This is an archive of the discontinued LLVM Phabricator instance.

Differential D68976

[ARM] Add qadd lowering from a sadd_sat
ClosedPublic

Authored by dmgreen on Oct 15 2019, 2:40 AM.

Details

Reviewers
samparker
SjoerdMeijer
efriedma
Commits
rGd7b77f2203dd: [ARM] Add qadd lowering from a sadd_sat
rL375411: [ARM] Add qadd lowering from a sadd_sat
Summary

This lowers a sadd_sat to a qadd by treating it as legal. Also adds qsub at the same time.

The qadd instruction sets the q flag, but we already have many cases where we do not model this explicitly.

Diff Detail

Event Timeline

dmgreen created this revision.Oct 15 2019, 2:40 AM
Herald added a project: Restricted Project. · View Herald TranscriptOct 15 2019, 2:40 AM
Herald added subscribers: hiraditya, kristof.beyls. · View Herald Transcript
samparker added a comment.Oct 15 2019, 6:16 AM

How about adding support for the QD* versions as well?

samparker added inline comments.Oct 15 2019, 8:00 AM
llvm/lib/Target/ARM/ARMInstrInfo.td
3758

I think these should be V5TE patterns.

dmgreen updated this revision to Diff 225074.Oct 15 2019, 10:23 AM
dmgreen marked 2 inline comments as done.
dmgreen added inline comments.
llvm/lib/Target/ARM/ARMInstrInfo.td
3758

I've added a hasBaseDSP method to do the check in ISel. Let me know if there's a better name for it.

samparker added inline comments.Oct 16 2019, 2:34 AM
llvm/test/CodeGen/ARM/qdadd.ll
45

Still need the tablegen patterns!

dmgreen marked an inline comment as done.Oct 16 2019, 6:00 AM
dmgreen added inline comments.
llvm/test/CodeGen/ARM/qdadd.ll
45

D68999. I thought I had put that somewhere here, but perhaps I had just presumed you would get the notification.

I rebased these tests, which is why they appear here too. That patch adds the patterns.

samparker added a comment.Oct 16 2019, 6:09 AM

LGTM.

llvm/test/CodeGen/ARM/qdadd.ll
45

Ah!

This revision was not accepted when it landed; it landed in state Needs Review.Oct 21 2019, 5:39 AM
Closed by commit rGd7b77f2203dd: [ARM] Add qadd lowering from a sadd_sat (authored by dmgreen). · Explain Why
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
llvm/
lib/
Target/
ARM/
9 lines
4 lines
4 lines
6 lines
test/
CodeGen/
ARM/
440 lines
154 lines
91 lines
400 lines
98 lines

Diff 225860

llvm/lib/Target/ARM/ARMISelLowering.cpp

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 1,021 Lines • ▼ Show 20 LinesARMTargetLowering::ARMTargetLowering(const TargetMachine &TM,
setOperationAction(ISD::ADDCARRY, MVT::i32, Custom); setOperationAction(ISD::ADDCARRY, MVT::i32, Custom);
setOperationAction(ISD::SUBCARRY, MVT::i32, Custom); setOperationAction(ISD::SUBCARRY, MVT::i32, Custom);
if (Subtarget->hasDSP()) { if (Subtarget->hasDSP()) {
setOperationAction(ISD::SADDSAT, MVT::i8, Custom); setOperationAction(ISD::SADDSAT, MVT::i8, Custom);
setOperationAction(ISD::SSUBSAT, MVT::i8, Custom); setOperationAction(ISD::SSUBSAT, MVT::i8, Custom);
setOperationAction(ISD::SADDSAT, MVT::i16, Custom); setOperationAction(ISD::SADDSAT, MVT::i16, Custom);
setOperationAction(ISD::SSUBSAT, MVT::i16, Custom); setOperationAction(ISD::SSUBSAT, MVT::i16, Custom);
} }
if (Subtarget->hasBaseDSP()) {
setOperationAction(ISD::SADDSAT, MVT::i32, Legal);
setOperationAction(ISD::SSUBSAT, MVT::i32, Legal);
}
// i64 operation support. // i64 operation support.
setOperationAction(ISD::MUL, MVT::i64, Expand); setOperationAction(ISD::MUL, MVT::i64, Expand);
setOperationAction(ISD::MULHU, MVT::i32, Expand); setOperationAction(ISD::MULHU, MVT::i32, Expand);
if (Subtarget->isThumb1Only()) { if (Subtarget->isThumb1Only()) {
setOperationAction(ISD::UMUL_LOHI, MVT::i32, Expand); setOperationAction(ISD::UMUL_LOHI, MVT::i32, Expand);
setOperationAction(ISD::SMUL_LOHI, MVT::i32, Expand); setOperationAction(ISD::SMUL_LOHI, MVT::i32, Expand);
} }
▲ Show 20 Lines • Show All 10,047 Lines • ▼ Show 20 Lines if (V->getOpcode() == ISD::UMUL_LOHI ||
V->getOpcode() == ISD::SMUL_LOHI) V->getOpcode() == ISD::SMUL_LOHI)
return V; return V;
return SDValue(); return SDValue();
} }
static SDValue AddCombineTo64BitSMLAL16(SDNode *AddcNode, SDNode *AddeNode, static SDValue AddCombineTo64BitSMLAL16(SDNode *AddcNode, SDNode *AddeNode,
TargetLowering::DAGCombinerInfo &DCI, TargetLowering::DAGCombinerInfo &DCI,
const ARMSubtarget *Subtarget) { const ARMSubtarget *Subtarget) {
if (Subtarget->isThumb()) { if (!Subtarget->hasBaseDSP())
if (!Subtarget->hasDSP())
return SDValue();
} else if (!Subtarget->hasV5TEOps())
return SDValue(); return SDValue();
// SMLALBB, SMLALBT, SMLALTB, SMLALTT multiply two 16-bit values and // SMLALBB, SMLALBT, SMLALTB, SMLALTT multiply two 16-bit values and
// accumulates the product into a 64-bit value. The 16-bit values will // accumulates the product into a 64-bit value. The 16-bit values will
// be sign extended somehow or SRA'd into 32-bit values // be sign extended somehow or SRA'd into 32-bit values
// (addc (adde (mul 16bit, 16bit), lo), hi) // (addc (adde (mul 16bit, 16bit), lo), hi)
SDValue Mul = AddcNode->getOperand(0); SDValue Mul = AddcNode->getOperand(0);
SDValue Lo = AddcNode->getOperand(1); SDValue Lo = AddcNode->getOperand(1);
▲ Show 20 Lines • Show All 5,984 Lines • Show Last 20 Lines

llvm/lib/Target/ARM/ARMInstrInfo.td

Show First 20 Lines • Show All 3,749 Lines • ▼ Show 20 Linesdef QDSUB : AAIRevOpr<0b00010110, 0b00000101, "qdsub",
(int_arm_qadd GPRnopc:$Rn, GPRnopc:$Rn)))]>; (int_arm_qadd GPRnopc:$Rn, GPRnopc:$Rn)))]>;
def QSUB : AAIRevOpr<0b00010010, 0b00000101, "qsub", def QSUB : AAIRevOpr<0b00010010, 0b00000101, "qsub",
[(set GPRnopc:$Rd, (int_arm_qsub GPRnopc:$Rm, GPRnopc:$Rn))]>; [(set GPRnopc:$Rd, (int_arm_qsub GPRnopc:$Rm, GPRnopc:$Rn))]>;
let DecoderMethod = "DecodeQADDInstruction" in let DecoderMethod = "DecodeQADDInstruction" in
def QADD : AAIRevOpr<0b00010000, 0b00000101, "qadd", def QADD : AAIRevOpr<0b00010000, 0b00000101, "qadd",
[(set GPRnopc:$Rd, (int_arm_qadd GPRnopc:$Rm, GPRnopc:$Rn))]>; [(set GPRnopc:$Rd, (int_arm_qadd GPRnopc:$Rm, GPRnopc:$Rn))]>;
} }
def : ARMV5TEPat<(saddsat GPR:$a, GPR:$b),
samparkerUnsubmitted
Done
ReplyInline Actions

I think these should be V5TE patterns.

samparker: I think these should be V5TE patterns.
dmgreenAuthorUnsubmitted
Done
ReplyInline Actions

I've added a hasBaseDSP method to do the check in ISel. Let me know if there's a better name for it.

dmgreen: I've added a hasBaseDSP method to do the check in ISel. Let me know if there's a better name…
(QADD GPR:$a, GPR:$b)>;
def : ARMV5TEPat<(ssubsat GPR:$a, GPR:$b),
(QSUB GPR:$a, GPR:$b)>;
def : ARMV6Pat<(ARMqadd8b rGPR:$Rm, rGPR:$Rn), def : ARMV6Pat<(ARMqadd8b rGPR:$Rm, rGPR:$Rn),
(QADD8 rGPR:$Rm, rGPR:$Rn)>; (QADD8 rGPR:$Rm, rGPR:$Rn)>;
def : ARMV6Pat<(ARMqsub8b rGPR:$Rm, rGPR:$Rn), def : ARMV6Pat<(ARMqsub8b rGPR:$Rm, rGPR:$Rn),
(QSUB8 rGPR:$Rm, rGPR:$Rn)>; (QSUB8 rGPR:$Rm, rGPR:$Rn)>;
def : ARMV6Pat<(ARMqadd16b rGPR:$Rm, rGPR:$Rn), def : ARMV6Pat<(ARMqadd16b rGPR:$Rm, rGPR:$Rn),
(QADD16 rGPR:$Rm, rGPR:$Rn)>; (QADD16 rGPR:$Rm, rGPR:$Rn)>;
def : ARMV6Pat<(ARMqsub16b rGPR:$Rm, rGPR:$Rn), def : ARMV6Pat<(ARMqsub16b rGPR:$Rm, rGPR:$Rn),
(QSUB16 rGPR:$Rm, rGPR:$Rn)>; (QSUB16 rGPR:$Rm, rGPR:$Rn)>;
▲ Show 20 Lines • Show All 2,449 Lines • Show Last 20 Lines

llvm/lib/Target/ARM/ARMInstrThumb2.td

Show First 20 Lines • Show All 2,389 Lines • ▼ Show 20 Linesdef : Thumb2DSPPat<(int_arm_qadd rGPR:$Rm, rGPR:$Rn),
(t2QADD rGPR:$Rm, rGPR:$Rn)>; (t2QADD rGPR:$Rm, rGPR:$Rn)>;
def : Thumb2DSPPat<(int_arm_qsub rGPR:$Rm, rGPR:$Rn), def : Thumb2DSPPat<(int_arm_qsub rGPR:$Rm, rGPR:$Rn),
(t2QSUB rGPR:$Rm, rGPR:$Rn)>; (t2QSUB rGPR:$Rm, rGPR:$Rn)>;
def : Thumb2DSPPat<(int_arm_qadd(int_arm_qadd rGPR:$Rm, rGPR:$Rm), rGPR:$Rn), def : Thumb2DSPPat<(int_arm_qadd(int_arm_qadd rGPR:$Rm, rGPR:$Rm), rGPR:$Rn),
(t2QDADD rGPR:$Rm, rGPR:$Rn)>; (t2QDADD rGPR:$Rm, rGPR:$Rn)>;
def : Thumb2DSPPat<(int_arm_qsub rGPR:$Rm, (int_arm_qadd rGPR:$Rn, rGPR:$Rn)), def : Thumb2DSPPat<(int_arm_qsub rGPR:$Rm, (int_arm_qadd rGPR:$Rn, rGPR:$Rn)),
(t2QDSUB rGPR:$Rm, rGPR:$Rn)>; (t2QDSUB rGPR:$Rm, rGPR:$Rn)>;
def : Thumb2DSPPat<(saddsat rGPR:$Rm, rGPR:$Rn),
(t2QADD rGPR:$Rm, rGPR:$Rn)>;
def : Thumb2DSPPat<(ssubsat rGPR:$Rm, rGPR:$Rn),
(t2QSUB rGPR:$Rm, rGPR:$Rn)>;
def : Thumb2DSPPat<(ARMqadd8b rGPR:$Rm, rGPR:$Rn), def : Thumb2DSPPat<(ARMqadd8b rGPR:$Rm, rGPR:$Rn),
(t2QADD8 rGPR:$Rm, rGPR:$Rn)>; (t2QADD8 rGPR:$Rm, rGPR:$Rn)>;
def : Thumb2DSPPat<(ARMqsub8b rGPR:$Rm, rGPR:$Rn), def : Thumb2DSPPat<(ARMqsub8b rGPR:$Rm, rGPR:$Rn),
(t2QSUB8 rGPR:$Rm, rGPR:$Rn)>; (t2QSUB8 rGPR:$Rm, rGPR:$Rn)>;
def : Thumb2DSPPat<(ARMqadd16b rGPR:$Rm, rGPR:$Rn), def : Thumb2DSPPat<(ARMqadd16b rGPR:$Rm, rGPR:$Rn),
(t2QADD16 rGPR:$Rm, rGPR:$Rn)>; (t2QADD16 rGPR:$Rm, rGPR:$Rn)>;
def : Thumb2DSPPat<(ARMqsub16b rGPR:$Rm, rGPR:$Rn), def : Thumb2DSPPat<(ARMqsub16b rGPR:$Rm, rGPR:$Rn),
(t2QSUB16 rGPR:$Rm, rGPR:$Rn)>; (t2QSUB16 rGPR:$Rm, rGPR:$Rn)>;
▲ Show 20 Lines • Show All 2,902 Lines • Show Last 20 Lines

