[SKX] Extended non-temporal load/store instructions for AVX512VL subsets.
authorRobert Khasanov <rob.khasanov@gmail.com>
Wed, 13 Aug 2014 10:46:00 +0000 (10:46 +0000)
committerRobert Khasanov <rob.khasanov@gmail.com>
Wed, 13 Aug 2014 10:46:00 +0000 (10:46 +0000)
Added avx512_movnt_vl multiclass for handling 256/128-bit forms of instruction.
Added encoding and lowering tests.

Reviewed by Elena Demikhovsky <elena.demikhovsky@intel.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@215536 91177308-0d34-0410-b5e6-96231b3b80d8

include/llvm/IR/IntrinsicsX86.td
lib/Target/X86/X86InstrAVX512.td
lib/Target/X86/X86InstrInfo.td
lib/Target/X86/X86InstrSSE.td
test/CodeGen/X86/avx512vl-nontemporal.ll [new file with mode: 0644]
test/MC/X86/avx512-encodings.s
test/MC/X86/x86-64-avx512f_vl.s

index 5f8ae27..1f5ad9d 100644 (file)
@@ -1954,8 +1954,6 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
                          llvm_i32_ty], [IntrNoMem, Commutative]>;
   def int_x86_avx2_movntdqa : GCCBuiltin<"__builtin_ia32_movntdqa256">,
               Intrinsic<[llvm_v4i64_ty], [llvm_ptr_ty], [IntrReadMem]>;
-  def int_x86_avx512_movntdqa : GCCBuiltin<"__builtin_ia32_movntdqa512">,
-              Intrinsic<[llvm_v8i64_ty], [llvm_ptr_ty], [IntrReadMem]>;
 }
 
 //===----------------------------------------------------------------------===//
@@ -3219,6 +3217,8 @@ let TargetPrefix = "x86" in {
             Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty,
                                         llvm_v8i64_ty, llvm_i8_ty],
                       [IntrNoMem]>;
+  def int_x86_avx512_movntdqa : GCCBuiltin<"__builtin_ia32_movntdqa512">,
+            Intrinsic<[llvm_v8i64_ty], [llvm_ptr_ty], [IntrReadMem]>;
 }
 
 //===----------------------------------------------------------------------===//
index 02be95f..3fe62d6 100644 (file)
@@ -2090,43 +2090,73 @@ def : Pat<(v8i64 (X86Vinsert undef, GR64:$src2, (iPTR 0))),
 //===----------------------------------------------------------------------===//
 // AVX-512 - Non-temporals
 //===----------------------------------------------------------------------===//
+let SchedRW = [WriteLoad] in {
+  def VMOVNTDQAZrm : AVX512PI<0x2A, MRMSrcMem, (outs VR512:$dst),
+                        (ins i512mem:$src), "vmovntdqa\t{$src, $dst|$dst, $src}",
+                        [(set VR512:$dst, (int_x86_avx512_movntdqa addr:$src))],
+                        SSEPackedInt>, EVEX, T8PD, EVEX_V512,
+                        EVEX_CD8<64, CD8VF>;
+
+  let Predicates = [HasAVX512, HasVLX] in {
+    def VMOVNTDQAZ256rm : AVX512PI<0x2A, MRMSrcMem, (outs VR256X:$dst),
+                             (ins i256mem:$src),
+                             "vmovntdqa\t{$src, $dst|$dst, $src}", [],
