[llvm] [LoongArch] make ABDS/ABDU legal for lsx/lasx (PR #134190)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Apr 2 19:05:26 PDT 2025
https://github.com/tangaac created https://github.com/llvm/llvm-project/pull/134190
None
>From c7d2307e6f084ecfdbe1e5062385b6d154ee1ff1 Mon Sep 17 00:00:00 2001
From: tangaac <tangyan01 at loongson.cn>
Date: Thu, 3 Apr 2025 09:47:12 +0800
Subject: [PATCH] make ABDS/ABDU legal for lsx/lasx
---
.../LoongArch/LoongArchISelLowering.cpp | 4 +
.../LoongArch/LoongArchLASXInstrInfo.td | 4 +
.../Target/LoongArch/LoongArchLSXInstrInfo.td | 4 +
.../LoongArch/lasx/ir-instruction/absd.ll | 151 +++++-------------
.../LoongArch/lsx/ir-instruction/absd.ll | 151 +++++-------------
5 files changed, 86 insertions(+), 228 deletions(-)
diff --git a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
index 900775eedfa7b..edb74eeb84b16 100644
--- a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
+++ b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
@@ -272,6 +272,8 @@ LoongArchTargetLowering::LoongArchTargetLowering(const TargetMachine &TM,
{ISD::SETNE, ISD::SETGE, ISD::SETGT, ISD::SETUGE, ISD::SETUGT}, VT,
Expand);
setOperationAction(ISD::SCALAR_TO_VECTOR, VT, Custom);
+ setOperationAction(ISD::ABDS, VT, Legal);
+ setOperationAction(ISD::ABDU, VT, Legal);
}
for (MVT VT : {MVT::v16i8, MVT::v8i16, MVT::v4i32})
setOperationAction(ISD::BITREVERSE, VT, Custom);
@@ -336,6 +338,8 @@ LoongArchTargetLowering::LoongArchTargetLowering(const TargetMachine &TM,
{ISD::SETNE, ISD::SETGE, ISD::SETGT, ISD::SETUGE, ISD::SETUGT}, VT,
Expand);
setOperationAction(ISD::SCALAR_TO_VECTOR, VT, Custom);
+ setOperationAction(ISD::ABDS, VT, Legal);
+ setOperationAction(ISD::ABDU, VT, Legal);
}
for (MVT VT : {MVT::v32i8, MVT::v16i16, MVT::v8i32})
setOperationAction(ISD::BITREVERSE, VT, Custom);
diff --git a/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td b/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td
index 7022fddf34100..415aa5f7820d2 100644
--- a/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td
+++ b/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td
@@ -1817,6 +1817,10 @@ def : Pat<(vt (concat_vectors LSX128:$vd, LSX128:$vj)),
(XVPERMI_Q (SUBREG_TO_REG (i64 0), LSX128:$vd, sub_128),
(SUBREG_TO_REG (i64 0), LSX128:$vj, sub_128), 2)>;
+// XVABSD_{B/H/W/D}[U]
+defm : PatXrXr<abds, "XVABSD">;
+defm : PatXrXrU<abdu, "XVABSD">;
+
} // Predicates = [HasExtLASX]
/// Intrinsic pattern
diff --git a/llvm/lib/Target/LoongArch/LoongArchLSXInstrInfo.td b/llvm/lib/Target/LoongArch/LoongArchLSXInstrInfo.td
index e37de4f545a2a..3e7aee04c11f8 100644
--- a/llvm/lib/Target/LoongArch/LoongArchLSXInstrInfo.td
+++ b/llvm/lib/Target/LoongArch/LoongArchLSXInstrInfo.td
@@ -1973,6 +1973,10 @@ def : Pat<(f32 f32imm_vldi:$in),
def : Pat<(f64 f64imm_vldi:$in),
(f64 (EXTRACT_SUBREG (VLDI (to_f64imm_vldi f64imm_vldi:$in)), sub_64))>;
+// VABSD_{B/H/W/D}[U]
+defm : PatVrVr<abds, "VABSD">;
+defm : PatVrVrU<abdu, "VABSD">;
+
} // Predicates = [HasExtLSX]
/// Intrinsic pattern
diff --git a/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/absd.ll b/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/absd.ll
index bd5b16f5147a2..c5df9f8420837 100644
--- a/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/absd.ll
+++ b/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/absd.ll
@@ -1,7 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=loongarch64 -mattr=+lasx < %s | FileCheck %s
-;; TODO: Currently LoongArch generates sub-optimal code for these cases
;; 1. trunc(abs(sub(sext(a),sext(b)))) -> abds(a,b) or abdu(a,b)
;; 2. abs(sub_nsw(x, y)) -> abds(a,b)
;; 3. sub(smax(a,b),smin(a,b)) -> abds(a,b) or abdu(a,b)
@@ -9,16 +8,12 @@
;; 5. sub(select(icmp(a,b),a,b),select(icmp(a,b),b,a)) -> abds(a,b) or abdu(a,b)
;;
;; abds / abdu can be lowered to xvabsd.{b/h/w/d} / xvabsd.{b/h/w/d}u instruction.
-;;
-;; Later patch will address it.
