AVX-512: BLENDM - fixed encoding of the broadcast version

Added more intrinsics and encoding tests.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@224760 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Elena Demikhovsky 2014-12-23 09:36:28 +00:00
parent 1f0ddef593
commit 6709428067
5 changed files with 348 additions and 3 deletions

View File

@ -19923,6 +19923,7 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
case X86ISD::XTEST: return "X86ISD::XTEST";
case X86ISD::COMPRESS: return "X86ISD::COMPRESS";
case X86ISD::EXPAND: return "X86ISD::EXPAND";
case X86ISD::SELECT: return "X86ISD::SELECT";
}
}

View File

@ -1133,14 +1133,14 @@ multiclass avx512_blendmask_rmb<bits<8> opc, string OpcodeStr, X86VectorVTInfo _
"$dst {${mask}}, $src1, ${src2}", _.BroadcastStr, "}"),
[(set _.RC:$dst,(X86select _.KRCWM:$mask, (_.VT _.RC:$src1),
(X86VBroadcast (_.ScalarLdFrag addr:$src2))))]>,
EVEX_4V, EVEX_K, EVEX_B;
EVEX_4V, EVEX_K, EVEX_B, EVEX_CD8<_.EltSize, CD8VF>;
def rmb : AVX5128I<opc, MRMSrcMem, (outs _.RC:$dst),
(ins _.RC:$src1, _.ScalarMemOp:$src2),
!strconcat(OpcodeStr,
"\t{${src2}", _.BroadcastStr, ", $src1, $dst|",
"$dst, $src1, ${src2}", _.BroadcastStr, "}"),
[]>, EVEX_4V, EVEX_B;
[]>, EVEX_4V, EVEX_B, EVEX_CD8<_.EltSize, CD8VF>;
}

View File

@ -1,4 +1,4 @@
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512bw --show-mc-encoding| FileCheck %s
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx --show-mc-encoding| FileCheck %s
define i64 @test_pcmpeq_b(<64 x i8> %a, <64 x i8> %b) {
; CHECK-LABEL: test_pcmpeq_b
@ -303,3 +303,51 @@ define <8 x i32> @test_mask_ucmp_w_512(<32 x i16> %a0, <32 x i16> %a1, i32 %mask
}
declare i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16>, <32 x i16>, i32, i32) nounwind readnone
; CHECK-LABEL: test_x86_mask_blend_b_256
; CHECK: vpblendmb
define <32 x i8> @test_x86_mask_blend_b_256(i32 %a0, <32 x i8> %a1, <32 x i8> %a2) {
%res = call <32 x i8> @llvm.x86.avx512.mask.blend.b.256(<32 x i8> %a1, <32 x i8> %a2, i32 %a0) ; <<32 x i8>> [#uses=1]
ret <32 x i8> %res
}
declare <32 x i8> @llvm.x86.avx512.mask.blend.b.256(<32 x i8>, <32 x i8>, i32) nounwind readonly
; CHECK-LABEL: test_x86_mask_blend_w_256
define <16 x i16> @test_x86_mask_blend_w_256(i16 %mask, <16 x i16> %a1, <16 x i16> %a2) {
; CHECK: vpblendmw
%res = call <16 x i16> @llvm.x86.avx512.mask.blend.w.256(<16 x i16> %a1, <16 x i16> %a2, i16 %mask) ; <<16 x i16>> [#uses=1]
ret <16 x i16> %res
}
declare <16 x i16> @llvm.x86.avx512.mask.blend.w.256(<16 x i16>, <16 x i16>, i16) nounwind readonly
; CHECK-LABEL: test_x86_mask_blend_b_512
; CHECK: vpblendmb
define <64 x i8> @test_x86_mask_blend_b_512(i64 %a0, <64 x i8> %a1, <64 x i8> %a2) {
%res = call <64 x i8> @llvm.x86.avx512.mask.blend.b.512(<64 x i8> %a1, <64 x i8> %a2, i64 %a0) ; <<64 x i8>> [#uses=1]
ret <64 x i8> %res
}
declare <64 x i8> @llvm.x86.avx512.mask.blend.b.512(<64 x i8>, <64 x i8>, i64) nounwind readonly
; CHECK-LABEL: test_x86_mask_blend_w_512
