[llvm] [InstSimplify] Add constant folding support for `ucmp`/`scmp` intrinsics (PR #93730)
via llvm-commits
llvm-commits at lists.llvm.org
Wed May 29 13:09:47 PDT 2024
https://github.com/Poseydon42 updated https://github.com/llvm/llvm-project/pull/93730
>From a21ce63ed6e0cf2e7eec47c7c9eb906e77b3afd5 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 | 46 ++++++++++++++++++++++
1 file changed, 46 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..9869129dec945
--- /dev/null
+++ b/llvm/test/Transforms/InstSimplify/uscmp.ll
@@ -0,0 +1,46 @@
+; 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 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 <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 28f7512e65459371c0e61e7122db37475bc95fd6 Mon Sep 17 00:00:00 2001
From: Poseydon42 <vvmposeydon at gmail.com>
Date: Wed, 29 May 2024 21:08:59 +0100
Subject: [PATCH 2/2] [InstSimplify] Implement constant folding for UCMP/SCMP
intrinsics
---
llvm/lib/Analysis/ConstantFolding.cpp | 29 ++++++++++++++++++++++
llvm/test/Transforms/InstSimplify/uscmp.ll | 15 ++++-------
2 files changed, 34 insertions(+), 10 deletions(-)
diff --git a/llvm/lib/Analysis/ConstantFolding.cpp b/llvm/lib/Analysis/ConstantFolding.cpp
index 046a769453808..787c2057184cb 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 +2766,33 @@ 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 UndefValue::get(Ty);
+
+ if (ICmpInst::compare(*C0, *C1,
+ IntrinsicID == Intrinsic::scmp
+ ? ICmpInst::ICMP_SLT
+ : ICmpInst::ICMP_ULT)) {
+ return ConstantInt::get(Ty, -1, true);
+ }
+ if (ICmpInst::compare(*C0, *C1, ICmpInst::ICMP_EQ)) {
+ return ConstantInt::get(Ty, 0);
+ }
+ if (ICmpInst::compare(*C0, *C1,
+ IntrinsicID == Intrinsic::scmp
+ ? ICmpInst::ICMP_SGT
+ : ICmpInst::ICMP_UGT)) {
+ return ConstantInt::get(Ty, 1);
+ }
+
+ assert(false && "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 9869129dec945..c4e9f7bd54fd7 100644
--- a/llvm/test/Transforms/InstSimplify/uscmp.ll
+++ b/llvm/test/Transforms/InstSimplify/uscmp.ll
@@ -2,8 +2,7 @@
; 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]]
+; CHECK-NEXT: ret i8 -1
;
%1 = call i8 @llvm.scmp(i32 -7, i32 3)
ret i8 %1
@@ -11,8 +10,7 @@ define i8 @scmp_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
@@ -20,8 +18,7 @@ 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
@@ -29,8 +26,7 @@ define i100 @ucmp_gt() {
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
@@ -38,8 +34,7 @@ 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