;; trunc(abs(sub(sext(a),sext(b)))) -> abds(a,b)
define <32 x i8> @xvabsd_b(<32 x i8> %a, <32 x i8> %b) {
; CHECK-LABEL: xvabsd_b:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.b $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.b $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.b $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.b $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a.sext = sext <32 x i8> %a to <32 x i16>
%b.sext = sext <32 x i8> %b to <32 x i16>
@@ -31,9 +26,7 @@ define <32 x i8> @xvabsd_b(<32 x i8> %a, <32 x i8> %b) {
define <16 x i16> @xvabsd_h(<16 x i16> %a, <16 x i16> %b) {
; CHECK-LABEL: xvabsd_h:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.h $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.h $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.h $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.h $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a.sext = sext <16 x i16> %a to <16 x i32>
%b.sext = sext <16 x i16> %b to <16 x i32>
@@ -46,9 +39,7 @@ define <16 x i16> @xvabsd_h(<16 x i16> %a, <16 x i16> %b) {
define <8 x i32> @xvabsd_w(<8 x i32> %a, <8 x i32> %b) {
; CHECK-LABEL: xvabsd_w:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.w $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.w $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.w $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.w $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a.sext = sext <8 x i32> %a to <8 x i64>
%b.sext = sext <8 x i32> %b to <8 x i64>
@@ -61,9 +52,7 @@ define <8 x i32> @xvabsd_w(<8 x i32> %a, <8 x i32> %b) {
define <4 x i64> @xvabsd_d(<4 x i64> %a, <4 x i64> %b) {
; CHECK-LABEL: xvabsd_d:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.d $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.d $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.d $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.d $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a.sext = sext <4 x i64> %a to <4 x i128>
%b.sext = sext <4 x i64> %b to <4 x i128>
@@ -76,9 +65,7 @@ define <4 x i64> @xvabsd_d(<4 x i64> %a, <4 x i64> %b) {
define <32 x i8> @xvabsd_bu(<32 x i8> %a, <32 x i8> %b) {
; CHECK-LABEL: xvabsd_bu:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.bu $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.bu $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.b $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.bu $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a.zext = zext <32 x i8> %a to <32 x i16>
%b.zext = zext <32 x i8> %b to <32 x i16>
@@ -91,9 +78,7 @@ define <32 x i8> @xvabsd_bu(<32 x i8> %a, <32 x i8> %b) {
define <16 x i16> @xvabsd_hu(<16 x i16> %a, <16 x i16> %b) {
; CHECK-LABEL: xvabsd_hu:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.hu $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.hu $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.h $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.hu $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a.zext = zext <16 x i16> %a to <16 x i32>
%b.zext = zext <16 x i16> %b to <16 x i32>
@@ -106,9 +91,7 @@ define <16 x i16> @xvabsd_hu(<16 x i16> %a, <16 x i16> %b) {
define <8 x i32> @xvabsd_wu(<8 x i32> %a, <8 x i32> %b) {
; CHECK-LABEL: xvabsd_wu:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.wu $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.wu $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.w $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.wu $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a.zext = zext <8 x i32> %a to <8 x i64>
%b.zext = zext <8 x i32> %b to <8 x i64>
@@ -121,9 +104,7 @@ define <8 x i32> @xvabsd_wu(<8 x i32> %a, <8 x i32> %b) {
define <4 x i64> @xvabsd_du(<4 x i64> %a, <4 x i64> %b) {
; CHECK-LABEL: xvabsd_du:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.du $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.du $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.d $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.du $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a.zext = zext <4 x i64> %a to <4 x i128>
%b.zext = zext <4 x i64> %b to <4 x i128>
@@ -137,9 +118,7 @@ define <4 x i64> @xvabsd_du(<4 x i64> %a, <4 x i64> %b) {
define <32 x i8> @xvabsd_b_nsw(<32 x i8> %a, <32 x i8> %b) {
; CHECK-LABEL: xvabsd_b_nsw:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvsub.b $xr0, $xr0, $xr1
-; CHECK-NEXT: xvneg.b $xr1, $xr0
-; CHECK-NEXT: xvmax.b $xr0, $xr0, $xr1
+; CHECK-NEXT: xvabsd.b $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%sub = sub nsw <32 x i8> %a, %b
%abs = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %sub, i1 true)
@@ -149,9 +128,7 @@ define <32 x i8> @xvabsd_b_nsw(<32 x i8> %a, <32 x i8> %b) {
define <16 x i16> @xvabsd_h_nsw(<16 x i16> %a, <16 x i16> %b) {
; CHECK-LABEL: xvabsd_h_nsw:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvsub.h $xr0, $xr0, $xr1
-; CHECK-NEXT: xvneg.h $xr1, $xr0
-; CHECK-NEXT: xvmax.h $xr0, $xr0, $xr1
+; CHECK-NEXT: xvabsd.h $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%sub = sub nsw <16 x i16> %a, %b
%abs = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %sub, i1 true)
@@ -161,9 +138,7 @@ define <16 x i16> @xvabsd_h_nsw(<16 x i16> %a, <16 x i16> %b) {
define <8 x i32> @xvabsd_w_nsw(<8 x i32> %a, <8 x i32> %b) {
; CHECK-LABEL: xvabsd_w_nsw:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvsub.w $xr0, $xr0, $xr1
-; CHECK-NEXT: xvneg.w $xr1, $xr0
-; CHECK-NEXT: xvmax.w $xr0, $xr0, $xr1
+; CHECK-NEXT: xvabsd.w $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%sub = sub nsw <8 x i32> %a, %b
%abs = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %sub, i1 true)
@@ -173,9 +148,7 @@ define <8 x i32> @xvabsd_w_nsw(<8 x i32> %a, <8 x i32> %b) {
define <4 x i64> @xvabsd_d_nsw(<4 x i64> %a, <4 x i64> %b) {
