[llvm] r344447 - [X86][SSE] Change CTTZ vector lowering to cttz(x) = ctpop(~x & (x - 1))
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Sat Oct 13 05:12:06 PDT 2018
Author: rksimon
Date: Sat Oct 13 05:12:06 2018
New Revision: 344447
URL: http://llvm.org/viewvc/llvm-project?rev=344447&view=rev
Log:
[X86][SSE] Change CTTZ vector lowering to cttz(x) = ctpop(~x & (x - 1))
This patch changes the vector CTTZ lowering from:
cttz(x) = ctpop((x & -x) - 1)
to:
cttz(x) = ctpop(~x & (x - 1))
Not only does this make better use of the PANDN instruction, but it also matches the LegalizeDAG method which should allow us to remove the x86 specific code at some point in the future (we need to fix some issues with the bitcasted logic ops and CTPOP lowering first).
Differential Revision: https://reviews.llvm.org/D53214
Modified:
llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
llvm/trunk/test/CodeGen/X86/vec_ctbits.ll
llvm/trunk/test/CodeGen/X86/vector-tzcnt-128.ll
llvm/trunk/test/CodeGen/X86/vector-tzcnt-256.ll
llvm/trunk/test/CodeGen/X86/vector-tzcnt-512.ll
Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=344447&r1=344446&r2=344447&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Sat Oct 13 05:12:06 2018
@@ -22968,7 +22968,8 @@ static SDValue LowerCTLZ(SDValue Op, con
return Op;
}
-static SDValue LowerCTTZ(SDValue Op, SelectionDAG &DAG) {
+static SDValue LowerCTTZ(SDValue Op, const X86Subtarget &Subtarget,
+ SelectionDAG &DAG) {
MVT VT = Op.getSimpleValueType();
unsigned NumBits = VT.getScalarSizeInBits();
SDLoc dl(Op);
@@ -22977,21 +22978,24 @@ static SDValue LowerCTTZ(SDValue Op, Sel
SDValue N0 = Op.getOperand(0);
SDValue Zero = DAG.getConstant(0, dl, VT);
- // lsb(x) = (x & -x)
- SDValue LSB = DAG.getNode(ISD::AND, dl, VT, N0,
- DAG.getNode(ISD::SUB, dl, VT, Zero, N0));
+ // Decompose 256-bit ops into smaller 128-bit ops.
+ if (VT.is256BitVector() && !Subtarget.hasInt256())
+ return Lower256IntUnary(Op, DAG);
- // cttz_undef(x) = (width - 1) - ctlz(lsb)
+ // cttz_undef(x) = (width - 1) - ctlz(x & -x)
if (Op.getOpcode() == ISD::CTTZ_ZERO_UNDEF) {
SDValue WidthMinusOne = DAG.getConstant(NumBits - 1, dl, VT);
+ SDValue LSB = DAG.getNode(ISD::AND, dl, VT, N0,
+ DAG.getNode(ISD::SUB, dl, VT, Zero, N0));
return DAG.getNode(ISD::SUB, dl, VT, WidthMinusOne,
DAG.getNode(ISD::CTLZ, dl, VT, LSB));
}
- // cttz(x) = ctpop(lsb - 1)
+ // cttz(x) = ctpop(~x & (x - 1))
SDValue One = DAG.getConstant(1, dl, VT);
return DAG.getNode(ISD::CTPOP, dl, VT,
- DAG.getNode(ISD::SUB, dl, VT, LSB, One));
+ DAG.getNode(ISD::AND, dl, VT, DAG.getNOT(dl, N0, VT),
+ DAG.getNode(ISD::SUB, dl, VT, N0, One)));
}
assert(Op.getOpcode() == ISD::CTTZ &&
@@ -25918,7 +25922,7 @@ SDValue X86TargetLowering::LowerOperatio
case ISD::CTLZ:
case ISD::CTLZ_ZERO_UNDEF: return LowerCTLZ(Op, Subtarget, DAG);
case ISD::CTTZ:
- case ISD::CTTZ_ZERO_UNDEF: return LowerCTTZ(Op, DAG);
+ case ISD::CTTZ_ZERO_UNDEF: return LowerCTTZ(Op, Subtarget, DAG);
case ISD::MUL: return LowerMUL(Op, Subtarget, DAG);
case ISD::MULHS:
case ISD::MULHU: return LowerMULH(Op, Subtarget, DAG);
Modified: llvm/trunk/test/CodeGen/X86/vec_ctbits.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vec_ctbits.ll?rev=344447&r1=344446&r2=344447&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vec_ctbits.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vec_ctbits.ll Sat Oct 13 05:12:06 2018
@@ -8,27 +8,26 @@ declare <2 x i64> @llvm.ctpop.v2i64(<2 x
define <2 x i64> @footz(<2 x i64> %a) nounwind {
; CHECK-LABEL: footz:
; CHECK: # %bb.0:
-; CHECK-NEXT: pxor %xmm1, %xmm1
