1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 11:13:28 +01:00

[InstSimplify] More fcmp cases when comparing against negative constants.

Summary:
For known positive non-zero value X:
    fcmp uge X, -C => true
    fcmp ugt X, -C => true
    fcmp une X, -C => true
    fcmp oeq X, -C => false
    fcmp ole X, -C => false
    fcmp olt X, -C => false


Patch by Paul Walker.

Reviewers: majnemer, t.p.northover, spatel, RKSimon

Reviewed By: spatel

Subscribers: fhahn, llvm-commits

Differential Revision: https://reviews.llvm.org/D40012

llvm-svn: 319538
This commit is contained in:
Florian Hahn 2017-12-01 12:34:16 +00:00
parent 6555f7902b
commit be95d4e7f1
3 changed files with 54 additions and 48 deletions

View File

@ -3378,6 +3378,28 @@ static Value *SimplifyFCmpInst(unsigned Predicate, Value *LHS, Value *RHS,
default:
break;
}
} else if (C->isNegative()) {
assert(!C->isNaN() && "Unexpected NaN constant!");
// TODO: We can catch more cases by using a range check rather than
// relying on CannotBeOrderedLessThanZero.
switch (Pred) {
case FCmpInst::FCMP_UGE:
case FCmpInst::FCMP_UGT:
case FCmpInst::FCMP_UNE:
// (X >= 0) implies (X > C) when (C < 0)
if (CannotBeOrderedLessThanZero(LHS, Q.TLI))
return getTrue(RetTy);
break;
case FCmpInst::FCMP_OEQ:
case FCmpInst::FCMP_OLE:
case FCmpInst::FCMP_OLT:
// (X >= 0) implies !(X < C) when (C < 0)
if (CannotBeOrderedLessThanZero(LHS, Q.TLI))
return getFalse(RetTy);
break;
default:
break;
}
}
}

View File

@ -198,14 +198,6 @@ define i1 @i32_cast_cmp_ole_int_0_sitofp(i32 %i) {
ret i1 %cmp
}
; CHECK-LABEL: @i32_cast_cmp_olt_int_0_uitofp(
; CHECK: ret i1 false
define i1 @i32_cast_cmp_olt_int_0_uitofp(i32 %i) {
%f = uitofp i32 %i to float
%cmp = fcmp olt float %f, 0.0
ret i1 %cmp
}
; CHECK-LABEL: @i32_cast_cmp_olt_int_0_sitofp(
; CHECK: icmp slt i32 %i, 0
; CHECK-NEXT: ret
@ -339,16 +331,6 @@ define i1 @i32_cast_cmp_oeq_int_i32umax_sitofp(i32 %i) {
ret i1 %cmp
}
; CHECK-LABEL: @i32_cast_cmp_oeq_int_i32imin_uitofp(
; CHECK: uitofp
; CHECK: fcmp oeq
; CHECK-NEXT: ret
define i1 @i32_cast_cmp_oeq_int_i32imin_uitofp(i32 %i) {
%f = uitofp i32 %i to float
%cmp = fcmp oeq float %f, 0xC1E0000000000000
ret i1 %cmp
}
; CHECK-LABEL: @i32_cast_cmp_oeq_int_i32imin_sitofp(
; CHECK: sitofp
; CHECK: fcmp oeq
@ -379,16 +361,6 @@ define i1 @i32_cast_cmp_oeq_int_i32imax_sitofp(i32 %i) {
ret i1 %cmp
}
; CHECK-LABEL: @i32_cast_cmp_oeq_int_negi32umax_uitofp(
; CHECK: uitofp
; CHECK: fcmp oeq
; CHECK-NEXT: ret
define i1 @i32_cast_cmp_oeq_int_negi32umax_uitofp(i32 %i) {
%f = uitofp i32 %i to float
%cmp = fcmp oeq float %f, 0xC1F0000000000000
ret i1 %cmp
}
; 32-bit signed integer cannot possibly round to -1<<32
; CHECK-LABEL: @i32_cast_cmp_oeq_int_negi32umax_sitofp(
; CHECK-NEXT: ret i1 false

