[llvm] [SDAG] Fix llvm.modf for ppc_fp128 (attempt two) (PR #127976)

Benjamin Maxwell via llvm-commits llvm-commits at lists.llvm.org
Thu Feb 20 06:21:33 PST 2025


https://github.com/MacDue updated https://github.com/llvm/llvm-project/pull/127976

>From be65788b0b8440d4a731374f04ee97a94caf9a47 Mon Sep 17 00:00:00 2001
From: Benjamin Maxwell <benjamin.maxwell at arm.com>
Date: Thu, 20 Feb 2025 09:16:42 +0000
Subject: [PATCH 1/3] [SDAG] Fix llvm.modf for ppc_fp128 (attempt two)

Apparently `DAG.getRoot()` can return null, so we need to check that
case. Hopefully fixes: https://lab.llvm.org/buildbot/#/builders/72/builds/8406
---
 .../lib/CodeGen/SelectionDAG/SelectionDAG.cpp |  4 +++-
 llvm/test/CodeGen/PowerPC/llvm.modf.ll        | 22 +++++++++++++++++++
 2 files changed, 25 insertions(+), 1 deletion(-)

diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 0a3210a10d394..8fd0de8d2fc44 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -2644,8 +2644,10 @@ bool SelectionDAG::expandMultipleResultFPLibCall(
     // optimized out. This prevents an FP stack pop from being emitted for it.
     // Setting the root like this ensures there will be a use of the
     // `CopyFromReg` chain, and ensures the FP pop will be emitted.
+    SDValue OldRoot = getRoot();
     SDValue NewRoot =
-        getNode(ISD::TokenFactor, DL, MVT::Other, getRoot(), CallChain);
+        OldRoot ? getNode(ISD::TokenFactor, DL, MVT::Other, OldRoot, CallChain)
+                : CallChain;
     setRoot(NewRoot);
     // Ensure the new root is reachable from the results.
     Results[0] = getMergeValues({Results[0], NewRoot}, DL);
diff --git a/llvm/test/CodeGen/PowerPC/llvm.modf.ll b/llvm/test/CodeGen/PowerPC/llvm.modf.ll
index 69e3b22c7352c..a3f8a9907a46a 100644
--- a/llvm/test/CodeGen/PowerPC/llvm.modf.ll
+++ b/llvm/test/CodeGen/PowerPC/llvm.modf.ll
@@ -328,3 +328,25 @@ define { ppc_fp128, ppc_fp128 } @test_modf_ppcf128(ppc_fp128 %a) {
   %result = call { ppc_fp128, ppc_fp128 } @llvm.modf.ppcf128(ppc_fp128 %a)
   ret { ppc_fp128, ppc_fp128 } %result
 }
+
+define ppc_fp128 @test_modf_ppcf128_only_use_intergral(ppc_fp128 %a) {
+; CHECK-LABEL: test_modf_ppcf128_only_use_intergral:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    mflr r0
+; CHECK-NEXT:    stdu r1, -48(r1)
+; CHECK-NEXT:    std r0, 64(r1)
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-NEXT:    .cfi_offset lr, 16
+; CHECK-NEXT:    addi r5, r1, 32
+; CHECK-NEXT:    bl modfl
+; CHECK-NEXT:    nop
+; CHECK-NEXT:    lfd f1, 32(r1)
+; CHECK-NEXT:    lfd f2, 40(r1)
+; CHECK-NEXT:    addi r1, r1, 48
+; CHECK-NEXT:    ld r0, 16(r1)
+; CHECK-NEXT:    mtlr r0
+; CHECK-NEXT:    blr
+  %result = call { ppc_fp128, ppc_fp128 } @llvm.modf.ppcf128(ppc_fp128 %a)
+  %result.1 = extractvalue { ppc_fp128, ppc_fp128 } %result, 1
+  ret ppc_fp128 %result.1
+}

>From ae1d93f2e51fd76b0eb20b2bb4a2a810180b639c Mon Sep 17 00:00:00 2001
From: Benjamin Maxwell <benjamin.maxwell at arm.com>
Date: Thu, 20 Feb 2025 12:19:46 +0000
Subject: [PATCH 2/3] Fixups

---
 .../CodeGen/SelectionDAG/LegalizeTypes.cpp    | 17 +++-
 llvm/test/CodeGen/PowerPC/llvm.modf.ll        | 83 +++++++++++++++++++
 2 files changed, 98 insertions(+), 2 deletions(-)

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp
index b6abad830c371..3a2d69d8a8eca 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.cpp
@@ -435,8 +435,21 @@ bool DAGTypeLegalizer::run() {
 #endif
     PerformExpensiveChecks();
 
-  // If the root changed (e.g. it was a dead load) update the root.
-  DAG.setRoot(Dummy.getValue());
+  // Get the value of the original root after type legalization.
+  SDValue Root = Dummy.getValue();
+
+  // Get the current root value, if it's not null combine it with the original
+  // root to prevent it being removed as a dead node.
+  if (SDValue LegalRoot = DAG.getRoot()) {
+    Root = DAG.getNode(ISD::TokenFactor, SDLoc(LegalRoot), MVT::Other, Root,
+                       LegalRoot);
+    // The token_factor should not need any legalization (as both inputs have
+    // already been legalized).
+    Root->setNodeId(Processed);
+  }
+
+  // Restore the root.
+  DAG.setRoot(Root);
 