define <32 x i16> @test_x86_mask_blend_w_512(i32 %mask, <32 x i16> %a1, <32 x i16> %a2) {
; CHECK: vpblendmw
%res = call <32 x i16> @llvm.x86.avx512.mask.blend.w.512(<32 x i16> %a1, <32 x i16> %a2, i32 %mask) ; <<32 x i16>> [#uses=1]
ret <32 x i16> %res
}
declare <32 x i16> @llvm.x86.avx512.mask.blend.w.512(<32 x i16>, <32 x i16>, i32) nounwind readonly
; CHECK-LABEL: test_x86_mask_blend_b_128
; CHECK: vpblendmb
define <16 x i8> @test_x86_mask_blend_b_128(i16 %a0, <16 x i8> %a1, <16 x i8> %a2) {
%res = call <16 x i8> @llvm.x86.avx512.mask.blend.b.128(<16 x i8> %a1, <16 x i8> %a2, i16 %a0) ; <<16 x i8>> [#uses=1]
ret <16 x i8> %res
}
declare <16 x i8> @llvm.x86.avx512.mask.blend.b.128(<16 x i8>, <16 x i8>, i16) nounwind readonly
; CHECK-LABEL: test_x86_mask_blend_w_128
define <8 x i16> @test_x86_mask_blend_w_128(i8 %mask, <8 x i16> %a1, <8 x i16> %a2) {
; CHECK: vpblendmw
%res = call <8 x i16> @llvm.x86.avx512.mask.blend.w.128(<8 x i16> %a1, <8 x i16> %a2, i8 %mask) ; <<8 x i16>> [#uses=1]
ret <8 x i16> %res
}
declare <8 x i16> @llvm.x86.avx512.mask.blend.w.128(<8 x i16>, <8 x i16>, i8) nounwind readonly

View File

@ -0,0 +1,73 @@
// RUN: llvm-mc -triple x86_64-unknown-unknown -mcpu=skx --show-encoding %s | FileCheck %s
// CHECK: vpblendmb %zmm25, %zmm18, %zmm17
// CHECK: encoding: [0x62,0x82,0x6d,0x40,0x66,0xc9]
vpblendmb %zmm25, %zmm18, %zmm17
// CHECK: vpblendmb %zmm25, %zmm18, %zmm17 {%k5}
// CHECK: encoding: [0x62,0x82,0x6d,0x45,0x66,0xc9]
vpblendmb %zmm25, %zmm18, %zmm17 {%k5}
// CHECK: vpblendmb %zmm25, %zmm18, %zmm17 {%k5} {z}
// CHECK: encoding: [0x62,0x82,0x6d,0xc5,0x66,0xc9]
vpblendmb %zmm25, %zmm18, %zmm17 {%k5} {z}
// CHECK: vpblendmb (%rcx), %zmm18, %zmm17
// CHECK: encoding: [0x62,0xe2,0x6d,0x40,0x66,0x09]
vpblendmb (%rcx), %zmm18, %zmm17
// CHECK: vpblendmb 291(%rax,%r14,8), %zmm18, %zmm17
// CHECK: encoding: [0x62,0xa2,0x6d,0x40,0x66,0x8c,0xf0,0x23,0x01,0x00,0x00]
vpblendmb 291(%rax,%r14,8), %zmm18, %zmm17
// CHECK: vpblendmb 8128(%rdx), %zmm18, %zmm17
// CHECK: encoding: [0x62,0xe2,0x6d,0x40,0x66,0x4a,0x7f]
vpblendmb 8128(%rdx), %zmm18, %zmm17
// CHECK: vpblendmb 8192(%rdx), %zmm18, %zmm17
// CHECK: encoding: [0x62,0xe2,0x6d,0x40,0x66,0x8a,0x00,0x20,0x00,0x00]
vpblendmb 8192(%rdx), %zmm18, %zmm17
// CHECK: vpblendmb -8192(%rdx), %zmm18, %zmm17
// CHECK: encoding: [0x62,0xe2,0x6d,0x40,0x66,0x4a,0x80]
vpblendmb -8192(%rdx), %zmm18, %zmm17
// CHECK: vpblendmb -8256(%rdx), %zmm18, %zmm17
// CHECK: encoding: [0x62,0xe2,0x6d,0x40,0x66,0x8a,0xc0,0xdf,0xff,0xff]
vpblendmb -8256(%rdx), %zmm18, %zmm17
// CHECK: vpblendmw %zmm17, %zmm20, %zmm26
// CHECK: encoding: [0x62,0x22,0xdd,0x40,0x66,0xd1]
vpblendmw %zmm17, %zmm20, %zmm26
// CHECK: vpblendmw %zmm17, %zmm20, %zmm26 {%k7}
// CHECK: encoding: [0x62,0x22,0xdd,0x47,0x66,0xd1]
vpblendmw %zmm17, %zmm20, %zmm26 {%k7}
// CHECK: vpblendmw %zmm17, %zmm20, %zmm26 {%k7} {z}
// CHECK: encoding: [0x62,0x22,0xdd,0xc7,0x66,0xd1]
vpblendmw %zmm17, %zmm20, %zmm26 {%k7} {z}
