[llvm] r238578 - Add support for VSX FMA single-precision instructions to the PPC back end
Nemanja Ivanovic
nemanja.i.ibm at gmail.com
Fri May 29 10:13:25 PDT 2015
Author: nemanjai
Date: Fri May 29 12:13:25 2015
New Revision: 238578
URL: http://llvm.org/viewvc/llvm-project?rev=238578&view=rev
Log:
Add support for VSX FMA single-precision instructions to the PPC back end
This patch corresponds to review:
http://reviews.llvm.org/D9941
It adds the various FMA instructions introduced in the version 2.07 of
the ISA along with the testing for them. These are operations on single
precision scalar values in VSX registers.
Added:
llvm/trunk/test/CodeGen/PowerPC/vsx-fma-sp.ll
Modified:
llvm/trunk/lib/Target/PowerPC/PPCFastISel.cpp
llvm/trunk/lib/Target/PowerPC/PPCInstrVSX.td
llvm/trunk/test/CodeGen/PowerPC/fma.ll
llvm/trunk/test/MC/Disassembler/PowerPC/vsx.txt
llvm/trunk/test/MC/PowerPC/vsx.s
Modified: llvm/trunk/lib/Target/PowerPC/PPCFastISel.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPCFastISel.cpp?rev=238578&r1=238577&r2=238578&view=diff
==============================================================================
--- llvm/trunk/lib/Target/PowerPC/PPCFastISel.cpp (original)
+++ llvm/trunk/lib/Target/PowerPC/PPCFastISel.cpp Fri May 29 12:13:25 2015
@@ -148,6 +148,9 @@ class PPCFastISel final : public FastISe
bool isVSFRCRegister(unsigned Register) const {
return MRI.getRegClass(Register)->getID() == PPC::VSFRCRegClassID;
}
+ bool isVSSRCRegister(unsigned Register) const {
+ return MRI.getRegClass(Register)->getID() == PPC::VSSRCRegClassID;
+ }
bool PPCEmitCmp(const Value *Src1Value, const Value *Src2Value,
bool isZExt, unsigned DestReg);
bool PPCEmitLoad(MVT VT, unsigned &ResultReg, Address &Addr,
@@ -503,8 +506,11 @@ bool PPCFastISel::PPCEmitLoad(MVT VT, un
// If this is a potential VSX load with an offset of 0, a VSX indexed load can
// be used.
+ bool IsVSSRC = (ResultReg != 0) && isVSSRCRegister(ResultReg);
bool IsVSFRC = (ResultReg != 0) && isVSFRCRegister(ResultReg);
- if (IsVSFRC && (Opc == PPC::LFD) &&
+ bool Is32VSXLoad = IsVSSRC && Opc == PPC::LFS;
+ bool Is64VSXLoad = IsVSSRC && Opc == PPC::LFD;
+ if ((Is32VSXLoad || Is64VSXLoad) &&
(Addr.BaseType != Address::FrameIndexBase) && UseOffset &&
(Addr.Offset == 0)) {
UseOffset = false;
@@ -518,7 +524,7 @@ bool PPCFastISel::PPCEmitLoad(MVT VT, un
// into a RegBase.
if (Addr.BaseType == Address::FrameIndexBase) {
// VSX only provides an indexed load.
- if (IsVSFRC && Opc == PPC::LFD) return false;
+ if (Is32VSXLoad || Is64VSXLoad) return false;
MachineMemOperand *MMO =
FuncInfo.MF->getMachineMemOperand(
@@ -532,7 +538,7 @@ bool PPCFastISel::PPCEmitLoad(MVT VT, un
// Base reg with offset in range.
} else if (UseOffset) {
// VSX only provides an indexed load.
