diff --git a/projects/rocprofiler-compute/CHANGELOG.md b/projects/rocprofiler-compute/CHANGELOG.md index c7d4873edb..c58b948f1d 100644 --- a/projects/rocprofiler-compute/CHANGELOG.md +++ b/projects/rocprofiler-compute/CHANGELOG.md @@ -34,6 +34,8 @@ Full documentation for ROCm Compute Profiler is available at [https://rocm.docs. * Fix the functioning of --dispatch option to act as 1-based index and ensure that correct kernel iterations are being profiled +* Corrected peak VALU Roofline profiling and analysis by removing `FP8` VALU and `BF16` VALU benchmarking. + ## ROCm Compute Profiler 3.4.0 for ROCm 7.2.0 ### Added @@ -96,7 +98,6 @@ Full documentation for ROCm Compute Profiler is available at [https://rocm.docs. * Improved Roofline Benchmarking by updating the `flops_benchmark` calculation. ### Resolved issues - * Bugfixes for stability ## ROCm Compute Profiler 3.3.0 for ROCm 7.1.0 @@ -254,7 +255,7 @@ Full documentation for ROCm Compute Profiler is available at [https://rocm.docs. * A workaround has been implemented using max(0, calculated_value) to prevent negative display values while the root cause is under investigation. * The profile mode crashes when `--format-rocprof-output json` is selected. - * As a workaround, this option should either not be provided or should be set to `csv` instead of `json`. This issue does not affect the profiling results since both `csv` and `json` output formats lead to the same profiling data. + * As a workaround, this option should either not be provided or should be set to `csv` instead of `json`. This issue does not affect the profiling results since both `csv` and `json` output formats lead to the same profiling data. ### Upcoming changes diff --git a/projects/rocprofiler-compute/src/utils/roofline_calc.py b/projects/rocprofiler-compute/src/utils/roofline_calc.py index 187fb8dc51..05eeb25ca2 100644 --- a/projects/rocprofiler-compute/src/utils/roofline_calc.py +++ b/projects/rocprofiler-compute/src/utils/roofline_calc.py @@ -101,7 +101,7 @@ SUPPORTED_DATATYPES: dict[str, list[str]] = { ], # Unsupported: } -PEAK_OPS_DATATYPES = ["FP8", "FP16", "BF16", "FP32", "FP64", "I8", "I32", "I64"] +PEAK_OPS_DATATYPES = ["FP16", "FP32", "FP64", "I8", "I32", "I64"] MFMA_DATATYPES = ["FP4", "FP6", "FP8", "FP16", "BF16", "FP32", "FP64", "I8"] CACHE_HIERARCHY = ["HBM", "L2", "L1", "LDS"] diff --git a/projects/rocprofiler-compute/src/utils/rooflines/roofline-azurelinux3-rocm7 b/projects/rocprofiler-compute/src/utils/rooflines/roofline-azurelinux3-rocm7 index c31ea03aee..4e38a84874 100755 Binary files a/projects/rocprofiler-compute/src/utils/rooflines/roofline-azurelinux3-rocm7 and b/projects/rocprofiler-compute/src/utils/rooflines/roofline-azurelinux3-rocm7 differ diff --git a/projects/rocprofiler-compute/src/utils/rooflines/roofline-rhel8-rocm7 b/projects/rocprofiler-compute/src/utils/rooflines/roofline-rhel8-rocm7 index 751568d192..57cc8ae1ce 100755 Binary files a/projects/rocprofiler-compute/src/utils/rooflines/roofline-rhel8-rocm7 and b/projects/rocprofiler-compute/src/utils/rooflines/roofline-rhel8-rocm7 differ diff --git a/projects/rocprofiler-compute/src/utils/rooflines/roofline-sles15sp6-rocm7 b/projects/rocprofiler-compute/src/utils/rooflines/roofline-sles15sp6-rocm7 index 2c8581da8a..895207bdda 100755 Binary files a/projects/rocprofiler-compute/src/utils/rooflines/roofline-sles15sp6-rocm7 and b/projects/rocprofiler-compute/src/utils/rooflines/roofline-sles15sp6-rocm7 differ diff --git a/projects/rocprofiler-compute/src/utils/rooflines/roofline-ubuntu22_04-rocm7 b/projects/rocprofiler-compute/src/utils/rooflines/roofline-ubuntu22_04-rocm7 index bdf1497e0e..768940c737 100755 Binary files a/projects/rocprofiler-compute/src/utils/rooflines/roofline-ubuntu22_04-rocm7 and b/projects/rocprofiler-compute/src/utils/rooflines/roofline-ubuntu22_04-rocm7 differ