From ad1839e9eeaf15274d766d49ed89b0fcf791b2fc Mon Sep 17 00:00:00 2001
From: foreman <dl.constructicon@amd.com>
Date: Wed, 29 Aug 2018 18:57:40 -0400
Subject: [PATCH] P4 to Git Change 1599701 by gandryey@gera-w8 on 2018/08/29
 18:48:36

	SWDEV-79445 - OCL generic changes and code clean-up
	- Use SDMA staging transfers for data upload if pinning fails. Fixes HIP failure in a test that uses the code segment data for uppload.

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palblit.cpp#26 edit


[ROCm/clr commit: 45d50bb7384a18f7b9a04e5be01b097115a22a83]
---
 .../clr/rocclr/runtime/device/pal/palblit.cpp    | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)

diff --git a/projects/clr/rocclr/runtime/device/pal/palblit.cpp b/projects/clr/rocclr/runtime/device/pal/palblit.cpp
index cf691c8598..43b0ecd860 100644
--- a/projects/clr/rocclr/runtime/device/pal/palblit.cpp
+++ b/projects/clr/rocclr/runtime/device/pal/palblit.cpp
@@ -270,17 +270,18 @@ bool DmaBlitManager::writeMemoryStaged(const void* srcHost, Memory& dstMemory, M
     chunkSize = std::max(chunkSize, 64 * Ki);
     bool flushDMA = true;
   }
-
+  size_t srcOffset = 0;
+  uint32_t flags = Resource::NoWait;
   while (xferSize != 0) {
     // Find the partial transfer size
     size_t tmpSize = std::min(chunkSize, xferSize);
-    amd::Coord3D src(offset, 0, 0);
+    amd::Coord3D src(srcOffset, 0, 0);
     amd::Coord3D dst(origin + offset, 0, 0);
     amd::Coord3D copySize(tmpSize, 0, 0);
 
     // Copy data into the temporary buffer, using CPU
     if (!xferBuf.hostWrite(&gpu(), reinterpret_cast<const char*>(srcHost) + offset,
-        src, copySize, Resource::NoWait)) {
+        src, copySize, flags)) {
       return false;
     }
 
@@ -292,6 +293,13 @@ bool DmaBlitManager::writeMemoryStaged(const void* srcHost, Memory& dstMemory, M
     totalSize -= tmpSize;
     offset += tmpSize;
     xferSize -= tmpSize;
+    srcOffset += tmpSize;
+    if ((srcOffset + tmpSize) > gpu().xferWrite().MaxSize()) {
+      srcOffset = 0;
+      flags =  0;
+    } else {
+      flags = Resource::NoWait;
+    }
   }
   return true;
 }
@@ -1950,7 +1958,7 @@ bool KernelBlitManager::writeBuffer(const void* srcHost, device::Memory& dstMemo
 
       if (amdMemory == NULL) {
         // Force SW copy
-        result = HostBlitManager::writeBuffer(srcHost, dstMemory, origin, size, entire);
+        result = DmaBlitManager::writeBuffer(srcHost, dstMemory, origin, size, entire);
         synchronize();
         return result;
       }