AVX-512: Added VMOVD, VMOVQ, VMOVSS, VMOVSD instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188637 91177308-0d34-0410-b5e6-96231b3b80d8

6 files changed, 335 insertions, 44 deletions

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 312c8db..4e71f36 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -1434,8 +1434,6 @@ void X86TargetLowering::resetOperationActions() {
       if (!VT.is512BitVector())
         continue;
 
-      setOperationAction(ISD::LOAD,   VT, Promote);
-      AddPromotedToType (ISD::LOAD,   VT, MVT::v8i64);
       setOperationAction(ISD::SELECT, VT, Promote);
       AddPromotedToType (ISD::SELECT, VT, MVT::v8i64);
     }
diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index d100e88..760b4ed 100644
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -555,7 +555,7 @@ let Constraints = "$src1 = $dst" in {
                       (bitconvert (mem_frag addr:$src3)))))]>, EVEX_4V;
   }
 }
-defm VPERMI2D  : avx512_perm_3src<0x76, "vpermi2d",  VR512, memopv8i64, i512mem,
+defm VPERMI2D  : avx512_perm_3src<0x76, "vpermi2d",  VR512, memopv16i32, i512mem, 
                                v16i32>, EVEX_V512, EVEX_CD8<32, CD8VF>;
 defm VPERMI2Q  : avx512_perm_3src<0x76, "vpermi2q",  VR512, memopv8i64, i512mem, 
                                v8i64>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
@@ -1107,7 +1107,7 @@ def VMOVDQA64rm  : AVX512BI<0x6F, MRMSrcMem, (outs VR512:$dst),
 }
 
 multiclass avx512_mov_int<bits<8> opc, string asm, RegisterClass RC,
-                          RegisterClass KRC, PatFrag bc_frag,
+                          RegisterClass KRC,
                           PatFrag ld_frag, X86MemOperand x86memop> {
 let neverHasSideEffects = 1 in
   def rr : AVX512XSI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
@@ -1116,7 +1116,7 @@ let neverHasSideEffects = 1 in
 let canFoldAsLoad = 1 in
   def rm : AVX512XSI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
               !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
-               [(set RC:$dst, (bc_frag (ld_frag addr:$src)))]>,
+               [(set RC:$dst, (ld_frag addr:$src))]>,
                EVEX;
 let Constraints = "$src1 = $dst" in {
   def rrk : AVX512XSI<opc, MRMSrcReg, (outs RC:$dst),
@@ -1132,10 +1132,10 @@ let Constraints = "$src1 = $dst" in {
 }
 }
 
-defm VMOVDQU32 : avx512_mov_int<0x6F, "vmovdqu32", VR512, VK16WM, bc_v16i32,
-                           memopv8i64, i512mem>, EVEX_V512, EVEX_CD8<32, CD8VF>;
-defm VMOVDQU64 : avx512_mov_int<0x6F, "vmovdqu64", VR512, VK8WM, bc_v8i64,
-                    memopv8i64, i512mem>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
+defm VMOVDQU32 : avx512_mov_int<0x6F, "vmovdqu32", VR512, VK16WM, memopv16i32, i512mem>,
+                                EVEX_V512, EVEX_CD8<32, CD8VF>;
+defm VMOVDQU64 : avx512_mov_int<0x6F, "vmovdqu64", VR512, VK8WM, memopv8i64, i512mem>,
+                                EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
 
