tritonparse 0.3.1.dev20251029071541__tar.gz → 0.3.1.dev20251030071508__tar.gz

{tritonparse-0.3.1.dev20251029071541/tritonparse.egg-info → tritonparse-0.3.1.dev20251030071508}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tritonparse
-Version: 0.3.1.dev20251029071541
+Version: 0.3.1.dev20251030071508
 Summary: TritonParse: A Compiler Tracer, Visualizer, and mini-Reproducer Generator for Triton Kernels
 Author-email: Yueming Hao <yhao@meta.com>
 License-Expression: BSD-3-Clause

{tritonparse-0.3.1.dev20251029071541 → tritonparse-0.3.1.dev20251030071508}/tests/test_tritonparse.py

@@ -139,6 +139,63 @@ def clear_all_caches(*kernels):
 class TestTritonparseCPU(unittest.TestCase):
     """CPU-only tests (no CUDA required)"""
 
+    def test_callsite_parsing(self):
+        """Test parsing of callsite locations in TTIR/TTGIR"""
+        from tritonparse.ir_parser import extract_loc_definitions
+        from tritonparse.trace_processor import generate_source_mappings
+
+        # Test MLIR callsite location definitions
+        ir_with_callsite = """
+module {
+  #loc7 = loc("/tmp/test.py":1091:8)
+  #loc57 = loc("/tmp/test.py":421:16)
+  #loc58 = loc("/tmp/test.py":853:16)
+  #loc190 = loc(callsite(#loc58 at #loc7))
+  #loc220 = loc(callsite(#loc57 at #loc190))
+  %0 = tt.load %ptr loc(#loc220)
+}
+"""
+        # Extract loc definitions
+        locs = extract_loc_definitions(ir_with_callsite)
+
+        # Verify loc220 (nested callsite)
+        self.assertIn("220", locs)
+        self.assertEqual(locs["220"]["file"], "/tmp/test.py")
+        self.assertEqual(locs["220"]["line"], 421)  # Inherited from callee loc57
+        self.assertEqual(locs["220"]["column"], 16)
+        self.assertTrue(locs["220"].get("is_callsite"))
+        self.assertEqual(locs["220"]["callsite_callee"], "57")
+        self.assertEqual(locs["220"]["callsite_caller"], "190")
+
+        # Verify loc190 (simple callsite)
+        self.assertIn("190", locs)
+        self.assertEqual(locs["190"]["line"], 853)  # Inherited from callee loc58
+        self.assertTrue(locs["190"].get("is_callsite"))
+        self.assertEqual(locs["190"]["callsite_callee"], "58")
+        self.assertEqual(locs["190"]["callsite_caller"], "7")
+
+        # Test source mappings generation
+        mappings = generate_source_mappings(ir_with_callsite, "ttir")
+
+        # Find the line with tt.load
+        line_with_load = None
+        for line_num, content in enumerate(ir_with_callsite.split("\n"), start=1):
+            if "tt.load" in content:
+                line_with_load = str(line_num)
+                break
+
+        self.assertIsNotNone(line_with_load)
+        self.assertIn(line_with_load, mappings)
+
+        mapping = mappings[line_with_load]
+        self.assertEqual(mapping["file"], "/tmp/test.py")
+        self.assertEqual(mapping["line"], 421)  # From loc220 -> loc57
+        self.assertTrue(mapping.get("is_callsite"))
+        self.assertEqual(mapping["callsite_callee"], "57")
+        self.assertEqual(mapping["callsite_caller"], "190")
+
+        print("✓ Callsite parsing tests passed")
+
     def test_convert(self):
         """Test convert function with various data types"""
         # Test with primitive types

