gpu-compute: Fix valgrind memleak complaints

Fixes several memory leaks, mostly of small and medium severity. Fixes
mismatched new/new[] and delete/delete[] calls.

Change-Id: Iedafc409389bd94e45f330bc587d6d72d1971219

src/dev/amdgpu/amdgpu_device.cc

@@ -453,6 +453,8 @@ AMDGPUDevice::writeFrame(PacketPtr pkt, Addr offset)
 
     auto system = cp->shader()->gpuCmdProc.system();
     system->getDeviceMemory(writePkt)->access(writePkt);
+
+    delete writePkt;
 }
 
 void
@@ -671,7 +673,10 @@ AMDGPUDevice::getRegVal(uint64_t addr)
     DPRINTF(AMDGPUDevice, "Getting register 0x%lx = %x\n",
             fixup_addr, pkt->getLE<uint32_t>());
 
-    return pkt->getLE<uint32_t>();
+    pkt_data = pkt->getLE<uint32_t>();
+    delete pkt;
+
+    return pkt_data;
 }
 
 void
@@ -686,6 +691,7 @@ AMDGPUDevice::setRegVal(uint64_t addr, uint32_t value)
     PacketPtr pkt = Packet::createWrite(request);
     pkt->dataStatic((uint8_t *)&pkt_data);
     writeMMIO(pkt, addr);
+    delete pkt;
 }
 
 void

src/dev/amdgpu/interrupt_handler.cc

@@ -130,6 +130,10 @@ AMDGPUInterruptHandler::DmaEvent::process()
     } else {
         fatal("Interrupt Handler DMA event returned bad value: %d\n", data);
     }
+
+    if (dataPtr) {
+        delete [] dataPtr;
+    }
 }
 
 void
@@ -140,7 +144,7 @@ AMDGPUInterruptHandler::submitWritePointer()
     Addr paddr = regs.WptrAddr;
     std::memcpy(dataPtr, &regs.IH_Wptr, sizeof(uint32_t));
 
-    dmaEvent = new AMDGPUInterruptHandler::DmaEvent(this, 2);
+    dmaEvent = new AMDGPUInterruptHandler::DmaEvent(this, 2, dataPtr);
     dmaWrite(paddr, sizeof(uint32_t), dmaEvent, dataPtr);
 }
 
@@ -157,7 +161,7 @@ AMDGPUInterruptHandler::submitInterruptCookie()
 
     DPRINTF(AMDGPUDevice, "InterruptHandler rptr: 0x%x wptr: 0x%x\n",
                           regs.IH_Rptr, regs.IH_Wptr);
-    dmaEvent = new AMDGPUInterruptHandler::DmaEvent(this, 1);
+    dmaEvent = new AMDGPUInterruptHandler::DmaEvent(this, 1, dataPtr);
     dmaWrite(paddr, cookieSize, dmaEvent, dataPtr);
 
     interruptQueue.pop();

src/dev/amdgpu/interrupt_handler.hh

@@ -136,10 +136,12 @@ class AMDGPUInterruptHandler : public DmaDevice
       private:
         AMDGPUInterruptHandler *deviceIh;
         uint32_t data;
+        uint8_t *dataPtr;
 
       public:
-        DmaEvent(AMDGPUInterruptHandler *deviceIh, uint32_t data)
-            : Event(), deviceIh(deviceIh), data(data)
+        DmaEvent(AMDGPUInterruptHandler *deviceIh, uint32_t data,
+                 uint8_t* _dataPtr)
+            : Event(), deviceIh(deviceIh), data(data), dataPtr(_dataPtr)
         {
             setFlags(Event::AutoDelete);
         }

src/dev/amdgpu/pm4_packet_processor.cc

@@ -456,8 +456,6 @@ PM4PacketProcessor::mapQueues(PM4Queue *q, PM4MapQueues *pkt)
     } else {
         panic("Unknown engine for MQD: %d\n", pkt->engineSel);
     }
-
-    decodeNext(q);
 }
 
 void
@@ -494,6 +492,9 @@ PM4PacketProcessor::processMQD(PM4MapQueues *pkt, PM4Queue *q, Addr addr,
             "hqdAQL %d.\n", mqd->base, mqd->mqdBase, mqd->aql);
 
