This is an archive of the discontinued LLVM Phabricator instance.

Differential D89228

[VE] Support copysign math function
ClosedPublic

Authored by kaz7 on Oct 12 2020, 3:22 AM.

Details

Reviewers
simoll
k-ishizaka
Commits
rG66be2e00ef29: [VE] Support copysign math function
Summary

VE doesn't have instruction for copysign, so expand it. Add a
regression test also.

Diff Detail

Event Timeline

kaz7 created this revision.Oct 12 2020, 3:22 AM
Herald added subscribers: llvm-commits, hiraditya. · View Herald TranscriptOct 12 2020, 3:22 AM
kaz7 requested review of this revision.Oct 12 2020, 3:22 AM
Harbormaster completed remote builds in B74770: Diff 297535.Oct 12 2020, 3:58 AM
simoll accepted this revision.Oct 12 2020, 4:57 AM
This revision is now accepted and ready to land.Oct 12 2020, 4:57 AM
This revision was landed with ongoing or failed builds.Oct 12 2020, 5:06 AM
Closed by commit rG66be2e00ef29: [VE] Support copysign math function (authored by kaz7). · Explain Why
This revision was automatically updated to reflect the committed changes.
kaz7 added a commit: rG66be2e00ef29: [VE] Support copysign math function.

Revision Contents

PathSize
llvm/
lib/
Target/
VE/
9 lines
test/
CodeGen/
VE/
195 lines

Diff 297535

llvm/lib/Target/VE/VEISelLowering.cpp

Show First 20 Lines • Show All 752 Lines • ▼ Show 20 LinesVETargetLowering::VETargetLowering(const TargetMachine &TM,
setOperationAction(ISD::FDIV, MVT::f128, Expand); setOperationAction(ISD::FDIV, MVT::f128, Expand);
for (MVT FPVT : {MVT::f32, MVT::f64}) { for (MVT FPVT : {MVT::f32, MVT::f64}) {
// f32 and f64 uses ConstantFP. f128 uses ConstantPool. // f32 and f64 uses ConstantFP. f128 uses ConstantPool.
setOperationAction(ISD::ConstantFP, FPVT, Legal); setOperationAction(ISD::ConstantFP, FPVT, Legal);
} }
/// } Floating-point Ops /// } Floating-point Ops
/// Floating-point math functions {
// VE doesn't have following floating point math functions.
for (MVT VT : MVT::fp_valuetypes()) {
setOperationAction(ISD::FCOPYSIGN, VT, Expand);
}
/// } Floating-point math functions
setStackPointerRegisterToSaveRestore(VE::SX11); setStackPointerRegisterToSaveRestore(VE::SX11);
// We have target-specific dag combine patterns for the following nodes: // We have target-specific dag combine patterns for the following nodes:
setTargetDAGCombine(ISD::TRUNCATE); setTargetDAGCombine(ISD::TRUNCATE);
// Set function alignment to 16 bytes // Set function alignment to 16 bytes
setMinFunctionAlignment(Align(16)); setMinFunctionAlignment(Align(16));
▲ Show 20 Lines • Show All 586 Lines • Show Last 20 Lines