llvm/lib/Target/ARM/ARMSubtarget.h

Show First 20 Lines • Show All 666 Lines • ▼ Show 20 Linespublic:
bool hasBranchPredictor() const { return HasBranchPredictor; } bool hasBranchPredictor() const { return HasBranchPredictor; }
bool hasMPExtension() const { return HasMPExtension; } bool hasMPExtension() const { return HasMPExtension; }
bool hasDSP() const { return HasDSP; } bool hasDSP() const { return HasDSP; }
bool useNaClTrap() const { return UseNaClTrap; } bool useNaClTrap() const { return UseNaClTrap; }
bool useSjLjEH() const { return UseSjLjEH; } bool useSjLjEH() const { return UseSjLjEH; }
bool hasSB() const { return HasSB; } bool hasSB() const { return HasSB; }
bool genLongCalls() const { return GenLongCalls; } bool genLongCalls() const { return GenLongCalls; }
bool genExecuteOnly() const { return GenExecuteOnly; } bool genExecuteOnly() const { return GenExecuteOnly; }
bool hasBaseDSP() const {
if (isThumb())
return hasDSP();
else
return hasV5TEOps();
}
bool hasFP16() const { return HasFP16; } bool hasFP16() const { return HasFP16; }
bool hasD32() const { return HasD32; } bool hasD32() const { return HasD32; }
bool hasFullFP16() const { return HasFullFP16; } bool hasFullFP16() const { return HasFullFP16; }
bool hasFP16FML() const { return HasFP16FML; } bool hasFP16FML() const { return HasFP16FML; }
bool hasFuseAES() const { return HasFuseAES; } bool hasFuseAES() const { return HasFuseAES; }
bool hasFuseLiterals() const { return HasFuseLiterals; } bool hasFuseLiterals() const { return HasFuseLiterals; }
▲ Show 20 Lines • Show All 189 Lines • Show Last 20 Lines

llvm/test/CodeGen/ARM/qdadd.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=thumbv7m-none-eabi | FileCheck %s --check-prefix=CHECK-T2 --check-prefix=CHECK-T2NODSP ; RUN: llc < %s -mtriple=thumbv7m-none-eabi | FileCheck %s --check-prefix=CHECK-T2 --check-prefix=CHECK-T2NODSP
; RUN: llc < %s -mtriple=thumbv7em-none-eabi | FileCheck %s --check-prefix=CHECK-T2 --check-prefix=CHECK-T2DSP ; RUN: llc < %s -mtriple=thumbv7em-none-eabi | FileCheck %s --check-prefix=CHECK-T2 --check-prefix=CHECK-T2DSP
; RUN: llc < %s -mtriple=armv5te-none-none-eabi | FileCheck %s --check-prefix=CHECK-ARM --check-prefix=CHECK-ARM6 ; RUN: llc < %s -mtriple=armv5te-none-none-eabi | FileCheck %s --check-prefix=CHECK-ARM --check-prefix=CHECK-ARM6
; RUN: llc < %s -mtriple=armv8a-none-eabi | FileCheck %s --check-prefix=CHECK-ARM --check-prefix=CHECK-ARM8 ; RUN: llc < %s -mtriple=armv8a-none-eabi | FileCheck %s --check-prefix=CHECK-ARM --check-prefix=CHECK-ARM8
define i32 @qdadd(i32 %x, i32 %y) nounwind { define i32 @qdadd(i32 %x, i32 %y) nounwind {
; CHECK-T2-LABEL: qdadd: ; CHECK-T2NODSP-LABEL: qdadd:
; CHECK-T2: @ %bb.0: ; CHECK-T2NODSP: @ %bb.0:
; CHECK-T2-NEXT: .save {r7, lr} ; CHECK-T2NODSP-NEXT: .save {r7, lr}
; CHECK-T2-NEXT: push {r7, lr} ; CHECK-T2NODSP-NEXT: push {r7, lr}
; CHECK-T2-NEXT: movs r3, #0 ; CHECK-T2NODSP-NEXT: movs r3, #0
; CHECK-T2-NEXT: adds.w r12, r0, r0 ; CHECK-T2NODSP-NEXT: adds.w r12, r0, r0
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi r3, #1 ; CHECK-T2NODSP-NEXT: movmi r3, #1
; CHECK-T2-NEXT: cmp r3, #0 ; CHECK-T2NODSP-NEXT: cmp r3, #0
; CHECK-T2-NEXT: mov.w r3, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w r3, #-2147483648
; CHECK-T2-NEXT: mov.w lr, #0 ; CHECK-T2NODSP-NEXT: mov.w lr, #0
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne r3, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne r3, #-2147483648
; CHECK-T2-NEXT: cmp r12, r0 ; CHECK-T2NODSP-NEXT: cmp r12, r0
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc r3, r12 ; CHECK-T2NODSP-NEXT: movvc r3, r12
; CHECK-T2-NEXT: adds r0, r3, r1 ; CHECK-T2NODSP-NEXT: adds r0, r3, r1
; CHECK-T2-NEXT: mov.w r2, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w r2, #-2147483648
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi.w lr, #1 ; CHECK-T2NODSP-NEXT: movmi.w lr, #1
; CHECK-T2-NEXT: cmp.w lr, #0 ; CHECK-T2NODSP-NEXT: cmp.w lr, #0
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne r2, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne r2, #-2147483648
; CHECK-T2-NEXT: cmp r0, r3 ; CHECK-T2NODSP-NEXT: cmp r0, r3
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc r2, r0 ; CHECK-T2NODSP-NEXT: movvc r2, r0
; CHECK-T2-NEXT: mov r0, r2 ; CHECK-T2NODSP-NEXT: mov r0, r2
; CHECK-T2-NEXT: pop {r7, pc} ; CHECK-T2NODSP-NEXT: pop {r7, pc}
; ;
; CHECK-ARM6-LABEL: qdadd: ; CHECK-T2DSP-LABEL: qdadd:
; CHECK-ARM6: @ %bb.0: ; CHECK-T2DSP: @ %bb.0:
; CHECK-ARM6-NEXT: .save {r11, lr} ; CHECK-T2DSP-NEXT: qadd r0, r0, r0
; CHECK-ARM6-NEXT: push {r11, lr} ; CHECK-T2DSP-NEXT: qadd r0, r0, r1
; CHECK-ARM6-NEXT: adds r12, r0, r0 ; CHECK-T2DSP-NEXT: bx lr
; CHECK-ARM6-NEXT: mov r3, #0 ;
; CHECK-ARM6-NEXT: movmi r3, #1 ; CHECK-ARM-LABEL: qdadd:
; CHECK-ARM6-NEXT: cmp r3, #0 ; CHECK-ARM: @ %bb.0:
; CHECK-ARM6-NEXT: mov r3, #-2147483648 ; CHECK-ARM-NEXT: qadd r0, r0, r0
samparkerUnsubmitted
Not Done
ReplyInline Actions

Still need the tablegen patterns!

samparker: Still need the tablegen patterns!
dmgreenAuthorUnsubmitted
Done
ReplyInline Actions

D68999. I thought I had put that somewhere here, but perhaps I had just presumed you would get the notification.

I rebased these tests, which is why they appear here too. That patch adds the patterns.

dmgreen: D68999. I thought I had put that somewhere here, but perhaps I had just presumed you would get…
samparkerUnsubmitted
Not Done
ReplyInline Actions

Ah!

