Support inline asm with multiple 64bit output in 32bit GPR

[thopre]

Summary

Extend fix for PR34170 to support inline assembly with multiple output operands that do not naturally go in the register class it is constrained to (eg. double in a 32-bit GPR as in the PR).

[thopre]

Ping?

[thopre]

Rebase

[thopre]

Rename testcase and test both soft and hard float ABI

[thopre]

Rebase on latest version of previous patch in stack

[thopre]

Rebase and rename testcase for consistency

[thopre]

Ping? This fix the bug reported in https://reviews.llvm.org/rL337903

[efriedma]

This fix the bug reported in https://reviews.llvm.org/rL337903

Do we have test coverage for that issue reported in that email? If not, please add a testcase.

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

7843

isSized()? How can you end up with an unsized value here?

7863

This part of the patch is just whitespace changes?

test/CodeGen/ARM/inlineasm-operand-implicit-cast.ll

175

Please add a test for multiple return values with matching inputs.

[thopre]

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

7843

It was in the original code so I kept it. I've now changed it to an assert and can confirm the testsuite shows no regression.

7863

The comment yes, the rest is s/ResultType/ResultVT

[thopre]

• Change test for isSized() into assert
• Add test for multiple return values with matching input
• Rewrite hardfloat tests to return a struct to make register used predictable and not dependent on register allocator

[efriedma]

LGTM

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

void SelectionDAGBuilder::visitInlineAsm(ImmutableCallSite CS) {
   Flag = Chain.getValue(1);
 
   // If this asm returns a register value, copy the result from that register
   // and set it as the value of the call.
   if (!RetValRegs.Regs.empty()) {
     SDValue Val = RetValRegs.getCopyFromRegs(DAG, FuncInfo, getCurSDLoc(),
                                              Chain, &Flag, CS.getInstruction());
 
-    // FIXME: Why don't we do this for inline asms with MRVs?
-    if (CS.getType()->isSingleValueType() && CS.getType()->isSized()) {
-      EVT ResultType = TLI.getValueType(DAG.getDataLayout(), CS.getType());
-
+    llvm::Type *CSResultType = CS.getType();
+    unsigned numRet;
+    ArrayRef<Type *> ResultTypes;
+    SmallVector<SDValue, 1> ResultValues(1);
+    if (CSResultType->isSingleValueType()) {
+      numRet = 1;
+      ResultValues[0] = Val;
+      ResultTypes = makeArrayRef(CSResultType);
+    } else {
+      numRet = CSResultType->getNumContainedTypes();
+      assert(Val->getNumOperands() == numRet &&
+             "Mismatch in number of output operands in asm result");
+      ResultTypes = CSResultType->subtypes();
+      ArrayRef<SDUse> ValueUses = Val->ops();
+      ResultValues.resize(numRet);
+      std::transform(ValueUses.begin(), ValueUses.end(), ResultValues.begin(),
+                     [](const SDUse &u) -> SDValue { return u.get(); });
+    }
+    SmallVector<EVT, 1> ResultVTs(numRet);
+    for (unsigned i = 0; i < numRet; i++) {
+      EVT ResultVT = TLI.getValueType(DAG.getDataLayout(), ResultTypes[i]);
+      SDValue Val = ResultValues[i];
+      assert(ResultTypes[i]->isSized() && "Unexpected unsized type");
       // If the type of the inline asm call site return value is different but
       // has same size as the type of the asm output bitcast it.  One example
       // of this is for vectors with different width / number of elements.
       // This can happen for register classes that can contain multiple
       // different value types.  The preg or vreg allocated may not have the
       // same VT as was expected.
       //
       // This can also happen for a return value that disagrees with the
       // register class it is put in, eg. a double in a general-purpose
       // register on a 32-bit machine.
-      if (ResultType != Val.getValueType() &&
-          ResultType.getSizeInBits() == Val.getValueSizeInBits()) {
-        Val = DAG.getNode(ISD::BITCAST, getCurSDLoc(),
-                          ResultType, Val);
-
-      } else if (ResultType != Val.getValueType() &&
-                 ResultType.isInteger() && Val.getValueType().isInteger()) {
-        // If a result value was tied to an input value, the computed result may
-        // have a wider width than the expected result.  Extract the relevant
-        // portion.
-        Val = DAG.getNode(ISD::TRUNCATE, getCurSDLoc(), ResultType, Val);
-      }
-
-      assert(ResultType == Val.getValueType() && "Asm result value mismatch!");
+      if (ResultVT != Val.getValueType() &&
+          ResultVT.getSizeInBits() == Val.getValueSizeInBits())
+        Val = DAG.getNode(ISD::BITCAST, getCurSDLoc(), ResultVT, Val);
+      else if (ResultVT != Val.getValueType() && ResultVT.isInteger() &&
+               Val.getValueType().isInteger()) {
+        // If a result value was tied to an input value, the computed result
+        // may have a wider width than the expected result.  Extract the
+        // relevant portion.
+        Val = DAG.getNode(ISD::TRUNCATE, getCurSDLoc(), ResultVT, Val);
+      }
+
+      assert(ResultVT == Val.getValueType() && "Asm result value mismatch!");
+      ResultVTs[i] = ResultVT;
+      ResultValues[i] = Val;
     }
 
