Add 256-bit isel for movsldup/movshdup

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@136051 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 62 insertions, 21 deletions

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 6e46b3f56d..b060e557a2 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -3222,32 +3222,31 @@ def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
 //===---------------------------------------------------------------------===//
 // Replicate Single FP - MOVSHDUP and MOVSLDUP
 //
-multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr> {
-def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
+                              ValueType vt, RegisterClass RC, PatFrag mem_frag,
+                              X86MemOperand x86memop> {
+def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
-                      [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
-def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
+                      [(set RC:$dst, (vt (OpNode RC:$src)))]>;
+def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
-                      [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
-}
-
-multiclass sse3_replicate_sfp_y<bits<8> op, SDNode OpNode,
-                                string OpcodeStr> {
-def rr : S3SI<op, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
-              !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
-def rm : S3SI<op, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
-              !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
+                      [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
 }
 
 let Predicates = [HasAVX] in {
-  // FIXME: Merge above classes when we have patterns for the ymm version
-  defm VMOVSHDUP  : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup">, VEX;
-  defm VMOVSLDUP  : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup">, VEX;
-  defm VMOVSHDUPY : sse3_replicate_sfp_y<0x16, X86Movshdup, "vmovshdup">, VEX;
-  defm VMOVSLDUPY : sse3_replicate_sfp_y<0x12, X86Movsldup, "vmovsldup">, VEX;
-}
-defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup">;
-defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup">;
+  defm VMOVSHDUP  : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
+                                       v4f32, VR128, memopv4f32, f128mem>, VEX;
+  defm VMOVSLDUP  : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
+                                       v4f32, VR128, memopv4f32, f128mem>, VEX;
+  defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
+                                       v8f32, VR256, memopv8f32, f256mem>, VEX;
+  defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
+                                       v8f32, VR256, memopv8f32, f256mem>, VEX;
+}
+defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
+                                   memopv4f32, f128mem>;
+defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
+                                   memopv4f32, f128mem>;
 
 let Predicates = [HasSSE3] in {
   def : Pat<(v4i32 (X86Movshdup VR128:$src)),
@@ -3269,6 +3268,14 @@ let Predicates = [HasAVX] in {
             (VMOVSLDUPrr VR128:$src)>;
   def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
             (VMOVSLDUPrm addr:$src)>;
+  def : Pat<(v8i32 (X86Movshdup VR256:$src)),
+            (VMOVSHDUPYrr VR256:$src)>;
+  def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
+            (VMOVSHDUPYrm addr:$src)>;
+  def : Pat<(v8i32 (X86Movsldup VR256:$src)),
+            (VMOVSLDUPYrr VR256:$src)>;
+  def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
+            (VMOVSLDUPYrm addr:$src)>;
 }
 
 //===---------------------------------------------------------------------===//
diff --git a/test/CodeGen/X86/avx-256-movdup.ll b/test/CodeGen/X86/avx-256-movdup.ll
new file mode 100644
index 0000000000..42d84def98
--- /dev/null
+++ b/test/CodeGen/X86/avx-256-movdup.ll
@@ -0,0 +1,34 @@
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7-avx -mattr=+avx | FileCheck %s
+
+; CHECK: vmovsldup
+define <8 x float> @movdupA(<8 x float> %src) nounwind uwtable readnone ssp {
+entry:
+  %shuffle.i = shufflevector <8 x float> %src, <8 x float> undef, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6>
+  ret <8 x float> %shuffle.i
+}
+
+; CHECK: vmovshdup
+define <8 x float> @movdupB(<8 x float> %src) nounwind uwtable readnone ssp {
+entry:
+  %shuffle.i = shufflevector <8 x float> %src, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7>
+  ret <8 x float> %shuffle.i
+}
+
+; CHECK: vmovsldup
+define <4 x i64> @movdupC(<4 x i64> %src) nounwind uwtable readnone ssp {
+entry:
+  %0 = bitcast <4 x i64> %src to <8 x float>
+  %shuffle.i = shufflevector <8 x float> %0, <8 x float> undef, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6>
+  %1 = bitcast <8 x float> %shuffle.i to <4 x i64>
+  ret <4 x i64> %1
+}
+
+; CHECK: vmovshdup
+define <4 x i64> @movdupD(<4 x i64> %src) nounwind uwtable readnone ssp {
+entry:
+  %0 = bitcast <4 x i64> %src to <8 x float>
+  %shuffle.i = shufflevector <8 x float> %0, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7>
+  %1 = bitcast <8 x float> %shuffle.i to <4 x i64>
+  ret <4 x i64> %1
+}
+