     gpuDevice->processPendingDoorbells(offset);
+
+    delete pkt;
+    decodeNext(q);
 }
 
 void
@@ -524,6 +525,9 @@ PM4PacketProcessor::processSDMAMQD(PM4MapQueues *pkt, PM4Queue *q, Addr addr,
     gpuDevice->setDoorbellType(pkt->doorbellOffset << 2, RLC, getIpId());
 
     gpuDevice->processPendingDoorbells(pkt->doorbellOffset << 2);
+
+    delete pkt;
+    decodeNext(q);
 }
 
 void
@@ -656,6 +660,7 @@ PM4PacketProcessor::unmapQueues(PM4Queue *q, PM4UnmapQueues *pkt)
                 dmaWriteVirt(addr, sizeof(QueueDesc), cb, mqd);
                 queues.erase(id);
                 hsa_pp.unsetDeviceQueueDesc(id, 8);
+                delete mqd;
             }
         }
         gpuDevice->deallocateAllQueues();
@@ -754,6 +759,7 @@ PM4PacketProcessor::indirectBuffer(PM4Queue *q, PM4IndirectBuf *pkt)
     q->ibBase(pkt->ibBase);
     q->wptr(pkt->ibSize * sizeof(uint32_t));
 
+    delete pkt;
     decodeNext(q);
 }
 
@@ -766,6 +772,7 @@ PM4PacketProcessor::switchBuffer(PM4Queue *q, PM4SwitchBuf *pkt)
     DPRINTF(PM4PacketProcessor, "PM4 switching buffer, rptr: %p.\n",
             q->wptr());
 
+    delete pkt;
     decodeNext(q);
 }
 
@@ -784,6 +791,7 @@ PM4PacketProcessor::setUconfigReg(PM4Queue *q, PM4SetUconfigReg *pkt)
     reg_addr += 0x40000 * getIpId();
     gpuDevice->setRegVal(reg_addr, pkt->data);
 
+    delete pkt;
     decodeNext(q);
 }
 
@@ -800,6 +808,7 @@ PM4PacketProcessor::waitRegMem(PM4Queue *q, PM4WaitRegMem *pkt)
     DPRINTF(PM4PacketProcessor, "    Mask: %lx\n", pkt->mask);
     DPRINTF(PM4PacketProcessor, "    Poll Interval: %lx\n", pkt->pollInterval);
 
+    delete pkt;
     decodeNext(q);
 }
 

src/dev/amdgpu/sdma_engine.cc

@@ -1132,7 +1132,7 @@ SDMAEngine::constFillDone(SDMAQueue *q, sdmaConstFill *pkt, uint8_t *fill_data)
 {
     DPRINTF(SDMAEngine, "ConstFill to %lx done\n", pkt->addr);
 
-    delete fill_data;
+    delete [] fill_data;
     delete pkt;
     decodeNext(q);
 }

src/gpu-compute/compute_unit.cc

@@ -1735,7 +1735,7 @@ ComputeUnit::DataPort::processMemReqEvent(PacketPtr pkt)
         SystemHubEvent *resp_event = new SystemHubEvent(pkt, this);
         compute_unit->shader->systemHub->sendRequest(pkt, resp_event);
     } else if (!(sendTimingReq(pkt))) {
-        retries.push_back(std::make_pair(pkt, gpuDynInst));
+        retries.emplace_back(pkt, gpuDynInst);
 
         if (gpuDynInst) {
             DPRINTF(GPUPort,
@@ -1772,7 +1772,7 @@ ComputeUnit::ScalarDataPort::MemReqEvent::process()
         SystemHubEvent *resp_event = new SystemHubEvent(pkt, &scalarDataPort);
         compute_unit->shader->systemHub->sendRequest(pkt, resp_event);
     } else if (!(scalarDataPort.sendTimingReq(pkt))) {
-        scalarDataPort.retries.push_back(pkt);
+        scalarDataPort.retries.emplace_back(pkt);
 
         DPRINTF(GPUPort,
                 "CU%d: WF[%d][%d]: addr %#x data req failed!\n",