+                             SSEPackedInt>, EVEX, T8PD, EVEX_V256,
+                             EVEX_CD8<64, CD8VF>;
+
+    def VMOVNTDQAZ128rm : AVX512PI<0x2A, MRMSrcMem, (outs VR128X:$dst),
+                             (ins i128mem:$src),
+                             "vmovntdqa\t{$src, $dst|$dst, $src}", [],
+                             SSEPackedInt>, EVEX, T8PD, EVEX_V128,
+                             EVEX_CD8<64, CD8VF>;
+  }
+}
 
-def VMOVNTDQAZrm : AVX5128I<0x2A, MRMSrcMem, (outs VR512:$dst),
-                            (ins i512mem:$src),
-                            "vmovntdqa\t{$src, $dst|$dst, $src}",
-                            [(set VR512:$dst,
-                              (int_x86_avx512_movntdqa addr:$src))]>,
-                   EVEX, EVEX_V512, EVEX_CD8<64, CD8VF>;
-
-// Prefer non-temporal over temporal versions
-let AddedComplexity = 400, SchedRW = [WriteStore] in {
-
-def VMOVNTPSZmr : AVX512PSI<0x2B, MRMDestMem, (outs),
-                            (ins f512mem:$dst, VR512:$src),
-                            "vmovntps\t{$src, $dst|$dst, $src}",
-                            [(alignednontemporalstore (v16f32 VR512:$src),
-                                                      addr:$dst)],
-                            IIC_SSE_MOVNT>,
-                  EVEX, EVEX_V512, EVEX_CD8<32, CD8VF>;
-
-def VMOVNTPDZmr : AVX512PDI<0x2B, MRMDestMem, (outs),
-                            (ins f512mem:$dst, VR512:$src),
-                            "vmovntpd\t{$src, $dst|$dst, $src}",
-                            [(alignednontemporalstore (v8f64 VR512:$src),
-                                                      addr:$dst)],
-                           IIC_SSE_MOVNT>,
-                  EVEX, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
-
-
-def VMOVNTDQZmr : AVX512BI<0xE7, MRMDestMem, (outs),
-                           (ins i512mem:$dst, VR512:$src),
-                           "vmovntdq\t{$src, $dst|$dst, $src}",
-                           [(alignednontemporalstore (v8i64 VR512:$src),
-                                                     addr:$dst)],
-                           IIC_SSE_MOVNT>,
-                  EVEX, EVEX_V512, EVEX_CD8<64, CD8VF>;
+multiclass avx512_movnt<bits<8> opc, string OpcodeStr, PatFrag st_frag,
+                        ValueType OpVT, RegisterClass RC, X86MemOperand memop,
+                        Domain d, InstrItinClass itin = IIC_SSE_MOVNT> {
+  let SchedRW = [WriteStore], mayStore = 1,
+      AddedComplexity = 400 in
+  def mr : AVX512PI<opc, MRMDestMem, (outs), (ins memop:$dst, RC:$src),
+                    !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+                    [(st_frag (OpVT RC:$src), addr:$dst)], d, itin>, EVEX;
