GlobalISel: Handle EVT argument lowering correctly

handleAssignments was assuming every argument type is an MVT, and
assignArg would always fail. This fixes one of the hacks in the
current AMDGPU calling convention code that pre-processes the
arguments.

diff --git a/llvm/lib/CodeGen/GlobalISel/CallLowering.cpp b/llvm/lib/CodeGen/GlobalISel/CallLowering.cpp
index c2294fa..a714651 100644 (file)
--- a/llvm/lib/CodeGen/GlobalISel/CallLowering.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/CallLowering.cpp
@@ -194,11 +194,11 @@ bool CallLowering::handleAssignments(CCState &CCInfo,
 
   unsigned NumArgs = Args.size();
   for (unsigned i = 0; i != NumArgs; ++i) {
-    MVT CurVT = MVT::getVT(Args[i].Ty);
-    if (Handler.assignArg(i, CurVT, CurVT, CCValAssign::Full, Args[i],
-                          Args[i].Flags[0], CCInfo)) {
-      if (!CurVT.isValid())
-        return false;
+    EVT CurVT = EVT::getEVT(Args[i].Ty);
+    if (!CurVT.isSimple() ||
+        Handler.assignArg(i, CurVT.getSimpleVT(), CurVT.getSimpleVT(),
+                          CCValAssign::Full, Args[i], Args[i].Flags[0],
+                          CCInfo)) {
       MVT NewVT = TLI->getRegisterTypeForCallingConv(
           F.getContext(), F.getCallingConv(), EVT(CurVT));
 
@@ -309,8 +309,10 @@ bool CallLowering::handleAssignments(CCState &CCInfo,
     // FIXME: Pack registers if we have more than one.
     Register ArgReg = Args[i].Regs[0];
 
-    MVT OrigVT = MVT::getVT(Args[i].Ty);
-    MVT VAVT = VA.getValVT();
+    EVT OrigVT = EVT::getEVT(Args[i].Ty);
+    EVT VAVT = VA.getValVT();
+    const LLT OrigTy = getLLTForType(*Args[i].Ty, DL);
+
     if (VA.isRegLoc()) {
       if (Handler.isIncomingArgumentHandler() && VAVT != OrigVT) {
         if (VAVT.getSizeInBits() < OrigVT.getSizeInBits()) {
@@ -332,7 +334,7 @@ bool CallLowering::handleAssignments(CCState &CCInfo,
           MIRBuilder.buildMerge(Args[i].OrigRegs[0], Args[i].Regs);
           continue;
         }
-        const LLT VATy(VAVT);
+        const LLT VATy(VAVT.getSimpleVT());
         Register NewReg =
             MIRBuilder.getMRI()->createGenericVirtualRegister(VATy);
         Handler.assignValueToReg(NewReg, VA.getLocReg(), VA);
@@ -340,7 +342,6 @@ bool CallLowering::handleAssignments(CCState &CCInfo,
         // or do an unmerge to get the lower block of elements.
         if (VATy.isVector() &&
             VATy.getNumElements() > OrigVT.getVectorNumElements()) {
-          const LLT OrigTy(OrigVT);
           // Just handle the case where the VA type is 2 * original type.
           if (VATy.getNumElements() != OrigVT.getVectorNumElements() * 2) {
             LLVM_DEBUG(dbgs()

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp
index d170185..05a4e34 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp
@@ -226,19 +226,6 @@ void AMDGPUCallLowering::splitToValueTypes(
     MVT RegVT = TLI.getRegisterTypeForCallingConv(Ctx, CallConv, VT);
 
     if (NumParts == 1) {
-      // Fixup EVTs to an MVT.
-      //
-      // FIXME: This is pretty hacky. Why do we have to split the type
-      // legalization logic between here and handleAssignments?
-      if (OrigArgIdx != AttributeList::ReturnIndex && VT != RegVT) {
-        assert(VT.getSizeInBits() < 32 &&
-               "unexpected illegal type");
-        Ty = Type::getInt32Ty(Ctx);
-        Register OrigReg = Reg;
-        Reg = B.getMRI()->createGenericVirtualRegister(LLT::scalar(32));
-        B.buildTrunc(OrigReg, Reg);
-      }
-
       // No splitting to do, but we want to replace the original type (e.g. [1 x
       // double] -> double).
       SplitArgs.emplace_back(Reg, Ty, OrigArg.Flags, OrigArg.IsFixed);

