From d1728f89878cf70dc87fbd407ee68d866e0c5ad9 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@intel.com>
Date: Mon, 1 Jul 2019 07:09:23 +0000
Subject: [PATCH] [X86] Add MOVHPDrm/MOVLPDrm patterns that use VZEXT_LOAD.

We already had patterns that used scalar_to_vector+load. But we can
also have a vzload.

Found while investigating combining scalar_to_vector+load to vzload.

llvm-svn: 364726
---
 llvm/lib/Target/X86/X86InstrAVX512.td                |  6 ++++++
 llvm/lib/Target/X86/X86InstrSSE.td                   | 12 ++++++++++++
 llvm/test/CodeGen/X86/merge-consecutive-loads-128.ll |  3 +--
 3 files changed, 19 insertions(+), 2 deletions(-)

diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index b334321..6452e63 100644
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -6349,6 +6349,12 @@ let Predicates = [HasAVX512] in {
   def : Pat<(v2f64 (X86Unpckl VR128X:$src1,
                     (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src2)))))),
            (VMOVHPDZ128rm VR128X:$src1, addr:$src2)>;
+  def : Pat<(v2f64 (X86Unpckl VR128X:$src1, (X86vzload addr:$src2))),
+            (VMOVHPDZ128rm VR128X:$src1, addr:$src2)>;
+
+  // VMOVLPD patterns
+  def : Pat<(v2f64 (X86Movsd VR128X:$src1, (X86vzload addr:$src2))),
+            (VMOVLPDZ128rm VR128X:$src1, addr:$src2)>;
 }
 
 let SchedRW = [WriteFStore] in {
diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td
index ebc284c..f1728d8 100644
--- a/llvm/lib/Target/X86/X86InstrSSE.td
+++ b/llvm/lib/Target/X86/X86InstrSSE.td
@@ -699,11 +699,17 @@ let Predicates = [UseAVX] in {
   def : Pat<(v2f64 (X86Unpckl VR128:$src1,
                       (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src2)))))),
             (VMOVHPDrm VR128:$src1, addr:$src2)>;
+  def : Pat<(v2f64 (X86Unpckl VR128:$src1, (X86vzload addr:$src2))),
+            (VMOVHPDrm VR128:$src1, addr:$src2)>;
 
   def : Pat<(store (f64 (extractelt
                           (v2f64 (X86VPermilpi VR128:$src, (i8 1))),
                           (iPTR 0))), addr:$dst),
             (VMOVHPDmr addr:$dst, VR128:$src)>;
+
+  // MOVLPD patterns
+  def : Pat<(v2f64 (X86Movsd VR128:$src1, (X86vzload addr:$src2))),
+            (VMOVLPDrm VR128:$src1, addr:$src2)>;
 }
 
 let Predicates = [UseSSE1] in {
@@ -722,11 +728,17 @@ let Predicates = [UseSSE2] in {
   def : Pat<(v2f64 (X86Unpckl VR128:$src1,
                       (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src2)))))),
             (MOVHPDrm VR128:$src1, addr:$src2)>;
+  def : Pat<(v2f64 (X86Unpckl VR128:$src1, (X86vzload addr:$src2))),
+            (MOVHPDrm VR128:$src1, addr:$src2)>;
 
   def : Pat<(store (f64 (extractelt
                           (v2f64 (X86Shufp VR128:$src, VR128:$src, (i8 1))),
                           (iPTR 0))), addr:$dst),
             (MOVHPDmr addr:$dst, VR128:$src)>;
+
+  // MOVLPD patterns
+  def : Pat<(v2f64 (X86Movsd VR128:$src1, (X86vzload addr:$src2))),
+            (MOVLPDrm VR128:$src1, addr:$src2)>;
 }
 
 //===----------------------------------------------------------------------===//
diff --git a/llvm/test/CodeGen/X86/merge-consecutive-loads-128.ll b/llvm/test/CodeGen/X86/merge-consecutive-loads-128.ll
index b7c8820..679b575 100644
--- a/llvm/test/CodeGen/X86/merge-consecutive-loads-128.ll
+++ b/llvm/test/CodeGen/X86/merge-consecutive-loads-128.ll
@@ -1173,8 +1173,7 @@ define <4 x float> @merge_4f32_f32_2345_volatile(float* %ptr) nounwind uwtable n
 ; SSE2-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; SSE2-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
 ; SSE2-NEXT:    unpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE2-NEXT:    movsd {{.*#+}} xmm1 = mem[0],zero
-; SSE2-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE2-NEXT:    movhpd {{.*#+}} xmm0 = xmm0[0],mem[0]
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: merge_4f32_f32_2345_volatile:
-- 
2.7.4