AVX-512: Implemented VFIXUPIMMPD and VFIXUPIMMPS instructions for KNL and SKX

Implemented DAG lowering for all these forms.
Added tests for encoding.

by Igor Breger (igor.breger@intel.com)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@238728 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index aab0087ed7a5c097c0d5319680bfde3b07c2a696..655fd45c1f357239b7fc35e00c46911cf31c883f 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -18221,6 +18221,7 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::VPERMV3:            return "X86ISD::VPERMV3";
   case X86ISD::VPERMIV3:           return "X86ISD::VPERMIV3";
   case X86ISD::VPERMI:             return "X86ISD::VPERMI";
+  case X86ISD::VFIXUPIMM:          return "X86ISD::VFIXUPIMM";
   case X86ISD::PMULUDQ:            return "X86ISD::PMULUDQ";
   case X86ISD::PMULDQ:             return "X86ISD::PMULDQ";
   case X86ISD::PSADBW:             return "X86ISD::PSADBW";

diff --git a/lib/Target/X86/X86ISelLowering.h b/lib/Target/X86/X86ISelLowering.h
index f48cb52b54782254ebdcba55091c34fdfb6a8ec7..970b3fd0a983a4ce0adac95665b4f1cba7f36c4d 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.h
+++ b/lib/Target/X86/X86ISelLowering.h
@@ -377,6 +377,8 @@ namespace llvm {
       VPERMIV3,
       VPERMI,
       VPERM2X128,
+      //Fix Up Special Packed Float32/64 values
+      VFIXUPIMM,
       // Broadcast scalar to vector
       VBROADCAST,
       // Broadcast subvector to vector

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index 9d11d3c7050f8dd72a455f9068e71283f6d3fa83..7db7fa466dec8885df23edbddbcdab52d174e1b4 100644 (file)
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -5979,3 +5979,70 @@ defm VEXPANDPS : expand_by_elt_width <0x88, "vexpandps", avx512vl_f32_info>,
                                          EVEX;
 defm VEXPANDPD : expand_by_elt_width <0x88, "vexpandpd", avx512vl_f64_info>,
                                          EVEX, VEX_W;
+
+//handle instruction  reg_vec1 = op(reg_vec2,reg_vec3,imm)
+//                               op(reg_vec2,mem_vec,imm)
+//                               op(reg_vec2,broadcast(eltVt),imm)
+//all instruction created with FROUND_CURRENT
+multiclass avx512_fp_packed_imm<bits<8> opc, string OpcodeStr, SDNode OpNode,
+                                                            X86VectorVTInfo _>{
+  defm rri : AVX512_maskable<opc, MRMSrcReg, _, (outs _.RC:$dst),
+                      (ins _.RC:$src1, _.RC:$src2, u8imm:$src3),
+                      OpcodeStr, "$src3, $src2, $src1", "$src1, $src2, $src3",
+                      (OpNode (_.VT _.RC:$src1),
+                              (_.VT _.RC:$src2),
+                              (i8 imm:$src3),
+                              (i32 FROUND_CURRENT))>;
+  let mayLoad = 1 in {
+    defm rmi : AVX512_maskable<opc, MRMSrcMem, _, (outs _.RC:$dst),
+                      (ins _.RC:$src1, _.MemOp:$src2, u8imm:$src3),
+                      OpcodeStr, "$src3, $src2, $src1", "$src1, $src2, $src3",
+                      (OpNode (_.VT _.RC:$src1),
+                              (_.VT (bitconvert (_.LdFrag addr:$src2))),