- if (IsVSFRC && Opc == PPC::LFD) return false;
+ if (Is32VSXLoad || Is64VSXLoad) return false;
BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg)
.addImm(Addr.Offset).addReg(Addr.Base.Reg);
@@ -555,7 +561,7 @@ bool PPCFastISel::PPCEmitLoad(MVT VT, un
case PPC::LWA: Opc = PPC::LWAX; break;
case PPC::LWA_32: Opc = PPC::LWAX_32; break;
case PPC::LD: Opc = PPC::LDX; break;
- case PPC::LFS: Opc = PPC::LFSX; break;
+ case PPC::LFS: Opc = IsVSSRC ? PPC::LXSSPX : PPC::LFSX; break;
case PPC::LFD: Opc = IsVSFRC ? PPC::LXSDX : PPC::LFDX; break;
}
BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg)
@@ -636,9 +642,12 @@ bool PPCFastISel::PPCEmitStore(MVT VT, u
// If this is a potential VSX store with an offset of 0, a VSX indexed store
// can be used.
+ bool IsVSSRC = isVSSRCRegister(SrcReg);
bool IsVSFRC = isVSFRCRegister(SrcReg);
- if (IsVSFRC && (Opc == PPC::STFD) &&
- (Addr.BaseType != Address::FrameIndexBase) && UseOffset &&
+ bool Is32VSXStore = IsVSSRC && Opc == PPC::STFS;
+ bool Is64VSXStore = IsVSFRC && Opc == PPC::STFD;
+ if ((Is32VSXStore || Is64VSXStore) &&
+ (Addr.BaseType != Address::FrameIndexBase) && UseOffset &&
(Addr.Offset == 0)) {
UseOffset = false;
}
@@ -648,7 +657,7 @@ bool PPCFastISel::PPCEmitStore(MVT VT, u
// into a RegBase.
if (Addr.BaseType == Address::FrameIndexBase) {
// VSX only provides an indexed store.
- if (IsVSFRC && Opc == PPC::STFD) return false;
+ if (Is32VSXStore || Is64VSXStore) return false;
MachineMemOperand *MMO =
FuncInfo.MF->getMachineMemOperand(
@@ -665,7 +674,7 @@ bool PPCFastISel::PPCEmitStore(MVT VT, u
// Base reg with offset in range.
} else if (UseOffset) {
// VSX only provides an indexed store.
- if (IsVSFRC && Opc == PPC::STFD) return false;
+ if (Is32VSXStore || Is64VSXStore) return false;
BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc))
.addReg(SrcReg).addImm(Addr.Offset).addReg(Addr.Base.Reg);
@@ -684,7 +693,7 @@ bool PPCFastISel::PPCEmitStore(MVT VT, u
case PPC::STH8: Opc = PPC::STHX8; break;
case PPC::STW8: Opc = PPC::STWX8; break;
case PPC::STD: Opc = PPC::STDX; break;
- case PPC::STFS: Opc = PPC::STFSX; break;
+ case PPC::STFS: Opc = IsVSSRC ? PPC::STXSSPX : PPC::STFSX; break;
case PPC::STFD: Opc = IsVSFRC ? PPC::STXSDX : PPC::STFDX; break;
}
Modified: llvm/trunk/lib/Target/PowerPC/PPCInstrVSX.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPCInstrVSX.td?rev=238578&r1=238577&r2=238578&view=diff
==============================================================================
--- llvm/trunk/lib/Target/PowerPC/PPCInstrVSX.td (original)
+++ llvm/trunk/lib/Target/PowerPC/PPCInstrVSX.td Fri May 29 12:13:25 2015
@@ -1078,6 +1078,82 @@ let AddedComplexity = 400 in { // Prefer
(outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
"xssubsp $XT, $XA, $XB", IIC_VecFP,
[(set f32:$XT, (fsub f32:$XA, f32:$XB))]>;
+
+ // FMA Instructions
+ let BaseName = "XSMADDASP" in {
+ let isCommutable = 1 in
+ def XSMADDASP : XX3Form<60, 1,
+ (outs vssrc:$XT),
+ (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
+ "xsmaddasp $XT, $XA, $XB", IIC_VecFP,
+ [(set f32:$XT, (fma f32:$XA, f32:$XB, f32:$XTi))]>,
+ RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
+ AltVSXFMARel;
+ let IsVSXFMAAlt = 1 in
+ def XSMADDMSP : XX3Form<60, 9,
