Make constant memory opt-in, spill large statics to global memory

crates/rustc_codegen_nvvm/src/builder.rs

@@ -1154,18 +1154,7 @@ impl<'ll, 'tcx, 'a> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
 
 impl<'ll> StaticBuilderMethods for Builder<'_, 'll, '_> {
     fn get_static(&mut self, def_id: DefId) -> &'ll Value {
-        unsafe {
-            let mut g = self.cx.get_static(def_id);
-            let llty = self.val_ty(g);
-            let addrspace = AddressSpace(llvm::LLVMGetPointerAddressSpace(llty));
-            if addrspace != AddressSpace::DATA {
-                trace!("Remapping global address space of global {:?}", g);
-                let llty = llvm::LLVMGetElementType(llty);
-                let ty = self.type_ptr_to_ext(llty, AddressSpace::DATA);
-                g = llvm::LLVMBuildAddrSpaceCast(self.llbuilder, g, ty, unnamed());
-            }
-            g
-        }
+        self.cx.get_static(def_id)
     }
 }
 

crates/rustc_codegen_nvvm/src/context.rs

@@ -22,7 +22,7 @@ use rustc_errors::DiagMessage;
 use rustc_hash::FxHashMap;
 use rustc_middle::dep_graph::DepContext;
 use rustc_middle::ty::layout::{
-    FnAbiError, FnAbiOf, FnAbiRequest, HasTyCtxt, HasTypingEnv, LayoutError,
+    FnAbiError, FnAbiOf, FnAbiRequest, HasTyCtxt, HasTypingEnv, LayoutError, LayoutOf,
 };
 use rustc_middle::ty::layout::{FnAbiOfHelpers, LayoutOfHelpers};
 use rustc_middle::ty::{Ty, TypeVisitableExt};
@@ -40,6 +40,10 @@ use rustc_target::callconv::FnAbi;
 use rustc_target::spec::{HasTargetSpec, Target};
 use tracing::{debug, trace};
 
+/// "There is a total of 64 KB constant memory on a device."
+/// <https://docs.nvidia.com/cuda/archive/12.8.1/pdf/CUDA_C_Best_Practices_Guide.pdf>
+const CONSTANT_MEMORY_SIZE_LIMIT_BYTES: u64 = 64 * 1024;
+
 pub(crate) struct CodegenCx<'ll, 'tcx> {
     pub tcx: TyCtxt<'tcx>,
 
@@ -267,7 +271,18 @@ impl<'ll, 'tcx> CodegenCx<'ll, 'tcx> {
         }
 
         if !is_mutable && self.type_is_freeze(ty) {
-            AddressSpace(4)
+            let layout = self.layout_of(ty);
+            if layout.size.bytes() > CONSTANT_MEMORY_SIZE_LIMIT_BYTES {
+                self.tcx.sess.dcx().warn(format!(
+                    "static `{}` exceeds the constant-memory limit; placing in global memory (performance may be reduced)",
+                    instance
+                ));
+                // Global memory
+                AddressSpace(1)
+            } else {
+                // Constant memory
+                AddressSpace(4)
+            }
         } else {
             AddressSpace::DATA
         }