[PATCHES] R600/SI: New V_FRACT fix, intrinsic for S_FLBIT_I32, and more

Tom Stellard tom at stellard.net
Mon Mar 23 08:25:19 PDT 2015


LGTM.

On Fri, Mar 13, 2015 at 07:13:11PM +0100, Marek Olšák wrote:
> New patches are attached. I only added -enable-unsafe-fp-math to the
> tests, so that folding NEG modifiers can be tested. Please review.
> 
> Marek
> 
> On Thu, Mar 12, 2015 at 4:48 PM, Marek Olšák <maraeo at gmail.com> wrote:
> > So can I consider that patches 2 & 3 have your "LGTM"?
> >
> > Marek
> >
> > On Tue, Mar 10, 2015 at 11:28 PM, Matt Arsenault
> > <Matthew.Arsenault at amd.com> wrote:
> >> On 03/10/2015 03:21 PM, Marek Olšák wrote:
> >>>
> >>> If we didn't have to deal with fsub, only one pattern would be needed.
> >>> A possible solution is to expand fsub, so that it's translated into
> >>> (v_add_f32 a, -b), and then convert it to v_sub_f32 in the shrinking
> >>> pass if it's possible.
> >>>
> >>> The hardware internally expands (v_sub_f32 a, b) into (v_add_f32 a, -b)
> >>> anyway.
> >>>
> >>> Marek
> >>
> >> When I removed the fsub pseudo, I was originally going to handle it with the
> >> expanded pattern, but then I noticed the expansion for fsub did the same
> >> thing anyway and was slightly easier. Another option might be to do that

> From bd77870e24d575ed59ea591fb2d7c83ac98cadf0 Mon Sep 17 00:00:00 2001
> From: =?UTF-8?q?Marek=20Ol=C5=A1=C3=A1k?= <marek.olsak at amd.com>
> Date: Sun, 1 Mar 2015 23:07:48 +0100
> Subject: [PATCH 1/2] R600/SI: Expand fract to floor, then only select V_FRACT
>  on CI
> 
> V_FRACT is buggy on SI.
> 
> R600-specific code is left intact.
> 
> v2: drop the multiclass, use complex VOP3 patterns
> ---
>  lib/Target/R600/AMDGPUISelLowering.cpp |  3 ---
>  lib/Target/R600/R600ISelLowering.cpp   |  4 +++
>  lib/Target/R600/SIISelLowering.cpp     |  6 +++++
>  lib/Target/R600/SIInstructions.td      | 22 ++++++++++++++++
>  test/CodeGen/R600/llvm.AMDGPU.fract.ll | 47 ++++++++++++++++++++++++++++++----
>  5 files changed, 74 insertions(+), 8 deletions(-)
> 
> diff --git a/lib/Target/R600/AMDGPUISelLowering.cpp b/lib/Target/R600/AMDGPUISelLowering.cpp
> index 4707279..62a33fa 100644
> --- a/lib/Target/R600/AMDGPUISelLowering.cpp
> +++ b/lib/Target/R600/AMDGPUISelLowering.cpp
> @@ -885,9 +885,6 @@ SDValue AMDGPUTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
>        return LowerIntrinsicIABS(Op, DAG);
>      case AMDGPUIntrinsic::AMDGPU_lrp:
>        return LowerIntrinsicLRP(Op, DAG);
> -    case AMDGPUIntrinsic::AMDGPU_fract:
> -    case AMDGPUIntrinsic::AMDIL_fraction: // Legacy name.
> -      return DAG.getNode(AMDGPUISD::FRACT, DL, VT, Op.getOperand(1));
>  
>      case AMDGPUIntrinsic::AMDGPU_clamp:
>      case AMDGPUIntrinsic::AMDIL_clamp: // Legacy name.