+ (outs vssrc:$XT),
+ (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
+ "xsmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
+ RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
+ AltVSXFMARel;
+ }
+
+ let BaseName = "XSMSUBASP" in {
+ let isCommutable = 1 in
+ def XSMSUBASP : XX3Form<60, 17,
+ (outs vssrc:$XT),
+ (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
+ "xsmsubasp $XT, $XA, $XB", IIC_VecFP,
+ [(set f32:$XT, (fma f32:$XA, f32:$XB,
+ (fneg f32:$XTi)))]>,
+ RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
+ AltVSXFMARel;
+ let IsVSXFMAAlt = 1 in
+ def XSMSUBMSP : XX3Form<60, 25,
+ (outs vssrc:$XT),
+ (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
+ "xsmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
+ RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
+ AltVSXFMARel;
+ }
+
+ let BaseName = "XSNMADDASP" in {
+ let isCommutable = 1 in
+ def XSNMADDASP : XX3Form<60, 129,
+ (outs vssrc:$XT),
+ (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
+ "xsnmaddasp $XT, $XA, $XB", IIC_VecFP,
+ [(set f32:$XT, (fneg (fma f32:$XA, f32:$XB,
+ f32:$XTi)))]>,
+ RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
+ AltVSXFMARel;
+ let IsVSXFMAAlt = 1 in
+ def XSNMADDMSP : XX3Form<60, 137,
+ (outs vssrc:$XT),
+ (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
+ "xsnmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
+ RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
+ AltVSXFMARel;
+ }
+
+ let BaseName = "XSNMSUBASP" in {
+ let isCommutable = 1 in
+ def XSNMSUBASP : XX3Form<60, 145,
+ (outs vssrc:$XT),
+ (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
+ "xsnmsubasp $XT, $XA, $XB", IIC_VecFP,
+ [(set f32:$XT, (fneg (fma f32:$XA, f32:$XB,
+ (fneg f32:$XTi))))]>,
+ RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
+ AltVSXFMARel;
+ let IsVSXFMAAlt = 1 in
+ def XSNMSUBMSP : XX3Form<60, 153,
+ (outs vssrc:$XT),
+ (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
+ "xsnmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
+ RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
+ AltVSXFMARel;
+ }
} // AddedComplexity = 400
} // HasP8Vector
Modified: llvm/trunk/test/CodeGen/PowerPC/fma.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/PowerPC/fma.ll?rev=238578&r1=238577&r2=238578&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/PowerPC/fma.ll (original)
+++ llvm/trunk/test/CodeGen/PowerPC/fma.ll Fri May 29 12:13:25 2015
@@ -1,9 +1,12 @@
; RUN: llc < %s -march=ppc32 -fp-contract=fast -mattr=-vsx | FileCheck %s
; RUN: llc < %s -mtriple=powerpc64-unknown-linux-gnu -fp-contract=fast -mattr=+vsx -mcpu=pwr7 | FileCheck -check-prefix=CHECK-VSX %s
+; RUN: llc < %s -mtriple=powerpc64-unknown-linux-gnu -fp-contract=fast -mcpu=pwr8 | FileCheck -check-prefix=CHECK-P8 %s
+; RUN: llc < %s -mtriple=powerpc64le-unknown-linux-gnu -fp-contract=fast -mcpu=pwr8 | FileCheck -check-prefix=CHECK-P8 %s
declare double @dummy1(double) #0
declare double @dummy2(double, double) #0
declare double @dummy3(double, double, double) #0
+declare float @dummy4(float, float) #0
define double @test_FMADD1(double %A, double %B, double %C) {
%D = fmul double %A, %B ; <double> [#uses=1]