-; CHECK-NEXT: pxor %xmm2, %xmm2
-; CHECK-NEXT: psubq %xmm0, %xmm2
-; CHECK-NEXT: pand %xmm0, %xmm2
-; CHECK-NEXT: pcmpeqd %xmm3, %xmm3
-; CHECK-NEXT: paddq %xmm2, %xmm3
-; CHECK-NEXT: movdqa %xmm3, %xmm0
-; CHECK-NEXT: psrlw $1, %xmm0
-; CHECK-NEXT: pand {{.*}}(%rip), %xmm0
-; CHECK-NEXT: psubb %xmm0, %xmm3
-; CHECK-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; CHECK-NEXT: movdqa %xmm3, %xmm2
-; CHECK-NEXT: pand %xmm0, %xmm2
-; CHECK-NEXT: psrlw $2, %xmm3
-; CHECK-NEXT: pand %xmm0, %xmm3
-; CHECK-NEXT: paddb %xmm2, %xmm3
-; CHECK-NEXT: movdqa %xmm3, %xmm0
-; CHECK-NEXT: psrlw $4, %xmm0
-; CHECK-NEXT: paddb %xmm3, %xmm0
-; CHECK-NEXT: pand {{.*}}(%rip), %xmm0
-; CHECK-NEXT: psadbw %xmm1, %xmm0
+; CHECK-NEXT: pcmpeqd %xmm1, %xmm1
+; CHECK-NEXT: paddq %xmm0, %xmm1
+; CHECK-NEXT: pandn %xmm1, %xmm0
+; CHECK-NEXT: movdqa %xmm0, %xmm1
+; CHECK-NEXT: psrlw $1, %xmm1
+; CHECK-NEXT: pand {{.*}}(%rip), %xmm1
+; CHECK-NEXT: psubb %xmm1, %xmm0
+; CHECK-NEXT: movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; CHECK-NEXT: movdqa %xmm0, %xmm2
+; CHECK-NEXT: pand %xmm1, %xmm2
+; CHECK-NEXT: psrlw $2, %xmm0
+; CHECK-NEXT: pand %xmm1, %xmm0
+; CHECK-NEXT: paddb %xmm2, %xmm0
+; CHECK-NEXT: movdqa %xmm0, %xmm1
+; CHECK-NEXT: psrlw $4, %xmm1
+; CHECK-NEXT: paddb %xmm0, %xmm1
+; CHECK-NEXT: pand {{.*}}(%rip), %xmm1
+; CHECK-NEXT: pxor %xmm0, %xmm0
+; CHECK-NEXT: psadbw %xmm0, %xmm1
+; CHECK-NEXT: movdqa %xmm1, %xmm0
; CHECK-NEXT: retq
%c = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
ret <2 x i64> %c
@@ -112,27 +111,26 @@ define <2 x i32> @promtz(<2 x i32> %a) n
; CHECK-LABEL: promtz:
; CHECK: # %bb.0:
; CHECK-NEXT: por {{.*}}(%rip), %xmm0
-; CHECK-NEXT: pxor %xmm1, %xmm1
-; CHECK-NEXT: pxor %xmm2, %xmm2
-; CHECK-NEXT: psubq %xmm0, %xmm2
-; CHECK-NEXT: pand %xmm0, %xmm2
-; CHECK-NEXT: pcmpeqd %xmm3, %xmm3
-; CHECK-NEXT: paddq %xmm2, %xmm3
-; CHECK-NEXT: movdqa %xmm3, %xmm0
-; CHECK-NEXT: psrlw $1, %xmm0
-; CHECK-NEXT: pand {{.*}}(%rip), %xmm0
-; CHECK-NEXT: psubb %xmm0, %xmm3
-; CHECK-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; CHECK-NEXT: movdqa %xmm3, %xmm2
-; CHECK-NEXT: pand %xmm0, %xmm2
-; CHECK-NEXT: psrlw $2, %xmm3
-; CHECK-NEXT: pand %xmm0, %xmm3
-; CHECK-NEXT: paddb %xmm2, %xmm3
-; CHECK-NEXT: movdqa %xmm3, %xmm0
-; CHECK-NEXT: psrlw $4, %xmm0
-; CHECK-NEXT: paddb %xmm3, %xmm0
-; CHECK-NEXT: pand {{.*}}(%rip), %xmm0
-; CHECK-NEXT: psadbw %xmm1, %xmm0
+; CHECK-NEXT: pcmpeqd %xmm1, %xmm1
+; CHECK-NEXT: paddq %xmm0, %xmm1
+; CHECK-NEXT: pandn %xmm1, %xmm0
+; CHECK-NEXT: movdqa %xmm0, %xmm1
+; CHECK-NEXT: psrlw $1, %xmm1
+; CHECK-NEXT: pand {{.*}}(%rip), %xmm1
+; CHECK-NEXT: psubb %xmm1, %xmm0
+; CHECK-NEXT: movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; CHECK-NEXT: movdqa %xmm0, %xmm2
+; CHECK-NEXT: pand %xmm1, %xmm2
+; CHECK-NEXT: psrlw $2, %xmm0
+; CHECK-NEXT: pand %xmm1, %xmm0
+; CHECK-NEXT: paddb %xmm2, %xmm0
+; CHECK-NEXT: movdqa %xmm0, %xmm1
+; CHECK-NEXT: psrlw $4, %xmm1
+; CHECK-NEXT: paddb %xmm0, %xmm1
+; CHECK-NEXT: pand {{.*}}(%rip), %xmm1
+; CHECK-NEXT: pxor %xmm0, %xmm0
+; CHECK-NEXT: psadbw %xmm0, %xmm1
+; CHECK-NEXT: movdqa %xmm1, %xmm0
; CHECK-NEXT: retq
%c = call <2 x i32> @llvm.cttz.v2i32(<2 x i32> %a, i1 false)
ret <2 x i32> %c
Modified: llvm/trunk/test/CodeGen/X86/vector-tzcnt-128.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-tzcnt-128.ll?rev=344447&r1=344446&r2=344447&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vector-tzcnt-128.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vector-tzcnt-128.ll Sat Oct 13 05:12:06 2018
