[llvm] [ARM] Allow usubo and uaddo to happen for any legal type (PR #185494)
via llvm-commits
llvm-commits at lists.llvm.org
Fri Apr 10 15:54:17 PDT 2026
https://github.com/SiliconA-Z updated https://github.com/llvm/llvm-project/pull/185494
>From 2cefe0114e26053b629f0c4a9d5427b32a43f6c5 Mon Sep 17 00:00:00 2001
From: AZero13 <gfunni234 at gmail.com>
Date: Mon, 9 Mar 2026 15:04:31 -0400
Subject: [PATCH] [ARM] Allow usubo and uaddo to happen for any legal type
With XOR 1, instead of sub 1, we can have these cancel each other out better, which prevents regressions.
---
llvm/lib/Target/ARM/ARMISelLowering.cpp | 11 ++---
llvm/lib/Target/ARM/ARMISelLowering.h | 5 +-
.../CodeGen/ARM/atomicrmw-uinc-udec-wrap.ll | 8 ++--
llvm/test/CodeGen/ARM/select_const.ll | 47 +++++++------------
.../CodeGen/Thumb/scheduler-clone-cpsr-def.ll | 33 +++++--------
5 files changed, 43 insertions(+), 61 deletions(-)
diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
index 0ad6c07f36868..574de121451f9 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -4728,8 +4728,8 @@ static SDValue valueToCarryFlag(SDValue Value, SelectionDAG &DAG, bool Invert) {
EVT VT = Value.getValueType();
if (Invert)
- Value = DAG.getNode(ISD::SUB, DL, MVT::i32,
- DAG.getConstant(1, DL, MVT::i32), Value);
+ Value =
+ DAG.getNode(ISD::XOR, DL, VT, Value, DAG.getConstant(1, DL, MVT::i32));
SDValue Cmp = DAG.getNode(ARMISD::SUBC, DL, DAG.getVTList(VT, MVT::i32),
Value, DAG.getConstant(1, DL, VT));
@@ -4745,7 +4745,7 @@ static SDValue carryFlagToValue(SDValue Flags, EVT VT, SelectionDAG &DAG,
SDValue BoolCarry = DAG.getNode(
ARMISD::ADDE, DL, DAG.getVTList(VT, MVT::i32),
DAG.getConstant(0, DL, VT), DAG.getConstant(0, DL, VT), Flags);
- return DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(1, DL, VT), BoolCarry);
+ return DAG.getNode(ISD::XOR, DL, VT, BoolCarry, DAG.getConstant(1, DL, VT));
}
// Now convert the carry flag into a boolean carry. We do this
@@ -18525,9 +18525,8 @@ ARMTargetLowering::PerformCMOVCombine(SDNode *N, SelectionDAG &DAG) const {
SDValue Neg = DAG.getNode(ISD::USUBO, dl, VTs, FalseVal, Sub);
// ISD::USUBO_CARRY returns a borrow but we want the carry here
// actually.
- SDValue Carry =
- DAG.getNode(ISD::SUB, dl, MVT::i32,
- DAG.getConstant(1, dl, MVT::i32), Neg.getValue(1));
+ SDValue Carry = DAG.getNode(ISD::XOR, dl, MVT::i32, Neg.getValue(1),
+ DAG.getConstant(1, dl, MVT::i32));
Res = DAG.getNode(ISD::UADDO_CARRY, dl, VTs, Sub, Neg, Carry);
}
} else if (CC == ARMCC::NE && !isNullConstant(RHS) &&
diff --git a/llvm/lib/Target/ARM/ARMISelLowering.h b/llvm/lib/Target/ARM/ARMISelLowering.h
index e58d872c548e4..2e268cf4f7ce2 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.h
+++ b/llvm/lib/Target/ARM/ARMISelLowering.h
@@ -333,8 +333,9 @@ class VectorType;
bool shouldFormOverflowOp(unsigned Opcode, EVT VT,
bool MathUsed) const override {
- // Using overflow ops for overflow checks only should beneficial on ARM.