 }
 
+multiclass avx512_movnt_vl<bits<8> opc, string OpcodeStr, PatFrag st_frag,
+                           string elty, string elsz, string vsz512,
+                           string vsz256, string vsz128, Domain d,
+                           Predicate prd, InstrItinClass itin = IIC_SSE_MOVNT> {
+  let Predicates = [prd] in
+  defm Z : avx512_movnt<opc, OpcodeStr, st_frag,
+                        !cast<ValueType>("v"##vsz512##elty##elsz), VR512,
+                        !cast<X86MemOperand>(elty##"512mem"), d, itin>,
+                        EVEX_V512;
+
+  let Predicates = [prd, HasVLX] in {
+    defm Z256 : avx512_movnt<opc, OpcodeStr, st_frag,
+                             !cast<ValueType>("v"##vsz256##elty##elsz), VR256X,
+                             !cast<X86MemOperand>(elty##"256mem"), d, itin>,
+                             EVEX_V256;
+
+    defm Z128 : avx512_movnt<opc, OpcodeStr, st_frag,
+                             !cast<ValueType>("v"##vsz128##elty##elsz), VR128X,
+                             !cast<X86MemOperand>(elty##"128mem"), d, itin>,
+                             EVEX_V128;
+  }
+}
+
+defm VMOVNTDQ : avx512_movnt_vl<0xE7, "vmovntdq", alignednontemporalstore,
+                                "i", "64", "8", "4", "2", SSEPackedInt,
+                                HasAVX512>, PD, EVEX_CD8<64, CD8VF>;
+
+defm VMOVNTPD : avx512_movnt_vl<0x2B, "vmovntpd", alignednontemporalstore,
+                                "f", "64", "8", "4", "2", SSEPackedDouble,
+                                HasAVX512>, PD, VEX_W, EVEX_CD8<64, CD8VF>;
+
+defm VMOVNTPS : avx512_movnt_vl<0x2B, "vmovntps", alignednontemporalstore,
+                                "f", "32", "16", "8", "4", SSEPackedSingle,
+                                HasAVX512>, PS, EVEX_CD8<32, CD8VF>;
+
 //===----------------------------------------------------------------------===//
 // AVX-512 - Integer arithmetic
 //
index f2f53f3..b262ec2 100644 (file)
@@ -727,6 +727,7 @@ def HasDQI       : Predicate<"Subtarget->hasDQI()">;
 def HasBWI       : Predicate<"Subtarget->hasBWI()">;
 def HasVLX       : Predicate<"Subtarget->hasVLX()">,
                      AssemblerPredicate<"FeatureVLX", "AVX-512 VLX ISA">;
+def NoVLX        : Predicate<"!Subtarget->hasVLX()">;
 
 def HasPOPCNT    : Predicate<"Subtarget->hasPOPCNT()">;
 def HasAES       : Predicate<"Subtarget->hasAES()">;
index ea8b6c7..2189d14 100644 (file)
@@ -3697,6 +3697,7 @@ let Predicates = [UseSSE1] in {
 
 let AddedComplexity = 400 in { // Prefer non-temporal versions
 let SchedRW = [WriteStore] in {
+let Predicates = [HasAVX, NoVLX] in {
 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
                      (ins f128mem:$dst, VR128:$src),
