RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-01-31 12:41:49 +01:00
Code Issues Projects Releases Wiki Activity

[InstCombine] use m_APInt to allow icmp ult X, C folds for splat constant vectors

Browse Source 
llvm-svn: 281107
This commit is contained in:
Sanjay Patel 2016-09-09 21:59:37 +00:00
parent f3e050c251
commit 7ac4d9da2f
5 changed files with 27 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
lib/Transforms/InstCombine/InstCombineCompares.cpp
View File

@ -3540,24 +3540,29 @@ Instruction *InstCombiner::visitICmpInst(ICmpInst &I) {
}
break;
}
case ICmpInst::ICMP_ULT:
case ICmpInst::ICMP_ULT: {
if (Op0Max.ult(Op1Min)) // A <u B -> true if max(A) < min(B)
return replaceInstUsesWith(I, ConstantInt::getTrue(I.getType()));
if (Op0Min.uge(Op1Max)) // A <u B -> false if min(A) >= max(B)
return replaceInstUsesWith(I, ConstantInt::getFalse(I.getType()));
if (Op1Min == Op0Max) // A <u B -> A != B if max(A) == min(B)
return new ICmpInst(ICmpInst::ICMP_NE, Op0, Op1);
if (ConstantInt *CI = dyn_cast<ConstantInt>(Op1)) {
if (Op1Max == Op0Min+1) // A <u C -> A == C-1 if min(A)+1 == C
return new ICmpInst(ICmpInst::ICMP_EQ, Op0,
Builder->getInt(CI->getValue()-1));
const APInt *CmpC;
if (match(Op1, m_APInt(CmpC))) {
// A <u C -> A == C-1 if min(A)+1 == C
if (Op1Max == Op0Min + 1) {
Constant *CMinus1 = ConstantInt::get(Op0->getType(), *CmpC - 1);
return new ICmpInst(ICmpInst::ICMP_EQ, Op0, CMinus1);
}
// (x <u 2147483648) -> (x >s -1) -> true if sign bit clear
if (CI->isMinValue(true))
return new ICmpInst(ICmpInst::ICMP_SGT, Op0,
Constant::getAllOnesValue(Op0->getType()));
if (CmpC->isMinSignedValue()) {
Constant *AllOnes = Constant::getAllOnesValue(Op0->getType());
return new ICmpInst(ICmpInst::ICMP_SGT, Op0, AllOnes);
}
}
break;
}
case ICmpInst::ICMP_UGT: {
if (Op0Min.ugt(Op1Max)) // A >u B -> true if min(A) > max(B)
return replaceInstUsesWith(I, ConstantInt::getTrue(I.getType()));

9
test/Transforms/InstCombine/exact.ll
View File

@ -175,10 +175,9 @@ define i1 @udiv_icmp2(i64 %X) {
ret i1 %B
}
; FIXME: missing vector fold for ult 1 -> eq 0
define <2 x i1> @udiv_icmp2_vec(<2 x i64> %X) {
; CHECK-LABEL: @udiv_icmp2_vec(
; CHECK-NEXT: [[TMP1:%.*]] = icmp ult <2 x i64> %X, <i64 1, i64 1>
; CHECK-NEXT: [[TMP1:%.*]] = icmp eq <2 x i64> %X, zeroinitializer
; CHECK-NEXT: ret <2 x i1> [[TMP1]]
;
%A = udiv exact <2 x i64> %X, <i64 5, i64 5>
@ -196,10 +195,9 @@ define i1 @sdiv_icmp1(i64 %X) {
ret i1 %B
}
; FIXME: missing vector fold for ult 1 -> eq 0
define <2 x i1> @sdiv_icmp1_vec(<2 x i64> %X) {
; CHECK-LABEL: @sdiv_icmp1_vec(
; CHECK-NEXT: [[TMP1:%.*]] = icmp ult <2 x i64> %X, <i64 1, i64 1>
; CHECK-NEXT: [[TMP1:%.*]] = icmp eq <2 x i64> %X, zeroinitializer
; CHECK-NEXT: ret <2 x i1> [[TMP1]]
;
%A = sdiv exact <2 x i64> %X, <i64 5, i64 5>
@ -257,10 +255,9 @@ define i1 @sdiv_icmp4(i64 %X) {
ret i1 %B
}
; FIXME: missing vector fold for ult 1 -> eq 0
define <2 x i1> @sdiv_icmp4_vec(<2 x i64> %X) {
; CHECK-LABEL: @sdiv_icmp4_vec(
; CHECK-NEXT: [[TMP1:%.*]] = icmp ult <2 x i64> %X, <i64 1, i64 1>
; CHECK-NEXT: [[TMP1:%.*]] = icmp eq <2 x i64> %X, zeroinitializer
; CHECK-NEXT: ret <2 x i1> [[TMP1]]
;
%A = sdiv exact <2 x i64> %X, <i64 -5, i64 -5>

11
test/Transforms/InstCombine/icmp-vec.ll
View File

@ -42,6 +42,15 @@ define <2 x i1> @ule(<2 x i8> %x) {
ret <2 x i1> %cmp
}
define <2 x i1> @ult_min_signed_value(<2 x i8> %x) {
; CHECK-LABEL: @ult_min_signed_value(
; CHECK-NEXT: [[CMP:%.*]] = icmp sgt <2 x i8> %x, <i8 -1, i8 -1>
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%cmp = icmp ult <2 x i8> %x, <i8 128, i8 128>
ret <2 x i1> %cmp
}
; Zeros are special: they're ConstantAggregateZero.
define <2 x i1> @sge_zero(<2 x i8> %x) {
@ -72,7 +81,7 @@ define <2 x i1> @sle_zero(<2 x i8> %x) {
define <2 x i1> @ule_zero(<2 x i8> %x) {
; CHECK-LABEL: @ule_zero(
; CHECK-NEXT: [[CMP:%.*]] = icmp ult <2 x i8> %x, <i8 1, i8 1>
; CHECK-NEXT: [[CMP:%.*]] = icmp eq <2 x i8> %x, zeroinitializer
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%cmp = icmp ule <2 x i8> %x, <i8 0, i8 0>

1
test/Transforms/InstCombine/icmp.ll
View File

@ -1377,7 +1377,6 @@ define i1 @icmp_mul_neq0(i32 %x) {
ret i1 %cmp
}
; FIXME: Vectors should fold the same way.
define <2 x i1> @icmp_mul_neq0_vec(<2 x i32> %x) {
; CHECK-LABEL: @icmp_mul_neq0_vec(
; CHECK-NEXT: [[CMP:%.*]] = icmp ne <2 x i32> %x, zeroinitializer

3
test/Transforms/InstCombine/pr17827.ll
View File

@ -80,11 +80,10 @@ define i1 @test_shift_and_cmp_changed2(i8 %p) {
ret i1 %cmp
}
; FIXME: icmp ult X, 1 -> icmp eq X, 0
define <2 x i1> @test_shift_and_cmp_changed2_vec(<2 x i8> %p) {
; CHECK-LABEL: @test_shift_and_cmp_changed2_vec(
; CHECK-NEXT: [[ANDP:%.*]] = and <2 x i8> %p, <i8 6, i8 6>
; CHECK-NEXT: [[CMP:%.*]] = icmp ult <2 x i8> [[ANDP]], <i8 1, i8 1>
; CHECK-NEXT: [[CMP:%.*]] = icmp eq <2 x i8> [[ANDP]], zeroinitializer
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%shlp = shl <2 x i8> %p, <i8 5, i8 5>