[llvm] [InstSimplify] Add constant folding support for `ucmp`/`scmp` intrinsics (PR #93730)
via llvm-commits
llvm-commits at lists.llvm.org
Wed May 29 14:35:29 PDT 2024
https://github.com/Poseydon42 updated https://github.com/llvm/llvm-project/pull/93730
>From 19610f5f8420c7431d47fe118d991f66dcf27115 Mon Sep 17 00:00:00 2001
From: Poseydon42 <vvmposeydon at gmail.com>
Date: Wed, 29 May 2024 20:36:09 +0100
Subject: [PATCH 1/2] [InstSimplify] Add constant folding tests for UCMP/SCMP
intrinsics
---
llvm/test/Transforms/InstSimplify/uscmp.ll | 92 ++++++++++++++++++++++
1 file changed, 92 insertions(+)
create mode 100644 llvm/test/Transforms/InstSimplify/uscmp.ll
diff --git a/llvm/test/Transforms/InstSimplify/uscmp.ll b/llvm/test/Transforms/InstSimplify/uscmp.ll
new file mode 100644
index 0000000000000..e97947773886d
--- /dev/null
+++ b/llvm/test/Transforms/InstSimplify/uscmp.ll
@@ -0,0 +1,92 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -passes=instsimplify -S | FileCheck %s
+
+define i8 @scmp_lt() {
+; CHECK-LABEL: define i8 @scmp_lt() {
+; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.scmp.i8.i32(i32 -7, i32 3)
+; CHECK-NEXT: ret i8 [[TMP1]]
+;
+ %1 = call i8 @llvm.scmp(i32 -7, i32 3)
+ ret i8 %1
+}
+
+define i8 @scmp_eq() {
+; CHECK-LABEL: define i8 @scmp_eq() {
+; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.scmp.i8.i32(i32 2, i32 2)
+; CHECK-NEXT: ret i8 [[TMP1]]
+;
+ %1 = call i8 @llvm.scmp(i32 2, i32 2)
+ ret i8 %1
+}
+
+define i8 @scmp_gt() {
+; CHECK-LABEL: define i8 @scmp_gt() {
+; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.scmp.i8.i32(i32 2, i32 -7)
+; CHECK-NEXT: ret i8 [[TMP1]]
+;
+ %1 = call i8 @llvm.scmp(i32 2, i32 -7)
+ ret i8 %1
+}
+
+define i8 @ucmp_lt() {
+; CHECK-LABEL: define i8 @ucmp_lt() {
+; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.ucmp.i8.i32(i32 7, i32 12)
+; CHECK-NEXT: ret i8 [[TMP1]]
+;
+ %1 = call i8 @llvm.ucmp(i32 7, i32 12)
+ ret i8 %1
+}
+
+define i2 @ucmp_eq() {
+; CHECK-LABEL: define i2 @ucmp_eq() {
+; CHECK-NEXT: [[TMP1:%.*]] = call i2 @llvm.ucmp.i2.i32(i32 12, i32 12)
+; CHECK-NEXT: ret i2 [[TMP1]]
+;
+ %1 = call i2 @llvm.ucmp(i32 12, i32 12)
+ ret i2 %1
+}
+
+define i100 @ucmp_gt() {
+; CHECK-LABEL: define i100 @ucmp_gt() {
+; CHECK-NEXT: [[TMP1:%.*]] = call i100 @llvm.ucmp.i100.i32(i32 7, i32 3)
+; CHECK-NEXT: ret i100 [[TMP1]]
+;
+ %1 = call i100 @llvm.ucmp(i32 7, i32 3)
+ ret i100 %1
+}
+
+define i8 @ucmp_poison() {
+; CHECK-LABEL: define i8 @ucmp_poison() {
+; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.ucmp.i8.i32(i32 poison, i32 5)
+; CHECK-NEXT: ret i8 [[TMP1]]
+;
+ %1 = call i8 @llvm.ucmp(i32 poison, i32 5)
+ ret i8 %1
+}
+
+define i8 @scmp_undef() {
+; CHECK-LABEL: define i8 @scmp_undef() {
+; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.scmp.i8.i32(i32 undef, i32 -12)
+; CHECK-NEXT: ret i8 [[TMP1]]
+;
+ %1 = call i8 @llvm.scmp(i32 undef, i32 -12)
+ ret i8 %1
+}
+
+define <4 x i8> @ucmp_lt_splat() {
+; CHECK-LABEL: define <4 x i8> @ucmp_lt_splat() {
+; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i8> @llvm.ucmp.v4i8.v4i32(<4 x i32> <i32 1, i32 1, i32 1, i32 1>, <4 x i32> <i32 3, i32 3, i32 3, i32 3>)
+; CHECK-NEXT: ret <4 x i8> [[TMP1]]
+;
+ %1 = call <4 x i8> @llvm.ucmp(<4 x i32> splat(i32 1), <4 x i32> splat(i32 3))
+ ret <4 x i8> %1
+}
+
+define <4 x i8> @scmp_nonsplat() {
+; CHECK-LABEL: define <4 x i8> @scmp_nonsplat() {
+; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i8> @llvm.scmp.v4i8.v4i32(<4 x i32> <i32 0, i32 1, i32 2, i32 3>, <4 x i32> <i32 -1, i32 1, i32 -2, i32 4>)
+; CHECK-NEXT: ret <4 x i8> [[TMP1]]
+;
+ %1 = call <4 x i8> @llvm.scmp(<4 x i32> <i32 0, i32 1, i32 2, i32 3>, <4 x i32> <i32 -1, i32 1, i32 -2, i32 4>)
+ ret <4 x i8> %1
+}
>From d926bc10f81190ab3b9a285dc0d06d900ab454d5 Mon Sep 17 00:00:00 2001
From: Poseydon42 <vvmposeydon at gmail.com>
Date: Wed, 29 May 2024 22:34:39 +0100
Subject: [PATCH 2/2] [InstSimplify] Implement constant folding for UCMP/SCMP
intrinsics
---
llvm/lib/Analysis/ConstantFolding.cpp | 26 +++++++++++++++++++
llvm/test/Transforms/InstSimplify/uscmp.ll | 30 ++++++++--------------
2 files changed, 36 insertions(+), 20 deletions(-)
diff --git a/llvm/lib/Analysis/ConstantFolding.cpp b/llvm/lib/Analysis/ConstantFolding.cpp
index 046a769453808..3b1b3e37babe7 100644
--- a/llvm/lib/Analysis/ConstantFolding.cpp
+++ b/llvm/lib/Analysis/ConstantFolding.cpp
@@ -1506,6 +1506,8 @@ bool llvm::canConstantFoldCallTo(const CallBase *Call, const Function *F) {
case Intrinsic::smin:
case Intrinsic::umax:
case Intrinsic::umin:
+ case Intrinsic::scmp:
+ case Intrinsic::ucmp:
case Intrinsic::sadd_with_overflow:
case Intrinsic::uadd_with_overflow:
case Intrinsic::ssub_with_overflow:
@@ -2764,6 +2766,30 @@ static Constant *ConstantFoldIntrinsicCall2(Intrinsic::ID IntrinsicID, Type *Ty,
? *C0
: *C1);
+ case Intrinsic::scmp:
+ case Intrinsic::ucmp:
+ if (isa<PoisonValue>(Operands[0]) || isa<PoisonValue>(Operands[1]))
+ return PoisonValue::get(Ty);
+
+ if (!C0 || !C1)
+ return ConstantInt::get(Ty, 0);
+
+ if (ICmpInst::compare(*C0, *C1,
+ IntrinsicID == Intrinsic::scmp
+ ? ICmpInst::ICMP_SLT
+ : ICmpInst::ICMP_ULT))
+ return ConstantInt::get(Ty, -1, true);
+ if (*C0 == *C1)
+ return ConstantInt::get(Ty, 0);
+ if (ICmpInst::compare(*C0, *C1,
+ IntrinsicID == Intrinsic::scmp
+ ? ICmpInst::ICMP_SGT
+ : ICmpInst::ICMP_UGT))
+ return ConstantInt::get(Ty, 1);
+
+ llvm_unreachable("Integer values must compare as equal, or one must be "
+ "less than the other");
+
case Intrinsic::usub_with_overflow:
case Intrinsic::ssub_with_overflow:
// X - undef -> { 0, false }
diff --git a/llvm/test/Transforms/InstSimplify/uscmp.ll b/llvm/test/Transforms/InstSimplify/uscmp.ll
index e97947773886d..af55bb0a12e35 100644
--- a/llvm/test/Transforms/InstSimplify/uscmp.ll
+++ b/llvm/test/Transforms/InstSimplify/uscmp.ll
@@ -3,8 +3,7 @@
define i8 @scmp_lt() {
; CHECK-LABEL: define i8 @scmp_lt() {
-; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.scmp.i8.i32(i32 -7, i32 3)
-; CHECK-NEXT: ret i8 [[TMP1]]
+; CHECK-NEXT: ret i8 -1
;
%1 = call i8 @llvm.scmp(i32 -7, i32 3)
ret i8 %1
@@ -12,8 +11,7 @@ define i8 @scmp_lt() {
