[FPEnv] [PowerPC] Lower ppc_fp128 StrictFP Nodes to libcalls

[ajwock]

Summary

ppc_libcalls

This change allows strictfp nodes with the type ppc_fp128 to be lowered to libcalls in the same manner that regular ppc_fp128 opcodes are. Unlike regular opcodes, whose expansions chain to the entry node and typically leave a dangling chain, these libcalls use the chain input of the original strictfp node and their chain outputs are used by the users of the original node.

A couple of opcodes are not lowered to libcalls- STRICT_FP_ROUND and STRICT_FP_EXTEND. Since the ppc_fp128 type is an f64 pair, truncation or extension from/to the ppc_fp128 type requires disposing one of the doubles or pairing with a new one. The code behind these also parallels normal opcodes, with a little chain finangling.

One caveat to this change is that hardware looping in the IR optimization stage does not know to expect these expansions yet, particularly when checking to see if the intrinsic will clobber the CTR register. I have a fix for this locally to be added later.

[ajwock]

Sorry Hal, I must have accidentally released an incomplete draft yesterday instead of simply deleting it like I meant to.

[hfinkel]

In D64662#1583456, @ajwock wrote:

Sorry Hal, I must have accidentally released an incomplete draft yesterday instead of simply deleting it like I meant to.

Can you please upload a patch with full context?

[ajwock]

Full context.

[ajwock]

Ping.

[ajwock]

ping

[ajwock]

Ping.

[kbarton]

include/llvm/CodeGen/TargetLowering.h

2969 ↗

(On Diff #209923)

Is there a specific reason you used the _ in the function name?
I think with camel case this can be makeLibCallChained, which is still easy to read and follows the coding guidelines.

lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp

1140

Similar comment here with the choice of _STRICT_ vs Strict_ (i.e., ExpandFloatResStrict_FMINNUM(...)).
I think the use of Strict_ follows the coding guidelines, while _STRICT_ deviates from it.

1242

Was this formatted with clang-format?

1931

Can you put parenthesis around this to make it clear the order of evaluation between the && and ||?

lib/CodeGen/SelectionDAG/LegalizeTypes.cpp

1029 ↗

(On Diff #209923)

Can you please make this into a doxygen comment .

1031 ↗

(On Diff #209923)

Same comment about the use of underscores. I think this should be LibCallifyStrictFP.

lib/CodeGen/SelectionDAG/TargetLowering.cpp

160 ↗

(On Diff #209923)

This should be makeLibCallChained

test/CodeGen/PowerPC/ppcf128-constrained-fp-instrinsics.ll

8

Were these checks automatically generated by the python script?
It seems unusual that they are placed in the middle of the parameter list for the test case. I thought they usually got put immediately before the function.

13

Others may disagree with me, but I would recommend getting rid of the checks for cfi here.
I don't think the location of the cfi has any bearing on the correctness here. I also think the compiler should be free to move them with respect to the std and stdu instructions. If it does so, this test case will fail.

[kbarton]

Sorry, I clicked submit before adding general comments.
Thanks for working on this. I think this looks OK, just needs some minor cleaning wrt coding guidelines and the test cases.
I'd like other people weigh in though, as it's been a while since I looked at SD Chains.

[kbarton]

test/CodeGen/PowerPC/ppcf128-constrained-fp-instrinsics.ll

5

I also just realized we don't have any 32-bit build bots anymore.
Can you please change this to powerpc64-linux-gnu?

[ajwock]

Variety of adjustments, see comment replies (still incoming if I just submitted this)

[ajwock]

include/llvm/CodeGen/TargetLowering.h

2969 ↗

(On Diff #209923)

Fixed. This was matching how I added strict to other names, but it is not necessary.

lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp

1140

So, each of these functions capitalized names corresponds to an instruction opcode. In order to stay consistent with the rest of these functions, I have simply named these functions after their opcode rather than arbitrarily prepending _STRICT.

I believe that if I did not name these functions as such, it would either be inconsistent with existent code or require significant refactoring.

1242

Fixed? Reformatted with clang-format.

1931

Fixed, and switched order of evaluation to short circuit on the common case.

lib/CodeGen/SelectionDAG/LegalizeTypes.cpp

1031 ↗

(On Diff #209923)

Fixed.

lib/CodeGen/SelectionDAG/TargetLowering.cpp

160 ↗

(On Diff #209923)

Fixed.

test/CodeGen/PowerPC/ppcf128-constrained-fp-instrinsics.ll

5

Fixed.

8

Semi-fixed: They were- though as far as I've seen these checks are typically placed right below the function header. I can change this if needed.

13

Alright- loosened -NEXT restrictions around std/u instructions.

[jsji]

lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp

1574

Add comments about why this is special here, not lowering to calls. I know you have description in summary,
but it would be great to add comment here, so that readers don't have to look into commit message or patches.

1909

Any reason that we are expanding STRICT_FP_ROUND in ExpandFloatOperand, while all the others, especially STRICT_FP_EXTEND in ExpandFloatResult?

2009

Add comments about why this is special here, not lowering to calls too.

lib/CodeGen/SelectionDAG/LegalizeTypes.cpp

1039 ↗

(On Diff #209923)

Can we just reuse ExpandChainLibCall below?

lib/CodeGen/SelectionDAG/TargetLowering.cpp

162 ↗

(On Diff #209923)

If we are going to keep this API, should use MakeLibCallOptions introduced in https://reviews.llvm.org/D65795 instead.

181 ↗

(On Diff #209923)

Most of these code are command and copied from makeLibCall, can we maintain one copy only?

test/CodeGen/PowerPC/ppcf128-constrained-fp-instrinsics.ll

1975

These tests are all simple test with one call. Can we add one or more complex tests? eg: one call use the other calls res? so that we can test the input chain?

[ajwock]

Thanks to jsji for making me realize that a lot of the functionality I implemented in this patch already existed. I converted the ExpandFloatRes/Op_STRICT_* functions to use ExpandChainLibCall and eliminated the LibCallifyStrictFP and MakeLibCallChained functions from the previous diffs.

I also added a test that ensures that the serial properties of strictFP functions are maintained.

[jsji]

test/CodeGen/PowerPC/ppcf128-constrained-fp-instrinsics.ll

13

You should be able to avoid checking those by adding nounwind to all these functions, you can see examples in vector-constrained-fp-intrinsics.ll in https://reviews.llvm.org/D67016.
This way, you can still use the script to generate the check automatically.

[ajwock]

Still have some test changes that I have to make- slight diff change incoming soon

lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp

1574

I didn't find it necessary to add such comments- libcallification is one of the possible ways one might expand a floating point result, but it isn't something that is implied.

Another reason I didn't do this is that the author of the analog ExpandFloatRes_FP_EXTEND (not strict and right above this) didn't comment on this, and when I came across this code in my problem solving, I could reasonably intuit why FP_EXTEND wouldn't be a libcall.

I could add documentation that explains this issue with respect to both this function and the above function, but wouldn't that be more appropriate for a ticket specifically about existing documentation?

lib/CodeGen/SelectionDAG/LegalizeTypes.cpp

1039 ↗

(On Diff #209923)

Good catch. I wonder why this function was named so differently- that alone kept me from realizing it had anything to do with LibCallify.

Fixed, and removed LibCallifyStrictFP + makeLibCallChained.

[ajwock]

Updated tests to have nounwind attribute so that we don't have to worry about https://reviews.llvm.org/D67016 non-relevant .cfi.

[ajwock]

lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp

1909

I put STRICT_FP_ROUND here for a rather simple-minded reason- FP_ROUND (non strict) is dealt with here as well.

I couldn't find any documentation or commits on the reason that ExpandFloatOperand_FP_ROUND was here, but this particular addition was required to get clang with the https://reviews.llvm.org/D43142 forceconstrainedfp pass to successfully compile and pass the GNU Scientific Library's tests that involved the long long data type on a PowerPC server.

[ajwock]

lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp

1909

Edit: by long long I mean long-double.

[jsji]

https://reviews.llvm.org/rL370228 just landed, so I think you should include fptrunc and fpext intrinsics as well.

lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp

1909

I think this is because STRICT_FP_ROUND is used for fptrunc to f64,f32, the result type (f64,f32 is already legal here. However, we need to *chain* in the expansion of operands, instead of using default non-strict expansion here. -- I think that is why this expansion is required to get the tests passed.

1911

Should also ExpandFloatOp_STRICT here for STRICT_FP_TO_SINT/STRICT_FP_TO_UINT.

test/CodeGen/PowerPC/ppcf128-constrained-fp-instrinsics.ll

4

Most of the P8 LE and P9 LE code are actually common, I believe you can simplify the file using check-prefixes.

; RUN: llc -O3 -mtriple=powerpc64le-linux-gnu < %s | FileCheck --check-prefixes=LECOMM,PC64LE %s
; RUN: llc -O3 -mtriple=powerpc64le-linux-gnu -mcpu=pwr9 < %s | FileCheck --check-prefixes=LECOMM,PC64LE9 %s

then rerun the script should reduce the file.

[jsji]

lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp

1574

Fair enough, maybe a NFC patch for documenting these in the future. Thanks.

[ajwock]

Added libcall lowering for fptosi and fptoui constrained intrinsics. They take a similar lowering route to STRICT_FP_ROUND.

