From 1919b650523282c550536b6b72eb4713cd6712f4 Mon Sep 17 00:00:00 2001
From: Sam Parker <sam.parker@arm.com>
Date: Wed, 9 Sep 2020 08:15:55 +0100
Subject: [PATCH] [ARM] Tail predicate VQDMULH and VQRDMULH

Mark the family of instructions as valid for tail predication.

Differential Revision: https://reviews.llvm.org/D87348
---
 llvm/lib/Target/ARM/ARMInstrMVE.td                     |  2 ++
 .../test/CodeGen/Thumb2/LowOverheadLoops/remat-vctp.ll | 18 +++---------------
 llvm/unittests/Target/ARM/MachineInstrTest.cpp         | 12 ++++++++++++
 3 files changed, 17 insertions(+), 15 deletions(-)

diff --git a/llvm/lib/Target/ARM/ARMInstrMVE.td b/llvm/lib/Target/ARM/ARMInstrMVE.td
index 2287ede..1d562c5 100644
--- a/llvm/lib/Target/ARM/ARMInstrMVE.td
+++ b/llvm/lib/Target/ARM/ARMInstrMVE.td
@@ -1918,6 +1918,7 @@ class MVE_VQxDMULH_Base<string iname, string suffix, bits<2> size, bit rounding,
   let Inst{12-8} = 0b01011;
   let Inst{4} = 0b0;
   let Inst{0} = 0b0;
+  let validForTailPredication = 1;
 }
 
 multiclass MVE_VQxDMULH_m<string iname, MVEVectorVTInfo VTI,
@@ -5453,6 +5454,7 @@ class MVE_VxxMUL_qr<string iname, string suffix,
   let Inst{12} = 0b0;
   let Inst{8} = 0b0;
   let Inst{5} = 0b1;
+  let validForTailPredication = 1;
 }
 
 multiclass MVE_VxxMUL_qr_m<string iname, MVEVectorVTInfo VTI, bit bit_28,
diff --git a/llvm/test/CodeGen/Thumb2/LowOverheadLoops/remat-vctp.ll b/llvm/test/CodeGen/Thumb2/LowOverheadLoops/remat-vctp.ll
index 6ce2b9f5..198ec16 100644
--- a/llvm/test/CodeGen/Thumb2/LowOverheadLoops/remat-vctp.ll
+++ b/llvm/test/CodeGen/Thumb2/LowOverheadLoops/remat-vctp.ll
@@ -7,23 +7,13 @@ define void @remat_vctp(i32* %arg, i32* %arg1, i32* %arg2, i32* %arg3, i32* %arg
 ; CHECK-NEXT:    push {r4, r5, r7, lr}
 ; CHECK-NEXT:    vpush {d8, d9, d10, d11, d12, d13, d14, d15}
 ; CHECK-NEXT:    ldrd r5, r12, [sp, #80]
-; CHECK-NEXT:    cmp.w r12, #4
-; CHECK-NEXT:    mov r4, r12
 ; CHECK-NEXT:    vmvn.i32 q0, #0x80000000
-; CHECK-NEXT:    it ge
-; CHECK-NEXT:    movge r4, #4
 ; CHECK-NEXT:    vmov.i32 q1, #0x3f
-; CHECK-NEXT:    sub.w r4, r12, r4
 ; CHECK-NEXT:    vmov.i32 q2, #0x1
-; CHECK-NEXT:    add.w lr, r4, #3
-; CHECK-NEXT:    movs r4, #1
-; CHECK-NEXT:    add.w lr, r4, lr, lsr #2
-; CHECK-NEXT:    dls lr, lr
+; CHECK-NEXT:    dlstp.32 lr, r12
 ; CHECK-NEXT:  .LBB0_1: @ %bb6
 ; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
-; CHECK-NEXT:    vctp.32 r12
-; CHECK-NEXT:    vpst
-; CHECK-NEXT:    vldrwt.u32 q4, [r1], #16
+; CHECK-NEXT:    vldrw.u32 q4, [r1], #16
 ; CHECK-NEXT:    vabs.s32 q5, q4
 ; CHECK-NEXT:    vcls.s32 q3, q5
 ; CHECK-NEXT:    vshl.u32 q5, q5, q3
@@ -41,15 +31,13 @@ define void @remat_vctp(i32* %arg, i32* %arg1, i32* %arg2, i32* %arg3, i32* %arg
 ; CHECK-NEXT:    vqshl.s32 q5, q5, #1
 ; CHECK-NEXT:    vpt.s32 lt, q4, zr
 ; CHECK-NEXT:    vnegt.s32 q5, q5
-; CHECK-NEXT:    vctp.32 r12
-; CHECK-NEXT:    sub.w r12, r12, #4
 ; CHECK-NEXT:    vpst
 ; CHECK-NEXT:    vldrwt.u32 q4, [r0], #16
 ; CHECK-NEXT:    vqrdmulh.s32 q4, q4, q5
 ; CHECK-NEXT:    vpstt
 ; CHECK-NEXT:    vstrwt.32 q4, [r2], #16
 ; CHECK-NEXT:    vstrwt.32 q3, [r3], #16
-; CHECK-NEXT:    le lr, .LBB0_1
+; CHECK-NEXT:    letp lr, .LBB0_1
 ; CHECK-NEXT:  @ %bb.2: @ %bb44
 ; CHECK-NEXT:    vpop {d8, d9, d10, d11, d12, d13, d14, d15}
 ; CHECK-NEXT:    pop {r4, r5, r7, pc}
diff --git a/llvm/unittests/Target/ARM/MachineInstrTest.cpp b/llvm/unittests/Target/ARM/MachineInstrTest.cpp
index 876e011..bc37f99 100644
--- a/llvm/unittests/Target/ARM/MachineInstrTest.cpp
+++ b/llvm/unittests/Target/ARM/MachineInstrTest.cpp
@@ -754,6 +754,12 @@ TEST(MachineInstrValidTailPredication, IsCorrect) {
     case MVE_VQADDu16:
     case MVE_VQADDu32:
     case MVE_VQADDu8:
+    case MVE_VQDMULH_qr_s16:
+    case MVE_VQDMULH_qr_s32:
+    case MVE_VQDMULH_qr_s8:
+    case MVE_VQDMULHi16:
+    case MVE_VQDMULHi32:
+    case MVE_VQDMULHi8:
     case MVE_VQDMULL_qr_s16bh:
     case MVE_VQDMULL_qr_s16th:
     case MVE_VQDMULL_qr_s32bh:
@@ -762,6 +768,12 @@ TEST(MachineInstrValidTailPredication, IsCorrect) {
     case MVE_VQDMULLs16th:
     case MVE_VQDMULLs32bh:
     case MVE_VQDMULLs32th:
+    case MVE_VQRDMULH_qr_s16:
+    case MVE_VQRDMULH_qr_s32:
+    case MVE_VQRDMULH_qr_s8:
+    case MVE_VQRDMULHi16:
+    case MVE_VQRDMULHi32:
+    case MVE_VQRDMULHi8:
     case MVE_VQNEGs16:
     case MVE_VQNEGs32:
     case MVE_VQNEGs8:
-- 
2.7.4