tritonparse-0.3.1.dev20251030071508/tritonparse/ir_analysis.py

@@ -0,0 +1,427 @@
+#  Copyright (c) Meta Platforms, Inc. and affiliates.
+
+import logging
+
+from .sourcemap_utils import load_ir_contents
+
+
+logger = logging.getLogger("IRAnalysis")
+
+
+def process_amd_bufferop(ir_content: str, io_keys: list[str]) -> dict[str, int]:
+    def make_key(prefix: str) -> str:
+        return f"{prefix}_count"
+
+    io_keys = [(make_key(prefix), prefix) for prefix in io_keys]
+    output: dict[str, int] = {}
+    for dict_key, _ in io_keys:
+        output[dict_key] = 0
+    if ir_content:
+        for line in ir_content.split("\n"):
+            for dict_key, code_key in io_keys:
+                if code_key in line:
+                    output[dict_key] += 1
+    return output
+
+
+def process_amd_ttgir_bufferops(
+    key: str,
+    file_content: dict[str, str],
+    file_path: dict[str, str],
+) -> dict[str, int]:
+    ir_content = load_ir_contents(key, file_content, file_path)
+    # TODO: Add atomics
+    io_keys = ["tt.load", "tt.store", "amdgpu.buffer_load", "amdgpu.buffer_store"]
+    return process_amd_bufferop(ir_content, io_keys)
+
+
+def process_amd_gcn_bufferops(
+    key: str,
+    file_content: dict[str, str],
+    file_path: dict[str, str],
+) -> dict[str, int]:
+    ir_content = load_ir_contents(key, file_content, file_path)
+    # TODO: Add atomics
+    io_keys = ["global_load", "global_store", "buffer_load", "buffer_store"]
+    return process_amd_bufferop(ir_content, io_keys)
+
+
+def find_loop_bounds(ir_content: str) -> list[tuple[int, int]]:
+    """
+    Find the bounds of all scf.for loops in the IR content.
+    These are the only candidates for Software Pipelining (SWP).
+
+    A loop starts with 'scf.for' and ends when its closing brace '}' is found.
+    Brace counts are tracked to determine when each loop closes.
+
+    Args:
+        ir_content: The IR content as a string.
+
+    Returns:
+        A list of tuples (start_line, end_line) for each scf.for loop found.
+        Line numbers are 0-indexed.
+    """
+    if not ir_content:
+        return []
+
+    loop_bounds: list[tuple[int, int]] = []
+    lines = ir_content.split("\n")
+
+    # Stack to track loop starts and their brace counts
+    # Each entry is (start_line, brace_count_at_start)
+    loop_stack: list[tuple[int, int]] = []
+    current_brace_count = 0
+
+    for line_idx, line in enumerate(lines):
+        # Check if this line starts a new scf.for loop
+        if "scf.for" in line:
+            loop_stack.append((line_idx, current_brace_count))
+
+        # Count braces on this line
+        for char in line:
+            if char == "{":
+                current_brace_count += 1
+            elif char == "}":
+                current_brace_count -= 1
+
+        # Check if we've closed any loops
+        while loop_stack and current_brace_count <= loop_stack[-1][1]:
+            start_line, _start_brace_count = loop_stack.pop()
+            # The loop ends at this line
+            loop_bounds.append((start_line, line_idx))
+
+    return loop_bounds
+
+
+def find_inner_loop_bounds(ir_content: str) -> list[tuple[int, int]]:
+    """
+    Find the bounds of inner scf.for loops (loops without nested loops inside).
+
+    Inner loops are the primary candidates for Software Pipelining (SWP) as they
+    represent the innermost computation that can be optimized.
+
+    Args:
+        ir_content: The IR content as a string.
+
+    Returns:
+        A list of tuples (start_line, end_line) for each inner scf.for loop found.
+        Line numbers are 0-indexed.
+    """
+    all_loops = find_loop_bounds(ir_content)
+
+    if not all_loops:
+        return []
+
+    # Filter to keep only inner loops (loops that don't contain other loops)
+    inner_loops: list[tuple[int, int]] = []
+
+    for i, (start_i, end_i) in enumerate(all_loops):
+        # Check if any other loop is nested inside this loop
+        has_nested_loop = False
+        for j, (start_j, end_j) in enumerate(all_loops):
+            if i != j:
+                # Check if loop j is nested inside loop i
+                if start_i < start_j and end_j < end_i:
+                    has_nested_loop = True
+                    break
+