@@ -18,121 +18,112 @@
define <2 x i64> @testv2i64(<2 x i64> %in) nounwind {
; SSE2-LABEL: testv2i64:
; SSE2: # %bb.0:
-; SSE2-NEXT: pxor %xmm1, %xmm1
-; SSE2-NEXT: pxor %xmm2, %xmm2
-; SSE2-NEXT: psubq %xmm0, %xmm2
-; SSE2-NEXT: pand %xmm0, %xmm2
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm3
-; SSE2-NEXT: paddq %xmm2, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: psrlw $1, %xmm0
-; SSE2-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE2-NEXT: psubb %xmm0, %xmm3
-; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE2-NEXT: movdqa %xmm3, %xmm2
-; SSE2-NEXT: pand %xmm0, %xmm2
-; SSE2-NEXT: psrlw $2, %xmm3
-; SSE2-NEXT: pand %xmm0, %xmm3
-; SSE2-NEXT: paddb %xmm2, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: psrlw $4, %xmm0
-; SSE2-NEXT: paddb %xmm3, %xmm0
-; SSE2-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE2-NEXT: psadbw %xmm1, %xmm0
+; SSE2-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE2-NEXT: paddq %xmm0, %xmm1
+; SSE2-NEXT: pandn %xmm1, %xmm0
+; SSE2-NEXT: movdqa %xmm0, %xmm1
+; SSE2-NEXT: psrlw $1, %xmm1
+; SSE2-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT: psubb %xmm1, %xmm0
+; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; SSE2-NEXT: movdqa %xmm0, %xmm2
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: psrlw $2, %xmm0
+; SSE2-NEXT: pand %xmm1, %xmm0
+; SSE2-NEXT: paddb %xmm2, %xmm0
+; SSE2-NEXT: movdqa %xmm0, %xmm1
+; SSE2-NEXT: psrlw $4, %xmm1
+; SSE2-NEXT: paddb %xmm0, %xmm1
+; SSE2-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT: pxor %xmm0, %xmm0
+; SSE2-NEXT: psadbw %xmm0, %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE3-LABEL: testv2i64:
; SSE3: # %bb.0:
-; SSE3-NEXT: pxor %xmm1, %xmm1
-; SSE3-NEXT: pxor %xmm2, %xmm2
-; SSE3-NEXT: psubq %xmm0, %xmm2
-; SSE3-NEXT: pand %xmm0, %xmm2
-; SSE3-NEXT: pcmpeqd %xmm3, %xmm3
-; SSE3-NEXT: paddq %xmm2, %xmm3
-; SSE3-NEXT: movdqa %xmm3, %xmm0
-; SSE3-NEXT: psrlw $1, %xmm0
-; SSE3-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE3-NEXT: psubb %xmm0, %xmm3
-; SSE3-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE3-NEXT: movdqa %xmm3, %xmm2
-; SSE3-NEXT: pand %xmm0, %xmm2
-; SSE3-NEXT: psrlw $2, %xmm3
-; SSE3-NEXT: pand %xmm0, %xmm3
-; SSE3-NEXT: paddb %xmm2, %xmm3
-; SSE3-NEXT: movdqa %xmm3, %xmm0
-; SSE3-NEXT: psrlw $4, %xmm0
-; SSE3-NEXT: paddb %xmm3, %xmm0
-; SSE3-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE3-NEXT: psadbw %xmm1, %xmm0
+; SSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE3-NEXT: paddq %xmm0, %xmm1
+; SSE3-NEXT: pandn %xmm1, %xmm0
+; SSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE3-NEXT: psrlw $1, %xmm1
+; SSE3-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE3-NEXT: psubb %xmm1, %xmm0
+; SSE3-NEXT: movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; SSE3-NEXT: movdqa %xmm0, %xmm2
+; SSE3-NEXT: pand %xmm1, %xmm2
+; SSE3-NEXT: psrlw $2, %xmm0
+; SSE3-NEXT: pand %xmm1, %xmm0
+; SSE3-NEXT: paddb %xmm2, %xmm0
+; SSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE3-NEXT: psrlw $4, %xmm1
+; SSE3-NEXT: paddb %xmm0, %xmm1
+; SSE3-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE3-NEXT: pxor %xmm0, %xmm0
+; SSE3-NEXT: psadbw %xmm0, %xmm1
+; SSE3-NEXT: movdqa %xmm1, %xmm0
; SSE3-NEXT: retq
;
; SSSE3-LABEL: testv2i64:
; SSSE3: # %bb.0:
-; SSSE3-NEXT: pxor %xmm1, %xmm1
-; SSSE3-NEXT: pxor %xmm2, %xmm2
-; SSSE3-NEXT: psubq %xmm0, %xmm2
-; SSSE3-NEXT: pand %xmm0, %xmm2
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm3
-; SSSE3-NEXT: paddq %xmm2, %xmm3