 let AddedComplexity = 20 in {
 def : Pat<(v16f32 (vselect VK16WM:$mask, (v16f32 VR512:$src1), 
@@ -1151,4 +1151,267 @@ def : Pat<(v8i64 (vselect VK8WM:$mask, (v8i64 VR512:$src1),
                            (v8i64 VR512:$src2))),
                   (VMOVDQU64rrk VR512:$src2, VK8WM:$mask, VR512:$src1)>;
 }
+// Move Int Doubleword to Packed Double Int
+//
+def VMOVDI2PDIZrr : AVX512SI<0x6E, MRMSrcReg, (outs VR128X:$dst), (ins GR32:$src),
+                      "vmovd{z}\t{$src, $dst|$dst, $src}",
+                      [(set VR128X:$dst,
+                        (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
+                        EVEX, VEX_LIG;
+def VMOVDI2PDIZrm : AVX512SI<0x6E, MRMSrcMem, (outs VR128X:$dst), (ins i32mem:$src),
+                      "vmovd{z}\t{$src, $dst|$dst, $src}",
+                      [(set VR128X:$dst,
+                        (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
+                        IIC_SSE_MOVDQ>, EVEX, VEX_LIG, EVEX_CD8<32, CD8VT1>;
+def VMOV64toPQIZrr : AVX512SI<0x6E, MRMSrcReg, (outs VR128X:$dst), (ins GR64:$src),
+                      "vmovq{z}\t{$src, $dst|$dst, $src}",
+                        [(set VR128X:$dst,
+                          (v2i64 (scalar_to_vector GR64:$src)))],
+                          IIC_SSE_MOVDQ>, EVEX, VEX_W, VEX_LIG;
+def VMOV64toSDZrr : AVX512SI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
+                       "vmovq{z}\t{$src, $dst|$dst, $src}",
+                       [(set FR64:$dst, (bitconvert GR64:$src))],
+                       IIC_SSE_MOVDQ>, EVEX, VEX_W, Sched<[WriteMove]>;
+def VMOVSDto64Zrr : AVX512SI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
+                         "vmovq{z}\t{$src, $dst|$dst, $src}",
+                         [(set GR64:$dst, (bitconvert FR64:$src))],
+                         IIC_SSE_MOVDQ>, EVEX, VEX_W, Sched<[WriteMove]>;
+def VMOVSDto64Zmr : AVX512SI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
+                         "vmovq{z}\t{$src, $dst|$dst, $src}",
+                         [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
+                         IIC_SSE_MOVDQ>, EVEX, VEX_W, Sched<[WriteStore]>,
+                         EVEX_CD8<64, CD8VT1>;
+
+// Move Int Doubleword to Single Scalar
+//
+def VMOVDI2SSZrr  : AVX512SI<0x6E, MRMSrcReg, (outs FR32X:$dst), (ins GR32:$src),
+                      "vmovd{z}\t{$src, $dst|$dst, $src}",
+                      [(set FR32X:$dst, (bitconvert GR32:$src))],
+                      IIC_SSE_MOVDQ>, EVEX, VEX_LIG;
+
+def VMOVDI2SSZrm  : AVX512SI<0x6E, MRMSrcMem, (outs FR32X:$dst), (ins i32mem:$src),
+                      "vmovd{z}\t{$src, $dst|$dst, $src}",
+                      [(set FR32X:$dst, (bitconvert (loadi32 addr:$src)))],
+                      IIC_SSE_MOVDQ>, EVEX, VEX_LIG, EVEX_CD8<32, CD8VT1>;
+
+// Move Packed Doubleword Int to Packed Double Int
+//
+def VMOVPDI2DIZrr  : AVX512SI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128X:$src),
+                       "vmovd{z}\t{$src, $dst|$dst, $src}",
+                       [(set GR32:$dst, (vector_extract (v4i32 VR128X:$src),
+                                        (iPTR 0)))], IIC_SSE_MOVD_ToGP>,
+                       EVEX, VEX_LIG;
+def VMOVPDI2DIZmr  : AVX512SI<0x7E, MRMDestMem, (outs),
+                       (ins i32mem:$dst, VR128X:$src),
+                       "vmovd{z}\t{$src, $dst|$dst, $src}",
+                       [(store (i32 (vector_extract (v4i32 VR128X:$src),
+                                     (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
+                       EVEX, VEX_LIG, EVEX_CD8<32, CD8VT1>;
+
+// Move Packed Doubleword Int first element to Doubleword Int
+//
+def VMOVPQIto64Zrr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128X:$src),
+                      "vmovq{z}\t{$src, $dst|$dst, $src}",
+                      [(set GR64:$dst, (extractelt (v2i64 VR128X:$src),
+                                                   (iPTR 0)))],
+                      IIC_SSE_MOVD_ToGP>, TB, OpSize, EVEX, VEX_LIG, VEX_W,
+                      Requires<[HasAVX512, In64BitMode]>;
+
+def VMOVPQIto64Zmr : I<0x7E, MRMDestMem, (outs),
+                       (ins i64mem:$dst, VR128X:$src),
+                       "vmovq{z}\t{$src, $dst|$dst, $src}",
+                       [(store (extractelt (v2i64 VR128X:$src), (iPTR 0)),
+                               addr:$dst)], IIC_SSE_MOVDQ>,
+                       EVEX, VEX_LIG, VEX_W, EVEX_CD8<64, CD8VT1>,
+                       Sched<[WriteStore]>, Requires<[HasAVX512, In64BitMode]>;
+
+// Move Scalar Single to Double Int
+//
+def VMOVSS2DIZrr  : AVX512SI<0x7E, MRMDestReg, (outs GR32:$dst),
+                      (ins FR32X:$src),
+                      "vmovd{z}\t{$src, $dst|$dst, $src}",
+                      [(set GR32:$dst, (bitconvert FR32X:$src))],
+                      IIC_SSE_MOVD_ToGP>, EVEX, VEX_LIG;
+def VMOVSS2DIZmr  : AVX512SI<0x7E, MRMDestMem, (outs),
+                      (ins i32mem:$dst, FR32X:$src),
+                      "vmovd{z}\t{$src, $dst|$dst, $src}",
+                      [(store (i32 (bitconvert FR32X:$src)), addr:$dst)],
+                      IIC_SSE_MOVDQ>, EVEX, VEX_LIG, EVEX_CD8<32, CD8VT1>;
+
+// Move Quadword Int to Packed Quadword Int
+//
+def VMOVQI2PQIZrm : AVX512SI<0x7E, MRMSrcMem, (outs VR128X:$dst),
+                      (ins i64mem:$src),
+                      "vmovq{z}\t{$src, $dst|$dst, $src}",
+                      [(set VR128X:$dst,
+                        (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>,
+                      EVEX, VEX_LIG, VEX_W, EVEX_CD8<64, CD8VT1>;
+
+//===----------------------------------------------------------------------===//
+// AVX-512  MOVSS, MOVSD
+//===----------------------------------------------------------------------===//
+
+multiclass avx512_move_scalar <string asm, RegisterClass RC, 
+                              SDNode OpNode, ValueType vt,
+                              X86MemOperand x86memop, PatFrag mem_pat> {
+  def rr : SI<0x10, MRMSrcReg, (outs VR128X:$dst), (ins VR128X:$src1, RC:$src2), 
+              !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+              [(set VR128X:$dst, (vt (OpNode VR128X:$src1,
+                                      (scalar_to_vector RC:$src2))))],
+              IIC_SSE_MOV_S_RR>, EVEX_4V, VEX_LIG;
+  def rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
+              !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
+              [(set RC:$dst, (mem_pat addr:$src))], IIC_SSE_MOV_S_RM>,
+              EVEX, VEX_LIG;
+  def mr: SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
+             !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
+             [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
+             EVEX, VEX_LIG;
+}
+
+let ExeDomain = SSEPackedSingle in
+defm VMOVSSZ : avx512_move_scalar<"movss{z}", FR32X, X86Movss, v4f32, f32mem,
+                                 loadf32>, XS, EVEX_CD8<32, CD8VT1>;
+
+let ExeDomain = SSEPackedDouble in
+defm VMOVSDZ : avx512_move_scalar<"movsd{z}", FR64X, X86Movsd, v2f64, f64mem,