                      "movntps\t{$src, $dst|$dst, $src}",
@@ -3737,6 +3738,7 @@ def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
                     [(alignednontemporalstore (v4i64 VR256:$src),
                                               addr:$dst)],
                                               IIC_SSE_MOVNT>, VEX, VEX_L;
+}
 
 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
                     "movntps\t{$src, $dst|$dst, $src}",
diff --git a/test/CodeGen/X86/avx512vl-nontemporal.ll b/test/CodeGen/X86/avx512vl-nontemporal.ll
new file mode 100644 (file)
index 0000000..2ad9768
--- /dev/null
@@ -0,0 +1,34 @@
+; RUN: llc < %s  -march=x86-64 -mtriple=x86_64-apple-darwin -mcpu=skx --show-mc-encoding | FileCheck %s
+
+define void @f256(<8 x float> %A, <8 x float> %AA, i8* %B, <4 x double> %C, <4 x double> %CC, i32 %D, <4 x i64> %E, <4 x i64> %EE) {
+; CHECK: vmovntps %ymm{{.*}} ## encoding: [0x62
+  %cast = bitcast i8* %B to <8 x float>*
+  %A2 = fadd <8 x float> %A, %AA
+  store <8 x float> %A2, <8 x float>* %cast, align 64, !nontemporal !0
+; CHECK: vmovntdq %ymm{{.*}} ## encoding: [0x62
+  %cast1 = bitcast i8* %B to <4 x i64>*
+  %E2 = add <4 x i64> %E, %EE
+  store <4 x i64> %E2, <4 x i64>* %cast1, align 64, !nontemporal !0
+; CHECK: vmovntpd %ymm{{.*}} ## encoding: [0x62
+  %cast2 = bitcast i8* %B to <4 x double>*
+  %C2 = fadd <4 x double> %C, %CC
+  store <4 x double> %C2, <4 x double>* %cast2, align 64, !nontemporal !0
+  ret void
+}
+
+define void @f128(<4 x float> %A, <4 x float> %AA, i8* %B, <2 x double> %C, <2 x double> %CC, i32 %D, <2 x i64> %E, <2 x i64> %EE) {
+; CHECK: vmovntps %xmm{{.*}} ## encoding: [0x62
+  %cast = bitcast i8* %B to <4 x float>*
+  %A2 = fadd <4 x float> %A, %AA
+  store <4 x float> %A2, <4 x float>* %cast, align 64, !nontemporal !0
+; CHECK: vmovntdq %xmm{{.*}} ## encoding: [0x62
+  %cast1 = bitcast i8* %B to <2 x i64>*
+  %E2 = add <2 x i64> %E, %EE
+  store <2 x i64> %E2, <2 x i64>* %cast1, align 64, !nontemporal !0
+; CHECK: vmovntpd %xmm{{.*}} ## encoding: [0x62
+  %cast2 = bitcast i8* %B to <2 x double>*
+  %C2 = fadd <2 x double> %C, %CC
+  store <2 x double> %C2, <2 x double>* %cast2, align 64, !nontemporal !0
+  ret void
+}
+!0 = metadata !{i32 1}
index a1a94fb..8cd7aa4 100644 (file)
 // CHECK:  encoding: [0x62,0xf1,0xfe,0x48,0x6f,0xb2,0xc0,0xdf,0xff,0xff]
           vmovdqu64 -8256(%rdx), %zmm6
 
+// CHECK: vmovntdq %zmm24, (%rcx)
+// CHECK:  encoding: [0x62,0x61,0x7d,0x48,0xe7,0x01]
+          vmovntdq %zmm24, (%rcx)
+
+// CHECK: vmovntdq %zmm24, 291(%rax,%r14,8)
+// CHECK:  encoding: [0x62,0x21,0x7d,0x48,0xe7,0x84,0xf0,0x23,0x01,0x00,0x00]
+          vmovntdq %zmm24, 291(%rax,%r14,8)
+
+// CHECK: vmovntdq %zmm24, 8128(%rdx)