+        # If no nested loops found, this is an inner loop
+        if not has_nested_loop:
+            inner_loops.append((start_i, end_i))
+
+    return inner_loops
+
+
+def find_loop_pipelining(
+    ttir_content: str,
+    ttgir_content: str,
+    ttir_loop_start: int,
+    ttir_loop_end: int,
+    loop_index: int,
+    ttir_to_ttgir_mapping: dict[str, dict],
+    ttgir_to_source_mapping: dict[str, dict],
+    python_source_content: str | None,
+    python_source_start_line: int,
+) -> dict[str, list[str]]:
+    """
+    Find pipelining information for a specific loop by identifying tt.load and tt.dot operations
+    in TTIR and mapping them to their corresponding operations in the original Python source code.
+
+    For each tt.load or tt.dot operation found in the TTIR loop, this function uses source
+    mappings to find the corresponding operations in TTGIR, then maps them back to the original
+    Python source code. Operations are categorized into three sections:
+    - prologue: Operations that appear before the loop body
+    - loop_body: Operations that appear within the loop body
+    - epilogue: Operations that appear after the loop body
+
+    Operations are merged together (both loads and dots) and sorted in program order
+    within each section.
+
+    Args:
+        ttir_content: The TTIR content as a string.
+        ttgir_content: The TTGIR content as a string.
+        ttir_loop_start: The starting line number of the loop in TTIR (0-indexed).
+        ttir_loop_end: The ending line number of the loop in TTIR (0-indexed).
+        ttir_to_ttgir_mapping: Source mapping from TTIR lines to TTGIR lines.
+        ttgir_to_source_mapping: Source mapping from TTGIR lines to original Python source.
+        python_source_content: The original Python source code content.
+
+    Returns:
+        A dictionary containing:
+        - "prologue": List of Python source line strings in program order
+        - "loop_body": List of Python source line strings in program order
+        - "epilogue": List of Python source line strings in program order
+    """
+    if not ttir_content or not ttgir_content:
+        return {
+            "prologue": [],
+            "loop_body": [],
+            "epilogue": [],
+        }
+
+    ttir_lines = ttir_content.split("\n")
+    ttgir_lines = ttgir_content.split("\n")
+    python_lines = python_source_content.split("\n") if python_source_content else []
+
+    def apply_trailing_space(op: str) -> str:
+        """
+        Add a trailing space to all ops to avoid false positives like
+        warp_group_dot and warp_group_dot_wait.
+        """
+        return op + " "
+
+    # Step 1: Find tt.load and tt.dot operations in TTIR loop
+    ttir_pipeline_lines: list[int] = []
+    pipeline_tt_ops = ["tt.load", "tt.dot"]
+    pipeline_tt_ops = [apply_trailing_space(op) for op in pipeline_tt_ops]
+    pipeline_ttgir_ops = [
+        "tt.load",
+        "tt.dot",
+        "async_copy_global_to_local",
+        "warp_group_dot",
+    ]
+    pipeline_ttgir_ops = [apply_trailing_space(op) for op in pipeline_ttgir_ops]
+    for line_idx in range(ttir_loop_start, min(ttir_loop_end + 1, len(ttir_lines))):
+        line = ttir_lines[line_idx]
+        for op in pipeline_tt_ops:
+            if op in line:
+                ttir_pipeline_lines.append(line_idx)
+                break
+
+    # Step 2: Find the corresponding loop in TTGIR using source mappings
+    # Map the TTIR loop bounds to TTGIR using source mappings
+    ttgir_inner_loops = find_inner_loop_bounds(ttgir_content)
+
+    if not ttgir_inner_loops:
+        # No loop found in TTGIR, return empty results
+        return {
+            "prologue": [],
+            "loop_body": [],
+            "epilogue": [],
+        }
+
+    # Use the first inner loop as the reference
+    # TODO: Implement more sophisticated mapping logic to match TTIR loops to TTGIR loops
+    ttgir_loop_start, ttgir_loop_end = ttgir_inner_loops[loop_index]
+
+    # Step 3: Map TTIR operations to TTGIR operations using source mappings
+    # and categorize them by their position relative to the TTGIR loop
+    # Store as (line_number, source_line) to maintain order before extracting just the source
+    prologue_ops: list[tuple[int, str]] = []
+    loop_body_ops: list[tuple[int, str]] = []
