:

🧪 PR contains tests

⚡ Recommended focus areas for review

Missing function validation The code calls deviceAvailableSharedMemoryBytes() but there's no indication this function exists or is properly implemented. Need to verify this function is available and returns the correct value for the target device. int64_t available_shared_memory_bytes = deviceAvailableSharedMemoryBytes(); Potential division by zero While the code handles total == 0 case, it doesn't validate that total is positive before division. Consider adding a bounds check to ensure total > 0 before performing the division operation. debug() << "shared memory limited blocks per SM: " << available_shared_memory_bytes / total << std::endl;

[liqiangxl]

!build

[greptile-apps]

Greptile Overview

Greptile Summary

Added debug output for shared memory usage metrics when DebugDumpOption::DynamicSharedMemory is enabled.

• Prints total requested shared memory bytes
• Prints available shared memory bytes
• Calculates and prints shared memory limited blocks per SM (available/total)
• Handles zero division case with "unlimited" message when no shared memory is requested

Issue found:

• Type mismatch: deviceAvailableSharedMemoryBytes() returns size_t but is assigned to int64_t

Confidence Score: 4/5

• Safe to merge with one type mismatch fix needed
• The change adds useful debug output with proper division-by-zero handling. One minor type mismatch exists where size_t is assigned to int64_t, which should be corrected for type consistency with the function's return type
• csrc/runtime/allocations.cpp needs the type mismatch fixed on line 119

Important Files Changed

File Analysis

Filename	Score	Overview
csrc/runtime/allocations.cpp	4/5	Added debug output for shared memory usage metrics including blocks per SM calculation; minor type mismatch detected

Sequence Diagram

sequenceDiagram
    participant Caller
    participant computeSharedMemory
    participant deviceAvailableSharedMemoryBytes
    participant debug

    Caller->>computeSharedMemory: Request shared memory calculation
    computeSharedMemory->>computeSharedMemory: Calculate total from buffers
    
    alt DebugDumpOption::DynamicSharedMemory enabled
        computeSharedMemory->>deviceAvailableSharedMemoryBytes: Get available shared memory
        deviceAvailableSharedMemoryBytes-->>computeSharedMemory: Return size_t bytes
        computeSharedMemory->>debug: Print total requested bytes
        computeSharedMemory->>debug: Print available bytes
        
        alt total > 0
            computeSharedMemory->>computeSharedMemory: Calculate blocks per SM
            computeSharedMemory->>debug: Print blocks per SM (available/total)
        else total == 0
            computeSharedMemory->>debug: Print unlimited blocks per SM
        end
    end
    
    computeSharedMemory-->>Caller: Return total bytes

Loading

[greptile-apps]

_{1 file reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

deviceAvailableSharedMemoryBytes() returns size_t (unsigned), not int64_t (signed)

Suggested change

	int64_t available_shared_memory_bytes = deviceAvailableSharedMemoryBytes();
	size_t available_shared_memory_bytes = deviceAvailableSharedMemoryBytes();

[jacobhinkle]

LGTM

[jacobhinkle]

Suggested change

	int64_t available_shared_memory_bytes = deviceAvailableSharedMemoryBytes();
	const size_t available_shared_memory_bytes = deviceAvailableSharedMemoryBytes();