[jsji]

LGTM. May need @kbarton to have another look, as he should be in the "Must Review" for this.

lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp

1288

Unnecessary extra newline.

1931

Is this formatted by clang-format? Extra space at the end of line here.

[ajwock]

Ensured that all changes to LLVM source are clang-formatted.

[ajwock]

ping.

[ajwock]

ping

[kbarton]

LGTM.

lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp

1140

Yes, I see now.
I think this is OK. And it's consistent with the way it's done in SystemZ too.

1242

Yes, this looks better now (although I honestly don't remeber what it looked like before :( ).

[craig.topper]

I think this needs to be rebased. ExpandChainLibCall no longer exists. It was merged into TLI.makeLibcall which now has an optional Chain argument.

[ajwock]

Mooted by D70867.

lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp

void DAGTypeLegalizer::ExpandFloatResult(SDNode *N, unsigned ResNo) {
   switch (N->getOpcode()) {
   default:
 #ifndef NDEBUG
     dbgs() << "ExpandFloatResult #" << ResNo << ": ";
     N->dump(&DAG); dbgs() << "\n";
 #endif
     llvm_unreachable("Do not know how to expand the result of this operator!");
 
-  case ISD::UNDEF:        SplitRes_UNDEF(N, Lo, Hi); break;
-  case ISD::SELECT:       SplitRes_SELECT(N, Lo, Hi); break;
-  case ISD::SELECT_CC:    SplitRes_SELECT_CC(N, Lo, Hi); break;
+  case ISD::UNDEF:              SplitRes_UNDEF(N, Lo, Hi); break;
+  case ISD::SELECT:             SplitRes_SELECT(N, Lo, Hi); break;
+  case ISD::SELECT_CC:          SplitRes_SELECT_CC(N, Lo, Hi); break;
 
   case ISD::MERGE_VALUES:       ExpandRes_MERGE_VALUES(N, ResNo, Lo, Hi); break;
   case ISD::BITCAST:            ExpandRes_BITCAST(N, Lo, Hi); break;
   case ISD::BUILD_PAIR:         ExpandRes_BUILD_PAIR(N, Lo, Hi); break;
   case ISD::EXTRACT_ELEMENT:    ExpandRes_EXTRACT_ELEMENT(N, Lo, Hi); break;
   case ISD::EXTRACT_VECTOR_ELT: ExpandRes_EXTRACT_VECTOR_ELT(N, Lo, Hi); break;
   case ISD::VAARG:              ExpandRes_VAARG(N, Lo, Hi); break;
 
-  case ISD::ConstantFP: ExpandFloatRes_ConstantFP(N, Lo, Hi); break;
-  case ISD::FABS:       ExpandFloatRes_FABS(N, Lo, Hi); break;
-  case ISD::FMINNUM:    ExpandFloatRes_FMINNUM(N, Lo, Hi); break;
-  case ISD::FMAXNUM:    ExpandFloatRes_FMAXNUM(N, Lo, Hi); break;
-  case ISD::FADD:       ExpandFloatRes_FADD(N, Lo, Hi); break;
-  case ISD::FCEIL:      ExpandFloatRes_FCEIL(N, Lo, Hi); break;
-  case ISD::FCOPYSIGN:  ExpandFloatRes_FCOPYSIGN(N, Lo, Hi); break;
-  case ISD::FCOS:       ExpandFloatRes_FCOS(N, Lo, Hi); break;
-  case ISD::FDIV:       ExpandFloatRes_FDIV(N, Lo, Hi); break;
-  case ISD::FEXP:       ExpandFloatRes_FEXP(N, Lo, Hi); break;
-  case ISD::FEXP2:      ExpandFloatRes_FEXP2(N, Lo, Hi); break;
-  case ISD::FFLOOR:     ExpandFloatRes_FFLOOR(N, Lo, Hi); break;
-  case ISD::FLOG:       ExpandFloatRes_FLOG(N, Lo, Hi); break;
-  case ISD::FLOG2:      ExpandFloatRes_FLOG2(N, Lo, Hi); break;
-  case ISD::FLOG10:     ExpandFloatRes_FLOG10(N, Lo, Hi); break;
-  case ISD::FMA:        ExpandFloatRes_FMA(N, Lo, Hi); break;
-  case ISD::FMUL:       ExpandFloatRes_FMUL(N, Lo, Hi); break;
-  case ISD::FNEARBYINT: ExpandFloatRes_FNEARBYINT(N, Lo, Hi); break;
-  case ISD::FNEG:       ExpandFloatRes_FNEG(N, Lo, Hi); break;
-  case ISD::FP_EXTEND:  ExpandFloatRes_FP_EXTEND(N, Lo, Hi); break;
-  case ISD::FPOW:       ExpandFloatRes_FPOW(N, Lo, Hi); break;
-  case ISD::FPOWI:      ExpandFloatRes_FPOWI(N, Lo, Hi); break;
-  case ISD::FRINT:      ExpandFloatRes_FRINT(N, Lo, Hi); break;
-  case ISD::FROUND:     ExpandFloatRes_FROUND(N, Lo, Hi); break;
-  case ISD::FSIN:       ExpandFloatRes_FSIN(N, Lo, Hi); break;
-  case ISD::FSQRT:      ExpandFloatRes_FSQRT(N, Lo, Hi); break;
-  case ISD::FSUB:       ExpandFloatRes_FSUB(N, Lo, Hi); break;
-  case ISD::FTRUNC:     ExpandFloatRes_FTRUNC(N, Lo, Hi); break;
-  case ISD::LOAD:       ExpandFloatRes_LOAD(N, Lo, Hi); break;
+  case ISD::ConstantFP:         ExpandFloatRes_ConstantFP(N, Lo, Hi); break;
+  case ISD::FABS:               ExpandFloatRes_FABS(N, Lo, Hi); break;
+  case ISD::FMINNUM:            ExpandFloatRes_FMINNUM(N, Lo, Hi); break;
+  case ISD::STRICT_FMINNUM:     ExpandFloatRes_STRICT_FMINNUM(N, Lo, Hi); break;
+  case ISD::FMAXNUM:            ExpandFloatRes_FMAXNUM(N, Lo, Hi); break;
+  case ISD::STRICT_FMAXNUM:     ExpandFloatRes_STRICT_FMAXNUM(N, Lo, Hi); break;
+  case ISD::FADD:               ExpandFloatRes_FADD(N, Lo, Hi); break;
+  case ISD::STRICT_FADD:        ExpandFloatRes_STRICT_FADD(N, Lo, Hi); break;
+  case ISD::FCEIL:              ExpandFloatRes_FCEIL(N, Lo, Hi); break;
+  case ISD::STRICT_FCEIL:       ExpandFloatRes_STRICT_FCEIL(N, Lo, Hi); break;
+  case ISD::FCOPYSIGN:          ExpandFloatRes_FCOPYSIGN(N, Lo, Hi); break;
+  case ISD::FCOS:               ExpandFloatRes_FCOS(N, Lo, Hi); break;
+  case ISD::STRICT_FCOS:        ExpandFloatRes_STRICT_FCOS(N, Lo, Hi); break;
+  case ISD::FDIV:               ExpandFloatRes_FDIV(N, Lo, Hi); break;
+  case ISD::STRICT_FDIV:        ExpandFloatRes_STRICT_FDIV(N, Lo, Hi); break;
+  case ISD::FEXP:               ExpandFloatRes_FEXP(N, Lo, Hi); break;
+  case ISD::STRICT_FEXP:        ExpandFloatRes_STRICT_FEXP(N, Lo, Hi); break;
+  case ISD::FEXP2:              ExpandFloatRes_FEXP2(N, Lo, Hi); break;
+  case ISD::STRICT_FEXP2:       ExpandFloatRes_STRICT_FEXP2(N, Lo, Hi); break;
+  case ISD::FFLOOR:             ExpandFloatRes_FFLOOR(N, Lo, Hi); break;
+  case ISD::STRICT_FFLOOR:      ExpandFloatRes_STRICT_FFLOOR(N, Lo, Hi); break;
+  case ISD::FLOG:               ExpandFloatRes_FLOG(N, Lo, Hi); break;
+  case ISD::STRICT_FLOG:        ExpandFloatRes_STRICT_FLOG(N, Lo, Hi); break;
+  case ISD::FLOG2:              ExpandFloatRes_FLOG2(N, Lo, Hi); break;
+  case ISD::STRICT_FLOG2:       ExpandFloatRes_STRICT_FLOG2(N, Lo, Hi); break;
+  case ISD::FLOG10:             ExpandFloatRes_FLOG10(N, Lo, Hi); break;
+  case ISD::STRICT_FLOG10:      ExpandFloatRes_STRICT_FLOG10(N, Lo, Hi); break;
+  case ISD::FMA:                ExpandFloatRes_FMA(N, Lo, Hi); break;
+  case ISD::STRICT_FMA:         ExpandFloatRes_STRICT_FMA(N, Lo, Hi); break;
+  case ISD::FMUL:               ExpandFloatRes_FMUL(N, Lo, Hi); break;
+  case ISD::STRICT_FMUL:        ExpandFloatRes_STRICT_FMUL(N, Lo, Hi); break;
+  case ISD::FNEARBYINT:         ExpandFloatRes_FNEARBYINT(N, Lo, Hi); break;
+  case ISD::STRICT_FNEARBYINT:  ExpandFloatRes_STRICT_FNEARBYINT(N, Lo, Hi); break;
+  case ISD::FNEG:               ExpandFloatRes_FNEG(N, Lo, Hi); break;
+  case ISD::FP_EXTEND:          ExpandFloatRes_FP_EXTEND(N, Lo, Hi); break;
+  case ISD::STRICT_FP_EXTEND:   ExpandFloatRes_STRICT_FP_EXTEND(N, Lo, Hi); break;
+  case ISD::FPOW:               ExpandFloatRes_FPOW(N, Lo, Hi); break;
+  case ISD::STRICT_FPOW:        ExpandFloatRes_STRICT_FPOW(N, Lo, Hi); break;
+  case ISD::FPOWI:              ExpandFloatRes_FPOWI(N, Lo, Hi); break;
+  case ISD::STRICT_FPOWI:       ExpandFloatRes_STRICT_FPOWI(N, Lo, Hi); break;
+  case ISD::FRINT:              ExpandFloatRes_FRINT(N, Lo, Hi); break;
+  case ISD::STRICT_FRINT:       ExpandFloatRes_STRICT_FRINT(N, Lo, Hi); break;
+  case ISD::FROUND:             ExpandFloatRes_FROUND(N, Lo, Hi); break;
+  case ISD::STRICT_FROUND:      ExpandFloatRes_STRICT_FROUND(N, Lo, Hi); break;
+  case ISD::FSIN:               ExpandFloatRes_FSIN(N, Lo, Hi); break;
+  case ISD::STRICT_FSIN:        ExpandFloatRes_STRICT_FSIN(N, Lo, Hi); break;
+  case ISD::FSQRT:              ExpandFloatRes_FSQRT(N, Lo, Hi); break;
+  case ISD::STRICT_FSQRT:       ExpandFloatRes_STRICT_FSQRT(N, Lo, Hi); break;
+  case ISD::FSUB:               ExpandFloatRes_FSUB(N, Lo, Hi); break;
+  case ISD::STRICT_FSUB:        ExpandFloatRes_STRICT_FSUB(N, Lo, Hi); break;
+  case ISD::FTRUNC:             ExpandFloatRes_FTRUNC(N, Lo, Hi); break;
+  case ISD::STRICT_FTRUNC:      ExpandFloatRes_STRICT_FTRUNC(N, Lo, Hi); break;
+  case ISD::LOAD:               ExpandFloatRes_LOAD(N, Lo, Hi); break;
   case ISD::SINT_TO_FP:
-  case ISD::UINT_TO_FP: ExpandFloatRes_XINT_TO_FP(N, Lo, Hi); break;
-  case ISD::FREM:       ExpandFloatRes_FREM(N, Lo, Hi); break;
+  case ISD::UINT_TO_FP:         ExpandFloatRes_XINT_TO_FP(N, Lo, Hi); break;
+  case ISD::FREM:               ExpandFloatRes_FREM(N, Lo, Hi); break;
+  case ISD::STRICT_FREM:        ExpandFloatRes_STRICT_FREM(N, Lo, Hi); break;
   }
 