+    epilogue_ops: list[tuple[int, str]] = []
+
+    for ttir_line in ttir_pipeline_lines:
+        # Convert 0-indexed line to 1-indexed string key for mapping lookup
+        ttir_line_key = str(ttir_line + 1)
+
+        # Get the corresponding TTGIR lines from the source mapping
+        if ttir_line_key in ttir_to_ttgir_mapping:
+            ttgir_lines_list = ttir_to_ttgir_mapping[ttir_line_key].get(
+                "ttgir_lines", []
+            )
+
+            # For each mapped TTGIR line, categorize it
+            for ttgir_line in ttgir_lines_list:
+                # Convert back to 0-indexed
+                ttgir_line_idx = ttgir_line - 1
+
+                # Get the actual TTGIR line content to check if it's relevant
+                if ttgir_line_idx < len(ttgir_lines):
+                    ttgir_source_line = ttgir_lines[ttgir_line_idx].strip()
+
+                    # Only keep mappings to the "compute" op.
+                    if any(op in ttgir_source_line for op in pipeline_ttgir_ops):
+                        # Map TTGIR line back to Python source
+                        ttgir_line_key = str(ttgir_line)
+                        python_source_line = ttgir_source_line  # Default to TTGIR line
+
+                        if ttgir_line_key in ttgir_to_source_mapping:
+                            source_info = ttgir_to_source_mapping[ttgir_line_key]
+                            python_line_num = source_info.get("line")
+
+                            if python_line_num and python_lines:
+                                # Account for the offset: the Python source may not start at line 1
+                                # python_line_num is the absolute line number in the original file
+                                # python_source_start_line is where the extracted code starts
+                                # So we need to subtract the offset to get the index in our python_lines array
+                                python_line_idx = (
+                                    python_line_num - python_source_start_line
+                                )
+                                if 0 <= python_line_idx < len(python_lines):
+                                    python_source_line = python_lines[
+                                        python_line_idx
+                                    ].strip()
+
+                        if ttgir_line_idx < ttgir_loop_start:
+                            prologue_ops.append((ttgir_line_idx, python_source_line))
+                        elif ttgir_loop_start <= ttgir_line_idx <= ttgir_loop_end:
+                            loop_body_ops.append((ttgir_line_idx, python_source_line))
+                        else:
+                            epilogue_ops.append((ttgir_line_idx, python_source_line))
+
+    # Step 4: Sort each section by line number to maintain program order
+    prologue_ops.sort(key=lambda x: x[0])
+    loop_body_ops.sort(key=lambda x: x[0])
+    epilogue_ops.sort(key=lambda x: x[0])
+
+    # Extract just the source lines (without line numbers)
+    prologue_lines = [line for _, line in prologue_ops]
+    loop_body_lines = [line for _, line in loop_body_ops]
+    epilogue_lines = [line for _, line in epilogue_ops]
+
+    # Log the pipelining results
+    logger.info(
+        f"Loop pipelining results (TTIR lines {ttir_loop_start}-{ttir_loop_end}):"
+    )
+    logger.info(f"  Prologue ({len(prologue_lines)} ops):")
+    for line in prologue_lines:
+        logger.info(f"    {line}")
+    logger.info(f"  Loop Body ({len(loop_body_lines)} ops):")
+    for line in loop_body_lines:
+        logger.info(f"    {line}")
+    logger.info(f"  Epilogue ({len(epilogue_lines)} ops):")
+    for line in epilogue_lines:
+        logger.info(f"    {line}")
+
+    return {
+        "prologue": prologue_lines,
+        "loop_body": loop_body_lines,
+        "epilogue": epilogue_lines,
+    }
+
+
+def generate_loop_schedule(
+    ttir_key: str,
+    ttgir_key: str,
+    file_content: dict[str, str],
+    file_path: dict[str, str],
+    source_mappings: dict[str, dict],
+    python_source_content: str | None,
+    python_source_start_line: int,
+) -> list[dict]:
+    """
+    Generate loop schedule information by finding inner scf.for loops in TTIR
