Index: llvm/trunk/lib/Target/AMDGPU/SIFixSGPRCopies.cpp
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIFixSGPRCopies.cpp
+++ llvm/trunk/lib/Target/AMDGPU/SIFixSGPRCopies.cpp
@@ -215,7 +215,7 @@
 
   for (unsigned I = 1, N = MI.getNumOperands(); I != N; I += 2) {
     unsigned SrcReg = MI.getOperand(I).getReg();
-    unsigned SrcSubReg = MI.getOperand(I).getReg();
+    unsigned SrcSubReg = MI.getOperand(I).getSubReg();
 
     const TargetRegisterClass *SrcRC = MRI.getRegClass(SrcReg);
     assert(TRI->isSGPRClass(SrcRC) &&
Index: llvm/trunk/lib/Target/AMDGPU/SIFoldOperands.cpp
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIFoldOperands.cpp
+++ llvm/trunk/lib/Target/AMDGPU/SIFoldOperands.cpp
@@ -334,13 +334,10 @@
           !MRI.hasOneUse(MI.getOperand(0).getReg()))
         continue;
 
-      // FIXME: Fold operands with subregs.
       if (OpToFold.isReg() &&
-          (!TargetRegisterInfo::isVirtualRegister(OpToFold.getReg()) ||
-           OpToFold.getSubReg()))
+          !TargetRegisterInfo::isVirtualRegister(OpToFold.getReg()))
         continue;
 
-
       // We need mutate the operands of new mov instructions to add implicit
       // uses of EXEC, but adding them invalidates the use_iterator, so defer
       // this.
Index: llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.cpp
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -1777,6 +1777,10 @@
     MRI.getRegClass(Reg) :
     RI.getPhysRegClass(Reg);
 
+  const SIRegisterInfo *TRI =
+      static_cast<const SIRegisterInfo*>(MRI.getTargetRegisterInfo());
+  RC = TRI->getSubRegClass(RC, MO.getSubReg());
+
   // In order to be legal, the common sub-class must be equal to the
   // class of the current operand.  For example:
   //
Index: llvm/trunk/lib/Target/AMDGPU/SIRegisterInfo.cpp
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIRegisterInfo.cpp
+++ llvm/trunk/lib/Target/AMDGPU/SIRegisterInfo.cpp
@@ -464,12 +464,38 @@
   if (SubIdx == AMDGPU::NoSubRegister)
     return RC;
 
-  // If this register has a sub-register, we can safely assume it is a 32-bit
-  // register, because all of SI's sub-registers are 32-bit.
+  // We can assume that each lane corresponds to one 32-bit register.
+  unsigned Count = countPopulation(getSubRegIndexLaneMask(SubIdx));
   if (isSGPRClass(RC)) {
-    return &AMDGPU::SGPR_32RegClass;
+    switch (Count) {
+    case 1:
+      return &AMDGPU::SGPR_32RegClass;
+    case 2:
+      return &AMDGPU::SReg_64RegClass;
+    case 4:
+      return &AMDGPU::SReg_128RegClass;
+    case 8:
+      return &AMDGPU::SReg_256RegClass;
+    case 16: /* fall-through */
+    default:
+      llvm_unreachable("Invalid sub-register class size");
+    }
   } else {
-    return &AMDGPU::VGPR_32RegClass;
+    switch (Count) {
+    case 1:
+      return &AMDGPU::VGPR_32RegClass;
+    case 2:
+      return &AMDGPU::VReg_64RegClass;
+    case 3:
+      return &AMDGPU::VReg_96RegClass;
+    case 4:
+      return &AMDGPU::VReg_128RegClass;
+    case 8:
+      return &AMDGPU::VReg_256RegClass;
+    case 16: /* fall-through */
+    default:
+      llvm_unreachable("Invalid sub-register class size");
+    }
   }
 }
 
Index: llvm/trunk/test/CodeGen/AMDGPU/fmin_legacy.ll
===================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/fmin_legacy.ll
+++ llvm/trunk/test/CodeGen/AMDGPU/fmin_legacy.ll
@@ -8,8 +8,8 @@
 
 ; FUNC-LABEL: @test_fmin_legacy_f32
 ; EG: MIN *
-; SI-SAFE: v_min_legacy_f32_e32
-; SI-NONAN: v_min_f32_e32
+; SI-SAFE: v_min_legacy_f32_e64
+; SI-NONAN: v_min_f32_e64
 define void @test_fmin_legacy_f32(<4 x float> addrspace(1)* %out, <4 x float> inreg %reg0) #0 {
    %r0 = extractelement <4 x float> %reg0, i32 0
    %r1 = extractelement <4 x float> %reg0, i32 1
Index: llvm/trunk/test/CodeGen/AMDGPU/fsub.ll
===================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/fsub.ll
+++ llvm/trunk/test/CodeGen/AMDGPU/fsub.ll
@@ -32,9 +32,8 @@
 ; R600-DAG: ADD {{\** *}}T{{[0-9]+\.[XYZW]}}, KC0[3].X, -KC0[3].Z
 ; R600-DAG: ADD {{\** *}}T{{[0-9]+\.[XYZW]}}, KC0[2].W, -KC0[3].Y
 
-; FIXME: Should be using SGPR directly for first operand
-; SI: v_subrev_f32_e32 {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}
-; SI: v_subrev_f32_e32 {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}
+; SI: v_subrev_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, {{v[0-9]+}}
+; SI: v_subrev_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, {{v[0-9]+}}
 define void @fsub_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2 x float> %b) {
   %sub = fsub <2 x float> %a, %b
   store <2 x float> %sub, <2 x float> addrspace(1)* %out, align 8
@@ -60,13 +59,11 @@
   ret void
 }
 
-; FIXME: Should be using SGPR directly for first operand
-
 ; FUNC-LABEL: {{^}}s_fsub_v4f32:
-; SI: v_subrev_f32_e32 {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}
-; SI: v_subrev_f32_e32 {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}
-; SI: v_subrev_f32_e32 {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}
-; SI: v_subrev_f32_e32 {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}
+; SI: v_subrev_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, {{v[0-9]+}}
+; SI: v_subrev_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, {{v[0-9]+}}
+; SI: v_subrev_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, {{v[0-9]+}}
+; SI: v_subrev_f32_e32 {{v[0-9]+}}, {{s[0-9]+}}, {{v[0-9]+}}
 ; SI: s_endpgm
 define void @s_fsub_v4f32(<4 x float> addrspace(1)* %out, <4 x float> %a, <4 x float> %b) {
   %result = fsub <4 x float> %a, %b
Index: llvm/trunk/test/CodeGen/AMDGPU/llvm.round.f64.ll
===================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/llvm.round.f64.ll
+++ llvm/trunk/test/CodeGen/AMDGPU/llvm.round.f64.ll
@@ -21,7 +21,7 @@
 ; SI-DAG: v_cmp_eq_i32
 
 ; SI-DAG: s_mov_b32 [[BFIMASK:s[0-9]+]], 0x7fffffff
-; SI-DAG: v_cmp_gt_i32_e32
+; SI-DAG: v_cmp_gt_i32
 ; SI-DAG: v_bfi_b32 [[COPYSIGN:v[0-9]+]], [[BFIMASK]]
 
 ; SI: buffer_store_dwordx2