samparker: Ah!
; CHECK-ARM6-NEXT: mov lr, #0 ; CHECK-ARM-NEXT: qadd r0, r0, r1
; CHECK-ARM6-NEXT: mvnne r3, #-2147483648 ; CHECK-ARM-NEXT: bx lr
; CHECK-ARM6-NEXT: cmp r12, r0
; CHECK-ARM6-NEXT: movvc r3, r12
; CHECK-ARM6-NEXT: adds r0, r3, r1
; CHECK-ARM6-NEXT: movmi lr, #1
; CHECK-ARM6-NEXT: mov r2, #-2147483648
; CHECK-ARM6-NEXT: cmp lr, #0
; CHECK-ARM6-NEXT: mvnne r2, #-2147483648
; CHECK-ARM6-NEXT: cmp r0, r3
; CHECK-ARM6-NEXT: movvc r2, r0
; CHECK-ARM6-NEXT: mov r0, r2
; CHECK-ARM6-NEXT: pop {r11, pc}
;
; CHECK-ARM8-LABEL: qdadd:
; CHECK-ARM8: @ %bb.0:
; CHECK-ARM8-NEXT: .save {r11, lr}
; CHECK-ARM8-NEXT: push {r11, lr}
; CHECK-ARM8-NEXT: adds r12, r0, r0
; CHECK-ARM8-NEXT: mov r3, #0
; CHECK-ARM8-NEXT: movwmi r3, #1
; CHECK-ARM8-NEXT: cmp r3, #0
; CHECK-ARM8-NEXT: mov r3, #-2147483648
; CHECK-ARM8-NEXT: mov lr, #0
; CHECK-ARM8-NEXT: mvnne r3, #-2147483648
; CHECK-ARM8-NEXT: cmp r12, r0
; CHECK-ARM8-NEXT: movvc r3, r12
; CHECK-ARM8-NEXT: adds r0, r3, r1
; CHECK-ARM8-NEXT: movwmi lr, #1
; CHECK-ARM8-NEXT: mov r2, #-2147483648
; CHECK-ARM8-NEXT: cmp lr, #0
; CHECK-ARM8-NEXT: mvnne r2, #-2147483648
; CHECK-ARM8-NEXT: cmp r0, r3
; CHECK-ARM8-NEXT: movvc r2, r0
; CHECK-ARM8-NEXT: mov r0, r2
; CHECK-ARM8-NEXT: pop {r11, pc}
%z = call i32 @llvm.sadd.sat.i32(i32 %x, i32 %x) %z = call i32 @llvm.sadd.sat.i32(i32 %x, i32 %x)
%tmp = call i32 @llvm.sadd.sat.i32(i32 %z, i32 %y) %tmp = call i32 @llvm.sadd.sat.i32(i32 %z, i32 %y)
ret i32 %tmp ret i32 %tmp
} }
define i32 @qdadd_c(i32 %x, i32 %y) nounwind { define i32 @qdadd_c(i32 %x, i32 %y) nounwind {
; CHECK-T2-LABEL: qdadd_c: ; CHECK-T2NODSP-LABEL: qdadd_c:
; CHECK-T2: @ %bb.0: ; CHECK-T2NODSP: @ %bb.0:
; CHECK-T2-NEXT: .save {r7, lr} ; CHECK-T2NODSP-NEXT: .save {r7, lr}
; CHECK-T2-NEXT: push {r7, lr} ; CHECK-T2NODSP-NEXT: push {r7, lr}
; CHECK-T2-NEXT: movs r3, #0 ; CHECK-T2NODSP-NEXT: movs r3, #0
; CHECK-T2-NEXT: adds.w r12, r0, r0 ; CHECK-T2NODSP-NEXT: adds.w r12, r0, r0
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi r3, #1 ; CHECK-T2NODSP-NEXT: movmi r3, #1
; CHECK-T2-NEXT: cmp r3, #0 ; CHECK-T2NODSP-NEXT: cmp r3, #0
; CHECK-T2-NEXT: mov.w r3, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w r3, #-2147483648
; CHECK-T2-NEXT: mov.w lr, #0 ; CHECK-T2NODSP-NEXT: mov.w lr, #0
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne r3, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne r3, #-2147483648
; CHECK-T2-NEXT: cmp r12, r0 ; CHECK-T2NODSP-NEXT: cmp r12, r0
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc r3, r12 ; CHECK-T2NODSP-NEXT: movvc r3, r12
; CHECK-T2-NEXT: adds r0, r1, r3 ; CHECK-T2NODSP-NEXT: adds r0, r1, r3
; CHECK-T2-NEXT: mov.w r2, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w r2, #-2147483648
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi.w lr, #1 ; CHECK-T2NODSP-NEXT: movmi.w lr, #1
; CHECK-T2-NEXT: cmp.w lr, #0 ; CHECK-T2NODSP-NEXT: cmp.w lr, #0
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne r2, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne r2, #-2147483648
; CHECK-T2-NEXT: cmp r0, r1 ; CHECK-T2NODSP-NEXT: cmp r0, r1
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc r2, r0 ; CHECK-T2NODSP-NEXT: movvc r2, r0
; CHECK-T2-NEXT: mov r0, r2 ; CHECK-T2NODSP-NEXT: mov r0, r2
; CHECK-T2-NEXT: pop {r7, pc} ; CHECK-T2NODSP-NEXT: pop {r7, pc}
; ;
; CHECK-ARM6-LABEL: qdadd_c: ; CHECK-T2DSP-LABEL: qdadd_c:
; CHECK-ARM6: @ %bb.0: ; CHECK-T2DSP: @ %bb.0:
; CHECK-ARM6-NEXT: .save {r11, lr} ; CHECK-T2DSP-NEXT: qadd r0, r0, r0
; CHECK-ARM6-NEXT: push {r11, lr} ; CHECK-T2DSP-NEXT: qadd r0, r1, r0
; CHECK-ARM6-NEXT: adds r12, r0, r0 ; CHECK-T2DSP-NEXT: bx lr
; CHECK-ARM6-NEXT: mov r3, #0 ;
; CHECK-ARM6-NEXT: movmi r3, #1 ; CHECK-ARM-LABEL: qdadd_c:
; CHECK-ARM6-NEXT: cmp r3, #0 ; CHECK-ARM: @ %bb.0:
; CHECK-ARM6-NEXT: mov r3, #-2147483648 ; CHECK-ARM-NEXT: qadd r0, r0, r0
; CHECK-ARM6-NEXT: mov lr, #0 ; CHECK-ARM-NEXT: qadd r0, r1, r0
; CHECK-ARM6-NEXT: mvnne r3, #-2147483648 ; CHECK-ARM-NEXT: bx lr
; CHECK-ARM6-NEXT: cmp r12, r0
; CHECK-ARM6-NEXT: movvc r3, r12
; CHECK-ARM6-NEXT: adds r0, r1, r3
; CHECK-ARM6-NEXT: movmi lr, #1
; CHECK-ARM6-NEXT: mov r2, #-2147483648
; CHECK-ARM6-NEXT: cmp lr, #0
; CHECK-ARM6-NEXT: mvnne r2, #-2147483648
; CHECK-ARM6-NEXT: cmp r0, r1
; CHECK-ARM6-NEXT: movvc r2, r0
; CHECK-ARM6-NEXT: mov r0, r2
; CHECK-ARM6-NEXT: pop {r11, pc}
;
; CHECK-ARM8-LABEL: qdadd_c:
; CHECK-ARM8: @ %bb.0:
; CHECK-ARM8-NEXT: .save {r11, lr}
; CHECK-ARM8-NEXT: push {r11, lr}
; CHECK-ARM8-NEXT: adds r12, r0, r0
; CHECK-ARM8-NEXT: mov r3, #0
; CHECK-ARM8-NEXT: movwmi r3, #1
; CHECK-ARM8-NEXT: cmp r3, #0
; CHECK-ARM8-NEXT: mov r3, #-2147483648
; CHECK-ARM8-NEXT: mov lr, #0
; CHECK-ARM8-NEXT: mvnne r3, #-2147483648
; CHECK-ARM8-NEXT: cmp r12, r0
; CHECK-ARM8-NEXT: movvc r3, r12
; CHECK-ARM8-NEXT: adds r0, r1, r3
; CHECK-ARM8-NEXT: movwmi lr, #1
; CHECK-ARM8-NEXT: mov r2, #-2147483648
; CHECK-ARM8-NEXT: cmp lr, #0
; CHECK-ARM8-NEXT: mvnne r2, #-2147483648
; CHECK-ARM8-NEXT: cmp r0, r1
; CHECK-ARM8-NEXT: movvc r2, r0
; CHECK-ARM8-NEXT: mov r0, r2
; CHECK-ARM8-NEXT: pop {r11, pc}
%z = call i32 @llvm.sadd.sat.i32(i32 %x, i32 %x) %z = call i32 @llvm.sadd.sat.i32(i32 %x, i32 %x)
%tmp = call i32 @llvm.sadd.sat.i32(i32 %y, i32 %z) %tmp = call i32 @llvm.sadd.sat.i32(i32 %y, i32 %z)
ret i32 %tmp ret i32 %tmp
} }
define i32 @qdsub(i32 %x, i32 %y) nounwind { define i32 @qdsub(i32 %x, i32 %y) nounwind {
; CHECK-T2-LABEL: qdsub: ; CHECK-T2NODSP-LABEL: qdsub:
; CHECK-T2: @ %bb.0: ; CHECK-T2NODSP: @ %bb.0:
; CHECK-T2-NEXT: .save {r7, lr} ; CHECK-T2NODSP-NEXT: .save {r7, lr}
; CHECK-T2-NEXT: push {r7, lr} ; CHECK-T2NODSP-NEXT: push {r7, lr}
; CHECK-T2-NEXT: movs r3, #0 ; CHECK-T2NODSP-NEXT: movs r3, #0
; CHECK-T2-NEXT: adds.w r12, r0, r0 ; CHECK-T2NODSP-NEXT: adds.w r12, r0, r0
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi r3, #1 ; CHECK-T2NODSP-NEXT: movmi r3, #1
; CHECK-T2-NEXT: cmp r3, #0 ; CHECK-T2NODSP-NEXT: cmp r3, #0
; CHECK-T2-NEXT: mov.w r3, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w r3, #-2147483648
; CHECK-T2-NEXT: mov.w lr, #0 ; CHECK-T2NODSP-NEXT: mov.w lr, #0
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne r3, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne r3, #-2147483648
; CHECK-T2-NEXT: cmp r12, r0 ; CHECK-T2NODSP-NEXT: cmp r12, r0
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc r3, r12 ; CHECK-T2NODSP-NEXT: movvc r3, r12
; CHECK-T2-NEXT: subs r0, r1, r3 ; CHECK-T2NODSP-NEXT: subs r0, r1, r3
; CHECK-T2-NEXT: mov.w r2, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w r2, #-2147483648
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi.w lr, #1 ; CHECK-T2NODSP-NEXT: movmi.w lr, #1
; CHECK-T2-NEXT: cmp.w lr, #0 ; CHECK-T2NODSP-NEXT: cmp.w lr, #0
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne r2, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne r2, #-2147483648
; CHECK-T2-NEXT: cmp r1, r3 ; CHECK-T2NODSP-NEXT: cmp r1, r3
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc r2, r0 ; CHECK-T2NODSP-NEXT: movvc r2, r0
; CHECK-T2-NEXT: mov r0, r2 ; CHECK-T2NODSP-NEXT: mov r0, r2
; CHECK-T2-NEXT: pop {r7, pc} ; CHECK-T2NODSP-NEXT: pop {r7, pc}
; ;
; CHECK-ARM6-LABEL: qdsub: ; CHECK-T2DSP-LABEL: qdsub:
; CHECK-ARM6: @ %bb.0: ; CHECK-T2DSP: @ %bb.0:
; CHECK-ARM6-NEXT: .save {r11, lr} ; CHECK-T2DSP-NEXT: qadd r0, r0, r0
; CHECK-ARM6-NEXT: push {r11, lr} ; CHECK-T2DSP-NEXT: qsub r0, r1, r0
; CHECK-ARM6-NEXT: adds r12, r0, r0 ; CHECK-T2DSP-NEXT: bx lr
; CHECK-ARM6-NEXT: mov r3, #0 ;
; CHECK-ARM6-NEXT: movmi r3, #1 ; CHECK-ARM-LABEL: qdsub:
; CHECK-ARM6-NEXT: cmp r3, #0 ; CHECK-ARM: @ %bb.0:
; CHECK-ARM6-NEXT: mov r3, #-2147483648 ; CHECK-ARM-NEXT: qadd r0, r0, r0
; CHECK-ARM6-NEXT: mov lr, #0 ; CHECK-ARM-NEXT: qsub r0, r1, r0
; CHECK-ARM6-NEXT: mvnne r3, #-2147483648 ; CHECK-ARM-NEXT: bx lr
; CHECK-ARM6-NEXT: cmp r12, r0
; CHECK-ARM6-NEXT: movvc r3, r12
; CHECK-ARM6-NEXT: subs r0, r1, r3
; CHECK-ARM6-NEXT: movmi lr, #1
; CHECK-ARM6-NEXT: mov r2, #-2147483648
; CHECK-ARM6-NEXT: cmp lr, #0
; CHECK-ARM6-NEXT: mvnne r2, #-2147483648
; CHECK-ARM6-NEXT: cmp r1, r3
; CHECK-ARM6-NEXT: movvc r2, r0
; CHECK-ARM6-NEXT: mov r0, r2
; CHECK-ARM6-NEXT: pop {r11, pc}
;
; CHECK-ARM8-LABEL: qdsub:
; CHECK-ARM8: @ %bb.0:
; CHECK-ARM8-NEXT: .save {r11, lr}
; CHECK-ARM8-NEXT: push {r11, lr}
; CHECK-ARM8-NEXT: adds r12, r0, r0
; CHECK-ARM8-NEXT: mov r3, #0
; CHECK-ARM8-NEXT: movwmi r3, #1
; CHECK-ARM8-NEXT: cmp r3, #0
; CHECK-ARM8-NEXT: mov r3, #-2147483648
; CHECK-ARM8-NEXT: mov lr, #0
; CHECK-ARM8-NEXT: mvnne r3, #-2147483648
; CHECK-ARM8-NEXT: cmp r12, r0
; CHECK-ARM8-NEXT: movvc r3, r12
; CHECK-ARM8-NEXT: subs r0, r1, r3
; CHECK-ARM8-NEXT: movwmi lr, #1
; CHECK-ARM8-NEXT: mov r2, #-2147483648
; CHECK-ARM8-NEXT: cmp lr, #0
; CHECK-ARM8-NEXT: mvnne r2, #-2147483648
; CHECK-ARM8-NEXT: cmp r1, r3
; CHECK-ARM8-NEXT: movvc r2, r0
; CHECK-ARM8-NEXT: mov r0, r2
; CHECK-ARM8-NEXT: pop {r11, pc}
%z = call i32 @llvm.sadd.sat.i32(i32 %x, i32 %x) %z = call i32 @llvm.sadd.sat.i32(i32 %x, i32 %x)
%tmp = call i32 @llvm.ssub.sat.i32(i32 %y, i32 %z) %tmp = call i32 @llvm.ssub.sat.i32(i32 %y, i32 %z)
ret i32 %tmp ret i32 %tmp
} }
define i32 @qdsub_c(i32 %x, i32 %y) nounwind { define i32 @qdsub_c(i32 %x, i32 %y) nounwind {
; CHECK-T2-LABEL: qdsub_c: ; CHECK-T2NODSP-LABEL: qdsub_c:
; CHECK-T2: @ %bb.0: ; CHECK-T2NODSP: @ %bb.0:
; CHECK-T2-NEXT: .save {r7, lr} ; CHECK-T2NODSP-NEXT: .save {r7, lr}
; CHECK-T2-NEXT: push {r7, lr} ; CHECK-T2NODSP-NEXT: push {r7, lr}
; CHECK-T2-NEXT: movs r3, #0 ; CHECK-T2NODSP-NEXT: movs r3, #0
; CHECK-T2-NEXT: adds.w r12, r0, r0 ; CHECK-T2NODSP-NEXT: adds.w r12, r0, r0
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi r3, #1 ; CHECK-T2NODSP-NEXT: movmi r3, #1
; CHECK-T2-NEXT: cmp r3, #0 ; CHECK-T2NODSP-NEXT: cmp r3, #0
; CHECK-T2-NEXT: mov.w r3, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w r3, #-2147483648
; CHECK-T2-NEXT: mov.w lr, #0 ; CHECK-T2NODSP-NEXT: mov.w lr, #0
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne r3, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne r3, #-2147483648
; CHECK-T2-NEXT: cmp r12, r0 ; CHECK-T2NODSP-NEXT: cmp r12, r0
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc r3, r12 ; CHECK-T2NODSP-NEXT: movvc r3, r12
; CHECK-T2-NEXT: subs r0, r3, r1 ; CHECK-T2NODSP-NEXT: subs r0, r3, r1
; CHECK-T2-NEXT: mov.w r2, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w r2, #-2147483648
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi.w lr, #1 ; CHECK-T2NODSP-NEXT: movmi.w lr, #1
; CHECK-T2-NEXT: cmp.w lr, #0 ; CHECK-T2NODSP-NEXT: cmp.w lr, #0
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne r2, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne r2, #-2147483648
; CHECK-T2-NEXT: cmp r3, r1 ; CHECK-T2NODSP-NEXT: cmp r3, r1
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc r2, r0 ; CHECK-T2NODSP-NEXT: movvc r2, r0
; CHECK-T2-NEXT: mov r0, r2 ; CHECK-T2NODSP-NEXT: mov r0, r2
; CHECK-T2-NEXT: pop {r7, pc} ; CHECK-T2NODSP-NEXT: pop {r7, pc}
; ;
; CHECK-ARM6-LABEL: qdsub_c: ; CHECK-T2DSP-LABEL: qdsub_c:
; CHECK-ARM6: @ %bb.0: ; CHECK-T2DSP: @ %bb.0:
; CHECK-ARM6-NEXT: .save {r11, lr} ; CHECK-T2DSP-NEXT: qadd r0, r0, r0
; CHECK-ARM6-NEXT: push {r11, lr} ; CHECK-T2DSP-NEXT: qsub r0, r0, r1
; CHECK-ARM6-NEXT: adds r12, r0, r0 ; CHECK-T2DSP-NEXT: bx lr
; CHECK-ARM6-NEXT: mov r3, #0 ;
; CHECK-ARM6-NEXT: movmi r3, #1 ; CHECK-ARM-LABEL: qdsub_c:
; CHECK-ARM6-NEXT: cmp r3, #0 ; CHECK-ARM: @ %bb.0:
; CHECK-ARM6-NEXT: mov r3, #-2147483648 ; CHECK-ARM-NEXT: qadd r0, r0, r0
; CHECK-ARM6-NEXT: mov lr, #0 ; CHECK-ARM-NEXT: qsub r0, r0, r1
; CHECK-ARM6-NEXT: mvnne r3, #-2147483648 ; CHECK-ARM-NEXT: bx lr
; CHECK-ARM6-NEXT: cmp r12, r0
; CHECK-ARM6-NEXT: movvc r3, r12
; CHECK-ARM6-NEXT: subs r0, r3, r1
; CHECK-ARM6-NEXT: movmi lr, #1
; CHECK-ARM6-NEXT: mov r2, #-2147483648
; CHECK-ARM6-NEXT: cmp lr, #0
; CHECK-ARM6-NEXT: mvnne r2, #-2147483648
; CHECK-ARM6-NEXT: cmp r3, r1
; CHECK-ARM6-NEXT: movvc r2, r0
; CHECK-ARM6-NEXT: mov r0, r2
; CHECK-ARM6-NEXT: pop {r11, pc}
;
; CHECK-ARM8-LABEL: qdsub_c:
; CHECK-ARM8: @ %bb.0:
; CHECK-ARM8-NEXT: .save {r11, lr}
; CHECK-ARM8-NEXT: push {r11, lr}
; CHECK-ARM8-NEXT: adds r12, r0, r0
; CHECK-ARM8-NEXT: mov r3, #0
; CHECK-ARM8-NEXT: movwmi r3, #1
; CHECK-ARM8-NEXT: cmp r3, #0
; CHECK-ARM8-NEXT: mov r3, #-2147483648
; CHECK-ARM8-NEXT: mov lr, #0
; CHECK-ARM8-NEXT: mvnne r3, #-2147483648
; CHECK-ARM8-NEXT: cmp r12, r0
; CHECK-ARM8-NEXT: movvc r3, r12
; CHECK-ARM8-NEXT: subs r0, r3, r1
; CHECK-ARM8-NEXT: movwmi lr, #1
; CHECK-ARM8-NEXT: mov r2, #-2147483648
; CHECK-ARM8-NEXT: cmp lr, #0
; CHECK-ARM8-NEXT: mvnne r2, #-2147483648
; CHECK-ARM8-NEXT: cmp r3, r1
; CHECK-ARM8-NEXT: movvc r2, r0
; CHECK-ARM8-NEXT: mov r0, r2
; CHECK-ARM8-NEXT: pop {r11, pc}
%z = call i32 @llvm.sadd.sat.i32(i32 %x, i32 %x) %z = call i32 @llvm.sadd.sat.i32(i32 %x, i32 %x)
%tmp = call i32 @llvm.ssub.sat.i32(i32 %z, i32 %y) %tmp = call i32 @llvm.ssub.sat.i32(i32 %z, i32 %y)
ret i32 %tmp ret i32 %tmp
} }
declare i32 @llvm.sadd.sat.i32(i32, i32) declare i32 @llvm.sadd.sat.i32(i32, i32)
declare i32 @llvm.ssub.sat.i32(i32, i32) declare i32 @llvm.ssub.sat.i32(i32, i32)