+                                 loadf64>, XD, VEX_W, EVEX_CD8<64, CD8VT1>;
+
+
+// For the disassembler
+let isCodeGenOnly = 1 in {
+  def VMOVSSZrr_REV : SI<0x11, MRMDestReg, (outs VR128X:$dst),
+                        (ins VR128X:$src1, FR32X:$src2),
+                        "movss{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
+                        IIC_SSE_MOV_S_RR>,
+                        XS, EVEX_4V, VEX_LIG;
+  def VMOVSDZrr_REV : SI<0x11, MRMDestReg, (outs VR128X:$dst),
+                        (ins VR128X:$src1, FR64X:$src2),
+                        "movsd{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
+                        IIC_SSE_MOV_S_RR>,
+                        XD, EVEX_4V, VEX_LIG, VEX_W;
+}
+
+let Predicates = [HasAVX512] in {
+  let AddedComplexity = 20 in {
+  // MOVSSrm zeros the high parts of the register; represent this
+  // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
+  def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
+            (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128X)>;
+  def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
+            (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128X)>;
+  def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
+            (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128X)>;
+
+  // MOVSDrm zeros the high parts of the register; represent this
+  // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
+  def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
+            (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128X)>;
+  def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
+            (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128X)>;
+  def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
+            (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128X)>;
+  def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
+            (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128X)>;
+  def : Pat<(v2f64 (X86vzload addr:$src)),
+            (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128X)>;
+
+  // Represent the same patterns above but in the form they appear for
+  // 256-bit types
+  def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
+                   (v4i32 (scalar_to_vector (loadi32 addr:$src))), (iPTR 0)))),
+            (SUBREG_TO_REG (i32 0), (VMOVSSZrm addr:$src), sub_xmm)>;
+  def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
+                   (v4f32 (scalar_to_vector (loadf32 addr:$src))), (iPTR 0)))),
+            (SUBREG_TO_REG (i32 0), (VMOVSSZrm addr:$src), sub_xmm)>;
+  def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
+                   (v2f64 (scalar_to_vector (loadf64 addr:$src))), (iPTR 0)))),
+            (SUBREG_TO_REG (i32 0), (VMOVSDZrm addr:$src), sub_xmm)>;
+  }
+  def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
+                   (v2i64 (scalar_to_vector (loadi64 addr:$src))), (iPTR 0)))),
+            (SUBREG_TO_REG (i64 0), (VMOVSDZrm addr:$src), sub_xmm)>;
+
+  // Extract and store.
+  def : Pat<(store (f32 (vector_extract (v4f32 VR128X:$src), (iPTR 0))),
+                   addr:$dst),
+            (VMOVSSZmr addr:$dst, (COPY_TO_REGCLASS (v4f32 VR128X:$src), FR32X))>;
+  def : Pat<(store (f64 (vector_extract (v2f64 VR128X:$src), (iPTR 0))),
+                   addr:$dst),
+            (VMOVSDZmr addr:$dst, (COPY_TO_REGCLASS (v2f64 VR128X:$src), FR64X))>;
+
+  // Shuffle with VMOVSS
+  def : Pat<(v4i32 (X86Movss VR128X:$src1, VR128X:$src2)),
+            (VMOVSSZrr (v4i32 VR128X:$src1),
+                      (COPY_TO_REGCLASS (v4i32 VR128X:$src2), FR32X))>;
+  def : Pat<(v4f32 (X86Movss VR128X:$src1, VR128X:$src2)),
+            (VMOVSSZrr (v4f32 VR128X:$src1),
+                      (COPY_TO_REGCLASS (v4f32 VR128X:$src2), FR32X))>;
+
+  // 256-bit variants
+  def : Pat<(v8i32 (X86Movss VR256X:$src1, VR256X:$src2)),
+            (SUBREG_TO_REG (i32 0),
+              (VMOVSSZrr (EXTRACT_SUBREG (v8i32 VR256X:$src1), sub_xmm),
+                        (EXTRACT_SUBREG (v8i32 VR256X:$src2), sub_xmm)),
+              sub_xmm)>;
+  def : Pat<(v8f32 (X86Movss VR256X:$src1, VR256X:$src2)),
+            (SUBREG_TO_REG (i32 0),
+              (VMOVSSZrr (EXTRACT_SUBREG (v8f32 VR256X:$src1), sub_xmm),
+                        (EXTRACT_SUBREG (v8f32 VR256X:$src2), sub_xmm)),
+              sub_xmm)>;
+
+  // Shuffle with VMOVSD
+  def : Pat<(v2i64 (X86Movsd VR128X:$src1, VR128X:$src2)),
+            (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>;
+  def : Pat<(v2f64 (X86Movsd VR128X:$src1, VR128X:$src2)),
+            (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>;
+  def : Pat<(v4f32 (X86Movsd VR128X:$src1, VR128X:$src2)),
+            (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>;
+  def : Pat<(v4i32 (X86Movsd VR128X:$src1, VR128X:$src2)),
+            (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>;
+
+  // 256-bit variants
+  def : Pat<(v4i64 (X86Movsd VR256X:$src1, VR256X:$src2)),
+            (SUBREG_TO_REG (i32 0),
+              (VMOVSDZrr (EXTRACT_SUBREG (v4i64 VR256X:$src1), sub_xmm),
+                        (EXTRACT_SUBREG (v4i64 VR256X:$src2), sub_xmm)),
+              sub_xmm)>;
+  def : Pat<(v4f64 (X86Movsd VR256X:$src1, VR256X:$src2)),
+            (SUBREG_TO_REG (i32 0),
+              (VMOVSDZrr (EXTRACT_SUBREG (v4f64 VR256X:$src1), sub_xmm),
+                        (EXTRACT_SUBREG (v4f64 VR256X:$src2), sub_xmm)),
+              sub_xmm)>;
+
+  def : Pat<(v2f64 (X86Movlpd VR128X:$src1, VR128X:$src2)),
+            (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>;
+  def : Pat<(v2i64 (X86Movlpd VR128X:$src1, VR128X:$src2)),
+            (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>;
+  def : Pat<(v4f32 (X86Movlps VR128X:$src1, VR128X:$src2)),
+            (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>;
+  def : Pat<(v4i32 (X86Movlps VR128X:$src1, VR128X:$src2)),
+            (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>;
+}
+
+let AddedComplexity = 15 in
+def VMOVZPQILo2PQIZrr : AVX512XSI<0x7E, MRMSrcReg, (outs VR128X:$dst),
+                                (ins VR128X:$src),
+                                "vmovq{z}\t{$src, $dst|$dst, $src}",
+                                [(set VR128X:$dst, (v2i64 (X86vzmovl 
+                                                   (v2i64 VR128X:$src))))],
+                                IIC_SSE_MOVQ_RR>, EVEX, VEX_W;
+
+let AddedComplexity = 20 in
+def VMOVZPQILo2PQIZrm : AVX512XSI<0x7E, MRMSrcMem, (outs VR128X:$dst),
+                                 (ins i128mem:$src),
+                                 "vmovq{z}\t{$src, $dst|$dst, $src}",
+                                 [(set VR128X:$dst, (v2i64 (X86vzmovl
+                                                     (loadv2i64 addr:$src))))],
+                                 IIC_SSE_MOVDQ>, EVEX, VEX_W,
+                                 EVEX_CD8<8, CD8VT8>;
+let AddedComplexity = 20 in {
+  def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
+            (VMOVZPQILo2PQIZrm addr:$src)>;
+  def : Pat<(v2f64 (X86vzmovl (v2f64 VR128X:$src))),
+            (VMOVZPQILo2PQIZrr VR128X:$src)>;
+}
 