   // Remove dead nodes.  This is important to do for cleanliness but also before
   // the checking loop below.  Implicit folding by the DAG.getNode operators and
diff --git a/llvm/test/CodeGen/PowerPC/llvm.modf.ll b/llvm/test/CodeGen/PowerPC/llvm.modf.ll
index a3f8a9907a46a..1b137c786cc91 100644
--- a/llvm/test/CodeGen/PowerPC/llvm.modf.ll
+++ b/llvm/test/CodeGen/PowerPC/llvm.modf.ll
@@ -350,3 +350,86 @@ define ppc_fp128 @test_modf_ppcf128_only_use_intergral(ppc_fp128 %a) {
   %result.1 = extractvalue { ppc_fp128, ppc_fp128 } %result, 1
   ret ppc_fp128 %result.1
 }
+
+define ppc_fp128 @test_modf_ppcf128_only_use_fractional(ppc_fp128 %a) {
+; CHECK-LABEL: test_modf_ppcf128_only_use_fractional:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    mflr r0
+; CHECK-NEXT:    stdu r1, -48(r1)
+; CHECK-NEXT:    std r0, 64(r1)
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-NEXT:    .cfi_offset lr, 16
+; CHECK-NEXT:    addi r5, r1, 32
+; CHECK-NEXT:    bl modfl
+; CHECK-NEXT:    nop
+; CHECK-NEXT:    addi r1, r1, 48
+; CHECK-NEXT:    ld r0, 16(r1)
+; CHECK-NEXT:    mtlr r0
+; CHECK-NEXT:    blr
+  %result = call { ppc_fp128, ppc_fp128 } @llvm.modf.ppcf128(ppc_fp128 %a)
+  %result.1 = extractvalue { ppc_fp128, ppc_fp128 } %result, 0
+  ret ppc_fp128 %result.1
+}
+
+define { ppc_fp128, ppc_fp128 } @test_modf_ppcf128_tail_call(ppc_fp128 %a) {
+; CHECK-LABEL: test_modf_ppcf128_tail_call:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    mflr r0
+; CHECK-NEXT:    stdu r1, -48(r1)
+; CHECK-NEXT:    std r0, 64(r1)
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-NEXT:    .cfi_offset lr, 16
+; CHECK-NEXT:    addi r5, r1, 32
+; CHECK-NEXT:    bl modfl
+; CHECK-NEXT:    nop
+; CHECK-NEXT:    lfd f3, 32(r1)
+; CHECK-NEXT:    lfd f4, 40(r1)
+; CHECK-NEXT:    addi r1, r1, 48
+; CHECK-NEXT:    ld r0, 16(r1)
+; CHECK-NEXT:    mtlr r0
+; CHECK-NEXT:    blr
+  %result = tail call { ppc_fp128, ppc_fp128 } @llvm.modf.ppcf128(ppc_fp128 %a)
+  ret { ppc_fp128, ppc_fp128 } %result
+}
+
+define ppc_fp128 @test_modf_ppcf128_only_use_intergral_tail_call(ppc_fp128 %a) {
+; CHECK-LABEL: test_modf_ppcf128_only_use_intergral_tail_call:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    mflr r0
+; CHECK-NEXT:    stdu r1, -48(r1)
+; CHECK-NEXT:    std r0, 64(r1)
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-NEXT:    .cfi_offset lr, 16
+; CHECK-NEXT:    addi r5, r1, 32
+; CHECK-NEXT:    bl modfl
+; CHECK-NEXT:    nop
+; CHECK-NEXT:    lfd f1, 32(r1)
+; CHECK-NEXT:    lfd f2, 40(r1)
+; CHECK-NEXT:    addi r1, r1, 48
+; CHECK-NEXT:    ld r0, 16(r1)
+; CHECK-NEXT:    mtlr r0
+; CHECK-NEXT:    blr
+  %result = tail call { ppc_fp128, ppc_fp128 } @llvm.modf.ppcf128(ppc_fp128 %a)
+  %result.1 = extractvalue { ppc_fp128, ppc_fp128 } %result, 1
+  ret ppc_fp128 %result.1
+}
+
+define ppc_fp128 @test_modf_ppcf128_only_use_fractional_tail_call(ppc_fp128 %a) {
+; CHECK-LABEL: test_modf_ppcf128_only_use_fractional_tail_call:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    mflr r0
+; CHECK-NEXT:    stdu r1, -48(r1)
+; CHECK-NEXT:    std r0, 64(r1)
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-NEXT:    .cfi_offset lr, 16
+; CHECK-NEXT:    addi r5, r1, 32
+; CHECK-NEXT:    bl modfl
+; CHECK-NEXT:    nop
+; CHECK-NEXT:    addi r1, r1, 48
+; CHECK-NEXT:    ld r0, 16(r1)
+; CHECK-NEXT:    mtlr r0
+; CHECK-NEXT:    blr
+  %result = tail call { ppc_fp128, ppc_fp128 } @llvm.modf.ppcf128(ppc_fp128 %a)
+  %result.1 = extractvalue { ppc_fp128, ppc_fp128 } %result, 0
+  ret ppc_fp128 %result.1
+}

>From b5226578e5c169102fca3a7dada865b31301e428 Mon Sep 17 00:00:00 2001
From: Benjamin Maxwell <benjamin.maxwell at arm.com>
Date: Thu, 20 Feb 2025 14:20:20 +0000
Subject: [PATCH 3/3] Add missing soften for modf

---
 .../SelectionDAG/LegalizeFloatTypes.cpp       |  44 +-
 llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h |   4 +-
 llvm/test/CodeGen/ARM/llvm.modf.ll            | 504 ++++++++++++++++++
 3 files changed, 539 insertions(+), 13 deletions(-)
 create mode 100644 llvm/test/CodeGen/ARM/llvm.modf.ll

