R600: Implement f64 ftrunc, ffloor and fceil.

CI has instructions for these, so this fixes them for older hardware.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211183 91177308-0d34-0410-b5e6-96231b3b80d8

5 files changed, 198 insertions, 22 deletions

diff --git a/lib/Target/R600/AMDGPUISelLowering.cpp b/lib/Target/R600/AMDGPUISelLowering.cpp
index e09db6429b..1e6f38ffc3 100644
--- a/lib/Target/R600/AMDGPUISelLowering.cpp
+++ b/lib/Target/R600/AMDGPUISelLowering.cpp
@@ -219,7 +219,10 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::BR_CC, MVT::i1, Expand);
 
   if (Subtarget->getGeneration() < AMDGPUSubtarget::SEA_ISLANDS) {
+    setOperationAction(ISD::FCEIL, MVT::f64, Custom);
+    setOperationAction(ISD::FTRUNC, MVT::f64, Custom);
     setOperationAction(ISD::FRINT, MVT::f64, Custom);
+    setOperationAction(ISD::FFLOOR, MVT::f64, Custom);
   }
 
   if (!Subtarget->hasBFI()) {
@@ -494,7 +497,10 @@ SDValue AMDGPUTargetLowering::LowerOperation(SDValue Op,
   case ISD::SDIV: return LowerSDIV(Op, DAG);
   case ISD::SREM: return LowerSREM(Op, DAG);
   case ISD::UDIVREM: return LowerUDIVREM(Op, DAG);
+  case ISD::FCEIL: return LowerFCEIL(Op, DAG);
+  case ISD::FTRUNC: return LowerFTRUNC(Op, DAG);
   case ISD::FRINT: return LowerFRINT(Op, DAG);
+  case ISD::FFLOOR: return LowerFFLOOR(Op, DAG);
   case ISD::UINT_TO_FP: return LowerUINT_TO_FP(Op, DAG);
 
   // AMDIL DAG lowering.
@@ -1571,6 +1577,84 @@ SDValue AMDGPUTargetLowering::LowerUDIVREM(SDValue Op,
   return DAG.getMergeValues(Ops, DL);
 }
 