diff --git a/lib/Target/X86/X86InstrFormats.td b/lib/Target/X86/X86InstrFormats.td
index 64018b3..b50706c 100644
--- a/lib/Target/X86/X86InstrFormats.td
+++ b/lib/Target/X86/X86InstrFormats.td
@@ -339,10 +339,11 @@ def __xd : XD;
 class SI<bits<8> o, Format F, dag outs, dag ins, string asm,
          list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin> {
-  let Predicates = !if(hasVEXPrefix /* VEX */, [HasAVX],
+  let Predicates = !if(hasEVEXPrefix /* EVEX */, [HasAVX512],
+                   !if(hasVEXPrefix /* VEX */, [UseAVX],
                    !if(!eq(Prefix, __xs.Prefix), [UseSSE1],
                    !if(!eq(Prefix, __xd.Prefix), [UseSSE2],
-                   !if(hasOpSizePrefix, [UseSSE2], [UseSSE1]))));
+                   !if(hasOpSizePrefix, [UseSSE2], [UseSSE1])))));
 
   // AVX instructions have a 'v' prefix in the mnemonic
   let AsmString = !if(hasVEXPrefix, !strconcat("v", asm), asm);
@@ -352,8 +353,9 @@ class SI<bits<8> o, Format F, dag outs, dag ins, string asm,
 class SIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : Ii8<o, F, outs, ins, asm, pattern, itin> {
-  let Predicates = !if(hasVEXPrefix /* VEX */, [HasAVX],
-            !if(!eq(Prefix, __xs.Prefix), [UseSSE1], [UseSSE2]));
+  let Predicates = !if(hasEVEXPrefix /* EVEX */, [HasAVX512],
+                   !if(hasVEXPrefix /* VEX */, [UseAVX],
+                   !if(!eq(Prefix, __xs.Prefix), [UseSSE1], [UseSSE2])));
 
