AVX512: Implemented encoding and intrinsics for VMOVSHDUP/VMOVSLDUP instructions.

[oota-llvm.git] / lib / Target / X86 / X86InstrSSE.td

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index f390d8fd34f0ee131856b93fab4a0e911cb0c529..f39fa3f617aa4ed492f41bc12cbdb0d4a8cf5c0b 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -1172,12 +1172,13 @@ multiclass sse12_mov_hilo_packed_base<bits<8>opc, SDNode psnode, SDNode pdnode,
 
 multiclass sse12_mov_hilo_packed<bits<8>opc, SDNode psnode, SDNode pdnode,
                                  string base_opc, InstrItinClass itin> {
-  defm V#NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
+  let Predicates = [UseAVX] in
+    defm V#NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
                                     "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                                     itin>, VEX_4V;
 
-let Constraints = "$src1 = $dst" in
-  defm NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
+  let Constraints = "$src1 = $dst" in
+    defm NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
                                     "\t{$src2, $dst|$dst, $src2}",
                                     itin>;
 }
@@ -1188,6 +1189,7 @@ let AddedComplexity = 20 in {
 }
 
 let SchedRW = [WriteStore] in {
+let Predicates = [UseAVX] in {
 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                    "movlps\t{$src, $dst|$dst, $src}",
                    [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
@@ -1198,6 +1200,7 @@ def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                    [(store (f64 (vector_extract (v2f64 VR128:$src),
                                  (iPTR 0))), addr:$dst)],
                                  IIC_SSE_MOV_LH>, VEX;
+}// UseAVX
 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                    "movlps\t{$src, $dst|$dst, $src}",
                    [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
@@ -1210,7 +1213,7 @@ def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                                  IIC_SSE_MOV_LH>;
 } // SchedRW
 
-let Predicates = [HasAVX] in {
+let Predicates = [UseAVX] in {
   // Shuffle with VMOVLPS
   def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
             (VMOVLPSrm VR128:$src1, addr:$src2)>;
@@ -1297,6 +1300,7 @@ let AddedComplexity = 20 in {
 let SchedRW = [WriteStore] in {
 // v2f64 extract element 1 is always custom lowered to unpack high to low
 // and extract element 0 so the non-store version isn't too horrible.
+let Predicates = [UseAVX] in {
 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                    "movhps\t{$src, $dst|$dst, $src}",
                    [(store (f64 (vector_extract
@@ -1308,6 +1312,7 @@ def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                    [(store (f64 (vector_extract
                                  (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
                                  (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
+} // UseAVX
 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                    "movhps\t{$src, $dst|$dst, $src}",
                    [(store (f64 (vector_extract
@@ -1321,7 +1326,7 @@ def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                                  (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
 } // SchedRW
 
-let Predicates = [HasAVX] in {
+let Predicates = [UseAVX] in {
   // VMOVHPS patterns
   def : Pat<(X86Movlhps VR128:$src1,
                  (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
@@ -4611,7 +4616,7 @@ multiclass sse2_pinsrw<bit Is2Addr = 1> {
 }
 
 // Extract
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoBWI] in
 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
                     (outs GR32orGR64:$dst), (ins VR128:$src1, u8imm:$src2),
                     "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
@@ -4626,7 +4631,7 @@ def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
                Sched<[WriteShuffleLd, ReadAfterLd]>;
 
 // Insert
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoBWI] in
 defm VPINSRW : sse2_pinsrw<0>, PD, VEX_4V;
 
 let Predicates = [UseSSE2], Constraints = "$src1 = $dst" in
@@ -5130,7 +5135,7 @@ def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
                       IIC_SSE_MOV_LH>, Sched<[WriteLoad]>;
 }
 
-let Predicates = [HasAVX] in {
+let Predicates = [HasAVX, NoVLX] in {
   defm VMOVSHDUP  : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
                                        v4f32, VR128, loadv4f32, f128mem>, VEX;
   defm VMOVSLDUP  : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
@@ -5145,7 +5150,7 @@ defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
                                    memopv4f32, f128mem>;
 
-let Predicates = [HasAVX] in {
+let Predicates = [HasAVX, NoVLX] in {
   def : Pat<(v4i32 (X86Movshdup VR128:$src)),
             (VMOVSHDUPrr VR128:$src)>;
   def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (loadv2i64 addr:$src)))),
@@ -6156,7 +6161,7 @@ multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
                                                  imm:$src2)))), addr:$dst)]>;
 }
 
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoBWI] in
   defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
 
 defm PEXTRB      : SS41I_extract8<0x14, "pextrb">;
@@ -6181,7 +6186,7 @@ multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
                                                   imm:$src2)))), addr:$dst)]>;
 }
 
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoBWI] in
   defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
 
 defm PEXTRW      : SS41I_extract16<0x15, "pextrw">;
@@ -6205,7 +6210,7 @@ multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
                           addr:$dst)]>;
 }
 
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoDQI] in
   defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
 
 defm PEXTRD      : SS41I_extract32<0x16, "pextrd">;
@@ -6228,7 +6233,7 @@ multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
                           addr:$dst)]>, REX_W;
 }
 
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoDQI] in
   defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
 
 defm PEXTRQ      : SS41I_extract64<0x16, "pextrq">;
@@ -6296,7 +6301,7 @@ multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
                    imm:$src3))]>, Sched<[WriteShuffleLd, ReadAfterLd]>;
 }
 
