SWDEV-363074 - Adjust staging copy limits in Windows

Pinned copy can cause big performance drops, because slow pinning under Windows.
Use up to 128MB for staging transfers. Change staging buffer size to 4MB.
Linux path should still have the old defaults.

Change-Id: I954edceb3ec89e8e670be116aa2d0a9564c8b11c


[ROCm/clr commit: 79d12df147]

projects/clr/rocclr/device/pal/palsettings.cpp

@@ -54,7 +54,7 @@ Settings::Settings() {
 
   stagedXferRead_ = true;
   stagedXferWrite_ = true;
-  stagedXferSize_ = GPU_STAGING_BUFFER_SIZE * Ki;
+  stagedXferSize_ = GPU_STAGING_BUFFER_SIZE * Mi;
 
   // We will enable staged read/write if we use local memory
   disablePersistent_ = false;
@@ -78,14 +78,9 @@ Settings::Settings() {
 
   // By default use host blit
   blitEngine_ = BlitEngineHost;
-  constexpr size_t MaxPinnedXferSize = 64;
-  pinnedXferSize_ = std::min(GPU_PINNED_XFER_SIZE, MaxPinnedXferSize) * Mi;
-
-  constexpr size_t PinnedMinXferSize = 4 * Mi;
+  pinnedXferSize_ = GPU_PINNED_MIN_XFER_SIZE * Mi;
   pinnedMinXferSize_ = flagIsDefault(GPU_PINNED_MIN_XFER_SIZE)
-      ? PinnedMinXferSize
-      : GPU_PINNED_MIN_XFER_SIZE * Ki;
-  pinnedMinXferSize_ = std::min(pinnedMinXferSize_, pinnedXferSize_);
+    ? 128 * Mi : GPU_PINNED_MIN_XFER_SIZE * Mi;
 
   // Disable FP_FAST_FMA defines by default
   reportFMAF_ = false;

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

@@ -62,14 +62,15 @@ Settings::Settings() {
 
   stagedXferRead_ = true;
   stagedXferWrite_ = true;
-  stagedXferSize_ = GPU_STAGING_BUFFER_SIZE * Ki;
+  stagedXferSize_ = flagIsDefault(GPU_STAGING_BUFFER_SIZE)
+      ? 1 * Mi : GPU_STAGING_BUFFER_SIZE * Mi;
 
   // Initialize transfer buffer size to 1MB by default
   xferBufSize_ = 1024 * Ki;
 
-  const static size_t MaxPinnedXferSize = 128;
-  pinnedXferSize_ = std::min(GPU_PINNED_XFER_SIZE, MaxPinnedXferSize) * Mi;
-  pinnedMinXferSize_ = std::min(GPU_PINNED_MIN_XFER_SIZE * Ki, pinnedXferSize_);
+  pinnedXferSize_ = GPU_PINNED_MIN_XFER_SIZE * Mi;
+  pinnedMinXferSize_ = flagIsDefault(GPU_PINNED_MIN_XFER_SIZE)
+    ? 1 * Mi : GPU_PINNED_MIN_XFER_SIZE * Mi;
 
   sdmaCopyThreshold_ = GPU_FORCE_BLIT_COPY_SIZE * Ki;
 

projects/clr/rocclr/utils/flags.hpp

@@ -66,8 +66,8 @@ release(uint, GPU_CP_DMA_COPY_SIZE, 1,                                        \
         "Set maximum size of CP DMA copy in KiB")                             \
 release(uint, GPU_MAX_HEAP_SIZE, 100,                                         \
         "Set maximum size of the GPU heap to % of board memory")              \
-release(uint, GPU_STAGING_BUFFER_SIZE, 1024,                                  \
-        "Size of the GPU staging buffer in KiB")                              \
+release(uint, GPU_STAGING_BUFFER_SIZE, 4,                                     \
+        "Size of the GPU staging buffer in MiB")                              \
 release(bool, GPU_DUMP_BLIT_KERNELS, false,                                   \
         "Dump the kernels for blit manager")                                  \
 release(uint, GPU_BLIT_ENGINE_TYPE, 0x0,                                      \
@@ -93,9 +93,9 @@ debug(cstring, AMD_OCL_SUBST_OBJFILE, 0,                                      \
 debug(bool, AMD_OCL_ENABLE_MESSAGE_BOX, false,                                \
         "Enable the error dialog on Windows")                                 \
 release(size_t, GPU_PINNED_XFER_SIZE, 32,                                     \
-        "The pinned buffer size for pinning in read/write transfers")         \
-release(size_t, GPU_PINNED_MIN_XFER_SIZE, 1024,                               \
-        "The minimal buffer size for pinned read/write transfers in KBytes")  \
+        "The pinned buffer size for pinning in read/write transfers in MiB")  \
+release(size_t, GPU_PINNED_MIN_XFER_SIZE, 128,                                \
+        "The minimal buffer size for pinned read/write transfers in MiB")     \
 release(size_t, GPU_RESOURCE_CACHE_SIZE, 64,                                  \
         "The resource cache size in MB")                                      \
 release(size_t, GPU_MAX_SUBALLOC_SIZE, 4096,                                  \