+                              (i8 imm:$src3),
+                              (i32 FROUND_CURRENT))>;
+    defm rmbi : AVX512_maskable<opc, MRMSrcMem, _, (outs _.RC:$dst),
+                      (ins _.RC:$src1, _.ScalarMemOp:$src2, u8imm:$src3),
+                      OpcodeStr, "$src3, ${src2}"##_.BroadcastStr##", $src1",
+                      "$src1, ${src2}"##_.BroadcastStr##", $src3",
+                      (OpNode (_.VT _.RC:$src1),
+                              (_.VT (X86VBroadcast(_.ScalarLdFrag addr:$src2))),
+                              (i8 imm:$src3),
+                              (i32 FROUND_CURRENT))>, EVEX_B;
+  }
+}
+
+//handle instruction  reg_vec1 = op(reg_vec2,reg_vec3,imm),{sae}
+multiclass avx512_fp_sae_packed_imm<bits<8> opc, string OpcodeStr,
+                                             SDNode OpNode, X86VectorVTInfo _>{
+  defm rrib : AVX512_maskable<opc, MRMSrcReg, _, (outs _.RC:$dst),
+                      (ins _.RC:$src1, _.RC:$src2, u8imm:$src3),
+                      OpcodeStr, "$src3,{sae}, $src2, $src1",
+                      "$src1, $src2,{sae}, $src3",
+                      (OpNode (_.VT _.RC:$src1),
+                              (_.VT _.RC:$src2),
+                              (i8 imm:$src3),
+                              (i32 FROUND_NO_EXC))>, EVEX_B;
+}
+
+multiclass avx512_vfixupimm_float<string OpcodeStr, AVX512VLVectorVTInfo _,
+                          bits<8> opc = 0x54, SDNode OpNode = X86VFixupimm>{
+    let Predicates = [HasAVX512] in {
+      defm Z    : avx512_fp_packed_imm<opc, OpcodeStr, OpNode, _.info512>,
+                  avx512_fp_sae_packed_imm<opc, OpcodeStr, OpNode, _.info512>,
+                                  EVEX_V512;
+
+    }
+    let Predicates = [HasAVX512, HasVLX] in {
+     defm Z128 : avx512_fp_packed_imm<opc, OpcodeStr, OpNode, _.info128>,
+                                  EVEX_V128;
+     defm Z256 : avx512_fp_packed_imm<opc, OpcodeStr, OpNode, _.info256>,
+                                  EVEX_V256;
+    }
+}
+
+defm VFIXUPIMMPD : avx512_vfixupimm_float<"vfixupimmpd", avx512vl_f64_info>,
+      AVX512AIi8Base, EVEX_4V, EVEX_CD8<64, CD8VF>, VEX_W;
+defm VFIXUPIMMPS : avx512_vfixupimm_float<"vfixupimmps", avx512vl_f32_info>,
+      AVX512AIi8Base, EVEX_4V, EVEX_CD8<32, CD8VF>;
+

diff --git a/lib/Target/X86/X86InstrFragmentsSIMD.td b/lib/Target/X86/X86InstrFragmentsSIMD.td
index 05dbf477bb3ce74ed05d4bdae893d26bbd40ad8b..0b91f79bf3bb5594c47c13db1802462d914d95bd 100644 (file)
--- a/lib/Target/X86/X86InstrFragmentsSIMD.td
+++ b/lib/Target/X86/X86InstrFragmentsSIMD.td
@@ -222,6 +222,8 @@ def SDTShuff2OpI : SDTypeProfile<1, 2, [SDTCisVec<0>,
                                  SDTCisSameAs<0,1>, SDTCisInt<2>]>;
 def SDTShuff3OpI : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0,1>,
                                  SDTCisSameAs<0,2>, SDTCisInt<3>]>;
+def SDTFPBinOpImmRound: SDTypeProfile<1, 4, [SDTCisVec<0>, SDTCisSameAs<0,1>,
+                             SDTCisSameAs<0,2>, SDTCisInt<3>, SDTCisInt<4>]>;
 
 def SDTVBroadcast  : SDTypeProfile<1, 1, [SDTCisVec<0>]>;
 def SDTVBroadcastm : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>]>;