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-function-args.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-function-args.ll
index c5bea01..340392e 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-function-args.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-function-args.ll
@@ -123,10 +123,11 @@ define void @void_func_i8(i8 %arg0) #0 {
   ; CHECK: bb.1 (%ir-block.0):
   ; CHECK:   liveins: $vgpr0, $sgpr30_sgpr31
   ; CHECK:   [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; CHECK:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+  ; CHECK:   [[TRUNC1:%[0-9]+]]:_(s8) = G_TRUNC [[TRUNC]](s16)
   ; CHECK:   [[COPY1:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
-  ; CHECK:   [[TRUNC:%[0-9]+]]:_(s8) = G_TRUNC [[COPY]](s32)
   ; CHECK:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
-  ; CHECK:   G_STORE [[TRUNC]](s8), [[DEF]](p1) :: (store 1 into `i8 addrspace(1)* undef`, addrspace 1)
+  ; CHECK:   G_STORE [[TRUNC1]](s8), [[DEF]](p1) :: (store 1 into `i8 addrspace(1)* undef`, addrspace 1)
   ; CHECK:   [[COPY2:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY1]]
   ; CHECK:   S_SETPC_B64_return [[COPY2]]
   store i8 %arg0, i8 addrspace(1)* undef
@@ -138,8 +139,8 @@ define void @void_func_i8_zeroext(i8 zeroext %arg0) #0 {
   ; CHECK: bb.1 (%ir-block.0):
   ; CHECK:   liveins: $vgpr0, $sgpr30_sgpr31
   ; CHECK:   [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[TRUNC:%[0-9]+]]:_(s8) = G_TRUNC [[COPY]](s32)
+  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
   ; CHECK:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; CHECK:   [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[TRUNC]](s8)
@@ -158,8 +159,8 @@ define void @void_func_i8_signext(i8 signext %arg0) #0 {
   ; CHECK: bb.1 (%ir-block.0):
   ; CHECK:   liveins: $vgpr0, $sgpr30_sgpr31
   ; CHECK:   [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[TRUNC:%[0-9]+]]:_(s8) = G_TRUNC [[COPY]](s32)
+  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
   ; CHECK:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; CHECK:   [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[TRUNC]](s8)
@@ -233,8 +234,8 @@ define void @void_func_i24(i24 %arg0) #0 {
   ; CHECK: bb.1 (%ir-block.0):
   ; CHECK:   liveins: $vgpr0, $sgpr30_sgpr31
   ; CHECK:   [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[TRUNC:%[0-9]+]]:_(s24) = G_TRUNC [[COPY]](s32)
+  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; CHECK:   G_STORE [[TRUNC]](s24), [[DEF]](p1) :: (store 3 into `i24 addrspace(1)* undef`, align 4, addrspace 1)
   ; CHECK:   [[COPY2:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY1]]
@@ -248,8 +249,8 @@ define void @void_func_i24_zeroext(i24 zeroext %arg0) #0 {
   ; CHECK: bb.1 (%ir-block.0):
   ; CHECK:   liveins: $vgpr0, $sgpr30_sgpr31
   ; CHECK:   [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[TRUNC:%[0-9]+]]:_(s24) = G_TRUNC [[COPY]](s32)
+  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; CHECK:   G_STORE [[TRUNC]](s24), [[DEF]](p1) :: (store 3 into `i24 addrspace(1)* undef`, align 4, addrspace 1)
   ; CHECK:   [[COPY2:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY1]]
@@ -263,8 +264,8 @@ define void @void_func_i24_signext(i24 signext %arg0) #0 {
   ; CHECK: bb.1 (%ir-block.0):
   ; CHECK:   liveins: $vgpr0, $sgpr30_sgpr31
   ; CHECK:   [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[TRUNC:%[0-9]+]]:_(s24) = G_TRUNC [[COPY]](s32)
+  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; CHECK:   G_STORE [[TRUNC]](s24), [[DEF]](p1) :: (store 3 into `i24 addrspace(1)* undef`, align 4, addrspace 1)
   ; CHECK:   [[COPY2:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY1]]
@@ -1569,11 +1570,12 @@ define void @void_func_struct_i8_i32({ i8, i32 } %arg0) #0 {
   ; CHECK: bb.1 (%ir-block.0):
   ; CHECK:   liveins: $vgpr0, $vgpr1, $sgpr30_sgpr31
   ; CHECK:   [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; CHECK:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+  ; CHECK:   [[TRUNC1:%[0-9]+]]:_(s8) = G_TRUNC [[TRUNC]](s16)
   ; CHECK:   [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; CHECK:   [[COPY2:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
-  ; CHECK:   [[TRUNC:%[0-9]+]]:_(s8) = G_TRUNC [[COPY]](s32)
   ; CHECK:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
-  ; CHECK:   G_STORE [[TRUNC]](s8), [[DEF]](p1) :: (store 1 into `{ i8, i32 } addrspace(1)* undef`, align 4, addrspace 1)
+  ; CHECK:   G_STORE [[TRUNC1]](s8), [[DEF]](p1) :: (store 1 into `{ i8, i32 } addrspace(1)* undef`, align 4, addrspace 1)
   ; CHECK:   [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 4
   ; CHECK:   [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[DEF]], [[C]](s64)
   ; CHECK:   G_STORE [[COPY1]](s32), [[PTR_ADD]](p1) :: (store 4 into `{ i8, i32 } addrspace(1)* undef` + 4, addrspace 1)
@@ -1766,14 +1768,13 @@ define void @void_func_v32i32_i1_i8_i16(<32 x i32> %arg0, i1 %arg1, i8 %arg2, i1
   ; CHECK:   [[FRAME_INDEX:%[0-9]+]]:_(p5) = G_FRAME_INDEX %fixed-stack.3
   ; CHECK:   [[LOAD:%[0-9]+]]:_(s1) = G_LOAD [[FRAME_INDEX]](p5) :: (invariant load 1 from %fixed-stack.3, align 16, addrspace 5)
   ; CHECK:   [[FRAME_INDEX1:%[0-9]+]]:_(p5) = G_FRAME_INDEX %fixed-stack.2
-  ; CHECK:   [[LOAD1:%[0-9]+]]:_(s32) = G_LOAD [[FRAME_INDEX1]](p5) :: (invariant load 4 from %fixed-stack.2, addrspace 5)
+  ; CHECK:   [[LOAD1:%[0-9]+]]:_(s8) = G_LOAD [[FRAME_INDEX1]](p5) :: (invariant load 1 from %fixed-stack.2, align 4, addrspace 5)
   ; CHECK:   [[FRAME_INDEX2:%[0-9]+]]:_(p5) = G_FRAME_INDEX %fixed-stack.1
   ; CHECK:   [[LOAD2:%[0-9]+]]:_(s16) = G_LOAD [[FRAME_INDEX2]](p5) :: (invariant load 2 from %fixed-stack.1, align 8, addrspace 5)
   ; CHECK:   [[FRAME_INDEX3:%[0-9]+]]:_(p5) = G_FRAME_INDEX %fixed-stack.0
   ; CHECK:   [[LOAD3:%[0-9]+]]:_(s16) = G_LOAD [[FRAME_INDEX3]](p5) :: (invariant load 2 from %fixed-stack.0, align 4, addrspace 5)
   ; CHECK:   [[COPY32:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[BUILD_VECTOR:%[0-9]+]]:_(<32 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32), [[COPY12]](s32), [[COPY13]](s32), [[COPY14]](s32), [[COPY15]](s32), [[COPY16]](s32), [[COPY17]](s32), [[COPY18]](s32), [[COPY19]](s32), [[COPY20]](s32), [[COPY21]](s32), [[COPY22]](s32), [[COPY23]](s32), [[COPY24]](s32), [[COPY25]](s32), [[COPY26]](s32), [[COPY27]](s32), [[COPY28]](s32), [[COPY29]](s32), [[COPY30]](s32), [[COPY31]](s32)
-  ; CHECK:   [[TRUNC:%[0-9]+]]:_(s8) = G_TRUNC [[LOAD1]](s32)
   ; CHECK:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; CHECK:   [[COPY33:%[0-9]+]]:_(p1) = COPY [[DEF]](p1)
   ; CHECK:   [[COPY34:%[0-9]+]]:_(p1) = COPY [[DEF]](p1)
@@ -1781,7 +1782,7 @@ define void @void_func_v32i32_i1_i8_i16(<32 x i32> %arg0, i1 %arg1, i8 %arg2, i1
   ; CHECK:   [[COPY36:%[0-9]+]]:_(p1) = COPY [[DEF]](p1)
   ; CHECK:   G_STORE [[BUILD_VECTOR]](<32 x s32>), [[DEF]](p1) :: (volatile store 128 into `<32 x i32> addrspace(1)* undef`, addrspace 1)
   ; CHECK:   G_STORE [[LOAD]](s1), [[COPY33]](p1) :: (volatile store 1 into `i1 addrspace(1)* undef`, addrspace 1)
-  ; CHECK:   G_STORE [[TRUNC]](s8), [[COPY34]](p1) :: (volatile store 1 into `i8 addrspace(1)* undef`, addrspace 1)
+  ; CHECK:   G_STORE [[LOAD1]](s8), [[COPY34]](p1) :: (volatile store 1 into `i8 addrspace(1)* undef`, addrspace 1)
   ; CHECK:   G_STORE [[LOAD2]](s16), [[COPY35]](p1) :: (volatile store 2 into `i16 addrspace(1)* undef`, addrspace 1)
   ; CHECK:   G_STORE [[LOAD3]](s16), [[COPY36]](p1) :: (volatile store 2 into `half addrspace(1)* undef`, addrspace 1)
   ; CHECK:   [[COPY37:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY32]]

