From b69d7e5c93a6fa95b7aa5be4060dbc890531f3ea Mon Sep 17 00:00:00 2001 From: Sjoerd Meijer Date: Fri, 13 Apr 2018 15:34:26 +0000 Subject: [PATCH] [ARM] FP16 vmaxnm/vminnm scalar instructions This adds code generation support for the FP16 vmaxnm/vminnm scalar instructions. Differential Revision: https://reviews.llvm.org/D44675 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@330034 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Target/ARM/ARMISelLowering.cpp | 5 + lib/Target/ARM/ARMInstrNEON.td | 13 + lib/Target/ARM/ARMInstrVFP.td | 4 +- test/CodeGen/ARM/fp16-instructions.ll | 7 +- test/CodeGen/ARM/fp16-vminmaxnm-safe.ll | 366 ++++++++++++++++++++++++++++ test/CodeGen/ARM/fp16-vminmaxnm.ll | 418 ++++++++++++++++++++++++++++++++ 6 files changed, 810 insertions(+), 3 deletions(-) create mode 100644 test/CodeGen/ARM/fp16-vminmaxnm-safe.ll create mode 100644 test/CodeGen/ARM/fp16-vminmaxnm.ll diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp index 7f0222019f2..1e032640c55 100644 --- a/lib/Target/ARM/ARMISelLowering.cpp +++ b/lib/Target/ARM/ARMISelLowering.cpp @@ -527,6 +527,9 @@ ARMTargetLowering::ARMTargetLowering(const TargetMachine &TM, setOperationAction(ISD::BITCAST, MVT::i16, Custom); setOperationAction(ISD::BITCAST, MVT::i32, Custom); setOperationAction(ISD::BITCAST, MVT::f16, Custom); + + setOperationAction(ISD::FMINNUM, MVT::f16, Legal); + setOperationAction(ISD::FMAXNUM, MVT::f16, Legal); } for (MVT VT : MVT::vector_valuetypes()) { @@ -1147,6 +1150,8 @@ ARMTargetLowering::ARMTargetLowering(const TargetMachine &TM, if (Subtarget->hasNEON()) { // vmin and vmax aren't available in a scalar form, so we use // a NEON instruction with an undef lane instead. + setOperationAction(ISD::FMINNAN, MVT::f16, Legal); + setOperationAction(ISD::FMAXNAN, MVT::f16, Legal); setOperationAction(ISD::FMINNAN, MVT::f32, Legal); setOperationAction(ISD::FMAXNAN, MVT::f32, Legal); setOperationAction(ISD::FMINNAN, MVT::v2f32, Legal); diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td index 8757520c10c..fa2b83581f7 100644 --- a/lib/Target/ARM/ARMInstrNEON.td +++ b/lib/Target/ARM/ARMInstrNEON.td @@ -6870,6 +6870,17 @@ class N3VSPat (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)), SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>; +class N3VSPatFP16 + : NEONFPPat<(f16 (OpNode HPR:$a, HPR:$b)), + (EXTRACT_SUBREG + (v4f16 (COPY_TO_REGCLASS (Inst + (INSERT_SUBREG + (v4f16 (COPY_TO_REGCLASS (v4f16 (IMPLICIT_DEF)), DPR_VFP2)), + HPR:$a, ssub_0), + (INSERT_SUBREG + (v4f16 (COPY_TO_REGCLASS (v4f16 (IMPLICIT_DEF)), DPR_VFP2)), + HPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>; + class N3VSMulOpPat : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))), (EXTRACT_SUBREG @@ -6912,6 +6923,8 @@ def : N3VSMulOpPat, Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>; def : N2VSPat; def : N2VSPat; +def : N3VSPatFP16, Requires<[HasFullFP16]>; +def : N3VSPatFP16, Requires<[HasFullFP16]>; def : N3VSPat, Requires<[HasNEON]>; def : N3VSPat, Requires<[HasNEON]>; def : NVCVTFIPat; diff --git a/lib/Target/ARM/ARMInstrVFP.td b/lib/Target/ARM/ARMInstrVFP.td index 34945a23c03..98b780f1459 100644 --- a/lib/Target/ARM/ARMInstrVFP.td +++ b/lib/Target/ARM/ARMInstrVFP.td @@ -482,9 +482,9 @@ defm VSELVS : vsel_inst<"vs", 0b01, 6>; multiclass vmaxmin_inst { let DecoderNamespace = "VFPV8", PostEncoderMethod = "" in { def H : AHbInp<0b11101, 0b00, opc, - (outs SPR:$Sd), (ins SPR:$Sn, SPR:$Sm), + (outs HPR:$Sd), (ins HPR:$Sn, HPR:$Sm), NoItinerary, !strconcat(op, ".f16\t$Sd, $Sn, $Sm"), - []>, + [(set HPR:$Sd, (SD HPR:$Sn, HPR:$Sm))]>, Requires<[HasFullFP16]>; def S : ASbInp<0b11101, 0b00, opc, diff --git a/test/CodeGen/ARM/fp16-instructions.ll b/test/CodeGen/ARM/fp16-instructions.ll index d61a631da3d..03deb2e3ef2 100644 --- a/test/CodeGen/ARM/fp16-instructions.ll +++ b/test/CodeGen/ARM/fp16-instructions.ll @@ -28,6 +28,10 @@ ; RUN: llc < %s -mtriple=arm-none-eabihf -mattr=+fullfp16 -fp-contract=fast | FileCheck %s --check-prefixes=CHECK,CHECK-HARDFP-FULLFP16-FAST ; RUN: llc < %s -mtriple=thumbv7-none-eabihf -mattr=+fullfp16 -fp-contract=fast | FileCheck %s --check-prefixes=CHECK,CHECK-HARDFP-FULLFP16-FAST +; TODO: we can't pass half-precision arguments as "half" types yet. We do +; that for the time being by passing "float %f.coerce" and the necessary +; bitconverts/truncates. But when we can pass half types, we do want to use +; and test that here. define float @RetValBug(float %A.coerce) { entry: @@ -477,9 +481,10 @@ entry: ; CHECK-HARDFP-FULLFP16-FAST-NEXT: vmov.f32 s0, s2 } -; TODO: ; 17. VMAXNM ; 18. VMINNM +; Tested in fp16-vminmaxnm.ll and fp16-vminmaxnm-safe.ll + ; 19. VMLA define float @VMLA(float %a.coerce, float %b.coerce, float %c.coerce) { diff --git a/test/CodeGen/ARM/fp16-vminmaxnm-safe.ll b/test/CodeGen/ARM/fp16-vminmaxnm-safe.ll new file mode 100644 index 00000000000..9c2405a1426 --- /dev/null +++ b/test/CodeGen/ARM/fp16-vminmaxnm-safe.ll @@ -0,0 +1,366 @@ +; RUN: llc < %s -mtriple=armv8-eabi -mattr=+fullfp16 | FileCheck %s +; RUN: llc < %s -mtriple thumbv7a -mattr=+fullfp16 | FileCheck %s + +; TODO: we can't pass half-precision arguments as "half" types yet. We do +; that for the time being by passing "float %f.coerce" and the necessary +; bitconverts/truncates. In these tests we pass i16 and use 1 bitconvert, which +; is the shortest way to get a half type. But when we can pass half types, we +; want to use that here. + +define half @fp16_vminnm_o(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_o: +; CHECK-NOT: vminnm.f16 +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp olt half %0, %1 + %cond = select i1 %cmp, half %0, half %1 + ret half %cond +} + +define half @fp16_vminnm_o_rev(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_o_rev: +; CHECK-NOT: vminnm.f16 +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp ogt half %0, %1 + %cond = select i1 %cmp, half %0, half %1 + ret half %cond +} + +define half @fp16_vminnm_u(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_u: +; CHECK-NOT: vminnm.f16 +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp ult half %0, %1 + %cond = select i1 %cmp, half %0, half %1 + ret half %cond +} + +define half @fp16_vminnm_ule(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_ule: +; CHECK-NOT: vminnm.f16 +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp ule half %0, %1 + %cond = select i1 %cmp, half %0, half %1 + ret half %cond +} + +define half @fp16_vminnm_u_rev(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_u_rev: +; CHECK-NOT: vminnm.f16 +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp ugt