arch: Bump MaxVecRegLenInBytes to 4096

The GPU model uses the generic vector register
containers, however the maximum vector register
length is fixed at 256, which is an invalid assumption
for the GPU model as it can operate on vectors up
to 4096B.

Change-Id: Id85e0ed45c9a9c1a4bb6e712c44eaeec2d628fce
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17908
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>

diff --git a/src/arch/generic/vec_reg.hh b/src/arch/generic/vec_reg.hh
index ed2545c64dd6c4fbb041ff4e70c00efa32f5b130..aab307b42fca0ec942295e861f9e31a064fba0b4 100644 (file)
--- a/src/arch/generic/vec_reg.hh
+++ b/src/arch/generic/vec_reg.hh
@@ -154,7 +154,7 @@
 #include "base/cprintf.hh"
 #include "base/logging.hh"
 
-constexpr unsigned MaxVecRegLenInBytes = 256;
+constexpr unsigned MaxVecRegLenInBytes = 4096;
 
 template <size_t Sz>
 class VecRegContainer;
@@ -523,6 +523,7 @@ class VecLaneT
     friend class VecRegContainer<32>;
     friend class VecRegContainer<64>;
     friend class VecRegContainer<128>;
+    friend class VecRegContainer<256>;
     friend class VecRegContainer<MaxVecRegLenInBytes>;
 
     /** My type alias. */