// CHECK: vpblendmw (%rcx), %zmm20, %zmm26
// CHECK: encoding: [0x62,0x62,0xdd,0x40,0x66,0x11]
vpblendmw (%rcx), %zmm20, %zmm26
// CHECK: vpblendmw 291(%rax,%r14,8), %zmm20, %zmm26
// CHECK: encoding: [0x62,0x22,0xdd,0x40,0x66,0x94,0xf0,0x23,0x01,0x00,0x00]
vpblendmw 291(%rax,%r14,8), %zmm20, %zmm26
// CHECK: vpblendmw 8128(%rdx), %zmm20, %zmm26
// CHECK: encoding: [0x62,0x62,0xdd,0x40,0x66,0x52,0x7f]
vpblendmw 8128(%rdx), %zmm20, %zmm26
// CHECK: vpblendmw 8192(%rdx), %zmm20, %zmm26
// CHECK: encoding: [0x62,0x62,0xdd,0x40,0x66,0x92,0x00,0x20,0x00,0x00]
vpblendmw 8192(%rdx), %zmm20, %zmm26
// CHECK: vpblendmw -8192(%rdx), %zmm20, %zmm26
// CHECK: encoding: [0x62,0x62,0xdd,0x40,0x66,0x52,0x80]
vpblendmw -8192(%rdx), %zmm20, %zmm26
// CHECK: vpblendmw -8256(%rdx), %zmm20, %zmm26
// CHECK: encoding: [0x62,0x62,0xdd,0x40,0x66,0x92,0xc0,0xdf,0xff,0xff]
vpblendmw -8256(%rdx), %zmm20, %zmm26

View File

@ -1,226 +1,449 @@
// RUN: llvm-mc -triple x86_64-unknown-unknown -mcpu=skx --show-encoding %s | FileCheck %s
// CHECK: vblendmpd %xmm19, %xmm20, %xmm27
// CHECK: encoding: [0x62,0x22,0xdd,0x00,0x65,0xdb]
vblendmpd %xmm19, %xmm20, %xmm27
// CHECK: vblendmpd %xmm19, %xmm20, %xmm27 {%k7}
// CHECK: encoding: [0x62,0x22,0xdd,0x07,0x65,0xdb]
vblendmpd %xmm19, %xmm20, %xmm27 {%k7}
// CHECK: vblendmpd %xmm19, %xmm20, %xmm27 {%k7} {z}
// CHECK: encoding: [0x62,0x22,0xdd,0x87,0x65,0xdb]
vblendmpd %xmm19, %xmm20, %xmm27 {%k7} {z}
// CHECK: vblendmpd (%rcx), %xmm20, %xmm27
// CHECK: encoding: [0x62,0x62,0xdd,0x00,0x65,0x19]
vblendmpd (%rcx), %xmm20, %xmm27
// CHECK: vblendmpd 291(%rax,%r14,8), %xmm20, %xmm27
// CHECK: encoding: [0x62,0x22,0xdd,0x00,0x65,0x9c,0xf0,0x23,0x01,0x00,0x00]
vblendmpd 291(%rax,%r14,8), %xmm20, %xmm27
// CHECK: vblendmpd (%rcx){1to2}, %xmm20, %xmm27
// CHECK: encoding: [0x62,0x62,0xdd,0x10,0x65,0x19]
vblendmpd (%rcx){1to2}, %xmm20, %xmm27
// CHECK: vblendmpd 2032(%rdx), %xmm20, %xmm27
// CHECK: encoding: [0x62,0x62,0xdd,0x00,0x65,0x5a,0x7f]
vblendmpd 2032(%rdx), %xmm20, %xmm27
// CHECK: vblendmpd 2048(%rdx), %xmm20, %xmm27
// CHECK: encoding: [0x62,0x62,0xdd,0x00,0x65,0x9a,0x00,0x08,0x00,0x00]
vblendmpd 2048(%rdx), %xmm20, %xmm27
// CHECK: vblendmpd -2048(%rdx), %xmm20, %xmm27
// CHECK: encoding: [0x62,0x62,0xdd,0x00,0x65,0x5a,0x80]
vblendmpd -2048(%rdx), %xmm20, %xmm27
// CHECK: vblendmpd -2064(%rdx), %xmm20, %xmm27
// CHECK: encoding: [0x62,0x62,0xdd,0x00,0x65,0x9a,0xf0,0xf7,0xff,0xff]
vblendmpd -2064(%rdx), %xmm20, %xmm27
// CHECK: vblendmpd 1016(%rdx){1to2}, %xmm20, %xmm27
// CHECK: encoding: [0x62,0x62,0xdd,0x10,0x65,0x5a,0x7f]
vblendmpd 1016(%rdx){1to2}, %xmm20, %xmm27
// CHECK: vblendmpd 1024(%rdx){1to2}, %xmm20, %xmm27
// CHECK: encoding: [0x62,0x62,0xdd,0x10,0x65,0x9a,0x00,0x04,0x00,0x00]
vblendmpd 1024(%rdx){1to2}, %xmm20, %xmm27
// CHECK: vblendmpd -1024(%rdx){1to2}, %xmm20, %xmm27
// CHECK: encoding: [0x62,0x62,0xdd,0x10,0x65,0x5a,0x80]
vblendmpd -1024(%rdx){1to2}, %xmm20, %xmm27