   // AVX instructions have a 'v' prefix in the mnemonic
   let AsmString = !if(hasVEXPrefix, !strconcat("v", asm), asm);
@@ -363,8 +365,9 @@ class SIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
 class PI<bits<8> o, Format F, dag outs, dag ins, string asm, list<dag> pattern,
          InstrItinClass itin, Domain d>
       : I<o, F, outs, ins, asm, pattern, itin, d> {
-  let Predicates = !if(hasVEXPrefix /* VEX */, [HasAVX],
-        !if(hasOpSizePrefix /* OpSize */, [UseSSE2], [UseSSE1]));
+  let Predicates = !if(hasEVEXPrefix /* EVEX */, [HasAVX512],
+                   !if(hasVEXPrefix /* VEX */, [HasAVX],
+                   !if(hasOpSizePrefix /* OpSize */, [UseSSE2], [UseSSE1])));
 
   // AVX instructions have a 'v' prefix in the mnemonic
   let AsmString = !if(hasVEXPrefix, !strconcat("v", asm), asm);
@@ -381,11 +384,12 @@ class MMXPI<bits<8> o, Format F, dag outs, dag ins, string asm, list<dag> patter
 class PIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
            list<dag> pattern, InstrItinClass itin, Domain d>
       : Ii8<o, F, outs, ins, asm, pattern, itin, d> {
-  let Predicates = !if(hasVEX_4VPrefix /* VEX */, [HasAVX],
-        !if(hasOpSizePrefix /* OpSize */, [UseSSE2], [UseSSE1]));
+  let Predicates = !if(hasEVEXPrefix /* EVEX */, [HasAVX512],
+                   !if(hasVEXPrefix /* VEX */, [HasAVX],
+                   !if(hasOpSizePrefix /* OpSize */, [UseSSE2], [UseSSE1])));
 
   // AVX instructions have a 'v' prefix in the mnemonic
-  let AsmString = !if(hasVEX_4VPrefix, !strconcat("v", asm), asm);
+  let AsmString = !if(hasVEXPrefix, !strconcat("v", asm), asm);
 }
 