; CHECK-LABEL: xvabsd_d_nsw:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvsub.d $xr0, $xr0, $xr1
-; CHECK-NEXT: xvneg.d $xr1, $xr0
-; CHECK-NEXT: xvmax.d $xr0, $xr0, $xr1
+; CHECK-NEXT: xvabsd.d $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%sub = sub nsw <4 x i64> %a, %b
%abs = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %sub, i1 true)
@@ -186,9 +159,7 @@ define <4 x i64> @xvabsd_d_nsw(<4 x i64> %a, <4 x i64> %b) {
define <32 x i8> @maxmin_b(<32 x i8> %0, <32 x i8> %1) {
; CHECK-LABEL: maxmin_b:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.b $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.b $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.b $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.b $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a = tail call <32 x i8> @llvm.smax.v32i8(<32 x i8> %0, <32 x i8> %1)
%b = tail call <32 x i8> @llvm.smin.v32i8(<32 x i8> %0, <32 x i8> %1)
@@ -199,9 +170,7 @@ define <32 x i8> @maxmin_b(<32 x i8> %0, <32 x i8> %1) {
define <16 x i16> @maxmin_h(<16 x i16> %0, <16 x i16> %1) {
; CHECK-LABEL: maxmin_h:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.h $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.h $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.h $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.h $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a = tail call <16 x i16> @llvm.smax.v16i16(<16 x i16> %0, <16 x i16> %1)
%b = tail call <16 x i16> @llvm.smin.v16i16(<16 x i16> %0, <16 x i16> %1)
@@ -212,9 +181,7 @@ define <16 x i16> @maxmin_h(<16 x i16> %0, <16 x i16> %1) {
define <8 x i32> @maxmin_w(<8 x i32> %0, <8 x i32> %1) {
; CHECK-LABEL: maxmin_w:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.w $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.w $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.w $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.w $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a = tail call <8 x i32> @llvm.smax.v8i32(<8 x i32> %0, <8 x i32> %1)
%b = tail call <8 x i32> @llvm.smin.v8i32(<8 x i32> %0, <8 x i32> %1)
@@ -225,9 +192,7 @@ define <8 x i32> @maxmin_w(<8 x i32> %0, <8 x i32> %1) {
define <4 x i64> @maxmin_d(<4 x i64> %0, <4 x i64> %1) {
; CHECK-LABEL: maxmin_d:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.d $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.d $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.d $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.d $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a = tail call <4 x i64> @llvm.smax.v4i64(<4 x i64> %0, <4 x i64> %1)
%b = tail call <4 x i64> @llvm.smin.v4i64(<4 x i64> %0, <4 x i64> %1)
@@ -238,9 +203,7 @@ define <4 x i64> @maxmin_d(<4 x i64> %0, <4 x i64> %1) {
define <32 x i8> @maxmin_bu(<32 x i8> %0, <32 x i8> %1) {
; CHECK-LABEL: maxmin_bu:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.bu $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.bu $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.b $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.bu $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a = tail call <32 x i8> @llvm.umax.v32i8(<32 x i8> %0, <32 x i8> %1)
%b = tail call <32 x i8> @llvm.umin.v32i8(<32 x i8> %0, <32 x i8> %1)
@@ -251,9 +214,7 @@ define <32 x i8> @maxmin_bu(<32 x i8> %0, <32 x i8> %1) {
define <16 x i16> @maxmin_hu(<16 x i16> %0, <16 x i16> %1) {
; CHECK-LABEL: maxmin_hu:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.hu $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.hu $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.h $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.hu $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a = tail call <16 x i16> @llvm.umax.v16i16(<16 x i16> %0, <16 x i16> %1)
%b = tail call <16 x i16> @llvm.umin.v16i16(<16 x i16> %0, <16 x i16> %1)
@@ -264,9 +225,7 @@ define <16 x i16> @maxmin_hu(<16 x i16> %0, <16 x i16> %1) {
define <8 x i32> @maxmin_wu(<8 x i32> %0, <8 x i32> %1) {
; CHECK-LABEL: maxmin_wu:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.wu $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.wu $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.w $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.wu $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a = tail call <8 x i32> @llvm.umax.v8i32(<8 x i32> %0, <8 x i32> %1)
%b = tail call <8 x i32> @llvm.umin.v8i32(<8 x i32> %0, <8 x i32> %1)
@@ -277,9 +236,7 @@ define <8 x i32> @maxmin_wu(<8 x i32> %0, <8 x i32> %1) {
define <4 x i64> @maxmin_du(<4 x i64> %0, <4 x i64> %1) {
; CHECK-LABEL: maxmin_du:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.du $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.du $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.d $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.du $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a = tail call <4 x i64> @llvm.umax.v4i64(<4 x i64> %0, <4 x i64> %1)
%b = tail call <4 x i64> @llvm.umin.v4i64(<4 x i64> %0, <4 x i64> %1)
@@ -290,9 +247,7 @@ define <4 x i64> @maxmin_du(<4 x i64> %0, <4 x i64> %1) {
define <32 x i8> @maxmin_bu_com1(<32 x i8> %0, <32 x i8> %1) {
; CHECK-LABEL: maxmin_bu_com1:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.bu $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.bu $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.b $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.bu $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%a = tail call <32 x i8> @llvm.umax.v32i8(<32 x i8> %0, <32 x i8> %1)
%b = tail call <32 x i8> @llvm.umin.v32i8(<32 x i8> %1, <32 x i8> %0)
@@ -304,9 +259,7 @@ define <32 x i8> @maxmin_bu_com1(<32 x i8> %0, <32 x i8> %1) {
define <32 x i8> @xvabsd_b_cmp(<32 x i8> %a, <32 x i8> %b) nounwind {
