This is an archive of the discontinued LLVM Phabricator instance.

Differential D134108

[VE] Add maxnum and minnum
ClosedPublic

Authored by kaz7 on Sep 17 2022, 4:33 AM.

Details

Reviewers
efocht
Commits
rG337e54ec956c: [VE] Add maxnum and minnum
Summary

Add maxnum and minnum for float and double. Lowering is already
implemented, so this patch changes them legal and adds regression
tests.

Diff Detail

Event Timeline

kaz7 created this revision.Sep 17 2022, 4:33 AM
Herald added a project: Restricted Project. · View Herald TranscriptSep 17 2022, 4:33 AM
Herald added a subscriber: hiraditya. · View Herald Transcript
kaz7 requested review of this revision.Sep 17 2022, 4:33 AM
Herald added a subscriber: llvm-commits. · View Herald TranscriptSep 17 2022, 4:33 AM
Harbormaster completed remote builds in B187321: Diff 460989.Sep 17 2022, 5:48 AM
efocht accepted this revision.Sep 21 2022, 1:15 AM

LGTM

This revision is now accepted and ready to land.Sep 21 2022, 1:15 AM
Closed by commit rG337e54ec956c: [VE] Add maxnum and minnum (authored by kaz7). · Explain WhySep 21 2022, 2:04 AM
This revision was automatically updated to reflect the committed changes.
kaz7 added a commit: rG337e54ec956c: [VE] Add maxnum and minnum.

Revision Contents

PathSize
llvm/
lib/
Target/
VE/
5 lines
test/
CodeGen/
VE/
Scalar/
162 lines
162 lines

Diff 461818

llvm/lib/Target/VE/VEISelLowering.cpp

Show First 20 Lines • Show All 243 Lines • ▼ Show 20 Lines for (MVT VT : MVT::fp_valuetypes()) {
setOperationAction(ISD::FABS, VT, Expand); setOperationAction(ISD::FABS, VT, Expand);
setOperationAction(ISD::FCOPYSIGN, VT, Expand); setOperationAction(ISD::FCOPYSIGN, VT, Expand);
setOperationAction(ISD::FCOS, VT, Expand); setOperationAction(ISD::FCOS, VT, Expand);
setOperationAction(ISD::FMA, VT, Expand); setOperationAction(ISD::FMA, VT, Expand);
setOperationAction(ISD::FSIN, VT, Expand); setOperationAction(ISD::FSIN, VT, Expand);
setOperationAction(ISD::FSQRT, VT, Expand); setOperationAction(ISD::FSQRT, VT, Expand);
} }
// VE has single and double FMINNUM and FMAXNUM
for (MVT VT : {MVT::f32, MVT::f64}) {
setOperationAction({ISD::FMAXNUM, ISD::FMINNUM}, VT, Legal);
}
/// } Floating-point math functions /// } Floating-point math functions
/// Atomic instructions { /// Atomic instructions {
setMaxAtomicSizeInBitsSupported(64); setMaxAtomicSizeInBitsSupported(64);
setMinCmpXchgSizeInBits(32); setMinCmpXchgSizeInBits(32);
setSupportsUnalignedAtomics(false); setSupportsUnalignedAtomics(false);
▲ Show 20 Lines • Show All 2,714 Lines • Show Last 20 Lines