// CHECK: vblendmpd -1032(%rdx){1to2}, %xmm20, %xmm27
// CHECK: encoding: [0x62,0x62,0xdd,0x10,0x65,0x9a,0xf8,0xfb,0xff,0xff]
vblendmpd -1032(%rdx){1to2}, %xmm20, %xmm27
// CHECK: vblendmpd %ymm23, %ymm21, %ymm28
// CHECK: encoding: [0x62,0x22,0xd5,0x20,0x65,0xe7]
vblendmpd %ymm23, %ymm21, %ymm28
// CHECK: vblendmpd %ymm23, %ymm21, %ymm28 {%k3}
// CHECK: encoding: [0x62,0x22,0xd5,0x23,0x65,0xe7]
vblendmpd %ymm23, %ymm21, %ymm28 {%k3}
// CHECK: vblendmpd %ymm23, %ymm21, %ymm28 {%k3} {z}
// CHECK: encoding: [0x62,0x22,0xd5,0xa3,0x65,0xe7]
vblendmpd %ymm23, %ymm21, %ymm28 {%k3} {z}
// CHECK: vblendmpd (%rcx), %ymm21, %ymm28
// CHECK: encoding: [0x62,0x62,0xd5,0x20,0x65,0x21]
vblendmpd (%rcx), %ymm21, %ymm28
// CHECK: vblendmpd 291(%rax,%r14,8), %ymm21, %ymm28
// CHECK: encoding: [0x62,0x22,0xd5,0x20,0x65,0xa4,0xf0,0x23,0x01,0x00,0x00]
vblendmpd 291(%rax,%r14,8), %ymm21, %ymm28
// CHECK: vblendmpd (%rcx){1to4}, %ymm21, %ymm28
// CHECK: encoding: [0x62,0x62,0xd5,0x30,0x65,0x21]
vblendmpd (%rcx){1to4}, %ymm21, %ymm28
// CHECK: vblendmpd 4064(%rdx), %ymm21, %ymm28
// CHECK: encoding: [0x62,0x62,0xd5,0x20,0x65,0x62,0x7f]
vblendmpd 4064(%rdx), %ymm21, %ymm28
// CHECK: vblendmpd 4096(%rdx), %ymm21, %ymm28
// CHECK: encoding: [0x62,0x62,0xd5,0x20,0x65,0xa2,0x00,0x10,0x00,0x00]
vblendmpd 4096(%rdx), %ymm21, %ymm28
// CHECK: vblendmpd -4096(%rdx), %ymm21, %ymm28
// CHECK: encoding: [0x62,0x62,0xd5,0x20,0x65,0x62,0x80]
vblendmpd -4096(%rdx), %ymm21, %ymm28
// CHECK: vblendmpd -4128(%rdx), %ymm21, %ymm28
// CHECK: encoding: [0x62,0x62,0xd5,0x20,0x65,0xa2,0xe0,0xef,0xff,0xff]
vblendmpd -4128(%rdx), %ymm21, %ymm28
// CHECK: vblendmpd 1016(%rdx){1to4}, %ymm21, %ymm28
// CHECK: encoding: [0x62,0x62,0xd5,0x30,0x65,0x62,0x7f]
vblendmpd 1016(%rdx){1to4}, %ymm21, %ymm28
// CHECK: vblendmpd 1024(%rdx){1to4}, %ymm21, %ymm28
// CHECK: encoding: [0x62,0x62,0xd5,0x30,0x65,0xa2,0x00,0x04,0x00,0x00]
vblendmpd 1024(%rdx){1to4}, %ymm21, %ymm28
// CHECK: vblendmpd -1024(%rdx){1to4}, %ymm21, %ymm28
// CHECK: encoding: [0x62,0x62,0xd5,0x30,0x65,0x62,0x80]
vblendmpd -1024(%rdx){1to4}, %ymm21, %ymm28
// CHECK: vblendmpd -1032(%rdx){1to4}, %ymm21, %ymm28
// CHECK: encoding: [0x62,0x62,0xd5,0x30,0x65,0xa2,0xf8,0xfb,0xff,0xff]
vblendmpd -1032(%rdx){1to4}, %ymm21, %ymm28
// CHECK: vblendmps %xmm20, %xmm20, %xmm24
// CHECK: encoding: [0x62,0x22,0x5d,0x00,0x65,0xc4]
vblendmps %xmm20, %xmm20, %xmm24
// CHECK: vblendmps %xmm20, %xmm20, %xmm24 {%k1}
// CHECK: encoding: [0x62,0x22,0x5d,0x01,0x65,0xc4]
vblendmps %xmm20, %xmm20, %xmm24 {%k1}
// CHECK: vblendmps %xmm20, %xmm20, %xmm24 {%k1} {z}
// CHECK: encoding: [0x62,0x22,0x5d,0x81,0x65,0xc4]
vblendmps %xmm20, %xmm20, %xmm24 {%k1} {z}
// CHECK: vblendmps (%rcx), %xmm20, %xmm24
// CHECK: encoding: [0x62,0x62,0x5d,0x00,0x65,0x01]
vblendmps (%rcx), %xmm20, %xmm24
// CHECK: vblendmps 291(%rax,%r14,8), %xmm20, %xmm24
// CHECK: encoding: [0x62,0x22,0x5d,0x00,0x65,0x84,0xf0,0x23,0x01,0x00,0x00]
vblendmps 291(%rax,%r14,8), %xmm20, %xmm24
