maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[InstCombine] fold copysign with constant sign argument to (fneg+)fabs 

If the sign of the sign argument is known (this could be extended to use ValueTracking),
then we can use fneg+fabs to clear/set the sign bit of the magnitude argument.
http://llvm.org/docs/LangRef.html#llvm-copysign-intrinsic

This transform is already done in DAGCombiner, but we can do it sooner in IR as
suggested in PR44153:
https://bugs.llvm.org/show_bug.cgi?id=44153

We have effectively no analysis for copysign in IR, so we are taking the unusual step
of increasing the number of IR instructions for the negative constant case.

Differential Revision: https://reviews.llvm.org/D70792
This commit is contained in:
Sanjay Patel 2019-12-02 09:21:59 -05:00
parent 6236496561
commit af0babc90a
2 changed files with 25 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
View File

@ -2279,6 +2279,21 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
break;
}
case Intrinsic::copysign: {
const APFloat *C;
if (match(II->getArgOperand(1), m_APFloat(C))) {
// If we know the sign bit of the sign argument, reduce to FABS/FNABS:
// copysign X, PosC --> fabs X
// copysign X, NegC --> fneg (fabs X)
Value *Fabs = Builder.CreateUnaryIntrinsic(Intrinsic::fabs,
II->getArgOperand(0), II);
if (C->isNegative())
Fabs = Builder.CreateFNegFMF(Fabs, II);
return replaceInstUsesWith(*II, Fabs);
}
break;
}
case Intrinsic::fabs: {
Value *Cond;
Constant *LHS, *RHS;

18
llvm/test/Transforms/InstCombine/copysign.ll
View File

@ -6,8 +6,8 @@ declare <3 x double> @llvm.copysign.v3f64(<3 x double>, <3 x double>)
define float @positive_sign_arg(float %x) {
; CHECK-LABEL: @positive_sign_arg(
; CHECK-NEXT: [[R:%.*]] = call arcp float @llvm.copysign.f32(float [[X:%.*]], float 0.000000e+00)
; CHECK-NEXT: ret float [[R]]
; CHECK-NEXT: [[TMP1:%.*]] = call arcp float @llvm.fabs.f32(float [[X:%.*]])
; CHECK-NEXT: ret float [[TMP1]]
;
%r = call arcp float @llvm.copysign.f32(float %x, float 0.0)
ret float %r
@ -15,8 +15,8 @@ define float @positive_sign_arg(float %x) {
define <3 x double> @positive_sign_arg_vec_splat(<3 x double> %x) {
; CHECK-LABEL: @positive_sign_arg_vec_splat(
; CHECK-NEXT: [[R:%.*]] = call ninf <3 x double> @llvm.copysign.v3f64(<3 x double> [[X:%.*]], <3 x double> <double 4.200000e+01, double 4.200000e+01, double 4.200000e+01>)
; CHECK-NEXT: ret <3 x double> [[R]]
; CHECK-NEXT: [[TMP1:%.*]] = call ninf <3 x double> @llvm.fabs.v3f64(<3 x double> [[X:%.*]])
; CHECK-NEXT: ret <3 x double> [[TMP1]]
;
%r = call ninf <3 x double> @llvm.copysign.v3f64(<3 x double> %x, <3 x double> <double 42.0, double 42.0, double 42.0>)
ret <3 x double> %r
@ -24,8 +24,9 @@ define <3 x double> @positive_sign_arg_vec_splat(<3 x double> %x) {
define float @negative_sign_arg(float %x) {
; CHECK-LABEL: @negative_sign_arg(
; CHECK-NEXT: [[R:%.*]] = call nnan float @llvm.copysign.f32(float [[X:%.*]], float -0.000000e+00)
; CHECK-NEXT: ret float [[R]]
; CHECK-NEXT: [[TMP1:%.*]] = call nnan float @llvm.fabs.f32(float [[X:%.*]])
; CHECK-NEXT: [[TMP2:%.*]] = fneg nnan float [[TMP1]]
; CHECK-NEXT: ret float [[TMP2]]
;
%r = call nnan float @llvm.copysign.f32(float %x, float -0.0)
ret float %r
@ -33,8 +34,9 @@ define float @negative_sign_arg(float %x) {
define <3 x double> @negative_sign_arg_vec_splat(<3 x double> %x) {
; CHECK-LABEL: @negative_sign_arg_vec_splat(
; CHECK-NEXT: [[R:%.*]] = call fast <3 x double> @llvm.copysign.v3f64(<3 x double> [[X:%.*]], <3 x double> <double -4.200000e+01, double -4.200000e+01, double -4.200000e+01>)
; CHECK-NEXT: ret <3 x double> [[R]]
; CHECK-NEXT: [[TMP1:%.*]] = call fast <3 x double> @llvm.fabs.v3f64(<3 x double> [[X:%.*]])
; CHECK-NEXT: [[TMP2:%.*]] = fneg fast <3 x double> [[TMP1]]
; CHECK-NEXT: ret <3 x double> [[TMP2]]
;
%r = call fast <3 x double> @llvm.copysign.v3f64(<3 x double> %x, <3 x double> <double -42.0, double -42.0, double -42.0>)
ret <3 x double> %r