> diff --git a/lib/Target/R600/R600ISelLowering.cpp b/lib/Target/R600/R600ISelLowering.cpp
> index d4f3145..06b7217 100644
> --- a/lib/Target/R600/R600ISelLowering.cpp
> +++ b/lib/Target/R600/R600ISelLowering.cpp
> @@ -837,6 +837,10 @@ SDValue R600TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const
>      case Intrinsic::AMDGPU_rsq:
>        // XXX - I'm assuming SI's RSQ_LEGACY matches R600's behavior.
>        return DAG.getNode(AMDGPUISD::RSQ_LEGACY, DL, VT, Op.getOperand(1));
> +
> +    case AMDGPUIntrinsic::AMDGPU_fract:
> +    case AMDGPUIntrinsic::AMDIL_fraction: // Legacy name.
> +      return DAG.getNode(AMDGPUISD::FRACT, DL, VT, Op.getOperand(1));
>      }
>      // break out of case ISD::INTRINSIC_WO_CHAIN in switch(Op.getOpcode())
>      break;
> diff --git a/lib/Target/R600/SIISelLowering.cpp b/lib/Target/R600/SIISelLowering.cpp
> index af38c94..80074c0 100644
> --- a/lib/Target/R600/SIISelLowering.cpp
> +++ b/lib/Target/R600/SIISelLowering.cpp
> @@ -928,6 +928,12 @@ SDValue SITargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
>                         Op.getOperand(1),
>                         Op.getOperand(2),
>                         Op.getOperand(3));
> +
> +  case AMDGPUIntrinsic::AMDGPU_fract:
> +  case AMDGPUIntrinsic::AMDIL_fraction: // Legacy name.
> +    return DAG.getNode(ISD::FSUB, DL, VT, Op.getOperand(1),
> +                       DAG.getNode(ISD::FFLOOR, DL, VT, Op.getOperand(1)));
> +
>    default:
>      return AMDGPUTargetLowering::LowerOperation(Op, DAG);
>    }
> diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
> index 5f02a31..1c9e11c 100644
> --- a/lib/Target/R600/SIInstructions.td
> +++ b/lib/Target/R600/SIInstructions.td
> @@ -3344,6 +3344,28 @@ def : Pat <
>    (V_CNDMASK_B32_e64 $src0, $src1, $src2)
>  >;
>  
> +//===----------------------------------------------------------------------===//
> +// Fract Patterns
> +//===----------------------------------------------------------------------===//
> +
> +let Predicates = [isCI] in {
> +
> +// Convert (x - floor(x)) to fract(x)
> +def : Pat <
> +  (f32 (fsub (f32 (VOP3Mods f32:$x, i32:$mods)),
> +             (f32 (ffloor (f32 (VOP3Mods f32:$x, i32:$mods)))))),
> +  (V_FRACT_F32_e64 $mods, $x, DSTCLAMP.NONE, DSTOMOD.NONE)
> +>;
> +
> +// Convert (x + (-floor(x))) to fract(x)
> +def : Pat <
> +  (f64 (fadd (f64 (VOP3Mods f64:$x, i32:$mods)),
> +             (f64 (fneg (f64 (ffloor (f64 (VOP3Mods f64:$x, i32:$mods)))))))),
> +  (V_FRACT_F64_e64 $mods, $x, DSTCLAMP.NONE, DSTOMOD.NONE)
> +>;
> +
> +} // End Predicates = [isCI]
> +
>  //============================================================================//
>  // Miscellaneous Optimization Patterns
>  //============================================================================//
> diff --git a/test/CodeGen/R600/llvm.AMDGPU.fract.ll b/test/CodeGen/R600/llvm.AMDGPU.fract.ll
> index f4cf7fc..7501b4b 100644
> --- a/test/CodeGen/R600/llvm.AMDGPU.fract.ll
> +++ b/test/CodeGen/R600/llvm.AMDGPU.fract.ll
> @@ -1,14 +1,19 @@
> -; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
> -; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
> -; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
> +; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GCN -check-prefix=SI -check-prefix=FUNC %s