- return TargetLowering::shouldFormOverflowOp(Opcode, VT, true);
+ if (VT.isVector())
+ return false;
+ return !isOperationExpand(Opcode, VT);
}
bool shouldReassociateReduction(unsigned Opc, EVT VT) const override {
diff --git a/llvm/test/CodeGen/ARM/atomicrmw-uinc-udec-wrap.ll b/llvm/test/CodeGen/ARM/atomicrmw-uinc-udec-wrap.ll
index 433fb325a7349..c37afeeea375d 100644
--- a/llvm/test/CodeGen/ARM/atomicrmw-uinc-udec-wrap.ll
+++ b/llvm/test/CodeGen/ARM/atomicrmw-uinc-udec-wrap.ll
@@ -147,11 +147,11 @@ define i32 @atomicrmw_udec_wrap_i32(ptr %ptr, i32 %val) {
; CHECK-NEXT: .LBB6_1: @ %atomicrmw.start
; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
; CHECK-NEXT: ldrex r12, [r0]
-; CHECK-NEXT: mov r3, r1
; CHECK-NEXT: cmp r12, r1
-; CHECK-NEXT: subls r3, r12, #1
-; CHECK-NEXT: cmp r12, #0
-; CHECK-NEXT: moveq r3, r1
+; CHECK-NEXT: sub r3, r12, #1
+; CHECK-NEXT: movhi r3, r1
+; CHECK-NEXT: cmp r12, #1
+; CHECK-NEXT: movlo r3, r1
; CHECK-NEXT: strex r2, r3, [r0]
; CHECK-NEXT: cmp r2, #0
; CHECK-NEXT: bne .LBB6_1
diff --git a/llvm/test/CodeGen/ARM/select_const.ll b/llvm/test/CodeGen/ARM/select_const.ll
index 77b67cc40f707..9b896eb7672d9 100644
--- a/llvm/test/CodeGen/ARM/select_const.ll
+++ b/llvm/test/CodeGen/ARM/select_const.ll
@@ -756,46 +756,35 @@ define i64 @opaque_constant2(i1 %cond, i64 %x) {
define i64 @func(i64 %arg) {
; ARM-LABEL: func:
; ARM: @ %bb.0: @ %entry
-; ARM-NEXT: adds r0, r0, #1
-; ARM-NEXT: mov r2, #0
-; ARM-NEXT: adcs r0, r1, #0
+; ARM-NEXT: and r0, r0, r1
; ARM-NEXT: mov r1, #0
-; ARM-NEXT: adcs r0, r2, #0
-; ARM-NEXT: movne r0, #8
+; ARM-NEXT: cmn r0, #1
+; ARM-NEXT: mov r0, #0
+; ARM-NEXT: moveq r0, #8
; ARM-NEXT: mov pc, lr
;
; THUMB2-LABEL: func:
; THUMB2: @ %bb.0: @ %entry
+; THUMB2-NEXT: ands r0, r1
+; THUMB2-NEXT: movs r1, #0
; THUMB2-NEXT: adds r0, #1
-; THUMB2-NEXT: mov.w r2, #0
-; THUMB2-NEXT: adcs r0, r1, #0
-; THUMB2-NEXT: mov.w r1, #0
-; THUMB2-NEXT: adcs r0, r2, #0
-; THUMB2-NEXT: it ne
-; THUMB2-NEXT: movne r0, #8
+; THUMB2-NEXT: mov.w r0, #0
+; THUMB2-NEXT: it eq
+; THUMB2-NEXT: moveq r0, #8
; THUMB2-NEXT: bx lr
;
; THUMB-LABEL: func:
; THUMB: @ %bb.0: @ %entry
-; THUMB-NEXT: .save {r4, lr}
-; THUMB-NEXT: push {r4, lr}
