[PATCH 1/2] R600: Promote i64 loads to v2i32

[Tom Stellard]

---
 lib/Target/R600/AMDGPUISelLowering.cpp | 3 +++
 lib/Target/R600/SIISelLowering.cpp     | 9 ++++++++-
 lib/Target/R600/SIInstructions.td      | 7 +------
 test/CodeGen/R600/add_i64.ll           | 6 +++---
 test/CodeGen/R600/load.ll              | 3 +--
 test/CodeGen/R600/or.ll                | 6 +++---
 6 files changed, 19 insertions(+), 15 deletions(-)

diff --git a/lib/Target/R600/AMDGPUISelLowering.cpp b/lib/Target/R600/AMDGPUISelLowering.cpp
index 84e78ba..354d4d0 100644
--- a/lib/Target/R600/AMDGPUISelLowering.cpp
+++ b/lib/Target/R600/AMDGPUISelLowering.cpp
@@ -185,6 +185,9 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::LOAD, MVT::v2f32, Promote);
   AddPromotedToType(ISD::LOAD, MVT::v2f32, MVT::v2i32);
 
+  setOperationAction(ISD::LOAD, MVT::i64, Promote);
+  AddPromotedToType(ISD::LOAD, MVT::i64, MVT::v2i32);
+
   setOperationAction(ISD::LOAD, MVT::v4f32, Promote);
   AddPromotedToType(ISD::LOAD, MVT::v4f32, MVT::v4i32);
 
diff --git a/lib/Target/R600/SIISelLowering.cpp b/lib/Target/R600/SIISelLowering.cpp
index 29e4b98..f491116 100644
--- a/lib/Target/R600/SIISelLowering.cpp
+++ b/lib/Target/R600/SIISelLowering.cpp
@@ -91,7 +91,6 @@ SITargetLowering::SITargetLowering(TargetMachine &TM) :
 
   // We need to custom lower loads/stores from private memory
   setOperationAction(ISD::LOAD, MVT::i32, Custom);
-  setOperationAction(ISD::LOAD, MVT::i64, Custom);
   setOperationAction(ISD::LOAD, MVT::v2i32, Custom);
   setOperationAction(ISD::LOAD, MVT::v4i32, Custom);
   setOperationAction(ISD::LOAD, MVT::v8i32, Custom);
@@ -602,6 +601,14 @@ SDValue SITargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   case ISD::BRCOND: return LowerBRCOND(Op, DAG);
   case ISD::LOAD: {
     LoadSDNode *Load = dyn_cast<LoadSDNode>(Op);
+    EVT VT = Op.getValueType();
+
+    // These loads are legal.
+    if (Load->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS &&
+        VT.isVector() && VT.getVectorNumElements() == 2 &&
+        VT.getVectorElementType() == MVT::i32)
+      return SDValue();
+
     if (Op.getValueType().isVector() &&
         (Load->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS ||
          Load->getAddressSpace() == AMDGPUAS::PRIVATE_ADDRESS ||
diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
index d4b93d1..7e71bbd 100644
--- a/lib/Target/R600/SIInstructions.td
+++ b/lib/Target/R600/SIInstructions.td
@@ -1724,7 +1724,6 @@ multiclass SMRD_Pattern <SMRD Instr_IMM, SMRD Instr_SGPR, ValueType vt> {
 
 defm : SMRD_Pattern <S_LOAD_DWORD_IMM, S_LOAD_DWORD_SGPR, f32>;
 defm : SMRD_Pattern <S_LOAD_DWORD_IMM, S_LOAD_DWORD_SGPR, i32>;
-defm : SMRD_Pattern <S_LOAD_DWORDX2_IMM, S_LOAD_DWORDX2_SGPR, i64>;
 defm : SMRD_Pattern <S_LOAD_DWORDX2_IMM, S_LOAD_DWORDX2_SGPR, v2i32>;
 defm : SMRD_Pattern <S_LOAD_DWORDX4_IMM, S_LOAD_DWORDX4_SGPR, v4i32>;
 defm : SMRD_Pattern <S_LOAD_DWORDX8_IMM, S_LOAD_DWORDX8_SGPR, v32i8>;
@@ -2372,7 +2371,7 @@ defm : DSReadPat <DS_READ_U8,  i32, az_extloadi8_local>;
 defm : DSReadPat <DS_READ_I16, i32, sextloadi16_local>;
 defm : DSReadPat <DS_READ_U16, i32, az_extloadi16_local>;
 defm : DSReadPat <DS_READ_B32, i32, local_load>;
-defm : DSReadPat <DS_READ_B64, i64, local_load>;
+defm : DSReadPat <DS_READ_B64, v2i32, local_load>;
 