+    Val = DAG.getNode(ISD::MERGE_VALUES, getCurSDLoc(),
+                      DAG.getVTList(ResultVTs), ResultValues);
     setValue(CS.getInstruction(), Val);
     // Don't need to use this as a chain in this case.
     if (!IA->hasSideEffects() && !hasMemory && IndirectStoresToEmit.empty())
       return;
   }
 
   std::vector<std::pair<SDValue, const Value *>> StoresToEmit;
 
   // Process indirect outputs, first output all of the flagged copies out of
   // physregs.
   for (unsigned i = 0, e = IndirectStoresToEmit.size(); i != e; ++i) {
     RegsForValue &OutRegs = IndirectStoresToEmit[i].first;
     const Value *Ptr = IndirectStoresToEmit[i].second;
     SDValue OutVal = OutRegs.getCopyFromRegs(DAG, FuncInfo, getCurSDLoc(),
                                              Chain, &Flag, IA);
     StoresToEmit.push_back(std::make_pair(OutVal, Ptr));
   }
 
   // Emit the non-flagged stores from the physregs.
   SmallVector<SDValue, 8> OutChains;
   for (unsigned i = 0, e = StoresToEmit.size(); i != e; ++i) {
     SDValue Val = DAG.getStore(Chain, getCurSDLoc(), StoresToEmit[i].first,
                                getValue(StoresToEmit[i].second),
                                MachinePointerInfo(StoresToEmit[i].second));
     OutChains.push_back(Val);
   }
 
   if (!OutChains.empty())
     Chain = DAG.getNode(ISD::TokenFactor, getCurSDLoc(), MVT::Other, OutChains);
 
   DAG.setRoot(Chain);
 }
 
 void SelectionDAGBuilder::emitInlineAsmError(ImmutableCallSite CS,
                                              const Twine &Message) {
   LLVMContext &Ctx = *DAG.getContext();
   Ctx.emitError(CS.getInstruction(), Message);
 
   // Make sure we leave the DAG in a valid state
   const TargetLowering &TLI = DAG.getTargetLoweringInfo();
   SmallVector<EVT, 1> ValueVTs;
   ComputeValueVTs(TLI, DAG.getDataLayout(), CS->getType(), ValueVTs);