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSSE3-NEXT: paddq %xmm0, %xmm1
+; SSSE3-NEXT: pandn %xmm1, %xmm0
+; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSSE3-NEXT: movdqa %xmm0, %xmm2
+; SSSE3-NEXT: pand %xmm1, %xmm2
+; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
; SSSE3-NEXT: movdqa %xmm3, %xmm4
-; SSSE3-NEXT: pand %xmm2, %xmm4
-; SSSE3-NEXT: movdqa {{.*#+}} xmm0 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSSE3-NEXT: movdqa %xmm0, %xmm5
-; SSSE3-NEXT: pshufb %xmm4, %xmm5
-; SSSE3-NEXT: psrlw $4, %xmm3
-; SSSE3-NEXT: pand %xmm2, %xmm3
-; SSSE3-NEXT: pshufb %xmm3, %xmm0
-; SSSE3-NEXT: paddb %xmm5, %xmm0
-; SSSE3-NEXT: psadbw %xmm1, %xmm0
+; SSSE3-NEXT: pshufb %xmm2, %xmm4
+; SSSE3-NEXT: psrlw $4, %xmm0
+; SSSE3-NEXT: pand %xmm1, %xmm0
+; SSSE3-NEXT: pshufb %xmm0, %xmm3
+; SSSE3-NEXT: paddb %xmm4, %xmm3
+; SSSE3-NEXT: pxor %xmm0, %xmm0
+; SSSE3-NEXT: psadbw %xmm3, %xmm0
; SSSE3-NEXT: retq
;
; SSE41-LABEL: testv2i64:
; SSE41: # %bb.0:
-; SSE41-NEXT: pxor %xmm1, %xmm1
-; SSE41-NEXT: pxor %xmm2, %xmm2
-; SSE41-NEXT: psubq %xmm0, %xmm2
-; SSE41-NEXT: pand %xmm0, %xmm2
-; SSE41-NEXT: pcmpeqd %xmm3, %xmm3
-; SSE41-NEXT: paddq %xmm2, %xmm3
-; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSE41-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT: paddq %xmm0, %xmm1
+; SSE41-NEXT: pandn %xmm1, %xmm0
+; SSE41-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSE41-NEXT: movdqa %xmm0, %xmm2
+; SSE41-NEXT: pand %xmm1, %xmm2
+; SSE41-NEXT: movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
; SSE41-NEXT: movdqa %xmm3, %xmm4
-; SSE41-NEXT: pand %xmm2, %xmm4
-; SSE41-NEXT: movdqa {{.*#+}} xmm0 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSE41-NEXT: movdqa %xmm0, %xmm5
-; SSE41-NEXT: pshufb %xmm4, %xmm5
-; SSE41-NEXT: psrlw $4, %xmm3
-; SSE41-NEXT: pand %xmm2, %xmm3
-; SSE41-NEXT: pshufb %xmm3, %xmm0
-; SSE41-NEXT: paddb %xmm5, %xmm0
-; SSE41-NEXT: psadbw %xmm1, %xmm0
+; SSE41-NEXT: pshufb %xmm2, %xmm4
+; SSE41-NEXT: psrlw $4, %xmm0
+; SSE41-NEXT: pand %xmm1, %xmm0
+; SSE41-NEXT: pshufb %xmm0, %xmm3
+; SSE41-NEXT: paddb %xmm4, %xmm3
+; SSE41-NEXT: pxor %xmm0, %xmm0
+; SSE41-NEXT: psadbw %xmm3, %xmm0
; SSE41-NEXT: retq
;
; AVX-LABEL: testv2i64:
; AVX: # %bb.0:
-; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT: vpsubq %xmm0, %xmm1, %xmm2
-; AVX-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; AVX-NEXT: vpaddq %xmm2, %xmm0, %xmm0
-; AVX-NEXT: vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX-NEXT: vpand %xmm2, %xmm0, %xmm3
-; AVX-NEXT: vmovdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX-NEXT: vpshufb %xmm3, %xmm4, %xmm3
+; AVX-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX-NEXT: vpaddq %xmm1, %xmm0, %xmm1
+; AVX-NEXT: vpandn %xmm1, %xmm0, %xmm0
+; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX-NEXT: vpand %xmm1, %xmm0, %xmm2
+; AVX-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX-NEXT: vpshufb %xmm2, %xmm3, %xmm2
; AVX-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX-NEXT: vpshufb %xmm0, %xmm4, %xmm0
-; AVX-NEXT: vpaddb %xmm3, %xmm0, %xmm0
+; AVX-NEXT: vpand %xmm1, %xmm0, %xmm0
+; AVX-NEXT: vpshufb %xmm0, %xmm3, %xmm0
+; AVX-NEXT: vpaddb %xmm2, %xmm0, %xmm0
+; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
; AVX512VPOPCNTDQ-LABEL: testv2i64:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubq %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpaddq %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQ-NEXT: vpaddq %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpopcntq %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512VPOPCNTDQ-NEXT: vzeroupper
