From f3a24ccf8b3f3805475334de2fd92fd2f99ca98b Mon Sep 17 00:00:00 2001
From: foreman <dl.constructicon@amd.com>
Date: Mon, 13 Jun 2016 15:58:22 -0400
Subject: [PATCH] P4 to Git Change 1279201 by gandryey@gera-w8 on 2016/06/13
 15:49:49

	SWDEV-95914 - OpenCL on PAL - Liquid Flash
	- Add liquid flash extension

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.cpp#5 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#10 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.hpp#5 edit


[ROCm/clr commit: c2aa3ca6f8206480dc91183e1f1ef94efda153a1]
---
 .../rocclr/runtime/device/pal/palsettings.cpp |  8 ++--
 .../rocclr/runtime/device/pal/palvirtual.cpp  | 39 ++++++++++++++++---
 .../rocclr/runtime/device/pal/palvirtual.hpp  |  1 +
 3 files changed, 39 insertions(+), 9 deletions(-)

diff --git a/projects/clr/rocclr/runtime/device/pal/palsettings.cpp b/projects/clr/rocclr/runtime/device/pal/palsettings.cpp
index e29b86741d..eb4e4dc907 100644
--- a/projects/clr/rocclr/runtime/device/pal/palsettings.cpp
+++ b/projects/clr/rocclr/runtime/device/pal/palsettings.cpp
@@ -281,6 +281,10 @@ Settings::create(
     enableExtension(ClAmdVec3);
     enableExtension(ClAmdPrintf);
     enableExtension(ClKhrImage2dFromBuffer);
+    // Enable some platform extensions
+    enableExtension(ClAmdDeviceAttributeQuery);
+    enableExtension(ClKhrSpir);
+    enableExtension(ClAMDLiquidFlash);
 
     hwLDSSize_      = 32 * Ki;
 
@@ -324,10 +328,6 @@ Settings::create(
 
     svmAtomics_ = (calAttr.svmAtomics || calAttr.isSVMFineGrainSystem) ? true : false;
 */
-    // Enable some platform extensions
-    enableExtension(ClAmdDeviceAttributeQuery);
-
-    enableExtension(ClKhrSpir);
 
     // SVM is not currently supported for DX Interop
 #if defined(_WIN32)
diff --git a/projects/clr/rocclr/runtime/device/pal/palvirtual.cpp b/projects/clr/rocclr/runtime/device/pal/palvirtual.cpp
index 5954396bed..a0972dcd0c 100644
--- a/projects/clr/rocclr/runtime/device/pal/palvirtual.cpp
+++ b/projects/clr/rocclr/runtime/device/pal/palvirtual.cpp
@@ -833,11 +833,10 @@ VirtualGPU::allocHsaQueueMem()
 
     // Provide private and local heap addresses
     const static uint addressShift = LP64_SWITCH(0, 32);
-    LogWarning("Private/Shared aperture isn't set");
-/*    queue->private_segment_aperture_base_hi =
-        static_cast<uint32>(dev().gslCtx()->getPrivateApertureBase()>>addressShift);
-    queue->group_segment_aperture_base_hi =
-        static_cast<uint32>(dev().gslCtx()->getSharedApertureBase()>>addressShift);
+/*    queue->private_segment_aperture_base_hi = static_cast<uint32_t>(
+        dev().properties().gpuMemoryProperties.privateApertureBase >> addressShift);
+    queue->group_segment_aperture_base_hi = static_cast<uint32_t>(
+        dev().properties().gpuMemoryProperties.sharedApertureBase >> addressShift);
 */
     hsaQueueMem_->unmap(nullptr);
     return true;
@@ -3438,4 +3437,34 @@ VirtualGPU::validateSdmaOverlap(const Resource& src, const Resource& dst)
     return false;
 }
 
+void
+VirtualGPU::submitTransferBufferFromFile(amd::TransferBufferFileCommand& cmd)
+{
+    size_t copySize = cmd.size()[0];
+    size_t fileOffset = cmd.fileOffset();
+    size_t srcDstOffset = cmd.origin()[0];
+    Memory* mem = dev().getGpuMemory(&cmd.memory());
+    uint    idx = 0;
+
+    assert((cmd.type() == CL_COMMAND_WRITE_BUFFER_FROM_FILE_AMD) ||
+        (cmd.type() == CL_COMMAND_READ_BUFFER_FROM_FILE_AMD));
+    bool writeBuffer(cmd.type() == CL_COMMAND_WRITE_BUFFER_FROM_FILE_AMD);
+
+    while (copySize > 0) {
+        Memory* staging = dev().getGpuMemory(&cmd.staging(idx));
+        size_t srcDstSize = amd::TransferBufferFileCommand::StagingBufferSize;
+        srcDstSize = std::min(srcDstSize, copySize);
+        void* srcDstBuffer = staging->cpuMap(*this);
+        if (!cmd.file()->transferBlock(writeBuffer, srcDstBuffer, fileOffset, 0, srcDstSize)) {
+            return;
+        }
+        staging->cpuUnmap(*this);
+
+        bool result = blitMgr().copyBuffer(*staging, *mem,
+            fileOffset, srcDstOffset, srcDstSize, false);
+        flushDMA(getGpuEvent(staging->iMem())->engineId_);
+        srcDstOffset += srcDstSize;
+        copySize -= srcDstSize;
+    }
+}
 } // namespace pal
diff --git a/projects/clr/rocclr/runtime/device/pal/palvirtual.hpp b/projects/clr/rocclr/runtime/device/pal/palvirtual.hpp
index 98c23c79d4..c566a3f646 100644
--- a/projects/clr/rocclr/runtime/device/pal/palvirtual.hpp
+++ b/projects/clr/rocclr/runtime/device/pal/palvirtual.hpp
@@ -253,6 +253,7 @@ public:
     virtual void submitSvmFillMemory(amd::SvmFillMemoryCommand& cmd);
     virtual void submitSvmMapMemory(amd::SvmMapMemoryCommand& cmd);
     virtual void submitSvmUnmapMemory(amd::SvmUnmapMemoryCommand& cmd);
+    virtual void submitTransferBufferFromFile(amd::TransferBufferFileCommand& cmd);
 
     void releaseMemory(Pal::IGpuMemory* iMem, bool wait = true);