 multiclass DSWritePat <DS inst, ValueType vt, PatFrag frag> {
   def : Pat <
@@ -2520,10 +2519,6 @@ defm : MUBUFLoad_Pattern <BUFFER_LOAD_USHORT_ADDR64, i32,
                           az_extloadi16_global, az_extloadi16_constant>;
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORD_ADDR64, i32,
                           global_load, constant_load>;
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, i64,
-                          global_load, constant_load>;
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, i64,
-                          az_extloadi32_global, az_extloadi32_constant>;
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, v2i32,
                           global_load, constant_load>;
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX4_ADDR64, v4i32,
diff --git a/test/CodeGen/R600/add_i64.ll b/test/CodeGen/R600/add_i64.ll
index c9eaeda..f733d90 100644
--- a/test/CodeGen/R600/add_i64.ll
+++ b/test/CodeGen/R600/add_i64.ll
@@ -70,9 +70,9 @@ define void @test_v2i64_vreg(<2 x i64> addrspace(1)* noalias %out, <2 x i64> add
 }
 
 ; SI-LABEL: @trunc_i64_add_to_i32
-; SI: S_LOAD_DWORD [[SREG0:s[0-9]+]],
-; SI: S_LOAD_DWORD [[SREG1:s[0-9]+]],
-; SI: S_ADD_I32 [[SRESULT:s[0-9]+]], [[SREG1]], [[SREG0]]
+; SI: S_LOAD_DWORDX2 s{{\[}}[[SREG0:[0-9]+]]
+; SI: S_LOAD_DWORDX2 s{{\[}}[[SREG1:[0-9]+]]
+; SI: S_ADD_I32 [[SRESULT:s[0-9]+]], s[[SREG1]], s[[SREG0]]
 ; SI-NOT: ADDC
 ; SI: V_MOV_B32_e32 [[VRESULT:v[0-9]+]], [[SRESULT]]
 ; SI: BUFFER_STORE_DWORD [[VRESULT]],
diff --git a/test/CodeGen/R600/load.ll b/test/CodeGen/R600/load.ll
index 1486c4d..a57df5c 100644
--- a/test/CodeGen/R600/load.ll
+++ b/test/CodeGen/R600/load.ll
@@ -696,8 +696,7 @@ entry:
 ; R600-CHECK: LDS_READ_RET
 ; R600-CHECK: LDS_READ_RET
 ; SI-CHECK: S_MOV_B32 m0
-; SI-CHECK: DS_READ_B32
-; SI-CHECK: DS_READ_B32
+; SI-CHECK: DS_READ_B64
 define void @load_v2f32_local(<2 x float> addrspace(1)* %out, <2 x float> addrspace(3)* %in) {
 entry:
   %0 = load <2 x float> addrspace(3)* %in
diff --git a/test/CodeGen/R600/or.ll b/test/CodeGen/R600/or.ll
index 9878366..91a70b7 100644
--- a/test/CodeGen/R600/or.ll
+++ b/test/CodeGen/R600/or.ll
@@ -116,9 +116,9 @@ define void @vector_or_i64_imm(i64 addrspace(1)* %out, i64 addrspace(1)* %a, i64
 }
 
 ; SI-LABEL: @trunc_i64_or_to_i32
-; SI: S_LOAD_DWORD [[SREG0:s[0-9]+]],
-; SI: S_LOAD_DWORD [[SREG1:s[0-9]+]],
-; SI: S_OR_B32 [[SRESULT:s[0-9]+]], [[SREG1]], [[SREG0]]
+; SI: S_LOAD_DWORDX2 s{{\[}}[[SREG0:[0-9]+]]
+; SI: S_LOAD_DWORDX2 s{{\[}}[[SREG1:[0-9]+]]
+; SI: S_OR_B32 [[SRESULT:s[0-9]+]], s[[SREG1]], s[[SREG0]]
 ; SI: V_MOV_B32_e32 [[VRESULT:v[0-9]+]], [[SRESULT]]
 ; SI: BUFFER_STORE_DWORD [[VRESULT]],
 define void @trunc_i64_or_to_i32(i32 addrspace(1)* %out, i64 %a, i64 %b) {
-- 
1.8.1.5