@@ -140,55 +131,50 @@ define <2 x i64> @testv2i64(<2 x i64> %i
;
; AVX512VPOPCNTDQVL-LABEL: testv2i64:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubq %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddq %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddq %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpopcntq %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: retq
;
; BITALG_NOVLX-LABEL: testv2i64:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubq %xmm0, %xmm1, %xmm2
-; BITALG_NOVLX-NEXT: vpand %xmm2, %xmm0, %xmm0
-; BITALG_NOVLX-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; BITALG_NOVLX-NEXT: vpaddq %xmm2, %xmm0, %xmm0
+; BITALG_NOVLX-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; BITALG_NOVLX-NEXT: vpaddq %xmm1, %xmm0, %xmm1
+; BITALG_NOVLX-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG_NOVLX-NEXT: vpopcntb %zmm0, %zmm0
+; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG_NOVLX-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
; BITALG_NOVLX-NEXT: vzeroupper
; BITALG_NOVLX-NEXT: retq
;
; BITALG-LABEL: testv2i64:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubq %xmm0, %xmm1, %xmm2
-; BITALG-NEXT: vpand %xmm2, %xmm0, %xmm0
-; BITALG-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; BITALG-NEXT: vpaddq %xmm2, %xmm0, %xmm0
+; BITALG-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; BITALG-NEXT: vpaddq %xmm1, %xmm0, %xmm1
+; BITALG-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG-NEXT: vpopcntb %xmm0, %xmm0
+; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
; BITALG-NEXT: retq
;
; X32-SSE-LABEL: testv2i64:
; X32-SSE: # %bb.0:
-; X32-SSE-NEXT: pxor %xmm1, %xmm1
-; X32-SSE-NEXT: pxor %xmm2, %xmm2
-; X32-SSE-NEXT: psubq %xmm0, %xmm2
-; X32-SSE-NEXT: pand %xmm0, %xmm2
-; X32-SSE-NEXT: psubq {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT: movdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-SSE-NEXT: movdqa %xmm2, %xmm4
-; X32-SSE-NEXT: pand %xmm3, %xmm4
-; X32-SSE-NEXT: movdqa {{.*#+}} xmm0 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-SSE-NEXT: movdqa %xmm0, %xmm5
-; X32-SSE-NEXT: pshufb %xmm4, %xmm5
-; X32-SSE-NEXT: psrlw $4, %xmm2
-; X32-SSE-NEXT: pand %xmm3, %xmm2
-; X32-SSE-NEXT: pshufb %xmm2, %xmm0
-; X32-SSE-NEXT: paddb %xmm5, %xmm0
-; X32-SSE-NEXT: psadbw %xmm1, %xmm0
+; X32-SSE-NEXT: movdqa %xmm0, %xmm1
+; X32-SSE-NEXT: psubq {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT: pandn %xmm1, %xmm0
+; X32-SSE-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X32-SSE-NEXT: movdqa %xmm0, %xmm2
+; X32-SSE-NEXT: pand %xmm1, %xmm2
+; X32-SSE-NEXT: movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X32-SSE-NEXT: movdqa %xmm3, %xmm4
+; X32-SSE-NEXT: pshufb %xmm2, %xmm4
+; X32-SSE-NEXT: psrlw $4, %xmm0
+; X32-SSE-NEXT: pand %xmm1, %xmm0
+; X32-SSE-NEXT: pshufb %xmm0, %xmm3
+; X32-SSE-NEXT: paddb %xmm4, %xmm3
+; X32-SSE-NEXT: pxor %xmm0, %xmm0
+; X32-SSE-NEXT: psadbw %xmm3, %xmm0
; X32-SSE-NEXT: retl
%out = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %in, i1 0)
ret <2 x i64> %out
@@ -197,129 +183,121 @@ define <2 x i64> @testv2i64(<2 x i64> %i
define <2 x i64> @testv2i64u(<2 x i64> %in) nounwind {
; SSE2-LABEL: testv2i64u:
; SSE2: # %bb.0:
-; SSE2-NEXT: pxor %xmm1, %xmm1
-; SSE2-NEXT: pxor %xmm2, %xmm2
-; SSE2-NEXT: psubq %xmm0, %xmm2
-; SSE2-NEXT: pand %xmm0, %xmm2
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm3