@@ -282,6 +284,8 @@ def X86VPermiv3   : SDNode<"X86ISD::VPERMIV3",  SDTShuff3Op>;
 
 def X86VPerm2x128 : SDNode<"X86ISD::VPERM2X128", SDTShuff3OpI>;
 
+def X86VFixupimm       : SDNode<"X86ISD::VFIXUPIMM", SDTFPBinOpImmRound>;
+
 def X86SubVBroadcast : SDNode<"X86ISD::SUBV_BROADCAST",
                     SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVec<1>,
                                          SDTCisSubVecOfVec<1, 0>]>, []>;

diff --git a/test/MC/X86/avx512-encodings.s b/test/MC/X86/avx512-encodings.s
index cb71a00381b13bcbd338119714aa0a0f157b3d9b..fe6e943596c636d8109f9c7f0ac6b729a5739c9b 100644 (file)
--- a/test/MC/X86/avx512-encodings.s
+++ b/test/MC/X86/avx512-encodings.s
@@ -7724,3 +7724,138 @@ vpermilpd $0x23, 0x400(%rbx), %zmm2
 // CHECK:  encoding: [0x62,0xf1,0x95,0x50,0x72,0xa2,0xf8,0xfb,0xff,0xff,0x7b]
           vpsraq $123, -1032(%rdx){1to8}, %zmm29
 
+// CHECK: vfixupimmps $171, %zmm2, %zmm26, %zmm15
+// CHECK:  encoding: [0x62,0x73,0x2d,0x40,0x54,0xfa,0xab]
+          vfixupimmps $0xab, %zmm2, %zmm26, %zmm15
+
+// CHECK: vfixupimmps $171, %zmm2, %zmm26, %zmm15 {%k4}
+// CHECK:  encoding: [0x62,0x73,0x2d,0x44,0x54,0xfa,0xab]
+          vfixupimmps $0xab, %zmm2, %zmm26, %zmm15 {%k4}
+
+// CHECK: vfixupimmps $171, %zmm2, %zmm26, %zmm15 {%k4} {z}
+// CHECK:  encoding: [0x62,0x73,0x2d,0xc4,0x54,0xfa,0xab]
+          vfixupimmps $0xab, %zmm2, %zmm26, %zmm15 {%k4} {z}
+
+// CHECK: vfixupimmps $171,{sae}, %zmm2, %zmm26, %zmm15
+// CHECK:  encoding: [0x62,0x73,0x2d,0x10,0x54,0xfa,0xab]
+          vfixupimmps $0xab,{sae}, %zmm2, %zmm26, %zmm15
+
+// CHECK: vfixupimmps $123, %zmm2, %zmm26, %zmm15
+// CHECK:  encoding: [0x62,0x73,0x2d,0x40,0x54,0xfa,0x7b]
+          vfixupimmps $0x7b, %zmm2, %zmm26, %zmm15
+
+// CHECK: vfixupimmps $123,{sae}, %zmm2, %zmm26, %zmm15
+// CHECK:  encoding: [0x62,0x73,0x2d,0x10,0x54,0xfa,0x7b]
+          vfixupimmps $0x7b,{sae}, %zmm2, %zmm26, %zmm15
+
+// CHECK: vfixupimmps $123, (%rcx), %zmm26, %zmm15
+// CHECK:  encoding: [0x62,0x73,0x2d,0x40,0x54,0x39,0x7b]
+          vfixupimmps $0x7b, (%rcx), %zmm26, %zmm15
+
+// CHECK: vfixupimmps $123, 291(%rax,%r14,8), %zmm26, %zmm15
+// CHECK:  encoding: [0x62,0x33,0x2d,0x40,0x54,0xbc,0xf0,0x23,0x01,0x00,0x00,0x7b]
+          vfixupimmps $0x7b, 291(%rax,%r14,8), %zmm26, %zmm15
+
+// CHECK: vfixupimmps $123, (%rcx){1to16}, %zmm26, %zmm15
+// CHECK:  encoding: [0x62,0x73,0x2d,0x50,0x54,0x39,0x7b]
+          vfixupimmps $0x7b, (%rcx){1to16}, %zmm26, %zmm15
+
+// CHECK: vfixupimmps $123, 8128(%rdx), %zmm26, %zmm15
+// CHECK:  encoding: [0x62,0x73,0x2d,0x40,0x54,0x7a,0x7f,0x7b]
+          vfixupimmps $0x7b, 8128(%rdx), %zmm26, %zmm15
+
+// CHECK: vfixupimmps $123, 8192(%rdx), %zmm26, %zmm15
+// CHECK:  encoding: [0x62,0x73,0x2d,0x40,0x54,0xba,0x00,0x20,0x00,0x00,0x7b]
+          vfixupimmps $0x7b, 8192(%rdx), %zmm26, %zmm15
+
+// CHECK: vfixupimmps $123, -8192(%rdx), %zmm26, %zmm15
+// CHECK:  encoding: [0x62,0x73,0x2d,0x40,0x54,0x7a,0x80,0x7b]
+          vfixupimmps $0x7b, -8192(%rdx), %zmm26, %zmm15
+
+// CHECK: vfixupimmps $123, -8256(%rdx), %zmm26, %zmm15