   // If Lo/Hi is null, the sub-method took care of registering results etc.
   if (Lo.getNode())
     SetExpandedFloat(SDValue(N, ResNo), Lo, Hi);
 }
 
 void DAGTypeLegalizer::ExpandFloatRes_ConstantFP(SDNode *N, SDValue &Lo,
void DAGTypeLegalizer::ExpandFloatRes_FMINNUM(SDNode *N, SDValue &Lo,
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::FMIN_F32, RTLIB::FMIN_F64,
                                          RTLIB::FMIN_F80, RTLIB::FMIN_F128,
                                          RTLIB::FMIN_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FMINNUM(SDNode *N, SDValue &Lo,
+                                                     SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::FMIN_F32, RTLIB::FMIN_F64,
+                   RTLIB::FMIN_F80, RTLIB::FMIN_F128, RTLIB::FMIN_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FMAXNUM(SDNode *N, SDValue &Lo,
                                               SDValue &Hi) {
-  SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
-                                         RTLIB::FMAX_F32, RTLIB::FMAX_F64,
-                                         RTLIB::FMAX_F80, RTLIB::FMAX_F128,
-                                         RTLIB::FMAX_PPCF128),
+  SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0), RTLIB::FMAX_F32,
+                                         RTLIB::FMAX_F64, RTLIB::FMAX_F80,
+                                         RTLIB::FMAX_F128, RTLIB::FMAX_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FMAXNUM(SDNode *N, SDValue &Lo,
+                                                     SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::FMAX_F32, RTLIB::FMAX_F64,
+                   RTLIB::FMAX_F80, RTLIB::FMAX_F128, RTLIB::FMAX_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FADD(SDNode *N, SDValue &Lo,
                                            SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::ADD_F32, RTLIB::ADD_F64,
                                          RTLIB::ADD_F80, RTLIB::ADD_F128,
                                          RTLIB::ADD_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FADD(SDNode *N, SDValue &Lo,
+                                                  SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::ADD_F32, RTLIB::ADD_F64,
+                   RTLIB::ADD_F80, RTLIB::ADD_F128, RTLIB::ADD_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+                                               
+
 void DAGTypeLegalizer::ExpandFloatRes_FCEIL(SDNode *N,
                                             SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::CEIL_F32, RTLIB::CEIL_F64,
                                          RTLIB::CEIL_F80, RTLIB::CEIL_F128,
                                          RTLIB::CEIL_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FCEIL(SDNode *N, SDValue &Lo,
+                                                   SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::CEIL_F32, RTLIB::CEIL_F64,
+                   RTLIB::CEIL_F80, RTLIB::CEIL_F128, RTLIB::CEIL_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FCOPYSIGN(SDNode *N,
                                                 SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::COPYSIGN_F32,
                                          RTLIB::COPYSIGN_F64,
                                          RTLIB::COPYSIGN_F80,
                                          RTLIB::COPYSIGN_F128,
                                          RTLIB::COPYSIGN_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
 void DAGTypeLegalizer::ExpandFloatRes_FCOS(SDNode *N,
                                            SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::COS_F32, RTLIB::COS_F64,
                                          RTLIB::COS_F80, RTLIB::COS_F128,
                                          RTLIB::COS_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FCOS(SDNode *N, SDValue &Lo,
+                                                  SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::COS_F32, RTLIB::COS_F64,
+                   RTLIB::COS_F80, RTLIB::COS_F128, RTLIB::COS_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FDIV(SDNode *N, SDValue &Lo,
                                            SDValue &Hi) {
   SDValue Ops[2] = { N->getOperand(0), N->getOperand(1) };
   SDValue Call = TLI.makeLibCall(DAG, GetFPLibCall(N->getValueType(0),
                                                    RTLIB::DIV_F32,
                                                    RTLIB::DIV_F64,
                                                    RTLIB::DIV_F80,
                                                    RTLIB::DIV_F128,
                                                    RTLIB::DIV_PPCF128),
                                  N->getValueType(0), Ops, false,
                                  SDLoc(N)).first;
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FDIV(SDNode *N, SDValue &Lo,
+                                                  SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::DIV_F32, RTLIB::DIV_F64,
+                   RTLIB::DIV_F80, RTLIB::DIV_F128, RTLIB::DIV_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FEXP(SDNode *N,
                                            SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::EXP_F32, RTLIB::EXP_F64,
                                          RTLIB::EXP_F80, RTLIB::EXP_F128,
                                          RTLIB::EXP_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FEXP(SDNode *N, SDValue &Lo,
+                                                  SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::EXP_F32, RTLIB::EXP_F64,
+                   RTLIB::EXP_F80, RTLIB::EXP_F128, RTLIB::EXP_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FEXP2(SDNode *N,
                                             SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::EXP2_F32, RTLIB::EXP2_F64,
                                          RTLIB::EXP2_F80, RTLIB::EXP2_F128,
                                          RTLIB::EXP2_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FEXP2(SDNode *N, SDValue &Lo,
+                                                   SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::EXP2_F32, RTLIB::EXP2_F64,
+                   RTLIB::EXP2_F80, RTLIB::EXP2_F128, RTLIB::EXP2_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FFLOOR(SDNode *N,
                                              SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::FLOOR_F32, RTLIB::FLOOR_F64,
                                          RTLIB::FLOOR_F80, RTLIB::FLOOR_F128,
                                          RTLIB::FLOOR_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FFLOOR(SDNode *N, SDValue &Lo,
+                                                    SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::FLOOR_F32, RTLIB::FLOOR_F64,
+                   RTLIB::FLOOR_F80, RTLIB::FLOOR_F128, RTLIB::FLOOR_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FLOG(SDNode *N,
                                            SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::LOG_F32, RTLIB::LOG_F64,
                                          RTLIB::LOG_F80, RTLIB::LOG_F128,
                                          RTLIB::LOG_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FLOG(SDNode *N, SDValue &Lo,
+                                                  SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::LOG_F32, RTLIB::LOG_F64,
+                   RTLIB::LOG_F80, RTLIB::LOG_F128, RTLIB::LOG_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FLOG2(SDNode *N,
                                             SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::LOG2_F32, RTLIB::LOG2_F64,
                                          RTLIB::LOG2_F80, RTLIB::LOG2_F128,
                                          RTLIB::LOG2_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FLOG2(SDNode *N, SDValue &Lo,
+                                                   SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::LOG2_F32, RTLIB::LOG2_F64,
+                   RTLIB::LOG2_F80, RTLIB::LOG2_F128, RTLIB::LOG2_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FLOG10(SDNode *N,
                                              SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::LOG10_F32, RTLIB::LOG10_F64,
                                          RTLIB::LOG10_F80, RTLIB::LOG10_F128,
                                          RTLIB::LOG10_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FLOG10(SDNode *N, SDValue &Lo,
+                                                    SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::LOG10_F32, RTLIB::LOG10_F64,
+                   RTLIB::LOG10_F80, RTLIB::LOG10_F128, RTLIB::LOG10_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FMA(SDNode *N, SDValue &Lo,
                                           SDValue &Hi) {
   SDValue Ops[3] = { N->getOperand(0), N->getOperand(1), N->getOperand(2) };
   SDValue Call = TLI.makeLibCall(DAG, GetFPLibCall(N->getValueType(0),
                                                    RTLIB::FMA_F32,
                                                    RTLIB::FMA_F64,
                                                    RTLIB::FMA_F80,
                                                    RTLIB::FMA_F128,
                                                    RTLIB::FMA_PPCF128),
                                  N->getValueType(0), Ops, false,
                                  SDLoc(N)).first;
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FMA(SDNode *N, SDValue &Lo,
+                                                 SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::FMA_F32, RTLIB::FMA_F64,
+                   RTLIB::FMA_F80, RTLIB::FMA_F128, RTLIB::FMA_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FMUL(SDNode *N, SDValue &Lo,
                                            SDValue &Hi) {
   SDValue Ops[2] = { N->getOperand(0), N->getOperand(1) };
   SDValue Call = TLI.makeLibCall(DAG, GetFPLibCall(N->getValueType(0),
                                                    RTLIB::MUL_F32,
                                                    RTLIB::MUL_F64,
                                                    RTLIB::MUL_F80,
                                                    RTLIB::MUL_F128,
                                                    RTLIB::MUL_PPCF128),
                                  N->getValueType(0), Ops, false,
                                  SDLoc(N)).first;
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FMUL(SDNode *N, SDValue &Lo,
+                                                  SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::MUL_F32, RTLIB::MUL_F64,
+                   RTLIB::MUL_F80, RTLIB::MUL_F128, RTLIB::MUL_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FNEARBYINT(SDNode *N,
                                                  SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::NEARBYINT_F32,
                                          RTLIB::NEARBYINT_F64,
                                          RTLIB::NEARBYINT_F80,
                                          RTLIB::NEARBYINT_F128,
                                          RTLIB::NEARBYINT_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FNEARBYINT(SDNode *N, SDValue &Lo,
+                                                        SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::NEARBYINT_F32,
+                   RTLIB::NEARBYINT_F64, RTLIB::NEARBYINT_F80,
+                   RTLIB::NEARBYINT_F128, RTLIB::NEARBYINT_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FNEG(SDNode *N, SDValue &Lo,
                                            SDValue &Hi) {
   SDLoc dl(N);
   GetExpandedFloat(N->getOperand(0), Lo, Hi);
   Lo = DAG.getNode(ISD::FNEG, dl, Lo.getValueType(), Lo);
   Hi = DAG.getNode(ISD::FNEG, dl, Hi.getValueType(), Hi);
 }
 