half %0, %1 + %cond = select i1 %cmp, half %1, half %0 + ret half %cond +} + +define half @fp16_vmaxnm_o(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_o: +; CHECK-NOT: vmaxnm.f16 +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp ogt half %0, %1 + %cond = select i1 %cmp, half %0, half %1 + ret half %cond +} + +define half @fp16_vmaxnm_oge(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_oge: +; CHECK-NOT: vmaxnm.f16 +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp oge half %0, %1 + %cond = select i1 %cmp, half %0, half %1 + ret half %cond +} + +define half @fp16_vmaxnm_o_rev(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_o_rev: +; CHECK-NOT: vmaxnm.f16 +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp olt half %0, %1 + %cond = select i1 %cmp, half %1, half %0 + ret half %cond +} + +define half @fp16_vmaxnm_ole_rev(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_ole_rev: +; CHECK-NOT: vmaxnm.f16 +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp ole half %0, %1 + %cond = select i1 %cmp, half %1, half %0 + ret half %cond +} + +define half @fp16_vmaxnm_u(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_u: +; CHECK-NOT: vmaxnm.f16 +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp ugt half %0, %1 + %cond = select i1 %cmp, half %0, half %1 + ret half %cond +} + +define half @fp16_vmaxnm_uge(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_uge: +; CHECK-NOT: vmaxnm.f16 +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp uge half %0, %1 + %cond = select i1 %cmp, half %0, half %1 + ret half %cond +} + +define half @fp16_vmaxnm_u_rev(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_u_rev: +; CHECK-NOT: vmaxnm.f16 +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp ult half %0, %1 + %cond = select i1 %cmp, half %1, half %0 + ret half %cond +} + +; known non-NaNs + +define half @fp16_vminnm_NNNo(i16 signext %a) { +; CHECK-LABEL: fp16_vminnm_NNNo: +; CHECK: vldr.16 [[S0:s[0-9]]], .LCPI{{.*}} +; CHECK: vmov.f16 [[S2:s[0-9]]], #1.200000e+01 +; CHECK: vmov.f16 [[S4:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s2, [[S4]], [[S2]] +; CHECK: vmin.f16 d0, d1, d0 +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp olt half %0, 12. + %cond1 = select i1 %cmp1, half %0, half 12. + %cmp2 = fcmp olt half 34., %cond1 + %cond2 = select i1 %cmp2, half 34., half %cond1 + ret half %cond2 +} + +define half @fp16_vminnm_NNNo_rev(i16 signext %a) { +; CHECK-LABEL: fp16_vminnm_NNNo_rev: +; CHECK: vldr.16 [[S0:s[0-9]]], .LCPI{{.*}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vmin.f16 d0, d1, d0 +; CHECK: vldr.16 [[S2:s[0-9]]], .LCPI{{.*}} +; CHECK: vminnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp ogt half %0, 56. + %cond1 = select i1 %cmp1, half 56., half %0 + %cmp2 = fcmp ogt half 78., %cond1 + %cond2 = select i1 %cmp2, half %cond1, half 78. + ret half %cond2 +} + +define half @fp16_vminnm_NNNu(i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_NNNu: +; CHECK: vldr.16 [[S0:s[0-9]]], .LCPI{{.*}} +; CHECK: vmov.f16 [[S2:s[0-9]]], #1.200000e+01 +; CHECK: vmov.f16 [[S4:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s2, [[S4]], [[S2]] +; CHECK: vmin.f16 d0, d1, d0 +entry: + %0 = bitcast i16 %b to half + %cmp1 = fcmp ult half 12., %0 + %cond1 = select i1 %cmp1, half 12., half %0 + %cmp2 = fcmp ult half %cond1, 34. + %cond2 = select i1 %cmp2, half %cond1, half 34. + ret half %cond2 +} + +define half @fp16_vminnm_NNNule(i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_NNNule: +; CHECK: vldr.16 [[S2:s[0-9]]], .LCPI{{.