+SDValue AMDGPUTargetLowering::LowerFCEIL(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc SL(Op);
+  SDValue Src = Op.getOperand(0);
+
+  // result = trunc(src)
+  // if (src > 0.0 && src != result)
+  //   result += 1.0
+
+  SDValue Trunc = DAG.getNode(ISD::FTRUNC, SL, MVT::f64, Src);
+
+  const SDValue Zero = DAG.getConstantFP(0.0, MVT::f64);
+  const SDValue One = DAG.getConstantFP(1.0, MVT::f64);
+
+  EVT SetCCVT = getSetCCResultType(*DAG.getContext(), MVT::f64);
+
+  SDValue Lt0 = DAG.getSetCC(SL, SetCCVT, Src, Zero, ISD::SETOGT);
+  SDValue NeTrunc = DAG.getSetCC(SL, SetCCVT, Src, Trunc, ISD::SETONE);
+  SDValue And = DAG.getNode(ISD::AND, SL, SetCCVT, Lt0, NeTrunc);
+
+  SDValue Add = DAG.getNode(ISD::SELECT, SL, MVT::f64, And, One, Zero);
+  return DAG.getNode(ISD::FADD, SL, MVT::f64, Trunc, Add);
+}
+
+SDValue AMDGPUTargetLowering::LowerFTRUNC(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc SL(Op);
+  SDValue Src = Op.getOperand(0);
+
+  assert(Op.getValueType() == MVT::f64);
+
+  const SDValue Zero = DAG.getConstant(0, MVT::i32);
+  const SDValue One = DAG.getConstant(1, MVT::i32);
+
+  SDValue VecSrc = DAG.getNode(ISD::BITCAST, SL, MVT::v2i32, Src);
+
+  // Extract the upper half, since this is where we will find the sign and
+  // exponent.
+  SDValue Hi = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SL, MVT::i32, VecSrc, One);
+
+  const unsigned FractBits = 52;
+  const unsigned ExpBits = 11;
+
+  // Extract the exponent.
+  SDValue ExpPart = DAG.getNode(AMDGPUISD::BFE_I32, SL, MVT::i32,
+                                Hi,
+                                DAG.getConstant(FractBits - 32, MVT::i32),
+                                DAG.getConstant(ExpBits, MVT::i32));
+  SDValue Exp = DAG.getNode(ISD::SUB, SL, MVT::i32, ExpPart,
+                            DAG.getConstant(1023, MVT::i32));
+
+  // Extract the sign bit.
+  const SDValue SignBitMask = DAG.getConstant(1ul << 31, MVT::i32);
+  SDValue SignBit = DAG.getNode(ISD::AND, SL, MVT::i32, Hi, SignBitMask);
+
+  // Extend back to to 64-bits.
+  SDValue SignBit64 = DAG.getNode(ISD::BUILD_VECTOR, SL, MVT::v2i32,
+                                  Zero, SignBit);
+  SignBit64 = DAG.getNode(ISD::BITCAST, SL, MVT::i64, SignBit64);
+
+  SDValue BcInt = DAG.getNode(ISD::BITCAST, SL, MVT::i64, Src);
+  const SDValue FractMask = DAG.getConstant((1L << FractBits) - 1, MVT::i64);
+
+  SDValue Shr = DAG.getNode(ISD::SRA, SL, MVT::i64, FractMask, Exp);
+  SDValue Not = DAG.getNOT(SL, Shr, MVT::i64);
+  SDValue Tmp0 = DAG.getNode(ISD::AND, SL, MVT::i64, BcInt, Not);
+
+  EVT SetCCVT = getSetCCResultType(*DAG.getContext(), MVT::i32);
+
+  const SDValue FiftyOne = DAG.getConstant(FractBits - 1, MVT::i32);
+
+  SDValue ExpLt0 = DAG.getSetCC(SL, SetCCVT, Exp, Zero, ISD::SETLT);
+  SDValue ExpGt51 = DAG.getSetCC(SL, SetCCVT, Exp, FiftyOne, ISD::SETGT);
+
+  SDValue Tmp1 = DAG.getNode(ISD::SELECT, SL, MVT::i64, ExpLt0, SignBit64, Tmp0);
+  SDValue Tmp2 = DAG.getNode(ISD::SELECT, SL, MVT::i64, ExpGt51, BcInt, Tmp1);
+
+  return DAG.getNode(ISD::BITCAST, SL, MVT::f64, Tmp2);
+}
+
 SDValue AMDGPUTargetLowering::LowerFRINT(SDValue Op, SelectionDAG &DAG) const {
   SDLoc SL(Op);
   SDValue Src = Op.getOperand(0);
@@ -1592,6 +1676,29 @@ SDValue AMDGPUTargetLowering::LowerFRINT(SDValue Op, SelectionDAG &DAG) const {
   return DAG.getSelect(SL, MVT::f64, Cond, Src, Tmp2);
 }
 
+SDValue AMDGPUTargetLowering::LowerFFLOOR(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc SL(Op);
+  SDValue Src = Op.getOperand(0);
+
+  // result = trunc(src);
+  // if (src < 0.0 && src != result)
+  //   result += -1.0.
+
+  SDValue Trunc = DAG.getNode(ISD::FTRUNC, SL, MVT::f64, Src);
+
+  const SDValue Zero = DAG.getConstantFP(0.0, MVT::f64);
+  const SDValue NegOne = DAG.getConstantFP(-1.0, MVT::f64);
+
+  EVT SetCCVT = getSetCCResultType(*DAG.getContext(), MVT::f64);
+
+  SDValue Lt0 = DAG.getSetCC(SL, SetCCVT, Src, Zero, ISD::SETOLT);
+  SDValue NeTrunc = DAG.getSetCC(SL, SetCCVT, Src, Trunc, ISD::SETONE);
+  SDValue And = DAG.getNode(ISD::AND, SL, SetCCVT, Lt0, NeTrunc);
+
+  SDValue Add = DAG.getNode(ISD::SELECT, SL, MVT::f64, And, NegOne, Zero);
+  return DAG.getNode(ISD::FADD, SL, MVT::f64, Trunc, Add);
+}
+
 SDValue AMDGPUTargetLowering::LowerUINT_TO_FP(SDValue Op,
                                                SelectionDAG &DAG) const {
   SDValue S0 = Op.getOperand(0);
diff --git a/lib/Target/R600/AMDGPUISelLowering.h b/lib/Target/R600/AMDGPUISelLowering.h
index 9b54022e15..34e36d886b 100644
--- a/lib/Target/R600/AMDGPUISelLowering.h
+++ b/lib/Target/R600/AMDGPUISelLowering.h
@@ -51,7 +51,11 @@ private:
   SDValue LowerSREM32(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerSREM64(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerUDIVREM(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFCEIL(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFTRUNC(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerFRINT(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFFLOOR(SDValue Op, SelectionDAG &DAG) const;
+
   SDValue LowerUINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
 
