From 59767cea79e9ca727b5d2de8a6aa2318d671afd5 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Fri, 25 Mar 2016 00:14:11 +0000
Subject: [PATCH] AMDGPU: TTI: Make insertelement free.

We don't want to have a cost to scalarizing operations.

llvm-svn: 264364
---
 .../Target/AMDGPU/AMDGPUTargetTransformInfo.cpp    |  5 +++
 .../Analysis/CostModel/AMDGPU/insertelement.ll     | 37 ++++++++++++++++++++++
 2 files changed, 42 insertions(+)
 create mode 100644 llvm/test/Analysis/CostModel/AMDGPU/insertelement.ll

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
index 9cd3c39..39be33e 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
@@ -99,6 +99,11 @@ int AMDGPUTTIImpl::getVectorInstrCost(unsigned Opcode, Type *ValTy,
                                       unsigned Index) {
   switch (Opcode) {
   case Instruction::ExtractElement:
+  case Instruction::InsertElement:
+    // Extracts are just reads of a subregister, so are free. Inserts are
+    // considered free because we don't want to have any cost for scalarizing
+    // operations, and we don't have to copy into a different register class.
+
     // Dynamic indexing isn't free and is best avoided.
     return Index == ~0u ? 2 : 0;
   default:
diff --git a/llvm/test/Analysis/CostModel/AMDGPU/insertelement.ll b/llvm/test/Analysis/CostModel/AMDGPU/insertelement.ll
new file mode 100644
index 0000000..1765afe
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/AMDGPU/insertelement.ll
@@ -0,0 +1,37 @@
+; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa < %s | FileCheck %s
+
+; CHECK: 'insertelement_v2i32'
+; CHECK: estimated cost of 0 for {{.*}} insertelement <2 x i32>
+define void @insertelement_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %vaddr) {
+  %vec = load <2 x i32>, <2 x i32> addrspace(1)* %vaddr
+  %insert = insertelement <2 x i32> %vec, i32 1, i32 123
+  store <2 x i32> %insert, <2 x i32> addrspace(1)* %out
+  ret void
+}
+
+; CHECK: 'insertelement_v2i64'
+; CHECK: estimated cost of 0 for {{.*}} insertelement <2 x i64>
+define void @insertelement_v2i64(<2 x i64> addrspace(1)* %out, <2 x i64> addrspace(1)* %vaddr) {
+  %vec = load <2 x i64>, <2 x i64> addrspace(1)* %vaddr
+  %insert = insertelement <2 x i64> %vec, i64 1, i64 123
+  store <2 x i64> %insert, <2 x i64> addrspace(1)* %out
+  ret void
+}
+
+; CHECK: 'insertelement_v2i16'
+; CHECK: estimated cost of 0 for {{.*}} insertelement <2 x i16>
+define void @insertelement_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %vaddr) {
+  %vec = load <2 x i16>, <2 x i16> addrspace(1)* %vaddr
+  %insert = insertelement <2 x i16> %vec, i16 1, i16 123
+  store <2 x i16> %insert, <2 x i16> addrspace(1)* %out
+  ret void
+}
+
+; CHECK: 'insertelement_v2i8'
+; CHECK: estimated cost of 0 for {{.*}} insertelement <2 x i8>
+define void @insertelement_v2i8(<2 x i8> addrspace(1)* %out, <2 x i8> addrspace(1)* %vaddr) {
+  %vec = load <2 x i8>, <2 x i8> addrspace(1)* %vaddr
+  %insert = insertelement <2 x i8> %vec, i8 1, i8 123
+  store <2 x i8> %insert, <2 x i8> addrspace(1)* %out
+  ret void
+}
-- 
2.7.4