 // SSE1 Instruction Templates:
@@ -460,7 +464,7 @@ class PDIi8<bits<8> o, Format F, dag outs, dag ins, string asm,
 class VSDI<bits<8> o, Format F, dag outs, dag ins, string asm,
            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, !strconcat("v", asm), pattern, itin>, XD,
-        Requires<[HasAVX]>;
+        Requires<[UseAVX]>;
 class VS2SI<bits<8> o, Format F, dag outs, dag ins, string asm,
             list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, !strconcat("v", asm), pattern, itin>, XS,
@@ -472,7 +476,7 @@ class VPDI<bits<8> o, Format F, dag outs, dag ins, string asm,
 class VS2I<bits<8> o, Format F, dag outs, dag ins, string asm,
            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, !strconcat("v", asm), pattern, itin>, TB,
-        OpSize, Requires<[HasAVX]>;
+        OpSize, Requires<[UseAVX]>;
 class S2I<bits<8> o, Format F, dag outs, dag ins, string asm,
            list<dag> pattern, InstrItinClass itin = NoItinerary>
       : I<o, F, outs, ins, asm, pattern, itin>, TB,
diff --git a/lib/Target/X86/X86InstrFragmentsSIMD.td b/lib/Target/X86/X86InstrFragmentsSIMD.td
index fe35393..e6460e9 100644
--- a/lib/Target/X86/X86InstrFragmentsSIMD.td
+++ b/lib/Target/X86/X86InstrFragmentsSIMD.td
@@ -348,6 +348,8 @@ def alignedloadv4i64 : PatFrag<(ops node:$ptr),
 // 512-bit aligned load pattern fragments
 def alignedloadv16f32 : PatFrag<(ops node:$ptr),
                                 (v16f32 (alignedload512 node:$ptr))>;
+def alignedloadv16i32 : PatFrag<(ops node:$ptr),
+                                (v16i32 (alignedload512 node:$ptr))>;
 def alignedloadv8f64  : PatFrag<(ops node:$ptr),
                                 (v8f64  (alignedload512 node:$ptr))>;
 def alignedloadv8i64  : PatFrag<(ops node:$ptr),
@@ -382,6 +384,7 @@ def memopv4i64 : PatFrag<(ops node:$ptr), (v4i64 (memop node:$ptr))>;
 // 512-bit memop pattern fragments
 def memopv16f32 : PatFrag<(ops node:$ptr), (v16f32 (memop node:$ptr))>;
 def memopv8f64 : PatFrag<(ops node:$ptr), (v8f64 (memop node:$ptr))>;
+def memopv16i32 : PatFrag<(ops node:$ptr), (v16i32 (memop node:$ptr))>;
 def memopv8i64 : PatFrag<(ops node:$ptr), (v8i64 (memop node:$ptr))>;
 
 // SSSE3 uses MMX registers for some instructions. They aren't aligned on a
@@ -437,7 +440,6 @@ def bc_v4i64 : PatFrag<(ops node:$in), (v4i64 (bitconvert node:$in))>;
 def bc_v16i32 : PatFrag<(ops node:$in), (v16i32 (bitconvert node:$in))>;
 def bc_v8i64 : PatFrag<(ops node:$in), (v8i64 (bitconvert node:$in))>;
 
-
 def vzmovl_v2i64 : PatFrag<(ops node:$src),
                            (bitconvert (v2i64 (X86vzmovl
                              (v2i64 (scalar_to_vector (loadi64 node:$src))))))>;
diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index b773768..71df2bb 100644
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -2910,17 +2910,20 @@ static unsigned CopyToFromAsymmetricReg(unsigned DestReg, unsigned SrcReg,
   // SrcReg(GR64)  -> DestReg(VR64)
 
   bool HasAVX = Subtarget.hasAVX();
+  bool HasAVX512 = Subtarget.hasAVX512();
   if (X86::GR64RegClass.contains(DestReg)) {
-    if (X86::VR128RegClass.contains(SrcReg))
+    if (X86::VR128XRegClass.contains(SrcReg))
       // Copy from a VR128 register to a GR64 register.
-      return HasAVX ? X86::VMOVPQIto64rr : X86::MOVPQIto64rr;
+      return HasAVX512 ? X86::VMOVPQIto64Zrr: (HasAVX ? X86::VMOVPQIto64rr :
+                                               X86::MOVPQIto64rr);
     if (X86::VR64RegClass.contains(SrcReg))
       // Copy from a VR64 register to a GR64 register.
       return X86::MOVSDto64rr;
   } else if (X86::GR64RegClass.contains(SrcReg)) {
     // Copy from a GR64 register to a VR128 register.
-    if (X86::VR128RegClass.contains(DestReg))
-      return HasAVX ? X86::VMOV64toPQIrr : X86::MOV64toPQIrr;
+    if (X86::VR128XRegClass.contains(DestReg))
+      return HasAVX512 ? X86::VMOV64toPQIZrr: (HasAVX ? X86::VMOV64toPQIrr :
+                                               X86::MOV64toPQIrr);
     // Copy from a GR64 register to a VR64 register.
     if (X86::VR64RegClass.contains(DestReg))
       return X86::MOV64toSDrr;
@@ -2929,13 +2932,13 @@ static unsigned CopyToFromAsymmetricReg(unsigned DestReg, unsigned SrcReg,
   // SrcReg(FR32) -> DestReg(GR32)
   // SrcReg(GR32) -> DestReg(FR32)
 
