replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[AVX512] Add 512b integer shift by variable intrinsics and patterns. 

llvm-svn: 222786
This commit is contained in:
Cameron McInally 2014-11-25 20:41:51 +00:00
parent f7f156ffc3
commit 9b7c15a364
5 changed files with 200 additions and 48 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
llvm/include/llvm/IR/IntrinsicsX86.td
View File

@ -1603,6 +1603,25 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
def int_x86_avx512_mask_psrai_q : GCCBuiltin<"__builtin_ia32_psraqi512">,
Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty,
llvm_i32_ty, llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>;
def int_x86_avx512_mask_psll_d : GCCBuiltin<"__builtin_ia32_pslld512_mask">,
Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty,
llvm_v4i32_ty, llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>;
def int_x86_avx512_mask_psll_q : GCCBuiltin<"__builtin_ia32_psllq512_mask">,
Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty,
llvm_v2i64_ty, llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>;
def int_x86_avx512_mask_psrl_d : GCCBuiltin<"__builtin_ia32_psrld512_mask">,
Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty,
llvm_v4i32_ty, llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>;
def int_x86_avx512_mask_psrl_q : GCCBuiltin<"__builtin_ia32_psrlq512_mask">,
Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty,
llvm_v2i64_ty, llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>;
def int_x86_avx512_mask_psra_d : GCCBuiltin<"__builtin_ia32_psrad512_mask">,
Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty,
llvm_v4i32_ty, llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>;
def int_x86_avx512_mask_psra_q : GCCBuiltin<"__builtin_ia32_psraq512_mask">,
Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty,
llvm_v2i64_ty, llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>;
}
// Pack ops.

8
llvm/lib/Target/X86/X86ISelLowering.cpp
View File

@ -16872,7 +16872,11 @@ static SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, const X86Subtarget *Subtarget
RoundingMode),
Mask, Src0, Subtarget, DAG);
}
case INTR_TYPE_2OP_MASK: {
return getVectorMaskingNode(DAG.getNode(IntrData->Opc0, dl, VT, Op.getOperand(1),
Op.getOperand(2)),
Op.getOperand(4), Op.getOperand(3), Subtarget, DAG);
}
case CMP_MASK:
case CMP_MASK_CC: {
// Comparison intrinsics with masks.
@ -16924,7 +16928,7 @@ static SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, const X86Subtarget *Subtarget
case VSHIFT_MASK:
return getVectorMaskingNode(getTargetVShiftNode(IntrData->Opc0, dl, Op.getSimpleValueType(),
Op.getOperand(1), Op.getOperand(2), DAG),
Op.getOperand(4), Op.getOperand(3), Subtarget, DAG);;
Op.getOperand(4), Op.getOperand(3), Subtarget, DAG);
default:
break;
}