   SDValue ExpandSIGN_EXTEND_INREG(SDValue Op,
diff --git a/test/CodeGen/R600/fceil.ll b/test/CodeGen/R600/fceil.ll
index b8b945f46f..b42aefa173 100644
--- a/test/CodeGen/R600/fceil.ll
+++ b/test/CodeGen/R600/fceil.ll
@@ -1,4 +1,5 @@
-; RUN: llc -march=r600 -mcpu=bonaire < %s | FileCheck -check-prefix=CI %s
+; RUN: llc -march=r600 -mcpu=bonaire < %s | FileCheck -check-prefix=CI -check-prefix=FUNC %s
+; RUN: llc -march=r600 -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
 
 declare double @llvm.ceil.f64(double) nounwind readnone
 declare <2 x double> @llvm.ceil.v2f64(<2 x double>) nounwind readnone
@@ -7,15 +8,33 @@ declare <4 x double> @llvm.ceil.v4f64(<4 x double>) nounwind readnone
 declare <8 x double> @llvm.ceil.v8f64(<8 x double>) nounwind readnone
 declare <16 x double> @llvm.ceil.v16f64(<16 x double>) nounwind readnone
 
-; CI-LABEL: @fceil_f64:
-; CI: V_CEIL_F64_e32
+; FUNC-LABEL: @fceil_f64:
+; CI: V_CEIL_F64_e32
+; SI: S_BFE_I32 [[SEXP:s[0-9]+]], {{s[0-9]+}}, 0xb0014
+; SI: S_ADD_I32 s{{[0-9]+}}, [[SEXP]], 0xfffffc01
+; SI: S_LSHR_B64
+; SI: S_NOT_B64
+; SI: S_AND_B64
+; SI: S_AND_B32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80000000
+; SI: CMP_LT_I32
+; SI: CNDMASK_B32
+; SI: CNDMASK_B32
+; SI: CMP_GT_I32
+; SI: CNDMASK_B32
+; SI: CNDMASK_B32
+; SI: CMP_GT_F64
+; SI: CNDMASK_B32
+; SI: CMP_NE_I32
+; SI: CNDMASK_B32
+; SI: CNDMASK_B32
+; SI: V_ADD_F64
 define void @fceil_f64(double addrspace(1)* %out, double %x) {
   %y = call double @llvm.ceil.f64(double %x) nounwind readnone
   store double %y, double addrspace(1)* %out
   ret void
 }
 
-; CI-LABEL: @fceil_v2f64:
+; FUNC-LABEL: @fceil_v2f64:
 ; CI: V_CEIL_F64_e32
 ; CI: V_CEIL_F64_e32
 define void @fceil_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %x) {
@@ -24,7 +43,7 @@ define void @fceil_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %x) {
   ret void
 }
 
-; FIXME-CI-LABEL: @fceil_v3f64:
+; FIXME-FUNC-LABEL: @fceil_v3f64:
 ; FIXME-CI: V_CEIL_F64_e32
 ; FIXME-CI: V_CEIL_F64_e32
 ; FIXME-CI: V_CEIL_F64_e32
@@ -34,7 +53,7 @@ define void @fceil_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %x) {
 ;   ret void
 ; }
 
-; CI-LABEL: @fceil_v4f64:
+; FUNC-LABEL: @fceil_v4f64:
 ; CI: V_CEIL_F64_e32
 ; CI: V_CEIL_F64_e32
 ; CI: V_CEIL_F64_e32
@@ -45,7 +64,7 @@ define void @fceil_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %x) {
   ret void
 }
 
-; CI-LABEL: @fceil_v8f64:
+; FUNC-LABEL: @fceil_v8f64:
 ; CI: V_CEIL_F64_e32
 ; CI: V_CEIL_F64_e32
 ; CI: V_CEIL_F64_e32
@@ -60,7 +79,7 @@ define void @fceil_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %x) {
   ret void
 }
 
