SWDEV-79445 OCL generic changes and code clean-up

Remove a workaround to CS_PARTIAL_FLUSH added in CL#1495187,
since PAL is no longer uses CS_PARTIAL_FLUSH.

Change-Id: I03edc7595459e19aad33b2b0901f0ebe4754d310


[ROCm/clr commit: 1d25343af8]

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

@@ -17,7 +17,7 @@
  LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  THE SOFTWARE. */
-
+ 
 #include "os/os.hpp"
 #include "platform/perfctr.hpp"
 #include "device/pal/paldefs.hpp"
@@ -36,7 +36,7 @@ TimeStamp::~TimeStamp() {}
 
 void TimeStamp::begin() {
   if (!flags_.beginIssued_) {
-    gpu().iCmd()->CmdWriteTimestamp(Pal::HwPipePoint::HwPipeTop, *iMem_,
+    gpu().iCmd()->CmdWriteTimestamp(Pal::HwPipePoint::HwPipeBottom, *iMem_,
                                     memOffset_ + CommandStartTime * sizeof(uint64_t));
     flags_.beginIssued_ = true;
   }
@@ -44,7 +44,7 @@ void TimeStamp::begin() {
 
 void TimeStamp::end() {
   CondLog(!flags_.beginIssued_, "We didn't issue a begin operation!");
-  gpu().iCmd()->CmdWriteTimestamp(Pal::HwPipePoint::HwPipeTop, *iMem_,
+  gpu().iCmd()->CmdWriteTimestamp(Pal::HwPipePoint::HwPipeBottom, *iMem_,
                                   memOffset_ + CommandEndTime * sizeof(uint64_t));
   flags_.endIssued_ = true;
 }

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

@@ -588,9 +588,8 @@ void VirtualGPU::MemoryDependency::validate(VirtualGPU& gpu, const Memory* memor
 
   if (flushL1Cache) {
     // Flush cache
-    if (!gpu.profiling()) {
-      gpu.addBarrier(RgpSqqtBarrierReason::MemDependency);
-    }
+    gpu.addBarrier(RgpSqqtBarrierReason::MemDependency);
+
     // Clear memory dependency state
     const static bool All = true;
     clear(!All);
@@ -2514,13 +2513,7 @@ bool VirtualGPU::submitKernelInternal(const amd::NDRangeContainer& sizes, const
     // Run AQL dispatch in HW
     eventBegin(MainEngine);
     iCmd()->CmdDispatchAql(dispatchParam);
-    // Note: This a workaround for incorrect results reported with release_mem packet,
-    // when the packet can be processed later after this dispatch and including extra time
-    if (profiling() || state_.profileEnabled_) {
-      addBarrier(RgpSqqtBarrierReason::ProfilingControl);
-      // Clear memory dependency to avoid the second L1 invalidation
-      memoryDependency().clear();
-    }
+
     if (id != gpuEvent.id_) {
       LogError("Something is wrong. ID mismatch!\n");
     }