; CHECK-LABEL: xvabsd_b_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.b $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.b $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.b $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.b $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp slt <32 x i8> %a, %b
%ab = sub <32 x i8> %a, %b
@@ -318,9 +271,7 @@ define <32 x i8> @xvabsd_b_cmp(<32 x i8> %a, <32 x i8> %b) nounwind {
define <16 x i16> @xvabsd_h_cmp(<16 x i16> %a, <16 x i16> %b) nounwind {
; CHECK-LABEL: xvabsd_h_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.h $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.h $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.h $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.h $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp slt <16 x i16> %a, %b
%ab = sub <16 x i16> %a, %b
@@ -332,9 +283,7 @@ define <16 x i16> @xvabsd_h_cmp(<16 x i16> %a, <16 x i16> %b) nounwind {
define <8 x i32> @xvabsd_w_cmp(<8 x i32> %a, <8 x i32> %b) nounwind {
; CHECK-LABEL: xvabsd_w_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.w $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.w $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.w $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.w $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp slt <8 x i32> %a, %b
%ab = sub <8 x i32> %a, %b
@@ -346,9 +295,7 @@ define <8 x i32> @xvabsd_w_cmp(<8 x i32> %a, <8 x i32> %b) nounwind {
define <4 x i64> @xvabsd_d_cmp(<4 x i64> %a, <4 x i64> %b) nounwind {
; CHECK-LABEL: xvabsd_d_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.d $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.d $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.d $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.d $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp slt <4 x i64> %a, %b
%ab = sub <4 x i64> %a, %b
@@ -360,9 +307,7 @@ define <4 x i64> @xvabsd_d_cmp(<4 x i64> %a, <4 x i64> %b) nounwind {
define <32 x i8> @xvabsd_bu_cmp(<32 x i8> %a, <32 x i8> %b) nounwind {
; CHECK-LABEL: xvabsd_bu_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.bu $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.bu $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.b $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.bu $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp ult <32 x i8> %a, %b
%ab = sub <32 x i8> %a, %b
@@ -374,9 +319,7 @@ define <32 x i8> @xvabsd_bu_cmp(<32 x i8> %a, <32 x i8> %b) nounwind {
define <16 x i16> @xvabsd_hu_cmp(<16 x i16> %a, <16 x i16> %b) nounwind {
; CHECK-LABEL: xvabsd_hu_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.hu $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.hu $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.h $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.hu $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp ult <16 x i16> %a, %b
%ab = sub <16 x i16> %a, %b
@@ -388,9 +331,7 @@ define <16 x i16> @xvabsd_hu_cmp(<16 x i16> %a, <16 x i16> %b) nounwind {
define <8 x i32> @xvabsd_wu_cmp(<8 x i32> %a, <8 x i32> %b) nounwind {
; CHECK-LABEL: xvabsd_wu_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.wu $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.wu $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.w $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.wu $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp ult <8 x i32> %a, %b
%ab = sub <8 x i32> %a, %b
@@ -402,9 +343,7 @@ define <8 x i32> @xvabsd_wu_cmp(<8 x i32> %a, <8 x i32> %b) nounwind {
define <4 x i64> @xvabsd_du_cmp(<4 x i64> %a, <4 x i64> %b) nounwind {
; CHECK-LABEL: xvabsd_du_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.du $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.du $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.d $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.du $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp ult <4 x i64> %a, %b
%ab = sub <4 x i64> %a, %b
@@ -417,9 +356,7 @@ define <4 x i64> @xvabsd_du_cmp(<4 x i64> %a, <4 x i64> %b) nounwind {
define <32 x i8> @xvabsd_b_select(<32 x i8> %a, <32 x i8> %b) nounwind {
; CHECK-LABEL: xvabsd_b_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.b $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.b $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.b $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.b $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp slt <32 x i8> %a, %b
%ab = select <32 x i1> %cmp, <32 x i8> %a, <32 x i8> %b
@@ -431,9 +368,7 @@ define <32 x i8> @xvabsd_b_select(<32 x i8> %a, <32 x i8> %b) nounwind {
define <16 x i16> @xvabsd_h_select(<16 x i16> %a, <16 x i16> %b) nounwind {
; CHECK-LABEL: xvabsd_h_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.h $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.h $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.h $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.h $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp sle <16 x i16> %a, %b
%ab = select <16 x i1> %cmp, <16 x i16> %a, <16 x i16> %b
@@ -445,9 +380,7 @@ define <16 x i16> @xvabsd_h_select(<16 x i16> %a, <16 x i16> %b) nounwind {
define <8 x i32> @xvabsd_w_select(<8 x i32> %a, <8 x i32> %b) nounwind {
; CHECK-LABEL: xvabsd_w_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.w $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.w $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.w $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.w $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp sgt <8 x i32> %a, %b
%ab = select <8 x i1> %cmp, <8 x i32> %a, <8 x i32> %b
@@ -459,9 +392,7 @@ define <8 x i32> @xvabsd_w_select(<8 x i32> %a, <8 x i32> %b) nounwind {
define <4 x i64> @xvabsd_d_select(<4 x i64> %a, <4 x i64> %b) nounwind {