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
index 0244c170a2123..b6fff948f89d5 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
@@ -132,6 +132,7 @@ void DAGTypeLegalizer::SoftenFloatResult(SDNode *N, unsigned ResNo) {
     case ISD::STRICT_FLDEXP: R = SoftenFloatRes_ExpOp(N); break;
     case ISD::FFREXP:        R = SoftenFloatRes_FFREXP(N); break;
     case ISD::FSINCOS:       R = SoftenFloatRes_FSINCOS(N); break;
+    case ISD::FMODF:         R = SoftenFloatRes_FMODF(N); break;
     case ISD::STRICT_FREM:
     case ISD::FREM:        R = SoftenFloatRes_FREM(N); break;
     case ISD::STRICT_FRINT:
@@ -791,27 +792,35 @@ SDValue DAGTypeLegalizer::SoftenFloatRes_FFREXP(SDNode *N) {
   return ReturnVal;
 }
 
-SDValue
-DAGTypeLegalizer::SoftenFloatRes_UnaryWithTwoFPResults(SDNode *N,
-                                                       RTLIB::Libcall LC) {
+SDValue DAGTypeLegalizer::SoftenFloatRes_UnaryWithTwoFPResults(
+    SDNode *N, RTLIB::Libcall LC, std::optional<unsigned> CallRetResNo) {
   assert(!N->isStrictFPOpcode() && "strictfp not implemented");
   EVT VT = N->getValueType(0);
 
+  assert(VT == N->getValueType(1) &&
+         "expected both return values to have the same type");
+
   if (!TLI.getLibcallName(LC))
     return SDValue();
 
   EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), VT);
-  SDValue FirstResultSlot = DAG.CreateStackTemporary(NVT);
-  SDValue SecondResultSlot = DAG.CreateStackTemporary(NVT);
 
   SDLoc DL(N);
 
-  TargetLowering::MakeLibCallOptions CallOptions;
-  std::array Ops{GetSoftenedFloat(N->getOperand(0)), FirstResultSlot,
-                 SecondResultSlot};
-  std::array OpsVT{VT, FirstResultSlot.getValueType(),
-                   SecondResultSlot.getValueType()};
+  SmallVector<SDValue, 3> Ops = {GetSoftenedFloat(N->getOperand(0))};
+  SmallVector<EVT, 3> OpsVT = {VT};
+
+  std::array<SDValue, 2> StackSlots;
+  for (auto [ResNum, _] : enumerate(N->values())) {
+    if (ResNum == CallRetResNo)
+      continue;
+    SDValue StackSlot = DAG.CreateStackTemporary(NVT);
+    Ops.push_back(StackSlot);
+    OpsVT.push_back(StackSlot.getValueType());
+    StackSlots[ResNum] = StackSlot;
+  }
 
+  TargetLowering::MakeLibCallOptions CallOptions;
   // TODO: setTypeListBeforeSoften can't properly express multiple return types,
   // but since both returns have the same type it should be okay.
   CallOptions.setTypeListBeforeSoften({OpsVT}, VT, true);
@@ -825,8 +834,14 @@ DAGTypeLegalizer::SoftenFloatRes_UnaryWithTwoFPResults(SDNode *N,
         MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FrameIdx);
     return DAG.getLoad(NVT, DL, Chain, StackSlot, PtrInfo);
   };
-  SetSoftenedFloat(SDValue(N, 0), CreateStackLoad(FirstResultSlot));
-  SetSoftenedFloat(SDValue(N, 1), CreateStackLoad(SecondResultSlot));
+
+  for (auto [ResNum, SlackSlot] : enumerate(StackSlots)) {
+    if (CallRetResNo == ResNum) {
+      SetSoftenedFloat(SDValue(N, ResNum), ReturnVal);
+      continue;
+    }
+    SetSoftenedFloat(SDValue(N, ResNum), CreateStackLoad(SlackSlot));
+  }
 
   return SDValue();
 }
@@ -836,6 +851,11 @@ SDValue DAGTypeLegalizer::SoftenFloatRes_FSINCOS(SDNode *N) {
       N, RTLIB::getSINCOS(N->getValueType(0)));
 }
 