View File

@ -169,8 +169,10 @@ define i1 @one_with_self(double %arg) {
; and between uge and olt, to give reasonble coverage
; without combinatorial explosion.
declare half @llvm.fabs.f16(half)
declare float @llvm.fabs.f32(float)
declare double @llvm.fabs.f64(double)
declare <2 x float> @llvm.fabs.v2f32(<2 x float>)
declare <2 x double> @llvm.fabs.v2f64(<2 x double>)
declare float @llvm.sqrt.f32(float)
declare double @llvm.powi.f64(double,i32)
@ -276,45 +278,55 @@ define i1 @orderedLessZeroMaxNum(float, float) {
define i1 @known_positive_olt_with_negative_constant(double %a) {
; CHECK-LABEL: @known_positive_olt_with_negative_constant(
; CHECK-NEXT: [[CALL:%.*]] = call double @llvm.fabs.f64(double %a)
; CHECK-NEXT: [[CMP:%.*]] = fcmp olt double [[CALL]], -1.000000e+00
; CHECK-NEXT: ret i1 [[CMP]]
; CHECK-NEXT: ret i1 false
;
%call = call double @llvm.fabs.f64(double %a)
%cmp = fcmp olt double %call, -1.0
ret i1 %cmp
}
define <2 x i1> @known_positive_ole_with_negative_constant_splat_vec(<2 x double> %a) {
define <2 x i1> @known_positive_ole_with_negative_constant_splat_vec(<2 x i32> %a) {
; CHECK-LABEL: @known_positive_ole_with_negative_constant_splat_vec(
; CHECK-NEXT: [[CALL:%.*]] = call <2 x double> @llvm.fabs.v2f64(<2 x double> %a)
; CHECK-NEXT: [[CMP:%.*]] = fcmp ole <2 x double> [[CALL]], <double -2.000000e+00, double -2.000000e+00>
; CHECK-NEXT: ret <2 x i1> [[CMP]]
; CHECK-NEXT: ret <2 x i1> zeroinitializer
;
%call = call <2 x double> @llvm.fabs.v2f64(<2 x double> %a)
%call = uitofp <2 x i32> %a to <2 x double>
%cmp = fcmp ole <2 x double> %call, <double -2.0, double -2.0>
ret <2 x i1> %cmp
}
define i1 @known_positive_ugt_with_negative_constant(double %a) {
define i1 @known_positive_ugt_with_negative_constant(i32 %a) {
; CHECK-LABEL: @known_positive_ugt_with_negative_constant(
; CHECK-NEXT: [[CALL:%.*]] = call double @llvm.fabs.f64(double %a)
; CHECK-NEXT: [[CMP:%.*]] = fcmp ugt double [[CALL]], -3.000000e+00
; CHECK-NEXT: ret i1 [[CMP]]
; CHECK-NEXT: ret i1 true
;
%call = call double @llvm.fabs.f64(double %a)
%cmp = fcmp ugt double %call, -3.0
%call = uitofp i32 %a to float
%cmp = fcmp ugt float %call, -3.0
ret i1 %cmp
}
define <2 x i1> @known_positive_uge_with_negative_constant_splat_vec(<2 x double> %a) {
define <2 x i1> @known_positive_uge_with_negative_constant_splat_vec(<2 x float> %a) {
; CHECK-LABEL: @known_positive_uge_with_negative_constant_splat_vec(
; CHECK-NEXT: [[CALL:%.*]] = call <2 x double> @llvm.fabs.v2f64(<2 x double> %a)
; CHECK-NEXT: [[CMP:%.*]] = fcmp uge <2 x double> [[CALL]], <double -4.000000e+00, double -4.000000e+00>
; CHECK-NEXT: ret <2 x i1> [[CMP]]
; CHECK-NEXT: ret <2 x i1> <i1 true, i1 true>
;
%call = call <2 x double> @llvm.fabs.v2f64(<2 x double> %a)
%cmp = fcmp uge <2 x double> %call, <double -4.0, double -4.0>
%call = call <2 x float> @llvm.fabs.v2f32(<2 x float> %a)
%cmp = fcmp uge <2 x float> %call, <float -4.0, float -4.0>
ret <2 x i1> %cmp
}
define i1 @known_positive_oeq_with_negative_constant(half %a) {
; CHECK-LABEL: @known_positive_oeq_with_negative_constant(
; CHECK-NEXT: ret i1 false
;
%call = call half @llvm.fabs.f16(half %a)
%cmp = fcmp oeq half %call, -5.0
ret i1 %cmp
}
define <2 x i1> @known_positive_une_with_negative_constant_splat_vec(<2 x i32> %a) {
; CHECK-LABEL: @known_positive_une_with_negative_constant_splat_vec(
; CHECK-NEXT: ret <2 x i1> <i1 true, i1 true>
;
%call = uitofp <2 x i32> %a to <2 x half>
%cmp = fcmp une <2 x half> %call, <half -6.0, half -6.0>
ret <2 x i1> %cmp
}