> +; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GCN -check-prefix=CI -check-prefix=FUNC %s
> +; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GCN -check-prefix=CI -check-prefix=FUNC %s
> +; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
>  
> +declare float @llvm.fabs.f32(float  %Val)
>  declare float @llvm.AMDGPU.fract.f32(float) nounwind readnone
>  
>  ; Legacy name
>  declare float @llvm.AMDIL.fraction.f32(float) nounwind readnone
>  
>  ; FUNC-LABEL: {{^}}fract_f32:
> -; SI: v_fract_f32
> +; CI: v_fract_f32_e32 [[RESULT:v[0-9]+]], [[INPUT:v[0-9]+]]
> +; SI: v_floor_f32_e32 [[FLR:v[0-9]+]], [[INPUT:v[0-9]+]]
> +; SI: v_subrev_f32_e32 [[RESULT:v[0-9]+]], [[FLR]], [[INPUT]]
> +; GCN: buffer_store_dword [[RESULT]]
>  ; EG: FRACT
>  define void @fract_f32(float addrspace(1)* %out, float addrspace(1)* %src) nounwind {
>    %val = load float, float addrspace(1)* %src, align 4
> @@ -18,7 +23,10 @@ define void @fract_f32(float addrspace(1)* %out, float addrspace(1)* %src) nounw
>  }
>  
>  ; FUNC-LABEL: {{^}}fract_f32_legacy_amdil:
> -; SI: v_fract_f32
> +; CI: v_fract_f32_e32 [[RESULT:v[0-9]+]], [[INPUT:v[0-9]+]]
> +; SI: v_floor_f32_e32 [[FLR:v[0-9]+]], [[INPUT:v[0-9]+]]
> +; SI: v_subrev_f32_e32 [[RESULT:v[0-9]+]], [[FLR]], [[INPUT]]
> +; GCN: buffer_store_dword [[RESULT]]
>  ; EG: FRACT
>  define void @fract_f32_legacy_amdil(float addrspace(1)* %out, float addrspace(1)* %src) nounwind {
>    %val = load float, float addrspace(1)* %src, align 4
> @@ -26,3 +34,32 @@ define void @fract_f32_legacy_amdil(float addrspace(1)* %out, float addrspace(1)
>    store float %fract, float addrspace(1)* %out, align 4
>    ret void
>  }
> +
> +; FUNC-LABEL: {{^}}fract_f32_neg:
> +; CI: v_fract_f32_e64 [[RESULT:v[0-9]+]], -[[INPUT:v[0-9]+]]
> +; SI: v_floor_f32_e64 [[FLR:v[0-9]+]], -[[INPUT:v[0-9]+]]
> +; SI: v_sub_f32_e64 [[RESULT:v[0-9]+]], -[[INPUT]], [[FLR]]
> +; GCN: buffer_store_dword [[RESULT]]
> +; EG: FRACT
> +define void @fract_f32_neg(float addrspace(1)* %out, float addrspace(1)* %src) nounwind {
> +  %val = load float, float addrspace(1)* %src, align 4
> +  %neg = fsub float 0.0, %val
> +  %fract = call float @llvm.AMDGPU.fract.f32(float %neg) nounwind readnone
> +  store float %fract, float addrspace(1)* %out, align 4
> +  ret void
> +}
> +
> +; FUNC-LABEL: {{^}}fract_f32_neg_abs:
> +; CI: v_fract_f32_e64 [[RESULT:v[0-9]+]], -|[[INPUT:v[0-9]+]]|
> +; SI: v_floor_f32_e64 [[FLR:v[0-9]+]], -|[[INPUT:v[0-9]+]]|
> +; SI: v_sub_f32_e64 [[RESULT:v[0-9]+]], -|[[INPUT]]|, [[FLR]]
> +; GCN: buffer_store_dword [[RESULT]]
> +; EG: FRACT
> +define void @fract_f32_neg_abs(float addrspace(1)* %out, float addrspace(1)* %src) nounwind {
> +  %val = load float, float addrspace(1)* %src, align 4
> +  %abs = call float @llvm.fabs.f32(float %val)
> +  %neg = fsub float 0.0, %abs
> +  %fract = call float @llvm.AMDGPU.fract.f32(float %neg) nounwind readnone
> +  store float %fract, float addrspace(1)* %out, align 4
> +  ret void
> +}
> -- 
> 2.1.0
> 

> From 7400ce890a5d648bc347c1bf7eb996d2f8189fe2 Mon Sep 17 00:00:00 2001
> From: =?UTF-8?q?Marek=20Ol=C5=A1=C3=A1k?= <marek.olsak at amd.com>
> Date: Sun, 1 Mar 2015 23:39:34 +0100
> Subject: [PATCH 2/2] R600/SI: Use V_FRACT_F64 for faster 64-bit floor on SI
> 
> Other f64 opcodes not supported on SI can be lowered in a similar way.
