P4 to Git Change 1288136 by jsjodin@jsjodin-git2p4-llvm on 2016/07/06 19:09:01

SWDEV-3 - [X86][AVX512] Add support for lowering shuffles to MOVDDUP/MOVSLDUP/MOVSHDUP

	git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@274436 91177308-0d34-0410-b5e6-96231b3b80d8

	GitHash: 2e1720f4556e47b98353c3a901913d0f1ba54944

Affected files ...

... //depot/stg/opencl/drivers/opencl/compiler/llvm.git/lib/Target/X86/X86ISelLowering.cpp#10 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm.git/test/CodeGen/X86/vector-shuffle-512-v16.ll#3 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm.git/test/CodeGen/X86/vector-shuffle-512-v8.ll#2 edit

rocclr/runtime/device/pal/palresource.cpp

@@ -1072,9 +1072,10 @@ Resource::free()
 void
 Resource::writeRawData(
     VirtualGPU& gpu,
-    size_t size,
+    size_t      offset,
+    size_t      size,
     const void* data,
-    bool waitForEvent) const
+    bool        waitForEvent) const
 {
     GpuEvent    event;
 
@@ -1082,11 +1083,8 @@ Resource::writeRawData(
     // size needs to be DWORD aligned
     assert((size & 3) == 0);
     gpu.eventBegin(MainEngine);
-    //! @todo Remove cache flush
-    //! It's a workaround for a PAL crash with embedded data, allocated before any command
-    gpu.flushCUCaches();
     gpu.queue(MainEngine).addCmdMemRef(iMem());
-    gpu.iCmd()->CmdUpdateMemory(*iMem(), 0, size, reinterpret_cast<const uint32_t*>(data));
+    gpu.iCmd()->CmdUpdateMemory(*iMem(), offset, size, reinterpret_cast<const uint32_t*>(data));
     gpu.eventEnd(MainEngine, event);
 
     setBusy(gpu, event);
@@ -1938,7 +1936,7 @@ Resource::warmUpRenames(VirtualGPU& gpu)
         uint    dummy = 0;
         const bool NoWait = false;
         // Write 0 for the buffer paging by VidMM
-        writeRawData(gpu, sizeof(dummy), &dummy, NoWait);
+        writeRawData(gpu, 0, sizeof(dummy), &dummy, NoWait);
         const bool Force = true;
         rename(gpu, Force);
     }