[AVX512] Bring back vector-shuffle lowering support through broadcasts

[oota-llvm.git] / lib / Target / X86 / X86InstrAVX512.td

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index be161e655cbecfb4054d81f1e2ec8b551f0ba46a..3dbc3d2abd8fdd62775ee5af007321afb70f48bb 100644 (file)
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -711,6 +711,16 @@ def : Pat<(v16f32 (X86VBroadcast (v4f32 VR128X:$src))),
 def : Pat<(v8f64 (X86VBroadcast (v2f64 VR128X:$src))),
           (VBROADCASTSDZrr VR128X:$src)>;
 
+def : Pat<(v16f32 (X86VBroadcast (v16f32 VR512:$src))),
+          (VBROADCASTSSZrr (EXTRACT_SUBREG (v16f32 VR512:$src), sub_xmm))>;
+def : Pat<(v8f64 (X86VBroadcast (v8f64 VR512:$src))),
+          (VBROADCASTSDZrr (EXTRACT_SUBREG (v8f64 VR512:$src), sub_xmm))>;
+
+def : Pat<(v16i32 (X86VBroadcast (v16i32 VR512:$src))),
+          (VPBROADCASTDZrr (EXTRACT_SUBREG (v16i32 VR512:$src), sub_xmm))>;
+def : Pat<(v8i64 (X86VBroadcast (v8i64 VR512:$src))),
+          (VPBROADCASTQZrr (EXTRACT_SUBREG (v8i64 VR512:$src), sub_xmm))>;
+
 def : Pat<(v16f32 (int_x86_avx512_vbroadcast_ss_ps_512 (v4f32 VR128X:$src))),
           (VBROADCASTSSZrr VR128X:$src)>;
 def : Pat<(v8f64 (int_x86_avx512_vbroadcast_sd_pd_512 (v2f64 VR128X:$src))),
@@ -762,30 +772,64 @@ defm VPBROADCASTMB2Q : avx512_mask_broadcast<0x2A, "vpbroadcastmb2q",
 // AVX-512 - VPERM
 //
 // -- immediate form --
-multiclass avx512_perm_imm<bits<8> opc, string OpcodeStr, RegisterClass RC,
-                         SDNode OpNode, PatFrag mem_frag, 
-                         X86MemOperand x86memop, ValueType OpVT> {
-  def ri : AVX512AIi8<opc, MRMSrcReg, (outs RC:$dst),
-                     (ins RC:$src1, i8imm:$src2),
+multiclass avx512_perm_imm<bits<8> opc, string OpcodeStr, SDNode OpNode,
+                           X86VectorVTInfo _> {
+  let ExeDomain = _.ExeDomain in {
+  def ri : AVX512AIi8<opc, MRMSrcReg, (outs _.RC:$dst),
+                     (ins _.RC:$src1, i8imm:$src2),
                      !strconcat(OpcodeStr,
                          " \t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-                     [(set RC:$dst,
-                       (OpVT (OpNode RC:$src1, (i8 imm:$src2))))]>,
+                     [(set _.RC:$dst,
+                       (_.VT (OpNode _.RC:$src1, (i8 imm:$src2))))]>,
                      EVEX;
-  def mi : AVX512AIi8<opc, MRMSrcMem, (outs RC:$dst),
-                     (ins x86memop:$src1, i8imm:$src2),
+  def mi : AVX512AIi8<opc, MRMSrcMem, (outs _.RC:$dst),
+                     (ins _.MemOp:$src1, i8imm:$src2),
                      !strconcat(OpcodeStr,
                          " \t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-                     [(set RC:$dst,
-                       (OpVT (OpNode (mem_frag addr:$src1),
-                              (i8 imm:$src2))))]>, EVEX;
+                     [(set _.RC:$dst,
+                       (_.VT (OpNode (_.MemOpFrag addr:$src1),
+                              (i8 imm:$src2))))]>,
+           EVEX, EVEX_CD8<_.EltSize, CD8VF>;
+}
 }
 