+SDValue DAGTypeLegalizer::SoftenFloatRes_FMODF(SDNode *N) {
+  return SoftenFloatRes_UnaryWithTwoFPResults(
+      N, RTLIB::getMODF(N->getValueType(0)), /*CallRetResNo=*/0);
+}
+
 SDValue DAGTypeLegalizer::SoftenFloatRes_FREM(SDNode *N) {
   return SoftenFloatRes_Binary(N, GetFPLibCall(N->getValueType(0),
                                                RTLIB::REM_F32,
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
index cac969f7e2185..5cbec5cf409cf 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
@@ -562,7 +562,8 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
   // Convert Float Results to Integer.
   void SoftenFloatResult(SDNode *N, unsigned ResNo);
   SDValue SoftenFloatRes_Unary(SDNode *N, RTLIB::Libcall LC);
-  SDValue SoftenFloatRes_UnaryWithTwoFPResults(SDNode *N, RTLIB::Libcall LC);
+  SDValue SoftenFloatRes_UnaryWithTwoFPResults(
+      SDNode *N, RTLIB::Libcall LC, std::optional<unsigned> CallRetResNo = {});
   SDValue SoftenFloatRes_Binary(SDNode *N, RTLIB::Libcall LC);
   SDValue SoftenFloatRes_MERGE_VALUES(SDNode *N, unsigned ResNo);
   SDValue SoftenFloatRes_ARITH_FENCE(SDNode *N);
@@ -608,6 +609,7 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
   SDValue SoftenFloatRes_ExpOp(SDNode *N);
   SDValue SoftenFloatRes_FFREXP(SDNode *N);
   SDValue SoftenFloatRes_FSINCOS(SDNode *N);
+  SDValue SoftenFloatRes_FMODF(SDNode *N);
   SDValue SoftenFloatRes_FREEZE(SDNode *N);
   SDValue SoftenFloatRes_FREM(SDNode *N);
   SDValue SoftenFloatRes_FRINT(SDNode *N);
diff --git a/llvm/test/CodeGen/ARM/llvm.modf.ll b/llvm/test/CodeGen/ARM/llvm.modf.ll
new file mode 100644
index 0000000000000..66f6c9b9383a7
--- /dev/null
+++ b/llvm/test/CodeGen/ARM/llvm.modf.ll
@@ -0,0 +1,504 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=thumbv7-gnu-linux < %s | FileCheck -check-prefixes=CHECK %s
+; RUN: llc -mtriple=armv6m < %s | FileCheck %s --check-prefix=THUMB
+
+define { half, half } @test_modf_f16(half %a) {
+; CHECK-LABEL: test_modf_f16:
+; CHECK:       @ %bb.0:
+; CHECK-NEXT:    push {r4, lr}
+; CHECK-NEXT:    sub sp, #8
+; CHECK-NEXT:    bl __gnu_h2f_ieee
+; CHECK-NEXT:    add r1, sp, #4
+; CHECK-NEXT:    bl modff
+; CHECK-NEXT:    bl __gnu_f2h_ieee
+; CHECK-NEXT:    mov r4, r0
+; CHECK-NEXT:    ldr r0, [sp, #4]
+; CHECK-NEXT:    bl __gnu_f2h_ieee
+; CHECK-NEXT:    mov r1, r0
+; CHECK-NEXT:    mov r0, r4
+; CHECK-NEXT:    add sp, #8
+; CHECK-NEXT:    pop {r4, pc}
+;
+; THUMB-LABEL: test_modf_f16:
+; THUMB:       @ %bb.0:
+; THUMB-NEXT:    push {r4, lr}
+; THUMB-NEXT:    sub sp, #8
+; THUMB-NEXT:    uxth r0, r0
+; THUMB-NEXT:    bl __gnu_h2f_ieee
+; THUMB-NEXT:    add r1, sp, #4
+; THUMB-NEXT:    bl modff
+; THUMB-NEXT:    bl __gnu_f2h_ieee
+; THUMB-NEXT:    mov r4, r0
+; THUMB-NEXT:    ldr r0, [sp, #4]
+; THUMB-NEXT:    bl __gnu_f2h_ieee
+; THUMB-NEXT:    mov r1, r0
+; THUMB-NEXT:    mov r0, r4
+; THUMB-NEXT:    add sp, #8
+; THUMB-NEXT:    pop {r4, pc}
+  %result = call { half, half } @llvm.modf.f16(half %a)
+  ret { half, half } %result
+}
+
+define half @test_modf_f16_only_use_fractional_part(half %a) {
+; CHECK-LABEL: test_modf_f16_only_use_fractional_part:
+; CHECK:       @ %bb.0:
+; CHECK-NEXT:    push {r7, lr}
+; CHECK-NEXT:    sub sp, #8
+; CHECK-NEXT:    bl __gnu_h2f_ieee
+; CHECK-NEXT:    add r1, sp, #4
+; CHECK-NEXT:    bl modff
+; CHECK-NEXT:    bl __gnu_f2h_ieee
+; CHECK-NEXT:    add sp, #8
+; CHECK-NEXT:    pop {r7, pc}
+;
+; THUMB-LABEL: test_modf_f16_only_use_fractional_part:
+; THUMB:       @ %bb.0:
+; THUMB-NEXT:    push {r7, lr}
+; THUMB-NEXT:    sub sp, #8
+; THUMB-NEXT:    uxth r0, r0