*}} +; CHECK: vmov.f16 [[S4:s[0-9]]], r{{.}} +; CHECK: vldr.16 [[S0:s[0-9]]], .LCPI{{.*}} +; CHECK: vminnm.f16 s2, [[S4]], [[S2]] +; CHECK: vmin.f16 d0, d1, d0 + +entry: + %0 = bitcast i16 %b to half + %cmp1 = fcmp ule half 34., %0 + %cond1 = select i1 %cmp1, half 34., half %0 + %cmp2 = fcmp ule half %cond1, 56. + %cond2 = select i1 %cmp2, half %cond1, half 56. + ret half %cond2 +} + +define half @fp16_vminnm_NNNu_rev(i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_NNNu_rev: + +; CHECK: vldr.16 [[S0:s[0-9]]], .LCPI{{.*}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vmin.f16 d0, d1, d0 +; CHECK: vldr.16 [[S2:s[0-9]]], .LCPI{{.*}} +; CHECK: vminnm.f16 s0, [[S0]], [[S2]] + +entry: + %0 = bitcast i16 %b to half + %cmp1 = fcmp ugt half 56., %0 + %cond1 = select i1 %cmp1, half %0, half 56. + %cmp2 = fcmp ugt half %cond1, 78. + %cond2 = select i1 %cmp2, half 78., half %cond1 + ret half %cond2 +} + +define half @fp16_vmaxnm_NNNo(i16 signext %a) { +; CHECK-LABEL: fp16_vmaxnm_NNNo: +; CHECK: vldr.16 [[S0:s[0-9]]], .LCPI{{.*}} +; CHECK: vmov.f16 [[S2:s[0-9]]], #1.200000e+01 +; CHECK: vmov.f16 [[S4:s[0-9]]], r{{.}} +; CHECK: vmaxnm.f16 s2, [[S4]], [[S2]] +; CHECK: vmax.f16 d0, d1, d0 +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp ogt half %0, 12. + %cond1 = select i1 %cmp1, half %0, half 12. + %cmp2 = fcmp ogt half 34., %cond1 + %cond2 = select i1 %cmp2, half 34., half %cond1 + ret half %cond2 +} + +define half @fp16_vmaxnm_NNNoge(i16 signext %a) { +; CHECK-LABEL: fp16_vmaxnm_NNNoge: +; CHECK: vldr.16 [[S2:s[0-9]]], .LCPI{{.*}} +; CHECK: vmov.f16 [[S4:s[0-9]]], r{{.}} +; CHECK: vldr.16 [[S0:s[0-9]]], .LCPI{{.*}} +; CHECK: vmaxnm.f16 s2, [[S4]], [[S2]] +; CHECK: vmax.f16 d0, d1, d0 +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp oge half %0, 34. + %cond1 = select i1 %cmp1, half %0, half 34. + %cmp2 = fcmp oge half 56., %cond1 + %cond2 = select i1 %cmp2, half 56., half %cond1 + ret half %cond2 +} + +define half @fp16_vmaxnm_NNNo_rev(i16 signext %a) { +; CHECK-LABEL: fp16_vmaxnm_NNNo_rev: +; CHECK: vldr.16 [[S0:s[0-9]]], .LCPI{{.*}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vmax.f16 d0, d1, d0 +; CHECK: vldr.16 [[S2:s[0-9]]], .LCPI{{.*}} +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp olt half %0, 56. + %cond1 = select i1 %cmp1, half 56., half %0 + %cmp2 = fcmp olt half 78., %cond1 + %cond2 = select i1 %cmp2, half %cond1, half 78. + ret half %cond2 +} + +define half @fp16_vmaxnm_NNNole_rev(i16 signext %a) { +; CHECK-LABEL: fp16_vmaxnm_NNNole_rev: +; CHECK: vldr.16 [[S0:s[0-9]]], .LCPI{{.*}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vmax.f16 d0, d1, d0 +; CHECK: vldr.16 [[S2:s[0-9]]], .LCPI{{.*}} +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp ole half %0, 78. + %cond1 = select i1 %cmp1, half 78., half %0 + %cmp2 = fcmp ole half 90., %cond1 + %cond2 = select i1 %cmp2, half %cond1, half 90. + ret half %cond2 +} + +define half @fp16_vmaxnm_NNNu(i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_NNNu: +; CHECK: vldr.16 [[S0:s[0-9]]], .LCPI{{.