test/CodeGen/ARM/inline-asm-operand-implicit-cast.ll

-; RUN: llc -mtriple armv7-arm-linux-gnueabihf -O2 -mcpu=cortex-a7 < %s | FileCheck %s
-
-; Check support for returning a float in GPR with soft float ABI
-define arm_aapcscc float @zerobits_float_soft() #0 {
-; CHECK-LABEL: zerobits_float_soft
-; CHECK: mov r0, #0
-  %1 = tail call float asm "mov ${0}, #0", "=&r"()
-  ret float %1
-}
-
-; Check support for returning a double in GPR with soft float ABI
-define arm_aapcscc double @zerobits_double_soft() #0 {
-; CHECK-LABEL: zerobits_double_soft
-; CHECK: mov r0, #0
-; CHECK-NEXT: mov r1, #0
-  %1 = tail call double asm "mov ${0:Q}, #0\0Amov ${0:R}, #0", "=&r"()
-  ret double %1
-}
-
-; Check support for returning a float in GPR with matching float input with
-; soft float ABI
-define arm_aapcscc float @flt_gpr_matching_in_op_soft(float %f) #0 {
-; CHECK-LABEL: flt_gpr_matching_in_op_soft
-; CHECK: mov r0, r0
-  %1 = call float asm "mov $0, $1", "=&r,0"(float %f)
-  ret float %1
-}
-
-; Check support for returning a double in GPR with matching double input with
-; soft float ABI
-define arm_aapcscc double @dbl_gpr_matching_in_op_soft(double %d) #0 {
-; CHECK-LABEL: dbl_gpr_matching_in_op_soft
-; CHECK: mov r1, r0
-  %1 = call double asm "mov ${0:R}, ${1:Q}", "=&r,0"(double %d)
-  ret double %1
-}
-
-; Check support for returning a float in specific GPR with matching float input
-; with soft float ABI
-define arm_aapcscc float @flt_gpr_matching_spec_reg_in_op_soft(float %f) #0 {
-; CHECK-LABEL: flt_gpr_matching_spec_reg_in_op_soft
-; CHECK: mov r3, r3
-  %1 = call float asm "mov $0, $1", "=&{r3},0"(float %f)
-  ret float %1
-}
-
-; Check support for returning a double in specific GPR with matching double
-; input with soft float ABI
-define arm_aapcscc double @dbl_gpr_matching_spec_reg_in_op_soft(double %d) #0 {
-; CHECK-LABEL: dbl_gpr_matching_spec_reg_in_op_soft
-; CHECK: mov r3, r2
-  %1 = call double asm "mov ${0:R}, ${1:Q}", "=&{r2},0"(double %d)
-  ret double %1
-}
-
-attributes #0 = { nounwind "target-features"="+d16,+vfp2,+vfp3,-fp-only-sp" "use-soft-float"="true" }
-
-
-; Check support for returning a float in GPR with hard float ABI
-define float @zerobits_float_hard() #1 {
-; CHECK-LABEL: zerobits_float_hard
-; CHECK: mov r0, #0
-; CHECK: vmov s0, r0
-  %1 = tail call float asm "mov ${0}, #0", "=&r"()
-  ret float %1
-}
-
-; Check support for returning a double in GPR with hard float ABI
-define double @zerobits_double_hard() #1 {
-; CHECK-LABEL: zerobits_double_hard
-; CHECK: mov r0, #0
-; CHECK-NEXT: mov r1, #0
-; CHECK: vmov d0, r0, r1
-  %1 = tail call double asm "mov ${0:Q}, #0\0Amov ${0:R}, #0", "=&r"()
-  ret double %1
-}
-
-; Check support for returning a float in GPR with matching float input with
-; hard float ABI
-define float @flt_gpr_matching_in_op_hard(float %f) #1 {
-; CHECK-LABEL: flt_gpr_matching_in_op_hard
-; CHECK: vmov r0, s0
-; CHECK: mov r0, r0
-; CHECK: vmov s0, r0
-  %1 = call float asm "mov $0, $1", "=&r,0"(float %f)
-  ret float %1
-}
-
-; Check support for returning a double in GPR with matching double input with
-; hard float ABI
-define double @dbl_gpr_matching_in_op_hard(double %d) #1 {
-; CHECK-LABEL: dbl_gpr_matching_in_op_hard
-; CHECK: vmov r0, r1, d0
-; CHECK: mov r1, r0
-; CHECK: vmov d0, r0, r1
-  %1 = call double asm "mov ${0:R}, ${1:Q}", "=&r,0"(double %d)
-  ret double %1
-}
-
-; Check support for returning a float in specific GPR with matching float
-; input with hard float ABI
-define float @flt_gpr_matching_spec_reg_in_op_hard(float %f) #1 {
-; CHECK-LABEL: flt_gpr_matching_spec_reg_in_op_hard
-; CHECK: vmov r3, s0
-; CHECK: mov r3, r3
-; CHECK: vmov s0, r3
-  %1 = call float asm "mov $0, $1", "=&{r3},0"(float %f)
-  ret float %1
-}
-
-; Check support for returning a double in specific GPR with matching double
-; input with hard float ABI
-define double @dbl_gpr_matching_spec_reg_in_op_hard(double %d) #1 {
-; CHECK-LABEL: dbl_gpr_matching_spec_reg_in_op_hard
-; CHECK: vmov r2, r3, d0
-; CHECK: mov r3, r2
-; CHECK: vmov d0, r2, r3
-  %1 = call double asm "mov ${0:R}, ${1:Q}", "=&{r2},0"(double %d)
-  ret double %1
-}
-
-attributes #1 = { nounwind "target-features"="+d16,+vfp2,+vfp3,-fp-only-sp" "use-soft-float"="false" }