-; SSE2-NEXT: paddq %xmm2, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: psrlw $1, %xmm0
-; SSE2-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE2-NEXT: psubb %xmm0, %xmm3
-; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE2-NEXT: movdqa %xmm3, %xmm2
-; SSE2-NEXT: pand %xmm0, %xmm2
-; SSE2-NEXT: psrlw $2, %xmm3
-; SSE2-NEXT: pand %xmm0, %xmm3
-; SSE2-NEXT: paddb %xmm2, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: psrlw $4, %xmm0
-; SSE2-NEXT: paddb %xmm3, %xmm0
-; SSE2-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE2-NEXT: psadbw %xmm1, %xmm0
+; SSE2-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE2-NEXT: paddq %xmm0, %xmm1
+; SSE2-NEXT: pandn %xmm1, %xmm0
+; SSE2-NEXT: movdqa %xmm0, %xmm1
+; SSE2-NEXT: psrlw $1, %xmm1
+; SSE2-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT: psubb %xmm1, %xmm0
+; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; SSE2-NEXT: movdqa %xmm0, %xmm2
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: psrlw $2, %xmm0
+; SSE2-NEXT: pand %xmm1, %xmm0
+; SSE2-NEXT: paddb %xmm2, %xmm0
+; SSE2-NEXT: movdqa %xmm0, %xmm1
+; SSE2-NEXT: psrlw $4, %xmm1
+; SSE2-NEXT: paddb %xmm0, %xmm1
+; SSE2-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT: pxor %xmm0, %xmm0
+; SSE2-NEXT: psadbw %xmm0, %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE3-LABEL: testv2i64u:
; SSE3: # %bb.0:
-; SSE3-NEXT: pxor %xmm1, %xmm1
-; SSE3-NEXT: pxor %xmm2, %xmm2
-; SSE3-NEXT: psubq %xmm0, %xmm2
-; SSE3-NEXT: pand %xmm0, %xmm2
-; SSE3-NEXT: pcmpeqd %xmm3, %xmm3
-; SSE3-NEXT: paddq %xmm2, %xmm3
-; SSE3-NEXT: movdqa %xmm3, %xmm0
-; SSE3-NEXT: psrlw $1, %xmm0
-; SSE3-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE3-NEXT: psubb %xmm0, %xmm3
-; SSE3-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE3-NEXT: movdqa %xmm3, %xmm2
-; SSE3-NEXT: pand %xmm0, %xmm2
-; SSE3-NEXT: psrlw $2, %xmm3
-; SSE3-NEXT: pand %xmm0, %xmm3
-; SSE3-NEXT: paddb %xmm2, %xmm3
-; SSE3-NEXT: movdqa %xmm3, %xmm0
-; SSE3-NEXT: psrlw $4, %xmm0
-; SSE3-NEXT: paddb %xmm3, %xmm0
-; SSE3-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE3-NEXT: psadbw %xmm1, %xmm0
+; SSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE3-NEXT: paddq %xmm0, %xmm1
+; SSE3-NEXT: pandn %xmm1, %xmm0
+; SSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE3-NEXT: psrlw $1, %xmm1
+; SSE3-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE3-NEXT: psubb %xmm1, %xmm0
+; SSE3-NEXT: movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; SSE3-NEXT: movdqa %xmm0, %xmm2
+; SSE3-NEXT: pand %xmm1, %xmm2
+; SSE3-NEXT: psrlw $2, %xmm0
+; SSE3-NEXT: pand %xmm1, %xmm0
+; SSE3-NEXT: paddb %xmm2, %xmm0
+; SSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE3-NEXT: psrlw $4, %xmm1
+; SSE3-NEXT: paddb %xmm0, %xmm1
+; SSE3-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE3-NEXT: pxor %xmm0, %xmm0
+; SSE3-NEXT: psadbw %xmm0, %xmm1
+; SSE3-NEXT: movdqa %xmm1, %xmm0
; SSE3-NEXT: retq
;
; SSSE3-LABEL: testv2i64u:
; SSSE3: # %bb.0:
-; SSSE3-NEXT: pxor %xmm1, %xmm1
-; SSSE3-NEXT: pxor %xmm2, %xmm2
-; SSSE3-NEXT: psubq %xmm0, %xmm2
-; SSSE3-NEXT: pand %xmm0, %xmm2
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm3
-; SSSE3-NEXT: paddq %xmm2, %xmm3
-; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSSE3-NEXT: paddq %xmm0, %xmm1
+; SSSE3-NEXT: pandn %xmm1, %xmm0
+; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSSE3-NEXT: movdqa %xmm0, %xmm2
+; SSSE3-NEXT: pand %xmm1, %xmm2
+; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
; SSSE3-NEXT: movdqa %xmm3, %xmm4
-; SSSE3-NEXT: pand %xmm2, %xmm4