> 
> v2: use complex VOP3 patterns
> ---
>  lib/Target/R600/SIISelLowering.cpp         |  2 +-
>  lib/Target/R600/SIInstrInfo.cpp            | 20 ++++++++++
>  lib/Target/R600/SIInstrInfo.td             |  1 +
>  lib/Target/R600/SIInstructions.td          | 51 ++++++++++++++++++++++++
>  test/CodeGen/R600/ffloor.f64.ll            | 63 ++++++++++++++++++++----------
>  test/CodeGen/R600/llvm.AMDGPU.fract.f64.ll | 60 ++++++++++++++++++++++++++++
>  6 files changed, 175 insertions(+), 22 deletions(-)
>  create mode 100644 test/CodeGen/R600/llvm.AMDGPU.fract.f64.ll
> 
> diff --git a/lib/Target/R600/SIISelLowering.cpp b/lib/Target/R600/SIISelLowering.cpp
> index 80074c0..48f5ede 100644
> --- a/lib/Target/R600/SIISelLowering.cpp
> +++ b/lib/Target/R600/SIISelLowering.cpp
> @@ -202,10 +202,10 @@ SITargetLowering::SITargetLowering(TargetMachine &TM,
>    if (Subtarget->getGeneration() >= AMDGPUSubtarget::SEA_ISLANDS) {
>      setOperationAction(ISD::FTRUNC, MVT::f64, Legal);
>      setOperationAction(ISD::FCEIL, MVT::f64, Legal);
> -    setOperationAction(ISD::FFLOOR, MVT::f64, Legal);
>      setOperationAction(ISD::FRINT, MVT::f64, Legal);
>    }
>  
> +  setOperationAction(ISD::FFLOOR, MVT::f64, Legal);
>    setOperationAction(ISD::FDIV, MVT::f32, Custom);
>    setOperationAction(ISD::FDIV, MVT::f64, Custom);
>  
> diff --git a/lib/Target/R600/SIInstrInfo.cpp b/lib/Target/R600/SIInstrInfo.cpp
> index 95334c3..61a5057 100644
> --- a/lib/Target/R600/SIInstrInfo.cpp
> +++ b/lib/Target/R600/SIInstrInfo.cpp
> @@ -717,6 +717,26 @@ bool SIInstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MI) const {
>      MI->eraseFromParent();
>      break;
>    }
> +
> +  case AMDGPU::V_CNDMASK_B64_PSEUDO: {
> +    unsigned Dst = MI->getOperand(0).getReg();
> +    unsigned DstLo = RI.getSubReg(Dst, AMDGPU::sub0);
> +    unsigned DstHi = RI.getSubReg(Dst, AMDGPU::sub1);
> +    unsigned Src0 = MI->getOperand(1).getReg();
> +    unsigned Src1 = MI->getOperand(2).getReg();
> +    const MachineOperand &SrcCond = MI->getOperand(3);
> +
> +    BuildMI(MBB, MI, DL, get(AMDGPU::V_CNDMASK_B32_e64), DstLo)
> +        .addReg(RI.getSubReg(Src0, AMDGPU::sub0))
> +        .addReg(RI.getSubReg(Src1, AMDGPU::sub0))
