From 9f5d593c1f61db2d8a5fc52835435570d26f1118 Mon Sep 17 00:00:00 2001
From: Tom Stellard <thomas.stellard@amd.com>
Date: Wed, 11 Feb 2015 00:34:35 +0000
Subject: [PATCH] R600/SI: Store immediate offsets > 12-bits in soffset

This will save us from having to extend these offsets to 64-bits
and storing them in a pair of vgprs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@228776 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/R600/AMDGPUISelDAGToDAG.cpp | 32 +++++++++++++++++++-------------
 test/CodeGen/R600/mubuf.ll             | 21 ++++++++++++---------
 2 files changed, 31 insertions(+), 22 deletions(-)

diff --git a/lib/Target/R600/AMDGPUISelDAGToDAG.cpp b/lib/Target/R600/AMDGPUISelDAGToDAG.cpp
index ebe34431da2..ac05e4ae2ce 100644
--- a/lib/Target/R600/AMDGPUISelDAGToDAG.cpp
+++ b/lib/Target/R600/AMDGPUISelDAGToDAG.cpp
@@ -925,26 +925,32 @@ void AMDGPUDAGToDAGISel::SelectMUBUF(SDValue Addr, SDValue &Ptr,
     SDValue N1 = Addr.getOperand(1);
     ConstantSDNode *C1 = cast<ConstantSDNode>(N1);
 
-    if (isLegalMUBUFImmOffset(C1)) {
-
-      if (N0.getOpcode() == ISD::ADD) {
-        // (add (add N2, N3), C1) -> addr64
-        SDValue N2 = N0.getOperand(0);
-        SDValue N3 = N0.getOperand(1);
-        Addr64 = CurDAG->getTargetConstant(1, MVT::i1);
-        Ptr = N2;
-        VAddr = N3;
-        Offset = CurDAG->getTargetConstant(C1->getZExtValue(), MVT::i16);
-        return;
-      }
+    if (N0.getOpcode() == ISD::ADD) {
+      // (add (add N2, N3), C1) -> addr64
+      SDValue N2 = N0.getOperand(0);
+      SDValue N3 = N0.getOperand(1);
+      Addr64 = CurDAG->getTargetConstant(1, MVT::i1);
+      Ptr = N2;
+      VAddr = N3;
+    } else {
 
       // (add N0, C1) -> offset
       VAddr = CurDAG->getTargetConstant(0, MVT::i32);
       Ptr = N0;
-      Offset = CurDAG->getTargetConstant(C1->getZExtValue(), MVT::i16);
+    }
+
+    if (isLegalMUBUFImmOffset(C1)) {
+        Offset = CurDAG->getTargetConstant(C1->getZExtValue(), MVT::i16);
+        return;
+    } else if (isUInt<32>(C1->getZExtValue())) {
+      // Illegal offset, store it in soffset.
+      Offset = CurDAG->getTargetConstant(0, MVT::i16);
+      SOffset = SDValue(CurDAG->getMachineNode(AMDGPU::S_MOV_B32, DL, MVT::i32,
+                   CurDAG->getTargetConstant(C1->getZExtValue(), MVT::i32)), 0);
       return;
     }
   }
+
   if (Addr.getOpcode() == ISD::ADD) {
     // (add N0, N1) -> addr64
     SDValue N0 = Addr.getOperand(0);
diff --git a/test/CodeGen/R600/mubuf.ll b/test/CodeGen/R600/mubuf.ll
index 9c2a17ce04f..e43c83d0b40 100644
--- a/test/CodeGen/R600/mubuf.ll
+++ b/test/CodeGen/R600/mubuf.ll
@@ -30,7 +30,8 @@ entry:
 
 ; MUBUF load with an immediate byte offset that doesn't fit into 12-bits
 ; CHECK-LABEL: {{^}}mubuf_load2:
-; CHECK: buffer_load_dword v{{[0-9]}}, v[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], 0 addr64 ; encoding: [0x00,0x80,0x30,0xe0
+; CHECK: s_movk_i32 [[SOFFSET:s[0-9]+]], 0x1000
+; CHECK: buffer_load_dword v{{[0-9]}}, s[{{[0-9]+:[0-9]+}}], [[SOFFSET]] ; encoding: [0x00,0x00,0x30,0xe0
 define void @mubuf_load2(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
 entry:
   %0 = getelementptr i32 addrspace(1)* %in, i64 1024
@@ -85,12 +86,6 @@ main_body:
   ret void
 }
 
-declare i32 @llvm.SI.buffer.load.dword.i32.i32(<16 x i8>, i32, i32, i32, i32, i32, i32, i32, i32) #3
-declare void @llvm.SI.tbuffer.store.i32(<16 x i8>, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32)
-
-attributes #1 = { "ShaderType"="2" "unsafe-fp-math"="true" }
-attributes #3 = { nounwind readonly }
-
 ;;;==========================================================================;;;
 ;;; MUBUF STORE TESTS
 ;;;==========================================================================;;;
@@ -118,7 +113,8 @@ entry:
 
 ; MUBUF store with an immediate byte offset that doesn't fit into 12-bits
 ; CHECK-LABEL: {{^}}mubuf_store2:
-; CHECK: buffer_store_dword v{{[0-9]}}, v[{{[0-9]+:[0-9]+}}], s[{{[0-9]:[0-9]}}], 0 addr64 ; encoding: [0x00,0x80,0x70,0xe0
+; CHECK: s_movk_i32 [[SOFFSET:s[0-9]+]], 0x1000
+; CHECK: buffer_store_dword v{{[0-9]}}, s[{{[0-9]:[0-9]}}], [[SOFFSET]] ; encoding: [0x00,0x00,0x70,0xe0
 define void @mubuf_store2(i32 addrspace(1)* %out) {
 entry:
   %0 = getelementptr i32 addrspace(1)* %out, i64 1024
@@ -154,7 +150,8 @@ define void @store_sgpr_ptr_offset(i32 addrspace(1)* %out) #0 {
 }
 
 ; CHECK-LABEL: {{^}}store_sgpr_ptr_large_offset:
-; CHECK: buffer_store_dword v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0 addr64
+; CHECK: s_mov_b32 [[SOFFSET:s[0-9]+]], 0x20000
+; CHECK: buffer_store_dword v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}}, [[SOFFSET]]
 define void @store_sgpr_ptr_large_offset(i32 addrspace(1)* %out) #0 {
   %out.gep = getelementptr i32 addrspace(1)* %out, i32 32768
   store i32 99, i32 addrspace(1)* %out.gep, align 4
@@ -169,3 +166,9 @@ define void @store_vgpr_ptr(i32 addrspace(1)* %out) #0 {
   store i32 99, i32 addrspace(1)* %out.gep, align 4
   ret void
 }
+
+declare i32 @llvm.SI.buffer.load.dword.i32.i32(<16 x i8>, i32, i32, i32, i32, i32, i32, i32, i32) #3
+declare void @llvm.SI.tbuffer.store.i32(<16 x i8>, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32)
+
+attributes #1 = { "ShaderType"="2" "unsafe-fp-math"="true" }
+attributes #3 = { nounwind readonly }
-- 
2.11.0