[X86][SSE] Add SSE_UNPCK/SSE_PUNPCK OpndItins

Update multi-classes to take the scheduling OpndItins instead of hard coding it.

Will be reused in the AVX512 equivalents.

llvm-svn: 319245

1 files changed, 61 insertions, 49 deletions

diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td
index cb7e1974d95..01770af0090 100644
--- a/llvm/lib/Target/X86/X86InstrSSE.td
+++ b/llvm/lib/Target/X86/X86InstrSSE.td
@@ -2432,66 +2432,72 @@ let Constraints = "$src1 = $dst" in {
 // SSE 1 & 2 - Unpack FP Instructions
 //===----------------------------------------------------------------------===//
 
+let Sched = WriteFShuffle in
+def SSE_UNPCK : OpndItins<
+  IIC_SSE_UNPCK, IIC_SSE_UNPCK
+>;
+
 /// sse12_unpack_interleave - sse 1 & 2 fp unpack and interleave
 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
                                    PatFrag mem_frag, RegisterClass RC,
                                    X86MemOperand x86memop, string asm,
-                                   Domain d, bit IsCommutable = 0> {
+                                   OpndItins itins, Domain d, bit IsCommutable = 0> {
     let isCommutable = IsCommutable in
     def rr : PI<opc, MRMSrcReg,
                 (outs RC:$dst), (ins RC:$src1, RC:$src2),
                 asm, [(set RC:$dst,
                            (vt (OpNode RC:$src1, RC:$src2)))],
-                           IIC_SSE_UNPCK, d>, Sched<[WriteFShuffle]>;
+                           itins.rr, d>, Sched<[itins.Sched]>;
     def rm : PI<opc, MRMSrcMem,
                 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
                 asm, [(set RC:$dst,
                            (vt (OpNode RC:$src1,
                                        (mem_frag addr:$src2))))],
-                                       IIC_SSE_UNPCK, d>,
-             Sched<[WriteFShuffleLd, ReadAfterLd]>;
+                                       itins.rm, d>,
+             Sched<[itins.Sched.Folded, ReadAfterLd]>;
 }
 
 let Predicates = [HasAVX, NoVLX] in {
 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, loadv4f32,
       VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
-                     SSEPackedSingle>, PS, VEX_4V, VEX_WIG;
+                     SSE_UNPCK, SSEPackedSingle>, PS, VEX_4V, VEX_WIG;
 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, loadv2f64,
       VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
-                     SSEPackedDouble>, PD, VEX_4V, VEX_WIG;
+                     SSE_UNPCK, SSEPackedDouble>, PD, VEX_4V, VEX_WIG;
 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, loadv4f32,
       VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
-                     SSEPackedSingle>, PS, VEX_4V, VEX_WIG;
+                     SSE_UNPCK, SSEPackedSingle>, PS, VEX_4V, VEX_WIG;
 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, loadv2f64,
       VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
-                     SSEPackedDouble>, PD, VEX_4V, VEX_WIG;
+                     SSE_UNPCK, SSEPackedDouble>, PD, VEX_4V, VEX_WIG;
 