@@ -126,3 +129,83 @@ define float @test_FNMSUBS(float %A, flo
; CHECK-VSX: fnmsubs
; CHECK-VSX-NEXT: blr
}
+
+define float @test_XSMADDMSP(float %A, float %B, float %C) {
+ %D = fmul float %A, %B ; <float> [#uses=1]
+ %E = fadd float %C, %D ; <float> [#uses=1]
+ ret float %E
+; CHECK-P8-LABEL: test_XSMADDMSP:
+; CHECK-P8: xsmaddmsp
+; CHECK-P8-NEXT: blr
+}
+
+define float @test_XSMSUBMSP(float %A, float %B, float %C) {
+ %D = fmul float %A, %B ; <float> [#uses=1]
+ %E = fsub float %D, %C ; <float> [#uses=1]
+ ret float %E
+; CHECK-P8-LABEL: test_XSMSUBMSP:
+; CHECK-P8: xsmsubmsp
+; CHECK-P8-NEXT: blr
+}
+
+define float @test_XSMADDASP(float %A, float %B, float %C, float %D) {
+ %E = fmul float %A, %B ; <float> [#uses=2]
+ %F = fadd float %E, %C ; <float> [#uses=1]
+ %G = fsub float %E, %D ; <float> [#uses=1]
+ %H = call float @dummy4(float %F, float %G) ; <float> [#uses=1]
+ ret float %H
+; CHECK-P8-LABEL: test_XSMADDASP:
+; CHECK-P8: xsmaddasp
+; CHECK-P8-NEXT: xsmsubmsp
+}
+
+define float @test_XSMSUBASP(float %A, float %B, float %C, float %D) {
+ %E = fmul float %A, %B ; <float> [#uses=2]
+ %F = fsub float %E, %C ; <float> [#uses=1]
+ %G = fsub float %E, %D ; <float> [#uses=1]
+ %H = call float @dummy4(float %F, float %G) ; <float> [#uses=1]
+ ret float %H
+; CHECK-P8-LABEL: test_XSMSUBASP:
+; CHECK-P8: xsmsubasp
+; CHECK-P8-NEXT: xsmsubmsp
+}
+
+define float @test_XSNMADDMSP(float %A, float %B, float %C) {
+ %D = fmul float %A, %B ; <float> [#uses=1]
+ %E = fadd float %D, %C ; <float> [#uses=1]
+ %F = fsub float -0.000000e+00, %E ; <float> [#uses=1]
+ ret float %F
+; CHECK-P8-LABEL: test_XSNMADDMSP:
+; CHECK-P8: xsnmaddmsp
+; CHECK-P8-NEXT: blr
+}
+
+define float @test_XSNMSUBMSP(float %A, float %B, float %C) {
+ %D = fmul float %A, %B ; <float> [#uses=1]
+ %E = fsub float %D, %C ; <float> [#uses=1]
+ %F = fsub float -0.000000e+00, %E ; <float> [#uses=1]
+ ret float %F
+; CHECK-P8-LABEL: test_XSNMSUBMSP:
+; CHECK-P8: xsnmsubmsp
+; CHECK-P8-NEXT: blr
+}
+
+define float @test_XSNMADDASP(float %A, float %B, float %C) {
+ %D = fmul float %A, %B ; <float> [#uses=1]
+ %E = fadd float %D, %C ; <float> [#uses=1]
+ %F = fsub float -0.000000e+00, %E ; <float> [#uses=1]
+ %H = call float @dummy4(float %E, float %F) ; <float> [#uses=1]
+ ret float %F
+; CHECK-P8-LABEL: test_XSNMADDASP:
+; CHECK-P8: xsnmaddasp
+}
+
+define float @test_XSNMSUBASP(float %A, float %B, float %C) {
+ %D = fmul float %A, %B ; <float> [#uses=1]
+ %E = fsub float %D, %C ; <float> [#uses=1]
+ %F = fsub float -0.000000e+00, %E ; <float> [#uses=1]
+ %H = call float @dummy4(float %E, float %F) ; <float> [#uses=1]
+ ret float %F
+; CHECK-P8-LABEL: test_XSNMSUBASP:
+; CHECK-P8: xsnmsubasp
+}
Added: llvm/trunk/test/CodeGen/PowerPC/vsx-fma-sp.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/PowerPC/vsx-fma-sp.ll?rev=238578&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/PowerPC/vsx-fma-sp.ll (added)
+++ llvm/trunk/test/CodeGen/PowerPC/vsx-fma-sp.ll Fri May 29 12:13:25 2015
@@ -0,0 +1,167 @@
+; RUN: llc < %s -mtriple=powerpc64le-unknown-linux-gnu -mcpu=pwr8 -mattr=+vsx | FileCheck %s
+; RUN: llc < %s -mtriple=powerpc64le-unknown-linux-gnu -mcpu=pwr8 -mattr=+vsx -fast-isel -O0 | FileCheck -check-prefix=CHECK-FISL %s