llvm/test/CodeGen/VE/fp_copysign.ll

  • This file was added.
; RUN: llc < %s -mtriple=ve | FileCheck %s
;;; Test ‘llvm.copysign.*’ Intrinsic
;;;
;;; Syntax:
;;; This is an overloaded intrinsic. You can use llvm.copysign on any
;;; floating-point or vector of floating-point type. Not all targets
;;; support all types however.
;;;
;;; declare float @llvm.copysign.f32(float %Mag, float %Sgn)
;;; declare double @llvm.copysign.f64(double %Mag, double %Sgn)
;;; declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
;;; declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
;;; declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
;;;
;;; Overview:
;;; The ‘llvm.copysign.*’ intrinsics return a value with the magnitude of
;;; the first operand and the sign of the second operand.
;;;
;;; Arguments:
;;; The arguments and return value are floating-point numbers of the same
;;; type.
;;;
;;; Semantics:
;;; This function returns the same values as the libm copysign functions
;;; would, and handles error conditions in the same way.
;;;
;;; Note:
;;; We test only float/double/fp128.
; Function Attrs: nounwind readnone
define float @copysign_float_var(float %0, float %1) {
; CHECK-LABEL: copysign_float_var:
; CHECK: .LBB{{[0-9]+}}_2:
; CHECK-NEXT: sra.l %s1, %s1, 32
; CHECK-NEXT: lea %s2, -2147483648
; CHECK-NEXT: and %s2, %s2, (32)0
; CHECK-NEXT: and %s1, %s1, %s2
; CHECK-NEXT: sra.l %s0, %s0, 32
; CHECK-NEXT: and %s0, %s0, (33)0
; CHECK-NEXT: or %s0, %s0, %s1
; CHECK-NEXT: sll %s0, %s0, 32
; CHECK-NEXT: or %s11, 0, %s9
%3 = tail call float @llvm.copysign.f32(float %0, float %1)
ret float %3
}
; Function Attrs: nounwind readnone speculatable willreturn
declare float @llvm.copysign.f32(float, float)
; Function Attrs: nounwind readnone
define double @copysign_double_var(double %0, double %1) {
; CHECK-LABEL: copysign_double_var:
; CHECK: .LBB{{[0-9]+}}_2:
; CHECK-NEXT: and %s1, %s1, (1)1
; CHECK-NEXT: and %s0, %s0, (1)0
; CHECK-NEXT: or %s0, %s0, %s1
; CHECK-NEXT: or %s11, 0, %s9
%3 = tail call double @llvm.copysign.f64(double %0, double %1)
ret double %3
}
; Function Attrs: nounwind readnone speculatable willreturn
declare double @llvm.copysign.f64(double, double)
; Function Attrs: nounwind readnone
define fp128 @copysign_quad_var(fp128 %0, fp128 %1) {
; CHECK-LABEL: copysign_quad_var:
; CHECK: .LBB{{[0-9]+}}_2:
; CHECK-NEXT: st %s3, 192(, %s11)
; CHECK-NEXT: st %s2, 200(, %s11)
; CHECK-NEXT: st %s1, 176(, %s11)
; CHECK-NEXT: st %s0, 184(, %s11)
; CHECK-NEXT: ld1b.zx %s0, 207(, %s11)
; CHECK-NEXT: ld1b.zx %s1, 191(, %s11)
; CHECK-NEXT: lea %s2, 128
; CHECK-NEXT: and %s0, %s0, %s2
; CHECK-NEXT: and %s1, %s1, (57)0
; CHECK-NEXT: or %s0, %s1, %s0
; CHECK-NEXT: st1b %s0, 191(, %s11)
; CHECK-NEXT: ld %s1, 176(, %s11)
; CHECK-NEXT: ld %s0, 184(, %s11)
; CHECK-NEXT: or %s11, 0, %s9
%3 = tail call fp128 @llvm.copysign.f128(fp128 %0, fp128 %1)
ret fp128 %3
}
; Function Attrs: nounwind readnone speculatable willreturn
declare fp128 @llvm.copysign.f128(fp128, fp128)
; Function Attrs: nounwind readnone
define float @copysign_float_zero(float %0) {
; CHECK-LABEL: copysign_float_zero:
; CHECK: .LBB{{[0-9]+}}_2:
; CHECK-NEXT: sra.l %s0, %s0, 32
; CHECK-NEXT: lea %s1, -2147483648