+// CHECK:  encoding: [0x62,0x61,0x7d,0x48,0xe7,0x42,0x7f]
+          vmovntdq %zmm24, 8128(%rdx)
+
+// CHECK: vmovntdq %zmm24, 8192(%rdx)
+// CHECK:  encoding: [0x62,0x61,0x7d,0x48,0xe7,0x82,0x00,0x20,0x00,0x00]
+          vmovntdq %zmm24, 8192(%rdx)
+
+// CHECK: vmovntdq %zmm24, -8192(%rdx)
+// CHECK:  encoding: [0x62,0x61,0x7d,0x48,0xe7,0x42,0x80]
+          vmovntdq %zmm24, -8192(%rdx)
+
+// CHECK: vmovntdq %zmm24, -8256(%rdx)
+// CHECK:  encoding: [0x62,0x61,0x7d,0x48,0xe7,0x82,0xc0,0xdf,0xff,0xff]
+          vmovntdq %zmm24, -8256(%rdx)
+
+// CHECK: vmovntdqa (%rcx), %zmm17
+// CHECK:  encoding: [0x62,0xe2,0x7d,0x48,0x2a,0x09]
+          vmovntdqa (%rcx), %zmm17
+
+// CHECK: vmovntdqa 291(%rax,%r14,8), %zmm17
+// CHECK:  encoding: [0x62,0xa2,0x7d,0x48,0x2a,0x8c,0xf0,0x23,0x01,0x00,0x00]
+          vmovntdqa 291(%rax,%r14,8), %zmm17
+
+// CHECK: vmovntdqa 8128(%rdx), %zmm17
+// CHECK:  encoding: [0x62,0xe2,0x7d,0x48,0x2a,0x4a,0x7f]
+          vmovntdqa 8128(%rdx), %zmm17
+
+// CHECK: vmovntdqa 8192(%rdx), %zmm17
+// CHECK:  encoding: [0x62,0xe2,0x7d,0x48,0x2a,0x8a,0x00,0x20,0x00,0x00]
+          vmovntdqa 8192(%rdx), %zmm17
+
+// CHECK: vmovntdqa -8192(%rdx), %zmm17
+// CHECK:  encoding: [0x62,0xe2,0x7d,0x48,0x2a,0x4a,0x80]
+          vmovntdqa -8192(%rdx), %zmm17
+
+// CHECK: vmovntdqa -8256(%rdx), %zmm17
+// CHECK:  encoding: [0x62,0xe2,0x7d,0x48,0x2a,0x8a,0xc0,0xdf,0xff,0xff]
+          vmovntdqa -8256(%rdx), %zmm17
+
+// CHECK: vmovntpd %zmm17, (%rcx)
+// CHECK:  encoding: [0x62,0xe1,0xfd,0x48,0x2b,0x09]
+          vmovntpd %zmm17, (%rcx)
+
+// CHECK: vmovntpd %zmm17, 291(%rax,%r14,8)
+// CHECK:  encoding: [0x62,0xa1,0xfd,0x48,0x2b,0x8c,0xf0,0x23,0x01,0x00,0x00]
+          vmovntpd %zmm17, 291(%rax,%r14,8)
+
+// CHECK: vmovntpd %zmm17, 8128(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0xfd,0x48,0x2b,0x4a,0x7f]
+          vmovntpd %zmm17, 8128(%rdx)
+
+// CHECK: vmovntpd %zmm17, 8192(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0xfd,0x48,0x2b,0x8a,0x00,0x20,0x00,0x00]
+          vmovntpd %zmm17, 8192(%rdx)
+
+// CHECK: vmovntpd %zmm17, -8192(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0xfd,0x48,0x2b,0x4a,0x80]
+          vmovntpd %zmm17, -8192(%rdx)
+
+// CHECK: vmovntpd %zmm17, -8256(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0xfd,0x48,0x2b,0x8a,0xc0,0xdf,0xff,0xff]
+          vmovntpd %zmm17, -8256(%rdx)
+
+// CHECK: vmovntps %zmm5, (%rcx)
+// CHECK:  encoding: [0x62,0xf1,0x7c,0x48,0x2b,0x29]
+          vmovntps %zmm5, (%rcx)
+
+// CHECK: vmovntps %zmm5, 291(%rax,%r14,8)
+// CHECK:  encoding: [0x62,0xb1,0x7c,0x48,0x2b,0xac,0xf0,0x23,0x01,0x00,0x00]
+          vmovntps %zmm5, 291(%rax,%r14,8)
+
+// CHECK: vmovntps %zmm5, 8128(%rdx)