+    and analyzing their pipelining potential using source mappings.
+
+    Only inner loops (loops without nested loops) are considered as they are
+    the primary candidates for Software Pipelining (SWP).
+
+    Args:
+        ttir_key: Key for the TTIR file.
+        ttgir_key: Key for the TTGIR file.
+        file_content: Dictionary mapping file keys to content.
+        file_path: Dictionary mapping file keys to file paths.
+        source_mappings: Dictionary containing source mappings between IR stages.
+        python_source_content: The original Python source code content.
+        python_source_start_line: The starting line number of the Python source in the original file.
+
+    Returns:
+        A list of dictionaries, each containing:
+        - "loop_bounds": Tuple of (start_line, end_line) for the loop in TTIR
+        - "pipelining": Dictionary with Python source lines for operations
+    """
+    ttir_content = load_ir_contents(ttir_key, file_content, file_path)
+    ttgir_content = load_ir_contents(ttgir_key, file_content, file_path)
+
+    # Get the TTIR to TTGIR mapping and TTGIR to source mapping
+    ttir_to_ttgir_mapping = source_mappings.get("ttir", {})
+    ttgir_to_source_mapping = source_mappings.get("ttgir", {})
+
+    # Find only inner loops (loops without nested loops inside)
+    inner_loop_bounds = find_inner_loop_bounds(ttir_content)
+    # TODO: Fix loop mapping with multiple loops.
+    inner_loop_bounds = inner_loop_bounds[:1]
+
+    # For each inner loop, find pipelining information
+    loop_schedules = []
+    for i, (loop_start, loop_end) in enumerate(inner_loop_bounds):
+        pipelining_info = find_loop_pipelining(
+            ttir_content,
+            ttgir_content,
+            loop_start,
+            loop_end,
+            i,
+            ttir_to_ttgir_mapping,
+            ttgir_to_source_mapping,
+            python_source_content,
+            python_source_start_line,
+        )
+        loop_schedules.append(pipelining_info)
+
+    return loop_schedules
+
+
+def _generate_ir_analysis(entry: str):
+    payload = entry.setdefault("payload", {})
+    file_content = payload.get("file_content", {})
+    file_path = payload.get("file_path", {})
+    source_mappings = payload.get("source_mappings", {})
+
+    # Find the IR file keys
+    ttir_key = next((k for k in file_content if k.endswith(".ttir")), None)
+    ttgir_key = next((k for k in file_content if k.endswith(".ttgir")), None)
+    amdgcn_key = next((k for k in file_content if k.endswith(".amdgcn")), None)
+    # Skip if no IR files found
+    if not (ttir_key or ttgir_key or amdgcn_key):
+        logger.debug("No IR found")
+        return {}
+    ir_analysis = {}
+    if amdgcn_key and ttgir_key:
+        # Add BufferOps information
+        ttgir_bufferops_info = process_amd_ttgir_bufferops(
+            ttgir_key, file_content, file_path
+        )
+        gcn_bufferops_info = process_amd_gcn_bufferops(
+            amdgcn_key, file_content, file_path
+        )
+        io_counts = {}
+        # NDJSON format requires a newline at the end of each line
+        if ttgir_bufferops_info:
+            io_counts["amd_ttgir_bufferops_count"] = ttgir_bufferops_info
+        if gcn_bufferops_info:
+            io_counts["amd_gcn_bufferops_count"] = gcn_bufferops_info
+        if io_counts:
+            ir_analysis["io_counts"] = io_counts
+    if ttir_key and ttgir_key:
+        # Get Python source content and start line if available
+        python_source_content = None
+        python_source_start_line = 1  # Default to 1 if not available
+        python_source_info = payload.get("python_source")
+        if python_source_info:
+            python_source_content = python_source_info.get("code")
+            python_source_start_line = python_source_info.get("start_line", 1)
+
+        # Add loop schedule information
+        loop_schedule = generate_loop_schedule(
+            ttir_key,
+            ttgir_key,
+            file_content,
+            file_path,
+            source_mappings,
+            python_source_content,
+            python_source_start_line,
+        )
+        if loop_schedule:
+            ir_analysis["loop_schedules"] = loop_schedule
+    return ir_analysis