+// CHECK:  encoding: [0x62,0x73,0x2d,0x40,0x54,0xba,0xc0,0xdf,0xff,0xff,0x7b]
+          vfixupimmps $0x7b, -8256(%rdx), %zmm26, %zmm15
+
+// CHECK: vfixupimmps $123, 508(%rdx){1to16}, %zmm26, %zmm15
+// CHECK:  encoding: [0x62,0x73,0x2d,0x50,0x54,0x7a,0x7f,0x7b]
+          vfixupimmps $0x7b, 508(%rdx){1to16}, %zmm26, %zmm15
+
+// CHECK: vfixupimmps $123, 512(%rdx){1to16}, %zmm26, %zmm15
+// CHECK:  encoding: [0x62,0x73,0x2d,0x50,0x54,0xba,0x00,0x02,0x00,0x00,0x7b]
+          vfixupimmps $0x7b, 512(%rdx){1to16}, %zmm26, %zmm15
+
+// CHECK: vfixupimmps $123, -512(%rdx){1to16}, %zmm26, %zmm15
+// CHECK:  encoding: [0x62,0x73,0x2d,0x50,0x54,0x7a,0x80,0x7b]
+          vfixupimmps $0x7b, -512(%rdx){1to16}, %zmm26, %zmm15
+
+// CHECK: vfixupimmps $123, -516(%rdx){1to16}, %zmm26, %zmm15
+// CHECK:  encoding: [0x62,0x73,0x2d,0x50,0x54,0xba,0xfc,0xfd,0xff,0xff,0x7b]
+          vfixupimmps $0x7b, -516(%rdx){1to16}, %zmm26, %zmm15
+
+// CHECK: vfixupimmpd $171, %zmm19, %zmm21, %zmm9
+// CHECK:  encoding: [0x62,0x33,0xd5,0x40,0x54,0xcb,0xab]
+          vfixupimmpd $0xab, %zmm19, %zmm21, %zmm9
+
+// CHECK: vfixupimmpd $171, %zmm19, %zmm21, %zmm9 {%k2}
+// CHECK:  encoding: [0x62,0x33,0xd5,0x42,0x54,0xcb,0xab]
+          vfixupimmpd $0xab, %zmm19, %zmm21, %zmm9 {%k2}
+
+// CHECK: vfixupimmpd $171, %zmm19, %zmm21, %zmm9 {%k2} {z}
+// CHECK:  encoding: [0x62,0x33,0xd5,0xc2,0x54,0xcb,0xab]
+          vfixupimmpd $0xab, %zmm19, %zmm21, %zmm9 {%k2} {z}
+
+// CHECK: vfixupimmpd $171,{sae}, %zmm19, %zmm21, %zmm9
+// CHECK:  encoding: [0x62,0x33,0xd5,0x10,0x54,0xcb,0xab]
+          vfixupimmpd $0xab,{sae}, %zmm19, %zmm21, %zmm9
+
+// CHECK: vfixupimmpd $123, %zmm19, %zmm21, %zmm9
+// CHECK:  encoding: [0x62,0x33,0xd5,0x40,0x54,0xcb,0x7b]
+          vfixupimmpd $0x7b, %zmm19, %zmm21, %zmm9
+
+// CHECK: vfixupimmpd $123,{sae}, %zmm19, %zmm21, %zmm9
+// CHECK:  encoding: [0x62,0x33,0xd5,0x10,0x54,0xcb,0x7b]
+          vfixupimmpd $0x7b,{sae}, %zmm19, %zmm21, %zmm9
+
+// CHECK: vfixupimmpd $123, (%rcx), %zmm21, %zmm9
+// CHECK:  encoding: [0x62,0x73,0xd5,0x40,0x54,0x09,0x7b]
+          vfixupimmpd $0x7b, (%rcx), %zmm21, %zmm9
+
+// CHECK: vfixupimmpd $123, 291(%rax,%r14,8), %zmm21, %zmm9
+// CHECK:  encoding: [0x62,0x33,0xd5,0x40,0x54,0x8c,0xf0,0x23,0x01,0x00,0x00,0x7b]
+          vfixupimmpd $0x7b, 291(%rax,%r14,8), %zmm21, %zmm9
+
+// CHECK: vfixupimmpd $123, (%rcx){1to8}, %zmm21, %zmm9
+// CHECK:  encoding: [0x62,0x73,0xd5,0x50,0x54,0x09,0x7b]
+          vfixupimmpd $0x7b, (%rcx){1to8}, %zmm21, %zmm9
+
+// CHECK: vfixupimmpd $123, 8128(%rdx), %zmm21, %zmm9
+// CHECK:  encoding: [0x62,0x73,0xd5,0x40,0x54,0x4a,0x7f,0x7b]
+          vfixupimmpd $0x7b, 8128(%rdx), %zmm21, %zmm9
+
+// CHECK: vfixupimmpd $123, 8192(%rdx), %zmm21, %zmm9
+// CHECK:  encoding: [0x62,0x73,0xd5,0x40,0x54,0x8a,0x00,0x20,0x00,0x00,0x7b]
+          vfixupimmpd $0x7b, 8192(%rdx), %zmm21, %zmm9
+
+// CHECK: vfixupimmpd $123, -8192(%rdx), %zmm21, %zmm9
+// CHECK:  encoding: [0x62,0x73,0xd5,0x40,0x54,0x4a,0x80,0x7b]
+          vfixupimmpd $0x7b, -8192(%rdx), %zmm21, %zmm9
+
+// CHECK: vfixupimmpd $123, -8256(%rdx), %zmm21, %zmm9