test/CodeGen/ARM/inlineasm-operand-implicit-cast.ll

+; RUN: llc -mtriple armv7-arm-linux-gnueabihf -O2 -mcpu=cortex-a7 < %s | FileCheck %s
+
+%struct.twofloat = type { float, float }
+%struct.twodouble = type { double, double }
+
+; Check support for returning a float in GPR with soft float ABI
+define arm_aapcscc float @zerobits_float_soft() #0 {
+; CHECK-LABEL: zerobits_float_soft
+; CHECK: mov r0, #0
+  %1 = tail call float asm "mov ${0}, #0", "=&r"()
+  ret float %1
+}
+
+; Check support for returning a double in GPR with soft float ABI
+define arm_aapcscc double @zerobits_double_soft() #0 {
+; CHECK-LABEL: zerobits_double_soft
+; CHECK: mov r0, #0
+; CHECK-NEXT: mov r1, #0
+  %1 = tail call double asm "mov ${0:Q}, #0\0Amov ${0:R}, #0", "=&r"()
+  ret double %1
+}
+
+; Check support for returning a float in GPR with matching float input with
+; soft float ABI
+define arm_aapcscc float @flt_gpr_matching_in_op_soft(float %f) #0 {
+; CHECK-LABEL: flt_gpr_matching_in_op_soft
+; CHECK: mov r0, r0
+  %1 = call float asm "mov $0, $1", "=&r,0"(float %f)
+  ret float %1
+}
+
+; Check support for returning a double in GPR with matching double input with
+; soft float ABI
+define arm_aapcscc double @dbl_gpr_matching_in_op_soft(double %d) #0 {
+; CHECK-LABEL: dbl_gpr_matching_in_op_soft
+; CHECK: mov r1, r0
+  %1 = call double asm "mov ${0:R}, ${1:Q}", "=&r,0"(double %d)
+  ret double %1
+}
+
+; Check support for returning a float in specific GPR with matching float input
+; with soft float ABI
+define arm_aapcscc float @flt_gpr_matching_spec_reg_in_op_soft(float %f) #0 {
+; CHECK-LABEL: flt_gpr_matching_spec_reg_in_op_soft
+; CHECK: mov r3, r3
+  %1 = call float asm "mov $0, $1", "=&{r3},0"(float %f)
+  ret float %1
+}
+
+; Check support for returning a double in specific GPR with matching double
+; input with soft float ABI
+define arm_aapcscc double @dbl_gpr_matching_spec_reg_in_op_soft(double %d) #0 {
+; CHECK-LABEL: dbl_gpr_matching_spec_reg_in_op_soft
+; CHECK: mov r3, r2
+  %1 = call double asm "mov ${0:R}, ${1:Q}", "=&{r2},0"(double %d)
+  ret double %1
+}
+
+; Check support for returning several float in GPR
+define arm_aapcscc float @zerobits_float_convoluted_soft() #0 {
+; CHECK-LABEL: zerobits_float_convoluted_soft
+; CHECK: mov r0, #0
+; CHECK-NEXT: mov r1, #0
+  %1 = call { float, float } asm "mov $0, #0; mov $1, #0", "=r,=r"()
+  %asmresult = extractvalue { float, float } %1, 0
+  %asmresult1 = extractvalue { float, float } %1, 1
+  %add = fadd float %asmresult, %asmresult1
+  ret float %add
+}
+
+; Check support for returning several double in GPR
+define double @zerobits_double_convoluted_soft() #0 {
+; CHECK-LABEL: zerobits_double_convoluted_soft
+; CHECK: mov r0, #0
+; CHECK-NEXT: mov r1, #0
+; CHECK-NEXT: mov r2, #0
+; CHECK-NEXT: mov r3, #0
+  %1 = call { double, double } asm "mov ${0:Q}, #0; mov ${0:R}, #0; mov ${1:Q}, #0; mov ${1:R}, #0", "=r,=r"()
+  %asmresult = extractvalue { double, double } %1, 0
+  %asmresult1 = extractvalue { double, double } %1, 1
+  %add = fadd double %asmresult, %asmresult1