{tritonparse-0.3.1.dev20251029071541 → tritonparse-0.3.1.dev20251030071508}/tritonparse/ir_parser.py

@@ -44,6 +44,14 @@ ALIAS_WITH_NAME_PATTERN = re.compile(
 # Example: #loc20 = loc(#loc16)
 ALIAS_SIMPLE_PATTERN = re.compile(r"#loc(\d+)\s*=\s*loc\(\s*#loc(\d*)\s*\)")
 
+# Callsite loc definitions in TTIR/TTGIR
+# Example: #loc220 = loc(callsite(#loc57 at #loc190))
+# Captures: loc_id, callee_loc_id, caller_loc_id
+# Note: Uses (\d*) to match optional numbers (for bare #loc references)
+CALLSITE_PATTERN = re.compile(
+    r"#loc(\d+)\s*=\s*loc\(\s*callsite\(\s*#loc(\d*)\s+at\s+#loc(\d*)\s*\)\s*\)"
+)
+
 
 def extract_loc_definitions(ir_content: str) -> Dict[str, Dict[str, Any]]:
     """
@@ -141,6 +149,50 @@ def extract_loc_definitions(ir_content: str) -> Dict[str, Dict[str, Any]]:
     for alias_id, target_id in alias_map.items():
         if alias_id not in locations:
             resolve_alias(alias_id)
+
+    # Collect callsite definitions
+    callsite_defs = []
+    for i, line in enumerate(ir_content.split("\n"), start=1):
+        if m := CALLSITE_PATTERN.search(line):
+            loc_id, callee_id, caller_id = m.groups()
+            # Empty strings map to main loc key ""
+            callsite_defs.append((loc_id, callee_id or "", caller_id or "", i))
+
+    # Resolve callsite definitions
+    # A callsite inherits the location from its callee (the code being called)
+    # and stores a reference to its caller (the code doing the calling)
+    for loc_id, callee_id, caller_id, def_line in callsite_defs:
+        if loc_id not in locations:  # Avoid overwriting existing definitions
+            if callee_id in locations:
+                # Inherit location info from callee
+                callee_info = locations[callee_id]
+                locations[loc_id] = {
+                    "file": callee_info["file"],
+                    "line": callee_info["line"],
+                    "column": callee_info["column"],
+                    "def_line": def_line,
+                    "is_callsite": True,
+                    "callsite_callee": callee_id,
+                    "callsite_caller": caller_id,
+                }
+            else:
+                logger.warning(
+                    f"Callsite #loc{loc_id} references undefined callee #loc{callee_id}"
+                )
+                # Note: We don't add this callsite to locations since callee is missing
+
+    # Verify caller references (warning only, don't block)
+    for loc_id, _callee_id, caller_id, _def_line in callsite_defs:
+        if loc_id in locations and caller_id and caller_id not in locations:
+            logger.warning(
+                f"Callsite #loc{loc_id} references undefined caller #loc{caller_id}"
+            )
+
+    # Attach definition line and alias metadata
+    for k, v in def_line_map.items():
+        if k in locations:
+            locations[k]["def_line"] = v
+    for alias_id, target_id in alias_map.items():
         if alias_id in locations:
             locations[alias_id]["alias_of"] = target_id
             if alias_id in alias_name_map:

{tritonparse-0.3.1.dev20251029071541 → tritonparse-0.3.1.dev20251030071508}/tritonparse/trace_processor.py

@@ -77,6 +77,11 @@ def generate_source_mappings(
                 "column": info["column"],
                 f"{ir_type}_line": ln,
             }
+            # Propagate callsite metadata if present
+            if info.get("is_callsite"):
+                entry["is_callsite"] = True
+                entry["callsite_callee"] = info["callsite_callee"]
+                entry["callsite_caller"] = info["callsite_caller"]
             # Propagate alias metadata if present
             if "alias_name" in info:
                 entry["alias_name"] = info["alias_name"]