+// CHECK:  encoding: [0x62,0xf1,0x7c,0x48,0x2b,0x6a,0x7f]
+          vmovntps %zmm5, 8128(%rdx)
+
+// CHECK: vmovntps %zmm5, 8192(%rdx)
+// CHECK:  encoding: [0x62,0xf1,0x7c,0x48,0x2b,0xaa,0x00,0x20,0x00,0x00]
+          vmovntps %zmm5, 8192(%rdx)
+
+// CHECK: vmovntps %zmm5, -8192(%rdx)
+// CHECK:  encoding: [0x62,0xf1,0x7c,0x48,0x2b,0x6a,0x80]
+          vmovntps %zmm5, -8192(%rdx)
+
+// CHECK: vmovntps %zmm5, -8256(%rdx)
+// CHECK:  encoding: [0x62,0xf1,0x7c,0x48,0x2b,0xaa,0xc0,0xdf,0xff,0xff]
+          vmovntps %zmm5, -8256(%rdx)
+
 // CHECK: vmovupd %zmm9, %zmm27
 // CHECK:  encoding: [0x62,0x41,0xfd,0x48,0x10,0xd9]
           vmovupd %zmm9, %zmm27
index 83320df..b529c03 100644 (file)
 // CHECK:  encoding: [0x62,0x61,0xfe,0x28,0x6f,0xaa,0xe0,0xef,0xff,0xff]
           vmovdqu64 -4128(%rdx), %ymm29
 
+// CHECK: vmovntdq %xmm22, (%rcx)
+// CHECK:  encoding: [0x62,0xe1,0x7d,0x08,0xe7,0x31]
+          vmovntdq %xmm22, (%rcx)
+
+// CHECK: vmovntdq %xmm22, 291(%rax,%r14,8)
+// CHECK:  encoding: [0x62,0xa1,0x7d,0x08,0xe7,0xb4,0xf0,0x23,0x01,0x00,0x00]
+          vmovntdq %xmm22, 291(%rax,%r14,8)
+
+// CHECK: vmovntdq %xmm22, 2032(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0x7d,0x08,0xe7,0x72,0x7f]
+          vmovntdq %xmm22, 2032(%rdx)
+
+// CHECK: vmovntdq %xmm22, 2048(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0x7d,0x08,0xe7,0xb2,0x00,0x08,0x00,0x00]
+          vmovntdq %xmm22, 2048(%rdx)
+
+// CHECK: vmovntdq %xmm22, -2048(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0x7d,0x08,0xe7,0x72,0x80]
+          vmovntdq %xmm22, -2048(%rdx)
+
+// CHECK: vmovntdq %xmm22, -2064(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0x7d,0x08,0xe7,0xb2,0xf0,0xf7,0xff,0xff]
+          vmovntdq %xmm22, -2064(%rdx)
+
+// CHECK: vmovntdq %ymm19, (%rcx)
+// CHECK:  encoding: [0x62,0xe1,0x7d,0x28,0xe7,0x19]
+          vmovntdq %ymm19, (%rcx)
+
+// CHECK: vmovntdq %ymm19, 291(%rax,%r14,8)
+// CHECK:  encoding: [0x62,0xa1,0x7d,0x28,0xe7,0x9c,0xf0,0x23,0x01,0x00,0x00]
+          vmovntdq %ymm19, 291(%rax,%r14,8)
+
+// CHECK: vmovntdq %ymm19, 4064(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0x7d,0x28,0xe7,0x5a,0x7f]
+          vmovntdq %ymm19, 4064(%rdx)
+
+// CHECK: vmovntdq %ymm19, 4096(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0x7d,0x28,0xe7,0x9a,0x00,0x10,0x00,0x00]
+          vmovntdq %ymm19, 4096(%rdx)
+
+// CHECK: vmovntdq %ymm19, -4096(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0x7d,0x28,0xe7,0x5a,0x80]
+          vmovntdq %ymm19, -4096(%rdx)
+
+// CHECK: vmovntdq %ymm19, -4128(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0x7d,0x28,0xe7,0x9a,0xe0,0xef,0xff,0xff]
+          vmovntdq %ymm19, -4128(%rdx)
+
+// CHECK: vmovntdqa (%rcx), %xmm24
