arch-gcn3,arch-vega: Select proper data on misaligned access

req1->getSize() returns the size in bytes, but because we're using it
in an array index, we need to scale it by the size of the data type.

This ensures we give the second request the proper data.

Change-Id: I578665406762d5d0c95f2ea8297c362e1cc0620b
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/54503
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Maintainer: Matt Sinclair <mattdsinclair@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matthew Poremba <matthew.poremba@amd.com>

src/arch/amdgpu/gcn3/gpu_mem_helpers.hh

diff --git a/src/arch/amdgpu/gcn3/gpu_mem_helpers.hh b/src/arch/amdgpu/gcn3/gpu_mem_helpers.hh
index 009bb7c..05299e1 100644
--- a/src/arch/amdgpu/gcn3/gpu_mem_helpers.hh
+++ b/src/arch/amdgpu/gcn3/gpu_mem_helpers.hh
@@ -107,7 +107,8 @@
                 pkt1->dataStatic(&(reinterpret_cast<T*>(
                     gpuDynInst->d_data))[lane * N]);
                 pkt2->dataStatic(&(reinterpret_cast<T*>(
-                    gpuDynInst->d_data))[lane * N + req1->getSize()]);
+                    gpuDynInst->d_data))[lane * N +
+                                         req1->getSize()/sizeof(T)]);
                 DPRINTF(GPUMem, "CU%d: WF[%d][%d]: index: %d unaligned memory "
                         "request for %#x\n", gpuDynInst->cu_id,
                         gpuDynInst->simdId, gpuDynInst->wfSlotId, lane,

src/arch/amdgpu/vega/gpu_mem_helpers.hh

diff --git a/src/arch/amdgpu/vega/gpu_mem_helpers.hh b/src/arch/amdgpu/vega/gpu_mem_helpers.hh
index c60325d..a5a9ec9 100644
--- a/src/arch/amdgpu/vega/gpu_mem_helpers.hh
+++ b/src/arch/amdgpu/vega/gpu_mem_helpers.hh
@@ -107,7 +107,8 @@
                 pkt1->dataStatic(&(reinterpret_cast<T*>(
                     gpuDynInst->d_data))[lane * N]);
                 pkt2->dataStatic(&(reinterpret_cast<T*>(
-                    gpuDynInst->d_data))[lane * N + req1->getSize()]);
+                    gpuDynInst->d_data))[lane * N +
+                                         req1->getSize()/sizeof(T)]);
                 DPRINTF(GPUMem, "CU%d: WF[%d][%d]: index: %d unaligned memory "
                         "request for %#x\n", gpuDynInst->cu_id,
                         gpuDynInst->simdId, gpuDynInst->wfSlotId, lane,