{tritonparse-0.3.1.dev20251029071541 → tritonparse-0.3.1.dev20251030071508/tritonparse.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tritonparse
-Version: 0.3.1.dev20251029071541
+Version: 0.3.1.dev20251030071508
 Summary: TritonParse: A Compiler Tracer, Visualizer, and mini-Reproducer Generator for Triton Kernels
 Author-email: Yueming Hao <yhao@meta.com>
 License-Expression: BSD-3-Clause

{tritonparse-0.3.1.dev20251029071541 → tritonparse-0.3.1.dev20251030071508}/website/src/pages/IRAnalysis.tsx

@@ -27,6 +27,7 @@ const IRAnalysis: React.FC<IRAnalysisProps> = ({ kernels, selectedKernel }) => {
   const io_counts = kernel.ir_analysis?.io_counts;
   const ttgir_info = io_counts?.["amd_ttgir_bufferops_count"];
   const amdgcn_info = io_counts?.["amd_gcn_bufferops_count"];
+  const loop_schedule = kernel.ir_analysis?.loop_schedules;
   const getCount = (info: Record<string, number> | undefined, key: string): string => { return info?.[key]?.toString() ?? "N/A"; };
 
   return (
@@ -44,7 +45,7 @@ const IRAnalysis: React.FC<IRAnalysisProps> = ({ kernels, selectedKernel }) => {
               AMD BufferOps Information
             </h3>
 
-            <div className="bg-gray-50 p-4 rounded-md border border-gray-200">
+            <div className="bg-gray-50 p-4 rounded-md border border-gray-200 mb-6">
               <div className="grid grid-cols-[repeat(auto-fit,_minmax(180px,_1fr))] gap-3">
                 {ttgir_info && (
                   <>
@@ -90,6 +91,76 @@ const IRAnalysis: React.FC<IRAnalysisProps> = ({ kernels, selectedKernel }) => {
             </div>
           </>
         )}
+
+        {loop_schedule && loop_schedule.length > 0 && (
+          <>
+            <h3 className="text-lg font-medium mb-3 text-gray-800">
+              Software Pipelining Schedule
+            </h3>
+
+            {loop_schedule.map((schedule: any, loopIndex: number) => {
+              const prologue = schedule?.prologue || [];
+              const loopBody = schedule?.loop_body || [];
+              const epilogue = schedule?.epilogue || [];
+
+              return (
+                <div key={loopIndex} className="bg-gray-50 p-4 rounded-md border border-gray-200 mb-4">
+                  <h4 className="text-md font-semibold mb-2 text-gray-700">
+                    Software Pipelining for Loop {loopIndex + 1}
+                  </h4>
+
+                  {/* Prologue */}
+                  {prologue.length > 0 && (
+                    <div className="mb-3">
+                      <div className="text-sm font-medium text-gray-600 mb-1">Prologue:</div>
+                      <div className="bg-white p-2 rounded border border-gray-200 font-mono text-xs">
+                        {prologue.map((line: string, idx: number) => (
+                          <div key={idx} className="text-gray-700">
+                            {line}
+                          </div>
+                        ))}
+                      </div>
+                    </div>
+                  )}
+
+                  {/* Loop Body */}
+                  <div className="mb-3">
+                    <div className="text-sm font-medium text-gray-600 mb-1">Loop Body:</div>
+                    <div className="bg-white p-2 rounded border border-gray-200">
+                      <div className="font-mono text-xs text-gray-500 mb-1">for (...) {'{'}</div>
+                      <div className="pl-4 font-mono text-xs">
+                        {loopBody.length > 0 ? (
+                          loopBody.map((line: string, idx: number) => (
+                            <div key={idx} className="text-gray-700">
+                              {line}
+                            </div>
+                          ))
+                        ) : (
+                          <div className="text-gray-400 italic">No operations in loop body</div>
+                        )}
+                      </div>
+                      <div className="font-mono text-xs text-gray-500 mt-1">{'}'}</div>
+                    </div>
+                  </div>
+
+                  {/* Epilogue */}
+                  {epilogue.length > 0 && (
+                    <div>
+                      <div className="text-sm font-medium text-gray-600 mb-1">Epilogue:</div>
+                      <div className="bg-white p-2 rounded border border-gray-200 font-mono text-xs">
+                        {epilogue.map((line: string, idx: number) => (
+                          <div key={idx} className="text-gray-700">
+                            {line}
+                          </div>
+                        ))}
+                      </div>
+                    </div>
+                  )}
+                </div>
+              );
+            })}
+          </>
+        )}
       </div>
     </div>
   );