+  ret double %add
+}
+
+; Check support for returning several floats in GPRs with matching float inputs
+; with soft float ABI
+define arm_aapcscc float @flt_gprs_matching_in_op_soft(float %f1, float %f2) #0 {
+; CHECK-LABEL: flt_gprs_matching_in_op_soft
+; CHECK: mov r0, r0
+; CHECK-NEXT: mov r1, r1
+  %1 = call { float, float } asm "mov $0, $2; mov $1, $3", "=&r,=&r,0,1"(float %f1, float %f2)
+  %asmresult1 = extractvalue { float, float } %1, 0
+  %asmresult2 = extractvalue { float, float } %1, 1
+  %add = fadd float %asmresult1, %asmresult2
+  ret float %add
+}
+
+; Check support for returning several double in GPRs with matching double input
+; with soft float ABI
+define arm_aapcscc double @dbl_gprs_matching_in_op_soft(double %d1, double %d2) #0 {
+; CHECK-LABEL: dbl_gprs_matching_in_op_soft
+; CHECK: mov r1, r0
+; CHECK-NEXT: mov r3, r2
+  %1 = call { double, double } asm "mov ${0:R}, ${2:Q}; mov ${1:R}, ${3:Q}", "=&r,=&r,0,1"(double %d1, double %d2)
+  %asmresult1 = extractvalue { double, double } %1, 0
+  %asmresult2 = extractvalue { double, double } %1, 1
+  %add = fadd double %asmresult1, %asmresult2
+  ret double %add
+}
+
+; Check support for returning several float in specific GPRs with matching
+; float input with soft float ABI
+define arm_aapcscc float @flt_gprs_matching_spec_reg_in_op_soft(float %f1, float %f2) #0 {
+; CHECK-LABEL: flt_gprs_matching_spec_reg_in_op_soft
+; CHECK: mov r3, r3
+; CHECK-NEXT: mov r4, r4
+  %1 = call { float, float } asm "mov $0, $2; mov $1, $3", "=&{r3},=&{r4},0,1"(float %f1, float %f2)
+  %asmresult1 = extractvalue { float, float } %1, 0
+  %asmresult2 = extractvalue { float, float } %1, 1
+  %add = fadd float %asmresult1, %asmresult2
+  ret float %add
+}
+
+; Check support for returning several double in specific GPRs with matching
+; double input with soft float ABI
+define arm_aapcscc double @dbl_gprs_matching_spec_reg_in_op_soft(double %d1, double %d2) #0 {
+; CHECK-LABEL: dbl_gprs_matching_spec_reg_in_op_soft
+; CHECK: mov r3, r2
+; CHECK-NEXT: mov r5, r4
+  %1 = call { double, double } asm "mov ${0:R}, ${2:Q}; mov ${1:R}, ${3:Q}", "=&{r2},=&{r4},0,1"(double %d1, double %d2)
+  %asmresult1 = extractvalue { double, double } %1, 0
+  %asmresult2 = extractvalue { double, double } %1, 1
+  %add = fadd double %asmresult1, %asmresult2
+  ret double %add
+}
+
+attributes #0 = { nounwind "target-features"="+d16,+vfp2,+vfp3,-fp-only-sp" "use-soft-float"="true" }
+
+
+; Check support for returning a float in GPR with hard float ABI
+define float @zerobits_float_hard() #1 {
+; CHECK-LABEL: zerobits_float_hard
+; CHECK: mov r0, #0
+; CHECK: vmov s0, r0
+  %1 = tail call float asm "mov ${0}, #0", "=&r"()
+  ret float %1
+}
+
+; Check support for returning a double in GPR with hard float ABI
+define double @zerobits_double_hard() #1 {
+; CHECK-LABEL: zerobits_double_hard
+; CHECK: mov r0, #0
+; CHECK-NEXT: mov r1, #0
+; CHECK: vmov d0, r0, r1
+  %1 = tail call double asm "mov ${0:Q}, #0\0Amov ${0:R}, #0", "=&r"()
+  ret double %1
+}
+