> +        .addOperand(SrcCond);
> +    BuildMI(MBB, MI, DL, get(AMDGPU::V_CNDMASK_B32_e64), DstHi)
> +        .addReg(RI.getSubReg(Src0, AMDGPU::sub1))
> +        .addReg(RI.getSubReg(Src1, AMDGPU::sub1))
> +        .addOperand(SrcCond);
> +    MI->eraseFromParent();
> +    break;
> +  }
>    }
>    return true;
>  }
> diff --git a/lib/Target/R600/SIInstrInfo.td b/lib/Target/R600/SIInstrInfo.td
> index b557b06..902c17d 100644
> --- a/lib/Target/R600/SIInstrInfo.td
> +++ b/lib/Target/R600/SIInstrInfo.td
> @@ -320,6 +320,7 @@ def SIOperand {
>  
>  def SRCMODS {
>    int NONE = 0;
> +  int NEG = 1;
>  }
>  
>  def DSTCLAMP {
> diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
> index 1c9e11c..21f4efd 100644
> --- a/lib/Target/R600/SIInstructions.td
> +++ b/lib/Target/R600/SIInstructions.td
> @@ -28,6 +28,8 @@ def SendMsgImm : Operand<i32> {
>  
>  def isGCN : Predicate<"Subtarget->getGeneration() "
>                        ">= AMDGPUSubtarget::SOUTHERN_ISLANDS">;
> +def isSI : Predicate<"Subtarget->getGeneration() "
> +                      "== AMDGPUSubtarget::SOUTHERN_ISLANDS">;
>  def isSICI : Predicate<
>    "Subtarget->getGeneration() == AMDGPUSubtarget::SOUTHERN_ISLANDS ||"
>    "Subtarget->getGeneration() == AMDGPUSubtarget::SEA_ISLANDS"
> @@ -1902,6 +1904,11 @@ defm V_ASHRREV_I64 : VOP3Inst <vop3<0, 0x291>, "v_ashrrev_i64",
>  //===----------------------------------------------------------------------===//
>  let isCodeGenOnly = 1, isPseudo = 1 in {
>  
> +// For use in patterns
> +def V_CNDMASK_B64_PSEUDO : VOP3Common <(outs VReg_64:$dst),
> +  (ins VSrc_64:$src0, VSrc_64:$src1, SSrc_64:$src2), "", []
> +>;
> +
>  let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in {
>  // 64-bit vector move instruction.  This is mainly used by the SIFoldOperands
>  // pass to enable folding of inline immediates.
> @@ -3348,6 +3355,50 @@ def : Pat <
>  // Fract Patterns
>  //===----------------------------------------------------------------------===//
>  
> +let Predicates = [isSI] in {
> +
> +// V_FRACT is buggy on SI, so the F32 version is never used and (x-floor(x)) is
> +// used instead. However, SI doesn't have V_FLOOR_F64, so the most efficient
> +// way to implement it is using V_FRACT_F64.