; CHECK-LABEL: xvabsd_d_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.d $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.d $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.d $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.d $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp sge <4 x i64> %a, %b
%ab = select <4 x i1> %cmp, <4 x i64> %a, <4 x i64> %b
@@ -473,9 +404,7 @@ define <4 x i64> @xvabsd_d_select(<4 x i64> %a, <4 x i64> %b) nounwind {
define <32 x i8> @xvabsd_bu_select(<32 x i8> %a, <32 x i8> %b) nounwind {
; CHECK-LABEL: xvabsd_bu_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.bu $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.bu $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.b $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.bu $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp ult <32 x i8> %a, %b
%ab = select <32 x i1> %cmp, <32 x i8> %a, <32 x i8> %b
@@ -487,9 +416,7 @@ define <32 x i8> @xvabsd_bu_select(<32 x i8> %a, <32 x i8> %b) nounwind {
define <16 x i16> @xvabsd_hu_select(<16 x i16> %a, <16 x i16> %b) nounwind {
; CHECK-LABEL: xvabsd_hu_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.hu $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.hu $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.h $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.hu $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp ule <16 x i16> %a, %b
%ab = select <16 x i1> %cmp, <16 x i16> %a, <16 x i16> %b
@@ -501,9 +428,7 @@ define <16 x i16> @xvabsd_hu_select(<16 x i16> %a, <16 x i16> %b) nounwind {
define <8 x i32> @xvabsd_wu_select(<8 x i32> %a, <8 x i32> %b) nounwind {
; CHECK-LABEL: xvabsd_wu_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.wu $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.wu $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.w $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.wu $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp ugt <8 x i32> %a, %b
%ab = select <8 x i1> %cmp, <8 x i32> %a, <8 x i32> %b
@@ -515,9 +440,7 @@ define <8 x i32> @xvabsd_wu_select(<8 x i32> %a, <8 x i32> %b) nounwind {
define <4 x i64> @xvabsd_du_select(<4 x i64> %a, <4 x i64> %b) nounwind {
; CHECK-LABEL: xvabsd_du_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: xvmin.du $xr2, $xr0, $xr1
-; CHECK-NEXT: xvmax.du $xr0, $xr0, $xr1
-; CHECK-NEXT: xvsub.d $xr0, $xr0, $xr2
+; CHECK-NEXT: xvabsd.du $xr0, $xr0, $xr1
; CHECK-NEXT: ret
%cmp = icmp uge <4 x i64> %a, %b
%ab = select <4 x i1> %cmp, <4 x i64> %a, <4 x i64> %b
diff --git a/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/absd.ll b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/absd.ll
index 2cbd74204d5d6..f77a31b600761 100644
--- a/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/absd.ll
+++ b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/absd.ll
@@ -1,7 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=loongarch64 -mattr=+lsx < %s | FileCheck %s
-;; TODO: Currently LoongArch generates sub-optimal code for five cases
;; 1. trunc(abs(sub(sext(a),sext(b)))) -> abds(a,b) or abdu(a,b)
;; 2. abs(sub_nsw(x, y)) -> abds(a,b)
;; 3. sub(smax(a,b),smin(a,b)) -> abds(a,b) or abdu(a,b)
@@ -9,16 +8,12 @@
;; 5. sub(select(icmp(a,b),a,b),select(icmp(a,b),b,a)) -> abds(a,b) or abdu(a,b)
;;
;; abds / abdu can be lowered to vabsd.{b/h/w/d} / vabsd.{b/h/w/d}u instruction
-;;
-;; Later patch will address it.
;; trunc(abs(sub(sext(a),sext(b)))) -> abds(a,b)
define <16 x i8> @vabsd_b(<16 x i8> %a, <16 x i8> %b) {
; CHECK-LABEL: vabsd_b:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.b $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.b $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.b $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.b $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a.sext = sext <16 x i8> %a to <16 x i16>
%b.sext = sext <16 x i8> %b to <16 x i16>
@@ -31,9 +26,7 @@ define <16 x i8> @vabsd_b(<16 x i8> %a, <16 x i8> %b) {
define <8 x i16> @vabsd_h(<8 x i16> %a, <8 x i16> %b) {
; CHECK-LABEL: vabsd_h:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.h $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.h $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.h $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.h $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a.sext = sext <8 x i16> %a to <8 x i32>
%b.sext = sext <8 x i16> %b to <8 x i32>
@@ -46,9 +39,7 @@ define <8 x i16> @vabsd_h(<8 x i16> %a, <8 x i16> %b) {
define <4 x i32> @vabsd_w(<4 x i32> %a, <4 x i32> %b) {
; CHECK-LABEL: vabsd_w:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.w $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.w $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.w $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.w $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a.sext = sext <4 x i32> %a to <4 x i64>
%b.sext = sext <4 x i32> %b to <4 x i64>
@@ -61,9 +52,7 @@ define <4 x i32> @vabsd_w(<4 x i32> %a, <4 x i32> %b) {
define <2 x i64> @vabsd_d(<2 x i64> %a, <2 x i64> %b) {
; CHECK-LABEL: vabsd_d:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.d $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.d $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.d $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.d $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a.sext = sext <2 x i64> %a to <2 x i128>
%b.sext = sext <2 x i64> %b to <2 x i128>
@@ -76,9 +65,7 @@ define <2 x i64> @vabsd_d(<2 x i64> %a, <2 x i64> %b) {
define <16 x i8> @vabsd_bu(<16 x i8> %a, <16 x i8> %b) {
; CHECK-LABEL: vabsd_bu:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.bu $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.bu $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.b $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.bu $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a.zext = zext <16 x i8> %a to <16 x i16>