+; THUMB-NEXT:    bl __gnu_h2f_ieee
+; THUMB-NEXT:    add r1, sp, #4
+; THUMB-NEXT:    bl modff
+; THUMB-NEXT:    bl __gnu_f2h_ieee
+; THUMB-NEXT:    add sp, #8
+; THUMB-NEXT:    pop {r7, pc}
+  %result = call { half, half } @llvm.modf.f16(half %a)
+  %result.0 = extractvalue { half, half } %result, 0
+  ret half %result.0
+}
+
+define half @test_modf_f16_only_use_integral_part(half %a) {
+; CHECK-LABEL: test_modf_f16_only_use_integral_part:
+; CHECK:       @ %bb.0:
+; CHECK-NEXT:    push {r7, lr}
+; CHECK-NEXT:    sub sp, #8
+; CHECK-NEXT:    bl __gnu_h2f_ieee
+; CHECK-NEXT:    add r1, sp, #4
+; CHECK-NEXT:    bl modff
+; CHECK-NEXT:    ldr r0, [sp, #4]
+; CHECK-NEXT:    bl __gnu_f2h_ieee
+; CHECK-NEXT:    add sp, #8
+; CHECK-NEXT:    pop {r7, pc}
+;
+; THUMB-LABEL: test_modf_f16_only_use_integral_part:
+; THUMB:       @ %bb.0:
+; THUMB-NEXT:    push {r7, lr}
+; THUMB-NEXT:    sub sp, #8
+; THUMB-NEXT:    uxth r0, r0
+; THUMB-NEXT:    bl __gnu_h2f_ieee
+; THUMB-NEXT:    add r1, sp, #4
+; THUMB-NEXT:    bl modff
+; THUMB-NEXT:    ldr r0, [sp, #4]
+; THUMB-NEXT:    bl __gnu_f2h_ieee
+; THUMB-NEXT:    add sp, #8
+; THUMB-NEXT:    pop {r7, pc}
+  %result = call { half, half } @llvm.modf.f16(half %a)
+  %result.1 = extractvalue { half, half } %result, 1
+  ret half %result.1
+}
+
+define { <2 x half>, <2 x half> } @test_modf_v2f16(<2 x half> %a) {
+; CHECK-LABEL: test_modf_v2f16:
+; CHECK:       @ %bb.0:
+; CHECK-NEXT:    push {r4, lr}
+; CHECK-NEXT:    vpush {d8}
+; CHECK-NEXT:    sub sp, #16
+; CHECK-NEXT:    mov r4, r0
+; CHECK-NEXT:    mov r0, r1
+; CHECK-NEXT:    bl __gnu_h2f_ieee
+; CHECK-NEXT:    add r1, sp, #4
+; CHECK-NEXT:    bl modff
+; CHECK-NEXT:    bl __gnu_f2h_ieee
+; CHECK-NEXT:    strh.w r0, [sp, #14]
+; CHECK-NEXT:    mov r0, r4
+; CHECK-NEXT:    bl __gnu_h2f_ieee
+; CHECK-NEXT:    mov r1, sp
+; CHECK-NEXT:    bl modff
+; CHECK-NEXT:    bl __gnu_f2h_ieee
+; CHECK-NEXT:    strh.w r0, [sp, #12]
+; CHECK-NEXT:    add r0, sp, #12
+; CHECK-NEXT:    vld1.32 {d8[0]}, [r0:32]
+; CHECK-NEXT:    ldr r0, [sp, #4]
+; CHECK-NEXT:    bl __gnu_f2h_ieee
+; CHECK-NEXT:    ldr r1, [sp]
+; CHECK-NEXT:    strh.w r0, [sp, #10]
+; CHECK-NEXT:    mov r0, r1
+; CHECK-NEXT:    bl __gnu_f2h_ieee
+; CHECK-NEXT:    strh.w r0, [sp, #8]
+; CHECK-NEXT:    add r0, sp, #8
+; CHECK-NEXT:    vmovl.u16 q9, d8
+; CHECK-NEXT:    vld1.32 {d16[0]}, [r0:32]
+; CHECK-NEXT:    vmovl.u16 q8, d16
+; CHECK-NEXT:    vmov.32 r0, d18[0]
+; CHECK-NEXT:    vmov.32 r1, d18[1]
+; CHECK-NEXT:    vmov.32 r2, d16[0]
+; CHECK-NEXT:    vmov.32 r3, d16[1]
+; CHECK-NEXT:    add sp, #16
+; CHECK-NEXT:    vpop {d8}
+; CHECK-NEXT:    pop {r4, pc}
+;
+; THUMB-LABEL: test_modf_v2f16:
+; THUMB:       @ %bb.0:
+; THUMB-NEXT:    push {r4, r5, r6, lr}
+; THUMB-NEXT:    sub sp, #8
+; THUMB-NEXT:    mov r5, r1
+; THUMB-NEXT:    uxth r0, r0
+; THUMB-NEXT:    bl __gnu_h2f_ieee
+; THUMB-NEXT:    mov r1, sp
+; THUMB-NEXT:    bl modff
+; THUMB-NEXT:    bl __gnu_f2h_ieee
+; THUMB-NEXT:    mov r4, r0
+; THUMB-NEXT:    uxth r0, r5
+; THUMB-NEXT:    bl __gnu_h2f_ieee
+; THUMB-NEXT:    add r1, sp, #4
+; THUMB-NEXT:    bl modff
+; THUMB-NEXT:    bl __gnu_f2h_ieee
+; THUMB-NEXT:    mov r5, r0
+; THUMB-NEXT:    ldr r0, [sp]
+; THUMB-NEXT:    bl __gnu_f2h_ieee
+; THUMB-NEXT:    mov r6, r0
+; THUMB-NEXT:    ldr r0, [sp, #4]
+; THUMB-NEXT:    bl __gnu_f2h_ieee
+; THUMB-NEXT:    mov r3, r0
+; THUMB-NEXT:    mov r0, r4
+; THUMB-NEXT:    mov r1, r5
+; THUMB-NEXT:    mov r2, r6
+; THUMB-NEXT:    add sp, #8
+; THUMB-NEXT:    pop {r4, r5, r6, pc}