75
llvm/lib/Target/X86/X86InstrAVX512.td
View File

@ -3196,6 +3196,7 @@ def : Pat <(i16 (int_x86_avx512_mask_ptestm_d_512 (v16i32 VR512:$src1),
def : Pat <(i8 (int_x86_avx512_mask_ptestm_q_512 (v8i64 VR512:$src1),
(v8i64 VR512:$src2), (i8 -1))),
(COPY_TO_REGCLASS (VPTESTMQZrr VR512:$src1, VR512:$src2), GR8)>;
//===----------------------------------------------------------------------===//
// AVX-512 Shift instructions
//===----------------------------------------------------------------------===//
@ -3214,73 +3215,57 @@ multiclass avx512_shift_rmi<bits<8> opc, Format ImmFormR, Format ImmFormM,
}
multiclass avx512_shift_rrm<bits<8> opc, string OpcodeStr, SDNode OpNode,
RegisterClass RC, ValueType vt, ValueType SrcVT,
PatFrag bc_frag, RegisterClass KRC> {
// src2 is always 128-bit
def rr : AVX512BI<opc, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, VR128X:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst, (vt (OpNode RC:$src1, (SrcVT VR128X:$src2))))],
SSE_INTSHIFT_ITINS_P.rr>, EVEX_4V;
def rrk : AVX512BI<opc, MRMSrcReg, (outs RC:$dst),
(ins KRC:$mask, RC:$src1, VR128X:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2}"),
[], SSE_INTSHIFT_ITINS_P.rr>, EVEX_4V, EVEX_K;
def rm : AVX512BI<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, i128mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst, (vt (OpNode RC:$src1,
(bc_frag (memopv2i64 addr:$src2)))))],
SSE_INTSHIFT_ITINS_P.rm>, EVEX_4V;
def rmk : AVX512BI<opc, MRMSrcMem, (outs RC:$dst),
(ins KRC:$mask, RC:$src1, i128mem:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2}"),
[], SSE_INTSHIFT_ITINS_P.rm>, EVEX_4V, EVEX_K;
ValueType SrcVT, PatFrag bc_frag, X86VectorVTInfo _> {
// src2 is always 128-bit
defm rr : AVX512_maskable<opc, MRMSrcReg, _, (outs _.RC:$dst),
(ins _.RC:$src1, VR128X:$src2), OpcodeStr,
"$src2, $src1", "$src1, $src2",
(_.VT (OpNode _.RC:$src1, (SrcVT VR128X:$src2))),
" ", SSE_INTSHIFT_ITINS_P.rr>, AVX512BIBase, EVEX_4V;
defm rm : AVX512_maskable<opc, MRMSrcMem, _, (outs _.RC:$dst),
(ins _.RC:$src1, i128mem:$src2), OpcodeStr,
"$src2, $src1", "$src1, $src2",
(_.VT (OpNode _.RC:$src1, (bc_frag (memopv2i64 addr:$src2)))),
" ", SSE_INTSHIFT_ITINS_P.rm>, AVX512BIBase, EVEX_4V;
}
multiclass avx512_varshift_sizes<bits<8> opc, string OpcodeStr, SDNode OpNode,
ValueType SrcVT, PatFrag bc_frag, X86VectorVTInfo _> {
defm Z : avx512_shift_rrm<opc, OpcodeStr, OpNode, SrcVT, bc_frag, _>, EVEX_V512;
}
multiclass avx512_varshift_types<bits<8> opcd, bits<8> opcq, string OpcodeStr,
SDNode OpNode> {
defm D : avx512_varshift_sizes<opcd, OpcodeStr#"d", OpNode, v4i32, bc_v4i32,
v16i32_info>, EVEX_CD8<32, CD8VQ>;
defm Q : avx512_varshift_sizes<opcq, OpcodeStr#"q", OpNode, v2i64, bc_v2i64,
v8i64_info>, EVEX_CD8<64, CD8VQ>, VEX_W;
}
defm VPSRLDZ : avx512_shift_rmi<0x72, MRM2r, MRM2m, "vpsrld", X86vsrli,
v16i32_info>,
EVEX_V512, EVEX_CD8<32, CD8VF>;
defm VPSRLDZ : avx512_shift_rrm<0xD2, "vpsrld", X86vsrl,
VR512, v16i32, v4i32, bc_v4i32, VK16WM>, EVEX_V512,
EVEX_CD8<32, CD8VQ>;
defm VPSRLQZ : avx512_shift_rmi<0x73, MRM2r, MRM2m, "vpsrlq", X86vsrli,
v8i64_info>, EVEX_V512,
EVEX_CD8<64, CD8VF>, VEX_W;
defm VPSRLQZ : avx512_shift_rrm<0xD3, "vpsrlq", X86vsrl,
VR512, v8i64, v2i64, bc_v2i64, VK8WM>, EVEX_V512,
EVEX_CD8<64, CD8VQ>, VEX_W;
defm VPSLLDZ : avx512_shift_rmi<0x72, MRM6r, MRM6m, "vpslld", X86vshli,
v16i32_info>, EVEX_V512,
EVEX_CD8<32, CD8VF>;
defm VPSLLDZ : avx512_shift_rrm<0xF2, "vpslld", X86vshl,
VR512, v16i32, v4i32, bc_v4i32, VK16WM>, EVEX_V512,
EVEX_CD8<32, CD8VQ>;
defm VPSLLQZ : avx512_shift_rmi<0x73, MRM6r, MRM6m, "vpsllq", X86vshli,
v8i64_info>, EVEX_V512,
EVEX_CD8<64, CD8VF>, VEX_W;
defm VPSLLQZ : avx512_shift_rrm<0xF3, "vpsllq", X86vshl,
VR512, v8i64, v2i64, bc_v2i64, VK8WM>, EVEX_V512,
EVEX_CD8<64, CD8VQ>, VEX_W;
defm VPSRADZ : avx512_shift_rmi<0x72, MRM4r, MRM4m, "vpsrad", X86vsrai,
v16i32_info>,
EVEX_V512, EVEX_CD8<32, CD8VF>;
defm VPSRADZ : avx512_shift_rrm<0xE2, "vpsrad", X86vsra,
VR512, v16i32, v4i32, bc_v4i32, VK16WM>, EVEX_V512,
EVEX_CD8<32, CD8VQ>;
defm VPSRAQZ : avx512_shift_rmi<0x72, MRM4r, MRM4m, "vpsraq", X86vsrai,
v8i64_info>, EVEX_V512,
EVEX_CD8<64, CD8VF>, VEX_W;
defm VPSRAQZ : avx512_shift_rrm<0xE2, "vpsraq", X86vsra,
VR512, v8i64, v2i64, bc_v2i64, VK8WM>, EVEX_V512,
EVEX_CD8<64, CD8VQ>, VEX_W;
defm VPSRL : avx512_varshift_types<0xD2, 0xD3, "vpsrl", X86vsrl>;
defm VPSLL : avx512_varshift_types<0xF2, 0xF3, "vpsll", X86vshl>;
defm VPSRA : avx512_varshift_types<0xE2, 0xE2, "vpsra", X86vsra>;
//===-------------------------------------------------------------------===//
// Variable Bit Shifts