%b.zext = zext <16 x i8> %b to <16 x i16>
@@ -91,9 +78,7 @@ define <16 x i8> @vabsd_bu(<16 x i8> %a, <16 x i8> %b) {
define <8 x i16> @vabsd_hu(<8 x i16> %a, <8 x i16> %b) {
; CHECK-LABEL: vabsd_hu:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.hu $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.hu $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.h $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.hu $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a.zext = zext <8 x i16> %a to <8 x i32>
%b.zext = zext <8 x i16> %b to <8 x i32>
@@ -106,9 +91,7 @@ define <8 x i16> @vabsd_hu(<8 x i16> %a, <8 x i16> %b) {
define <4 x i32> @vabsd_wu(<4 x i32> %a, <4 x i32> %b) {
; CHECK-LABEL: vabsd_wu:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.wu $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.wu $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.w $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.wu $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a.zext = zext <4 x i32> %a to <4 x i64>
%b.zext = zext <4 x i32> %b to <4 x i64>
@@ -121,9 +104,7 @@ define <4 x i32> @vabsd_wu(<4 x i32> %a, <4 x i32> %b) {
define <2 x i64> @vabsd_du(<2 x i64> %a, <2 x i64> %b) {
; CHECK-LABEL: vabsd_du:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.du $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.du $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.d $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.du $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a.zext = zext <2 x i64> %a to <2 x i128>
%b.zext = zext <2 x i64> %b to <2 x i128>
@@ -137,9 +118,7 @@ define <2 x i64> @vabsd_du(<2 x i64> %a, <2 x i64> %b) {
define <16 x i8> @vabsd_b_nsw(<16 x i8> %a, <16 x i8> %b) {
; CHECK-LABEL: vabsd_b_nsw:
; CHECK: # %bb.0:
-; CHECK-NEXT: vsub.b $vr0, $vr0, $vr1
-; CHECK-NEXT: vneg.b $vr1, $vr0
-; CHECK-NEXT: vmax.b $vr0, $vr0, $vr1
+; CHECK-NEXT: vabsd.b $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%sub = sub nsw <16 x i8> %a, %b
%abs = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %sub, i1 true)
@@ -149,9 +128,7 @@ define <16 x i8> @vabsd_b_nsw(<16 x i8> %a, <16 x i8> %b) {
define <8 x i16> @vabsd_h_nsw(<8 x i16> %a, <8 x i16> %b) {
; CHECK-LABEL: vabsd_h_nsw:
; CHECK: # %bb.0:
-; CHECK-NEXT: vsub.h $vr0, $vr0, $vr1
-; CHECK-NEXT: vneg.h $vr1, $vr0
-; CHECK-NEXT: vmax.h $vr0, $vr0, $vr1
+; CHECK-NEXT: vabsd.h $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%sub = sub nsw <8 x i16> %a, %b
%abs = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %sub, i1 true)
@@ -161,9 +138,7 @@ define <8 x i16> @vabsd_h_nsw(<8 x i16> %a, <8 x i16> %b) {
define <4 x i32> @vabsd_w_nsw(<4 x i32> %a, <4 x i32> %b) {
; CHECK-LABEL: vabsd_w_nsw:
; CHECK: # %bb.0:
-; CHECK-NEXT: vsub.w $vr0, $vr0, $vr1
-; CHECK-NEXT: vneg.w $vr1, $vr0
-; CHECK-NEXT: vmax.w $vr0, $vr0, $vr1
+; CHECK-NEXT: vabsd.w $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%sub = sub nsw <4 x i32> %a, %b
%abs = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %sub, i1 true)
@@ -173,9 +148,7 @@ define <4 x i32> @vabsd_w_nsw(<4 x i32> %a, <4 x i32> %b) {
define <2 x i64> @vabsd_d_nsw(<2 x i64> %a, <2 x i64> %b) {
; CHECK-LABEL: vabsd_d_nsw:
; CHECK: # %bb.0:
-; CHECK-NEXT: vsub.d $vr0, $vr0, $vr1
-; CHECK-NEXT: vneg.d $vr1, $vr0
-; CHECK-NEXT: vmax.d $vr0, $vr0, $vr1
+; CHECK-NEXT: vabsd.d $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%sub = sub nsw <2 x i64> %a, %b
%abs = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %sub, i1 true)
@@ -186,9 +159,7 @@ define <2 x i64> @vabsd_d_nsw(<2 x i64> %a, <2 x i64> %b) {
define <16 x i8> @maxmin_b(<16 x i8> %0, <16 x i8> %1) {
; CHECK-LABEL: maxmin_b:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.b $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.b $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.b $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.b $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a = tail call <16 x i8> @llvm.smax.v16i8(<16 x i8> %0, <16 x i8> %1)
%b = tail call <16 x i8> @llvm.smin.v16i8(<16 x i8> %0, <16 x i8> %1)
@@ -199,9 +170,7 @@ define <16 x i8> @maxmin_b(<16 x i8> %0, <16 x i8> %1) {
define <8 x i16> @maxmin_h(<8 x i16> %0, <8 x i16> %1) {
; CHECK-LABEL: maxmin_h:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.h $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.h $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.h $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.h $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a = tail call <8 x i16> @llvm.smax.v8i16(<8 x i16> %0, <8 x i16> %1)
%b = tail call <8 x i16> @llvm.smin.v8i16(<8 x i16> %0, <8 x i16> %1)
@@ -212,9 +181,7 @@ define <8 x i16> @maxmin_h(<8 x i16> %0, <8 x i16> %1) {
define <4 x i32> @maxmin_w(<4 x i32> %0, <4 x i32> %1) {
; CHECK-LABEL: maxmin_w:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.w $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.w $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.w $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.w $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a = tail call <4 x i32> @llvm.smax.v4i32(<4 x i32> %0, <4 x i32> %1)
%b = tail call <4 x i32> @llvm.smin.v4i32(<4 x i32> %0, <4 x i32> %1)
@@ -225,9 +192,7 @@ define <4 x i32> @maxmin_w(<4 x i32> %0, <4 x i32> %1) {
define <2 x i64> @maxmin_d(<2 x i64> %0, <2 x i64> %1) {
; CHECK-LABEL: maxmin_d:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.d $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.d $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.d $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.d $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a = tail call <2 x i64> @llvm.smax.v2i64(<2 x i64> %0, <2 x i64> %1)
