From 655ba251b50346262bee125d6fb29bb086a883d2 Mon Sep 17 00:00:00 2001
From: Tom Stellard <thomas.stellard@amd.com>
Date: Wed, 22 Jan 2014 19:24:19 +0000
Subject: [PATCH] R600: Begin private memory at the second GPR.

This way private memory does not over-write work group information
stored in GPRs 0 and 1.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@199824 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/R600/AMDGPUFrameLowering.cpp |  5 ++++-
 test/CodeGen/R600/private-memory.ll     | 25 +++++++++++++++++++++++++
 2 files changed, 29 insertions(+), 1 deletion(-)

diff --git a/lib/Target/R600/AMDGPUFrameLowering.cpp b/lib/Target/R600/AMDGPUFrameLowering.cpp
index 40f14d2f67c..40cc9083c00 100644
--- a/lib/Target/R600/AMDGPUFrameLowering.cpp
+++ b/lib/Target/R600/AMDGPUFrameLowering.cpp
@@ -74,7 +74,10 @@ unsigned AMDGPUFrameLowering::getStackWidth(const MachineFunction &MF) const {
 int AMDGPUFrameLowering::getFrameIndexOffset(const MachineFunction &MF,
                                          int FI) const {
   const MachineFrameInfo *MFI = MF.getFrameInfo();
-  unsigned Offset = 0;
+  // Start the offset at 2 so we don't overwrite work group information.
+  // XXX: We should only do this when the shader actually uses this
+  // information.
+  unsigned Offset = 2;
   int UpperBound = FI == -1 ? MFI->getNumObjects() : FI;
 
   for (int i = MFI->getObjectIndexBegin(); i < UpperBound; ++i) {
diff --git a/test/CodeGen/R600/private-memory.ll b/test/CodeGen/R600/private-memory.ll
index 3fd67d75bb7..b25fc7ba572 100644
--- a/test/CodeGen/R600/private-memory.ll
+++ b/test/CodeGen/R600/private-memory.ll
@@ -156,3 +156,28 @@ entry:
   ret void
 
 }
+
+; Make sure we don't overwrite workitem information with private memory
+
+; FUNC-LABEL: @work_item_info
+; R600-CHECK-NOT: MOV T0.X
+; Additional check in case the move ends up in the last slot
+; R600-CHECK-NOT: MOV * TO.X
+
+; SI-CHECK-NOT: V_MOV_B32_e{{(32|64)}} v0
+define void @work_item_info(i32 addrspace(1)* %out, i32 %in) {
+entry:
+  %0 = alloca [2 x i32]
+  %1 = getelementptr [2 x i32]* %0, i32 0, i32 0
+  %2 = getelementptr [2 x i32]* %0, i32 0, i32 1
+  store i32 0, i32* %1
+  store i32 1, i32* %2
+  %3 = getelementptr [2 x i32]* %0, i32 0, i32 %in
+  %4 = load i32* %3
+  %5 = call i32 @llvm.r600.read.tidig.x()
+  %6 = add i32 %4, %5
+  store i32 %6, i32 addrspace(1)* %out
+  ret void
+}
+
+declare i32 @llvm.r600.read.tidig.x() nounwind readnone