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoBWI] in
   defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
 let Constraints = "$src1 = $dst" in
   defm PINSRB  : SS41I_insert8<0x20, "pinsrb">;
@@ -6322,7 +6327,7 @@ multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
                           imm:$src3)))]>, Sched<[WriteShuffleLd, ReadAfterLd]>;
 }
 
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoDQI] in
   defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
 let Constraints = "$src1 = $dst" in
   defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
@@ -6348,7 +6353,7 @@ multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
                           imm:$src3)))]>, Sched<[WriteShuffleLd, ReadAfterLd]>;
 }
 
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoDQI] in
   defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
 let Constraints = "$src1 = $dst" in
   defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
@@ -8318,29 +8323,45 @@ defm VPBLENDDY : AVX2_binop_rmi<0x02, "vpblendd", X86Blendi, v8i32,
 //
 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
                           X86MemOperand x86memop, PatFrag ld_frag,
-                          ValueType OpVT128, ValueType OpVT256> {
-  def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+                          ValueType OpVT128, ValueType OpVT256, Predicate prd> {
+  let Predicates = [HasAVX2, prd] in {
+    def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                   !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
-                  [(set VR128:$dst, (OpVT128 (X86VBroadcast (OpVT128 VR128:$src))))]>,
+                  [(set VR128:$dst,
+                   (OpVT128 (X86VBroadcast (OpVT128 VR128:$src))))]>,
                   Sched<[WriteShuffle]>, VEX;
-  def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
+    def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
                   !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
-                  [(set VR128:$dst, (OpVT128 (X86VBroadcast (ld_frag addr:$src))))]>,
+                  [(set VR128:$dst,
+                   (OpVT128 (X86VBroadcast (ld_frag addr:$src))))]>,
                   Sched<[WriteLoad]>, VEX;
-  def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
+    def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
                    !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
-                   [(set VR256:$dst, (OpVT256 (X86VBroadcast (OpVT128 VR128:$src))))]>,
+                   [(set VR256:$dst,
+                    (OpVT256 (X86VBroadcast (OpVT128 VR128:$src))))]>,
                    Sched<[WriteShuffle256]>, VEX, VEX_L;
-  def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
+    def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
                    !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
-                   [(set VR256:$dst, (OpVT256 (X86VBroadcast (ld_frag addr:$src))))]>,
+                   [(set VR256:$dst,
+                    (OpVT256 (X86VBroadcast (ld_frag addr:$src))))]>,
                    Sched<[WriteLoad]>, VEX, VEX_L;
+
+    // Provide aliases for broadcast from the same register class that
+    // automatically does the extract.
+    def : Pat<(OpVT256 (X86VBroadcast (OpVT256 VR256:$src))),
+              (!cast<Instruction>(NAME#"Yrr")
+                  (OpVT128 (EXTRACT_SUBREG (OpVT256 VR256:$src),sub_xmm)))>;
+  }
 }
 
-defm VPBROADCASTB  : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8, v16i8, v32i8>;
-defm VPBROADCASTW  : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16, v8i16, v16i16>;
-defm VPBROADCASTD  : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32, v4i32, v8i32>;
-defm VPBROADCASTQ  : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64, v2i64, v4i64>;
+defm VPBROADCASTB  : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
+                                    v16i8, v32i8, NoVLX_Or_NoBWI>;
+defm VPBROADCASTW  : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
+                                    v8i16, v16i16, NoVLX_Or_NoBWI>;
+defm VPBROADCASTD  : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
+                                    v4i32, v8i32, NoVLX>;
+defm VPBROADCASTQ  : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
+                                    v2i64, v4i64, NoVLX>;
 
 let Predicates = [HasAVX2] in {
   // loadi16 is tricky to fold, because !isTypeDesirableForOp, justifiably.
@@ -8349,21 +8370,15 @@ let Predicates = [HasAVX2] in {
             (VPBROADCASTWrm addr:$src)>;
   def : Pat<(v16i16 (X86VBroadcast (i16 (trunc (i32 (load addr:$src)))))),
             (VPBROADCASTWYrm addr:$src)>;
+  def : Pat<(v8i16 (X86VBroadcast
+              (i16 (trunc (i32 (zextloadi16 addr:$src)))))),
+            (VPBROADCASTWrm addr:$src)>;
+  def : Pat<(v16i16 (X86VBroadcast
+              (i16 (trunc (i32 (zextloadi16 addr:$src)))))),
+            (VPBROADCASTWYrm addr:$src)>;
 
   // Provide aliases for broadcast from the same register class that
   // automatically does the extract.
-  def : Pat<(v32i8 (X86VBroadcast (v32i8 VR256:$src))),
-            (VPBROADCASTBYrr (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src),
-                                                    sub_xmm)))>;
-  def : Pat<(v16i16 (X86VBroadcast (v16i16 VR256:$src))),
-            (VPBROADCASTWYrr (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src),
-                                                    sub_xmm)))>;
-  def : Pat<(v8i32 (X86VBroadcast (v8i32 VR256:$src))),
-            (VPBROADCASTDYrr (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src),
-                                                    sub_xmm)))>;
-  def : Pat<(v4i64 (X86VBroadcast (v4i64 VR256:$src))),
-            (VPBROADCASTQYrr (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src),
-                                                    sub_xmm)))>;
   def : Pat<(v8f32 (X86VBroadcast (v8f32 VR256:$src))),
             (VBROADCASTSSYrr (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src),
                                                     sub_xmm)))>;