 void DAGTypeLegalizer::ExpandFloatRes_FP_EXTEND(SDNode *N, SDValue &Lo,
                                                 SDValue &Hi) {
   EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
   SDLoc dl(N);
   Hi = DAG.getNode(ISD::FP_EXTEND, dl, NVT, N->getOperand(0));
   Lo = DAG.getConstantFP(APFloat(DAG.EVTToAPFloatSemantics(NVT),
                                  APInt(NVT.getSizeInBits(), 0)), dl, NVT);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FP_EXTEND(SDNode *N, SDValue &Lo,
+                                                       SDValue &Hi) {
+  EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
+  SDLoc dl(N);
+  Hi = DAG.getNode(ISD::STRICT_FP_EXTEND, dl, {NVT, MVT::Other},
+                   {N->getOperand(0), N->getOperand(1)});
+  Lo = DAG.getConstantFP(
+      APFloat(DAG.EVTToAPFloatSemantics(NVT), APInt(NVT.getSizeInBits(), 0)),
+      dl, NVT);
+  ReplaceValueWith(SDValue(N, 1), SDValue(Hi.getNode(), 1));
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FPOW(SDNode *N,
                                            SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::POW_F32, RTLIB::POW_F64,
                                          RTLIB::POW_F80, RTLIB::POW_F128,
                                          RTLIB::POW_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FPOW(SDNode *N, SDValue &Lo,
+                                                  SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::POW_F32, RTLIB::POW_F64,
+                   RTLIB::POW_F80, RTLIB::POW_F128, RTLIB::POW_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FPOWI(SDNode *N,
                                             SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::POWI_F32, RTLIB::POWI_F64,
                                          RTLIB::POWI_F80, RTLIB::POWI_F128,
                                          RTLIB::POWI_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FPOWI(SDNode *N, SDValue &Lo,
+                                                   SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::POWI_F32, RTLIB::POWI_F64,
+                   RTLIB::POWI_F80, RTLIB::POWI_F128, RTLIB::POWI_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FREM(SDNode *N,
                                            SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::REM_F32, RTLIB::REM_F64,
                                          RTLIB::REM_F80, RTLIB::REM_F128,
                                          RTLIB::REM_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FREM(SDNode *N, SDValue &Lo,
+                                                  SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::REM_F32, RTLIB::REM_F64,
+                   RTLIB::REM_F80, RTLIB::REM_F128, RTLIB::REM_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FRINT(SDNode *N,
                                             SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::RINT_F32, RTLIB::RINT_F64,
                                          RTLIB::RINT_F80, RTLIB::RINT_F128,
                                          RTLIB::RINT_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FRINT(SDNode *N, SDValue &Lo,
+                                                   SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::RINT_F32, RTLIB::RINT_F64,
+                   RTLIB::RINT_F80, RTLIB::RINT_F128, RTLIB::RINT_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FROUND(SDNode *N,
                                              SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::ROUND_F32,
                                          RTLIB::ROUND_F64,
                                          RTLIB::ROUND_F80,
                                          RTLIB::ROUND_F128,
                                          RTLIB::ROUND_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FROUND(SDNode *N, SDValue &Lo,
+                                                    SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::ROUND_F32, RTLIB::ROUND_F64,
+                   RTLIB::ROUND_F80, RTLIB::ROUND_F128, RTLIB::ROUND_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FSIN(SDNode *N,
                                            SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::SIN_F32, RTLIB::SIN_F64,
                                          RTLIB::SIN_F80, RTLIB::SIN_F128,
                                          RTLIB::SIN_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FSIN(SDNode *N, SDValue &Lo,
+                                                  SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::SIN_F32, RTLIB::SIN_F64,
+                   RTLIB::SIN_F80, RTLIB::SIN_F128, RTLIB::SIN_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FSQRT(SDNode *N,
                                             SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::SQRT_F32, RTLIB::SQRT_F64,
                                          RTLIB::SQRT_F80, RTLIB::SQRT_F128,
                                          RTLIB::SQRT_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FSQRT(SDNode *N, SDValue &Lo,
+                                                   SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::SQRT_F32, RTLIB::SQRT_F64,
+                   RTLIB::SQRT_F80, RTLIB::SQRT_F128, RTLIB::SQRT_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FSUB(SDNode *N, SDValue &Lo,
                                            SDValue &Hi) {
   SDValue Ops[2] = { N->getOperand(0), N->getOperand(1) };
   SDValue Call = TLI.makeLibCall(DAG, GetFPLibCall(N->getValueType(0),
                                                    RTLIB::SUB_F32,
                                                    RTLIB::SUB_F64,
                                                    RTLIB::SUB_F80,
                                                    RTLIB::SUB_F128,
                                                    RTLIB::SUB_PPCF128),
                                  N->getValueType(0), Ops, false,
                                  SDLoc(N)).first;
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FSUB(SDNode *N, SDValue &Lo,
+                                                  SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::SUB_F32, RTLIB::SUB_F64,
+                   RTLIB::SUB_F80, RTLIB::SUB_F128, RTLIB::SUB_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_FTRUNC(SDNode *N,
                                              SDValue &Lo, SDValue &Hi) {
   SDValue Call = LibCallify(GetFPLibCall(N->getValueType(0),
                                          RTLIB::TRUNC_F32, RTLIB::TRUNC_F64,
                                          RTLIB::TRUNC_F80, RTLIB::TRUNC_F128,
                                          RTLIB::TRUNC_PPCF128),
                             N, false);
   GetPairElements(Call, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandFloatRes_STRICT_FTRUNC(SDNode *N, SDValue &Lo,
+                                                    SDValue &Hi) {
+  std::pair<SDValue, SDValue> Call = ExpandChainLibCall(
+      GetFPLibCall(N->getValueType(0), RTLIB::TRUNC_F32, RTLIB::TRUNC_F64,
+                   RTLIB::TRUNC_F80, RTLIB::TRUNC_F128, RTLIB::TRUNC_PPCF128),
+      N, false);
+  DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Call.second);
+  GetPairElements(Call.first, Lo, Hi);
+}
+
 void DAGTypeLegalizer::ExpandFloatRes_LOAD(SDNode *N, SDValue &Lo,
                                            SDValue &Hi) {
   if (ISD::isNormalLoad(N)) {
     ExpandRes_NormalLoad(N, Lo, Hi);
     return;
   }
 
   assert(ISD::isUNINDEXEDLoad(N) && "Indexed load during type legalization!");
#ifndef NDEBUG
     N->dump(&DAG); dbgs() << "\n";
 #endif
     llvm_unreachable("Do not know how to expand this operator's operand!");
 
   case ISD::BITCAST:         Res = ExpandOp_BITCAST(N); break;
   case ISD::BUILD_VECTOR:    Res = ExpandOp_BUILD_VECTOR(N); break;
   case ISD::EXTRACT_ELEMENT: Res = ExpandOp_EXTRACT_ELEMENT(N); break;
 
