SWDEV-440746 - Limit WG for compute P2P

Use only 16 workgroups for compute P2P copies.
That should be enough to utilize XGMI bandwidth.

Change-Id: I60dfe019279bb95f93c8874244c1738aad1896d8


[ROCm/clr commit: 31101c6219]

projects/clr/rocclr/device/rocm/rocblit.cpp

@@ -2270,8 +2270,18 @@ bool KernelBlitManager::copyBuffer(device::Memory& srcMemory, device::Memory& ds
                                    amd::CopyMetadata copyMetadata) const {
   amd::ScopedLock k(lockXferOps_);
   bool result = false;
-  bool p2p = (&gpuMem(srcMemory).dev() != &gpuMem(dstMemory).dev()) &&
-             ((sizeIn[0] > ROC_P2P_SDMA_SIZE * Ki) || !gpu().IsPendingDispatch());
+  bool p2p = false;
+  uint32_t blit_wg_ = dev().settings().limit_blit_wg_;
+
+  if (&gpuMem(srcMemory).dev() != &gpuMem(dstMemory).dev()) {
+    if ((sizeIn[0] > dev().settings().sdma_p2p_threshold_) || !gpu().IsPendingDispatch()) {
+      p2p = true;
+    } else {
+      constexpr uint32_t kLimitWgForKernelP2p = 16;
+      blit_wg_ = kLimitWgForKernelP2p;
+    }
+  }
+
   bool asan = false;
   bool ipcShared = srcMemory.owner()->ipcShared() || dstMemory.owner()->ipcShared();
 #if defined(__clang__)
@@ -2317,7 +2327,7 @@ bool KernelBlitManager::copyBuffer(device::Memory& srcMemory, device::Memory& ds
 
     // Program the dispatch dimensions
     const size_t localWorkSize = (aligned) ? 512 : 1024;
-    size_t globalWorkSize = std::min(dev().settings().limit_blit_wg_ * localWorkSize, size[0]);
+    size_t globalWorkSize = std::min(blit_wg_ * localWorkSize, size[0]);
     globalWorkSize = amd::alignUp(globalWorkSize, localWorkSize);
 
     // Program kernels arguments for the blit operation

projects/clr/rocclr/device/rocm/rocsettings.cpp

@@ -81,6 +81,7 @@ Settings::Settings() {
   lcWavefrontSize64_ = true;
   imageBufferWar_ = false;
 
+  sdma_p2p_threshold_ = ROC_P2P_SDMA_SIZE * Ki;
   hmmFlags_ = (!flagIsDefault(ROC_HMM_FLAGS)) ? ROC_HMM_FLAGS : 0;
 
   rocr_backend_ = true;

projects/clr/rocclr/device/rocm/rocsettings.hpp

@@ -74,6 +74,7 @@ class Settings : public device::Settings {
   size_t pinnedMinXferSize_;  //!< Minimal buffer size for pinned transfer
 
   size_t sdmaCopyThreshold_;  //!< Use SDMA to copy above this size
+  size_t sdma_p2p_threshold_; //!< Use SDMA in P2P above this size
 
   uint32_t  hmmFlags_;        //!< HMM functionality control flags
   uint32_t  limit_blit_wg_;   //!< The number of workgroups for blit execution