gpu-compute: replace uint32_t* casts with bits API calls

The uint32_t* casting was challenging to fully understand what was
being done at a glance. Replaced with calls to various bits functions
as it's functionally equivalent and much more clear.

This also fixes a segfault in GPUInitAbi DPRINTFs from a mis-typed
uint32_t* cast.

Change-Id: Id5d1863942848dd7a9e5e17e8180c33adbc72f15
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/34677
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Matt Sinclair <mattdsinclair@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
diff --git a/src/gpu-compute/fetch_unit.cc b/src/gpu-compute/fetch_unit.cc
index 098b783..5d98288 100644
--- a/src/gpu-compute/fetch_unit.cc
+++ b/src/gpu-compute/fetch_unit.cc
@@ -33,6 +33,7 @@
 
 #include "gpu-compute/fetch_unit.hh"
 
+#include "base/bitfield.hh"
 #include "debug/GPUFetch.hh"
 #include "debug/GPUPort.hh"
 #include "debug/GPUTLB.hh"
@@ -576,7 +577,8 @@
     int num_dwords = sizeof(TheGpuISA::RawMachInst) / dword_size;
 
     for (int i = 0; i < num_dwords; ++i) {
-        ((uint32_t*)(&split_inst))[i] = *reinterpret_cast<uint32_t*>(readPtr);
+        replaceBits(split_inst, 32*(i+1)-1, 32*i,
+            *reinterpret_cast<uint32_t*>(readPtr));
         if (readPtr + dword_size >= bufEnd) {
             readPtr = bufStart;
         }
diff --git a/src/gpu-compute/wavefront.cc b/src/gpu-compute/wavefront.cc
index 0e737db..dd914ca 100644
--- a/src/gpu-compute/wavefront.cc
+++ b/src/gpu-compute/wavefront.cc
@@ -33,6 +33,7 @@
 
 #include "gpu-compute/wavefront.hh"
 
+#include "base/bitfield.hh"
 #include "debug/GPUExec.hh"
 #include "debug/GPUInitAbi.hh"
 #include "debug/WavefrontStack.hh"
@@ -257,23 +258,23 @@
                 physSgprIdx =
                     computeUnit->registerManager->mapSgpr(this, regInitIdx);
                 computeUnit->srf[simdId]->write(physSgprIdx,
-                        ((uint32_t*)&host_disp_pkt_addr)[0]);
+                        bits(host_disp_pkt_addr, 31, 0));
                 ++regInitIdx;
                 DPRINTF(GPUInitAbi, "CU%d: WF[%d][%d]: wave[%d] "
                         "Setting DispatchPtr: s[%d] = %x\n",
                         computeUnit->cu_id, simdId,
                         wfSlotId, wfDynId, physSgprIdx,
-                        ((uint32_t*)&host_disp_pkt_addr)[0]);
+                        bits(host_disp_pkt_addr, 31, 0));
 
                 physSgprIdx =
                     computeUnit->registerManager->mapSgpr(this, regInitIdx);
                 computeUnit->srf[simdId]->write(physSgprIdx,
-                        ((uint32_t*)&host_disp_pkt_addr)[1]);
+                        bits(host_disp_pkt_addr, 63, 32));
                 DPRINTF(GPUInitAbi, "CU%d: WF[%d][%d]: wave[%d] "
                         "Setting DispatchPtr: s[%d] = %x\n",
                         computeUnit->cu_id, simdId,
                         wfSlotId, wfDynId, physSgprIdx,
-                        ((uint32_t*)&host_disp_pkt_addr)[1]);
+                        bits(host_disp_pkt_addr, 63, 32));
 
                 ++regInitIdx;
                 break;
@@ -281,23 +282,23 @@
                 physSgprIdx =
                     computeUnit->registerManager->mapSgpr(this, regInitIdx);
                 computeUnit->srf[simdId]->write(physSgprIdx,
-                        ((uint32_t*)&task->hostAMDQueueAddr)[0]);
+                        bits(task->hostAMDQueueAddr, 31, 0));
                 ++regInitIdx;
                 DPRINTF(GPUInitAbi, "CU%d: WF[%d][%d]: wave[%d] "
                         "Setting QueuePtr: s[%d] = %x\n",
                         computeUnit->cu_id, simdId,
                         wfSlotId, wfDynId, physSgprIdx,
-                       ((uint32_t*)&task->hostAMDQueueAddr)[0]);
+                        bits(task->hostAMDQueueAddr, 31, 0));
 
                 physSgprIdx =
                     computeUnit->registerManager->mapSgpr(this, regInitIdx);
                 computeUnit->srf[simdId]->write(physSgprIdx,
-                        ((uint32_t*)&task->hostAMDQueueAddr)[1]);
+                        bits(task->hostAMDQueueAddr, 63, 32));
                 DPRINTF(GPUInitAbi, "CU%d: WF[%d][%d]: wave[%d] "
                         "Setting QueuePtr: s[%d] = %x\n",
                         computeUnit->cu_id, simdId,
                         wfSlotId, wfDynId, physSgprIdx,
-                       ((uint32_t*)&task->hostAMDQueueAddr)[1]);
+                        bits(task->hostAMDQueueAddr, 63, 32));
 
                 ++regInitIdx;
                 break;
@@ -305,23 +306,23 @@
                 physSgprIdx =
                     computeUnit->registerManager->mapSgpr(this, regInitIdx);
                 computeUnit->srf[simdId]->write(physSgprIdx,
-                        ((uint32_t*)&kernarg_addr)[0]);
+                        bits(kernarg_addr, 31, 0));
                 ++regInitIdx;
                 DPRINTF(GPUInitAbi, "CU%d: WF[%d][%d]: wave[%d] "
                         "Setting KernargSegPtr: s[%d] = %x\n",
                         computeUnit->cu_id, simdId,
                         wfSlotId, wfDynId, physSgprIdx,
-                       ((uint32_t*)kernarg_addr)[0]);
+                        bits(kernarg_addr, 31, 0));
 
                 physSgprIdx =
                     computeUnit->registerManager->mapSgpr(this, regInitIdx);
                 computeUnit->srf[simdId]->write(physSgprIdx,
-                        ((uint32_t*)&kernarg_addr)[1]);
+                        bits(kernarg_addr, 63, 32));
                 DPRINTF(GPUInitAbi, "CU%d: WF[%d][%d]: wave[%d] "
                         "Setting KernargSegPtr: s[%d] = %x\n",
                         computeUnit->cu_id, simdId,
                         wfSlotId, wfDynId, physSgprIdx,
-                       ((uint32_t*)kernarg_addr)[1]);
+                        bits(kernarg_addr, 63, 32));
 
                 ++regInitIdx;
                 break;