+// CHECK:  encoding: [0x62,0x73,0xd5,0x40,0x54,0x8a,0xc0,0xdf,0xff,0xff,0x7b]
+          vfixupimmpd $0x7b, -8256(%rdx), %zmm21, %zmm9
+
+// CHECK: vfixupimmpd $123, 1016(%rdx){1to8}, %zmm21, %zmm9
+// CHECK:  encoding: [0x62,0x73,0xd5,0x50,0x54,0x4a,0x7f,0x7b]
+          vfixupimmpd $0x7b, 1016(%rdx){1to8}, %zmm21, %zmm9
+
+// CHECK: vfixupimmpd $123, 1024(%rdx){1to8}, %zmm21, %zmm9
+// CHECK:  encoding: [0x62,0x73,0xd5,0x50,0x54,0x8a,0x00,0x04,0x00,0x00,0x7b]
+          vfixupimmpd $0x7b, 1024(%rdx){1to8}, %zmm21, %zmm9
+
+// CHECK: vfixupimmpd $123, -1024(%rdx){1to8}, %zmm21, %zmm9
+// CHECK:  encoding: [0x62,0x73,0xd5,0x50,0x54,0x4a,0x80,0x7b]
+          vfixupimmpd $0x7b, -1024(%rdx){1to8}, %zmm21, %zmm9
+
+// CHECK: vfixupimmpd $123, -1032(%rdx){1to8}, %zmm21, %zmm9
+// CHECK:  encoding: [0x62,0x73,0xd5,0x50,0x54,0x8a,0xf8,0xfb,0xff,0xff,0x7b]
+          vfixupimmpd $0x7b, -1032(%rdx){1to8}, %zmm21, %zmm9

diff --git a/test/MC/X86/x86-64-avx512f_vl.s b/test/MC/X86/x86-64-avx512f_vl.s
index 5007726817076ee88fc6c283569e6af9a834038a..137c327ffb051d0216f35ccea8aebea4e4408e9e 100644 (file)
--- a/test/MC/X86/x86-64-avx512f_vl.s
+++ b/test/MC/X86/x86-64-avx512f_vl.s
@@ -10188,6 +10188,246 @@ vaddpd  {rz-sae}, %zmm2, %zmm1, %zmm1
 // CHECK:  encoding: [0x62,0xe2,0x7d,0x28,0x5a,0x9a,0xf0,0xf7,0xff,0xff]
           vbroadcasti32x4 -2064(%rdx), %ymm19
 
+// CHECK: vfixupimmps $171, %xmm17, %xmm17, %xmm25
+// CHECK:  encoding: [0x62,0x23,0x75,0x00,0x54,0xc9,0xab]
+          vfixupimmps $0xab, %xmm17, %xmm17, %xmm25
+
+// CHECK: vfixupimmps $171, %xmm17, %xmm17, %xmm25 {%k3}
+// CHECK:  encoding: [0x62,0x23,0x75,0x03,0x54,0xc9,0xab]
+          vfixupimmps $0xab, %xmm17, %xmm17, %xmm25 {%k3}
+
+// CHECK: vfixupimmps $171, %xmm17, %xmm17, %xmm25 {%k3} {z}
+// CHECK:  encoding: [0x62,0x23,0x75,0x83,0x54,0xc9,0xab]
+          vfixupimmps $0xab, %xmm17, %xmm17, %xmm25 {%k3} {z}
+
+// CHECK: vfixupimmps $123, %xmm17, %xmm17, %xmm25
+// CHECK:  encoding: [0x62,0x23,0x75,0x00,0x54,0xc9,0x7b]
+          vfixupimmps $0x7b, %xmm17, %xmm17, %xmm25
+
+// CHECK: vfixupimmps $123, (%rcx), %xmm17, %xmm25
+// CHECK:  encoding: [0x62,0x63,0x75,0x00,0x54,0x09,0x7b]
+          vfixupimmps $0x7b, (%rcx), %xmm17, %xmm25
+
+// CHECK: vfixupimmps $123, 291(%rax,%r14,8), %xmm17, %xmm25
+// CHECK:  encoding: [0x62,0x23,0x75,0x00,0x54,0x8c,0xf0,0x23,0x01,0x00,0x00,0x7b]
+          vfixupimmps $0x7b, 291(%rax,%r14,8), %xmm17, %xmm25
+
+// CHECK: vfixupimmps $123, (%rcx){1to4}, %xmm17, %xmm25
+// CHECK:  encoding: [0x62,0x63,0x75,0x10,0x54,0x09,0x7b]
+          vfixupimmps $0x7b, (%rcx){1to4}, %xmm17, %xmm25
+
+// CHECK: vfixupimmps $123, 2032(%rdx), %xmm17, %xmm25
+// CHECK:  encoding: [0x62,0x63,0x75,0x00,0x54,0x4a,0x7f,0x7b]
+          vfixupimmps $0x7b, 2032(%rdx), %xmm17, %xmm25
+
+// CHECK: vfixupimmps $123, 2048(%rdx), %xmm17, %xmm25
+// CHECK:  encoding: [0x62,0x63,0x75,0x00,0x54,0x8a,0x00,0x08,0x00,0x00,0x7b]
+          vfixupimmps $0x7b, 2048(%rdx), %xmm17, %xmm25
+
+// CHECK: vfixupimmps $123, -2048(%rdx), %xmm17, %xmm25
+// CHECK:  encoding: [0x62,0x63,0x75,0x00,0x54,0x4a,0x80,0x7b]
+          vfixupimmps $0x7b, -2048(%rdx), %xmm17, %xmm25