 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, loadv8f32,
       VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
-                     SSEPackedSingle>, PS, VEX_4V, VEX_L, VEX_WIG;
+                     SSE_UNPCK, SSEPackedSingle>, PS, VEX_4V, VEX_L, VEX_WIG;
 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, loadv4f64,
       VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
-                     SSEPackedDouble>, PD, VEX_4V, VEX_L, VEX_WIG;
+                     SSE_UNPCK, SSEPackedDouble>, PD, VEX_4V, VEX_L, VEX_WIG;
 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, loadv8f32,
       VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
-                     SSEPackedSingle>, PS, VEX_4V, VEX_L, VEX_WIG;
+                     SSE_UNPCK, SSEPackedSingle>, PS, VEX_4V, VEX_L, VEX_WIG;
 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, loadv4f64,
       VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
-                     SSEPackedDouble>, PD, VEX_4V, VEX_L, VEX_WIG;
+                     SSE_UNPCK, SSEPackedDouble>, PD, VEX_4V, VEX_L, VEX_WIG;
 }// Predicates = [HasAVX, NoVLX]
+
 let Constraints = "$src1 = $dst" in {
   defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
         VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
-                       SSEPackedSingle>, PS;
+                       SSE_UNPCK, SSEPackedSingle>, PS;
   defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
         VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
-                       SSEPackedDouble, 1>, PD;
+                       SSE_UNPCK, SSEPackedDouble, 1>, PD;
   defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
         VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
-                       SSEPackedSingle>, PS;
+                       SSE_UNPCK, SSEPackedSingle>, PS;
   defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
         VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
-                       SSEPackedDouble>, PD;
+                       SSE_UNPCK, SSEPackedDouble>, PD;
 } // Constraints = "$src1 = $dst"
 
 let Predicates = [HasAVX1Only] in {
@@ -4062,16 +4068,22 @@ let Constraints = "$src1 = $dst" in {
 // SSE2 - Packed Integer Unpack Instructions
 //===---------------------------------------------------------------------===//
 
+let Sched = WriteShuffle in
+def SSE_PUNPCK : OpndItins<
+  IIC_SSE_UNPCK, IIC_SSE_UNPCK
+>;
+
 let ExeDomain = SSEPackedInt in {
 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
-                       SDNode OpNode, PatFrag ld_frag, bit Is2Addr = 1> {
+                       SDNode OpNode, OpndItins itins, PatFrag ld_frag,
+                       bit Is2Addr = 1> {
   def rr : PDI<opc, MRMSrcReg,
       (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
       !if(Is2Addr,
           !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
           !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
       [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
-      IIC_SSE_UNPCK>, Sched<[WriteShuffle]>;
+      itins.rr>, Sched<[itins.Sched]>;
   def rm : PDI<opc, MRMSrcMem,
       (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
       !if(Is2Addr,
@@ -4079,85 +4091,85 @@ multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
           !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
       [(set VR128:$dst, (vt (OpNode VR128:$src1,
                                   (bitconvert (ld_frag addr:$src2)))))],
-                                               IIC_SSE_UNPCK>,
-      Sched<[WriteShuffleLd, ReadAfterLd]>;
+                                               itins.rm>,
+      Sched<[itins.Sched.Folded, ReadAfterLd]>;
 }
 
 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
-                         SDNode OpNode> {
+                         SDNode OpNode, OpndItins itins> {
   def Yrr : PDI<opc, MRMSrcReg,
       (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
       !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-      [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>,
-      Sched<[WriteShuffle]>;
+      [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))],
+      itins.rr>, Sched<[itins.Sched]>;
   def Yrm : PDI<opc, MRMSrcMem,
       (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
       !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
       [(set VR256:$dst, (vt (OpNode VR256:$src1,
-                                  (bitconvert (loadv4i64 addr:$src2)))))]>,
-      Sched<[WriteShuffleLd, ReadAfterLd]>;
+                                  (bitconvert (loadv4i64 addr:$src2)))))],
+      itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
 }
 