-  case ISD::BR_CC:      Res = ExpandFloatOp_BR_CC(N); break;
-  case ISD::FCOPYSIGN:  Res = ExpandFloatOp_FCOPYSIGN(N); break;
-  case ISD::FP_ROUND:   Res = ExpandFloatOp_FP_ROUND(N); break;
-  case ISD::FP_TO_SINT: Res = ExpandFloatOp_FP_TO_SINT(N); break;
-  case ISD::FP_TO_UINT: Res = ExpandFloatOp_FP_TO_UINT(N); break;
-  case ISD::LROUND:     Res = ExpandFloatOp_LROUND(N); break;
-  case ISD::LLROUND:    Res = ExpandFloatOp_LLROUND(N); break;
-  case ISD::LRINT:      Res = ExpandFloatOp_LRINT(N); break;
-  case ISD::LLRINT:     Res = ExpandFloatOp_LLRINT(N); break;
-  case ISD::SELECT_CC:  Res = ExpandFloatOp_SELECT_CC(N); break;
-  case ISD::SETCC:      Res = ExpandFloatOp_SETCC(N); break;
-  case ISD::STORE:      Res = ExpandFloatOp_STORE(cast<StoreSDNode>(N),
+  case ISD::BR_CC:            Res = ExpandFloatOp_BR_CC(N); break;
+  case ISD::FCOPYSIGN:        Res = ExpandFloatOp_FCOPYSIGN(N); break;
+  case ISD::FP_ROUND:         Res = ExpandFloatOp_FP_ROUND(N); break;
+  case ISD::STRICT_FP_ROUND:  Res = ExpandFloatOp_STRICT_FP_ROUND(N); break;
+  case ISD::FP_TO_SINT:       Res = ExpandFloatOp_FP_TO_SINT(N); break;
+  case ISD::FP_TO_UINT:       Res = ExpandFloatOp_FP_TO_UINT(N); break;
+  case ISD::LROUND:           Res = ExpandFloatOp_LROUND(N); break;
+  case ISD::LLROUND:          Res = ExpandFloatOp_LLROUND(N); break;
+  case ISD::LRINT:            Res = ExpandFloatOp_LRINT(N); break;
+  case ISD::LLRINT:           Res = ExpandFloatOp_LLRINT(N); break;
+  case ISD::SELECT_CC:        Res = ExpandFloatOp_SELECT_CC(N); break;
+  case ISD::SETCC:            Res = ExpandFloatOp_SETCC(N); break;
+  case ISD::STORE:            Res = ExpandFloatOp_STORE(cast<StoreSDNode>(N),
                                                   OpNo); break;
   }
 
   // If the result is null, the sub-method took care of registering results etc.
   if (!Res.getNode()) return false;
 
   // If the result is N, the sub-method updated N in place.  Tell the legalizer
   // core about this.
   if (Res.getNode() == N)
     return true;
 
-  assert(Res.getValueType() == N->getValueType(0) && N->getNumValues() == 1 &&
+  assert(Res.getValueType() == N->getValueType(0) && 
+          (N->getNumValues() == 1 || N->isStrictFPOpcode()) &&
          "Invalid operand expansion");
 
   ReplaceValueWith(SDValue(N, 0), Res);
   return false;
 }
 
 /// FloatExpandSetCCOperands - Expand the operands of a comparison.  This code
 /// is shared among BR_CC, SELECT_CC, and SETCC handlers.
assert(N->getOperand(0).getValueType() == MVT::ppcf128 &&
          "Logic only correct for ppcf128!");
   SDValue Lo, Hi;
   GetExpandedFloat(N->getOperand(0), Lo, Hi);
   // Round it the rest of the way (e.g. to f32) if needed.
   return DAG.getNode(ISD::FP_ROUND, SDLoc(N),
                      N->getValueType(0), Hi, N->getOperand(1));
 }
 
+SDValue DAGTypeLegalizer::ExpandFloatOp_STRICT_FP_ROUND(SDNode *N) {
+  assert(N->getOperand(1).getValueType() == MVT::ppcf128 &&
+         "Logic only correct for ppcf128!");
+  SDValue Lo, Hi;
+  GetExpandedFloat(N->getOperand(1), Lo, Hi);
+
+  // Eliminate the node if the input float type is the same as the output float
+  // type.
+  if (Hi.getValueType() == N->getValueType(0)) {
+    // Connect the output chain to the input chain, unlinking the node.
+    ReplaceValueWith(SDValue(N, 1), N->getOperand(0));
+    return Hi;
+  }
+  // Round it the rest of the way (e.g. to f32) if needed.
+  SDValue Expansion = DAG.getNode(ISD::STRICT_FP_ROUND, SDLoc(N),
+                                  {N->getValueType(0), N->getValueType(1)},
+                                  {N->getOperand(0), Hi, N->getOperand(2)});
+  // Replace the old chain with the new chain.
+  ReplaceValueWith(SDValue(N, 1), SDValue(Expansion.getNode(), 1));
+  return Expansion;
+}
+
 SDValue DAGTypeLegalizer::ExpandFloatOp_FP_TO_SINT(SDNode *N) {
   EVT RVT = N->getValueType(0);
   SDLoc dl(N);
 
   RTLIB::Libcall LC = RTLIB::getFPTOSINT(N->getOperand(0).getValueType(), RVT);
   assert(LC != RTLIB::UNKNOWN_LIBCALL && "Unsupported FP_TO_SINT!");
   return TLI.makeLibCall(DAG, LC, RVT, N->getOperand(0), false, dl).first;
 }

lib/CodeGen/SelectionDAG/LegalizeTypes.h

private:
 
   /// Replace each result of the given MERGE_VALUES node with the corresponding
   /// input operand, except for the result 'ResNo', for which the corresponding
   /// input operand is returned.
   SDValue DisintegrateMERGE_VALUES(SDNode *N, unsigned ResNo);
 
   SDValue JoinIntegers(SDValue Lo, SDValue Hi);
   SDValue LibCallify(RTLIB::Libcall LC, SDNode *N, bool isSigned);
+  SDValue LibCallifyStrictFP(RTLIB::Libcall LC, SDNode *N, bool isSigned);
 
   std::pair<SDValue, SDValue> ExpandChainLibCall(RTLIB::Libcall LC,
                                                  SDNode *Node, bool isSigned);
   std::pair<SDValue, SDValue> ExpandAtomic(SDNode *Node);
 
   SDValue PromoteTargetBoolean(SDValue Bool, EVT ValVT);
 
   void ReplaceValueWith(SDValue From, SDValue To);
private:
   /// exactly equal Hi.  For example, if Op is a ppcf128 which was expanded
   /// into two f64's, then this method returns the two f64's, with Lo being
   /// equal to the lower 64 bits of Op, and Hi to the upper 64 bits.
   void GetExpandedFloat(SDValue Op, SDValue &Lo, SDValue &Hi);
   void SetExpandedFloat(SDValue Op, SDValue Lo, SDValue Hi);
 
   // Float Result Expansion.
   void ExpandFloatResult(SDNode *N, unsigned ResNo);
-  void ExpandFloatRes_ConstantFP(SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FABS      (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FMINNUM   (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FMAXNUM   (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FADD      (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FCEIL     (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FCOPYSIGN (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FCOS      (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FDIV      (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FEXP      (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FEXP2     (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FFLOOR    (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FLOG      (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FLOG2     (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FLOG10    (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FMA       (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FMUL      (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FNEARBYINT(SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FNEG      (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FP_EXTEND (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FPOW      (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FPOWI     (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FREM      (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FRINT     (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FROUND    (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FSIN      (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FSQRT     (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FSUB      (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_FTRUNC    (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_LOAD      (SDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandFloatRes_XINT_TO_FP(SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_ConstantFP        (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FABS              (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FMINNUM           (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FMINNUM    (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FMAXNUM           (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FMAXNUM    (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FADD              (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FADD       (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FCEIL             (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FCEIL      (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FCOPYSIGN         (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FCOS              (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FCOS       (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FDIV              (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FDIV       (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FEXP              (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FEXP       (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FEXP2             (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FEXP2      (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FFLOOR            (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FFLOOR     (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FLOG              (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FLOG       (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FLOG2             (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FLOG2      (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FLOG10            (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FLOG10     (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FMA               (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FMA        (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FMUL              (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FMUL       (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FNEARBYINT        (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FNEARBYINT (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FNEG              (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FP_EXTEND         (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FP_EXTEND  (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FPOW              (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FPOW       (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FPOWI             (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FPOWI      (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FREM              (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FREM       (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FRINT             (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FRINT      (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FROUND            (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FROUND     (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FSIN              (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FSIN       (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FSQRT             (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FSQRT      (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FSUB              (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FSUB       (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_FTRUNC            (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_STRICT_FTRUNC     (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_LOAD              (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandFloatRes_XINT_TO_FP        (SDNode *N, SDValue &Lo, SDValue &Hi);
 