> +// The workaround for the V_FRACT bug is:
> +//    fract(x) = isnan(x) ? x : min(V_FRACT(x), 0.99999999999999999)
> +
> +// Convert (x + (-floor(x)) to fract(x)
> +def : Pat <
> +  (f64 (fadd (f64 (VOP3Mods f64:$x, i32:$mods)),
> +             (f64 (fneg (f64 (ffloor (f64 (VOP3Mods f64:$x, i32:$mods)))))))),
> +  (V_CNDMASK_B64_PSEUDO
> +      $x,
> +      (V_MIN_F64
> +          SRCMODS.NONE,
> +          (V_FRACT_F64_e64 $mods, $x, DSTCLAMP.NONE, DSTOMOD.NONE),
> +          SRCMODS.NONE,
> +          (V_MOV_B64_PSEUDO 0x3fefffffffffffff),
> +          DSTCLAMP.NONE, DSTOMOD.NONE),
> +      (V_CMP_CLASS_F64_e64 SRCMODS.NONE, $x, 3/*NaN*/))
> +>;
> +
> +// Convert floor(x) to (x - fract(x))
> +def : Pat <
> +  (f64 (ffloor (f64 (VOP3Mods f64:$x, i32:$mods)))),
> +  (V_ADD_F64
> +      $mods,
> +      $x,
> +      SRCMODS.NEG,
> +      (V_CNDMASK_B64_PSEUDO
> +         $x,
> +         (V_MIN_F64
> +             SRCMODS.NONE,
> +             (V_FRACT_F64_e64 $mods, $x, DSTCLAMP.NONE, DSTOMOD.NONE),
> +             SRCMODS.NONE,
> +             (V_MOV_B64_PSEUDO 0x3fefffffffffffff),
> +             DSTCLAMP.NONE, DSTOMOD.NONE),
> +         (V_CMP_CLASS_F64_e64 SRCMODS.NONE, $x, 3/*NaN*/)),
> +      DSTCLAMP.NONE, DSTOMOD.NONE)
> +>;
> +
> +} // End Predicates = [isSI]
> +
>  let Predicates = [isCI] in {
>  
>  // Convert (x - floor(x)) to fract(x)
> diff --git a/test/CodeGen/R600/ffloor.f64.ll b/test/CodeGen/R600/ffloor.f64.ll
> index 745ad3b..45f8382 100644
> --- a/test/CodeGen/R600/ffloor.f64.ll
> +++ b/test/CodeGen/R600/ffloor.f64.ll
> @@ -1,7 +1,8 @@
> -; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
> -; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefix=CI -check-prefix=FUNC %s
> -; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=CI -check-prefix=FUNC %s
> +; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
> +; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=CI -check-prefix=FUNC %s
> +; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=CI -check-prefix=FUNC %s
>  
> +declare double @llvm.fabs.f64(double %Val)
>  declare double @llvm.floor.f64(double) nounwind readnone
>  declare <2 x double> @llvm.floor.v2f64(<2 x double>) nounwind readnone
>  declare <3 x double> @llvm.floor.v3f64(<3 x double>) nounwind readnone
> @@ -11,24 +12,11 @@ declare <16 x double> @llvm.floor.v16f64(<16 x double>) nounwind readnone
>  
>  ; FUNC-LABEL: {{^}}ffloor_f64:
>  ; CI: v_floor_f64_e32
> -
> -; SI: s_bfe_u32 [[SEXP:s[0-9]+]], {{s[0-9]+}}, 0xb0014
> -; SI: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80000000
> -; SI: s_add_i32 s{{[0-9]+}}, [[SEXP]], 0xfffffc01
> -; SI: s_lshr_b64
> -; SI: s_not_b64
> -; SI: s_and_b64
> -; SI: cmp_lt_i32
> -; SI: cndmask_b32
> -; SI: cndmask_b32
> -; SI: cmp_gt_i32
> -; SI: cndmask_b32