 
 let Predicates = [HasAVX, NoVLX_Or_NoBWI] in {
   defm VPUNPCKLBW  : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
-                                 loadv2i64, 0>, VEX_4V, VEX_WIG;
+                                 SSE_PUNPCK, loadv2i64, 0>, VEX_4V, VEX_WIG;
   defm VPUNPCKLWD  : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
-                                 loadv2i64, 0>, VEX_4V, VEX_WIG;
+                                 SSE_PUNPCK, loadv2i64, 0>, VEX_4V, VEX_WIG;
   defm VPUNPCKHBW  : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
-                                 loadv2i64, 0>, VEX_4V, VEX_WIG;
+                                 SSE_PUNPCK, loadv2i64, 0>, VEX_4V, VEX_WIG;
   defm VPUNPCKHWD  : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
-                                 loadv2i64, 0>, VEX_4V, VEX_WIG;
+                                 SSE_PUNPCK, loadv2i64, 0>, VEX_4V, VEX_WIG;
 }
 let Predicates = [HasAVX, NoVLX] in {
   defm VPUNPCKLDQ  : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
-                                 loadv2i64, 0>, VEX_4V, VEX_WIG;
+                                 SSE_PUNPCK, loadv2i64, 0>, VEX_4V, VEX_WIG;
   defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
-                                 loadv2i64, 0>, VEX_4V, VEX_WIG;
+                                 SSE_PUNPCK, loadv2i64, 0>, VEX_4V, VEX_WIG;
   defm VPUNPCKHDQ  : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
-                                 loadv2i64, 0>, VEX_4V, VEX_WIG;
+                                 SSE_PUNPCK, loadv2i64, 0>, VEX_4V, VEX_WIG;
   defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
-                                 loadv2i64, 0>, VEX_4V, VEX_WIG;
+                                 SSE_PUNPCK, loadv2i64, 0>, VEX_4V, VEX_WIG;
 }
 
 let Predicates = [HasAVX2, NoVLX_Or_NoBWI] in {
-  defm VPUNPCKLBW  : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl>,
+  defm VPUNPCKLBW  : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl, SSE_PUNPCK>,
                                    VEX_4V, VEX_L, VEX_WIG;
-  defm VPUNPCKLWD  : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl>,
+  defm VPUNPCKLWD  : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl, SSE_PUNPCK>,
                                    VEX_4V, VEX_L, VEX_WIG;
-  defm VPUNPCKHBW  : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh>,
+  defm VPUNPCKHBW  : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh, SSE_PUNPCK>,
                                    VEX_4V, VEX_L, VEX_WIG;
-  defm VPUNPCKHWD  : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh>,
+  defm VPUNPCKHWD  : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh, SSE_PUNPCK>,
                                    VEX_4V, VEX_L, VEX_WIG;
 }
 let Predicates = [HasAVX2, NoVLX] in {
-  defm VPUNPCKLDQ  : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl>,
+  defm VPUNPCKLDQ  : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl, SSE_PUNPCK>,
                                    VEX_4V, VEX_L, VEX_WIG;
-  defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl>,
+  defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl, SSE_PUNPCK>,
                                    VEX_4V, VEX_L, VEX_WIG;
-  defm VPUNPCKHDQ  : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh>,
+  defm VPUNPCKHDQ  : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh, SSE_PUNPCK>,
                                    VEX_4V, VEX_L, VEX_WIG;
-  defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh>,
+  defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh, SSE_PUNPCK>,
                                    VEX_4V, VEX_L, VEX_WIG;
 }
 
 let Constraints = "$src1 = $dst" in {
-  defm PUNPCKLBW  : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
+  defm PUNPCKLBW  : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl, SSE_PUNPCK,
                                 memopv2i64>;
-  defm PUNPCKLWD  : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
+  defm PUNPCKLWD  : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl, SSE_PUNPCK,
                                 memopv2i64>;
-  defm PUNPCKLDQ  : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
+  defm PUNPCKLDQ  : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl, SSE_PUNPCK,
                                 memopv2i64>;
-  defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
+  defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl, SSE_PUNPCK,
                                 memopv2i64>;
 
-  defm PUNPCKHBW  : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
+  defm PUNPCKHBW  : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh, SSE_PUNPCK,
                                 memopv2i64>;
-  defm PUNPCKHWD  : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
+  defm PUNPCKHWD  : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh, SSE_PUNPCK,
                                 memopv2i64>;
-  defm PUNPCKHDQ  : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
+  defm PUNPCKHDQ  : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh, SSE_PUNPCK,
                                 memopv2i64>;
-  defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
+  defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh, SSE_PUNPCK,
                                 memopv2i64>;
 }
 } // ExeDomain = SSEPackedInt