-defm VPERMQZ  : avx512_perm_imm<0x00, "vpermq", VR512, X86VPermi, memopv8i64,
-                        i512mem, v8i64>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
-let ExeDomain = SSEPackedDouble in 
-defm VPERMPDZ  : avx512_perm_imm<0x01, "vpermpd", VR512, X86VPermi, memopv8f64, 
-                        f512mem, v8f64>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
+multiclass avx512_permil<bits<8> OpcImm, bits<8> OpcVar, X86VectorVTInfo _,
+                         X86VectorVTInfo Ctrl> :
+     avx512_perm_imm<OpcImm, "vpermil" # _.Suffix, X86VPermilpi, _> {
+  let ExeDomain = _.ExeDomain in {
+    def rr : AVX5128I<OpcVar, MRMSrcReg, (outs _.RC:$dst),
+                     (ins _.RC:$src1, _.RC:$src2),
+                     !strconcat("vpermil" # _.Suffix,
+                         " \t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                     [(set _.RC:$dst,
+                         (_.VT (X86VPermilpv _.RC:$src1,
+                                  (Ctrl.VT Ctrl.RC:$src2))))]>,
+             EVEX_4V;
+    def rm : AVX5128I<OpcVar, MRMSrcMem, (outs _.RC:$dst),
+                     (ins _.RC:$src1, Ctrl.MemOp:$src2),
+                     !strconcat("vpermil" # _.Suffix,
+                         " \t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+                     [(set _.RC:$dst,
+                         (_.VT (X86VPermilpv _.RC:$src1,
+                                  (Ctrl.VT (Ctrl.MemOpFrag addr:$src2)))))]>,
+             EVEX_4V;
+  }
+}
+
+defm VPERMQZ :    avx512_perm_imm<0x00, "vpermq", X86VPermi, v8i64_info>,
+                  EVEX_V512, VEX_W;
+defm VPERMPDZ :   avx512_perm_imm<0x01, "vpermpd", X86VPermi, v8f64_info>,
+                  EVEX_V512, VEX_W;
+
+defm VPERMILPSZ : avx512_permil<0x04, 0x0C, v16f32_info, v16i32_info>,
+                  EVEX_V512;
+defm VPERMILPDZ : avx512_permil<0x05, 0x0D, v8f64_info, v8i64_info>,
+                  EVEX_V512, VEX_W;
+
+def : Pat<(v16i32 (X86VPermilpi VR512:$src1, (i8 imm:$imm))),
+          (VPERMILPSZri VR512:$src1, imm:$imm)>;
+def : Pat<(v8i64 (X86VPermilpi VR512:$src1, (i8 imm:$imm))),
+          (VPERMILPDZri VR512:$src1, imm:$imm)>;
 
 // -- VPERM - register form --
 multiclass avx512_perm<bits<8> opc, string OpcodeStr, RegisterClass RC, 
@@ -2897,20 +2941,6 @@ multiclass avx512_pshuf_imm<bits<8> opc, string OpcodeStr, RegisterClass RC,
 defm VPSHUFDZ : avx512_pshuf_imm<0x70, "vpshufd", VR512, X86PShufd, memopv16i32,
                       i512mem, v16i32>, PD, EVEX_V512, EVEX_CD8<32, CD8VF>;
 
-let ExeDomain = SSEPackedSingle in
-defm VPERMILPSZ : avx512_pshuf_imm<0x04, "vpermilps", VR512, X86VPermilpi,
-                      memopv16f32, f512mem, v16f32>, TAPD, EVEX_V512,
-                      EVEX_CD8<32, CD8VF>;
-let ExeDomain = SSEPackedDouble in
-defm VPERMILPDZ : avx512_pshuf_imm<0x05, "vpermilpd", VR512, X86VPermilpi,
-                      memopv8f64, f512mem, v8f64>, TAPD, EVEX_V512,
-                      VEX_W, EVEX_CD8<64, CD8VF>;
-
-def : Pat<(v16i32 (X86VPermilpi VR512:$src1, (i8 imm:$imm))),
-          (VPERMILPSZri VR512:$src1, imm:$imm)>;
-def : Pat<(v8i64 (X86VPermilpi VR512:$src1, (i8 imm:$imm))),
-          (VPERMILPDZri VR512:$src1, imm:$imm)>;
-
 //===----------------------------------------------------------------------===//
 // AVX-512  Logical Instructions
 //===----------------------------------------------------------------------===//