+  %result = call { <2 x half>, <2 x half> } @llvm.modf.v2f16(<2 x half> %a)
+  ret { <2 x half>, <2 x half> } %result
+}
+
+define { float, float } @test_modf_f32(float %a) {
+; CHECK-LABEL: test_modf_f32:
+; CHECK:       @ %bb.0:
+; CHECK-NEXT:    push {r7, lr}
+; CHECK-NEXT:    sub sp, #8
+; CHECK-NEXT:    add r1, sp, #4
+; CHECK-NEXT:    bl modff
+; CHECK-NEXT:    ldr r1, [sp, #4]
+; CHECK-NEXT:    add sp, #8
+; CHECK-NEXT:    pop {r7, pc}
+;
+; THUMB-LABEL: test_modf_f32:
+; THUMB:       @ %bb.0:
+; THUMB-NEXT:    push {r7, lr}
+; THUMB-NEXT:    sub sp, #8
+; THUMB-NEXT:    add r1, sp, #4
+; THUMB-NEXT:    bl modff
+; THUMB-NEXT:    ldr r1, [sp, #4]
+; THUMB-NEXT:    add sp, #8
+; THUMB-NEXT:    pop {r7, pc}
+  %result = call { float, float } @llvm.modf.f32(float %a)
+  ret { float, float } %result
+}
+
+define { <3 x float>, <3 x float> } @test_modf_v3f32(<3 x float> %a) {
+; CHECK-LABEL: test_modf_v3f32:
+; CHECK:       @ %bb.0:
+; CHECK-NEXT:    push {r4, r5, r6, lr}
+; CHECK-NEXT:    vpush {d8, d9}
+; CHECK-NEXT:    sub sp, #8
+; CHECK-NEXT:    vldr d9, [sp, #40]
+; CHECK-NEXT:    mov r4, r0
+; CHECK-NEXT:    mov r1, sp
+; CHECK-NEXT:    mov r0, r2
+; CHECK-NEXT:    mov r5, r3
+; CHECK-NEXT:    vmov d8, r2, r3
+; CHECK-NEXT:    bl modff
+; CHECK-NEXT:    add r1, sp, #4
+; CHECK-NEXT:    mov r6, r0
+; CHECK-NEXT:    mov r0, r5
+; CHECK-NEXT:    bl modff
+; CHECK-NEXT:    mov r5, r0
+; CHECK-NEXT:    vmov r0, s18
+; CHECK-NEXT:    vldmia sp, {s0, s1}
+; CHECK-NEXT:    add.w r1, r4, #16
+; CHECK-NEXT:    vst1.32 {d0}, [r1:64]!
+; CHECK-NEXT:    bl modff
+; CHECK-NEXT:    vmov s1, r5
+; CHECK-NEXT:    vmov s0, r6
+; CHECK-NEXT:    vst1.32 {d0}, [r4:64]!
+; CHECK-NEXT:    str r0, [r4]
+; CHECK-NEXT:    add sp, #8
+; CHECK-NEXT:    vpop {d8, d9}
+; CHECK-NEXT:    pop {r4, r5, r6, pc}
+;
+; THUMB-LABEL: test_modf_v3f32:
+; THUMB:       @ %bb.0:
+; THUMB-NEXT:    push {r4, r5, r6, r7, lr}
+; THUMB-NEXT:    sub sp, #12
+; THUMB-NEXT:    mov r7, r3
+; THUMB-NEXT:    mov r5, r2
+; THUMB-NEXT:    mov r4, r0
+; THUMB-NEXT:    ldr r0, [sp, #32]
+; THUMB-NEXT:    add r1, sp, #8
+; THUMB-NEXT:    bl modff
+; THUMB-NEXT:    mov r6, r0
+; THUMB-NEXT:    ldr r0, [sp, #8]
+; THUMB-NEXT:    str r0, [r4, #24]
+; THUMB-NEXT:    add r1, sp, #4
+; THUMB-NEXT:    mov r0, r7
+; THUMB-NEXT:    bl modff
+; THUMB-NEXT:    mov r7, r0
+; THUMB-NEXT:    ldr r0, [sp, #4]
+; THUMB-NEXT:    str r0, [r4, #20]
+; THUMB-NEXT:    mov r1, sp
+; THUMB-NEXT:    mov r0, r5
+; THUMB-NEXT:    bl modff
+; THUMB-NEXT:    ldr r1, [sp]
+; THUMB-NEXT:    str r1, [r4, #16]
+; THUMB-NEXT:    stm r4!, {r0, r7}
+; THUMB-NEXT:    str r6, [r4]
+; THUMB-NEXT:    add sp, #12
+; THUMB-NEXT:    pop {r4, r5, r6, r7, pc}
+  %result = call { <3 x float>, <3 x float> } @llvm.modf.v3f32(<3 x float> %a)
+  ret { <3 x float>, <3 x float> } %result
+}
+
+define { <2 x float>, <2 x float> } @test_modf_v2f32(<2 x float> %a) {
+; CHECK-LABEL: test_modf_v2f32:
+; CHECK:       @ %bb.0:
+; CHECK-NEXT:    push {r4, lr}
+; CHECK-NEXT:    vpush {d8}
+; CHECK-NEXT:    sub sp, #8
+; CHECK-NEXT:    vmov d8, r0, r1
+; CHECK-NEXT:    mov r1, sp
+; CHECK-NEXT:    vmov r0, s17
+; CHECK-NEXT:    bl modff
+; CHECK-NEXT:    mov r4, r0
+; CHECK-NEXT:    vmov r0, s16
+; CHECK-NEXT:    add r1, sp, #4
+; CHECK-NEXT:    bl modff
+; CHECK-NEXT:    vldr s1, [sp]
+; CHECK-NEXT:    mov r1, r4
+; CHECK-NEXT:    vldr s0, [sp, #4]
+; CHECK-NEXT:    vmov r2, r3, d0
+; CHECK-NEXT:    add sp, #8
+; CHECK-NEXT:    vpop {d8}
+; CHECK-NEXT:    pop {r4, pc}
+;
+; THUMB-LABEL: test_modf_v2f32:
+; THUMB:       @ %bb.0:
+; THUMB-NEXT:    push {r4, r5, r7, lr}