llvm/test/CodeGen/ARM/sadd_sat.ll

Show All 37 Lines
; CHECK-T1-NEXT: mov r0, r1 ; CHECK-T1-NEXT: mov r0, r1
; CHECK-T1-NEXT: .LBB0_6: ; CHECK-T1-NEXT: .LBB0_6:
; CHECK-T1-NEXT: bx lr ; CHECK-T1-NEXT: bx lr
; CHECK-T1-NEXT: .p2align 2 ; CHECK-T1-NEXT: .p2align 2
; CHECK-T1-NEXT: @ %bb.7: ; CHECK-T1-NEXT: @ %bb.7:
; CHECK-T1-NEXT: .LCPI0_0: ; CHECK-T1-NEXT: .LCPI0_0:
; CHECK-T1-NEXT: .long 2147483647 @ 0x7fffffff ; CHECK-T1-NEXT: .long 2147483647 @ 0x7fffffff
; ;
; CHECK-T2-LABEL: func: ; CHECK-T2NODSP-LABEL: func:
; CHECK-T2: @ %bb.0: ; CHECK-T2NODSP: @ %bb.0:
; CHECK-T2-NEXT: adds r2, r0, r1 ; CHECK-T2NODSP-NEXT: adds r2, r0, r1
; CHECK-T2-NEXT: mov.w r3, #0 ; CHECK-T2NODSP-NEXT: mov.w r3, #0
; CHECK-T2-NEXT: mov.w r1, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w r1, #-2147483648
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi r3, #1 ; CHECK-T2NODSP-NEXT: movmi r3, #1
; CHECK-T2-NEXT: cmp r3, #0 ; CHECK-T2NODSP-NEXT: cmp r3, #0
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne r1, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne r1, #-2147483648
; CHECK-T2-NEXT: cmp r2, r0 ; CHECK-T2NODSP-NEXT: cmp r2, r0
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc r1, r2 ; CHECK-T2NODSP-NEXT: movvc r1, r2
; CHECK-T2-NEXT: mov r0, r1 ; CHECK-T2NODSP-NEXT: mov r0, r1
; CHECK-T2-NEXT: bx lr ; CHECK-T2NODSP-NEXT: bx lr
; ;
; CHECK-ARM-LABEL: func: ; CHECK-T2DSP-LABEL: func:
; CHECK-ARM: @ %bb.0: ; CHECK-T2DSP: @ %bb.0:
; CHECK-ARM-NEXT: adds r2, r0, r1 ; CHECK-T2DSP-NEXT: qadd r0, r0, r1
; CHECK-ARM-NEXT: mov r3, #0 ; CHECK-T2DSP-NEXT: bx lr
; CHECK-ARM-NEXT: movmi r3, #1 ;
; CHECK-ARM-NEXT: mov r1, #-2147483648 ; CHECK-ARMNODPS-LABEL: func:
; CHECK-ARM-NEXT: cmp r3, #0 ; CHECK-ARMNODPS: @ %bb.0:
; CHECK-ARM-NEXT: mvnne r1, #-2147483648 ; CHECK-ARMNODPS-NEXT: adds r2, r0, r1
; CHECK-ARM-NEXT: cmp r2, r0 ; CHECK-ARMNODPS-NEXT: mov r3, #0
; CHECK-ARM-NEXT: movvc r1, r2 ; CHECK-ARMNODPS-NEXT: movmi r3, #1
; CHECK-ARM-NEXT: mov r0, r1 ; CHECK-ARMNODPS-NEXT: mov r1, #-2147483648
; CHECK-ARM-NEXT: bx lr ; CHECK-ARMNODPS-NEXT: cmp r3, #0
; CHECK-ARMNODPS-NEXT: mvnne r1, #-2147483648
; CHECK-ARMNODPS-NEXT: cmp r2, r0
; CHECK-ARMNODPS-NEXT: movvc r1, r2
; CHECK-ARMNODPS-NEXT: mov r0, r1
; CHECK-ARMNODPS-NEXT: bx lr
;
; CHECK-ARMBASEDSP-LABEL: func:
; CHECK-ARMBASEDSP: @ %bb.0:
; CHECK-ARMBASEDSP-NEXT: qadd r0, r0, r1
; CHECK-ARMBASEDSP-NEXT: bx lr
;
; CHECK-ARMDSP-LABEL: func:
; CHECK-ARMDSP: @ %bb.0:
; CHECK-ARMDSP-NEXT: qadd r0, r0, r1
; CHECK-ARMDSP-NEXT: bx lr
%tmp = call i32 @llvm.sadd.sat.i32(i32 %x, i32 %y) %tmp = call i32 @llvm.sadd.sat.i32(i32 %x, i32 %y)
ret i32 %tmp ret i32 %tmp
} }
define i64 @func2(i64 %x, i64 %y) nounwind { define i64 @func2(i64 %x, i64 %y) nounwind {
; CHECK-T1-LABEL: func2: ; CHECK-T1-LABEL: func2:
; CHECK-T1: @ %bb.0: ; CHECK-T1: @ %bb.0:
; CHECK-T1-NEXT: .save {r4, r5, r6, r7, lr} ; CHECK-T1-NEXT: .save {r4, r5, r6, r7, lr}
▲ Show 20 Lines • Show All 184 Lines • ▼ Show 20 Lines
; CHECK-ARMNODPS-NEXT: bx lr ; CHECK-ARMNODPS-NEXT: bx lr
; CHECK-ARMNODPS-NEXT: .p2align 2 ; CHECK-ARMNODPS-NEXT: .p2align 2
; CHECK-ARMNODPS-NEXT: @ %bb.1: ; CHECK-ARMNODPS-NEXT: @ %bb.1:
; CHECK-ARMNODPS-NEXT: .LCPI2_0: ; CHECK-ARMNODPS-NEXT: .LCPI2_0:
; CHECK-ARMNODPS-NEXT: .long 4294934528 @ 0xffff8000 ; CHECK-ARMNODPS-NEXT: .long 4294934528 @ 0xffff8000
; ;
; CHECK-ARMBASEDSP-LABEL: func16: ; CHECK-ARMBASEDSP-LABEL: func16:
; CHECK-ARMBASEDSP: @ %bb.0: ; CHECK-ARMBASEDSP: @ %bb.0:
; CHECK-ARMBASEDSP-NEXT: add r0, r0, r1 ; CHECK-ARMBASEDSP-NEXT: lsl r0, r0, #16
; CHECK-ARMBASEDSP-NEXT: mov r1, #255 ; CHECK-ARMBASEDSP-NEXT: lsl r1, r1, #16
; CHECK-ARMBASEDSP-NEXT: orr r1, r1, #32512 ; CHECK-ARMBASEDSP-NEXT: qadd r0, r0, r1
; CHECK-ARMBASEDSP-NEXT: cmp r0, r1 ; CHECK-ARMBASEDSP-NEXT: asr r0, r0, #16
; CHECK-ARMBASEDSP-NEXT: movlt r1, r0
; CHECK-ARMBASEDSP-NEXT: ldr r0, .LCPI2_0
; CHECK-ARMBASEDSP-NEXT: cmn r1, #32768
; CHECK-ARMBASEDSP-NEXT: movgt r0, r1
; CHECK-ARMBASEDSP-NEXT: bx lr ; CHECK-ARMBASEDSP-NEXT: bx lr
; CHECK-ARMBASEDSP-NEXT: .p2align 2
; CHECK-ARMBASEDSP-NEXT: @ %bb.1:
; CHECK-ARMBASEDSP-NEXT: .LCPI2_0:
; CHECK-ARMBASEDSP-NEXT: .long 4294934528 @ 0xffff8000
; ;
; CHECK-ARMDSP-LABEL: func16: ; CHECK-ARMDSP-LABEL: func16:
; CHECK-ARMDSP: @ %bb.0: ; CHECK-ARMDSP: @ %bb.0:
; CHECK-ARMDSP-NEXT: qadd16 r0, r0, r1 ; CHECK-ARMDSP-NEXT: qadd16 r0, r0, r1
; CHECK-ARMDSP-NEXT: sxth r0, r0 ; CHECK-ARMDSP-NEXT: sxth r0, r0
; CHECK-ARMDSP-NEXT: bx lr ; CHECK-ARMDSP-NEXT: bx lr
%tmp = call i16 @llvm.sadd.sat.i16(i16 %x, i16 %y) %tmp = call i16 @llvm.sadd.sat.i16(i16 %x, i16 %y)
ret i16 %tmp ret i16 %tmp
Show All 40 Lines
; CHECK-ARMNODPS-NEXT: cmp r0, #127 ; CHECK-ARMNODPS-NEXT: cmp r0, #127
; CHECK-ARMNODPS-NEXT: movge r0, #127 ; CHECK-ARMNODPS-NEXT: movge r0, #127
; CHECK-ARMNODPS-NEXT: cmn r0, #128 ; CHECK-ARMNODPS-NEXT: cmn r0, #128
; CHECK-ARMNODPS-NEXT: mvnle r0, #127 ; CHECK-ARMNODPS-NEXT: mvnle r0, #127
; CHECK-ARMNODPS-NEXT: bx lr ; CHECK-ARMNODPS-NEXT: bx lr
; ;
; CHECK-ARMBASEDSP-LABEL: func8: ; CHECK-ARMBASEDSP-LABEL: func8:
; CHECK-ARMBASEDSP: @ %bb.0: ; CHECK-ARMBASEDSP: @ %bb.0:
; CHECK-ARMBASEDSP-NEXT: add r0, r0, r1 ; CHECK-ARMBASEDSP-NEXT: lsl r0, r0, #24
; CHECK-ARMBASEDSP-NEXT: cmp r0, #127 ; CHECK-ARMBASEDSP-NEXT: lsl r1, r1, #24
; CHECK-ARMBASEDSP-NEXT: movge r0, #127 ; CHECK-ARMBASEDSP-NEXT: qadd r0, r0, r1
; CHECK-ARMBASEDSP-NEXT: cmn r0, #128 ; CHECK-ARMBASEDSP-NEXT: asr r0, r0, #24
; CHECK-ARMBASEDSP-NEXT: mvnle r0, #127
; CHECK-ARMBASEDSP-NEXT: bx lr ; CHECK-ARMBASEDSP-NEXT: bx lr
; ;
; CHECK-ARMDSP-LABEL: func8: ; CHECK-ARMDSP-LABEL: func8:
; CHECK-ARMDSP: @ %bb.0: ; CHECK-ARMDSP: @ %bb.0:
; CHECK-ARMDSP-NEXT: qadd8 r0, r0, r1 ; CHECK-ARMDSP-NEXT: qadd8 r0, r0, r1
; CHECK-ARMDSP-NEXT: sxtb r0, r0 ; CHECK-ARMDSP-NEXT: sxtb r0, r0
; CHECK-ARMDSP-NEXT: bx lr ; CHECK-ARMDSP-NEXT: bx lr
%tmp = call i8 @llvm.sadd.sat.i8(i8 %x, i8 %y) %tmp = call i8 @llvm.sadd.sat.i8(i8 %x, i8 %y)
Show All 13 Lines
; CHECK-T1-NEXT: mvns r1, r1 ; CHECK-T1-NEXT: mvns r1, r1
; CHECK-T1-NEXT: cmp r0, r1 ; CHECK-T1-NEXT: cmp r0, r1
; CHECK-T1-NEXT: bgt .LBB4_4 ; CHECK-T1-NEXT: bgt .LBB4_4
; CHECK-T1-NEXT: @ %bb.3: ; CHECK-T1-NEXT: @ %bb.3:
; CHECK-T1-NEXT: mov r0, r1 ; CHECK-T1-NEXT: mov r0, r1
; CHECK-T1-NEXT: .LBB4_4: ; CHECK-T1-NEXT: .LBB4_4:
; CHECK-T1-NEXT: bx lr ; CHECK-T1-NEXT: bx lr
; ;
; CHECK-T2-LABEL: func3: ; CHECK-T2NODSP-LABEL: func3:
; CHECK-T2: @ %bb.0: ; CHECK-T2NODSP: @ %bb.0:
; CHECK-T2-NEXT: add r0, r1 ; CHECK-T2NODSP-NEXT: add r0, r1
; CHECK-T2-NEXT: cmp r0, #7 ; CHECK-T2NODSP-NEXT: cmp r0, #7
; CHECK-T2-NEXT: it ge ; CHECK-T2NODSP-NEXT: it ge
; CHECK-T2-NEXT: movge r0, #7 ; CHECK-T2NODSP-NEXT: movge r0, #7
; CHECK-T2-NEXT: cmn.w r0, #8 ; CHECK-T2NODSP-NEXT: cmn.w r0, #8
; CHECK-T2-NEXT: it le ; CHECK-T2NODSP-NEXT: it le
; CHECK-T2-NEXT: mvnle r0, #7 ; CHECK-T2NODSP-NEXT: mvnle r0, #7
; CHECK-T2-NEXT: bx lr ; CHECK-T2NODSP-NEXT: bx lr
; ;
; CHECK-ARM-LABEL: func3: ; CHECK-T2DSP-LABEL: func3:
; CHECK-ARM: @ %bb.0: ; CHECK-T2DSP: @ %bb.0:
; CHECK-ARM-NEXT: add r0, r0, r1 ; CHECK-T2DSP-NEXT: lsls r1, r1, #28
; CHECK-ARM-NEXT: cmp r0, #7 ; CHECK-T2DSP-NEXT: lsls r0, r0, #28
; CHECK-ARM-NEXT: movge r0, #7 ; CHECK-T2DSP-NEXT: qadd r0, r0, r1
; CHECK-ARM-NEXT: cmn r0, #8 ; CHECK-T2DSP-NEXT: asrs r0, r0, #28
; CHECK-ARM-NEXT: mvnle r0, #7 ; CHECK-T2DSP-NEXT: bx lr
; CHECK-ARM-NEXT: bx lr ;
; CHECK-ARMNODPS-LABEL: func3:
; CHECK-ARMNODPS: @ %bb.0:
; CHECK-ARMNODPS-NEXT: add r0, r0, r1
; CHECK-ARMNODPS-NEXT: cmp r0, #7
; CHECK-ARMNODPS-NEXT: movge r0, #7
; CHECK-ARMNODPS-NEXT: cmn r0, #8
; CHECK-ARMNODPS-NEXT: mvnle r0, #7
; CHECK-ARMNODPS-NEXT: bx lr
;
; CHECK-ARMBASEDSP-LABEL: func3:
; CHECK-ARMBASEDSP: @ %bb.0:
; CHECK-ARMBASEDSP-NEXT: lsl r0, r0, #28
; CHECK-ARMBASEDSP-NEXT: lsl r1, r1, #28
; CHECK-ARMBASEDSP-NEXT: qadd r0, r0, r1
; CHECK-ARMBASEDSP-NEXT: asr r0, r0, #28
; CHECK-ARMBASEDSP-NEXT: bx lr
;
; CHECK-ARMDSP-LABEL: func3:
; CHECK-ARMDSP: @ %bb.0:
; CHECK-ARMDSP-NEXT: lsl r0, r0, #28
; CHECK-ARMDSP-NEXT: lsl r1, r1, #28
; CHECK-ARMDSP-NEXT: qadd r0, r0, r1
; CHECK-ARMDSP-NEXT: asr r0, r0, #28
; CHECK-ARMDSP-NEXT: bx lr
%tmp = call i4 @llvm.sadd.sat.i4(i4 %x, i4 %y) %tmp = call i4 @llvm.sadd.sat.i4(i4 %x, i4 %y)
ret i4 %tmp ret i4 %tmp
} }