; CHECK-NEXT: and %s1, %s1, (32)0
; CHECK-NEXT: and %s0, %s0, %s1
; CHECK-NEXT: sll %s0, %s0, 32
; CHECK-NEXT: or %s11, 0, %s9
%2 = tail call float @llvm.copysign.f32(float 0.000000e+00, float %0)
ret float %2
}
; Function Attrs: nounwind readnone
define double @copysign_double_zero(double %0) {
; CHECK-LABEL: copysign_double_zero:
; CHECK: .LBB{{[0-9]+}}_2:
; CHECK-NEXT: and %s0, %s0, (1)1
; CHECK-NEXT: or %s11, 0, %s9
%2 = tail call double @llvm.copysign.f64(double 0.000000e+00, double %0)
ret double %2
}
; Function Attrs: nounwind readnone
define fp128 @copysign_quad_zero(fp128 %0) {
; CHECK-LABEL: copysign_quad_zero:
; CHECK: .LBB{{[0-9]+}}_2:
; CHECK-NEXT: lea %s2, .LCPI{{[0-9]+}}_0@lo
; CHECK-NEXT: and %s2, %s2, (32)0
; CHECK-NEXT: lea.sl %s2, .LCPI{{[0-9]+}}_0@hi(, %s2)
; CHECK-NEXT: ld %s4, 8(, %s2)
; CHECK-NEXT: ld %s5, (, %s2)
; CHECK-NEXT: st %s1, 192(, %s11)
; CHECK-NEXT: st %s0, 200(, %s11)
; CHECK-NEXT: st %s5, 176(, %s11)
; CHECK-NEXT: st %s4, 184(, %s11)
; CHECK-NEXT: ld1b.zx %s0, 207(, %s11)
; CHECK-NEXT: ld1b.zx %s1, 191(, %s11)
; CHECK-NEXT: lea %s2, 128
; CHECK-NEXT: and %s0, %s0, %s2
; CHECK-NEXT: and %s1, %s1, (57)0
; CHECK-NEXT: or %s0, %s1, %s0
; CHECK-NEXT: st1b %s0, 191(, %s11)
; CHECK-NEXT: ld %s1, 176(, %s11)
; CHECK-NEXT: ld %s0, 184(, %s11)
; CHECK-NEXT: or %s11, 0, %s9
%2 = tail call fp128 @llvm.copysign.f128(fp128 0xL00000000000000000000000000000000, fp128 %0)
ret fp128 %2
}
; Function Attrs: nounwind readnone
define float @copysign_float_const(float %0) {
; CHECK-LABEL: copysign_float_const:
; CHECK: .LBB{{[0-9]+}}_2:
; CHECK-NEXT: sra.l %s0, %s0, 32
; CHECK-NEXT: lea %s1, -2147483648
; CHECK-NEXT: and %s1, %s1, (32)0
; CHECK-NEXT: and %s0, %s0, %s1
; CHECK-NEXT: lea %s1, 1073741824
; CHECK-NEXT: or %s0, %s0, %s1
; CHECK-NEXT: sll %s0, %s0, 32
; CHECK-NEXT: or %s11, 0, %s9
%2 = tail call float @llvm.copysign.f32(float -2.000000e+00, float %0)
ret float %2
}
; Function Attrs: nounwind readnone
define double @copysign_double_const(double %0) {
; CHECK-LABEL: copysign_double_const:
; CHECK: .LBB{{[0-9]+}}_2:
; CHECK-NEXT: and %s0, %s0, (1)1
; CHECK-NEXT: lea.sl %s1, 1073741824
; CHECK-NEXT: or %s0, %s0, %s1
; CHECK-NEXT: or %s11, 0, %s9
%2 = tail call double @llvm.copysign.f64(double -2.000000e+00, double %0)
ret double %2
}
; Function Attrs: nounwind readnone
define fp128 @copysign_quad_const(fp128 %0) {
; CHECK-LABEL: copysign_quad_const:
; CHECK: .LBB{{[0-9]+}}_2:
; CHECK-NEXT: lea %s2, .LCPI{{[0-9]+}}_0@lo
; CHECK-NEXT: and %s2, %s2, (32)0
; CHECK-NEXT: lea.sl %s2, .LCPI{{[0-9]+}}_0@hi(, %s2)
; CHECK-NEXT: ld %s4, 8(, %s2)
; CHECK-NEXT: ld %s5, (, %s2)
; CHECK-NEXT: st %s1, 192(, %s11)
; CHECK-NEXT: st %s0, 200(, %s11)
; CHECK-NEXT: st %s5, 176(, %s11)
; CHECK-NEXT: st %s4, 184(, %s11)
; CHECK-NEXT: ld1b.zx %s0, 207(, %s11)
; CHECK-NEXT: ld1b.zx %s1, 191(, %s11)
; CHECK-NEXT: lea %s2, 128
; CHECK-NEXT: and %s0, %s0, %s2
; CHECK-NEXT: and %s1, %s1, (57)0
; CHECK-NEXT: or %s0, %s1, %s0
; CHECK-NEXT: st1b %s0, 191(, %s11)
; CHECK-NEXT: ld %s1, 176(, %s11)
; CHECK-NEXT: ld %s0, 184(, %s11)
; CHECK-NEXT: or %s11, 0, %s9
%2 = tail call fp128 @llvm.copysign.f128(fp128 0xL0000000000000000C000000000000000, fp128 %0)
ret fp128 %2
}