+define void @test1sp(float %a, float %b, float %c, float %e, float* nocapture %d) #0 {
+entry:
+ %0 = tail call float @llvm.fma.f32(float %b, float %c, float %a)
+ store float %0, float* %d, align 4
+ %1 = tail call float @llvm.fma.f32(float %b, float %e, float %a)
+ %arrayidx1 = getelementptr inbounds float, float* %d, i64 1
+ store float %1, float* %arrayidx1, align 4
+ ret void
+
+; CHECK-LABEL: @test1sp
+; CHECK-DAG: li [[C1:[0-9]+]], 4
+; CHECK-DAG: xsmaddmsp 3, 2, 1
+; CHECK-DAG: xsmaddasp 1, 2, 4
+; CHECK-DAG: stxsspx 3, 0, 7
+; CHECK-DAG: stxsspx 1, 7, [[C1]]
+; CHECK: blr
+
+; CHECK-FISL-LABEL: @test1sp
+; CHECK-FISL-DAG: fmr 0, 1
+; CHECK-FISL-DAG: xsmaddasp 0, 2, 3
+; CHECK-FISL-DAG: stxsspx 0, 0, 7
+; CHECK-FISL-DAG: xsmaddasp 1, 2, 4
+; CHECK-FISL-DAG: li [[C1:[0-9]+]], 4
+; CHECK-FISL-DAG: stxsspx 1, 7, [[C1]]
+; CHECK-FISL: blr
+}
+
+define void @test2sp(float %a, float %b, float %c, float %e, float %f, float* nocapture %d) #0 {
+entry:
+ %0 = tail call float @llvm.fma.f32(float %b, float %c, float %a)
+ store float %0, float* %d, align 4
+ %1 = tail call float @llvm.fma.f32(float %b, float %e, float %a)
+ %arrayidx1 = getelementptr inbounds float, float* %d, i64 1
+ store float %1, float* %arrayidx1, align 4
+ %2 = tail call float @llvm.fma.f32(float %b, float %f, float %a)
+ %arrayidx2 = getelementptr inbounds float, float* %d, i64 2
+ store float %2, float* %arrayidx2, align 4
+ ret void
+
+; CHECK-LABEL: @test2sp
+; CHECK-DAG: li [[C1:[0-9]+]], 4
+; CHECK-DAG: li [[C2:[0-9]+]], 8
+; CHECK-DAG: xsmaddmsp 3, 2, 1
+; CHECK-DAG: xsmaddmsp 4, 2, 1
+; CHECK-DAG: xsmaddasp 1, 2, 5
+; CHECK-DAG: stxsspx 3, 0, 8
+; CHECK-DAG: stxsspx 4, 8, [[C1]]
+; CHECK-DAG: stxsspx 1, 8, [[C2]]
+; CHECK: blr
+
+; CHECK-FISL-LABEL: @test2sp
+; CHECK-FISL-DAG: fmr 0, 1
+; CHECK-FISL-DAG: xsmaddasp 0, 2, 3
+; CHECK-FISL-DAG: stxsspx 0, 0, 8
+; CHECK-FISL-DAG: fmr 0, 1
+; CHECK-FISL-DAG: xsmaddasp 0, 2, 4
+; CHECK-FISL-DAG: li [[C1:[0-9]+]], 4
+; CHECK-FISL-DAG: stxsspx 0, 8, [[C1]]
+; CHECK-FISL-DAG: xsmaddasp 1, 2, 5
+; CHECK-FISL-DAG: li [[C2:[0-9]+]], 8
+; CHECK-FISL-DAG: stxsspx 1, 8, [[C2]]
+; CHECK-FISL: blr
+}
+
+define void @test3sp(float %a, float %b, float %c, float %e, float %f, float* nocapture %d) #0 {
+entry:
+ %0 = tail call float @llvm.fma.f32(float %b, float %c, float %a)
+ store float %0, float* %d, align 4
+ %1 = tail call float @llvm.fma.f32(float %b, float %e, float %a)
+ %2 = tail call float @llvm.fma.f32(float %b, float %c, float %1)
+ %arrayidx1 = getelementptr inbounds float, float* %d, i64 3
+ store float %2, float* %arrayidx1, align 4
+ %3 = tail call float @llvm.fma.f32(float %b, float %f, float %a)
+ %arrayidx2 = getelementptr inbounds float, float* %d, i64 2
+ store float %3, float* %arrayidx2, align 4
+ %arrayidx3 = getelementptr inbounds float, float* %d, i64 1
+ store float %1, float* %arrayidx3, align 4
+ ret void
+
+; CHECK-LABEL: @test3sp
+; CHECK-DAG: fmr [[F1:[0-9]+]], 1
+; CHECK-DAG: li [[C1:[0-9]+]], 12
+; CHECK-DAG: li [[C2:[0-9]+]], 8
+; CHECK-DAG: li [[C3:[0-9]+]], 4
+; CHECK-DAG: xsmaddmsp 4, 2, 1
+; CHECK-DAG: xsmaddasp 1, 2, 5
+
+; Note: We could convert this next FMA to M-type as well, but it would require
+; re-ordering the instructions.
