[llvm] r269480 - AMDGPU/R600: Fold global address operand

[Jan Vesely via llvm-commits]

Author: jvesely
Date: Fri May 13 15:39:31 2016
New Revision: 269480

URL: http://llvm.org/viewvc/llvm-project?rev=269480&view=rev
Log:
AMDGPU/R600: Fold global address operand

Reviewers: tstellard

Subscribers: arsenm

Differential Revision: http://reviews.llvm.org/D19793

Modified:
    llvm/trunk/lib/Target/AMDGPU/R600ISelLowering.cpp
    llvm/trunk/test/CodeGen/AMDGPU/gv-const-addrspace.ll

Modified: llvm/trunk/lib/Target/AMDGPU/R600ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/R600ISelLowering.cpp?rev=269480&r1=269479&r2=269480&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/R600ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/AMDGPU/R600ISelLowering.cpp Fri May 13 15:39:31 2016
@@ -2250,6 +2250,13 @@ FoldOperand(SDNode *ParentNode, unsigned
     Src = DAG.getRegister(AMDGPU::ALU_CONST, MVT::f32);
     return true;
   }
+  case AMDGPU::MOV_IMM_GLOBAL_ADDR:
+    // Check if the Imm slot is used. Taken from below.
+    if (cast<ConstantSDNode>(Imm)->getZExtValue())
+      return false;
+    Imm = Src.getOperand(0);
+    Src = DAG.getRegister(AMDGPU::ALU_LITERAL_X, MVT::i32);
+    return true;
   case AMDGPU::MOV_IMM_I32:
   case AMDGPU::MOV_IMM_F32: {
     unsigned ImmReg = AMDGPU::ALU_LITERAL_X;

Modified: llvm/trunk/test/CodeGen/AMDGPU/gv-const-addrspace.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/gv-const-addrspace.ll?rev=269480&r1=269479&r2=269480&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/gv-const-addrspace.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/gv-const-addrspace.ll Fri May 13 15:39:31 2016
@@ -14,6 +14,7 @@
 ; EG: VTX_READ_32
 ; EG: @float_gv
 ; EG-NOT: MOVA_INT
+; EG-NOT: MOV
 define void @float(float addrspace(1)* %out, i32 %index) {
 entry:
   %0 = getelementptr inbounds [5 x float], [5 x float] addrspace(2)* @float_gv, i32 0, i32 %index
@@ -31,6 +32,7 @@ entry:
 ; EG: VTX_READ_32
 ; EG: @i32_gv
 ; EG-NOT: MOVA_INT
+; EG-NOT: MOV
 define void @i32(i32 addrspace(1)* %out, i32 %index) {
 entry:
   %0 = getelementptr inbounds [5 x i32], [5 x i32] addrspace(2)* @i32_gv, i32 0, i32 %index
@@ -50,6 +52,7 @@ entry:
 ; EG: VTX_READ_32
 ; EG: @struct_foo_gv
 ; EG-NOT: MOVA_INT
+; EG-NOT: MOV
 define void @struct_foo_gv_load(i32 addrspace(1)* %out, i32 %index) {
   %gep = getelementptr inbounds [1 x %struct.foo], [1 x %struct.foo] addrspace(2)* @struct_foo_gv, i32 0, i32 0, i32 1, i32 %index
   %load = load i32, i32 addrspace(2)* %gep, align 4
@@ -68,6 +71,7 @@ define void @struct_foo_gv_load(i32 addr
 ; EG: VTX_READ_32
 ; EG: @array_v1_gv
 ; EG-NOT: MOVA_INT
+; EG-NOT: MOV
 define void @array_v1_gv_load(<1 x i32> addrspace(1)* %out, i32 %index) {
   %gep = getelementptr inbounds [4 x <1 x i32>], [4 x <1 x i32>] addrspace(2)* @array_v1_gv, i32 0, i32 %index
   %load = load <1 x i32>, <1 x i32> addrspace(2)* %gep, align 4