+// CHECK:  encoding: [0x62,0x62,0x7d,0x08,0x2a,0x01]
+          vmovntdqa (%rcx), %xmm24
+
+// CHECK: vmovntdqa 291(%rax,%r14,8), %xmm24
+// CHECK:  encoding: [0x62,0x22,0x7d,0x08,0x2a,0x84,0xf0,0x23,0x01,0x00,0x00]
+          vmovntdqa 291(%rax,%r14,8), %xmm24
+
+// CHECK: vmovntdqa 2032(%rdx), %xmm24
+// CHECK:  encoding: [0x62,0x62,0x7d,0x08,0x2a,0x42,0x7f]
+          vmovntdqa 2032(%rdx), %xmm24
+
+// CHECK: vmovntdqa 2048(%rdx), %xmm24
+// CHECK:  encoding: [0x62,0x62,0x7d,0x08,0x2a,0x82,0x00,0x08,0x00,0x00]
+          vmovntdqa 2048(%rdx), %xmm24
+
+// CHECK: vmovntdqa -2048(%rdx), %xmm24
+// CHECK:  encoding: [0x62,0x62,0x7d,0x08,0x2a,0x42,0x80]
+          vmovntdqa -2048(%rdx), %xmm24
+
+// CHECK: vmovntdqa -2064(%rdx), %xmm24
+// CHECK:  encoding: [0x62,0x62,0x7d,0x08,0x2a,0x82,0xf0,0xf7,0xff,0xff]
+          vmovntdqa -2064(%rdx), %xmm24
+
+// CHECK: vmovntdqa (%rcx), %ymm28
+// CHECK:  encoding: [0x62,0x62,0x7d,0x28,0x2a,0x21]
+          vmovntdqa (%rcx), %ymm28
+
+// CHECK: vmovntdqa 291(%rax,%r14,8), %ymm28
+// CHECK:  encoding: [0x62,0x22,0x7d,0x28,0x2a,0xa4,0xf0,0x23,0x01,0x00,0x00]
+          vmovntdqa 291(%rax,%r14,8), %ymm28
+
+// CHECK: vmovntdqa 4064(%rdx), %ymm28
+// CHECK:  encoding: [0x62,0x62,0x7d,0x28,0x2a,0x62,0x7f]
+          vmovntdqa 4064(%rdx), %ymm28
+
+// CHECK: vmovntdqa 4096(%rdx), %ymm28
+// CHECK:  encoding: [0x62,0x62,0x7d,0x28,0x2a,0xa2,0x00,0x10,0x00,0x00]
+          vmovntdqa 4096(%rdx), %ymm28
+
+// CHECK: vmovntdqa -4096(%rdx), %ymm28
+// CHECK:  encoding: [0x62,0x62,0x7d,0x28,0x2a,0x62,0x80]
+          vmovntdqa -4096(%rdx), %ymm28
+
+// CHECK: vmovntdqa -4128(%rdx), %ymm28
+// CHECK:  encoding: [0x62,0x62,0x7d,0x28,0x2a,0xa2,0xe0,0xef,0xff,0xff]
+          vmovntdqa -4128(%rdx), %ymm28
+
+// CHECK: vmovntpd %xmm17, (%rcx)
+// CHECK:  encoding: [0x62,0xe1,0xfd,0x08,0x2b,0x09]
+          vmovntpd %xmm17, (%rcx)
+
+// CHECK: vmovntpd %xmm17, 291(%rax,%r14,8)
+// CHECK:  encoding: [0x62,0xa1,0xfd,0x08,0x2b,0x8c,0xf0,0x23,0x01,0x00,0x00]
+          vmovntpd %xmm17, 291(%rax,%r14,8)
+
+// CHECK: vmovntpd %xmm17, 2032(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0xfd,0x08,0x2b,0x4a,0x7f]
+          vmovntpd %xmm17, 2032(%rdx)
+
+// CHECK: vmovntpd %xmm17, 2048(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0xfd,0x08,0x2b,0x8a,0x00,0x08,0x00,0x00]
+          vmovntpd %xmm17, 2048(%rdx)
+
+// CHECK: vmovntpd %xmm17, -2048(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0xfd,0x08,0x2b,0x4a,0x80]
+          vmovntpd %xmm17, -2048(%rdx)
+
+// CHECK: vmovntpd %xmm17, -2064(%rdx)
+// CHECK:  encoding: [0x62,0xe1,0xfd,0x08,0x2b,0x8a,0xf0,0xf7,0xff,0xff]
+          vmovntpd %xmm17, -2064(%rdx)
+
+// CHECK: vmovntpd %ymm27, (%rcx)