llvm/test/CodeGen/VE/Scalar/maxnum.ll

  • This file was added.
; RUN: llc < %s -mtriple=ve | FileCheck %s
;;; Test ‘llvm.maxnum.*’ intrinsic
;;;
;;; Syntax:
;;; This is an overloaded intrinsic. You can use llvm.maxnum on any
;;; floating-point or vector of floating-point type. Not all targets
;;; support all types however.
;;;
;;; declare float @llvm.maxnum.f32(float %Val0, float %Val1)
;;; declare double @llvm.maxnum.f64(double %Val0, double %Val1)
;;; declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
;;; declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
;;; declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
;;;
;;; Overview:
;;; The ‘llvm.maxnum.*’ intrinsics return the maximum of the two arguments.
;;;
;;; Arguments:
;;; The arguments and return value are floating-point numbers of the same
;;; type.
;;;
;;; Semantics:
;;; Follows the IEEE-754 semantics for maxNum except for the handling of
;;; signaling NaNs. This matches the behavior of libm’s fmax.
;;;
;;; If either operand is a NaN, returns the other non-NaN operand.
;;; Returns NaN only if both operands are NaN. The returned NaN is
;;; always quiet. If the operands compare equal, returns a value
;;; that compares equal to both operands. This means that
;;; fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
;;;
;;; Unlike the IEEE-754 2008 behavior, this does not distinguish between
;;; signaling and quiet NaN inputs. If a target’s implementation follows
;;; the standard and returns a quiet NaN if either input is a signaling
;;; NaN, the intrinsic lowering is responsible for quieting the inputs
;;; to correctly return the non-NaN input (e.g. by using the equivalent
;;; of llvm.canonicalize).
;;;
;;; Note:
;;; We test only float/double/fp128.
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define float @func_fp_fmax_var_float(float noundef %0, float noundef %1) {
; CHECK-LABEL: func_fp_fmax_var_float:
; CHECK: # %bb.0:
; CHECK-NEXT: fmax.s %s0, %s0, %s1
; CHECK-NEXT: b.l.t (, %s10)
%3 = tail call fast float @llvm.maxnum.f32(float %0, float %1)
ret float %3
}
; Function Attrs: mustprogress nocallback nofree nosync nounwind readnone speculatable willreturn
declare float @llvm.maxnum.f32(float, float)
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define double @func_fp_fmax_var_double(double noundef %0, double noundef %1) {
; CHECK-LABEL: func_fp_fmax_var_double:
; CHECK: # %bb.0:
; CHECK-NEXT: fmax.d %s0, %s0, %s1
; CHECK-NEXT: b.l.t (, %s10)
%3 = tail call fast double @llvm.maxnum.f64(double %0, double %1)
ret double %3
}
; Function Attrs: mustprogress nocallback nofree nosync nounwind readnone speculatable willreturn
declare double @llvm.maxnum.f64(double, double)
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define fp128 @func_fp_fmax_var_quad(fp128 noundef %0, fp128 noundef %1) {
; CHECK-LABEL: func_fp_fmax_var_quad:
; CHECK: # %bb.0:
; CHECK-NEXT: fcmp.q %s4, %s0, %s2
; CHECK-NEXT: cmov.d.gt %s2, %s0, %s4
; CHECK-NEXT: cmov.d.gt %s3, %s1, %s4
; CHECK-NEXT: or %s0, 0, %s2
; CHECK-NEXT: or %s1, 0, %s3
; CHECK-NEXT: b.l.t (, %s10)
%3 = tail call fast fp128 @llvm.maxnum.f128(fp128 %0, fp128 %1)
ret fp128 %3
}
; Function Attrs: mustprogress nocallback nofree nosync nounwind readnone speculatable willreturn
declare fp128 @llvm.maxnum.f128(fp128, fp128)
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define float @func_fp_fmax_zero_float(float noundef %0) {
; CHECK-LABEL: func_fp_fmax_zero_float:
; CHECK: # %bb.0:
; CHECK-NEXT: fmax.s %s0, %s0, (0)1
; CHECK-NEXT: b.l.t (, %s10)
%2 = tail call fast float @llvm.maxnum.f32(float %0, float 0.000000e+00)
ret float %2
}
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define double @func_fp_fmax_zero_double(double noundef %0) {
; CHECK-LABEL: func_fp_fmax_zero_double:
; CHECK: # %bb.0:
; CHECK-NEXT: fmax.d %s0, %s0, (0)1
; CHECK-NEXT: b.l.t (, %s10)
%2 = tail call fast double @llvm.maxnum.f64(double %0, double 0.000000e+00)
ret double %2
}
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define fp128 @func_fp_fmax_zero_quad(fp128 noundef %0) {
; CHECK-LABEL: func_fp_fmax_zero_quad:
; CHECK: # %bb.0:
; CHECK-NEXT: lea %s2, .LCPI{{[0-9]+}}_0@lo
; CHECK-NEXT: and %s2, %s2, (32)0
; CHECK-NEXT: lea.sl %s4, .LCPI{{[0-9]+}}_0@hi(, %s2)
; CHECK-NEXT: ld %s2, 8(, %s4)
; CHECK-NEXT: ld %s3, (, %s4)
; CHECK-NEXT: fcmp.q %s4, %s0, %s2
; CHECK-NEXT: cmov.d.gt %s2, %s0, %s4
; CHECK-NEXT: cmov.d.gt %s3, %s1, %s4
; CHECK-NEXT: or %s0, 0, %s2
; CHECK-NEXT: or %s1, 0, %s3
; CHECK-NEXT: b.l.t (, %s10)
%2 = tail call fast fp128 @llvm.maxnum.f128(fp128 %0, fp128 0xL00000000000000000000000000000000)
ret fp128 %2
}
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define float @func_fp_fmax_const_float(float noundef %0) {
; CHECK-LABEL: func_fp_fmax_const_float:
; CHECK: # %bb.0:
; CHECK-NEXT: fmax.s %s0, %s0, (2)1
; CHECK-NEXT: b.l.t (, %s10)
%2 = tail call fast float @llvm.maxnum.f32(float %0, float -2.000000e+00)
ret float %2
}
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define double @func_fp_fmax_const_double(double noundef %0) {
; CHECK-LABEL: func_fp_fmax_const_double:
; CHECK: # %bb.0:
; CHECK-NEXT: fmax.d %s0, %s0, (2)1
; CHECK-NEXT: b.l.t (, %s10)
%2 = tail call fast double @llvm.maxnum.f64(double %0, double -2.000000e+00)
ret double %2
}
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define fp128 @func_fp_fmax_const_quad(fp128 noundef %0) {
; CHECK-LABEL: func_fp_fmax_const_quad:
; CHECK: # %bb.0:
; CHECK-NEXT: lea %s2, .LCPI{{[0-9]+}}_0@lo
; CHECK-NEXT: and %s2, %s2, (32)0
; CHECK-NEXT: lea.sl %s4, .LCPI{{[0-9]+}}_0@hi(, %s2)
; CHECK-NEXT: ld %s2, 8(, %s4)
; CHECK-NEXT: ld %s3, (, %s4)
; CHECK-NEXT: fcmp.q %s4, %s0, %s2
; CHECK-NEXT: cmov.d.gt %s2, %s0, %s4
; CHECK-NEXT: cmov.d.gt %s3, %s1, %s4
; CHECK-NEXT: or %s0, 0, %s2
; CHECK-NEXT: or %s1, 0, %s3
; CHECK-NEXT: b.l.t (, %s10)
%2 = tail call fast fp128 @llvm.maxnum.f128(fp128 %0, fp128 0xL0000000000000000C000000000000000)
ret fp128 %2
}