> -; SI: cndmask_b32
> -; SI-DAG: v_cmp_lt_f64
> -; SI-DAG: v_cmp_lg_f64
> -; SI-DAG: s_and_b64
> -; SI-DAG: v_cndmask_b32
> -; SI-DAG: v_cndmask_b32
> +; SI: v_fract_f64_e32
> +; SI: v_min_f64
> +; SI: v_cmp_class_f64_e64
> +; SI: v_cndmask_b32_e64
> +; SI: v_cndmask_b32_e64
>  ; SI: v_add_f64
>  ; SI: s_endpgm
>  define void @ffloor_f64(double addrspace(1)* %out, double %x) {
> @@ -37,6 +25,39 @@ define void @ffloor_f64(double addrspace(1)* %out, double %x) {
>    ret void
>  }
>  
> +; FUNC-LABEL: {{^}}ffloor_f64_neg:
> +; CI: v_floor_f64_e64
> +; SI: v_fract_f64_e64 {{v[[0-9]+:[0-9]+]}}, -[[INPUT:s[[0-9]+:[0-9]+]]]
> +; SI: v_min_f64
> +; SI: v_cmp_class_f64_e64
> +; SI: v_cndmask_b32_e64
> +; SI: v_cndmask_b32_e64
> +; SI: v_add_f64 {{v[[0-9]+:[0-9]+]}}, -[[INPUT]]
> +; SI: s_endpgm
> +define void @ffloor_f64_neg(double addrspace(1)* %out, double %x) {
> +  %neg = fsub double 0.0, %x
> +  %y = call double @llvm.floor.f64(double %neg) nounwind readnone
> +  store double %y, double addrspace(1)* %out
> +  ret void
> +}
> +
> +; FUNC-LABEL: {{^}}ffloor_f64_neg_abs:
> +; CI: v_floor_f64_e64
> +; SI: v_fract_f64_e64 {{v[[0-9]+:[0-9]+]}}, -|[[INPUT:s[[0-9]+:[0-9]+]]]|
> +; SI: v_min_f64
> +; SI: v_cmp_class_f64_e64
> +; SI: v_cndmask_b32_e64
> +; SI: v_cndmask_b32_e64
> +; SI: v_add_f64 {{v[[0-9]+:[0-9]+]}}, -|[[INPUT]]|
> +; SI: s_endpgm
> +define void @ffloor_f64_neg_abs(double addrspace(1)* %out, double %x) {
> +  %abs = call double @llvm.fabs.f64(double %x)
> +  %neg = fsub double 0.0, %abs
> +  %y = call double @llvm.floor.f64(double %neg) nounwind readnone
> +  store double %y, double addrspace(1)* %out
> +  ret void
> +}
> +
>  ; FUNC-LABEL: {{^}}ffloor_v2f64:
>  ; CI: v_floor_f64_e32
>  ; CI: v_floor_f64_e32
> diff --git a/test/CodeGen/R600/llvm.AMDGPU.fract.f64.ll b/test/CodeGen/R600/llvm.AMDGPU.fract.f64.ll
> new file mode 100644
> index 0000000..e098dd3
> --- /dev/null
> +++ b/test/CodeGen/R600/llvm.AMDGPU.fract.f64.ll
> @@ -0,0 +1,60 @@
> +; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GCN -check-prefix=SI -check-prefix=FUNC %s
> +; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GCN -check-prefix=CI -check-prefix=FUNC %s
> +; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GCN -check-prefix=CI -check-prefix=FUNC %s
> +
> +declare double @llvm.fabs.f64(double %Val)
> +declare double @llvm.AMDGPU.fract.f64(double) nounwind readnone
> +
> +; FUNC-LABEL: {{^}}fract_f64:
> +; GCN: v_fract_f64_e32 [[FRC:v\[[0-9]+:[0-9]+\]]], v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]]
> +; SI: v_mov_b32_e32 v[[UPLO:[0-9]+]], -1
> +; SI: v_mov_b32_e32 v[[UPHI:[0-9]+]], 0x3fefffff
> +; SI: v_min_f64 v{{\[}}[[MINLO:[0-9]+]]:[[MINHI:[0-9]+]]], v{{\[}}[[UPLO]]:[[UPHI]]], [[FRC]]
