From 648e48d02eafd538719f5115ed3bf8832383112a Mon Sep 17 00:00:00 2001
From: Evan Cheng <evan.cheng@apple.com>
Date: Tue, 29 Nov 2011 22:48:34 +0000
Subject: [PATCH] Add another missing pattern. llvm-gcc likes f64 but clang
 likes i64 so it was generating poor code for some SSE builtins.

llvm-svn: 145448
---
 llvm/lib/Target/X86/X86InstrSSE.td      |  6 ++++++
 llvm/test/CodeGen/X86/vec_shuffle-38.ll | 21 ++++++++++++++++++++-
 2 files changed, 26 insertions(+), 1 deletion(-)

diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td
index 22a7fa478509..0dca0007d247 100644
--- a/llvm/lib/Target/X86/X86InstrSSE.td
+++ b/llvm/lib/Target/X86/X86InstrSSE.td
@@ -1150,6 +1150,9 @@ let Predicates = [HasAVX] in {
   def : Pat<(X86Movlhps VR128:$src1,
                  (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
             (VMOVHPSrm VR128:$src1, addr:$src2)>;
+  def : Pat<(X86Movlhps VR128:$src1,
+                 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
+            (VMOVHPSrm VR128:$src1, addr:$src2)>;
   def : Pat<(X86Movlhps VR128:$src1,
                  (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
             (VMOVHPSrm VR128:$src1, addr:$src2)>;
@@ -1183,6 +1186,9 @@ let Predicates = [HasSSE1] in {
   def : Pat<(X86Movlhps VR128:$src1,
                  (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
             (MOVHPSrm VR128:$src1, addr:$src2)>;
+  def : Pat<(X86Movlhps VR128:$src1,
+                 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
+            (MOVHPSrm VR128:$src1, addr:$src2)>;
   def : Pat<(X86Movlhps VR128:$src1,
                  (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
             (MOVHPSrm VR128:$src1, addr:$src2)>;
diff --git a/llvm/test/CodeGen/X86/vec_shuffle-38.ll b/llvm/test/CodeGen/X86/vec_shuffle-38.ll
index 66da013665f6..96ef883c4e1e 100644
--- a/llvm/test/CodeGen/X86/vec_shuffle-38.ll
+++ b/llvm/test/CodeGen/X86/vec_shuffle-38.ll
@@ -46,7 +46,7 @@ entry:
 
 ; rdar://10119696
 ; CHECK: f
-define <4 x float> @f(<4 x float> %x, double* nocapture %y) nounwind uwtable readonly ssp {
+define <4 x float> @f(<4 x float> %x, double* nocapture %y) nounwind readonly ssp {
 entry:
   ; CHECK: movlps  (%{{rdi|rdx}}), %xmm0
   %u110.i = load double* %y, align 1
@@ -56,3 +56,22 @@ entry:
   ret <4 x float> %shuffle.i
 }
 
+define <4 x float> @loadhpi2(%struct.Float2* nocapture %vHiCoefPtr_0, %struct.Float2* nocapture %vLoCoefPtr_0, i32 %s) nounwind readonly ssp {
+entry:
+; CHECK: loadhpi2
+; CHECK: movhps (
+; CHECK-NOT: movlhps
+  %0 = bitcast %struct.Float2* %vHiCoefPtr_0 to <1 x i64>*
+  %idx.ext = sext i32 %s to i64
+  %add.ptr = getelementptr inbounds <1 x i64>* %0, i64 %idx.ext
+  %add.ptr.val = load <1 x i64>* %add.ptr, align 1
+  %1 = bitcast <1 x i64> %add.ptr.val to <2 x float>
+  %shuffle.i = shufflevector <2 x float> %1, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
+  %2 = bitcast %struct.Float2* %vLoCoefPtr_0 to <1 x i64>*
+  %add.ptr2 = getelementptr inbounds <1 x i64>* %2, i64 %idx.ext
+  %add.ptr2.val = load <1 x i64>* %add.ptr2, align 1
+  %3 = bitcast <1 x i64> %add.ptr2.val to <2 x float>
+  %shuffle.i4 = shufflevector <2 x float> %3, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
+  %shuffle1.i5 = shufflevector <4 x float> %shuffle.i, <4 x float> %shuffle.i4, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
+  ret <4 x float> %shuffle1.i5
+}