[X86][AVX] Move VPERM2F128 defs above VINSERTF128 defs. NFC.

This will be necessary for a future patch to lower VINSERTF128 custom folds to VPERM2F128

Author: RKSimon

llvm/lib/Target/X86/X86InstrSSE.td

@@ -7048,6 +7048,50 @@ def : Pat<(v32i8 (X86SubVBroadcastld128 addr:$src)),
           (VBROADCASTF128 addr:$src)>;
 }
 
+//===----------------------------------------------------------------------===//
+// VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
+//
+
+let ExeDomain = SSEPackedSingle in {
+let isCommutable = 1 in
+def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
+          (ins VR256:$src1, VR256:$src2, u8imm:$src3),
+          "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}", []>,
+          VEX_4V, VEX_L, Sched<[WriteFShuffle256]>;
+def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
+          (ins VR256:$src1, f256mem:$src2, u8imm:$src3),
+          "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}", []>,
+          VEX_4V, VEX_L, Sched<[WriteFShuffle256.Folded, WriteFShuffle256.ReadAfterFold]>;
+}
+
+// Immediate transform to help with commuting.
+def Perm2XCommuteImm : SDNodeXForm<timm, [{
+  return getI8Imm(N->getZExtValue() ^ 0x22, SDLoc(N));
+}]>;
+
+multiclass vperm2x128_lowering<string InstrStr, ValueType VT, PatFrag memop_frag> {
+  def : Pat<(VT (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 timm:$imm))),
+            (!cast<Instruction>(InstrStr#rr) VR256:$src1, VR256:$src2, timm:$imm)>;
+  def : Pat<(VT (X86VPerm2x128 VR256:$src1, (memop_frag addr:$src2), (i8 timm:$imm))),
+            (!cast<Instruction>(InstrStr#rm) VR256:$src1, addr:$src2, timm:$imm)>;
+  // Pattern with load in other operand.
+  def : Pat<(VT (X86VPerm2x128 (memop_frag addr:$src2), VR256:$src1, (i8 timm:$imm))),
+            (!cast<Instruction>(InstrStr#rm) VR256:$src1, addr:$src2,
+                                             (Perm2XCommuteImm timm:$imm))>;
+}
+
+let Predicates = [HasAVX] in {
+  defm : vperm2x128_lowering<"VPERM2F128", v4f64, loadv4f64>;
+  defm : vperm2x128_lowering<"VPERM2F128", v8f32, loadv8f32>;
+}
+
+let Predicates = [HasAVX1Only] in {
+  defm : vperm2x128_lowering<"VPERM2F128", v4i64,  loadv4i64>;
+  defm : vperm2x128_lowering<"VPERM2F128", v8i32,  loadv8i32>;
+  defm : vperm2x128_lowering<"VPERM2F128", v16i16, loadv16i16>;
+  defm : vperm2x128_lowering<"VPERM2F128", v32i8,  loadv32i8>;
+}
+
 //===----------------------------------------------------------------------===//
 // VINSERTF128 - Insert packed floating-point values
 //
@@ -7296,50 +7340,6 @@ let ExeDomain = SSEPackedDouble in {
                                SchedWriteFVarShuffle.YMM>, VEX_L;
 }
 
-//===----------------------------------------------------------------------===//
-// VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
-//
-
-let ExeDomain = SSEPackedSingle in {
-let isCommutable = 1 in
-def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
-          (ins VR256:$src1, VR256:$src2, u8imm:$src3),
-          "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}", []>,
-          VEX_4V, VEX_L, Sched<[WriteFShuffle256]>;
-def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
-          (ins VR256:$src1, f256mem:$src2, u8imm:$src3),
-          "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}", []>,
-          VEX_4V, VEX_L, Sched<[WriteFShuffle256.Folded, WriteFShuffle256.ReadAfterFold]>;
-}
-
-// Immediate transform to help with commuting.
-def Perm2XCommuteImm : SDNodeXForm<timm, [{
-  return getI8Imm(N->getZExtValue() ^ 0x22, SDLoc(N));
-}]>;
-
-multiclass vperm2x128_lowering<string InstrStr, ValueType VT, PatFrag memop_frag> {
-  def : Pat<(VT (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 timm:$imm))),
-            (!cast<Instruction>(InstrStr#rr) VR256:$src1, VR256:$src2, timm:$imm)>;
-  def : Pat<(VT (X86VPerm2x128 VR256:$src1, (memop_frag addr:$src2), (i8 timm:$imm))),
-            (!cast<Instruction>(InstrStr#rm) VR256:$src1, addr:$src2, timm:$imm)>;
-  // Pattern with load in other operand.
-  def : Pat<(VT (X86VPerm2x128 (memop_frag addr:$src2), VR256:$src1, (i8 timm:$imm))),
-            (!cast<Instruction>(InstrStr#rm) VR256:$src1, addr:$src2,
-                                             (Perm2XCommuteImm timm:$imm))>;
-}
-
-let Predicates = [HasAVX] in {
-  defm : vperm2x128_lowering<"VPERM2F128", v4f64, loadv4f64>;
-  defm : vperm2x128_lowering<"VPERM2F128", v8f32, loadv8f32>;
-}
-
-let Predicates = [HasAVX1Only] in {
-  defm : vperm2x128_lowering<"VPERM2F128", v4i64,  loadv4i64>;
-  defm : vperm2x128_lowering<"VPERM2F128", v8i32,  loadv8i32>;
-  defm : vperm2x128_lowering<"VPERM2F128", v16i16, loadv16i16>;
-  defm : vperm2x128_lowering<"VPERM2F128", v32i8,  loadv32i8>;
-}
-
 //===----------------------------------------------------------------------===//
 // VZERO - Zero YMM registers
 // Note: These instruction do not affect the YMM16-YMM31.