+
+// CHECK: vfixupimmps $123, -2064(%rdx), %xmm17, %xmm25
+// CHECK:  encoding: [0x62,0x63,0x75,0x00,0x54,0x8a,0xf0,0xf7,0xff,0xff,0x7b]
+          vfixupimmps $0x7b, -2064(%rdx), %xmm17, %xmm25
+
+// CHECK: vfixupimmps $123, 508(%rdx){1to4}, %xmm17, %xmm25
+// CHECK:  encoding: [0x62,0x63,0x75,0x10,0x54,0x4a,0x7f,0x7b]
+          vfixupimmps $0x7b, 508(%rdx){1to4}, %xmm17, %xmm25
+
+// CHECK: vfixupimmps $123, 512(%rdx){1to4}, %xmm17, %xmm25
+// CHECK:  encoding: [0x62,0x63,0x75,0x10,0x54,0x8a,0x00,0x02,0x00,0x00,0x7b]
+          vfixupimmps $0x7b, 512(%rdx){1to4}, %xmm17, %xmm25
+
+// CHECK: vfixupimmps $123, -512(%rdx){1to4}, %xmm17, %xmm25
+// CHECK:  encoding: [0x62,0x63,0x75,0x10,0x54,0x4a,0x80,0x7b]
+          vfixupimmps $0x7b, -512(%rdx){1to4}, %xmm17, %xmm25
+
+// CHECK: vfixupimmps $123, -516(%rdx){1to4}, %xmm17, %xmm25
+// CHECK:  encoding: [0x62,0x63,0x75,0x10,0x54,0x8a,0xfc,0xfd,0xff,0xff,0x7b]
+          vfixupimmps $0x7b, -516(%rdx){1to4}, %xmm17, %xmm25
+
+// CHECK: vfixupimmps $171, %ymm28, %ymm21, %ymm30
+// CHECK:  encoding: [0x62,0x03,0x55,0x20,0x54,0xf4,0xab]
+          vfixupimmps $0xab, %ymm28, %ymm21, %ymm30
+
+// CHECK: vfixupimmps $171, %ymm28, %ymm21, %ymm30 {%k4}
+// CHECK:  encoding: [0x62,0x03,0x55,0x24,0x54,0xf4,0xab]
+          vfixupimmps $0xab, %ymm28, %ymm21, %ymm30 {%k4}
+
+// CHECK: vfixupimmps $171, %ymm28, %ymm21, %ymm30 {%k4} {z}
+// CHECK:  encoding: [0x62,0x03,0x55,0xa4,0x54,0xf4,0xab]
+          vfixupimmps $0xab, %ymm28, %ymm21, %ymm30 {%k4} {z}
+
+// CHECK: vfixupimmps $123, %ymm28, %ymm21, %ymm30
+// CHECK:  encoding: [0x62,0x03,0x55,0x20,0x54,0xf4,0x7b]
+          vfixupimmps $0x7b, %ymm28, %ymm21, %ymm30
+
+// CHECK: vfixupimmps $123, (%rcx), %ymm21, %ymm30
+// CHECK:  encoding: [0x62,0x63,0x55,0x20,0x54,0x31,0x7b]
+          vfixupimmps $0x7b, (%rcx), %ymm21, %ymm30
+
+// CHECK: vfixupimmps $123, 291(%rax,%r14,8), %ymm21, %ymm30
+// CHECK:  encoding: [0x62,0x23,0x55,0x20,0x54,0xb4,0xf0,0x23,0x01,0x00,0x00,0x7b]
+          vfixupimmps $0x7b, 291(%rax,%r14,8), %ymm21, %ymm30
+
+// CHECK: vfixupimmps $123, (%rcx){1to8}, %ymm21, %ymm30
+// CHECK:  encoding: [0x62,0x63,0x55,0x30,0x54,0x31,0x7b]
+          vfixupimmps $0x7b, (%rcx){1to8}, %ymm21, %ymm30
+
+// CHECK: vfixupimmps $123, 4064(%rdx), %ymm21, %ymm30
+// CHECK:  encoding: [0x62,0x63,0x55,0x20,0x54,0x72,0x7f,0x7b]
+          vfixupimmps $0x7b, 4064(%rdx), %ymm21, %ymm30
+
+// CHECK: vfixupimmps $123, 4096(%rdx), %ymm21, %ymm30
+// CHECK:  encoding: [0x62,0x63,0x55,0x20,0x54,0xb2,0x00,0x10,0x00,0x00,0x7b]
+          vfixupimmps $0x7b, 4096(%rdx), %ymm21, %ymm30
+
+// CHECK: vfixupimmps $123, -4096(%rdx), %ymm21, %ymm30
+// CHECK:  encoding: [0x62,0x63,0x55,0x20,0x54,0x72,0x80,0x7b]
+          vfixupimmps $0x7b, -4096(%rdx), %ymm21, %ymm30
+
+// CHECK: vfixupimmps $123, -4128(%rdx), %ymm21, %ymm30
+// CHECK:  encoding: [0x62,0x63,0x55,0x20,0x54,0xb2,0xe0,0xef,0xff,0xff,0x7b]
+          vfixupimmps $0x7b, -4128(%rdx), %ymm21, %ymm30
+
+// CHECK: vfixupimmps $123, 508(%rdx){1to8}, %ymm21, %ymm30
+// CHECK:  encoding: [0x62,0x63,0x55,0x30,0x54,0x72,0x7f,0x7b]