   // Float Operand Expansion.
   bool ExpandFloatOperand(SDNode *N, unsigned OpNo);
   SDValue ExpandFloatOp_BR_CC(SDNode *N);
   SDValue ExpandFloatOp_FCOPYSIGN(SDNode *N);
   SDValue ExpandFloatOp_FP_ROUND(SDNode *N);
+  SDValue ExpandFloatOp_STRICT_FP_ROUND(SDNode *N);
   SDValue ExpandFloatOp_FP_TO_SINT(SDNode *N);
   SDValue ExpandFloatOp_FP_TO_UINT(SDNode *N);
   SDValue ExpandFloatOp_LROUND(SDNode *N);
   SDValue ExpandFloatOp_LLROUND(SDNode *N);
   SDValue ExpandFloatOp_LRINT(SDNode *N);
   SDValue ExpandFloatOp_LLRINT(SDNode *N);
   SDValue ExpandFloatOp_SELECT_CC(SDNode *N);
   SDValue ExpandFloatOp_SETCC(SDNode *N);

test/CodeGen/PowerPC/ppcf128-constrained-fp-instrinsics.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -O3 -mtriple=powerpc64le-linux-gnu < %s | FileCheck --check-prefix=PC64LE %s
+; RUN: llc -O3 -mtriple=powerpc64le-linux-gnu -mcpu=pwr9 < %s | FileCheck --check-prefix=PC64LE9 %s
+; RUN: llc -O3 -mtriple=powerpc64-linux-gnu < %s | FileCheck --check-prefix=PC64 %s
+
+define ppc_fp128 @test_fadd_ppc_fp128(ppc_fp128 %first, ppc_fp128 %second) {
+; PC64LE-LABEL: test_fadd_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl __gcc_qadd
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_fadd_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl __gcc_qadd
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_fadd_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl __gcc_qadd
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %add = call ppc_fp128 @llvm.experimental.constrained.fadd.ppcf128(
+                    ppc_fp128 %first,
+                    ppc_fp128 %second,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %add
+}
+
+define ppc_fp128 @test_fsub_ppc_fp128(ppc_fp128 %first, ppc_fp128 %second) {
+; PC64LE-LABEL: test_fsub_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl __gcc_qsub
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_fsub_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl __gcc_qsub
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_fsub_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl __gcc_qsub
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %sub = call ppc_fp128 @llvm.experimental.constrained.fsub.ppcf128(
+                    ppc_fp128 %first,
+                    ppc_fp128 %second,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %sub
+}
+
+define ppc_fp128 @test_fmul_ppc_fp128(ppc_fp128 %first, ppc_fp128 %second) {
+; PC64LE-LABEL: test_fmul_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl __gcc_qmul
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_fmul_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl __gcc_qmul
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_fmul_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl __gcc_qmul
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %mul = call ppc_fp128 @llvm.experimental.constrained.fmul.ppcf128(
+                    ppc_fp128 %first,
+                    ppc_fp128 %second,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %mul
+}
+
+define ppc_fp128 @test_fdiv_ppc_fp128(ppc_fp128 %first, ppc_fp128 %second) {
+; PC64LE-LABEL: test_fdiv_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl __gcc_qdiv
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_fdiv_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl __gcc_qdiv
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_fdiv_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl __gcc_qdiv
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %div = call ppc_fp128 @llvm.experimental.constrained.fdiv.ppcf128(
+                    ppc_fp128 %first,
+                    ppc_fp128 %second,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %div
+}
+
+define ppc_fp128 @test_frem_ppc_fp128(ppc_fp128 %first, ppc_fp128 %second) {
+; PC64LE-LABEL: test_frem_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl fmodl
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_frem_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl fmodl
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_frem_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl fmodl
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %rem = call ppc_fp128 @llvm.experimental.constrained.frem.ppcf128(
+                    ppc_fp128 %first,
+                    ppc_fp128 %second,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %rem
+}
+
+define ppc_fp128 @test_fma_ppc_fp128(ppc_fp128 %first, ppc_fp128 %second, ppc_fp128 %third) {
+; PC64LE-LABEL: test_fma_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl fmal
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_fma_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl fmal
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_fma_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl fmal
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %add = call ppc_fp128 @llvm.experimental.constrained.fma.ppcf128(
+                    ppc_fp128 %first,
+                    ppc_fp128 %second,
+                    ppc_fp128 %third,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %add
+}
+
+define ppc_fp128 @test_sqrt_ppc_fp128(ppc_fp128 %first) {
+; PC64LE-LABEL: test_sqrt_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl sqrtl
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_sqrt_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl sqrtl
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_sqrt_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl sqrtl
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %sqrt = call ppc_fp128 @llvm.experimental.constrained.sqrt.ppcf128(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %sqrt
+}
+
+define ppc_fp128 @test_pow_ppc_fp128(ppc_fp128 %first, ppc_fp128 %second) {
+; PC64LE-LABEL: test_pow_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl powl
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_pow_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl powl
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_pow_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl powl
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %pow = call ppc_fp128 @llvm.experimental.constrained.pow.ppcf128(
+                    ppc_fp128 %first,
+                    ppc_fp128 %second,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %pow
+}
+
+define ppc_fp128 @test_powi_ppc_fp128(ppc_fp128 %first, i32 %second) {
+; PC64LE-LABEL: test_powi_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    clrldi 5, 5, 32
+; PC64LE-NEXT:    bl __powitf2
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_powi_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    clrldi 5, 5, 32
+; PC64LE9-NEXT:    bl __powitf2
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_powi_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    clrldi 5, 5, 32
+; PC64-NEXT:    bl __powitf2
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+; PC64LE9     :    clrldi 5, 5, 32
+entry:
+  %powi = call ppc_fp128 @llvm.experimental.constrained.powi.ppcf128(
+                    ppc_fp128 %first,
+                    i32 %second,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %powi
+}
+
+define ppc_fp128 @test_sin_ppc_fp128(ppc_fp128 %first) {
+; PC64LE-LABEL: test_sin_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl sinl
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_sin_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl sinl
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_sin_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl sinl
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %sin = call ppc_fp128 @llvm.experimental.constrained.sin.ppcf128(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %sin
+}
+
+define ppc_fp128 @test_cos_ppc_fp128(ppc_fp128 %first) {
+; PC64LE-LABEL: test_cos_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl cosl
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_cos_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl cosl
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_cos_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl cosl
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %cos = call ppc_fp128 @llvm.experimental.constrained.cos.ppcf128(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %cos
+}
+
+define ppc_fp128 @test_exp_ppc_fp128(ppc_fp128 %first) {
+; PC64LE-LABEL: test_exp_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl expl
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_exp_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl expl
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_exp_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl expl
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %exp = call ppc_fp128 @llvm.experimental.constrained.exp.ppcf128(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %exp
+}
+
+define ppc_fp128 @test_exp2_ppc_fp128(ppc_fp128 %first) {
+; PC64LE-LABEL: test_exp2_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl exp2l
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_exp2_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl exp2l
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_exp2_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl exp2l
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %exp2 = call ppc_fp128 @llvm.experimental.constrained.exp2.ppcf128(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %exp2
+}
+
+define ppc_fp128 @test_log_ppc_fp128(ppc_fp128 %first) {
+; PC64LE-LABEL: test_log_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl logl
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_log_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl logl
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_log_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl logl
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %log = call ppc_fp128 @llvm.experimental.constrained.log.ppcf128(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %log
