TTI: Report 0 cost for free addrspacecasts

llvm-svn: 264369

2 files changed, 50 insertions, 0 deletions

diff --git a/llvm/include/llvm/CodeGen/BasicTTIImpl.h b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
index e614440dee0..db9960c1aa0 100644
--- a/llvm/include/llvm/CodeGen/BasicTTIImpl.h
+++ b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
@@ -359,6 +359,11 @@ public:
         TLI->isZExtFree(SrcLT.second, DstLT.second))
       return 0;
 
+    if (Opcode == Instruction::AddrSpaceCast &&
+        TLI->isNoopAddrSpaceCast(Src->getPointerAddressSpace(),
+                                 Dst->getPointerAddressSpace()))
+      return 0;
+
     // If the cast is marked as legal (or promote) then assume low cost.
     if (SrcLT.first == DstLT.first &&
         TLI->isOperationLegalOrPromote(ISD, DstLT.second))
diff --git a/llvm/test/Analysis/CostModel/AMDGPU/addrspacecast.ll b/llvm/test/Analysis/CostModel/AMDGPU/addrspacecast.ll
new file mode 100644
index 00000000000..1f6cb85a789
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/AMDGPU/addrspacecast.ll
@@ -0,0 +1,45 @@
+; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri < %s | FileCheck %s
+
+; CHECK: 'addrspacecast_global_to_flat'
+; CHECK: estimated cost of 0 for {{.*}} addrspacecast i8 addrspace(1)* %ptr to i8 addrspace(4)*
+define i8 addrspace(4)* @addrspacecast_global_to_flat(i8 addrspace(1)* %ptr) #0 {
+  %cast = addrspacecast i8 addrspace(1)* %ptr to i8 addrspace(4)*
+  ret i8 addrspace(4)* %cast
+}
+
+; CHECK: 'addrspacecast_global_to_flat_v2'
+; CHECK: estimated cost of 0 for {{.*}} addrspacecast <2 x i8 addrspace(1)*> %ptr to <2 x i8 addrspace(4)*>
+define <2 x i8 addrspace(4)*> @addrspacecast_global_to_flat_v2(<2 x i8 addrspace(1)*> %ptr) #0 {
+  %cast = addrspacecast <2 x i8 addrspace(1)*> %ptr to <2 x i8 addrspace(4)*>
+  ret <2 x i8 addrspace(4)*> %cast
+}
+
+; CHECK: 'addrspacecast_global_to_flat_v32'
+; CHECK: estimated cost of 0 for {{.*}} addrspacecast <32 x i8 addrspace(1)*> %ptr to <32 x i8 addrspace(4)*>
+define <32 x i8 addrspace(4)*> @addrspacecast_global_to_flat_v32(<32 x i8 addrspace(1)*> %ptr) #0 {
+  %cast = addrspacecast <32 x i8 addrspace(1)*> %ptr to <32 x i8 addrspace(4)*>
+  ret <32 x i8 addrspace(4)*> %cast
+}
+
+; CHECK: 'addrspacecast_local_to_flat'
+; CHECK: estimated cost of 1 for {{.*}} addrspacecast i8 addrspace(3)* %ptr to i8 addrspace(4)*
+define i8 addrspace(4)* @addrspacecast_local_to_flat(i8 addrspace(3)* %ptr) #0 {
+  %cast = addrspacecast i8 addrspace(3)* %ptr to i8 addrspace(4)*
+  ret i8 addrspace(4)* %cast
+}
+
+; CHECK: 'addrspacecast_local_to_flat_v2'
+; CHECK: estimated cost of 2 for {{.*}} addrspacecast <2 x i8 addrspace(3)*> %ptr to <2 x i8 addrspace(4)*>
+define <2 x i8 addrspace(4)*> @addrspacecast_local_to_flat_v2(<2 x i8 addrspace(3)*> %ptr) #0 {
+  %cast = addrspacecast <2 x i8 addrspace(3)*> %ptr to <2 x i8 addrspace(4)*>
+  ret <2 x i8 addrspace(4)*> %cast
+}
+
+; CHECK: 'addrspacecast_local_to_flat_v32'
+; CHECK: estimated cost of 32 for {{.*}} addrspacecast <32 x i8 addrspace(3)*> %ptr to <32 x i8 addrspace(4)*>
+define <32 x i8 addrspace(4)*> @addrspacecast_local_to_flat_v32(<32 x i8 addrspace(3)*> %ptr) #0 {
+  %cast = addrspacecast <32 x i8 addrspace(3)*> %ptr to <32 x i8 addrspace(4)*>
+  ret <32 x i8 addrspace(4)*> %cast
+}
+
+attributes #0 = { nounwind readnone }