+; Check support for returning a float in GPR with matching float input with
+; hard float ABI
+define float @flt_gpr_matching_in_op_hard(float %f) #1 {
+; CHECK-LABEL: flt_gpr_matching_in_op_hard
+; CHECK: vmov r0, s0
+; CHECK: mov r0, r0
+; CHECK: vmov s0, r0
+  %1 = call float asm "mov $0, $1", "=&r,0"(float %f)
+  ret float %1
+}
+
+; Check support for returning a double in GPR with matching double input with
+; hard float ABI
+define double @dbl_gpr_matching_in_op_hard(double %d) #1 {
+; CHECK-LABEL: dbl_gpr_matching_in_op_hard
+; CHECK: vmov r0, r1, d0
+; CHECK: mov r1, r0
+; CHECK: vmov d0, r0, r1
+  %1 = call double asm "mov ${0:R}, ${1:Q}", "=&r,0"(double %d)
+  ret double %1
+}
+
+; Check support for returning a float in specific GPR with matching float
+; input with hard float ABI
+define float @flt_gpr_matching_spec_reg_in_op_hard(float %f) #1 {
+; CHECK-LABEL: flt_gpr_matching_spec_reg_in_op_hard
+; CHECK: vmov r3, s0
+; CHECK: mov r3, r3
+; CHECK: vmov s0, r3
+  %1 = call float asm "mov $0, $1", "=&{r3},0"(float %f)
+  ret float %1
+}
+
+; Check support for returning a double in specific GPR with matching double
+; input with hard float ABI
+define double @dbl_gpr_matching_spec_reg_in_op_hard(double %d) #1 {
+; CHECK-LABEL: dbl_gpr_matching_spec_reg_in_op_hard
+; CHECK: vmov r2, r3, d0
+; CHECK: mov r3, r2
+; CHECK: vmov d0, r2, r3
+  %1 = call double asm "mov ${0:R}, ${1:Q}", "=&{r2},0"(double %d)
+  ret double %1
+}
+
+; Check support for returning several float in GPR
+define %struct.twofloat @zerobits_float_convoluted_hard() #1 {
+; CHECK-LABEL: zerobits_float_convoluted_hard
+; CHECK: mov r0, #0
+; CHECK-NEXT: mov r1, #0
+; CHECK: vmov s0, r0
+; CHECK-NEXT: vmov s1, r1
+  %1 = call { float, float } asm "mov $0, #0; mov $1, #0", "=r,=r"()
+  %asmresult1 = extractvalue { float, float } %1, 0
+  %asmresult2 = extractvalue { float, float } %1, 1
+  %partialres = insertvalue %struct.twofloat undef, float %asmresult1, 0
+  %res = insertvalue %struct.twofloat %partialres, float %asmresult2, 1
+  ret %struct.twofloat %res
+}
+
+; Check support for returning several double in GPR
+define %struct.twodouble @zerobits_double_convoluted_hard() #1 {
+; CHECK-LABEL: zerobits_double_convoluted_hard
+; CHECK: mov r0, #0
+; CHECK-NEXT: mov r1, #0
+; CHECK-NEXT: mov r2, #0
+; CHECK-NEXT: mov r3, #0
+; CHECK: vmov d0, r0, r1
+; CHECK-NEXT: vmov d1, r2, r3
+  %1 = call { double, double } asm "mov ${0:Q}, #0; mov ${0:R}, #0; mov ${1:Q}, #0; mov ${1:R}, #0", "=r,=r"()
+  %asmresult1 = extractvalue { double, double } %1, 0
+  %asmresult2 = extractvalue { double, double } %1, 1
+  %partialres = insertvalue %struct.twodouble undef, double %asmresult1, 0
+  %res = insertvalue %struct.twodouble %partialres, double %asmresult2, 1
+  ret %struct.twodouble %res
+}
+
+; Check support for returning several floats in GPRs with matching float inputs
+; with hard float ABI
+define %struct.twofloat @flt_gprs_matching_in_op_hard(float %f1, float %f2) #1 {
+; CHECK-LABEL: flt_gprs_matching_in_op_hard
+; CHECK: vmov r0, s0
