Review comments

* Clear CHANGELOG indicating we resolved roofline peaks for MI350

* Fix typo in pre-processor guard preventing roofline from running on
  MI300

* Ruff formatting

Author: vedithal-amd

projects/rocprofiler-compute/CHANGELOG.md

@@ -16,6 +16,8 @@ Full documentation for ROCm Compute Profiler is available at [https://rocm.docs.
 
 ### Resolved issues
 
+* Fixed roofline benchmark MFMA FP16/BF16/INT8 peaks for MI 350
+
 ### Upcoming changes
 
 ## ROCm Compute Profiler 3.5.0 for ROCm 7.12.0
@@ -55,8 +57,6 @@ Full documentation for ROCm Compute Profiler is available at [https://rocm.docs.
     * Processes attempting to benchmark on the same GPU will wait with user-visible feedback and execute sequentially.
     * Lock applies specifically to the roofline.csv file generated during benchmarking, not other files generated in profile mode.
 
-* Added proper support for `gfx950` in roofline benchmark.
-
 * Missing metric descriptions for gfx950 and gfx942 architecture.
 
 * Added `--membw-analysis` under experimental features to allow memory bandwidth specific profiling and analysis with metric block 30.

projects/rocprofiler-compute/src/utils/benchmark.py

@@ -1,7 +1,7 @@
 ##############################################################################
 # MIT License
 #
-# Copyright (c) 2026 Advanced Micro Devices, Inc. All Rights Reserved.
+# Copyright (c) 2025 - 2026 Advanced Micro Devices, Inc. All Rights Reserved.
 #
 # Permission is hereby granted, free of charge, to any person obtaining a copy
 # of this software and associated documentation files (the "Software"), to deal
@@ -753,7 +753,8 @@ def flops_bench(device: int, type: str, unit: str, rate: int) -> PerfMetrics:
 extern "C" __global__ void mfma_f16(int iter, float *dummy)
 {
     vec16<float> result = {0};
-#if defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942___)
+#if defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) || \
+    defined(__gfx941__) || defined(__gfx942__)
     vec4<__fp16> a;
     a[1] = a[0] = threadIdx.x;
     for(int i = 0; i < iter; ++i)