+}
+
+define ppc_fp128 @test_log2_ppc_fp128(ppc_fp128 %first) {
+; PC64LE-LABEL: test_log2_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl log2l
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_log2_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl log2l
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_log2_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl log2l
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %log2 = call ppc_fp128 @llvm.experimental.constrained.log2.ppcf128(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %log2
+}
+
+define ppc_fp128 @test_log10_ppc_fp128(ppc_fp128 %first) {
+; PC64LE-LABEL: test_log10_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl log10l
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_log10_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl log10l
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_log10_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl log10l
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %log10 = call ppc_fp128 @llvm.experimental.constrained.log10.ppcf128(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %log10
+}
+
+define ppc_fp128 @test_rint_ppc_fp128(ppc_fp128 %first) {
+; PC64LE-LABEL: test_rint_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl rintl
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_rint_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl rintl
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_rint_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl rintl
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %rint = call ppc_fp128 @llvm.experimental.constrained.rint.ppcf128(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %rint
+}
+
+define ppc_fp128 @test_nearbyint_ppc_fp128(ppc_fp128 %first) {
+; PC64LE-LABEL: test_nearbyint_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl nearbyintl
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_nearbyint_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl nearbyintl
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_nearbyint_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl nearbyintl
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %nearbyint = call ppc_fp128 @llvm.experimental.constrained.nearbyint.ppcf128(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %nearbyint
+}
+
+define ppc_fp128 @test_maxnum_ppc_fp128(ppc_fp128 %first, ppc_fp128 %second) {
+; PC64LE-LABEL: test_maxnum_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl fmaxl
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_maxnum_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl fmaxl
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_maxnum_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl fmaxl
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %maxnum = call ppc_fp128 @llvm.experimental.constrained.maxnum.ppcf128(
+                    ppc_fp128 %first,
+                    ppc_fp128 %second,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %maxnum
+}
+
+define ppc_fp128 @test_minnum_ppc_fp128(ppc_fp128 %first, ppc_fp128 %second) {
+; PC64LE-LABEL: test_minnum_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl fminl
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_minnum_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl fminl
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_minnum_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl fminl
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %minnum = call ppc_fp128 @llvm.experimental.constrained.minnum.ppcf128(
+                    ppc_fp128 %first,
+                    ppc_fp128 %second,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %minnum
+}
+
+define ppc_fp128 @test_ceil_ppc_fp128(ppc_fp128 %first) {
+; PC64LE-LABEL: test_ceil_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl ceill
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_ceil_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl ceill
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_ceil_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl ceill
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %ceil = call ppc_fp128 @llvm.experimental.constrained.ceil.ppcf128(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %ceil
+}
+
+define ppc_fp128 @test_floor_ppc_fp128(ppc_fp128 %first) {
+; PC64LE-LABEL: test_floor_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl floorl
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_floor_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl floorl
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_floor_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl floorl
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %floor = call ppc_fp128 @llvm.experimental.constrained.floor.ppcf128(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %floor
+}
+
+define ppc_fp128 @test_round_ppc_fp128(ppc_fp128 %first) {
+; PC64LE-LABEL: test_round_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl roundl
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_round_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl roundl
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_round_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl roundl
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %round = call ppc_fp128 @llvm.experimental.constrained.round.ppcf128(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %round
+}
+
+define ppc_fp128 @test_trunc_ppc_fp128(ppc_fp128 %first) {
+; PC64LE-LABEL: test_trunc_ppc_fp128:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -32(1)
+; PC64LE-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    bl truncl
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    addi 1, 1, 32
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_trunc_ppc_fp128:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -32(1)
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 32
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    bl truncl
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addi 1, 1, 32
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_trunc_ppc_fp128:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -112(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 112
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    bl truncl
+; PC64-NEXT:    nop
+; PC64-NEXT:    addi 1, 1, 112
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+entry:
+  %trunc = call ppc_fp128 @llvm.experimental.constrained.trunc.ppcf128(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %trunc
+}
+
+define float @test_fptrunc_ppc_fp128_f32(ppc_fp128 %first) {
+; PC64LE-LABEL: test_fptrunc_ppc_fp128_f32:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    frsp 1, 1
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_fptrunc_ppc_fp128_f32:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    frsp 1, 1
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_fptrunc_ppc_fp128_f32:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    frsp 1, 1
+; PC64-NEXT:    blr
+entry:
+  %fptrunc = call float @llvm.experimental.constrained.fptrunc.ppcf128.f32(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret float %fptrunc
+}
+
+define double @test_fptrunc_ppc_fp128_f64(ppc_fp128 %first) {
+; PC64LE-LABEL: test_fptrunc_ppc_fp128_f64:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_fptrunc_ppc_fp128_f64:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_fptrunc_ppc_fp128_f64:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    blr
+entry:
+  %fptrunc = call double @llvm.experimental.constrained.fptrunc.ppcf128.f64(
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  ret double %fptrunc
+}
+
+define ppc_fp128 @test_fpext_ppc_fp128_f32(float %first) {
+; PC64LE-LABEL: test_fpext_ppc_fp128_f32:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    xxlxor 2, 2, 2
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_fpext_ppc_fp128_f32:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    xxlxor 2, 2, 2
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_fpext_ppc_fp128_f32:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    addis 3, 2, .LCPI26_0@toc@ha
+; PC64-NEXT:    lfs 2, .LCPI26_0@toc@l(3)
+; PC64-NEXT:    blr
+entry:
+  %fpext = call ppc_fp128 @llvm.experimental.constrained.fpext.f32.ppcf128(
+                    float %first,
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %fpext
+}
+
+define ppc_fp128 @test_fpext_ppc_fp128_f64(double %first) {
+; PC64LE-LABEL: test_fpext_ppc_fp128_f64:
+; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    xxlxor 2, 2, 2
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_fpext_ppc_fp128_f64:
+; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    xxlxor 2, 2, 2
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_fpext_ppc_fp128_f64:
+; PC64:       # %bb.0: # %entry
+; PC64-NEXT:    addis 3, 2, .LCPI27_0@toc@ha
+; PC64-NEXT:    lfs 2, .LCPI27_0@toc@l(3)
+; PC64-NEXT:    blr
+entry:
+  %fpext = call ppc_fp128 @llvm.experimental.constrained.fpext.f64.ppcf128(
+                    double %first,
+                    metadata !"fpexcept.strict")
+  ret ppc_fp128 %fpext
+}
+
+; Test that resultant libcalls retain order even when their non-strict FLOP form could be
+; trivially optimized into differing sequences.
+define void @test_constrained_libcall_multichain(float* %firstptr, ppc_fp128* %result) {
+; PC64LE-LABEL: test_constrained_libcall_multichain:
+; PC64LE:       # %bb.0:
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    .cfi_def_cfa_offset 80
+; PC64LE-NEXT:    .cfi_offset lr, 16
+; PC64LE-NEXT:    .cfi_offset r29, -48
+; PC64LE-NEXT:    .cfi_offset r30, -40
+; PC64LE-NEXT:    .cfi_offset f29, -24
+; PC64LE-NEXT:    .cfi_offset f30, -16
+; PC64LE-NEXT:    .cfi_offset f31, -8
+; PC64LE-NEXT:    std 29, -48(1) # 8-byte Folded Spill