+          vfixupimmps $0x7b, 508(%rdx){1to8}, %ymm21, %ymm30
+
+// CHECK: vfixupimmps $123, 512(%rdx){1to8}, %ymm21, %ymm30
+// CHECK:  encoding: [0x62,0x63,0x55,0x30,0x54,0xb2,0x00,0x02,0x00,0x00,0x7b]
+          vfixupimmps $0x7b, 512(%rdx){1to8}, %ymm21, %ymm30
+
+// CHECK: vfixupimmps $123, -512(%rdx){1to8}, %ymm21, %ymm30
+// CHECK:  encoding: [0x62,0x63,0x55,0x30,0x54,0x72,0x80,0x7b]
+          vfixupimmps $0x7b, -512(%rdx){1to8}, %ymm21, %ymm30
+
+// CHECK: vfixupimmps $123, -516(%rdx){1to8}, %ymm21, %ymm30
+// CHECK:  encoding: [0x62,0x63,0x55,0x30,0x54,0xb2,0xfc,0xfd,0xff,0xff,0x7b]
+          vfixupimmps $0x7b, -516(%rdx){1to8}, %ymm21, %ymm30
+
+// CHECK: vfixupimmpd $171, %xmm25, %xmm18, %xmm24
+// CHECK:  encoding: [0x62,0x03,0xed,0x00,0x54,0xc1,0xab]
+          vfixupimmpd $0xab, %xmm25, %xmm18, %xmm24
+
+// CHECK: vfixupimmpd $171, %xmm25, %xmm18, %xmm24 {%k3}
+// CHECK:  encoding: [0x62,0x03,0xed,0x03,0x54,0xc1,0xab]
+          vfixupimmpd $0xab, %xmm25, %xmm18, %xmm24 {%k3}
+
+// CHECK: vfixupimmpd $171, %xmm25, %xmm18, %xmm24 {%k3} {z}
+// CHECK:  encoding: [0x62,0x03,0xed,0x83,0x54,0xc1,0xab]
+          vfixupimmpd $0xab, %xmm25, %xmm18, %xmm24 {%k3} {z}
+
+// CHECK: vfixupimmpd $123, %xmm25, %xmm18, %xmm24
+// CHECK:  encoding: [0x62,0x03,0xed,0x00,0x54,0xc1,0x7b]
+          vfixupimmpd $0x7b, %xmm25, %xmm18, %xmm24
+
+// CHECK: vfixupimmpd $123, (%rcx), %xmm18, %xmm24
+// CHECK:  encoding: [0x62,0x63,0xed,0x00,0x54,0x01,0x7b]
+          vfixupimmpd $0x7b, (%rcx), %xmm18, %xmm24
+
+// CHECK: vfixupimmpd $123, 291(%rax,%r14,8), %xmm18, %xmm24
+// CHECK:  encoding: [0x62,0x23,0xed,0x00,0x54,0x84,0xf0,0x23,0x01,0x00,0x00,0x7b]
+          vfixupimmpd $0x7b, 291(%rax,%r14,8), %xmm18, %xmm24
+
+// CHECK: vfixupimmpd $123, (%rcx){1to2}, %xmm18, %xmm24
+// CHECK:  encoding: [0x62,0x63,0xed,0x10,0x54,0x01,0x7b]
+          vfixupimmpd $0x7b, (%rcx){1to2}, %xmm18, %xmm24
+
+// CHECK: vfixupimmpd $123, 2032(%rdx), %xmm18, %xmm24
+// CHECK:  encoding: [0x62,0x63,0xed,0x00,0x54,0x42,0x7f,0x7b]
+          vfixupimmpd $0x7b, 2032(%rdx), %xmm18, %xmm24
+
+// CHECK: vfixupimmpd $123, 2048(%rdx), %xmm18, %xmm24
+// CHECK:  encoding: [0x62,0x63,0xed,0x00,0x54,0x82,0x00,0x08,0x00,0x00,0x7b]
+          vfixupimmpd $0x7b, 2048(%rdx), %xmm18, %xmm24
+
+// CHECK: vfixupimmpd $123, -2048(%rdx), %xmm18, %xmm24
+// CHECK:  encoding: [0x62,0x63,0xed,0x00,0x54,0x42,0x80,0x7b]
+          vfixupimmpd $0x7b, -2048(%rdx), %xmm18, %xmm24
+
+// CHECK: vfixupimmpd $123, -2064(%rdx), %xmm18, %xmm24
+// CHECK:  encoding: [0x62,0x63,0xed,0x00,0x54,0x82,0xf0,0xf7,0xff,0xff,0x7b]
+          vfixupimmpd $0x7b, -2064(%rdx), %xmm18, %xmm24
+
+// CHECK: vfixupimmpd $123, 1016(%rdx){1to2}, %xmm18, %xmm24
+// CHECK:  encoding: [0x62,0x63,0xed,0x10,0x54,0x42,0x7f,0x7b]
+          vfixupimmpd $0x7b, 1016(%rdx){1to2}, %xmm18, %xmm24
+
+// CHECK: vfixupimmpd $123, 1024(%rdx){1to2}, %xmm18, %xmm24
+// CHECK:  encoding: [0x62,0x63,0xed,0x10,0x54,0x82,0x00,0x04,0x00,0x00,0x7b]
+          vfixupimmpd $0x7b, 1024(%rdx){1to2}, %xmm18, %xmm24
+
+// CHECK: vfixupimmpd $123, -1024(%rdx){1to2}, %xmm18, %xmm24