// CHECK: vblendmps (%rcx){1to4}, %xmm20, %xmm24
// CHECK: encoding: [0x62,0x62,0x5d,0x10,0x65,0x01]
vblendmps (%rcx){1to4}, %xmm20, %xmm24
// CHECK: vblendmps 2032(%rdx), %xmm20, %xmm24
// CHECK: encoding: [0x62,0x62,0x5d,0x00,0x65,0x42,0x7f]
vblendmps 2032(%rdx), %xmm20, %xmm24
// CHECK: vblendmps 2048(%rdx), %xmm20, %xmm24
// CHECK: encoding: [0x62,0x62,0x5d,0x00,0x65,0x82,0x00,0x08,0x00,0x00]
vblendmps 2048(%rdx), %xmm20, %xmm24
// CHECK: vblendmps -2048(%rdx), %xmm20, %xmm24
// CHECK: encoding: [0x62,0x62,0x5d,0x00,0x65,0x42,0x80]
vblendmps -2048(%rdx), %xmm20, %xmm24
// CHECK: vblendmps -2064(%rdx), %xmm20, %xmm24
// CHECK: encoding: [0x62,0x62,0x5d,0x00,0x65,0x82,0xf0,0xf7,0xff,0xff]
vblendmps -2064(%rdx), %xmm20, %xmm24
// CHECK: vblendmps 508(%rdx){1to4}, %xmm20, %xmm24
// CHECK: encoding: [0x62,0x62,0x5d,0x10,0x65,0x42,0x7f]
vblendmps 508(%rdx){1to4}, %xmm20, %xmm24
// CHECK: vblendmps 512(%rdx){1to4}, %xmm20, %xmm24
// CHECK: encoding: [0x62,0x62,0x5d,0x10,0x65,0x82,0x00,0x02,0x00,0x00]
vblendmps 512(%rdx){1to4}, %xmm20, %xmm24
// CHECK: vblendmps -512(%rdx){1to4}, %xmm20, %xmm24
// CHECK: encoding: [0x62,0x62,0x5d,0x10,0x65,0x42,0x80]
vblendmps -512(%rdx){1to4}, %xmm20, %xmm24
// CHECK: vblendmps -516(%rdx){1to4}, %xmm20, %xmm24
// CHECK: encoding: [0x62,0x62,0x5d,0x10,0x65,0x82,0xfc,0xfd,0xff,0xff]
vblendmps -516(%rdx){1to4}, %xmm20, %xmm24
// CHECK: vblendmps %ymm24, %ymm23, %ymm17
// CHECK: encoding: [0x62,0x82,0x45,0x20,0x65,0xc8]
vblendmps %ymm24, %ymm23, %ymm17
// CHECK: vblendmps %ymm24, %ymm23, %ymm17 {%k6}
// CHECK: encoding: [0x62,0x82,0x45,0x26,0x65,0xc8]
vblendmps %ymm24, %ymm23, %ymm17 {%k6}
// CHECK: vblendmps %ymm24, %ymm23, %ymm17 {%k6} {z}
// CHECK: encoding: [0x62,0x82,0x45,0xa6,0x65,0xc8]
vblendmps %ymm24, %ymm23, %ymm17 {%k6} {z}
// CHECK: vblendmps (%rcx), %ymm23, %ymm17
// CHECK: encoding: [0x62,0xe2,0x45,0x20,0x65,0x09]
vblendmps (%rcx), %ymm23, %ymm17
// CHECK: vblendmps 291(%rax,%r14,8), %ymm23, %ymm17
// CHECK: encoding: [0x62,0xa2,0x45,0x20,0x65,0x8c,0xf0,0x23,0x01,0x00,0x00]
vblendmps 291(%rax,%r14,8), %ymm23, %ymm17
// CHECK: vblendmps (%rcx){1to8}, %ymm23, %ymm17
// CHECK: encoding: [0x62,0xe2,0x45,0x30,0x65,0x09]
vblendmps (%rcx){1to8}, %ymm23, %ymm17
// CHECK: vblendmps 4064(%rdx), %ymm23, %ymm17
// CHECK: encoding: [0x62,0xe2,0x45,0x20,0x65,0x4a,0x7f]
vblendmps 4064(%rdx), %ymm23, %ymm17
// CHECK: vblendmps 4096(%rdx), %ymm23, %ymm17
// CHECK: encoding: [0x62,0xe2,0x45,0x20,0x65,0x8a,0x00,0x10,0x00,0x00]
vblendmps 4096(%rdx), %ymm23, %ymm17
// CHECK: vblendmps -4096(%rdx), %ymm23, %ymm17
// CHECK: encoding: [0x62,0xe2,0x45,0x20,0x65,0x4a,0x80]
vblendmps -4096(%rdx), %ymm23, %ymm17
// CHECK: vblendmps -4128(%rdx), %ymm23, %ymm17
// CHECK: encoding: [0x62,0xe2,0x45,0x20,0x65,0x8a,0xe0,0xef,0xff,0xff]
vblendmps -4128(%rdx), %ymm23, %ymm17
// CHECK: vblendmps 508(%rdx){1to8}, %ymm23, %ymm17
// CHECK: encoding: [0x62,0xe2,0x45,0x30,0x65,0x4a,0x7f]