llvm/test/CodeGen/ARM/sadd_sat_plus.ll

Show All 36 Lines
; CHECK-T1-NEXT: mov r0, r1 ; CHECK-T1-NEXT: mov r0, r1
; CHECK-T1-NEXT: .LBB0_6: ; CHECK-T1-NEXT: .LBB0_6:
; CHECK-T1-NEXT: bx lr ; CHECK-T1-NEXT: bx lr
; CHECK-T1-NEXT: .p2align 2 ; CHECK-T1-NEXT: .p2align 2
; CHECK-T1-NEXT: @ %bb.7: ; CHECK-T1-NEXT: @ %bb.7:
; CHECK-T1-NEXT: .LCPI0_0: ; CHECK-T1-NEXT: .LCPI0_0:
; CHECK-T1-NEXT: .long 2147483647 @ 0x7fffffff ; CHECK-T1-NEXT: .long 2147483647 @ 0x7fffffff
; ;
; CHECK-T2-LABEL: func32: ; CHECK-T2NODSP-LABEL: func32:
; CHECK-T2: @ %bb.0: ; CHECK-T2NODSP: @ %bb.0:
; CHECK-T2-NEXT: mla r2, r1, r2, r0 ; CHECK-T2NODSP-NEXT: mla r2, r1, r2, r0
; CHECK-T2-NEXT: movs r3, #0 ; CHECK-T2NODSP-NEXT: movs r3, #0
; CHECK-T2-NEXT: mov.w r1, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w r1, #-2147483648
; CHECK-T2-NEXT: cmp r2, #0 ; CHECK-T2NODSP-NEXT: cmp r2, #0
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi r3, #1 ; CHECK-T2NODSP-NEXT: movmi r3, #1
; CHECK-T2-NEXT: cmp r3, #0 ; CHECK-T2NODSP-NEXT: cmp r3, #0
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne r1, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne r1, #-2147483648
; CHECK-T2-NEXT: cmp r2, r0 ; CHECK-T2NODSP-NEXT: cmp r2, r0
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc r1, r2 ; CHECK-T2NODSP-NEXT: movvc r1, r2
; CHECK-T2-NEXT: mov r0, r1 ; CHECK-T2NODSP-NEXT: mov r0, r1
; CHECK-T2-NEXT: bx lr ; CHECK-T2NODSP-NEXT: bx lr
;
; CHECK-T2DSP-LABEL: func32:
; CHECK-T2DSP: @ %bb.0:
; CHECK-T2DSP-NEXT: muls r1, r2, r1
; CHECK-T2DSP-NEXT: qadd r0, r0, r1
; CHECK-T2DSP-NEXT: bx lr
; ;
; CHECK-ARM-LABEL: func32: ; CHECK-ARM-LABEL: func32:
; CHECK-ARM: @ %bb.0: ; CHECK-ARM: @ %bb.0:
; CHECK-ARM-NEXT: mla r2, r1, r2, r0 ; CHECK-ARM-NEXT: mul r1, r1, r2
; CHECK-ARM-NEXT: mov r3, #0 ; CHECK-ARM-NEXT: qadd r0, r0, r1
; CHECK-ARM-NEXT: mov r1, #-2147483648
; CHECK-ARM-NEXT: cmp r2, #0
; CHECK-ARM-NEXT: movwmi r3, #1
; CHECK-ARM-NEXT: cmp r3, #0
; CHECK-ARM-NEXT: mvnne r1, #-2147483648
; CHECK-ARM-NEXT: cmp r2, r0
; CHECK-ARM-NEXT: movvc r1, r2
; CHECK-ARM-NEXT: mov r0, r1
; CHECK-ARM-NEXT: bx lr ; CHECK-ARM-NEXT: bx lr
%a = mul i32 %y, %z %a = mul i32 %y, %z
%tmp = call i32 @llvm.sadd.sat.i32(i32 %x, i32 %a) %tmp = call i32 @llvm.sadd.sat.i32(i32 %x, i32 %a)
ret i32 %tmp ret i32 %tmp
} }
define i64 @func64(i64 %x, i64 %y, i64 %z) nounwind { define i64 @func64(i64 %x, i64 %y, i64 %z) nounwind {
; CHECK-T1-LABEL: func64: ; CHECK-T1-LABEL: func64:
▲ Show 20 Lines • Show All 258 Lines • ▼ Show 20 Lines
; CHECK-T1-NEXT: mvns r1, r1 ; CHECK-T1-NEXT: mvns r1, r1
; CHECK-T1-NEXT: cmp r0, r1 ; CHECK-T1-NEXT: cmp r0, r1
; CHECK-T1-NEXT: bgt .LBB4_4 ; CHECK-T1-NEXT: bgt .LBB4_4
; CHECK-T1-NEXT: @ %bb.3: ; CHECK-T1-NEXT: @ %bb.3:
; CHECK-T1-NEXT: mov r0, r1 ; CHECK-T1-NEXT: mov r0, r1
; CHECK-T1-NEXT: .LBB4_4: ; CHECK-T1-NEXT: .LBB4_4:
; CHECK-T1-NEXT: bx lr ; CHECK-T1-NEXT: bx lr
; ;
; CHECK-T2-LABEL: func4: ; CHECK-T2NODSP-LABEL: func4:
; CHECK-T2: @ %bb.0: ; CHECK-T2NODSP: @ %bb.0:
; CHECK-T2-NEXT: muls r1, r2, r1 ; CHECK-T2NODSP-NEXT: muls r1, r2, r1
; CHECK-T2-NEXT: lsls r1, r1, #28 ; CHECK-T2NODSP-NEXT: lsls r1, r1, #28
; CHECK-T2-NEXT: add.w r0, r0, r1, asr #28 ; CHECK-T2NODSP-NEXT: add.w r0, r0, r1, asr #28
; CHECK-T2-NEXT: cmp r0, #7 ; CHECK-T2NODSP-NEXT: cmp r0, #7
; CHECK-T2-NEXT: it ge ; CHECK-T2NODSP-NEXT: it ge
; CHECK-T2-NEXT: movge r0, #7 ; CHECK-T2NODSP-NEXT: movge r0, #7
; CHECK-T2-NEXT: cmn.w r0, #8 ; CHECK-T2NODSP-NEXT: cmn.w r0, #8
; CHECK-T2-NEXT: it le ; CHECK-T2NODSP-NEXT: it le
; CHECK-T2-NEXT: mvnle r0, #7 ; CHECK-T2NODSP-NEXT: mvnle r0, #7
; CHECK-T2-NEXT: bx lr ; CHECK-T2NODSP-NEXT: bx lr
;
; CHECK-T2DSP-LABEL: func4:
; CHECK-T2DSP: @ %bb.0:
; CHECK-T2DSP-NEXT: muls r1, r2, r1
; CHECK-T2DSP-NEXT: lsls r0, r0, #28
; CHECK-T2DSP-NEXT: lsls r1, r1, #28
; CHECK-T2DSP-NEXT: qadd r0, r0, r1
; CHECK-T2DSP-NEXT: asrs r0, r0, #28
; CHECK-T2DSP-NEXT: bx lr
; ;
; CHECK-ARM-LABEL: func4: ; CHECK-ARM-LABEL: func4:
; CHECK-ARM: @ %bb.0: ; CHECK-ARM: @ %bb.0:
; CHECK-ARM-NEXT: smulbb r1, r1, r2 ; CHECK-ARM-NEXT: smulbb r1, r1, r2
; CHECK-ARM-NEXT: lsl r0, r0, #28
; CHECK-ARM-NEXT: lsl r1, r1, #28 ; CHECK-ARM-NEXT: lsl r1, r1, #28
; CHECK-ARM-NEXT: add r0, r0, r1, asr #28 ; CHECK-ARM-NEXT: qadd r0, r0, r1
; CHECK-ARM-NEXT: cmp r0, #7 ; CHECK-ARM-NEXT: asr r0, r0, #28
; CHECK-ARM-NEXT: movge r0, #7
; CHECK-ARM-NEXT: cmn r0, #8
; CHECK-ARM-NEXT: mvnle r0, #7
; CHECK-ARM-NEXT: bx lr ; CHECK-ARM-NEXT: bx lr
%a = mul i4 %y, %z %a = mul i4 %y, %z
%tmp = call i4 @llvm.sadd.sat.i4(i4 %x, i4 %a) %tmp = call i4 @llvm.sadd.sat.i4(i4 %x, i4 %a)
ret i4 %tmp ret i4 %tmp
} }