-; THUMB-NEXT: movs r2, #0
-; THUMB-NEXT: adds r3, r0, #1
-; THUMB-NEXT: mov r12, r1
-; THUMB-NEXT: mov r3, r12
-; THUMB-NEXT: adcs r3, r2
-; THUMB-NEXT: mov r12, r2
-; THUMB-NEXT: mov r3, r12
-; THUMB-NEXT: adcs r3, r2
-; THUMB-NEXT: subs r4, r3, #1
+; THUMB-NEXT: ands r0, r1
+; THUMB-NEXT: movs r1, #0
; THUMB-NEXT: adds r0, r0, #1
-; THUMB-NEXT: adcs r1, r2
-; THUMB-NEXT: sbcs r3, r4
-; THUMB-NEXT: lsls r0, r3, #3
-; THUMB-NEXT: movs r1, r2
-; THUMB-NEXT: pop {r4}
-; THUMB-NEXT: pop {r2}
-; THUMB-NEXT: bx r2
+; THUMB-NEXT: beq .LBB26_2
+; THUMB-NEXT: @ %bb.1: @ %entry
+; THUMB-NEXT: movs r0, r1
+; THUMB-NEXT: bx lr
+; THUMB-NEXT: .LBB26_2:
+; THUMB-NEXT: movs r0, #8
+; THUMB-NEXT: bx lr
entry:
%0 = add i64 %arg, 1
%1 = icmp ult i64 %0, 1
diff --git a/llvm/test/CodeGen/Thumb/scheduler-clone-cpsr-def.ll b/llvm/test/CodeGen/Thumb/scheduler-clone-cpsr-def.ll
index 31e54c43c1e5f..4c92a00020475 100644
--- a/llvm/test/CodeGen/Thumb/scheduler-clone-cpsr-def.ll
+++ b/llvm/test/CodeGen/Thumb/scheduler-clone-cpsr-def.ll
@@ -11,27 +11,20 @@
define i64 @f(i64 %x2, i32 %z) {
; CHECK-LABEL: f:
; CHECK: @ %bb.0:
-; CHECK-NEXT: .save {r4, r5, r7, lr}
-; CHECK-NEXT: push {r4, r5, r7, lr}
-; CHECK-NEXT: movs r2, #0
-; CHECK-NEXT: subs r3, r0, #1
-; CHECK-NEXT: mov r3, r1
-; CHECK-NEXT: sbcs r3, r2
-; CHECK-NEXT: mov r3, r2
+; CHECK-NEXT: .save {r4, lr}
+; CHECK-NEXT: push {r4, lr}
+; CHECK-NEXT: mov r2, r0
+; CHECK-NEXT: orrs r2, r1
+; CHECK-NEXT: rsbs r3, r2, #0
; CHECK-NEXT: adcs r3, r2
-; CHECK-NEXT: movs r4, #30
-; CHECK-NEXT: subs r5, r0, #1
-; CHECK-NEXT: mov r5, r1
-; CHECK-NEXT: sbcs r5, r2
-; CHECK-NEXT: adcs r4, r2
-; CHECK-NEXT: lsls r2, r1, #1
-; CHECK-NEXT: lsls r2, r4
-; CHECK-NEXT: movs r4, #1
-; CHECK-NEXT: eors r4, r3
-; CHECK-NEXT: lsrs r0, r4
-; CHECK-NEXT: orrs r0, r2
-; CHECK-NEXT: lsrs r1, r4
-; CHECK-NEXT: pop {r4, r5, r7, pc}
+; CHECK-NEXT: lsrs r0, r3
+; CHECK-NEXT: movs r2, #31
+; CHECK-NEXT: eors r2, r3
+; CHECK-NEXT: lsls r4, r1, #1
+; CHECK-NEXT: lsls r4, r2
+; CHECK-NEXT: orrs r0, r4
+; CHECK-NEXT: lsrs r1, r3
+; CHECK-NEXT: pop {r4, pc}
%x3 = add nsw i64 %x2, -1
%x8 = icmp ne i64 %x2, 0
%x9 = xor i1 %x8, true
More information about the llvm-commits
mailing list