vblendmps 508(%rdx){1to8}, %ymm23, %ymm17
// CHECK: vblendmps 512(%rdx){1to8}, %ymm23, %ymm17
// CHECK: encoding: [0x62,0xe2,0x45,0x30,0x65,0x8a,0x00,0x02,0x00,0x00]
vblendmps 512(%rdx){1to8}, %ymm23, %ymm17
// CHECK: vblendmps -512(%rdx){1to8}, %ymm23, %ymm17
// CHECK: encoding: [0x62,0xe2,0x45,0x30,0x65,0x4a,0x80]
vblendmps -512(%rdx){1to8}, %ymm23, %ymm17
// CHECK: vblendmps -516(%rdx){1to8}, %ymm23, %ymm17
// CHECK: encoding: [0x62,0xe2,0x45,0x30,0x65,0x8a,0xfc,0xfd,0xff,0xff]
vblendmps -516(%rdx){1to8}, %ymm23, %ymm17
// CHECK: vpblendmd %xmm26, %xmm25, %xmm17
// CHECK: encoding: [0x62,0x82,0x35,0x00,0x64,0xca]
vpblendmd %xmm26, %xmm25, %xmm17
// CHECK: vpblendmd %xmm26, %xmm25, %xmm17 {%k5}
// CHECK: encoding: [0x62,0x82,0x35,0x05,0x64,0xca]
vpblendmd %xmm26, %xmm25, %xmm17 {%k5}
// CHECK: vpblendmd %xmm26, %xmm25, %xmm17 {%k5} {z}
// CHECK: encoding: [0x62,0x82,0x35,0x85,0x64,0xca]
vpblendmd %xmm26, %xmm25, %xmm17 {%k5} {z}
// CHECK: vpblendmd (%rcx), %xmm25, %xmm17
// CHECK: encoding: [0x62,0xe2,0x35,0x00,0x64,0x09]
vpblendmd (%rcx), %xmm25, %xmm17
// CHECK: vpblendmd 291(%rax,%r14,8), %xmm25, %xmm17
// CHECK: encoding: [0x62,0xa2,0x35,0x00,0x64,0x8c,0xf0,0x23,0x01,0x00,0x00]
vpblendmd 291(%rax,%r14,8), %xmm25, %xmm17
// CHECK: vpblendmd (%rcx){1to4}, %xmm25, %xmm17
// CHECK: encoding: [0x62,0xe2,0x35,0x10,0x64,0x09]
vpblendmd (%rcx){1to4}, %xmm25, %xmm17
// CHECK: vpblendmd 2032(%rdx), %xmm25, %xmm17
// CHECK: encoding: [0x62,0xe2,0x35,0x00,0x64,0x4a,0x7f]
vpblendmd 2032(%rdx), %xmm25, %xmm17
// CHECK: vpblendmd 2048(%rdx), %xmm25, %xmm17
// CHECK: encoding: [0x62,0xe2,0x35,0x00,0x64,0x8a,0x00,0x08,0x00,0x00]
vpblendmd 2048(%rdx), %xmm25, %xmm17
// CHECK: vpblendmd -2048(%rdx), %xmm25, %xmm17
// CHECK: encoding: [0x62,0xe2,0x35,0x00,0x64,0x4a,0x80]
vpblendmd -2048(%rdx), %xmm25, %xmm17
// CHECK: vpblendmd -2064(%rdx), %xmm25, %xmm17
// CHECK: encoding: [0x62,0xe2,0x35,0x00,0x64,0x8a,0xf0,0xf7,0xff,0xff]
vpblendmd -2064(%rdx), %xmm25, %xmm17
// CHECK: vpblendmd 508(%rdx){1to4}, %xmm25, %xmm17
// CHECK: encoding: [0x62,0xe2,0x35,0x10,0x64,0x4a,0x7f]
vpblendmd 508(%rdx){1to4}, %xmm25, %xmm17
// CHECK: vpblendmd 512(%rdx){1to4}, %xmm25, %xmm17
// CHECK: encoding: [0x62,0xe2,0x35,0x10,0x64,0x8a,0x00,0x02,0x00,0x00]
vpblendmd 512(%rdx){1to4}, %xmm25, %xmm17
// CHECK: vpblendmd -512(%rdx){1to4}, %xmm25, %xmm17
// CHECK: encoding: [0x62,0xe2,0x35,0x10,0x64,0x4a,0x80]
vpblendmd -512(%rdx){1to4}, %xmm25, %xmm17
// CHECK: vpblendmd -516(%rdx){1to4}, %xmm25, %xmm17
// CHECK: encoding: [0x62,0xe2,0x35,0x10,0x64,0x8a,0xfc,0xfd,0xff,0xff]
vpblendmd -516(%rdx){1to4}, %xmm25, %xmm17
// CHECK: vpblendmd %ymm23, %ymm29, %ymm26
// CHECK: encoding: [0x62,0x22,0x15,0x20,0x64,0xd7]
vpblendmd %ymm23, %ymm29, %ymm26
// CHECK: vpblendmd %ymm23, %ymm29, %ymm26 {%k7}
// CHECK: encoding: [0x62,0x22,0x15,0x27,0x64,0xd7]
vpblendmd %ymm23, %ymm29, %ymm26 {%k7}
// CHECK: vpblendmd %ymm23, %ymm29, %ymm26 {%k7} {z}
// CHECK: encoding: [0x62,0x22,0x15,0xa7,0x64,0xd7]