define i8 @scmp_eq() {
; CHECK-LABEL: define i8 @scmp_eq() {
-; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.scmp.i8.i32(i32 2, i32 2)
-; CHECK-NEXT: ret i8 [[TMP1]]
+; CHECK-NEXT: ret i8 0
;
%1 = call i8 @llvm.scmp(i32 2, i32 2)
ret i8 %1
@@ -21,8 +19,7 @@ define i8 @scmp_eq() {
define i8 @scmp_gt() {
; CHECK-LABEL: define i8 @scmp_gt() {
-; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.scmp.i8.i32(i32 2, i32 -7)
-; CHECK-NEXT: ret i8 [[TMP1]]
+; CHECK-NEXT: ret i8 1
;
%1 = call i8 @llvm.scmp(i32 2, i32 -7)
ret i8 %1
@@ -30,8 +27,7 @@ define i8 @scmp_gt() {
define i8 @ucmp_lt() {
; CHECK-LABEL: define i8 @ucmp_lt() {
-; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.ucmp.i8.i32(i32 7, i32 12)
-; CHECK-NEXT: ret i8 [[TMP1]]
+; CHECK-NEXT: ret i8 -1
;
%1 = call i8 @llvm.ucmp(i32 7, i32 12)
ret i8 %1
@@ -39,8 +35,7 @@ define i8 @ucmp_lt() {
define i2 @ucmp_eq() {
; CHECK-LABEL: define i2 @ucmp_eq() {
-; CHECK-NEXT: [[TMP1:%.*]] = call i2 @llvm.ucmp.i2.i32(i32 12, i32 12)
-; CHECK-NEXT: ret i2 [[TMP1]]
+; CHECK-NEXT: ret i2 0
;
%1 = call i2 @llvm.ucmp(i32 12, i32 12)
ret i2 %1
@@ -48,8 +43,7 @@ define i2 @ucmp_eq() {
define i100 @ucmp_gt() {
; CHECK-LABEL: define i100 @ucmp_gt() {
-; CHECK-NEXT: [[TMP1:%.*]] = call i100 @llvm.ucmp.i100.i32(i32 7, i32 3)
-; CHECK-NEXT: ret i100 [[TMP1]]
+; CHECK-NEXT: ret i100 1
;
%1 = call i100 @llvm.ucmp(i32 7, i32 3)
ret i100 %1
@@ -57,8 +51,7 @@ define i100 @ucmp_gt() {
define i8 @ucmp_poison() {
; CHECK-LABEL: define i8 @ucmp_poison() {
-; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.ucmp.i8.i32(i32 poison, i32 5)
-; CHECK-NEXT: ret i8 [[TMP1]]
+; CHECK-NEXT: ret i8 poison
;
%1 = call i8 @llvm.ucmp(i32 poison, i32 5)
ret i8 %1
@@ -66,8 +59,7 @@ define i8 @ucmp_poison() {
define i8 @scmp_undef() {
; CHECK-LABEL: define i8 @scmp_undef() {
-; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.scmp.i8.i32(i32 undef, i32 -12)
-; CHECK-NEXT: ret i8 [[TMP1]]
+; CHECK-NEXT: ret i8 0
;
%1 = call i8 @llvm.scmp(i32 undef, i32 -12)
ret i8 %1
@@ -75,8 +67,7 @@ define i8 @scmp_undef() {
define <4 x i8> @ucmp_lt_splat() {
; CHECK-LABEL: define <4 x i8> @ucmp_lt_splat() {
-; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i8> @llvm.ucmp.v4i8.v4i32(<4 x i32> <i32 1, i32 1, i32 1, i32 1>, <4 x i32> <i32 3, i32 3, i32 3, i32 3>)
-; CHECK-NEXT: ret <4 x i8> [[TMP1]]
+; CHECK-NEXT: ret <4 x i8> <i8 -1, i8 -1, i8 -1, i8 -1>
;
%1 = call <4 x i8> @llvm.ucmp(<4 x i32> splat(i32 1), <4 x i32> splat(i32 3))
ret <4 x i8> %1
@@ -84,8 +75,7 @@ define <4 x i8> @ucmp_lt_splat() {
define <4 x i8> @scmp_nonsplat() {
; CHECK-LABEL: define <4 x i8> @scmp_nonsplat() {
-; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i8> @llvm.scmp.v4i8.v4i32(<4 x i32> <i32 0, i32 1, i32 2, i32 3>, <4 x i32> <i32 -1, i32 1, i32 -2, i32 4>)
-; CHECK-NEXT: ret <4 x i8> [[TMP1]]
+; CHECK-NEXT: ret <4 x i8> <i8 1, i8 0, i8 1, i8 -1>
;
%1 = call <4 x i8> @llvm.scmp(<4 x i32> <i32 0, i32 1, i32 2, i32 3>, <4 x i32> <i32 -1, i32 1, i32 -2, i32 4>)
ret <4 x i8> %1
More information about the llvm-commits
mailing list