+; PC64LE-NEXT:    std 30, -40(1) # 8-byte Folded Spill
+; PC64LE-NEXT:    stfd 29, -24(1) # 8-byte Folded Spill
+; PC64LE-NEXT:    stfd 30, -16(1) # 8-byte Folded Spill
+; PC64LE-NEXT:    stfd 31, -8(1) # 8-byte Folded Spill
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -80(1)
+; PC64LE-NEXT:    mr 29, 3
+; PC64LE-NEXT:    xxlxor 2, 2, 2
+; PC64LE-NEXT:    li 3, 0
+; PC64LE-NEXT:    mr 30, 4
+; PC64LE-NEXT:    lfsx 31, 0, 29
+; PC64LE-NEXT:    xxlxor 4, 4, 4
+; PC64LE-NEXT:    std 3, 8(4)
+; PC64LE-NEXT:    fmr 1, 31
+; PC64LE-NEXT:    fmr 3, 31
+; PC64LE-NEXT:    stfdx 31, 0, 4
+; PC64LE-NEXT:    bl __gcc_qadd
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    fmr 3, 1
+; PC64LE-NEXT:    fmr 4, 2
+; PC64LE-NEXT:    fmr 30, 1
+; PC64LE-NEXT:    fmr 29, 2
+; PC64LE-NEXT:    stfd 1, 16(30)
+; PC64LE-NEXT:    stfd 2, 24(30)
+; PC64LE-NEXT:    bl __gcc_qmul
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    fmr 1, 31
+; PC64LE-NEXT:    xxlxor 2, 2, 2
+; PC64LE-NEXT:    li 5, 2
+; PC64LE-NEXT:    stfd 30, 32(30)
+; PC64LE-NEXT:    stfd 29, 40(30)
+; PC64LE-NEXT:    bl __powitf2
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    frsp 0, 1
+; PC64LE-NEXT:    stfsx 0, 0, 29
+; PC64LE-NEXT:    stfd 2, -8(30)
+; PC64LE-NEXT:    stfd 1, -16(30)
+; PC64LE-NEXT:    addi 1, 1, 80
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
+; PC64LE-NEXT:    lfd 31, -8(1) # 8-byte Folded Reload
+; PC64LE-NEXT:    lfd 30, -16(1) # 8-byte Folded Reload
+; PC64LE-NEXT:    ld 30, -40(1) # 8-byte Folded Reload
+; PC64LE-NEXT:    ld 29, -48(1) # 8-byte Folded Reload
+; PC64LE-NEXT:    lfd 29, -24(1) # 8-byte Folded Reload
+; PC64LE-NEXT:    blr
+;
+; PC64LE9-LABEL: test_constrained_libcall_multichain:
+; PC64LE9:       # %bb.0:
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    .cfi_def_cfa_offset 80
+; PC64LE9-NEXT:    .cfi_offset lr, 16
+; PC64LE9-NEXT:    .cfi_offset r29, -48
+; PC64LE9-NEXT:    .cfi_offset r30, -40
+; PC64LE9-NEXT:    .cfi_offset f29, -24
+; PC64LE9-NEXT:    .cfi_offset f30, -16
+; PC64LE9-NEXT:    .cfi_offset f31, -8
+; PC64LE9-NEXT:    std 29, -48(1) # 8-byte Folded Spill
+; PC64LE9-NEXT:    std 30, -40(1) # 8-byte Folded Spill
+; PC64LE9-NEXT:    stfd 29, -24(1) # 8-byte Folded Spill
+; PC64LE9-NEXT:    stfd 30, -16(1) # 8-byte Folded Spill
+; PC64LE9-NEXT:    stfd 31, -8(1) # 8-byte Folded Spill
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -80(1)
+; PC64LE9-NEXT:    lfs 31, 0(3)
+; PC64LE9-NEXT:    mr 29, 3
+; PC64LE9-NEXT:    li 3, 0
+; PC64LE9-NEXT:    xxlxor 2, 2, 2
+; PC64LE9-NEXT:    xxlxor 4, 4, 4
+; PC64LE9-NEXT:    std 3, 8(4)
+; PC64LE9-NEXT:    fmr 1, 31
+; PC64LE9-NEXT:    fmr 3, 31
+; PC64LE9-NEXT:    mr 30, 4
+; PC64LE9-NEXT:    stfd 31, 0(4)
+; PC64LE9-NEXT:    bl __gcc_qadd
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    fmr 3, 1
+; PC64LE9-NEXT:    fmr 4, 2
+; PC64LE9-NEXT:    fmr 30, 2
+; PC64LE9-NEXT:    fmr 29, 1
+; PC64LE9-NEXT:    stfd 1, 16(30)
+; PC64LE9-NEXT:    stfd 2, 24(30)
+; PC64LE9-NEXT:    bl __gcc_qmul
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    fmr 1, 31
+; PC64LE9-NEXT:    xxlxor 2, 2, 2
+; PC64LE9-NEXT:    li 5, 2
+; PC64LE9-NEXT:    stfd 29, 32(30)
+; PC64LE9-NEXT:    stfd 30, 40(30)
+; PC64LE9-NEXT:    bl __powitf2
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    frsp 0, 1
+; PC64LE9-NEXT:    stfs 0, 0(29)
+; PC64LE9-NEXT:    stfd 2, -8(30)
+; PC64LE9-NEXT:    stfd 1, -16(30)
+; PC64LE9-NEXT:    addi 1, 1, 80
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
+; PC64LE9-NEXT:    lfd 31, -8(1) # 8-byte Folded Reload
+; PC64LE9-NEXT:    lfd 30, -16(1) # 8-byte Folded Reload
+; PC64LE9-NEXT:    lfd 29, -24(1) # 8-byte Folded Reload
+; PC64LE9-NEXT:    ld 30, -40(1) # 8-byte Folded Reload
+; PC64LE9-NEXT:    ld 29, -48(1) # 8-byte Folded Reload
+; PC64LE9-NEXT:    blr
+;
+; PC64-LABEL: test_constrained_libcall_multichain:
+; PC64:       # %bb.0:
+; PC64-NEXT:    mflr 0
+; PC64-NEXT:    std 0, 16(1)
+; PC64-NEXT:    stdu 1, -176(1)
+; PC64-NEXT:    .cfi_def_cfa_offset 176
+; PC64-NEXT:    .cfi_offset lr, 16
+; PC64-NEXT:    .cfi_offset r29, -56
+; PC64-NEXT:    .cfi_offset r30, -48
+; PC64-NEXT:    .cfi_offset f28, -32
+; PC64-NEXT:    .cfi_offset f29, -24
+; PC64-NEXT:    .cfi_offset f30, -16
+; PC64-NEXT:    .cfi_offset f31, -8
+; PC64-NEXT:    std 29, 120(1) # 8-byte Folded Spill
+; PC64-NEXT:    mr 29, 3
+; PC64-NEXT:    li 3, 0
+; PC64-NEXT:    stfd 31, 168(1) # 8-byte Folded Spill
+; PC64-NEXT:    stfd 30, 160(1) # 8-byte Folded Spill
+; PC64-NEXT:    std 30, 128(1) # 8-byte Folded Spill
+; PC64-NEXT:    stfd 28, 144(1) # 8-byte Folded Spill
+; PC64-NEXT:    stfd 29, 152(1) # 8-byte Folded Spill
+; PC64-NEXT:    mr 30, 4
+; PC64-NEXT:    lfs 31, 0(29)
+; PC64-NEXT:    std 3, 8(4)
+; PC64-NEXT:    addis 3, 2, .LCPI28_0@toc@ha
+; PC64-NEXT:    lfs 30, .LCPI28_0@toc@l(3)
+; PC64-NEXT:    fmr 1, 31
+; PC64-NEXT:    fmr 3, 31
+; PC64-NEXT:    fmr 2, 30
+; PC64-NEXT:    fmr 4, 30
+; PC64-NEXT:    stfd 31, 0(4)
+; PC64-NEXT:    bl __gcc_qadd
+; PC64-NEXT:    nop
+; PC64-NEXT:    fmr 3, 1
+; PC64-NEXT:    fmr 4, 2
+; PC64-NEXT:    fmr 29, 1
+; PC64-NEXT:    fmr 28, 2
+; PC64-NEXT:    stfd 1, 16(30)
+; PC64-NEXT:    stfd 2, 24(30)
+; PC64-NEXT:    bl __gcc_qmul
+; PC64-NEXT:    nop
+; PC64-NEXT:    fmr 1, 31
+; PC64-NEXT:    fmr 2, 30
+; PC64-NEXT:    li 5, 2
+; PC64-NEXT:    stfd 29, 32(30)
+; PC64-NEXT:    stfd 28, 40(30)
+; PC64-NEXT:    bl __powitf2
+; PC64-NEXT:    nop
+; PC64-NEXT:    frsp 0, 1
+; PC64-NEXT:    stfs 0, 0(29)
+; PC64-NEXT:    lfd 31, 168(1) # 8-byte Folded Reload
+; PC64-NEXT:    lfd 30, 160(1) # 8-byte Folded Reload
+; PC64-NEXT:    lfd 29, 152(1) # 8-byte Folded Reload
+; PC64-NEXT:    lfd 28, 144(1) # 8-byte Folded Reload
+; PC64-NEXT:    ld 29, 120(1) # 8-byte Folded Reload
+; PC64-NEXT:    stfd 2, -8(30)
+; PC64-NEXT:    stfd 1, -16(30)
+; PC64-NEXT:    ld 30, 128(1) # 8-byte Folded Reload
+; PC64-NEXT:    addi 1, 1, 176
+; PC64-NEXT:    ld 0, 16(1)
+; PC64-NEXT:    mtlr 0
+; PC64-NEXT:    blr
+  %load = load float, float* %firstptr
+  %first = call ppc_fp128 @llvm.experimental.constrained.fpext.f32.ppcf128(
+                    float %load,
+                    metadata !"fpexcept.strict")
+  store ppc_fp128 %first, ppc_fp128* %result
+
+  ; For unconstrained FLOPs, these next two FP instructions would necessarily
+  ; be executed in series with one another.
+  %fadd = call ppc_fp128 @llvm.experimental.constrained.fadd.ppcf128(
+                    ppc_fp128 %first,
+                    ppc_fp128 %first,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  %stridx1 = getelementptr ppc_fp128, ppc_fp128* %result, i32 1
+  store ppc_fp128 %fadd, ppc_fp128* %stridx1
+  %fmul = call ppc_fp128 @llvm.experimental.constrained.fmul.ppcf128(
+                    ppc_fp128 %fadd,
+                    ppc_fp128 %fadd,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  %stridx2 = getelementptr ppc_fp128, ppc_fp128* %stridx1, i32 1
+  store ppc_fp128 %fadd, ppc_fp128* %stridx2
+
+  ; For unconstrained FLOPs, these next two FP instructions could be reordered
+  ; or even executed in parallel with respect to the previous two instructions.
+  ; However, strict floating point rules would not allow this.
+  %powi = call ppc_fp128 @llvm.experimental.constrained.powi.ppcf128(
+                    ppc_fp128 %first,
+                    i32 2,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  %tinypow = call float @llvm.experimental.constrained.fptrunc.ppcf128.f32(
+                    ppc_fp128 %powi,
+                    metadata !"round.dynamic",
+                    metadata !"fpexcept.strict")
+  store float %tinypow, float* %firstptr
+  %stridxn1 = getelementptr ppc_fp128, ppc_fp128* %result, i32 -1
+  store ppc_fp128 %powi, ppc_fp128* %stridxn1
+  ret void
+}
+
+declare ppc_fp128 @llvm.experimental.constrained.fadd.ppcf128(ppc_fp128, ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.ceil.ppcf128(ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.cos.ppcf128(ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.fdiv.ppcf128(ppc_fp128, ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.exp.ppcf128(ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.exp2.ppcf128(ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.floor.ppcf128(ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.fma.ppcf128(ppc_fp128, ppc_fp128, ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.fpext.f32.ppcf128(float, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.fpext.f64.ppcf128(double, metadata)
+declare float @llvm.experimental.constrained.fptrunc.ppcf128.f32(ppc_fp128, metadata, metadata)
+declare double @llvm.experimental.constrained.fptrunc.ppcf128.f64(ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.log.ppcf128(ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.log10.ppcf128(ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.log2.ppcf128(ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.maxnum.ppcf128(ppc_fp128, ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.minnum.ppcf128(ppc_fp128, ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.fmul.ppcf128(ppc_fp128, ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.nearbyint.ppcf128(ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.pow.ppcf128(ppc_fp128, ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.powi.ppcf128(ppc_fp128, i32, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.frem.ppcf128(ppc_fp128, ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.rint.ppcf128(ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.round.ppcf128(ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.sin.ppcf128(ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.sqrt.ppcf128(ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.fsub.ppcf128(ppc_fp128, ppc_fp128, metadata, metadata)
+declare ppc_fp128 @llvm.experimental.constrained.trunc.ppcf128(ppc_fp128, metadata, metadata)