vpblendmd %ymm23, %ymm29, %ymm26 {%k7} {z}
// CHECK: vpblendmd (%rcx), %ymm29, %ymm26
// CHECK: encoding: [0x62,0x62,0x15,0x20,0x64,0x11]
vpblendmd (%rcx), %ymm29, %ymm26
// CHECK: vpblendmd 291(%rax,%r14,8), %ymm29, %ymm26
// CHECK: encoding: [0x62,0x22,0x15,0x20,0x64,0x94,0xf0,0x23,0x01,0x00,0x00]
vpblendmd 291(%rax,%r14,8), %ymm29, %ymm26
// CHECK: vpblendmd (%rcx){1to8}, %ymm29, %ymm26
// CHECK: encoding: [0x62,0x62,0x15,0x30,0x64,0x11]
vpblendmd (%rcx){1to8}, %ymm29, %ymm26
// CHECK: vpblendmd 4064(%rdx), %ymm29, %ymm26
// CHECK: encoding: [0x62,0x62,0x15,0x20,0x64,0x52,0x7f]
vpblendmd 4064(%rdx), %ymm29, %ymm26
// CHECK: vpblendmd 4096(%rdx), %ymm29, %ymm26
// CHECK: encoding: [0x62,0x62,0x15,0x20,0x64,0x92,0x00,0x10,0x00,0x00]
vpblendmd 4096(%rdx), %ymm29, %ymm26
// CHECK: vpblendmd -4096(%rdx), %ymm29, %ymm26
// CHECK: encoding: [0x62,0x62,0x15,0x20,0x64,0x52,0x80]
vpblendmd -4096(%rdx), %ymm29, %ymm26
// CHECK: vpblendmd -4128(%rdx), %ymm29, %ymm26
// CHECK: encoding: [0x62,0x62,0x15,0x20,0x64,0x92,0xe0,0xef,0xff,0xff]
vpblendmd -4128(%rdx), %ymm29, %ymm26
// CHECK: vpblendmd 508(%rdx){1to8}, %ymm29, %ymm26
// CHECK: encoding: [0x62,0x62,0x15,0x30,0x64,0x52,0x7f]
vpblendmd 508(%rdx){1to8}, %ymm29, %ymm26
// CHECK: vpblendmd 512(%rdx){1to8}, %ymm29, %ymm26
// CHECK: encoding: [0x62,0x62,0x15,0x30,0x64,0x92,0x00,0x02,0x00,0x00]
vpblendmd 512(%rdx){1to8}, %ymm29, %ymm26
// CHECK: vpblendmd -512(%rdx){1to8}, %ymm29, %ymm26
// CHECK: encoding: [0x62,0x62,0x15,0x30,0x64,0x52,0x80]
vpblendmd -512(%rdx){1to8}, %ymm29, %ymm26
// CHECK: vpblendmd -516(%rdx){1to8}, %ymm29, %ymm26
// CHECK: encoding: [0x62,0x62,0x15,0x30,0x64,0x92,0xfc,0xfd,0xff,0xff]
vpblendmd -516(%rdx){1to8}, %ymm29, %ymm26
// CHECK: vpblendmq %xmm17, %xmm27, %xmm29
// CHECK: encoding: [0x62,0x22,0xa5,0x00,0x64,0xe9]
vpblendmq %xmm17, %xmm27, %xmm29
// CHECK: vpblendmq %xmm17, %xmm27, %xmm29 {%k6}
// CHECK: encoding: [0x62,0x22,0xa5,0x06,0x64,0xe9]
vpblendmq %xmm17, %xmm27, %xmm29 {%k6}
// CHECK: vpblendmq %xmm17, %xmm27, %xmm29 {%k6} {z}
// CHECK: encoding: [0x62,0x22,0xa5,0x86,0x64,0xe9]
vpblendmq %xmm17, %xmm27, %xmm29 {%k6} {z}
// CHECK: vpblendmq (%rcx), %xmm27, %xmm29
// CHECK: encoding: [0x62,0x62,0xa5,0x00,0x64,0x29]
vpblendmq (%rcx), %xmm27, %xmm29
// CHECK: vpblendmq 291(%rax,%r14,8), %xmm27, %xmm29
// CHECK: encoding: [0x62,0x22,0xa5,0x00,0x64,0xac,0xf0,0x23,0x01,0x00,0x00]
vpblendmq 291(%rax,%r14,8), %xmm27, %xmm29
// CHECK: vpblendmq (%rcx){1to2}, %xmm27, %xmm29
// CHECK: encoding: [0x62,0x62,0xa5,0x10,0x64,0x29]
vpblendmq (%rcx){1to2}, %xmm27, %xmm29
// CHECK: vpblendmq 2032(%rdx), %xmm27, %xmm29
// CHECK: encoding: [0x62,0x62,0xa5,0x00,0x64,0x6a,0x7f]
vpblendmq 2032(%rdx), %xmm27, %xmm29
// CHECK: vpblendmq 2048(%rdx), %xmm27, %xmm29
// CHECK: encoding: [0x62,0x62,0xa5,0x00,0x64,0xaa,0x00,0x08,0x00,0x00]
vpblendmq 2048(%rdx), %xmm27, %xmm29
// CHECK: vpblendmq -2048(%rdx), %xmm27, %xmm29
// CHECK: encoding: [0x62,0x62,0xa5,0x00,0x64,0x6a,0x80]