llvm/test/CodeGen/VE/Scalar/minnum.ll

  • This file was added.
; RUN: llc < %s -mtriple=ve | FileCheck %s
;;; Test ‘llvm.minnum.*’ intrinsic
;;;
;;; Syntax:
;;; This is an overloaded intrinsic. You can use llvm.minnum on any
;;; floating-point or vector of floating-point type. Not all targets
;;; support all types however.
;;;
;;; declare float @llvm.minnum.f32(float %Val0, float %Val1)
;;; declare double @llvm.minnum.f64(double %Val0, double %Val1)
;;; declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
;;; declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
;;; declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
;;;
;;; Overview:
;;; The ‘llvm.minnum.*’ intrinsics return the minimum of the two arguments.
;;;
;;; Arguments:
;;; The arguments and return value are floating-point numbers of the same
;;; type.
;;;
;;; Semantics:
;;; Follows the IEEE-754 semantics for minNum, except for handling of
;;; signaling NaNs. This match’s the behavior of libm’s fmin.
;;;
;;; If either operand is a NaN, returns the other non-NaN operand.
;;; Returns NaN only if both operands are NaN. The returned NaN is
;;; always quiet. If the operands compare equal, returns a value
;;; that compares equal to both operands. This means that
;;; fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
;;;
;;; Unlike the IEEE-754 2008 behavior, this does not distinguish between
;;; signaling and quiet NaN inputs. If a target’s implementation follows
;;; the standard and returns a quiet NaN if either input is a signaling
;;; NaN, the intrinsic lowering is responsible for quieting the inputs
;;; to correctly return the non-NaN input (e.g. by using the equivalent
;;; of llvm.canonicalize).
;;;
;;; Note:
;;; We test only float/double/fp128.
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define float @func_fp_fmin_var_float(float noundef %0, float noundef %1) {
; CHECK-LABEL: func_fp_fmin_var_float:
; CHECK: # %bb.0:
; CHECK-NEXT: fmin.s %s0, %s0, %s1
; CHECK-NEXT: b.l.t (, %s10)
%3 = tail call fast float @llvm.minnum.f32(float %0, float %1)
ret float %3
}
; Function Attrs: mustprogress nocallback nofree nosync nounwind readnone speculatable willreturn
declare float @llvm.minnum.f32(float, float)
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define double @func_fp_fmin_var_double(double noundef %0, double noundef %1) {
; CHECK-LABEL: func_fp_fmin_var_double:
; CHECK: # %bb.0:
; CHECK-NEXT: fmin.d %s0, %s0, %s1
; CHECK-NEXT: b.l.t (, %s10)
%3 = tail call fast double @llvm.minnum.f64(double %0, double %1)
ret double %3
}
; Function Attrs: mustprogress nocallback nofree nosync nounwind readnone speculatable willreturn
declare double @llvm.minnum.f64(double, double)
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define fp128 @func_fp_fmin_var_quad(fp128 noundef %0, fp128 noundef %1) {
; CHECK-LABEL: func_fp_fmin_var_quad:
; CHECK: # %bb.0:
; CHECK-NEXT: fcmp.q %s4, %s0, %s2
; CHECK-NEXT: cmov.d.lt %s2, %s0, %s4
; CHECK-NEXT: cmov.d.lt %s3, %s1, %s4