+; CHECK-NEXT: vmov r1, s1
+; CHECK: mov r0, r0
+; CHECK-NEXT: mov r1, r1
+; CHECK: vmov s0, r0
+; CHECK-NEXT: vmov s1, r1
+  %1 = call { float, float } asm "mov $0, $2; mov $1, $3", "=&r,=&r,0,1"(float %f1, float %f2)
+  %asmresult1 = extractvalue { float, float } %1, 0
+  %asmresult2 = extractvalue { float, float } %1, 1
+  %partialres = insertvalue %struct.twofloat undef, float %asmresult1, 0
+  %res = insertvalue %struct.twofloat %partialres, float %asmresult2, 1
+  ret %struct.twofloat %res
+}
+
+; Check support for returning several double in GPRs with matching double input
+; with hard float ABI
+define %struct.twodouble @dbl_gprs_matching_in_op_hard(double %d1, double %d2) #1 {
+; CHECK-LABEL: dbl_gprs_matching_in_op_hard
+; CHECK: vmov r0, r1, d0
+; CHECK-NEXT: vmov r2, r3, d1
+; CHECK: mov r1, r0
+; CHECK-NEXT: mov r3, r2
+; CHECK: vmov d0, r0, r1
+; CHECK-NEXT: vmov d1, r2, r3
+  %1 = call { double, double } asm "mov ${0:R}, ${2:Q}; mov ${1:R}, ${3:Q}", "=&r,=&r,0,1"(double %d1, double %d2)
+  %asmresult1 = extractvalue { double, double } %1, 0
+  %asmresult2 = extractvalue { double, double } %1, 1
+  %partialres = insertvalue %struct.twodouble undef, double %asmresult1, 0
+  %res = insertvalue %struct.twodouble %partialres, double %asmresult2, 1
+  ret %struct.twodouble %res
+}
+
+; Check support for returning several float in specific GPRs with matching
+; float input with hard float ABI
+define %struct.twofloat @flt_gprs_matching_spec_reg_in_op_hard(float %f1, float %f2) #1 {
+; CHECK-LABEL: flt_gprs_matching_spec_reg_in_op_hard
+; CHECK: vmov r3, s0
+; CHECK-NEXT: vmov r4, s1
+; CHECK: mov r3, r3
+; CHECK-NEXT: mov r4, r4
+; CHECK: vmov s0, r3
+; CHECK-NEXT: vmov s1, r4
+  %1 = call { float, float } asm "mov $0, $2; mov $1, $3", "=&{r3},=&{r4},0,1"(float %f1, float %f2)
+  %asmresult1 = extractvalue { float, float } %1, 0
+  %asmresult2 = extractvalue { float, float } %1, 1
+  %partialres = insertvalue %struct.twofloat undef, float %asmresult1, 0
+  %res = insertvalue %struct.twofloat %partialres, float %asmresult2, 1
+  ret %struct.twofloat %res
+}
+
+; Check support for returning several double in specific GPRs with matching
+; double input with hard float ABI
+define %struct.twodouble @dbl_gprs_matching_spec_reg_in_op_hard(double %d1, double %d2) #1 {
+; CHECK-LABEL: dbl_gprs_matching_spec_reg_in_op_hard
+; CHECK: vmov r2, r3, d0
+; CHECK-NEXT: vmov r4, r5, d1
+; CHECK: mov r3, r2
+; CHECK-NEXT: mov r5, r4
+; CHECK: vmov d0, r2, r3
+; CHECK-NEXT: vmov d1, r4, r5
+  %1 = call { double, double } asm "mov ${0:R}, ${2:Q}; mov ${1:R}, ${3:Q}", "=&{r2},=&{r4},0,1"(double %d1, double %d2)
+  %asmresult1 = extractvalue { double, double } %1, 0
+  %asmresult2 = extractvalue { double, double } %1, 1
+  %partialres = insertvalue %struct.twodouble undef, double %asmresult1, 0
+  %res = insertvalue %struct.twodouble %partialres, double %asmresult2, 1
+  ret %struct.twodouble %res
+}
+
+attributes #1 = { nounwind "target-features"="+d16,+vfp2,+vfp3,-fp-only-sp" "use-soft-float"="false" }