+; CHECK-DAG: xsmaddasp [[F1]], 2, 3
+
+; CHECK-DAG: xsmaddmsp 3, 2, 4
+; CHECK-DAG: stxsspx [[F1]], 0, 8
+; CHECK-DAG: stxsspx 3, 8, [[C1]]
+; CHECK-DAG: stxsspx 1, 8, [[C2]]
+; CHECK-DAG: stxsspx 4, 8, [[C3]]
+; CHECK: blr
+
+; CHECK-FISL-LABEL: @test3sp
+; CHECK-FISL-DAG: fmr [[F1:[0-9]+]], 1
+; CHECK-FISL-DAG: xsmaddasp [[F1]], 2, 4
+; CHECK-FISL-DAG: fmr 4, [[F1]]
+; CHECK-FISL-DAG: xsmaddasp 4, 2, 3
+; CHECK-FISL-DAG: li [[C1:[0-9]+]], 12
+; CHECK-FISL-DAG: stxsspx 4, 8, [[C1]]
+; CHECK-FISL-DAG: xsmaddasp 1, 2, 5
+; CHECK-FISL-DAG: li [[C2:[0-9]+]], 8
+; CHECK-FISL-DAG: stxsspx 1, 8, [[C2]]
+; CHECK-FISL-DAG: li [[C3:[0-9]+]], 4
+; CHECK-FISL-DAG: stxsspx 0, 8, [[C3]]
+; CHECK-FISL: blr
+}
+
+define void @test4sp(float %a, float %b, float %c, float %e, float %f, float* nocapture %d) #0 {
+entry:
+ %0 = tail call float @llvm.fma.f32(float %b, float %c, float %a)
+ store float %0, float* %d, align 4
+ %1 = tail call float @llvm.fma.f32(float %b, float %e, float %a)
+ %arrayidx1 = getelementptr inbounds float, float* %d, i64 1
+ store float %1, float* %arrayidx1, align 4
+ %2 = tail call float @llvm.fma.f32(float %b, float %c, float %1)
+ %arrayidx3 = getelementptr inbounds float, float* %d, i64 3
+ store float %2, float* %arrayidx3, align 4
+ %3 = tail call float @llvm.fma.f32(float %b, float %f, float %a)
+ %arrayidx4 = getelementptr inbounds float, float* %d, i64 2
+ store float %3, float* %arrayidx4, align 4
+ ret void
+
+; CHECK-LABEL: @test4sp
+; CHECK-DAG: fmr [[F1:[0-9]+]], 1
+; CHECK-DAG: li [[C1:[0-9]+]], 4
+; CHECK-DAG: li [[C2:[0-9]+]], 8
+; CHECK-DAG: xsmaddmsp 4, 2, 1
+
+; Note: We could convert this next FMA to M-type as well, but it would require
+; re-ordering the instructions.