> +; SI: v_cmp_class_f64_e64 [[COND:s\[[0-9]+:[0-9]+\]]], v{{\[}}[[LO]]:[[HI]]], 3
> +; SI: v_cndmask_b32_e64 v[[RESLO:[0-9]+]], v[[LO]], v[[MINLO]], [[COND]]
> +; SI: v_cndmask_b32_e64 v[[RESHI:[0-9]+]], v[[HI]], v[[MINHI]], [[COND]]
> +; SI: buffer_store_dwordx2 v{{\[}}[[RESLO]]:[[RESHI]]]
> +; CI: buffer_store_dwordx2 [[FRC]]
> +define void @fract_f64(double addrspace(1)* %out, double addrspace(1)* %src) nounwind {
> +  %val = load double, double addrspace(1)* %src, align 4
> +  %fract = call double @llvm.AMDGPU.fract.f64(double %val) nounwind readnone
> +  store double %fract, double addrspace(1)* %out, align 4
> +  ret void
> +}
> +
> +; FUNC-LABEL: {{^}}fract_f64_neg:
> +; GCN: v_fract_f64_e64 [[FRC:v\[[0-9]+:[0-9]+\]]], -v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]]
> +; SI: v_mov_b32_e32 v[[UPLO:[0-9]+]], -1
> +; SI: v_mov_b32_e32 v[[UPHI:[0-9]+]], 0x3fefffff
> +; SI: v_min_f64 v{{\[}}[[MINLO:[0-9]+]]:[[MINHI:[0-9]+]]], v{{\[}}[[UPLO]]:[[UPHI]]], [[FRC]]
> +; SI: v_cmp_class_f64_e64 [[COND:s\[[0-9]+:[0-9]+\]]], v{{\[}}[[LO]]:[[HI]]], 3
> +; SI: v_cndmask_b32_e64 v[[RESLO:[0-9]+]], v[[LO]], v[[MINLO]], [[COND]]
> +; SI: v_cndmask_b32_e64 v[[RESHI:[0-9]+]], v[[HI]], v[[MINHI]], [[COND]]
> +; SI: buffer_store_dwordx2 v{{\[}}[[RESLO]]:[[RESHI]]]
> +; CI: buffer_store_dwordx2 [[FRC]]
> +define void @fract_f64_neg(double addrspace(1)* %out, double addrspace(1)* %src) nounwind {
> +  %val = load double, double addrspace(1)* %src, align 4
> +  %neg = fsub double 0.0, %val
> +  %fract = call double @llvm.AMDGPU.fract.f64(double %neg) nounwind readnone
> +  store double %fract, double addrspace(1)* %out, align 4
> +  ret void
> +}
> +
> +; FUNC-LABEL: {{^}}fract_f64_neg_abs:
> +; GCN: v_fract_f64_e64 [[FRC:v\[[0-9]+:[0-9]+\]]], -|v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]]|
> +; SI: v_mov_b32_e32 v[[UPLO:[0-9]+]], -1
> +; SI: v_mov_b32_e32 v[[UPHI:[0-9]+]], 0x3fefffff
> +; SI: v_min_f64 v{{\[}}[[MINLO:[0-9]+]]:[[MINHI:[0-9]+]]], v{{\[}}[[UPLO]]:[[UPHI]]], [[FRC]]
> +; SI: v_cmp_class_f64_e64 [[COND:s\[[0-9]+:[0-9]+\]]], v{{\[}}[[LO]]:[[HI]]], 3
> +; SI: v_cndmask_b32_e64 v[[RESLO:[0-9]+]], v[[LO]], v[[MINLO]], [[COND]]
> +; SI: v_cndmask_b32_e64 v[[RESHI:[0-9]+]], v[[HI]], v[[MINHI]], [[COND]]
> +; SI: buffer_store_dwordx2 v{{\[}}[[RESLO]]:[[RESHI]]]
> +; CI: buffer_store_dwordx2 [[FRC]]
> +define void @fract_f64_neg_abs(double addrspace(1)* %out, double addrspace(1)* %src) nounwind {
> +  %val = load double, double addrspace(1)* %src, align 4
> +  %abs = call double @llvm.fabs.f64(double %val)
> +  %neg = fsub double 0.0, %abs
> +  %fract = call double @llvm.AMDGPU.fract.f64(double %neg) nounwind readnone
> +  store double %fract, double addrspace(1)* %out, align 4
> +  ret void
> +}
> -- 
> 2.1.0
> 





More information about the llvm-commits mailing list