{tritonparse-0.3.1.dev20251029071541 → tritonparse-0.3.1.dev20251030071508}/website/src/utils/dataLoader.ts

@@ -176,6 +176,7 @@ export interface CompilationMetadata {
 export interface IRAnalysisData {
     // Mapping from IR stage -> <IO type -> count>
     io_counts?: Record<string, Record<string, number>>;
+    loop_schedules?: [Record<string, [string]>];
 }
 
 /**

tritonparse-0.3.1.dev20251029071541/tritonparse/ir_analysis.py

@@ -1,77 +0,0 @@
-#  Copyright (c) Meta Platforms, Inc. and affiliates.
-
-import logging
-
-from .sourcemap_utils import load_ir_contents
-
-
-logger = logging.getLogger("IRAnalysis")
-
-
-def process_amd_bufferop(ir_content: str, io_keys: list[str]) -> dict[str, int]:
-    def make_key(prefix: str) -> str:
-        return f"{prefix}_count"
-
-    io_keys = [(make_key(prefix), prefix) for prefix in io_keys]
-    output: dict[str, int] = {}
-    for dict_key, _ in io_keys:
-        output[dict_key] = 0
-    if ir_content:
-        for line in ir_content.split("\n"):
-            for dict_key, code_key in io_keys:
-                if code_key in line:
-                    output[dict_key] += 1
-    return output
-
-
-def process_amd_ttgir_bufferops(
-    key: str,
-    file_content: dict[str, str],
-    file_path: dict[str, str],
-) -> dict[str, int]:
-    ir_content = load_ir_contents(key, file_content, file_path)
-    # TODO: Add atomics
-    io_keys = ["tt.load", "tt.store", "amdgpu.buffer_load", "amdgpu.buffer_store"]
-    return process_amd_bufferop(ir_content, io_keys)
-
-
-def process_amd_gcn_bufferops(
-    key: str,
-    file_content: dict[str, str],
-    file_path: dict[str, str],
-) -> dict[str, int]:
-    ir_content = load_ir_contents(key, file_content, file_path)
-    # TODO: Add atomics
-    io_keys = ["global_load", "global_store", "buffer_load", "buffer_store"]
-    return process_amd_bufferop(ir_content, io_keys)
-
-
-def _generate_ir_analysis(entry: str):
-    payload = entry.setdefault("payload", {})
-    file_content = payload.get("file_content", {})
-    file_path = payload.get("file_path", {})
-
-    # Find the IR file keys
-    ttgir_key = next((k for k in file_content if k.endswith(".ttgir")), None)
-    amdgcn_key = next((k for k in file_content if k.endswith(".amdgcn")), None)
-    # Skip if no IR files found
-    if not (ttgir_key or amdgcn_key):
-        logger.debug("No AMD IR found")
-        return {}
-    ir_analysis = {}
-    if amdgcn_key:
-        ttgir_bufferops_info = process_amd_ttgir_bufferops(
-            ttgir_key, file_content, file_path
-        )
-        gcn_bufferops_info = process_amd_gcn_bufferops(
-            amdgcn_key, file_content, file_path
-        )
-        io_counts = {}
-        # NDJSON format requires a newline at the end of each line
-        if ttgir_bufferops_info:
-            io_counts["amd_ttgir_bufferops_count"] = ttgir_bufferops_info
-        if gcn_bufferops_info:
-            io_counts["amd_gcn_bufferops_count"] = gcn_bufferops_info
-        if io_counts:
-            ir_analysis["io_counts"] = io_counts
-    return ir_analysis