From 9820480cbd0d362473bf370051721284b1f0fdb3 Mon Sep 17 00:00:00 2001 From: Anusha GodavarthySurya Date: Fri, 8 Nov 2024 11:39:16 +0000 Subject: [PATCH] SWDEV-491643 - AQL packets are captured for kernels disable sdma profiling hsa_amd_profiling_async_copy_enable is taking 45us for the first call. Disable sdma profiling for enqueuing captured kernel packets and for accumulate command. Change-Id: I80b51a58c46bccc9c1025e9331515f57c97b5a2a --- rocclr/device/rocm/rocvirtual.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/rocclr/device/rocm/rocvirtual.cpp b/rocclr/device/rocm/rocvirtual.cpp index cfec1ecbd2..7f13ae5797 100644 --- a/rocclr/device/rocm/rocvirtual.cpp +++ b/rocclr/device/rocm/rocvirtual.cpp @@ -995,7 +995,7 @@ inline bool VirtualGPU::dispatchAqlPacket( vcmd->addKernelName(kernelName); amd::ScopedLock lock(execution()); - profilingBegin(*vcmd, true); + profilingBegin(*vcmd); dispatchBlockingWait(); auto packet = reinterpret_cast(aqlpacket); @@ -3659,7 +3659,7 @@ void VirtualGPU::submitMarker(amd::Marker& vcmd) { void VirtualGPU::submitAccumulate(amd::AccumulateCommand& vcmd) { // Make sure VirtualGPU has an exclusive access to the resources amd::ScopedLock lock(execution()); - profilingBegin(vcmd, true); + profilingBegin(vcmd); const Settings& settings = dev().settings(); if (settings.barrier_value_packet_) {