+; CHECK-DAG: xsmaddasp 1, 2, 5
+
+; CHECK-DAG: xsmaddasp [[F1]], 2, 3
+; CHECK-DAG: stxsspx [[F1]], 0, 8
+; CHECK-DAG: stxsspx 4, 8, [[C1]]
+; CHECK-DAG: li [[C3:[0-9]+]], 12
+; CHECK-DAG: xsmaddasp 4, 2, 3
+; CHECK-DAG: stxsspx 4, 8, [[C3]]
+; CHECK-DAG: stxsspx 1, 8, [[C2]]
+; CHECK: blr
+
+; CHECK-FISL-LABEL: @test4sp
+; CHECK-FISL-DAG: fmr [[F1:[0-9]+]], 1
+; CHECK-FISL-DAG: xsmaddasp [[F1]], 2, 3
+; CHECK-FISL-DAG: stxsspx 0, 0, 8
+; CHECK-FISL-DAG: fmr [[F1]], 1
+; CHECK-FISL-DAG: xsmaddasp [[F1]], 2, 4
+; CHECK-FISL-DAG: li [[C3:[0-9]+]], 4
+; CHECK-FISL-DAG: stxsspx 0, 8, [[C3]]
+; CHECK-FISL-DAG: xsmaddasp 0, 2, 3
+; CHECK-FISL-DAG: li [[C1:[0-9]+]], 12
+; CHECK-FISL-DAG: stxsspx 0, 8, [[C1]]
+; CHECK-FISL-DAG: xsmaddasp 1, 2, 5
+; CHECK-FISL-DAG: li [[C2:[0-9]+]], 8
+; CHECK-FISL-DAG: stxsspx 1, 8, [[C2]]
+; CHECK-FISL: blr
+}
+
+declare float @llvm.fma.f32(float, float, float) #0
Modified: llvm/trunk/test/MC/Disassembler/PowerPC/vsx.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/MC/Disassembler/PowerPC/vsx.txt?rev=238578&r1=238577&r2=238578&view=diff
==============================================================================
--- llvm/trunk/test/MC/Disassembler/PowerPC/vsx.txt (original)
+++ llvm/trunk/test/MC/Disassembler/PowerPC/vsx.txt Fri May 29 12:13:25 2015
@@ -90,6 +90,12 @@
# CHECK: xsmaddmdp 7, 63, 27
0xf0 0xff 0xd9 0x4c
+# CHECK: xsmaddasp 7, 63, 27
+0xf0 0xff 0xd8 0x0c
+
+# CHECK: xsmaddmsp 7, 63, 27
+0xf0 0xff 0xd8 0x4c
+
# CHECK: xsmaxdp 7, 63, 27
0xf0 0xff 0xdd 0x04
@@ -102,6 +108,12 @@
# CHECK: xsmsubmdp 7, 63, 27
0xf0 0xff 0xd9 0xcc
+# CHECK: xsmsubasp 7, 63, 27
+0xf0 0xff 0xd8 0x8c
+
+# CHECK: xsmsubmsp 7, 63, 27
+0xf0 0xff 0xd8 0xcc
+
# CHECK: xsmulsp 7, 63, 27
0xf0 0xff 0xd8 0x84
@@ -126,6 +138,18 @@
# CHECK: xsnmsubmdp 7, 63, 27
0xf0 0xff 0xdd 0xcc
+# CHECK: xsnmaddasp 7, 63, 27
+0xf0 0xff 0xdc 0x0c
+
+# CHECK: xsnmaddmsp 7, 63, 27
+0xf0 0xff 0xdc 0x4c
+
+# CHECK: xsnmsubasp 7, 63, 27
+0xf0 0xff 0xdc 0x8c
+
+# CHECK: xsnmsubmsp 7, 63, 27
+0xf0 0xff 0xdc 0xcc
+
# CHECK: xsrdpi 7, 27
0xf0 0xe0 0xd9 0x24
Modified: llvm/trunk/test/MC/PowerPC/vsx.s
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/MC/PowerPC/vsx.s?rev=238578&r1=238577&r2=238578&view=diff
==============================================================================
--- llvm/trunk/test/MC/PowerPC/vsx.s (original)