%b = tail call <2 x i64> @llvm.smin.v2i64(<2 x i64> %0, <2 x i64> %1)
@@ -238,9 +203,7 @@ define <2 x i64> @maxmin_d(<2 x i64> %0, <2 x i64> %1) {
define <16 x i8> @maxmin_bu(<16 x i8> %0, <16 x i8> %1) {
; CHECK-LABEL: maxmin_bu:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.bu $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.bu $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.b $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.bu $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a = tail call <16 x i8> @llvm.umax.v16i8(<16 x i8> %0, <16 x i8> %1)
%b = tail call <16 x i8> @llvm.umin.v16i8(<16 x i8> %0, <16 x i8> %1)
@@ -251,9 +214,7 @@ define <16 x i8> @maxmin_bu(<16 x i8> %0, <16 x i8> %1) {
define <8 x i16> @maxmin_hu(<8 x i16> %0, <8 x i16> %1) {
; CHECK-LABEL: maxmin_hu:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.hu $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.hu $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.h $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.hu $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a = tail call <8 x i16> @llvm.umax.v8i16(<8 x i16> %0, <8 x i16> %1)
%b = tail call <8 x i16> @llvm.umin.v8i16(<8 x i16> %0, <8 x i16> %1)
@@ -264,9 +225,7 @@ define <8 x i16> @maxmin_hu(<8 x i16> %0, <8 x i16> %1) {
define <4 x i32> @maxmin_wu(<4 x i32> %0, <4 x i32> %1) {
; CHECK-LABEL: maxmin_wu:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.wu $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.wu $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.w $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.wu $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a = tail call <4 x i32> @llvm.umax.v4i32(<4 x i32> %0, <4 x i32> %1)
%b = tail call <4 x i32> @llvm.umin.v4i32(<4 x i32> %0, <4 x i32> %1)
@@ -277,9 +236,7 @@ define <4 x i32> @maxmin_wu(<4 x i32> %0, <4 x i32> %1) {
define <2 x i64> @maxmin_du(<2 x i64> %0, <2 x i64> %1) {
; CHECK-LABEL: maxmin_du:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.du $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.du $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.d $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.du $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a = tail call <2 x i64> @llvm.umax.v2i64(<2 x i64> %0, <2 x i64> %1)
%b = tail call <2 x i64> @llvm.umin.v2i64(<2 x i64> %0, <2 x i64> %1)
@@ -290,9 +247,7 @@ define <2 x i64> @maxmin_du(<2 x i64> %0, <2 x i64> %1) {
define <16 x i8> @maxmin_bu_com1(<16 x i8> %0, <16 x i8> %1) {
; CHECK-LABEL: maxmin_bu_com1:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.bu $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.bu $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.b $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.bu $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%a = tail call <16 x i8> @llvm.umax.v16i8(<16 x i8> %0, <16 x i8> %1)
%b = tail call <16 x i8> @llvm.umin.v16i8(<16 x i8> %1, <16 x i8> %0)
@@ -304,9 +259,7 @@ define <16 x i8> @maxmin_bu_com1(<16 x i8> %0, <16 x i8> %1) {
define <16 x i8> @vabsd_b_cmp(<16 x i8> %a, <16 x i8> %b) nounwind {
; CHECK-LABEL: vabsd_b_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.b $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.b $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.b $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.b $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp slt <16 x i8> %a, %b
%ab = sub <16 x i8> %a, %b
@@ -318,9 +271,7 @@ define <16 x i8> @vabsd_b_cmp(<16 x i8> %a, <16 x i8> %b) nounwind {
define <8 x i16> @vabsd_h_cmp(<8 x i16> %a, <8 x i16> %b) nounwind {
; CHECK-LABEL: vabsd_h_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.h $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.h $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.h $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.h $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp slt <8 x i16> %a, %b
%ab = sub <8 x i16> %a, %b
@@ -332,9 +283,7 @@ define <8 x i16> @vabsd_h_cmp(<8 x i16> %a, <8 x i16> %b) nounwind {
define <4 x i32> @vabsd_w_cmp(<4 x i32> %a, <4 x i32> %b) nounwind {
; CHECK-LABEL: vabsd_w_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.w $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.w $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.w $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.w $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp slt <4 x i32> %a, %b
%ab = sub <4 x i32> %a, %b
@@ -346,9 +295,7 @@ define <4 x i32> @vabsd_w_cmp(<4 x i32> %a, <4 x i32> %b) nounwind {
define <2 x i64> @vabsd_d_cmp(<2 x i64> %a, <2 x i64> %b) nounwind {
; CHECK-LABEL: vabsd_d_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.d $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.d $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.d $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.d $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp slt <2 x i64> %a, %b
%ab = sub <2 x i64> %a, %b
@@ -360,9 +307,7 @@ define <2 x i64> @vabsd_d_cmp(<2 x i64> %a, <2 x i64> %b) nounwind {
define <16 x i8> @vabsd_bu_cmp(<16 x i8> %a, <16 x i8> %b) nounwind {
; CHECK-LABEL: vabsd_bu_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.bu $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.bu $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.b $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.bu $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp ult <16 x i8> %a, %b
%ab = sub <16 x i8> %a, %b
@@ -374,9 +319,7 @@ define <16 x i8> @vabsd_bu_cmp(<16 x i8> %a, <16 x i8> %b) nounwind {