vpblendmq -2048(%rdx), %xmm27, %xmm29
// CHECK: vpblendmq -2064(%rdx), %xmm27, %xmm29
// CHECK: encoding: [0x62,0x62,0xa5,0x00,0x64,0xaa,0xf0,0xf7,0xff,0xff]
vpblendmq -2064(%rdx), %xmm27, %xmm29
// CHECK: vpblendmq 1016(%rdx){1to2}, %xmm27, %xmm29
// CHECK: encoding: [0x62,0x62,0xa5,0x10,0x64,0x6a,0x7f]
vpblendmq 1016(%rdx){1to2}, %xmm27, %xmm29
// CHECK: vpblendmq 1024(%rdx){1to2}, %xmm27, %xmm29
// CHECK: encoding: [0x62,0x62,0xa5,0x10,0x64,0xaa,0x00,0x04,0x00,0x00]
vpblendmq 1024(%rdx){1to2}, %xmm27, %xmm29
// CHECK: vpblendmq -1024(%rdx){1to2}, %xmm27, %xmm29
// CHECK: encoding: [0x62,0x62,0xa5,0x10,0x64,0x6a,0x80]
vpblendmq -1024(%rdx){1to2}, %xmm27, %xmm29
// CHECK: vpblendmq -1032(%rdx){1to2}, %xmm27, %xmm29
// CHECK: encoding: [0x62,0x62,0xa5,0x10,0x64,0xaa,0xf8,0xfb,0xff,0xff]
vpblendmq -1032(%rdx){1to2}, %xmm27, %xmm29
// CHECK: vpblendmq %ymm21, %ymm23, %ymm21
// CHECK: encoding: [0x62,0xa2,0xc5,0x20,0x64,0xed]
vpblendmq %ymm21, %ymm23, %ymm21
// CHECK: vpblendmq %ymm21, %ymm23, %ymm21 {%k3}
// CHECK: encoding: [0x62,0xa2,0xc5,0x23,0x64,0xed]
vpblendmq %ymm21, %ymm23, %ymm21 {%k3}
// CHECK: vpblendmq %ymm21, %ymm23, %ymm21 {%k3} {z}
// CHECK: encoding: [0x62,0xa2,0xc5,0xa3,0x64,0xed]
vpblendmq %ymm21, %ymm23, %ymm21 {%k3} {z}
// CHECK: vpblendmq (%rcx), %ymm23, %ymm21
// CHECK: encoding: [0x62,0xe2,0xc5,0x20,0x64,0x29]
vpblendmq (%rcx), %ymm23, %ymm21
// CHECK: vpblendmq 291(%rax,%r14,8), %ymm23, %ymm21
// CHECK: encoding: [0x62,0xa2,0xc5,0x20,0x64,0xac,0xf0,0x23,0x01,0x00,0x00]
vpblendmq 291(%rax,%r14,8), %ymm23, %ymm21
// CHECK: vpblendmq (%rcx){1to4}, %ymm23, %ymm21
// CHECK: encoding: [0x62,0xe2,0xc5,0x30,0x64,0x29]
vpblendmq (%rcx){1to4}, %ymm23, %ymm21
// CHECK: vpblendmq 4064(%rdx), %ymm23, %ymm21
// CHECK: encoding: [0x62,0xe2,0xc5,0x20,0x64,0x6a,0x7f]
vpblendmq 4064(%rdx), %ymm23, %ymm21
// CHECK: vpblendmq 4096(%rdx), %ymm23, %ymm21
// CHECK: encoding: [0x62,0xe2,0xc5,0x20,0x64,0xaa,0x00,0x10,0x00,0x00]
vpblendmq 4096(%rdx), %ymm23, %ymm21
// CHECK: vpblendmq -4096(%rdx), %ymm23, %ymm21
// CHECK: encoding: [0x62,0xe2,0xc5,0x20,0x64,0x6a,0x80]
vpblendmq -4096(%rdx), %ymm23, %ymm21
// CHECK: vpblendmq -4128(%rdx), %ymm23, %ymm21
// CHECK: encoding: [0x62,0xe2,0xc5,0x20,0x64,0xaa,0xe0,0xef,0xff,0xff]
vpblendmq -4128(%rdx), %ymm23, %ymm21
// CHECK: vpblendmq 1016(%rdx){1to4}, %ymm23, %ymm21
// CHECK: encoding: [0x62,0xe2,0xc5,0x30,0x64,0x6a,0x7f]
vpblendmq 1016(%rdx){1to4}, %ymm23, %ymm21
// CHECK: vpblendmq 1024(%rdx){1to4}, %ymm23, %ymm21
// CHECK: encoding: [0x62,0xe2,0xc5,0x30,0x64,0xaa,0x00,0x04,0x00,0x00]
vpblendmq 1024(%rdx){1to4}, %ymm23, %ymm21
// CHECK: vpblendmq -1024(%rdx){1to4}, %ymm23, %ymm21
// CHECK: encoding: [0x62,0xe2,0xc5,0x30,0x64,0x6a,0x80]
vpblendmq -1024(%rdx){1to4}, %ymm23, %ymm21
// CHECK: vpblendmq -1032(%rdx){1to4}, %ymm23, %ymm21
// CHECK: encoding: [0x62,0xe2,0xc5,0x30,0x64,0xaa,0xf8,0xfb,0xff,0xff]
vpblendmq -1032(%rdx){1to4}, %ymm23, %ymm21