llvm/test/CodeGen/ARM/ssub_sat.ll

Show All 40 Lines
; CHECK-T1-NEXT: mov r0, r3 ; CHECK-T1-NEXT: mov r0, r3
; CHECK-T1-NEXT: .LBB0_6: ; CHECK-T1-NEXT: .LBB0_6:
; CHECK-T1-NEXT: pop {r4, pc} ; CHECK-T1-NEXT: pop {r4, pc}
; CHECK-T1-NEXT: .p2align 2 ; CHECK-T1-NEXT: .p2align 2
; CHECK-T1-NEXT: @ %bb.7: ; CHECK-T1-NEXT: @ %bb.7:
; CHECK-T1-NEXT: .LCPI0_0: ; CHECK-T1-NEXT: .LCPI0_0:
; CHECK-T1-NEXT: .long 2147483647 @ 0x7fffffff ; CHECK-T1-NEXT: .long 2147483647 @ 0x7fffffff
; ;
; CHECK-T2-LABEL: func: ; CHECK-T2NODSP-LABEL: func:
; CHECK-T2: @ %bb.0: ; CHECK-T2NODSP: @ %bb.0:
; CHECK-T2-NEXT: subs.w r12, r0, r1 ; CHECK-T2NODSP-NEXT: subs.w r12, r0, r1
; CHECK-T2-NEXT: mov.w r3, #0 ; CHECK-T2NODSP-NEXT: mov.w r3, #0
; CHECK-T2-NEXT: mov.w r2, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w r2, #-2147483648
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi r3, #1 ; CHECK-T2NODSP-NEXT: movmi r3, #1
; CHECK-T2-NEXT: cmp r3, #0 ; CHECK-T2NODSP-NEXT: cmp r3, #0
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne r2, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne r2, #-2147483648
; CHECK-T2-NEXT: cmp r0, r1 ; CHECK-T2NODSP-NEXT: cmp r0, r1
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc r2, r12 ; CHECK-T2NODSP-NEXT: movvc r2, r12
; CHECK-T2-NEXT: mov r0, r2 ; CHECK-T2NODSP-NEXT: mov r0, r2
; CHECK-T2-NEXT: bx lr ; CHECK-T2NODSP-NEXT: bx lr
; ;
; CHECK-ARM-LABEL: func: ; CHECK-T2DSP-LABEL: func:
; CHECK-ARM: @ %bb.0: ; CHECK-T2DSP: @ %bb.0:
; CHECK-ARM-NEXT: subs r12, r0, r1 ; CHECK-T2DSP-NEXT: qsub r0, r0, r1
; CHECK-ARM-NEXT: mov r3, #0 ; CHECK-T2DSP-NEXT: bx lr
; CHECK-ARM-NEXT: movmi r3, #1 ;
; CHECK-ARM-NEXT: mov r2, #-2147483648 ; CHECK-ARMNODPS-LABEL: func:
; CHECK-ARM-NEXT: cmp r3, #0 ; CHECK-ARMNODPS: @ %bb.0:
; CHECK-ARM-NEXT: mvnne r2, #-2147483648 ; CHECK-ARMNODPS-NEXT: subs r12, r0, r1
; CHECK-ARM-NEXT: cmp r0, r1 ; CHECK-ARMNODPS-NEXT: mov r3, #0
; CHECK-ARM-NEXT: movvc r2, r12 ; CHECK-ARMNODPS-NEXT: movmi r3, #1
; CHECK-ARM-NEXT: mov r0, r2 ; CHECK-ARMNODPS-NEXT: mov r2, #-2147483648
; CHECK-ARM-NEXT: bx lr ; CHECK-ARMNODPS-NEXT: cmp r3, #0
; CHECK-ARMNODPS-NEXT: mvnne r2, #-2147483648
; CHECK-ARMNODPS-NEXT: cmp r0, r1
; CHECK-ARMNODPS-NEXT: movvc r2, r12
; CHECK-ARMNODPS-NEXT: mov r0, r2
; CHECK-ARMNODPS-NEXT: bx lr
;
; CHECK-ARMBASEDSP-LABEL: func:
; CHECK-ARMBASEDSP: @ %bb.0:
; CHECK-ARMBASEDSP-NEXT: qsub r0, r0, r1
; CHECK-ARMBASEDSP-NEXT: bx lr
;
; CHECK-ARMDSP-LABEL: func:
; CHECK-ARMDSP: @ %bb.0:
; CHECK-ARMDSP-NEXT: qsub r0, r0, r1
; CHECK-ARMDSP-NEXT: bx lr
%tmp = call i32 @llvm.ssub.sat.i32(i32 %x, i32 %y) %tmp = call i32 @llvm.ssub.sat.i32(i32 %x, i32 %y)
ret i32 %tmp ret i32 %tmp
} }
define i64 @func2(i64 %x, i64 %y) nounwind { define i64 @func2(i64 %x, i64 %y) nounwind {
; CHECK-T1-LABEL: func2: ; CHECK-T1-LABEL: func2:
; CHECK-T1: @ %bb.0: ; CHECK-T1: @ %bb.0:
; CHECK-T1-NEXT: .save {r4, r5, r6, r7, lr} ; CHECK-T1-NEXT: .save {r4, r5, r6, r7, lr}
▲ Show 20 Lines • Show All 183 Lines • ▼ Show 20 Lines
; CHECK-ARMNODPS-NEXT: bx lr ; CHECK-ARMNODPS-NEXT: bx lr
; CHECK-ARMNODPS-NEXT: .p2align 2 ; CHECK-ARMNODPS-NEXT: .p2align 2
; CHECK-ARMNODPS-NEXT: @ %bb.1: ; CHECK-ARMNODPS-NEXT: @ %bb.1:
; CHECK-ARMNODPS-NEXT: .LCPI2_0: ; CHECK-ARMNODPS-NEXT: .LCPI2_0:
; CHECK-ARMNODPS-NEXT: .long 4294934528 @ 0xffff8000 ; CHECK-ARMNODPS-NEXT: .long 4294934528 @ 0xffff8000
; ;
; CHECK-ARMBASEDSP-LABEL: func16: ; CHECK-ARMBASEDSP-LABEL: func16:
; CHECK-ARMBASEDSP: @ %bb.0: ; CHECK-ARMBASEDSP: @ %bb.0:
; CHECK-ARMBASEDSP-NEXT: sub r0, r0, r1 ; CHECK-ARMBASEDSP-NEXT: lsl r0, r0, #16
; CHECK-ARMBASEDSP-NEXT: mov r1, #255 ; CHECK-ARMBASEDSP-NEXT: lsl r1, r1, #16
; CHECK-ARMBASEDSP-NEXT: orr r1, r1, #32512 ; CHECK-ARMBASEDSP-NEXT: qsub r0, r0, r1
; CHECK-ARMBASEDSP-NEXT: cmp r0, r1 ; CHECK-ARMBASEDSP-NEXT: asr r0, r0, #16
; CHECK-ARMBASEDSP-NEXT: movlt r1, r0
; CHECK-ARMBASEDSP-NEXT: ldr r0, .LCPI2_0
; CHECK-ARMBASEDSP-NEXT: cmn r1, #32768
; CHECK-ARMBASEDSP-NEXT: movgt r0, r1
; CHECK-ARMBASEDSP-NEXT: bx lr ; CHECK-ARMBASEDSP-NEXT: bx lr
; CHECK-ARMBASEDSP-NEXT: .p2align 2
; CHECK-ARMBASEDSP-NEXT: @ %bb.1:
; CHECK-ARMBASEDSP-NEXT: .LCPI2_0:
; CHECK-ARMBASEDSP-NEXT: .long 4294934528 @ 0xffff8000
; ;
; CHECK-ARMDSP-LABEL: func16: ; CHECK-ARMDSP-LABEL: func16:
; CHECK-ARMDSP: @ %bb.0: ; CHECK-ARMDSP: @ %bb.0:
; CHECK-ARMDSP-NEXT: qsub16 r0, r0, r1 ; CHECK-ARMDSP-NEXT: qsub16 r0, r0, r1
; CHECK-ARMDSP-NEXT: sxth r0, r0 ; CHECK-ARMDSP-NEXT: sxth r0, r0
; CHECK-ARMDSP-NEXT: bx lr ; CHECK-ARMDSP-NEXT: bx lr
%tmp = call i16 @llvm.ssub.sat.i16(i16 %x, i16 %y) %tmp = call i16 @llvm.ssub.sat.i16(i16 %x, i16 %y)
ret i16 %tmp ret i16 %tmp
Show All 40 Lines
; CHECK-ARMNODPS-NEXT: cmp r0, #127 ; CHECK-ARMNODPS-NEXT: cmp r0, #127
; CHECK-ARMNODPS-NEXT: movge r0, #127 ; CHECK-ARMNODPS-NEXT: movge r0, #127
; CHECK-ARMNODPS-NEXT: cmn r0, #128 ; CHECK-ARMNODPS-NEXT: cmn r0, #128
; CHECK-ARMNODPS-NEXT: mvnle r0, #127 ; CHECK-ARMNODPS-NEXT: mvnle r0, #127
; CHECK-ARMNODPS-NEXT: bx lr ; CHECK-ARMNODPS-NEXT: bx lr
; ;
; CHECK-ARMBASEDSP-LABEL: func8: ; CHECK-ARMBASEDSP-LABEL: func8:
; CHECK-ARMBASEDSP: @ %bb.0: ; CHECK-ARMBASEDSP: @ %bb.0:
; CHECK-ARMBASEDSP-NEXT: sub r0, r0, r1 ; CHECK-ARMBASEDSP-NEXT: lsl r0, r0, #24
; CHECK-ARMBASEDSP-NEXT: cmp r0, #127 ; CHECK-ARMBASEDSP-NEXT: lsl r1, r1, #24
; CHECK-ARMBASEDSP-NEXT: movge r0, #127 ; CHECK-ARMBASEDSP-NEXT: qsub r0, r0, r1
; CHECK-ARMBASEDSP-NEXT: cmn r0, #128 ; CHECK-ARMBASEDSP-NEXT: asr r0, r0, #24
; CHECK-ARMBASEDSP-NEXT: mvnle r0, #127
; CHECK-ARMBASEDSP-NEXT: bx lr ; CHECK-ARMBASEDSP-NEXT: bx lr
; ;
; CHECK-ARMDSP-LABEL: func8: ; CHECK-ARMDSP-LABEL: func8:
; CHECK-ARMDSP: @ %bb.0: ; CHECK-ARMDSP: @ %bb.0:
; CHECK-ARMDSP-NEXT: qsub8 r0, r0, r1 ; CHECK-ARMDSP-NEXT: qsub8 r0, r0, r1
; CHECK-ARMDSP-NEXT: sxtb r0, r0 ; CHECK-ARMDSP-NEXT: sxtb r0, r0
; CHECK-ARMDSP-NEXT: bx lr ; CHECK-ARMDSP-NEXT: bx lr
%tmp = call i8 @llvm.ssub.sat.i8(i8 %x, i8 %y) %tmp = call i8 @llvm.ssub.sat.i8(i8 %x, i8 %y)
Show All 13 Lines
; CHECK-T1-NEXT: mvns r1, r1 ; CHECK-T1-NEXT: mvns r1, r1
; CHECK-T1-NEXT: cmp r0, r1 ; CHECK-T1-NEXT: cmp r0, r1
; CHECK-T1-NEXT: bgt .LBB4_4 ; CHECK-T1-NEXT: bgt .LBB4_4
; CHECK-T1-NEXT: @ %bb.3: ; CHECK-T1-NEXT: @ %bb.3:
; CHECK-T1-NEXT: mov r0, r1 ; CHECK-T1-NEXT: mov r0, r1
; CHECK-T1-NEXT: .LBB4_4: ; CHECK-T1-NEXT: .LBB4_4:
; CHECK-T1-NEXT: bx lr ; CHECK-T1-NEXT: bx lr
; ;
; CHECK-T2-LABEL: func3: ; CHECK-T2NODSP-LABEL: func3:
; CHECK-T2: @ %bb.0: ; CHECK-T2NODSP: @ %bb.0:
; CHECK-T2-NEXT: subs r0, r0, r1 ; CHECK-T2NODSP-NEXT: subs r0, r0, r1
; CHECK-T2-NEXT: cmp r0, #7 ; CHECK-T2NODSP-NEXT: cmp r0, #7
; CHECK-T2-NEXT: it ge ; CHECK-T2NODSP-NEXT: it ge
; CHECK-T2-NEXT: movge r0, #7 ; CHECK-T2NODSP-NEXT: movge r0, #7
; CHECK-T2-NEXT: cmn.w r0, #8 ; CHECK-T2NODSP-NEXT: cmn.w r0, #8
; CHECK-T2-NEXT: it le ; CHECK-T2NODSP-NEXT: it le
; CHECK-T2-NEXT: mvnle r0, #7 ; CHECK-T2NODSP-NEXT: mvnle r0, #7
; CHECK-T2-NEXT: bx lr ; CHECK-T2NODSP-NEXT: bx lr
; ;
; CHECK-ARM-LABEL: func3: ; CHECK-T2DSP-LABEL: func3:
; CHECK-ARM: @ %bb.0: ; CHECK-T2DSP: @ %bb.0:
; CHECK-ARM-NEXT: sub r0, r0, r1 ; CHECK-T2DSP-NEXT: lsls r1, r1, #28
; CHECK-ARM-NEXT: cmp r0, #7 ; CHECK-T2DSP-NEXT: lsls r0, r0, #28
; CHECK-ARM-NEXT: movge r0, #7 ; CHECK-T2DSP-NEXT: qsub r0, r0, r1
; CHECK-ARM-NEXT: cmn r0, #8 ; CHECK-T2DSP-NEXT: asrs r0, r0, #28
; CHECK-ARM-NEXT: mvnle r0, #7 ; CHECK-T2DSP-NEXT: bx lr
; CHECK-ARM-NEXT: bx lr ;
; CHECK-ARMNODPS-LABEL: func3:
; CHECK-ARMNODPS: @ %bb.0:
; CHECK-ARMNODPS-NEXT: sub r0, r0, r1
; CHECK-ARMNODPS-NEXT: cmp r0, #7
; CHECK-ARMNODPS-NEXT: movge r0, #7
; CHECK-ARMNODPS-NEXT: cmn r0, #8
; CHECK-ARMNODPS-NEXT: mvnle r0, #7
; CHECK-ARMNODPS-NEXT: bx lr
;
; CHECK-ARMBASEDSP-LABEL: func3:
; CHECK-ARMBASEDSP: @ %bb.0:
; CHECK-ARMBASEDSP-NEXT: lsl r0, r0, #28
; CHECK-ARMBASEDSP-NEXT: lsl r1, r1, #28
; CHECK-ARMBASEDSP-NEXT: qsub r0, r0, r1
; CHECK-ARMBASEDSP-NEXT: asr r0, r0, #28
; CHECK-ARMBASEDSP-NEXT: bx lr
;
; CHECK-ARMDSP-LABEL: func3:
; CHECK-ARMDSP: @ %bb.0:
; CHECK-ARMDSP-NEXT: lsl r0, r0, #28
; CHECK-ARMDSP-NEXT: lsl r1, r1, #28
; CHECK-ARMDSP-NEXT: qsub r0, r0, r1
; CHECK-ARMDSP-NEXT: asr r0, r0, #28
; CHECK-ARMDSP-NEXT: bx lr
%tmp = call i4 @llvm.ssub.sat.i4(i4 %x, i4 %y) %tmp = call i4 @llvm.ssub.sat.i4(i4 %x, i4 %y)
ret i4 %tmp ret i4 %tmp
} }
define <4 x i32> @vec(<4 x i32> %x, <4 x i32> %y) nounwind { define <4 x i32> @vec(<4 x i32> %x, <4 x i32> %y) nounwind {
; CHECK-T1-LABEL: vec: ; CHECK-T1-LABEL: vec:
; CHECK-T1: @ %bb.0: ; CHECK-T1: @ %bb.0:
; CHECK-T1-NEXT: .save {r4, r5, r6, r7, lr} ; CHECK-T1-NEXT: .save {r4, r5, r6, r7, lr}
▲ Show 20 Lines • Show All 90 Lines • ▼ Show 20 Lines
; CHECK-T1-NEXT: cmp r6, r2 ; CHECK-T1-NEXT: cmp r6, r2
; CHECK-T1-NEXT: bvs .LBB5_21 ; CHECK-T1-NEXT: bvs .LBB5_21