*}} +; CHECK: vmov.f16 [[S2:s[0-9]]], #1.200000e+01 +; CHECK: vmov.f16 [[S4:s[0-9]]], r{{.}} +; CHECK: vmaxnm.f16 s2, [[S4]], [[S2]] +; CHECK: vmax.f16 d0, d1, d0 +entry: + %0 = bitcast i16 %b to half + %cmp1 = fcmp ugt half 12., %0 + %cond1 = select i1 %cmp1, half 12., half %0 + %cmp2 = fcmp ugt half %cond1, 34. + %cond2 = select i1 %cmp2, half %cond1, half 34. + ret half %cond2 +} + +define half @fp16_vmaxnm_NNNuge(i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_NNNuge: +; CHECK: vldr.16 [[S2:s[0-9]]], .LCPI{{.*}} +; CHECK: vmov.f16 [[S4:s[0-9]]], r{{.}} +; CHECK: vldr.16 [[S0:s[0-9]]], .LCPI{{.*}} +; CHECK: vmaxnm.f16 s2, [[S4]], [[S2]] +; CHECK: vmax.f16 d0, d1, d0 +entry: + %0 = bitcast i16 %b to half + %cmp1 = fcmp uge half 34., %0 + %cond1 = select i1 %cmp1, half 34., half %0 + %cmp2 = fcmp uge half %cond1, 56. + %cond2 = select i1 %cmp2, half %cond1, half 56. + ret half %cond2 +} + +define half @fp16_vminmaxnm_neg0(i16 signext %a) { +; CHECK-LABEL: fp16_vminmaxnm_neg0: +; CHECK: vldr.16 [[S0:s[0-9]]], .LCPI{{.*}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s2, [[S2]], [[S0]] +; CHECK: vmax.f16 d0, d1, d0 +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp olt half %0, -0. + %cond1 = select i1 %cmp1, half %0, half -0. + %cmp2 = fcmp ugt half %cond1, -0. + %cond2 = select i1 %cmp2, half %cond1, half -0. + ret half %cond2 +} + +define half @fp16_vminmaxnm_e_0(i16 signext %a) { +; CHECK-LABEL: fp16_vminmaxnm_e_0: +; CHECK: vldr.16 [[S2:s[0-9]]], .LCPI{{.*}} +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmin.f16 d0, d0, d1 +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp nsz ole half 0., %0 + %cond1 = select i1 %cmp1, half 0., half %0 + %cmp2 = fcmp nsz uge half 0., %cond1 + %cond2 = select i1 %cmp2, half 0., half %cond1 + ret half %cond2 +} + +define half @fp16_vminmaxnm_e_neg0(i16 signext %a) { +; CHECK-LABEL: fp16_vminmaxnm_e_neg0: +; CHECK: vldr.16 [[S0:s[0-9]]], .LCPI{{.*}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s2, [[S2]], [[S0]] +; CHECK: vmax.f16 d0, d1, d0 +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp nsz ule half -0., %0 + %cond1 = select i1 %cmp1, half -0., half %0 + %cmp2 = fcmp nsz oge half -0., %cond1 + %cond2 = select i1 %cmp2, half -0., half %cond1 + ret half %cond2 +} diff --git a/test/CodeGen/ARM/fp16-vminmaxnm.ll b/test/CodeGen/ARM/fp16-vminmaxnm.ll new file mode 100644 index 00000000000..525c27be4f9 --- /dev/null +++ b/test/CodeGen/ARM/fp16-vminmaxnm.ll @@ -0,0 +1,418 @@ +; RUN: llc < %s -mtriple=arm-eabi -mattr=+fullfp16 -enable-unsafe-fp-math -enable-no-nans-fp-math | FileCheck %s +; RUN: llc < %s -mtriple thumbv7a -mattr=+fullfp16 -enable-unsafe-fp-math -enable-no-nans-fp-math | FileCheck %s + +; TODO: we can't pass half-precision arguments as "half" types yet. We do +; that for the time being by passing "float %f.coerce" and the necessary +; bitconverts/truncates. In these tests we pass i16 and use 1 bitconvert, which +; is the shortest way to get a half type. But when we can pass half types, we +; want to use that here. + +define half @fp16_vminnm_o(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_o: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp fast olt half %0, %1 + %cond = select i1 %cmp, half %0, half %1 + ret half %cond +} + +define half @fp16_vminnm_o_rev(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_o_rev: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp fast ogt half %0, %1 + %cond = select i1 %cmp, half %1, half %0 + ret half %cond +} + +define half @fp16_vminnm_u(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_u: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp fast ult half %0, %1 + %cond = select i1 %cmp, half %0, half %1 + ret half %cond +} + +define half @fp16_vminnm_ule(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_ule: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp fast ule half %0, %1 + %cond = select i1 %cmp, half %0, half %1 + ret half %cond +} + +define half @fp16_vminnm_u_rev(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_u_rev: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp fast ugt half %0, %1 + %cond = select i1 %cmp, half %1, half %0 + ret half %cond +} + +define half @fp16_vmaxnm_o(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_o: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vmaxnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp fast ogt half %0, %1 + %cond = select i1 %cmp, half %0, half %1 + ret half %cond +} + +define half @fp16_vmaxnm_oge(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_oge: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vmaxnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp fast oge half %0, %1 + %cond = select i1 %cmp, half %0, half %1 + ret half %cond +} + +define half @fp16_vmaxnm_o_rev(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_o_rev: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vmaxnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp fast olt half %0, %1 + %cond = select i1 %cmp, half %1, half %0 + ret half %cond +} + +define half @fp16_vmaxnm_ole_rev(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_ole_rev: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vmaxnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp fast ole half %0, %1 + %cond = select i1 %cmp, half %1, half %0 + ret half %cond +} + +define half @fp16_vmaxnm_u(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_u: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vmaxnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp fast ugt half %0, %1 + %cond = select i1 %cmp, half %0, half %1 + ret half %cond +} + +define half @fp16_vmaxnm_uge(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_uge: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vmaxnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp fast uge half %0, %1 + %cond = select i1 %cmp, half %0, half %1 + ret half %cond +} + +define half @fp16_vmaxnm_u_rev(i16 signext %a, i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_u_rev: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vmaxnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %1 = bitcast i16 %b to half + %cmp = fcmp fast ult half %0, %1 + %cond = select i1 %cmp, half %1, half %0 + ret half %cond +} + +; known non-NaNs + +define half @fp16_vminnm_NNNo(i16 signext %a) { +; CHECK-LABEL: fp16_vminnm_NNNo: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], #1.200000e+01 +; CHECK: vminnm.f16 s0, [[S0]], [[S2]] +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vminnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp fast olt half %0, 