8
llvm/lib/Target/X86/X86IntrinsicsInfo.h
View File

@ -21,7 +21,7 @@ enum IntrinsicType {
GATHER, SCATTER, PREFETCH, RDSEED, RDRAND, RDPMC, RDTSC, XTEST, ADX,
INTR_TYPE_1OP, INTR_TYPE_2OP, INTR_TYPE_3OP,
CMP_MASK, CMP_MASK_CC, VSHIFT, VSHIFT_MASK, COMI,
INTR_TYPE_1OP_MASK_RM
INTR_TYPE_1OP_MASK_RM, INTR_TYPE_2OP_MASK
};
struct IntrinsicData {
@ -195,10 +195,16 @@ static const IntrinsicData IntrinsicsWithoutChain[] = {
X86_INTRINSIC_DATA(avx512_mask_pcmpgt_w_128, CMP_MASK, X86ISD::PCMPGTM, 0),
X86_INTRINSIC_DATA(avx512_mask_pcmpgt_w_256, CMP_MASK, X86ISD::PCMPGTM, 0),
X86_INTRINSIC_DATA(avx512_mask_pcmpgt_w_512, CMP_MASK, X86ISD::PCMPGTM, 0),
X86_INTRINSIC_DATA(avx512_mask_psll_d, INTR_TYPE_2OP_MASK, X86ISD::VSHL, 0),
X86_INTRINSIC_DATA(avx512_mask_psll_q, INTR_TYPE_2OP_MASK, X86ISD::VSHL, 0),
X86_INTRINSIC_DATA(avx512_mask_pslli_d, VSHIFT_MASK, X86ISD::VSHLI, 0),
X86_INTRINSIC_DATA(avx512_mask_pslli_q, VSHIFT_MASK, X86ISD::VSHLI, 0),
X86_INTRINSIC_DATA(avx512_mask_psra_d, INTR_TYPE_2OP_MASK, X86ISD::VSRA, 0),
X86_INTRINSIC_DATA(avx512_mask_psra_q, INTR_TYPE_2OP_MASK, X86ISD::VSRA, 0),
X86_INTRINSIC_DATA(avx512_mask_psrai_d, VSHIFT_MASK, X86ISD::VSRAI, 0),
X86_INTRINSIC_DATA(avx512_mask_psrai_q, VSHIFT_MASK, X86ISD::VSRAI, 0),
X86_INTRINSIC_DATA(avx512_mask_psrl_d, INTR_TYPE_2OP_MASK, X86ISD::VSRL, 0),
X86_INTRINSIC_DATA(avx512_mask_psrl_q, INTR_TYPE_2OP_MASK, X86ISD::VSRL, 0),
X86_INTRINSIC_DATA(avx512_mask_psrli_d, VSHIFT_MASK, X86ISD::VSRLI, 0),
X86_INTRINSIC_DATA(avx512_mask_psrli_q, VSHIFT_MASK, X86ISD::VSRLI, 0),
X86_INTRINSIC_DATA(avx512_mask_ucmp_b_128, CMP_MASK_CC, X86ISD::CMPMU, 0),