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-ptrmask.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-ptrmask.ll
index cc1c75e..4f56e1a 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-ptrmask.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-ptrmask.ll
@@ -48,9 +48,9 @@ define i8* @ptrmask_flat_i16(i8* %ptr, i16 %mask) {
   ; CHECK:   [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; CHECK:   [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; CHECK:   [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; CHECK:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
   ; CHECK:   [[COPY3:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[MV:%[0-9]+]]:_(p0) = G_MERGE_VALUES [[COPY]](s32), [[COPY1]](s32)
-  ; CHECK:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
   ; CHECK:   [[PTRMASK:%[0-9]+]]:_(p0) = G_PTRMASK [[MV]], [[TRUNC]](s16)
   ; CHECK:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[PTRMASK]](p0)
   ; CHECK:   $vgpr0 = COPY [[UV]](s32)
@@ -119,8 +119,8 @@ define i8 addrspace(3)* @ptrmask_local_i16(i8 addrspace(3)* %ptr, i16 %mask) {
   ; CHECK:   liveins: $vgpr0, $vgpr1, $sgpr30_sgpr31
   ; CHECK:   [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
   ; CHECK:   [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
-  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[PTRMASK:%[0-9]+]]:_(p3) = G_PTRMASK [[COPY]], [[TRUNC]](s16)
   ; CHECK:   $vgpr0 = COPY [[PTRMASK]](p3)
   ; CHECK:   [[COPY3:%[0-9]+]]:ccr_sgpr_64 = COPY [[COPY2]]