+++ llvm/trunk/test/MC/PowerPC/vsx.s Fri May 29 12:13:25 2015
@@ -95,6 +95,12 @@
# CHECK-BE: xsmaddmdp 7, 63, 27 # encoding: [0xf0,0xff,0xd9,0x4c]
# CHECK-LE: xsmaddmdp 7, 63, 27 # encoding: [0x4c,0xd9,0xff,0xf0]
xsmaddmdp 7, 63, 27
+# CHECK-BE: xsmaddasp 7, 63, 27 # encoding: [0xf0,0xff,0xd8,0x0c]
+# CHECK-LE: xsmaddasp 7, 63, 27 # encoding: [0x0c,0xd8,0xff,0xf0]
+ xsmaddasp 7, 63, 27
+# CHECK-BE: xsmaddmsp 7, 63, 27 # encoding: [0xf0,0xff,0xd8,0x4c]
+# CHECK-LE: xsmaddmsp 7, 63, 27 # encoding: [0x4c,0xd8,0xff,0xf0]
+ xsmaddmsp 7, 63, 27
# CHECK-BE: xsmaxdp 7, 63, 27 # encoding: [0xf0,0xff,0xdd,0x04]
# CHECK-LE: xsmaxdp 7, 63, 27 # encoding: [0x04,0xdd,0xff,0xf0]
xsmaxdp 7, 63, 27
@@ -107,6 +113,12 @@
# CHECK-BE: xsmsubmdp 7, 63, 27 # encoding: [0xf0,0xff,0xd9,0xcc]
# CHECK-LE: xsmsubmdp 7, 63, 27 # encoding: [0xcc,0xd9,0xff,0xf0]
xsmsubmdp 7, 63, 27
+# CHECK-BE: xsmsubasp 7, 63, 27 # encoding: [0xf0,0xff,0xd8,0x8c]
+# CHECK-LE: xsmsubasp 7, 63, 27 # encoding: [0x8c,0xd8,0xff,0xf0]
+ xsmsubasp 7, 63, 27
+# CHECK-BE: xsmsubmsp 7, 63, 27 # encoding: [0xf0,0xff,0xd8,0xcc]
+# CHECK-LE: xsmsubmsp 7, 63, 27 # encoding: [0xcc,0xd8,0xff,0xf0]
+ xsmsubmsp 7, 63, 27
# CHECK-BE: xsmulsp 7, 63, 27 # encoding: [0xf0,0xff,0xd8,0x84]
# CHECK-LE: xsmulsp 7, 63, 27 # encoding: [0x84,0xd8,0xff,0xf0]
xsmulsp 7, 63, 27
@@ -131,6 +143,18 @@
# CHECK-BE: xsnmsubmdp 7, 63, 27 # encoding: [0xf0,0xff,0xdd,0xcc]
# CHECK-LE: xsnmsubmdp 7, 63, 27 # encoding: [0xcc,0xdd,0xff,0xf0]
xsnmsubmdp 7, 63, 27
+# CHECK-BE: xsnmaddasp 7, 63, 27 # encoding: [0xf0,0xff,0xdc,0x0c]
+# CHECK-LE: xsnmaddasp 7, 63, 27 # encoding: [0x0c,0xdc,0xff,0xf0]
+ xsnmaddasp 7, 63, 27
+# CHECK-BE: xsnmaddmsp 7, 63, 27 # encoding: [0xf0,0xff,0xdc,0x4c]
+# CHECK-LE: xsnmaddmsp 7, 63, 27 # encoding: [0x4c,0xdc,0xff,0xf0]
+ xsnmaddmsp 7, 63, 27
+# CHECK-BE: xsnmsubasp 7, 63, 27 # encoding: [0xf0,0xff,0xdc,0x8c]
+# CHECK-LE: xsnmsubasp 7, 63, 27 # encoding: [0x8c,0xdc,0xff,0xf0]
+ xsnmsubasp 7, 63, 27
+# CHECK-BE: xsnmsubmsp 7, 63, 27 # encoding: [0xf0,0xff,0xdc,0xcc]
+# CHECK-LE: xsnmsubmsp 7, 63, 27 # encoding: [0xcc,0xdc,0xff,0xf0]
+ xsnmsubmsp 7, 63, 27
# CHECK-BE: xsrdpi 7, 27 # encoding: [0xf0,0xe0,0xd9,0x24]
# CHECK-LE: xsrdpi 7, 27 # encoding: [0x24,0xd9,0xe0,0xf0]
xsrdpi 7, 27
More information about the llvm-commits
mailing list