12. + %cond1 = select i1 %cmp1, half %0, half 12. + %cmp2 = fcmp fast olt half 34., %cond1 + %cond2 = select i1 %cmp2, half 34., half %cond1 + ret half %cond2 +} + +define half @fp16_vminnm_NNNo_rev(i16 signext %a) { +; CHECK-LABEL: fp16_vminnm_NNNo_rev: +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s0, [[S0]], [[S2]] +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vminnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp fast ogt half %0, 56. + %cond1 = select i1 %cmp1, half 56., half %0 + %cmp2 = fcmp fast ogt half 78., %cond1 + %cond2 = select i1 %cmp2, half %cond1, half 78. + ret half %cond2 +} + +define half @fp16_vminnm_NNNu(i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_NNNu: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], #1.200000e+01 +; CHECK: vminnm.f16 s0, [[S0]], [[S2]] +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vminnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %b to half + %cmp1 = fcmp fast ult half 12., %0 + %cond1 = select i1 %cmp1, half 12., half %0 + %cmp2 = fcmp fast ult half %cond1, 34. + %cond2 = select i1 %cmp2, half %cond1, half 34. + ret half %cond2 +} + +define half @fp16_vminnm_NNNule(i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_NNNule: +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s0, [[S0]], [[S2]] +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vminnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %b to half + %cmp1 = fcmp fast ule half 34., %0 + %cond1 = select i1 %cmp1, half 34., half %0 + %cmp2 = fcmp fast ule half %cond1, 56. + %cond2 = select i1 %cmp2, half %cond1, half 56. + ret half %cond2 +} + +define half @fp16_vminnm_NNNu_rev(i16 signext %b) { +; CHECK-LABEL: fp16_vminnm_NNNu_rev: +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s0, [[S0]], [[S2]] +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vminnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %b to half + %cmp1 = fcmp fast ugt half 56., %0 + %cond1 = select i1 %cmp1, half %0, half 56. + %cmp2 = fcmp fast ugt half %cond1, 78. + %cond2 = select i1 %cmp2, half 78., half %cond1 + ret half %cond2 +} + +define half @fp16_vmaxnm_NNNo(i16 signext %a) { +; CHECK-LABEL: fp16_vmaxnm_NNNo: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], #1.200000e+01 +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp fast ogt half %0, 12. + %cond1 = select i1 %cmp1, half %0, half 12. + %cmp2 = fcmp fast ogt half 34., %cond1 + %cond2 = select i1 %cmp2, half 34., half %cond1 + ret half %cond2 +} + +define half @fp16_vmaxnm_NNNoge(i16 signext %a) { +; CHECK-LABEL: fp16_vmaxnm_NNNoge: +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp fast oge half %0, 34. + %cond1 = select i1 %cmp1, half %0, half 34. + %cmp2 = fcmp fast oge half 56., %cond1 + %cond2 = select i1 %cmp2, half 56., half %cond1 + ret half %cond2 +} + +define half @fp16_vmaxnm_NNNo_rev(i16 signext %a) { +; CHECK-LABEL: fp16_vmaxnm_NNNo_rev: +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp fast olt half %0, 56. + %cond1 = select i1 %cmp1, half 56., half %0 + %cmp2 = fcmp fast olt half 78., %cond1 + %cond2 = select i1 %cmp2, half %cond1, half 78. + ret half %cond2 +} + +define half @fp16_vmaxnm_NNNole_rev(i16 signext %a) { +; CHECK-LABEL: fp16_vmaxnm_NNNole_rev: +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp fast ole half %0, 78. + %cond1 = select i1 %cmp1, half 78., half %0 + %cmp2 = fcmp fast ole half 90., %cond1 + %cond2 = select i1 %cmp2, half %cond1, half 90. + ret half %cond2 +} + +define half @fp16_vmaxnm_NNNu(i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_NNNu: +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmov.f16 [[S2:s[0-9]]], #1.200000e+01 +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %b to half + %cmp1 = fcmp fast ugt half 12., %0 + %cond1 = select i1 %cmp1, half 12., half %0 + %cmp2 = fcmp fast ugt half %cond1, 34. + %cond2 = select i1 %cmp2, half %cond1, half 34. + ret half %cond2 +} + +define half @fp16_vmaxnm_NNNuge(i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_NNNuge: +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %b to half + %cmp1 = fcmp fast uge half 34., %0 + %cond1 = select i1 %cmp1, half 34., half %0 + %cmp2 = fcmp fast uge half %cond1, 56. + %cond2 = select i1 %cmp2, half %cond1, half 56. + ret half %cond2 +} + +define half @fp16_vmaxnm_NNNu_rev(i16 signext %b) { +; CHECK-LABEL: fp16_vmaxnm_NNNu_rev: +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vmov.f16 [[S0:s[0-9]]], r{{.}} +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +; CHECK: vldr.16 s2, .LCPI{{.*}} +; CHECK: vmaxnm.f16 s0, [[S0]], [[S2]] +entry: + %0 = bitcast i16 %b to half + %cmp1 = fcmp fast ult half 56., %0 + %cond1 = select i1 %cmp1, half %0, half 56. + %cmp2 = fcmp fast ult half %cond1, 78. + %cond2 = select i1 %cmp2, half 78., half %cond1 + ret half %cond2 +} + +define half @fp16_vminmaxnm_0(i16 signext %a) { +; CHECK-LABEL: fp16_vminmaxnm_0: +; CHECK: vldr.16 s0, .LCPI{{.*}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s2, s2, s0 +; CHECK: vmaxnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp fast olt half %0, 0. + %cond1 = select i1 %cmp1, half %0, half 0. + %cmp2 = fcmp fast ogt half %cond1, 0. + %cond2 = select i1 %cmp2, half %cond1, half 0. + ret half %cond2 +} + +define half @fp16_vminmaxnm_neg0(i16 signext %a) { +; CHECK-LABEL: fp16_vminmaxnm_neg0: +; CHECK: vldr.16 s0, .LCPI{{.*}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s2, s2, s0 +; CHECK: vmaxnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp fast olt half %0, -0. + %cond1 = select i1 %cmp1, half %0, half -0. + %cmp2 = fcmp fast ugt half %cond1, -0. + %cond2 = select i1 %cmp2, half %cond1, half -0. + ret half %cond2 +} + +define half @fp16_vminmaxnm_e_0(i16 signext %a) { +; CHECK-LABEL: fp16_vminmaxnm_e_0: +; CHECK: vldr.16 s0, .LCPI{{.*}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s2, s2, s0 +; CHECK: vmaxnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp fast ule half 0., %0 + %cond1 = select i1 %cmp1, half 0., half %0 + %cmp2 = fcmp fast uge half 0., %cond1 + %cond2 = select i1 %cmp2, half 0., half %cond1 + ret half %cond2 +} + +define half @fp16_vminmaxnm_e_neg0(i16 signext %a) { +; CHECK-LABEL: fp16_vminmaxnm_e_neg0: +; CHECK: vldr.16 s0, .LCPI{{.*}} +; CHECK: vmov.f16 [[S2:s[0-9]]], r{{.}} +; CHECK: vminnm.f16 s2, s2, s0 +; CHECK: vmaxnm.f16 s0, [[S2]], [[S0]] +entry: + %0 = bitcast i16 %a to half + %cmp1 = fcmp fast ule half -0., %0 + %cond1 = select i1 %cmp1, half -0., half %0 + %cmp2 = fcmp fast oge half -0., %cond1 + %cond2 = select i1 %cmp2, half -0., half %cond1 + ret half %cond2 +} -- 2.11.0