138
llvm/test/CodeGen/X86/avx512-intrinsics.ll
View File

@ -1088,3 +1088,141 @@ define <8 x i64> @test_x86_avx512_maskz_psrai_q(<8 x i64> %a0, i8 %mask) {
}
declare <8 x i64> @llvm.x86.avx512.mask.psrai.q(<8 x i64>, i32, <8 x i64>, i8) nounwind readnone
define <16 x i32> @test_x86_avx512_psll_d(<16 x i32> %a0, <4 x i32> %a1) {
; CHECK-LABEL: test_x86_avx512_psll_d
; CHECK: vpslld
%res = call <16 x i32> @llvm.x86.avx512.mask.psll.d(<16 x i32> %a0, <4 x i32> %a1, <16 x i32> zeroinitializer, i16 -1)
ret <16 x i32> %res
}
define <16 x i32> @test_x86_avx512_mask_psll_d(<16 x i32> %a0, <4 x i32> %a1, <16 x i32> %a2, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psll_d
; CHECK: vpslld %xmm1, %zmm0, %zmm2 {%k1}
%res = call <16 x i32> @llvm.x86.avx512.mask.psll.d(<16 x i32> %a0, <4 x i32> %a1, <16 x i32> %a2, i16 %mask)
ret <16 x i32> %res
}
define <16 x i32> @test_x86_avx512_maskz_psll_d(<16 x i32> %a0, <4 x i32> %a1, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psll_d
; CHECK: vpslld %xmm1, %zmm0, %zmm0 {%k1} {z}
%res = call <16 x i32> @llvm.x86.avx512.mask.psll.d(<16 x i32> %a0, <4 x i32> %a1, <16 x i32> zeroinitializer, i16 %mask)
ret <16 x i32> %res
}
declare <16 x i32> @llvm.x86.avx512.mask.psll.d(<16 x i32>, <4 x i32>, <16 x i32>, i16) nounwind readnone
define <8 x i64> @test_x86_avx512_psll_q(<8 x i64> %a0, <2 x i64> %a1) {
; CHECK-LABEL: test_x86_avx512_psll_q
; CHECK: vpsllq
%res = call <8 x i64> @llvm.x86.avx512.mask.psll.q(<8 x i64> %a0, <2 x i64> %a1, <8 x i64> zeroinitializer, i8 -1)
ret <8 x i64> %res
}
define <8 x i64> @test_x86_avx512_mask_psll_q(<8 x i64> %a0, <2 x i64> %a1, <8 x i64> %a2, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psll_q
; CHECK: vpsllq %xmm1, %zmm0, %zmm2 {%k1}
%res = call <8 x i64> @llvm.x86.avx512.mask.psll.q(<8 x i64> %a0, <2 x i64> %a1, <8 x i64> %a2, i8 %mask)
ret <8 x i64> %res
}
define <8 x i64> @test_x86_avx512_maskz_psll_q(<8 x i64> %a0, <2 x i64> %a1, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psll_q
; CHECK: vpsllq %xmm1, %zmm0, %zmm0 {%k1} {z}
%res = call <8 x i64> @llvm.x86.avx512.mask.psll.q(<8 x i64> %a0, <2 x i64> %a1, <8 x i64> zeroinitializer, i8 %mask)
ret <8 x i64> %res
}
declare <8 x i64> @llvm.x86.avx512.mask.psll.q(<8 x i64>, <2 x i64>, <8 x i64>, i8) nounwind readnone
define <16 x i32> @test_x86_avx512_psrl_d(<16 x i32> %a0, <4 x i32> %a1) {
; CHECK-LABEL: test_x86_avx512_psrl_d
; CHECK: vpsrld
%res = call <16 x i32> @llvm.x86.avx512.mask.psrl.d(<16 x i32> %a0, <4 x i32> %a1, <16 x i32> zeroinitializer, i16 -1)
ret <16 x i32> %res
}
define <16 x i32> @test_x86_avx512_mask_psrl_d(<16 x i32> %a0, <4 x i32> %a1, <16 x i32> %a2, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psrl_d
; CHECK: vpsrld %xmm1, %zmm0, %zmm2 {%k1}
%res = call <16 x i32> @llvm.x86.avx512.mask.psrl.d(<16 x i32> %a0, <4 x i32> %a1, <16 x i32> %a2, i16 %mask)
ret <16 x i32> %res
}
define <16 x i32> @test_x86_avx512_maskz_psrl_d(<16 x i32> %a0, <4 x i32> %a1, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psrl_d
; CHECK: vpsrld %xmm1, %zmm0, %zmm0 {%k1} {z}
%res = call <16 x i32> @llvm.x86.avx512.mask.psrl.d(<16 x i32> %a0, <4 x i32> %a1, <16 x i32> zeroinitializer, i16 %mask)