+// CHECK:  encoding: [0x62,0x61,0xfd,0x28,0x2b,0x19]
+          vmovntpd %ymm27, (%rcx)
+
+// CHECK: vmovntpd %ymm27, 291(%rax,%r14,8)
+// CHECK:  encoding: [0x62,0x21,0xfd,0x28,0x2b,0x9c,0xf0,0x23,0x01,0x00,0x00]
+          vmovntpd %ymm27, 291(%rax,%r14,8)
+
+// CHECK: vmovntpd %ymm27, 4064(%rdx)
+// CHECK:  encoding: [0x62,0x61,0xfd,0x28,0x2b,0x5a,0x7f]
+          vmovntpd %ymm27, 4064(%rdx)
+
+// CHECK: vmovntpd %ymm27, 4096(%rdx)
+// CHECK:  encoding: [0x62,0x61,0xfd,0x28,0x2b,0x9a,0x00,0x10,0x00,0x00]
+          vmovntpd %ymm27, 4096(%rdx)
+
+// CHECK: vmovntpd %ymm27, -4096(%rdx)
+// CHECK:  encoding: [0x62,0x61,0xfd,0x28,0x2b,0x5a,0x80]
+          vmovntpd %ymm27, -4096(%rdx)
+
+// CHECK: vmovntpd %ymm27, -4128(%rdx)
+// CHECK:  encoding: [0x62,0x61,0xfd,0x28,0x2b,0x9a,0xe0,0xef,0xff,0xff]
+          vmovntpd %ymm27, -4128(%rdx)
+
+// CHECK: vmovntps %xmm26, (%rcx)
+// CHECK:  encoding: [0x62,0x61,0x7c,0x08,0x2b,0x11]
+          vmovntps %xmm26, (%rcx)
+
+// CHECK: vmovntps %xmm26, 291(%rax,%r14,8)
+// CHECK:  encoding: [0x62,0x21,0x7c,0x08,0x2b,0x94,0xf0,0x23,0x01,0x00,0x00]
+          vmovntps %xmm26, 291(%rax,%r14,8)
+
+// CHECK: vmovntps %xmm26, 2032(%rdx)
+// CHECK:  encoding: [0x62,0x61,0x7c,0x08,0x2b,0x52,0x7f]
+          vmovntps %xmm26, 2032(%rdx)
+
+// CHECK: vmovntps %xmm26, 2048(%rdx)
+// CHECK:  encoding: [0x62,0x61,0x7c,0x08,0x2b,0x92,0x00,0x08,0x00,0x00]
+          vmovntps %xmm26, 2048(%rdx)
+
+// CHECK: vmovntps %xmm26, -2048(%rdx)
+// CHECK:  encoding: [0x62,0x61,0x7c,0x08,0x2b,0x52,0x80]
+          vmovntps %xmm26, -2048(%rdx)
+
+// CHECK: vmovntps %xmm26, -2064(%rdx)
+// CHECK:  encoding: [0x62,0x61,0x7c,0x08,0x2b,0x92,0xf0,0xf7,0xff,0xff]
+          vmovntps %xmm26, -2064(%rdx)
+
+// CHECK: vmovntps %ymm28, (%rcx)
+// CHECK:  encoding: [0x62,0x61,0x7c,0x28,0x2b,0x21]
+          vmovntps %ymm28, (%rcx)
+
+// CHECK: vmovntps %ymm28, 291(%rax,%r14,8)
+// CHECK:  encoding: [0x62,0x21,0x7c,0x28,0x2b,0xa4,0xf0,0x23,0x01,0x00,0x00]
+          vmovntps %ymm28, 291(%rax,%r14,8)
+
+// CHECK: vmovntps %ymm28, 4064(%rdx)
+// CHECK:  encoding: [0x62,0x61,0x7c,0x28,0x2b,0x62,0x7f]
+          vmovntps %ymm28, 4064(%rdx)
+
+// CHECK: vmovntps %ymm28, 4096(%rdx)
+// CHECK:  encoding: [0x62,0x61,0x7c,0x28,0x2b,0xa2,0x00,0x10,0x00,0x00]
+          vmovntps %ymm28, 4096(%rdx)
+
+// CHECK: vmovntps %ymm28, -4096(%rdx)
+// CHECK:  encoding: [0x62,0x61,0x7c,0x28,0x2b,0x62,0x80]
+          vmovntps %ymm28, -4096(%rdx)
+
+// CHECK: vmovntps %ymm28, -4128(%rdx)
+// CHECK:  encoding: [0x62,0x61,0x7c,0x28,0x2b,0xa2,0xe0,0xef,0xff,0xff]
+          vmovntps %ymm28, -4128(%rdx)
+
 // CHECK: vmovupd %xmm22, %xmm24
 // CHECK:  encoding: [0x62,0x21,0xfd,0x08,0x10,0xc6]
           vmovupd %xmm22, %xmm24