-  if (X86::GR32RegClass.contains(DestReg) && X86::FR32RegClass.contains(SrcReg))
+  if (X86::GR32RegClass.contains(DestReg) && X86::FR32XRegClass.contains(SrcReg))
     // Copy from a FR32 register to a GR32 register.
-    return HasAVX ? X86::VMOVSS2DIrr : X86::MOVSS2DIrr;
+    return HasAVX512 ? X86::VMOVSS2DIZrr : (HasAVX ? X86::VMOVSS2DIrr : X86::MOVSS2DIrr);
 
-  if (X86::FR32RegClass.contains(DestReg) && X86::GR32RegClass.contains(SrcReg))
+  if (X86::FR32XRegClass.contains(DestReg) && X86::GR32RegClass.contains(SrcReg))
     // Copy from a GR32 register to a FR32 register.
-    return HasAVX ? X86::VMOVDI2SSrr : X86::MOVDI2SSrr;
+    return HasAVX512 ? X86::VMOVDI2SSZrr : (HasAVX ? X86::VMOVDI2SSrr : X86::MOVDI2SSrr);
 
   return 0;
 }
@@ -3040,6 +3043,21 @@ static unsigned getLoadStoreRegOpcode(unsigned Reg,
                                       bool isStackAligned,
                                       const TargetMachine &TM,
                                       bool load) {
+  if (TM.getSubtarget<X86Subtarget>().hasAVX512()) {
+    if (X86::VK8RegClass.hasSubClassEq(RC)  || 
+      X86::VK16RegClass.hasSubClassEq(RC))
+      return load ? X86::KMOVWkm : X86::KMOVWmk;
+
+    if (X86::FR32XRegClass.hasSubClassEq(RC))
+      return load ? X86::VMOVSSZrm : X86::VMOVSSZmr;
+    if (X86::FR64XRegClass.hasSubClassEq(RC))
+      return load ? X86::VMOVSDZrm : X86::VMOVSDZmr;
+    if (X86::VR128XRegClass.hasSubClassEq(RC) ||
+        X86::VR256XRegClass.hasSubClassEq(RC) ||
+        X86::VR512RegClass.hasSubClassEq(RC))
+      return load ? X86::VMOVUPSZrm : X86::VMOVUPSZmr;
+  }
+
   bool HasAVX = TM.getSubtarget<X86Subtarget>().hasAVX();
   switch (RC->getSize()) {
   default:
@@ -3099,6 +3117,12 @@ static unsigned getLoadStoreRegOpcode(unsigned Reg,
       return load ? X86::VMOVAPSYrm : X86::VMOVAPSYmr;
     else
       return load ? X86::VMOVUPSYrm : X86::VMOVUPSYmr;
+  case 64:
+    assert(X86::VR512RegClass.hasSubClassEq(RC) && "Unknown 64-byte regclass");
+    if (isStackAligned)
+      return load ? X86::VMOVAPSZrm : X86::VMOVAPSZmr;
+    else
+      return load ? X86::VMOVUPSZrm : X86::VMOVUPSZmr;
   }
 }
 
@@ -3125,7 +3149,7 @@ void X86InstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
   const MachineFunction &MF = *MBB.getParent();
   assert(MF.getFrameInfo()->getObjectSize(FrameIdx) >= RC->getSize() &&
          "Stack slot too small for store");
-  unsigned Alignment = RC->getSize() == 32 ? 32 : 16;
+  unsigned Alignment = std::max<uint32_t>(RC->getSize(), 16);
   bool isAligned = (TM.getFrameLowering()->getStackAlignment() >= Alignment) ||
     RI.canRealignStack(MF);
   unsigned Opc = getStoreRegOpcode(SrcReg, RC, isAligned, TM);
@@ -3141,7 +3165,7 @@ void X86InstrInfo::storeRegToAddr(MachineFunction &MF, unsigned SrcReg,
                                   MachineInstr::mmo_iterator MMOBegin,
                                   MachineInstr::mmo_iterator MMOEnd,
                                   SmallVectorImpl<MachineInstr*> &NewMIs) const {
-  unsigned Alignment = RC->getSize() == 32 ? 32 : 16;
+  unsigned Alignment = std::max<uint32_t>(RC->getSize(), 16);
   bool isAligned = MMOBegin != MMOEnd &&
                    (*MMOBegin)->getAlignment() >= Alignment;
   unsigned Opc = getStoreRegOpcode(SrcReg, RC, isAligned, TM);
@@ -3161,7 +3185,7 @@ void X86InstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
                                         const TargetRegisterClass *RC,
                                         const TargetRegisterInfo *TRI) const {
   const MachineFunction &MF = *MBB.getParent();
-  unsigned Alignment = RC->getSize() == 32 ? 32 : 16;
+  unsigned Alignment = std::max<uint32_t>(RC->getSize(), 16);
   bool isAligned = (TM.getFrameLowering()->getStackAlignment() >= Alignment) ||
     RI.canRealignStack(MF);
   unsigned Opc = getLoadRegOpcode(DestReg, RC, isAligned, TM);
@@ -3175,7 +3199,7 @@ void X86InstrInfo::loadRegFromAddr(MachineFunction &MF, unsigned DestReg,
                                  MachineInstr::mmo_iterator MMOBegin,
                                  MachineInstr::mmo_iterator MMOEnd,
                                  SmallVectorImpl<MachineInstr*> &NewMIs) const {
-  unsigned Alignment = RC->getSize() == 32 ? 32 : 16;
+  unsigned Alignment = std::max<uint32_t>(RC->getSize(), 16);
   bool isAligned = MMOBegin != MMOEnd &&
                    (*MMOBegin)->getAlignment() >= Alignment;
   unsigned Opc = getLoadRegOpcode(DestReg, RC, isAligned, TM);
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index a86006a..aa057db 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -4484,8 +4484,8 @@ def MOVPQIto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
 //===---------------------------------------------------------------------===//
 // Bitcast FR64 <-> GR64
 //
-let Predicates = [HasAVX] in
-def VMOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
+let Predicates = [UseAVX] in
+def VMOV64toSDrm : VS2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
                         "vmovq\t{$src, $dst|$dst, $src}",
                         [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
                         VEX, Sched<[WriteLoad]>;
@@ -4577,7 +4577,7 @@ def MOVZDI2PDIrm : S2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
                                                    IIC_SSE_MOVDQ>;
 } // AddedComplexity, SchedRW
 