-; CI-LABEL: @fceil_v16f64:
+; FUNC-LABEL: @fceil_v16f64:
 ; CI: V_CEIL_F64_e32
 ; CI: V_CEIL_F64_e32
 ; CI: V_CEIL_F64_e32
diff --git a/test/CodeGen/R600/ffloor.ll b/test/CodeGen/R600/ffloor.ll
index 51d2b89615..31c6116988 100644
--- a/test/CodeGen/R600/ffloor.ll
+++ b/test/CodeGen/R600/ffloor.ll
@@ -1,4 +1,5 @@
-; RUN: llc -march=r600 -mcpu=bonaire < %s | FileCheck -check-prefix=CI %s
+; RUN: llc -march=r600 -mcpu=bonaire < %s | FileCheck -check-prefix=CI -check-prefix=FUNC %s
+; RUN: llc -march=r600 -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
 
 declare double @llvm.floor.f64(double) nounwind readnone
 declare <2 x double> @llvm.floor.v2f64(<2 x double>) nounwind readnone
@@ -7,15 +8,34 @@ declare <4 x double> @llvm.floor.v4f64(<4 x double>) nounwind readnone
 declare <8 x double> @llvm.floor.v8f64(<8 x double>) nounwind readnone
 declare <16 x double> @llvm.floor.v16f64(<16 x double>) nounwind readnone
 
-; CI-LABEL: @ffloor_f64:
+; FUNC-LABEL: @ffloor_f64:
 ; CI: V_FLOOR_F64_e32
+
+; SI: S_BFE_I32 [[SEXP:s[0-9]+]], {{s[0-9]+}}, 0xb0014
+; SI: S_ADD_I32 s{{[0-9]+}}, [[SEXP]], 0xfffffc01
+; SI: S_LSHR_B64
+; SI: S_NOT_B64
+; SI: S_AND_B64
+; SI: S_AND_B32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80000000
+; SI: CMP_LT_I32
+; SI: CNDMASK_B32
+; SI: CNDMASK_B32
+; SI: CMP_GT_I32
+; SI: CNDMASK_B32
+; SI: CNDMASK_B32
+; SI: CMP_LT_F64
+; SI: CNDMASK_B32
+; SI: CMP_NE_I32
+; SI: CNDMASK_B32
+; SI: CNDMASK_B32
+; SI: V_ADD_F64
 define void @ffloor_f64(double addrspace(1)* %out, double %x) {
   %y = call double @llvm.floor.f64(double %x) nounwind readnone
   store double %y, double addrspace(1)* %out
   ret void
 }
 
-; CI-LABEL: @ffloor_v2f64:
+; FUNC-LABEL: @ffloor_v2f64:
 ; CI: V_FLOOR_F64_e32
 ; CI: V_FLOOR_F64_e32
 define void @ffloor_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %x) {
@@ -24,7 +44,7 @@ define void @ffloor_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %x) {
   ret void
 }
 
-; FIXME-CI-LABEL: @ffloor_v3f64:
+; FIXME-FUNC-LABEL: @ffloor_v3f64:
 ; FIXME-CI: V_FLOOR_F64_e32
 ; FIXME-CI: V_FLOOR_F64_e32
 ; FIXME-CI: V_FLOOR_F64_e32
@@ -34,7 +54,7 @@ define void @ffloor_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %x) {
 ;   ret void
 ; }
 
-; CI-LABEL: @ffloor_v4f64:
+; FUNC-LABEL: @ffloor_v4f64:
 ; CI: V_FLOOR_F64_e32
 ; CI: V_FLOOR_F64_e32
 ; CI: V_FLOOR_F64_e32
@@ -45,7 +65,7 @@ define void @ffloor_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %x) {
   ret void
 }
 
-; CI-LABEL: @ffloor_v8f64:
+; FUNC-LABEL: @ffloor_v8f64:
 ; CI: V_FLOOR_F64_e32
 ; CI: V_FLOOR_F64_e32
 ; CI: V_FLOOR_F64_e32
@@ -60,7 +80,7 @@ define void @ffloor_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %x) {
   ret void
 }
 