; CHECK-T1-NEXT: b .LBB5_22 ; CHECK-T1-NEXT: b .LBB5_22
; CHECK-T1-NEXT: .p2align 2 ; CHECK-T1-NEXT: .p2align 2
; CHECK-T1-NEXT: @ %bb.25: ; CHECK-T1-NEXT: @ %bb.25:
; CHECK-T1-NEXT: .LCPI5_0: ; CHECK-T1-NEXT: .LCPI5_0:
; CHECK-T1-NEXT: .long 2147483647 @ 0x7fffffff ; CHECK-T1-NEXT: .long 2147483647 @ 0x7fffffff
; ;
; CHECK-T2-LABEL: vec: ; CHECK-T2NODSP-LABEL: vec:
; CHECK-T2: @ %bb.0: ; CHECK-T2NODSP: @ %bb.0:
; CHECK-T2-NEXT: .save {r4, r5, r6, r7, lr} ; CHECK-T2NODSP-NEXT: .save {r4, r5, r6, r7, lr}
; CHECK-T2-NEXT: push {r4, r5, r6, r7, lr} ; CHECK-T2NODSP-NEXT: push {r4, r5, r6, r7, lr}
; CHECK-T2-NEXT: .pad #4 ; CHECK-T2NODSP-NEXT: .pad #4
; CHECK-T2-NEXT: sub sp, #4 ; CHECK-T2NODSP-NEXT: sub sp, #4
; CHECK-T2-NEXT: ldr r4, [sp, #24] ; CHECK-T2NODSP-NEXT: ldr r4, [sp, #24]
; CHECK-T2-NEXT: mov lr, r0 ; CHECK-T2NODSP-NEXT: mov lr, r0
; CHECK-T2-NEXT: ldr r7, [sp, #28] ; CHECK-T2NODSP-NEXT: ldr r7, [sp, #28]
; CHECK-T2-NEXT: movs r5, #0 ; CHECK-T2NODSP-NEXT: movs r5, #0
; CHECK-T2-NEXT: subs r6, r0, r4 ; CHECK-T2NODSP-NEXT: subs r6, r0, r4
; CHECK-T2-NEXT: mov.w r0, #0 ; CHECK-T2NODSP-NEXT: mov.w r0, #0
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi r0, #1 ; CHECK-T2NODSP-NEXT: movmi r0, #1
; CHECK-T2-NEXT: cmp r0, #0 ; CHECK-T2NODSP-NEXT: cmp r0, #0
; CHECK-T2-NEXT: mov.w r0, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w r0, #-2147483648
; CHECK-T2-NEXT: mov.w r12, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w r12, #-2147483648
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne r0, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne r0, #-2147483648
; CHECK-T2-NEXT: cmp lr, r4 ; CHECK-T2NODSP-NEXT: cmp lr, r4
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc r0, r6 ; CHECK-T2NODSP-NEXT: movvc r0, r6
; CHECK-T2-NEXT: subs r6, r1, r7 ; CHECK-T2NODSP-NEXT: subs r6, r1, r7
; CHECK-T2-NEXT: mov.w r4, #0 ; CHECK-T2NODSP-NEXT: mov.w r4, #0
; CHECK-T2-NEXT: mov.w lr, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w lr, #-2147483648
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi r4, #1 ; CHECK-T2NODSP-NEXT: movmi r4, #1
; CHECK-T2-NEXT: cmp r4, #0 ; CHECK-T2NODSP-NEXT: cmp r4, #0
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne lr, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne lr, #-2147483648
; CHECK-T2-NEXT: cmp r1, r7 ; CHECK-T2NODSP-NEXT: cmp r1, r7
; CHECK-T2-NEXT: ldr r1, [sp, #32] ; CHECK-T2NODSP-NEXT: ldr r1, [sp, #32]
; CHECK-T2-NEXT: mov.w r4, #0 ; CHECK-T2NODSP-NEXT: mov.w r4, #0
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc lr, r6 ; CHECK-T2NODSP-NEXT: movvc lr, r6
; CHECK-T2-NEXT: subs r6, r2, r1 ; CHECK-T2NODSP-NEXT: subs r6, r2, r1
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi r4, #1 ; CHECK-T2NODSP-NEXT: movmi r4, #1
; CHECK-T2-NEXT: cmp r4, #0 ; CHECK-T2NODSP-NEXT: cmp r4, #0
; CHECK-T2-NEXT: mov.w r4, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w r4, #-2147483648
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne r4, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne r4, #-2147483648
; CHECK-T2-NEXT: cmp r2, r1 ; CHECK-T2NODSP-NEXT: cmp r2, r1
; CHECK-T2-NEXT: ldr r1, [sp, #36] ; CHECK-T2NODSP-NEXT: ldr r1, [sp, #36]
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc r4, r6 ; CHECK-T2NODSP-NEXT: movvc r4, r6
; CHECK-T2-NEXT: subs r2, r3, r1 ; CHECK-T2NODSP-NEXT: subs r2, r3, r1
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi r5, #1 ; CHECK-T2NODSP-NEXT: movmi r5, #1
; CHECK-T2-NEXT: cmp r5, #0 ; CHECK-T2NODSP-NEXT: cmp r5, #0
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne r12, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne r12, #-2147483648
; CHECK-T2-NEXT: cmp r3, r1 ; CHECK-T2NODSP-NEXT: cmp r3, r1
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc r12, r2 ; CHECK-T2NODSP-NEXT: movvc r12, r2
; CHECK-T2-NEXT: mov r1, lr ; CHECK-T2NODSP-NEXT: mov r1, lr
; CHECK-T2-NEXT: mov r2, r4 ; CHECK-T2NODSP-NEXT: mov r2, r4
; CHECK-T2-NEXT: mov r3, r12 ; CHECK-T2NODSP-NEXT: mov r3, r12
; CHECK-T2-NEXT: add sp, #4 ; CHECK-T2NODSP-NEXT: add sp, #4
; CHECK-T2-NEXT: pop {r4, r5, r6, r7, pc} ; CHECK-T2NODSP-NEXT: pop {r4, r5, r6, r7, pc}
; ;
; CHECK-ARM-LABEL: vec: ; CHECK-T2DSP-LABEL: vec:
; CHECK-ARM: @ %bb.0: ; CHECK-T2DSP: @ %bb.0:
; CHECK-ARM-NEXT: .save {r4, r5, r6, r7, r11, lr} ; CHECK-T2DSP-NEXT: ldr.w r12, [sp]
; CHECK-ARM-NEXT: push {r4, r5, r6, r7, r11, lr} ; CHECK-T2DSP-NEXT: qsub r0, r0, r12
; CHECK-ARM-NEXT: ldr r4, [sp, #24] ; CHECK-T2DSP-NEXT: ldr.w r12, [sp, #4]
; CHECK-ARM-NEXT: mov lr, r0 ; CHECK-T2DSP-NEXT: qsub r1, r1, r12
; CHECK-ARM-NEXT: ldr r7, [sp, #28] ; CHECK-T2DSP-NEXT: ldr.w r12, [sp, #8]
; CHECK-ARM-NEXT: mov r5, #0 ; CHECK-T2DSP-NEXT: qsub r2, r2, r12
; CHECK-ARM-NEXT: subs r6, r0, r4 ; CHECK-T2DSP-NEXT: ldr.w r12, [sp, #12]
; CHECK-ARM-NEXT: mov r0, #0 ; CHECK-T2DSP-NEXT: qsub r3, r3, r12
; CHECK-ARM-NEXT: movmi r0, #1 ; CHECK-T2DSP-NEXT: bx lr
; CHECK-ARM-NEXT: cmp r0, #0 ;
; CHECK-ARM-NEXT: mov r0, #-2147483648 ; CHECK-ARMNODPS-LABEL: vec:
; CHECK-ARM-NEXT: mov r12, #-2147483648 ; CHECK-ARMNODPS: @ %bb.0:
; CHECK-ARM-NEXT: mvnne r0, #-2147483648 ; CHECK-ARMNODPS-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-ARM-NEXT: cmp lr, r4 ; CHECK-ARMNODPS-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-ARM-NEXT: movvc r0, r6 ; CHECK-ARMNODPS-NEXT: ldr r4, [sp, #24]
; CHECK-ARM-NEXT: subs r6, r1, r7 ; CHECK-ARMNODPS-NEXT: mov lr, r0
; CHECK-ARM-NEXT: mov r4, #0 ; CHECK-ARMNODPS-NEXT: ldr r7, [sp, #28]
; CHECK-ARM-NEXT: mov lr, #-2147483648 ; CHECK-ARMNODPS-NEXT: mov r5, #0
; CHECK-ARM-NEXT: movmi r4, #1 ; CHECK-ARMNODPS-NEXT: subs r6, r0, r4
; CHECK-ARM-NEXT: cmp r4, #0 ; CHECK-ARMNODPS-NEXT: mov r0, #0
; CHECK-ARM-NEXT: mvnne lr, #-2147483648 ; CHECK-ARMNODPS-NEXT: movmi r0, #1
; CHECK-ARM-NEXT: cmp r1, r7 ; CHECK-ARMNODPS-NEXT: cmp r0, #0
; CHECK-ARM-NEXT: ldr r1, [sp, #32] ; CHECK-ARMNODPS-NEXT: mov r0, #-2147483648
; CHECK-ARM-NEXT: movvc lr, r6 ; CHECK-ARMNODPS-NEXT: mov r12, #-2147483648
; CHECK-ARM-NEXT: mov r4, #0 ; CHECK-ARMNODPS-NEXT: mvnne r0, #-2147483648
; CHECK-ARM-NEXT: subs r6, r2, r1 ; CHECK-ARMNODPS-NEXT: cmp lr, r4
; CHECK-ARM-NEXT: movmi r4, #1 ; CHECK-ARMNODPS-NEXT: movvc r0, r6
; CHECK-ARM-NEXT: cmp r4, #0 ; CHECK-ARMNODPS-NEXT: subs r6, r1, r7
; CHECK-ARM-NEXT: mov r4, #-2147483648 ; CHECK-ARMNODPS-NEXT: mov r4, #0
; CHECK-ARM-NEXT: mvnne r4, #-2147483648 ; CHECK-ARMNODPS-NEXT: mov lr, #-2147483648
; CHECK-ARM-NEXT: cmp r2, r1 ; CHECK-ARMNODPS-NEXT: movmi r4, #1
; CHECK-ARM-NEXT: ldr r1, [sp, #36] ; CHECK-ARMNODPS-NEXT: cmp r4, #0
; CHECK-ARM-NEXT: movvc r4, r6 ; CHECK-ARMNODPS-NEXT: mvnne lr, #-2147483648
; CHECK-ARM-NEXT: subs r2, r3, r1 ; CHECK-ARMNODPS-NEXT: cmp r1, r7
; CHECK-ARM-NEXT: movmi r5, #1 ; CHECK-ARMNODPS-NEXT: ldr r1, [sp, #32]
; CHECK-ARM-NEXT: cmp r5, #0 ; CHECK-ARMNODPS-NEXT: movvc lr, r6
; CHECK-ARM-NEXT: mvnne r12, #-2147483648 ; CHECK-ARMNODPS-NEXT: mov r4, #0
; CHECK-ARM-NEXT: cmp r3, r1 ; CHECK-ARMNODPS-NEXT: subs r6, r2, r1
; CHECK-ARM-NEXT: movvc r12, r2 ; CHECK-ARMNODPS-NEXT: movmi r4, #1
; CHECK-ARM-NEXT: mov r1, lr ; CHECK-ARMNODPS-NEXT: cmp r4, #0
; CHECK-ARM-NEXT: mov r2, r4 ; CHECK-ARMNODPS-NEXT: mov r4, #-2147483648
; CHECK-ARM-NEXT: mov r3, r12 ; CHECK-ARMNODPS-NEXT: mvnne r4, #-2147483648
; CHECK-ARM-NEXT: pop {r4, r5, r6, r7, r11, pc} ; CHECK-ARMNODPS-NEXT: cmp r2, r1
; CHECK-ARMNODPS-NEXT: ldr r1, [sp, #36]
; CHECK-ARMNODPS-NEXT: movvc r4, r6
; CHECK-ARMNODPS-NEXT: subs r2, r3, r1
; CHECK-ARMNODPS-NEXT: movmi r5, #1
; CHECK-ARMNODPS-NEXT: cmp r5, #0
; CHECK-ARMNODPS-NEXT: mvnne r12, #-2147483648
; CHECK-ARMNODPS-NEXT: cmp r3, r1
; CHECK-ARMNODPS-NEXT: movvc r12, r2
; CHECK-ARMNODPS-NEXT: mov r1, lr
; CHECK-ARMNODPS-NEXT: mov r2, r4
; CHECK-ARMNODPS-NEXT: mov r3, r12
; CHECK-ARMNODPS-NEXT: pop {r4, r5, r6, r7, r11, pc}
;
; CHECK-ARMBASEDSP-LABEL: vec:
; CHECK-ARMBASEDSP: @ %bb.0:
; CHECK-ARMBASEDSP-NEXT: ldr r12, [sp]
; CHECK-ARMBASEDSP-NEXT: qsub r0, r0, r12
; CHECK-ARMBASEDSP-NEXT: ldr r12, [sp, #4]
; CHECK-ARMBASEDSP-NEXT: qsub r1, r1, r12
; CHECK-ARMBASEDSP-NEXT: ldr r12, [sp, #8]
; CHECK-ARMBASEDSP-NEXT: qsub r2, r2, r12
; CHECK-ARMBASEDSP-NEXT: ldr r12, [sp, #12]
; CHECK-ARMBASEDSP-NEXT: qsub r3, r3, r12
; CHECK-ARMBASEDSP-NEXT: bx lr
;
; CHECK-ARMDSP-LABEL: vec:
; CHECK-ARMDSP: @ %bb.0:
; CHECK-ARMDSP-NEXT: ldr r12, [sp]
; CHECK-ARMDSP-NEXT: qsub r0, r0, r12
; CHECK-ARMDSP-NEXT: ldr r12, [sp, #4]
; CHECK-ARMDSP-NEXT: qsub r1, r1, r12
; CHECK-ARMDSP-NEXT: ldr r12, [sp, #8]
; CHECK-ARMDSP-NEXT: qsub r2, r2, r12
; CHECK-ARMDSP-NEXT: ldr r12, [sp, #12]
; CHECK-ARMDSP-NEXT: qsub r3, r3, r12
; CHECK-ARMDSP-NEXT: bx lr
%tmp = call <4 x i32> @llvm.ssub.sat.v4i32(<4 x i32> %x, <4 x i32> %y) %tmp = call <4 x i32> @llvm.ssub.sat.v4i32(<4 x i32> %x, <4 x i32> %y)
ret <4 x i32> %tmp ret <4 x i32> %tmp
} }