+// CHECK:  encoding: [0x62,0x63,0xed,0x10,0x54,0x42,0x80,0x7b]
+          vfixupimmpd $0x7b, -1024(%rdx){1to2}, %xmm18, %xmm24
+
+// CHECK: vfixupimmpd $123, -1032(%rdx){1to2}, %xmm18, %xmm24
+// CHECK:  encoding: [0x62,0x63,0xed,0x10,0x54,0x82,0xf8,0xfb,0xff,0xff,0x7b]
+          vfixupimmpd $0x7b, -1032(%rdx){1to2}, %xmm18, %xmm24
+
+// CHECK: vfixupimmpd $171, %ymm28, %ymm22, %ymm18
+// CHECK:  encoding: [0x62,0x83,0xcd,0x20,0x54,0xd4,0xab]
+          vfixupimmpd $0xab, %ymm28, %ymm22, %ymm18
+
+// CHECK: vfixupimmpd $171, %ymm28, %ymm22, %ymm18 {%k2}
+// CHECK:  encoding: [0x62,0x83,0xcd,0x22,0x54,0xd4,0xab]
+          vfixupimmpd $0xab, %ymm28, %ymm22, %ymm18 {%k2}
+
+// CHECK: vfixupimmpd $171, %ymm28, %ymm22, %ymm18 {%k2} {z}
+// CHECK:  encoding: [0x62,0x83,0xcd,0xa2,0x54,0xd4,0xab]
+          vfixupimmpd $0xab, %ymm28, %ymm22, %ymm18 {%k2} {z}
+
+// CHECK: vfixupimmpd $123, %ymm28, %ymm22, %ymm18
+// CHECK:  encoding: [0x62,0x83,0xcd,0x20,0x54,0xd4,0x7b]
+          vfixupimmpd $0x7b, %ymm28, %ymm22, %ymm18
+
+// CHECK: vfixupimmpd $123, (%rcx), %ymm22, %ymm18
+// CHECK:  encoding: [0x62,0xe3,0xcd,0x20,0x54,0x11,0x7b]
+          vfixupimmpd $0x7b, (%rcx), %ymm22, %ymm18
+
+// CHECK: vfixupimmpd $123, 291(%rax,%r14,8), %ymm22, %ymm18
+// CHECK:  encoding: [0x62,0xa3,0xcd,0x20,0x54,0x94,0xf0,0x23,0x01,0x00,0x00,0x7b]
+          vfixupimmpd $0x7b, 291(%rax,%r14,8), %ymm22, %ymm18
+
+// CHECK: vfixupimmpd $123, (%rcx){1to4}, %ymm22, %ymm18
+// CHECK:  encoding: [0x62,0xe3,0xcd,0x30,0x54,0x11,0x7b]
+          vfixupimmpd $0x7b, (%rcx){1to4}, %ymm22, %ymm18
+
+// CHECK: vfixupimmpd $123, 4064(%rdx), %ymm22, %ymm18
+// CHECK:  encoding: [0x62,0xe3,0xcd,0x20,0x54,0x52,0x7f,0x7b]
+          vfixupimmpd $0x7b, 4064(%rdx), %ymm22, %ymm18
+
+// CHECK: vfixupimmpd $123, 4096(%rdx), %ymm22, %ymm18
+// CHECK:  encoding: [0x62,0xe3,0xcd,0x20,0x54,0x92,0x00,0x10,0x00,0x00,0x7b]
+          vfixupimmpd $0x7b, 4096(%rdx), %ymm22, %ymm18
+
+// CHECK: vfixupimmpd $123, -4096(%rdx), %ymm22, %ymm18
+// CHECK:  encoding: [0x62,0xe3,0xcd,0x20,0x54,0x52,0x80,0x7b]
+          vfixupimmpd $0x7b, -4096(%rdx), %ymm22, %ymm18
+
+// CHECK: vfixupimmpd $123, -4128(%rdx), %ymm22, %ymm18
+// CHECK:  encoding: [0x62,0xe3,0xcd,0x20,0x54,0x92,0xe0,0xef,0xff,0xff,0x7b]
+          vfixupimmpd $0x7b, -4128(%rdx), %ymm22, %ymm18
+
+// CHECK: vfixupimmpd $123, 1016(%rdx){1to4}, %ymm22, %ymm18
+// CHECK:  encoding: [0x62,0xe3,0xcd,0x30,0x54,0x52,0x7f,0x7b]
+          vfixupimmpd $0x7b, 1016(%rdx){1to4}, %ymm22, %ymm18
+
+// CHECK: vfixupimmpd $123, 1024(%rdx){1to4}, %ymm22, %ymm18
+// CHECK:  encoding: [0x62,0xe3,0xcd,0x30,0x54,0x92,0x00,0x04,0x00,0x00,0x7b]
+          vfixupimmpd $0x7b, 1024(%rdx){1to4}, %ymm22, %ymm18
+
+// CHECK: vfixupimmpd $123, -1024(%rdx){1to4}, %ymm22, %ymm18
+// CHECK:  encoding: [0x62,0xe3,0xcd,0x30,0x54,0x52,0x80,0x7b]
+          vfixupimmpd $0x7b, -1024(%rdx){1to4}, %ymm22, %ymm18
+
+// CHECK: vfixupimmpd $123, -1032(%rdx){1to4}, %ymm22, %ymm18
+// CHECK:  encoding: [0x62,0xe3,0xcd,0x30,0x54,0x92,0xf8,0xfb,0xff,0xff,0x7b]
+          vfixupimmpd $0x7b, -1032(%rdx){1to4}, %ymm22, %ymm18
+