-let Predicates = [HasAVX] in {
+let Predicates = [UseAVX] in {
   // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
   let AddedComplexity = 20 in {
     def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
@@ -4630,7 +4630,7 @@ def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
                     "vmovq\t{$src, $dst|$dst, $src}",
                     [(set VR128:$dst,
                       (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
-                    VEX, Requires<[HasAVX]>;
+                    VEX, Requires<[UseAVX]>;
 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
                     "movq\t{$src, $dst|$dst, $src}",
                     [(set VR128:$dst,
@@ -4674,7 +4674,7 @@ def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
                        (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
                                                  (loadi64 addr:$src))))))],
                                                  IIC_SSE_MOVDQ>,
-                     XS, VEX, Requires<[HasAVX]>, Sched<[WriteLoad]>;
+                     XS, VEX, Requires<[UseAVX]>, Sched<[WriteLoad]>;
 
 let AddedComplexity = 20 in
 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
@@ -4685,7 +4685,7 @@ def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
                                                  IIC_SSE_MOVDQ>,
                      XS, Requires<[UseSSE2]>, Sched<[WriteLoad]>;
 
-let Predicates = [HasAVX], AddedComplexity = 20 in {
+let Predicates = [UseAVX], AddedComplexity = 20 in {
   def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
             (VMOVZQI2PQIrm addr:$src)>;
   def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
@@ -4719,7 +4719,7 @@ def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                         "vmovq\t{$src, $dst|$dst, $src}",
                     [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
                     IIC_SSE_MOVQ_RR>,
-                      XS, VEX, Requires<[HasAVX]>;
+                      XS, VEX, Requires<[UseAVX]>;
 let AddedComplexity = 15 in
 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                         "movq\t{$src, $dst|$dst, $src}",
@@ -4735,7 +4735,7 @@ def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
                     [(set VR128:$dst, (v2i64 (X86vzmovl
                                              (loadv2i64 addr:$src))))],
                                              IIC_SSE_MOVDQ>,
-                      XS, VEX, Requires<[HasAVX]>;
+                      XS, VEX, Requires<[UseAVX]>;
 let AddedComplexity = 20 in {
 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
                         "movq\t{$src, $dst|$dst, $src}",
@@ -4747,7 +4747,7 @@ def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
 } // SchedRW
 
 let AddedComplexity = 20 in {
-  let Predicates = [HasAVX] in {
+  let Predicates = [UseAVX] in {
     def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
               (VMOVZPQILo2PQIrm addr:$src)>;
     def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
@@ -4771,7 +4771,7 @@ def VMOVQd64rr : VS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
                       IIC_SSE_MOVDQ>, VEX, VEX_W;
 // Recognize "movd" with GR64 destination, but encode as a "movq"
 def VMOVQd64rr_alt : VS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
-                          "movd\t{$src, $dst|$dst, $src}", [],
+                          "movq\t{$src, $dst|$dst, $src}", [],
                           IIC_SSE_MOVDQ>, VEX, VEX_W;
 } // SchedRW
 
@@ -4780,7 +4780,7 @@ def VMOVQd64rr_alt : VS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
 // xm = mem64
 
 let SchedRW = [WriteMove] in {
-let Predicates = [HasAVX] in
+let Predicates = [UseAVX] in
 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                  "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),