llvm/test/CodeGen/ARM/ssub_sat_plus.ll

Show All 38 Lines
; CHECK-T1-NEXT: mov r0, r2 ; CHECK-T1-NEXT: mov r0, r2
; CHECK-T1-NEXT: .LBB0_6: ; CHECK-T1-NEXT: .LBB0_6:
; CHECK-T1-NEXT: pop {r4, pc} ; CHECK-T1-NEXT: pop {r4, pc}
; CHECK-T1-NEXT: .p2align 2 ; CHECK-T1-NEXT: .p2align 2
; CHECK-T1-NEXT: @ %bb.7: ; CHECK-T1-NEXT: @ %bb.7:
; CHECK-T1-NEXT: .LCPI0_0: ; CHECK-T1-NEXT: .LCPI0_0:
; CHECK-T1-NEXT: .long 2147483647 @ 0x7fffffff ; CHECK-T1-NEXT: .long 2147483647 @ 0x7fffffff
; ;
; CHECK-T2-LABEL: func32: ; CHECK-T2NODSP-LABEL: func32:
; CHECK-T2: @ %bb.0: ; CHECK-T2NODSP: @ %bb.0:
; CHECK-T2-NEXT: .save {r7, lr} ; CHECK-T2NODSP-NEXT: .save {r7, lr}
; CHECK-T2-NEXT: push {r7, lr} ; CHECK-T2NODSP-NEXT: push {r7, lr}
; CHECK-T2-NEXT: mls r12, r1, r2, r0 ; CHECK-T2NODSP-NEXT: mls r12, r1, r2, r0
; CHECK-T2-NEXT: mov.w lr, #0 ; CHECK-T2NODSP-NEXT: mov.w lr, #0
; CHECK-T2-NEXT: mov.w r3, #-2147483648 ; CHECK-T2NODSP-NEXT: mov.w r3, #-2147483648
; CHECK-T2-NEXT: muls r1, r2, r1 ; CHECK-T2NODSP-NEXT: muls r1, r2, r1
; CHECK-T2-NEXT: cmp.w r12, #0 ; CHECK-T2NODSP-NEXT: cmp.w r12, #0
; CHECK-T2-NEXT: it mi ; CHECK-T2NODSP-NEXT: it mi
; CHECK-T2-NEXT: movmi.w lr, #1 ; CHECK-T2NODSP-NEXT: movmi.w lr, #1
; CHECK-T2-NEXT: cmp.w lr, #0 ; CHECK-T2NODSP-NEXT: cmp.w lr, #0
; CHECK-T2-NEXT: it ne ; CHECK-T2NODSP-NEXT: it ne
; CHECK-T2-NEXT: mvnne r3, #-2147483648 ; CHECK-T2NODSP-NEXT: mvnne r3, #-2147483648
; CHECK-T2-NEXT: cmp r0, r1 ; CHECK-T2NODSP-NEXT: cmp r0, r1
; CHECK-T2-NEXT: it vc ; CHECK-T2NODSP-NEXT: it vc
; CHECK-T2-NEXT: movvc r3, r12 ; CHECK-T2NODSP-NEXT: movvc r3, r12
; CHECK-T2-NEXT: mov r0, r3 ; CHECK-T2NODSP-NEXT: mov r0, r3
; CHECK-T2-NEXT: pop {r7, pc} ; CHECK-T2NODSP-NEXT: pop {r7, pc}
;
; CHECK-T2DSP-LABEL: func32:
; CHECK-T2DSP: @ %bb.0:
; CHECK-T2DSP-NEXT: muls r1, r2, r1
; CHECK-T2DSP-NEXT: qsub r0, r0, r1
; CHECK-T2DSP-NEXT: bx lr
; ;
; CHECK-ARM-LABEL: func32: ; CHECK-ARM-LABEL: func32:
; CHECK-ARM: @ %bb.0: ; CHECK-ARM: @ %bb.0:
; CHECK-ARM-NEXT: mls r3, r1, r2, r0 ; CHECK-ARM-NEXT: mul r1, r1, r2
; CHECK-ARM-NEXT: mul r12, r1, r2 ; CHECK-ARM-NEXT: qsub r0, r0, r1
; CHECK-ARM-NEXT: mov r2, #0
; CHECK-ARM-NEXT: mov r1, #-2147483648
; CHECK-ARM-NEXT: cmp r3, #0
; CHECK-ARM-NEXT: movwmi r2, #1
; CHECK-ARM-NEXT: cmp r2, #0
; CHECK-ARM-NEXT: mvnne r1, #-2147483648
; CHECK-ARM-NEXT: cmp r0, r12
; CHECK-ARM-NEXT: movvc r1, r3
; CHECK-ARM-NEXT: mov r0, r1
; CHECK-ARM-NEXT: bx lr ; CHECK-ARM-NEXT: bx lr
%a = mul i32 %y, %z %a = mul i32 %y, %z
%tmp = call i32 @llvm.ssub.sat.i32(i32 %x, i32 %a) %tmp = call i32 @llvm.ssub.sat.i32(i32 %x, i32 %a)
ret i32 %tmp ret i32 %tmp
} }
define i64 @func64(i64 %x, i64 %y, i64 %z) nounwind { define i64 @func64(i64 %x, i64 %y, i64 %z) nounwind {
; CHECK-T1-LABEL: func64: ; CHECK-T1-LABEL: func64:
▲ Show 20 Lines • Show All 258 Lines • ▼ Show 20 Lines
; CHECK-T1-NEXT: mvns r1, r1 ; CHECK-T1-NEXT: mvns r1, r1
; CHECK-T1-NEXT: cmp r0, r1 ; CHECK-T1-NEXT: cmp r0, r1
; CHECK-T1-NEXT: bgt .LBB4_4 ; CHECK-T1-NEXT: bgt .LBB4_4
; CHECK-T1-NEXT: @ %bb.3: ; CHECK-T1-NEXT: @ %bb.3:
; CHECK-T1-NEXT: mov r0, r1 ; CHECK-T1-NEXT: mov r0, r1
; CHECK-T1-NEXT: .LBB4_4: ; CHECK-T1-NEXT: .LBB4_4:
; CHECK-T1-NEXT: bx lr ; CHECK-T1-NEXT: bx lr
; ;
; CHECK-T2-LABEL: func4: ; CHECK-T2NODSP-LABEL: func4:
; CHECK-T2: @ %bb.0: ; CHECK-T2NODSP: @ %bb.0:
; CHECK-T2-NEXT: muls r1, r2, r1 ; CHECK-T2NODSP-NEXT: muls r1, r2, r1
; CHECK-T2-NEXT: lsls r1, r1, #28 ; CHECK-T2NODSP-NEXT: lsls r1, r1, #28
; CHECK-T2-NEXT: sub.w r0, r0, r1, asr #28 ; CHECK-T2NODSP-NEXT: sub.w r0, r0, r1, asr #28
; CHECK-T2-NEXT: cmp r0, #7 ; CHECK-T2NODSP-NEXT: cmp r0, #7
; CHECK-T2-NEXT: it ge ; CHECK-T2NODSP-NEXT: it ge
; CHECK-T2-NEXT: movge r0, #7 ; CHECK-T2NODSP-NEXT: movge r0, #7
; CHECK-T2-NEXT: cmn.w r0, #8 ; CHECK-T2NODSP-NEXT: cmn.w r0, #8
; CHECK-T2-NEXT: it le ; CHECK-T2NODSP-NEXT: it le
; CHECK-T2-NEXT: mvnle r0, #7 ; CHECK-T2NODSP-NEXT: mvnle r0, #7
; CHECK-T2-NEXT: bx lr ; CHECK-T2NODSP-NEXT: bx lr
;
; CHECK-T2DSP-LABEL: func4:
; CHECK-T2DSP: @ %bb.0:
; CHECK-T2DSP-NEXT: muls r1, r2, r1
; CHECK-T2DSP-NEXT: lsls r0, r0, #28
; CHECK-T2DSP-NEXT: lsls r1, r1, #28
; CHECK-T2DSP-NEXT: qsub r0, r0, r1
; CHECK-T2DSP-NEXT: asrs r0, r0, #28
; CHECK-T2DSP-NEXT: bx lr
; ;
; CHECK-ARM-LABEL: func4: ; CHECK-ARM-LABEL: func4:
; CHECK-ARM: @ %bb.0: ; CHECK-ARM: @ %bb.0:
; CHECK-ARM-NEXT: smulbb r1, r1, r2 ; CHECK-ARM-NEXT: smulbb r1, r1, r2
; CHECK-ARM-NEXT: lsl r0, r0, #28
; CHECK-ARM-NEXT: lsl r1, r1, #28 ; CHECK-ARM-NEXT: lsl r1, r1, #28
; CHECK-ARM-NEXT: sub r0, r0, r1, asr #28 ; CHECK-ARM-NEXT: qsub r0, r0, r1
; CHECK-ARM-NEXT: cmp r0, #7 ; CHECK-ARM-NEXT: asr r0, r0, #28
; CHECK-ARM-NEXT: movge r0, #7
; CHECK-ARM-NEXT: cmn r0, #8
; CHECK-ARM-NEXT: mvnle r0, #7
; CHECK-ARM-NEXT: bx lr ; CHECK-ARM-NEXT: bx lr
%a = mul i4 %y, %z %a = mul i4 %y, %z
%tmp = call i4 @llvm.ssub.sat.i4(i4 %x, i4 %a) %tmp = call i4 @llvm.ssub.sat.i4(i4 %x, i4 %a)
ret i4 %tmp ret i4 %tmp
} }