+; THUMB-NEXT:    sub sp, #8
+; THUMB-NEXT:    mov r4, r1
+; THUMB-NEXT:    mov r1, sp
+; THUMB-NEXT:    bl modff
+; THUMB-NEXT:    mov r5, r0
+; THUMB-NEXT:    add r1, sp, #4
+; THUMB-NEXT:    mov r0, r4
+; THUMB-NEXT:    bl modff
+; THUMB-NEXT:    mov r1, r0
+; THUMB-NEXT:    ldr r2, [sp]
+; THUMB-NEXT:    ldr r3, [sp, #4]
+; THUMB-NEXT:    mov r0, r5
+; THUMB-NEXT:    add sp, #8
+; THUMB-NEXT:    pop {r4, r5, r7, pc}
+  %result = call { <2 x float>, <2 x float> } @llvm.modf.v2f32(<2 x float> %a)
+  ret { <2 x float>, <2 x float> } %result
+}
+
+define { <2 x double>, <2 x double> } @test_modf_v2f64(<2 x double> %a) {
+; CHECK-LABEL: test_modf_v2f64:
+; CHECK:       @ %bb.0:
+; CHECK-NEXT:    push {r4, lr}
+; CHECK-NEXT:    vpush {d8, d9}
+; CHECK-NEXT:    sub sp, #16
+; CHECK-NEXT:    mov r1, r3
+; CHECK-NEXT:    mov r3, r2
+; CHECK-NEXT:    add r2, sp, #8
+; CHECK-NEXT:    mov r4, r0
+; CHECK-NEXT:    mov r0, r3
+; CHECK-NEXT:    bl modf
+; CHECK-NEXT:    ldrd r12, r3, [sp, #40]
+; CHECK-NEXT:    vmov d8, r0, r1
+; CHECK-NEXT:    mov r2, sp
+; CHECK-NEXT:    mov r1, r3
+; CHECK-NEXT:    mov r0, r12
+; CHECK-NEXT:    bl modf
+; CHECK-NEXT:    vmov d9, r0, r1
+; CHECK-NEXT:    vldr d17, [sp]
+; CHECK-NEXT:    vldr d16, [sp, #8]
+; CHECK-NEXT:    vst1.64 {d8, d9}, [r4]!
+; CHECK-NEXT:    vst1.64 {d16, d17}, [r4]
+; CHECK-NEXT:    add sp, #16
+; CHECK-NEXT:    vpop {d8, d9}
+; CHECK-NEXT:    pop {r4, pc}
+;
+; THUMB-LABEL: test_modf_v2f64:
+; THUMB:       @ %bb.0:
+; THUMB-NEXT:    push {r4, r5, r6, r7, lr}
+; THUMB-NEXT:    sub sp, #28
+; THUMB-NEXT:    str r3, [sp, #4] @ 4-byte Spill
+; THUMB-NEXT:    mov r7, r2
+; THUMB-NEXT:    mov r4, r0
+; THUMB-NEXT:    ldr r0, [sp, #48]
+; THUMB-NEXT:    ldr r1, [sp, #52]
+; THUMB-NEXT:    add r2, sp, #16
+; THUMB-NEXT:    bl modf
+; THUMB-NEXT:    mov r6, r0
+; THUMB-NEXT:    mov r5, r1
+; THUMB-NEXT:    ldr r0, [sp, #20]
+; THUMB-NEXT:    str r0, [r4, #28]
+; THUMB-NEXT:    ldr r0, [sp, #16]
+; THUMB-NEXT:    str r0, [r4, #24]
+; THUMB-NEXT:    add r2, sp, #8
+; THUMB-NEXT:    mov r0, r7
+; THUMB-NEXT:    ldr r1, [sp, #4] @ 4-byte Reload
+; THUMB-NEXT:    bl modf
+; THUMB-NEXT:    ldr r2, [sp, #12]
+; THUMB-NEXT:    str r2, [r4, #20]
+; THUMB-NEXT:    ldr r2, [sp, #8]
+; THUMB-NEXT:    str r2, [r4, #16]
+; THUMB-NEXT:    str r5, [r4, #12]
+; THUMB-NEXT:    stm r4!, {r0, r1, r6}
+; THUMB-NEXT:    add sp, #28
+; THUMB-NEXT:    pop {r4, r5, r6, r7, pc}
+  %result = call { <2 x double>, <2 x double> } @llvm.modf.v2f64(<2 x double> %a)
+  ret { <2 x double>, <2 x double> } %result
+}
+
+define { double, double } @test_modf_f64(double %a) {
+; CHECK-LABEL: test_modf_f64:
+; CHECK:       @ %bb.0:
+; CHECK-NEXT:    push {r7, lr}
+; CHECK-NEXT:    sub sp, #8
+; CHECK-NEXT:    mov r2, sp
+; CHECK-NEXT:    bl modf
+; CHECK-NEXT:    ldrd r2, r3, [sp], #8
+; CHECK-NEXT:    pop {r7, pc}
+;
+; THUMB-LABEL: test_modf_f64:
+; THUMB:       @ %bb.0:
+; THUMB-NEXT:    push {r7, lr}
+; THUMB-NEXT:    sub sp, #8
+; THUMB-NEXT:    mov r2, sp
+; THUMB-NEXT:    bl modf
+; THUMB-NEXT:    ldr r2, [sp]
+; THUMB-NEXT:    ldr r3, [sp, #4]
+; THUMB-NEXT:    add sp, #8
+; THUMB-NEXT:    pop {r7, pc}
+  %result = call { double, double } @llvm.modf.f64(double %a)
+  ret { double, double } %result
+}
+
+define double @test_modf_f64_only_use_intergral(double %a) {
+; CHECK-LABEL: test_modf_f64_only_use_intergral:
+; CHECK:       @ %bb.0:
+; CHECK-NEXT:    push {r7, lr}
+; CHECK-NEXT:    sub sp, #8
+; CHECK-NEXT:    mov r2, sp
+; CHECK-NEXT:    bl modf
+; CHECK-NEXT:    ldrd r0, r1, [sp], #8
