[dev/perf][Update] update dev/perf branch to main branch commit 1ec04f7

.github/scripts/build_aiter_triton.sh

@@ -12,6 +12,7 @@ pip install --upgrade pandas zmq einops numpy==1.26.2
 pip uninstall -y aiter || true
 pip install --upgrade "pybind11>=3.0.1"
 pip install --upgrade "ninja>=1.11.1"
+pip install tabulate
 python3 setup.py develop
 
 # Read BUILD_TRITON env var, default to 1. If 1, install Triton; if 0, skip installation.
@@ -25,6 +26,9 @@ if [[ "$BUILD_TRITON" == "1" ]]; then
     cd triton
     pip install -r python/requirements.txt
     pip install filecheck
+    # NetworkX is a dependency of Triton test selection script
+    # `.github/scripts/select_triton_tests.py`.
+    pip install networkx
     MAX_JOBS=64 pip --retries=10 --default-timeout=60 install .
     cd ..
 else

.github/scripts/collect_logs.py

@@ -0,0 +1,69 @@
+#!/usr/bin/env python3
+import re
+import sys
+from pathlib import Path
+
+
+def extract_markdown_blocks(path: Path):
+    """
+    Extract markdown blocks from a log file.
+    The blocks are defined as:
+    [aiter] <operator> summary (markdown):
+    | ... |
+    | ... |
+    ...
+    """
+
+    start_pattern = re.compile(r"^\[aiter\]\s+.*summary\s*\(markdown\):")
+    table_line_pattern = re.compile(r"^\|")
+    blocks = []
+
+    with path.open("r", encoding="utf-8", errors="ignore") as f:
+        in_block = False
+        current_block = []
+
+        for line in f:
+            stripped = line.rstrip("\n")
+
+            if not in_block:
+                if start_pattern.match(stripped):
+                    in_block = True
+                    current_block = [stripped]
+                continue
+            else:
+                if table_line_pattern.match(stripped):
+                    current_block.append(stripped)
+                    continue
+                else:
+                    blocks.append(current_block)
+                    in_block = False
+                    current_block = []
+
+        if in_block and current_block:
+            blocks.append(current_block)
+
+    return blocks
+
+
+def main():
+    if len(sys.argv) < 2:
+        print("Usage: collect_logs.py <log_file>", file=sys.stderr)
+        sys.exit(1)
+
+    log_path = Path(sys.argv[1])
+
+    if not log_path.exists():
+        print(f"File not found: {log_path}", file=sys.stderr)
+        sys.exit(1)
+
+    blocks = extract_markdown_blocks(log_path)
+
+    for i, block in enumerate(blocks):
+        for line in block:
+            print(line)
+        if i != len(blocks) - 1:
+            print()
+
+
+if __name__ == "__main__":
+    main()

.github/scripts/op_tune.sh

@@ -22,6 +22,7 @@ declare -a tune_jobs=(
   "ck_gemm_a8w8_blockscale:csrc/ck_gemm_a8w8_blockscale:op_tests/test_gemm_a8w8_blockscale.py:python3 csrc/ck_gemm_a8w8_blockscale/gemm_a8w8_blockscale_tune.py -i aiter/configs/a8w8_blockscale_untuned_gemm.csv -o aiter/configs/a8w8_blockscale_tuned_gemm.csv"
   "ck_gemm_a8w8_blockscale_bpreshuffle:csrc/ck_gemm_a8w8_blockscale_bpreshuffle:op_tests/test_gemm_a8w8_blockscale.py:python3 csrc/ck_gemm_a8w8_blockscale_bpreshuffle/gemm_a8w8_blockscale_bpreshuffle_tune.py -i aiter/configs/a8w8_blockscale_bpreshuffle_untuned_gemm.csv -o aiter/configs/a8w8_blockscale_bpreshuffle_tuned_gemm.csv"
   "ck_gemm_a8w8_bpreshuffle:csrc/ck_gemm_a8w8_bpreshuffle:op_tests/test_gemm_a8w8.py:python3 csrc/ck_gemm_a8w8_bpreshuffle/gemm_a8w8_bpreshuffle_tune.py -i aiter/configs/a8w8_bpreshuffle_untuned_gemm.csv -o aiter/configs/a8w8_bpreshuffle_tuned_gemm.csv"
+  "ck_gemm_moe_2stages_codegen:csrc/ck_gemm_moe_2stages_codegen:op_tests/test_moe.py:python3 csrc/ck_gemm_moe_2stages_codegen/gemm_moe_tune.py -i aiter/configs/untuned_fmoe.csv -o aiter/configs/tuned_fmoe.csv"
   #"ck_gemm_a4w4_blockscale:csrc/ck_gemm_a4w4_blockscale:op_tests/test_gemm_a4w4_blockscale.py:python3 csrc/ck_gemm_a4w4_blockscale/gemm_a4w4_blockscale_tune.py -i aiter/configs/a4w4_blockscale_untuned_gemm.csv -o aiter/configs/a4w4_blockscale_tuned_gemm.csv"
 )