define <8 x i16> @vabsd_hu_cmp(<8 x i16> %a, <8 x i16> %b) nounwind {
; CHECK-LABEL: vabsd_hu_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.hu $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.hu $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.h $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.hu $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp ult <8 x i16> %a, %b
%ab = sub <8 x i16> %a, %b
@@ -388,9 +331,7 @@ define <8 x i16> @vabsd_hu_cmp(<8 x i16> %a, <8 x i16> %b) nounwind {
define <4 x i32> @vabsd_wu_cmp(<4 x i32> %a, <4 x i32> %b) nounwind {
; CHECK-LABEL: vabsd_wu_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.wu $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.wu $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.w $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.wu $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp ult <4 x i32> %a, %b
%ab = sub <4 x i32> %a, %b
@@ -402,9 +343,7 @@ define <4 x i32> @vabsd_wu_cmp(<4 x i32> %a, <4 x i32> %b) nounwind {
define <2 x i64> @vabsd_du_cmp(<2 x i64> %a, <2 x i64> %b) nounwind {
; CHECK-LABEL: vabsd_du_cmp:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.du $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.du $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.d $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.du $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp ult <2 x i64> %a, %b
%ab = sub <2 x i64> %a, %b
@@ -417,9 +356,7 @@ define <2 x i64> @vabsd_du_cmp(<2 x i64> %a, <2 x i64> %b) nounwind {
define <16 x i8> @vabsd_b_select(<16 x i8> %a, <16 x i8> %b) nounwind {
; CHECK-LABEL: vabsd_b_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.b $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.b $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.b $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.b $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp slt <16 x i8> %a, %b
%ab = select <16 x i1> %cmp, <16 x i8> %a, <16 x i8> %b
@@ -431,9 +368,7 @@ define <16 x i8> @vabsd_b_select(<16 x i8> %a, <16 x i8> %b) nounwind {
define <8 x i16> @vabsd_h_select(<8 x i16> %a, <8 x i16> %b) nounwind {
; CHECK-LABEL: vabsd_h_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.h $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.h $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.h $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.h $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp sle <8 x i16> %a, %b
%ab = select <8 x i1> %cmp, <8 x i16> %a, <8 x i16> %b
@@ -445,9 +380,7 @@ define <8 x i16> @vabsd_h_select(<8 x i16> %a, <8 x i16> %b) nounwind {
define <4 x i32> @vabsd_w_select(<4 x i32> %a, <4 x i32> %b) nounwind {
; CHECK-LABEL: vabsd_w_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.w $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.w $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.w $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.w $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp sgt <4 x i32> %a, %b
%ab = select <4 x i1> %cmp, <4 x i32> %a, <4 x i32> %b
@@ -459,9 +392,7 @@ define <4 x i32> @vabsd_w_select(<4 x i32> %a, <4 x i32> %b) nounwind {
define <2 x i64> @vabsd_d_select(<2 x i64> %a, <2 x i64> %b) nounwind {
; CHECK-LABEL: vabsd_d_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.d $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.d $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.d $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.d $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp sge <2 x i64> %a, %b
%ab = select <2 x i1> %cmp, <2 x i64> %a, <2 x i64> %b
@@ -473,9 +404,7 @@ define <2 x i64> @vabsd_d_select(<2 x i64> %a, <2 x i64> %b) nounwind {
define <16 x i8> @vabsd_bu_select(<16 x i8> %a, <16 x i8> %b) nounwind {
; CHECK-LABEL: vabsd_bu_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.bu $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.bu $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.b $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.bu $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp ult <16 x i8> %a, %b
%ab = select <16 x i1> %cmp, <16 x i8> %a, <16 x i8> %b
@@ -487,9 +416,7 @@ define <16 x i8> @vabsd_bu_select(<16 x i8> %a, <16 x i8> %b) nounwind {
define <8 x i16> @vabsd_hu_select(<8 x i16> %a, <8 x i16> %b) nounwind {
; CHECK-LABEL: vabsd_hu_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.hu $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.hu $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.h $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.hu $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp ule <8 x i16> %a, %b
%ab = select <8 x i1> %cmp, <8 x i16> %a, <8 x i16> %b
@@ -501,9 +428,7 @@ define <8 x i16> @vabsd_hu_select(<8 x i16> %a, <8 x i16> %b) nounwind {
define <4 x i32> @vabsd_wu_select(<4 x i32> %a, <4 x i32> %b) nounwind {
; CHECK-LABEL: vabsd_wu_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.wu $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.wu $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.w $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.wu $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp ugt <4 x i32> %a, %b
%ab = select <4 x i1> %cmp, <4 x i32> %a, <4 x i32> %b
@@ -515,9 +440,7 @@ define <4 x i32> @vabsd_wu_select(<4 x i32> %a, <4 x i32> %b) nounwind {
define <2 x i64> @vabsd_du_select(<2 x i64> %a, <2 x i64> %b) nounwind {
; CHECK-LABEL: vabsd_du_select:
; CHECK: # %bb.0:
-; CHECK-NEXT: vmin.du $vr2, $vr0, $vr1
-; CHECK-NEXT: vmax.du $vr0, $vr0, $vr1
-; CHECK-NEXT: vsub.d $vr0, $vr0, $vr2
+; CHECK-NEXT: vabsd.du $vr0, $vr0, $vr1
; CHECK-NEXT: ret
%cmp = icmp uge <2 x i64> %a, %b
%ab = select <2 x i1> %cmp, <2 x i64> %a, <2 x i64> %b
More information about the llvm-commits
mailing list