+; CHECK-NEXT:    pop {r7, pc}
+;
+; THUMB-LABEL: test_modf_f64_only_use_intergral:
+; THUMB:       @ %bb.0:
+; THUMB-NEXT:    push {r7, lr}
+; THUMB-NEXT:    sub sp, #8
+; THUMB-NEXT:    mov r2, sp
+; THUMB-NEXT:    bl modf
+; THUMB-NEXT:    ldr r0, [sp]
+; THUMB-NEXT:    ldr r1, [sp, #4]
+; THUMB-NEXT:    add sp, #8
+; THUMB-NEXT:    pop {r7, pc}
+  %result = call { double, double } @llvm.modf.f64(double %a)
+  %result.1 = extractvalue { double, double } %result, 1
+  ret double %result.1
+}
+
+define double @test_modf_f64_only_use_fractional(double %a) {
+; CHECK-LABEL: test_modf_f64_only_use_fractional:
+; CHECK:       @ %bb.0:
+; CHECK-NEXT:    push {r7, lr}
+; CHECK-NEXT:    sub sp, #8
+; CHECK-NEXT:    mov r2, sp
+; CHECK-NEXT:    bl modf
+; CHECK-NEXT:    add sp, #8
+; CHECK-NEXT:    pop {r7, pc}
+;
+; THUMB-LABEL: test_modf_f64_only_use_fractional:
+; THUMB:       @ %bb.0:
+; THUMB-NEXT:    push {r7, lr}
+; THUMB-NEXT:    sub sp, #8
+; THUMB-NEXT:    mov r2, sp
+; THUMB-NEXT:    bl modf
+; THUMB-NEXT:    add sp, #8
+; THUMB-NEXT:    pop {r7, pc}
+  %result = call { double, double } @llvm.modf.f64(double %a)
+  %result.1 = extractvalue { double, double } %result, 0
+  ret double %result.1
+}
+
+define { double, double } @test_modf_f64_tail_call(double %a) {
+; CHECK-LABEL: test_modf_f64_tail_call:
+; CHECK:       @ %bb.0:
+; CHECK-NEXT:    push {r7, lr}
+; CHECK-NEXT:    sub sp, #8
+; CHECK-NEXT:    mov r2, sp
+; CHECK-NEXT:    bl modf
+; CHECK-NEXT:    ldrd r2, r3, [sp], #8
+; CHECK-NEXT:    pop {r7, pc}
+;
+; THUMB-LABEL: test_modf_f64_tail_call:
+; THUMB:       @ %bb.0:
+; THUMB-NEXT:    push {r7, lr}
+; THUMB-NEXT:    sub sp, #8
+; THUMB-NEXT:    mov r2, sp
+; THUMB-NEXT:    bl modf
+; THUMB-NEXT:    ldr r2, [sp]
+; THUMB-NEXT:    ldr r3, [sp, #4]
+; THUMB-NEXT:    add sp, #8
+; THUMB-NEXT:    pop {r7, pc}
+  %result = tail call { double, double } @llvm.modf.f64(double %a)
+  ret { double, double } %result
+}
+
+define double @test_modf_f64_only_use_intergral_tail_call(double %a) {
+; CHECK-LABEL: test_modf_f64_only_use_intergral_tail_call:
+; CHECK:       @ %bb.0:
+; CHECK-NEXT:    push {r7, lr}
+; CHECK-NEXT:    sub sp, #8
+; CHECK-NEXT:    mov r2, sp
+; CHECK-NEXT:    bl modf
+; CHECK-NEXT:    ldrd r0, r1, [sp], #8
+; CHECK-NEXT:    pop {r7, pc}
+;
+; THUMB-LABEL: test_modf_f64_only_use_intergral_tail_call:
+; THUMB:       @ %bb.0:
+; THUMB-NEXT:    push {r7, lr}
+; THUMB-NEXT:    sub sp, #8
+; THUMB-NEXT:    mov r2, sp
+; THUMB-NEXT:    bl modf
+; THUMB-NEXT:    ldr r0, [sp]
+; THUMB-NEXT:    ldr r1, [sp, #4]
+; THUMB-NEXT:    add sp, #8
+; THUMB-NEXT:    pop {r7, pc}
+  %result = tail call { double, double } @llvm.modf.f64(double %a)
+  %result.1 = extractvalue { double, double } %result, 1
+  ret double %result.1
+}
+
+define double @test_modf_f64_only_use_fractional_tail_call(double %a) {
+; CHECK-LABEL: test_modf_f64_only_use_fractional_tail_call:
+; CHECK:       @ %bb.0:
+; CHECK-NEXT:    push {r7, lr}
+; CHECK-NEXT:    sub sp, #8
+; CHECK-NEXT:    mov r2, sp
+; CHECK-NEXT:    bl modf
+; CHECK-NEXT:    add sp, #8
+; CHECK-NEXT:    pop {r7, pc}
+;
+; THUMB-LABEL: test_modf_f64_only_use_fractional_tail_call:
+; THUMB:       @ %bb.0:
+; THUMB-NEXT:    push {r7, lr}
+; THUMB-NEXT:    sub sp, #8
+; THUMB-NEXT:    mov r2, sp
+; THUMB-NEXT:    bl modf
+; THUMB-NEXT:    add sp, #8
+; THUMB-NEXT:    pop {r7, pc}
+  %result = tail call { double, double } @llvm.modf.f64(double %a)
+  %result.1 = extractvalue { double, double } %result, 0
+  ret double %result.1
+}



More information about the llvm-commits mailing list