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-sat.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-sat.ll
index 2d64858..efdecf7 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-sat.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-sat.ll
@@ -6,10 +6,10 @@ define i16 @uaddsat_i16(i16 %lhs, i16 %rhs) {
   ; CHECK: bb.1 (%ir-block.0):
   ; CHECK:   liveins: $vgpr0, $vgpr1, $sgpr30_sgpr31
   ; CHECK:   [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
-  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+  ; CHECK:   [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; CHECK:   [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[UADDSAT:%[0-9]+]]:_(s16) = G_UADDSAT [[TRUNC]], [[TRUNC1]]
   ; CHECK:   [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UADDSAT]](s16)
   ; CHECK:   $vgpr0 = COPY [[ANYEXT]](s32)
@@ -85,10 +85,10 @@ define i16 @saddsat_i16(i16 %lhs, i16 %rhs) {
   ; CHECK: bb.1 (%ir-block.0):
   ; CHECK:   liveins: $vgpr0, $vgpr1, $sgpr30_sgpr31
   ; CHECK:   [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
-  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+  ; CHECK:   [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; CHECK:   [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[SADDSAT:%[0-9]+]]:_(s16) = G_SADDSAT [[TRUNC]], [[TRUNC1]]
   ; CHECK:   [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SADDSAT]](s16)
   ; CHECK:   $vgpr0 = COPY [[ANYEXT]](s32)
@@ -164,10 +164,10 @@ define i16 @usubsat_i16(i16 %lhs, i16 %rhs) {
   ; CHECK: bb.1 (%ir-block.0):
   ; CHECK:   liveins: $vgpr0, $vgpr1, $sgpr30_sgpr31
   ; CHECK:   [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
-  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+  ; CHECK:   [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; CHECK:   [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[USUBSAT:%[0-9]+]]:_(s16) = G_USUBSAT [[TRUNC]], [[TRUNC1]]
   ; CHECK:   [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[USUBSAT]](s16)
   ; CHECK:   $vgpr0 = COPY [[ANYEXT]](s32)
@@ -243,10 +243,10 @@ define i16 @ssubsat_i16(i16 %lhs, i16 %rhs) {
   ; CHECK: bb.1 (%ir-block.0):
   ; CHECK:   liveins: $vgpr0, $vgpr1, $sgpr30_sgpr31
   ; CHECK:   [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
-  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
+  ; CHECK:   [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; CHECK:   [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr_64 = COPY $sgpr30_sgpr31
   ; CHECK:   [[SSUBSAT:%[0-9]+]]:_(s16) = G_SSUBSAT [[TRUNC]], [[TRUNC1]]
   ; CHECK:   [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SSUBSAT]](s16)
   ; CHECK:   $vgpr0 = COPY [[ANYEXT]](s32)