ret <16 x i32> %res
}
declare <16 x i32> @llvm.x86.avx512.mask.psrl.d(<16 x i32>, <4 x i32>, <16 x i32>, i16) nounwind readnone
define <8 x i64> @test_x86_avx512_psrl_q(<8 x i64> %a0, <2 x i64> %a1) {
; CHECK-LABEL: test_x86_avx512_psrl_q
; CHECK: vpsrlq
%res = call <8 x i64> @llvm.x86.avx512.mask.psrl.q(<8 x i64> %a0, <2 x i64> %a1, <8 x i64> zeroinitializer, i8 -1)
ret <8 x i64> %res
}
define <8 x i64> @test_x86_avx512_mask_psrl_q(<8 x i64> %a0, <2 x i64> %a1, <8 x i64> %a2, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psrl_q
; CHECK: vpsrlq %xmm1, %zmm0, %zmm2 {%k1}
%res = call <8 x i64> @llvm.x86.avx512.mask.psrl.q(<8 x i64> %a0, <2 x i64> %a1, <8 x i64> %a2, i8 %mask)
ret <8 x i64> %res
}
define <8 x i64> @test_x86_avx512_maskz_psrl_q(<8 x i64> %a0, <2 x i64> %a1, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psrl_q
; CHECK: vpsrlq %xmm1, %zmm0, %zmm0 {%k1} {z}
%res = call <8 x i64> @llvm.x86.avx512.mask.psrl.q(<8 x i64> %a0, <2 x i64> %a1, <8 x i64> zeroinitializer, i8 %mask)
ret <8 x i64> %res
}
declare <8 x i64> @llvm.x86.avx512.mask.psrl.q(<8 x i64>, <2 x i64>, <8 x i64>, i8) nounwind readnone
define <16 x i32> @test_x86_avx512_psra_d(<16 x i32> %a0, <4 x i32> %a1) {
; CHECK-LABEL: test_x86_avx512_psra_d
; CHECK: vpsrad
%res = call <16 x i32> @llvm.x86.avx512.mask.psra.d(<16 x i32> %a0, <4 x i32> %a1, <16 x i32> zeroinitializer, i16 -1)
ret <16 x i32> %res
}
define <16 x i32> @test_x86_avx512_mask_psra_d(<16 x i32> %a0, <4 x i32> %a1, <16 x i32> %a2, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psra_d
; CHECK: vpsrad %xmm1, %zmm0, %zmm2 {%k1}
%res = call <16 x i32> @llvm.x86.avx512.mask.psra.d(<16 x i32> %a0, <4 x i32> %a1, <16 x i32> %a2, i16 %mask)
ret <16 x i32> %res
}
define <16 x i32> @test_x86_avx512_maskz_psra_d(<16 x i32> %a0, <4 x i32> %a1, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psra_d
; CHECK: vpsrad %xmm1, %zmm0, %zmm0 {%k1} {z}
%res = call <16 x i32> @llvm.x86.avx512.mask.psra.d(<16 x i32> %a0, <4 x i32> %a1, <16 x i32> zeroinitializer, i16 %mask)
ret <16 x i32> %res
}
declare <16 x i32> @llvm.x86.avx512.mask.psra.d(<16 x i32>, <4 x i32>, <16 x i32>, i16) nounwind readnone
define <8 x i64> @test_x86_avx512_psra_q(<8 x i64> %a0, <2 x i64> %a1) {
; CHECK-LABEL: test_x86_avx512_psra_q
; CHECK: vpsraq
%res = call <8 x i64> @llvm.x86.avx512.mask.psra.q(<8 x i64> %a0, <2 x i64> %a1, <8 x i64> zeroinitializer, i8 -1)
ret <8 x i64> %res
}
define <8 x i64> @test_x86_avx512_mask_psra_q(<8 x i64> %a0, <2 x i64> %a1, <8 x i64> %a2, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psra_q
; CHECK: vpsraq %xmm1, %zmm0, %zmm2 {%k1}
%res = call <8 x i64> @llvm.x86.avx512.mask.psra.q(<8 x i64> %a0, <2 x i64> %a1, <8 x i64> %a2, i8 %mask)
ret <8 x i64> %res
}
define <8 x i64> @test_x86_avx512_maskz_psra_q(<8 x i64> %a0, <2 x i64> %a1, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psra_q
; CHECK: vpsraq %xmm1, %zmm0, %zmm0 {%k1} {z}
%res = call <8 x i64> @llvm.x86.avx512.mask.psra.q(<8 x i64> %a0, <2 x i64> %a1, <8 x i64> zeroinitializer, i8 %mask)
ret <8 x i64> %res
}
declare <8 x i64> @llvm.x86.avx512.mask.psra.q(<8 x i64>, <2 x i64>, <8 x i64>, i8) nounwind readnone