-; CI-LABEL: @ffloor_v16f64:
+; FUNC-LABEL: @ffloor_v16f64:
 ; CI: V_FLOOR_F64_e32
 ; CI: V_FLOOR_F64_e32
 ; CI: V_FLOOR_F64_e32
diff --git a/test/CodeGen/R600/ftrunc.ll b/test/CodeGen/R600/ftrunc.ll
index 6b235ffbd9..3cd1deb921 100644
--- a/test/CodeGen/R600/ftrunc.ll
+++ b/test/CodeGen/R600/ftrunc.ll
@@ -1,4 +1,5 @@
-; RUN: llc -march=r600 -mcpu=bonaire < %s | FileCheck -check-prefix=CI %s
+; RUN: llc -march=r600 -mcpu=bonaire < %s | FileCheck -check-prefix=CI -check-prefix=FUNC %s
+; RUN: llc -march=r600 -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
 
 declare double @llvm.trunc.f64(double) nounwind readnone
 declare <2 x double> @llvm.trunc.v2f64(<2 x double>) nounwind readnone
@@ -7,15 +8,40 @@ declare <4 x double> @llvm.trunc.v4f64(<4 x double>) nounwind readnone
 declare <8 x double> @llvm.trunc.v8f64(<8 x double>) nounwind readnone
 declare <16 x double> @llvm.trunc.v16f64(<16 x double>) nounwind readnone
 
-; CI-LABEL: @ftrunc_f64:
+; FUNC-LABEL: @v_ftrunc_f64:
 ; CI: V_TRUNC_F64_e32
+; SI: V_BFE_I32 {{v[0-9]+}}, {{v[0-9]+}}, 20, 11
+; SI: S_ENDPGM
+define void @v_ftrunc_f64(double addrspace(1)* %out, double addrspace(1)* %in) {
+  %x = load double addrspace(1)* %in, align 8
+  %y = call double @llvm.trunc.f64(double %x) nounwind readnone
+  store double %y, double addrspace(1)* %out, align 8
+  ret void
+}
+
+; FUNC-LABEL: @ftrunc_f64:
+; CI: V_TRUNC_F64_e32
+
+; SI: S_BFE_I32 [[SEXP:s[0-9]+]], {{s[0-9]+}}, 0xb0014
+; SI: S_ADD_I32 s{{[0-9]+}}, [[SEXP]], 0xfffffc01
+; SI: S_LSHR_B64
+; SI: S_NOT_B64
+; SI: S_AND_B64
+; SI: S_AND_B32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80000000
+; SI: CMP_LT_I32
+; SI: CNDMASK_B32
+; SI: CNDMASK_B32
+; SI: CMP_GT_I32
+; SI: CNDMASK_B32
+; SI: CNDMASK_B32
+; SI: S_ENDPGM
 define void @ftrunc_f64(double addrspace(1)* %out, double %x) {
   %y = call double @llvm.trunc.f64(double %x) nounwind readnone
   store double %y, double addrspace(1)* %out
   ret void
 }
 
-; CI-LABEL: @ftrunc_v2f64:
+; FUNC-LABEL: @ftrunc_v2f64:
 ; CI: V_TRUNC_F64_e32
 ; CI: V_TRUNC_F64_e32
 define void @ftrunc_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %x) {
@@ -24,7 +50,7 @@ define void @ftrunc_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %x) {
   ret void
 }
 
-; FIXME-CI-LABEL: @ftrunc_v3f64:
+; FIXME-FUNC-LABEL: @ftrunc_v3f64:
 ; FIXME-CI: V_TRUNC_F64_e32
 ; FIXME-CI: V_TRUNC_F64_e32
 ; FIXME-CI: V_TRUNC_F64_e32
@@ -34,7 +60,7 @@ define void @ftrunc_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %x) {
 ;   ret void
 ; }
 
-; CI-LABEL: @ftrunc_v4f64:
+; FUNC-LABEL: @ftrunc_v4f64:
 ; CI: V_TRUNC_F64_e32
 ; CI: V_TRUNC_F64_e32
 ; CI: V_TRUNC_F64_e32
@@ -45,7 +71,7 @@ define void @ftrunc_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %x) {
   ret void
 }
 
-; CI-LABEL: @ftrunc_v8f64:
+; FUNC-LABEL: @ftrunc_v8f64:
 ; CI: V_TRUNC_F64_e32
 ; CI: V_TRUNC_F64_e32
 ; CI: V_TRUNC_F64_e32
@@ -60,7 +86,7 @@ define void @ftrunc_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %x) {
   ret void
 }
 
-; CI-LABEL: @ftrunc_v16f64:
+; FUNC-LABEL: @ftrunc_v16f64:
 ; CI: V_TRUNC_F64_e32
 ; CI: V_TRUNC_F64_e32
 ; CI: V_TRUNC_F64_e32