; CHECK-NEXT: or %s0, 0, %s2
; CHECK-NEXT: or %s1, 0, %s3
; CHECK-NEXT: b.l.t (, %s10)
%3 = tail call fast fp128 @llvm.minnum.f128(fp128 %0, fp128 %1)
ret fp128 %3
}
; Function Attrs: mustprogress nocallback nofree nosync nounwind readnone speculatable willreturn
declare fp128 @llvm.minnum.f128(fp128, fp128)
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define float @func_fp_fmin_zero_float(float noundef %0) {
; CHECK-LABEL: func_fp_fmin_zero_float:
; CHECK: # %bb.0:
; CHECK-NEXT: fmin.s %s0, %s0, (0)1
; CHECK-NEXT: b.l.t (, %s10)
%2 = tail call fast float @llvm.minnum.f32(float %0, float 0.000000e+00)
ret float %2
}
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define double @func_fp_fmin_zero_double(double noundef %0) {
; CHECK-LABEL: func_fp_fmin_zero_double:
; CHECK: # %bb.0:
; CHECK-NEXT: fmin.d %s0, %s0, (0)1
; CHECK-NEXT: b.l.t (, %s10)
%2 = tail call fast double @llvm.minnum.f64(double %0, double 0.000000e+00)
ret double %2
}
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define fp128 @func_fp_fmin_zero_quad(fp128 noundef %0) {
; CHECK-LABEL: func_fp_fmin_zero_quad:
; CHECK: # %bb.0:
; CHECK-NEXT: lea %s2, .LCPI{{[0-9]+}}_0@lo
; CHECK-NEXT: and %s2, %s2, (32)0
; CHECK-NEXT: lea.sl %s4, .LCPI{{[0-9]+}}_0@hi(, %s2)
; CHECK-NEXT: ld %s2, 8(, %s4)
; CHECK-NEXT: ld %s3, (, %s4)
; CHECK-NEXT: fcmp.q %s4, %s0, %s2
; CHECK-NEXT: cmov.d.lt %s2, %s0, %s4
; CHECK-NEXT: cmov.d.lt %s3, %s1, %s4
; CHECK-NEXT: or %s0, 0, %s2
; CHECK-NEXT: or %s1, 0, %s3
; CHECK-NEXT: b.l.t (, %s10)
%2 = tail call fast fp128 @llvm.minnum.f128(fp128 %0, fp128 0xL00000000000000000000000000000000)
ret fp128 %2
}
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define float @func_fp_fmin_const_float(float noundef %0) {
; CHECK-LABEL: func_fp_fmin_const_float:
; CHECK: # %bb.0:
; CHECK-NEXT: fmin.s %s0, %s0, (2)1
; CHECK-NEXT: b.l.t (, %s10)
%2 = tail call fast float @llvm.minnum.f32(float %0, float -2.000000e+00)
ret float %2
}
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define double @func_fp_fmin_const_double(double noundef %0) {
; CHECK-LABEL: func_fp_fmin_const_double:
; CHECK: # %bb.0:
; CHECK-NEXT: fmin.d %s0, %s0, (2)1
; CHECK-NEXT: b.l.t (, %s10)
%2 = tail call fast double @llvm.minnum.f64(double %0, double -2.000000e+00)
ret double %2
}
; Function Attrs: mustprogress nofree nosync nounwind readnone willreturn
define fp128 @func_fp_fmin_const_quad(fp128 noundef %0) {
; CHECK-LABEL: func_fp_fmin_const_quad:
; CHECK: # %bb.0:
; CHECK-NEXT: lea %s2, .LCPI{{[0-9]+}}_0@lo
; CHECK-NEXT: and %s2, %s2, (32)0
; CHECK-NEXT: lea.sl %s4, .LCPI{{[0-9]+}}_0@hi(, %s2)
; CHECK-NEXT: ld %s2, 8(, %s4)
; CHECK-NEXT: ld %s3, (, %s4)
; CHECK-NEXT: fcmp.q %s4, %s0, %s2
; CHECK-NEXT: cmov.d.lt %s2, %s0, %s4
; CHECK-NEXT: cmov.d.lt %s3, %s1, %s4
; CHECK-NEXT: or %s0, 0, %s2
; CHECK-NEXT: or %s1, 0, %s3
; CHECK-NEXT: b.l.t (, %s10)
%2 = tail call fast fp128 @llvm.minnum.f128(fp128 %0, fp128 0xL0000000000000000C000000000000000)
ret fp128 %2
}