-; SSSE3-NEXT: movdqa {{.*#+}} xmm0 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSSE3-NEXT: movdqa %xmm0, %xmm5
-; SSSE3-NEXT: pshufb %xmm4, %xmm5
-; SSSE3-NEXT: psrlw $4, %xmm3
-; SSSE3-NEXT: pand %xmm2, %xmm3
-; SSSE3-NEXT: pshufb %xmm3, %xmm0
-; SSSE3-NEXT: paddb %xmm5, %xmm0
-; SSSE3-NEXT: psadbw %xmm1, %xmm0
+; SSSE3-NEXT: pshufb %xmm2, %xmm4
+; SSSE3-NEXT: psrlw $4, %xmm0
+; SSSE3-NEXT: pand %xmm1, %xmm0
+; SSSE3-NEXT: pshufb %xmm0, %xmm3
+; SSSE3-NEXT: paddb %xmm4, %xmm3
+; SSSE3-NEXT: pxor %xmm0, %xmm0
+; SSSE3-NEXT: psadbw %xmm3, %xmm0
; SSSE3-NEXT: retq
;
; SSE41-LABEL: testv2i64u:
; SSE41: # %bb.0:
-; SSE41-NEXT: pxor %xmm1, %xmm1
-; SSE41-NEXT: pxor %xmm2, %xmm2
-; SSE41-NEXT: psubq %xmm0, %xmm2
-; SSE41-NEXT: pand %xmm0, %xmm2
-; SSE41-NEXT: pcmpeqd %xmm3, %xmm3
-; SSE41-NEXT: paddq %xmm2, %xmm3
-; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSE41-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT: paddq %xmm0, %xmm1
+; SSE41-NEXT: pandn %xmm1, %xmm0
+; SSE41-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSE41-NEXT: movdqa %xmm0, %xmm2
+; SSE41-NEXT: pand %xmm1, %xmm2
+; SSE41-NEXT: movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
; SSE41-NEXT: movdqa %xmm3, %xmm4
-; SSE41-NEXT: pand %xmm2, %xmm4
-; SSE41-NEXT: movdqa {{.*#+}} xmm0 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSE41-NEXT: movdqa %xmm0, %xmm5
-; SSE41-NEXT: pshufb %xmm4, %xmm5
-; SSE41-NEXT: psrlw $4, %xmm3
-; SSE41-NEXT: pand %xmm2, %xmm3
-; SSE41-NEXT: pshufb %xmm3, %xmm0
-; SSE41-NEXT: paddb %xmm5, %xmm0
-; SSE41-NEXT: psadbw %xmm1, %xmm0
+; SSE41-NEXT: pshufb %xmm2, %xmm4
+; SSE41-NEXT: psrlw $4, %xmm0
+; SSE41-NEXT: pand %xmm1, %xmm0
+; SSE41-NEXT: pshufb %xmm0, %xmm3
+; SSE41-NEXT: paddb %xmm4, %xmm3
+; SSE41-NEXT: pxor %xmm0, %xmm0
+; SSE41-NEXT: psadbw %xmm3, %xmm0
; SSE41-NEXT: retq
;
; AVX1-LABEL: testv2i64u:
; AVX1: # %bb.0:
-; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT: vpsubq %xmm0, %xmm1, %xmm2
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; AVX1-NEXT: vpaddq %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm3
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT: vpshufb %xmm3, %xmm4, %xmm3
+; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT: vpaddq %xmm1, %xmm0, %xmm1
+; AVX1-NEXT: vpandn %xmm1, %xmm0, %xmm0
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm2
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX1-NEXT: vpshufb %xmm2, %xmm3, %xmm2
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vpshufb %xmm0, %xmm4, %xmm0
-; AVX1-NEXT: vpaddb %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm0
+; AVX1-NEXT: vpshufb %xmm0, %xmm3, %xmm0
+; AVX1-NEXT: vpaddb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: testv2i64u:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT: vpsubq %xmm0, %xmm1, %xmm2
-; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX2-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; AVX2-NEXT: vpaddq %xmm2, %xmm0, %xmm0
-; AVX2-NEXT: vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm3
-; AVX2-NEXT: vmovdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT: vpshufb %xmm3, %xmm4, %xmm3
+; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT: vpaddq %xmm1, %xmm0, %xmm1
+; AVX2-NEXT: vpandn %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm2
+; AVX2-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX2-NEXT: vpshufb %xmm2, %xmm3, %xmm2
; AVX2-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX2-NEXT: vpshufb %xmm0, %xmm4, %xmm0
-; AVX2-NEXT: vpaddb %xmm3, %xmm0, %xmm0
+; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vpshufb %xmm0, %xmm3, %xmm0
+; AVX2-NEXT: vpaddb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
;
@@ -346,11 +324,9 @@ define <2 x i64> @testv2i64u(<2 x i64> %
;
; AVX512VPOPCNTDQ-LABEL: testv2i64u:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubq %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpaddq %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQ-NEXT: vpaddq %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpopcntq %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512VPOPCNTDQ-NEXT: vzeroupper
@@ -358,55 +334,50 @@ define <2 x i64> @testv2i64u(<2 x i64> %
;
; AVX512VPOPCNTDQVL-LABEL: testv2i64u:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubq %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddq %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddq %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpopcntq %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: retq
;
; BITALG_NOVLX-LABEL: testv2i64u:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubq %xmm0, %xmm1, %xmm2
-; BITALG_NOVLX-NEXT: vpand %xmm2, %xmm0, %xmm0
-; BITALG_NOVLX-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; BITALG_NOVLX-NEXT: vpaddq %xmm2, %xmm0, %xmm0
+; BITALG_NOVLX-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; BITALG_NOVLX-NEXT: vpaddq %xmm1, %xmm0, %xmm1
+; BITALG_NOVLX-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG_NOVLX-NEXT: vpopcntb %zmm0, %zmm0
+; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG_NOVLX-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
; BITALG_NOVLX-NEXT: vzeroupper
; BITALG_NOVLX-NEXT: retq
;
; BITALG-LABEL: testv2i64u:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubq %xmm0, %xmm1, %xmm2
-; BITALG-NEXT: vpand %xmm2, %xmm0, %xmm0
-; BITALG-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; BITALG-NEXT: vpaddq %xmm2, %xmm0, %xmm0
+; BITALG-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; BITALG-NEXT: vpaddq %xmm1, %xmm0, %xmm1
+; BITALG-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG-NEXT: vpopcntb %xmm0, %xmm0
+; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
; BITALG-NEXT: retq
;
; X32-SSE-LABEL: testv2i64u:
; X32-SSE: # %bb.0:
-; X32-SSE-NEXT: pxor %xmm1, %xmm1
-; X32-SSE-NEXT: pxor %xmm2, %xmm2
-; X32-SSE-NEXT: psubq %xmm0, %xmm2
-; X32-SSE-NEXT: pand %xmm0, %xmm2
-; X32-SSE-NEXT: psubq {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT: movdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-SSE-NEXT: movdqa %xmm2, %xmm4
-; X32-SSE-NEXT: pand %xmm3, %xmm4
-; X32-SSE-NEXT: movdqa {{.*#+}} xmm0 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-SSE-NEXT: movdqa %xmm0, %xmm5
-; X32-SSE-NEXT: pshufb %xmm4, %xmm5
-; X32-SSE-NEXT: psrlw $4, %xmm2
-; X32-SSE-NEXT: pand %xmm3, %xmm2
-; X32-SSE-NEXT: pshufb %xmm2, %xmm0
-; X32-SSE-NEXT: paddb %xmm5, %xmm0
-; X32-SSE-NEXT: psadbw %xmm1, %xmm0
+; X32-SSE-NEXT: movdqa %xmm0, %xmm1
+; X32-SSE-NEXT: psubq {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT: pandn %xmm1, %xmm0
+; X32-SSE-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X32-SSE-NEXT: movdqa %xmm0, %xmm2
+; X32-SSE-NEXT: pand %xmm1, %xmm2
+; X32-SSE-NEXT: movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X32-SSE-NEXT: movdqa %xmm3, %xmm4
+; X32-SSE-NEXT: pshufb %xmm2, %xmm4
+; X32-SSE-NEXT: psrlw $4, %xmm0
+; X32-SSE-NEXT: pand %xmm1, %xmm0
+; X32-SSE-NEXT: pshufb %xmm0, %xmm3
+; X32-SSE-NEXT: paddb %xmm4, %xmm3
+; X32-SSE-NEXT: pxor %xmm0, %xmm0
+; X32-SSE-NEXT: psadbw %xmm3, %xmm0
; X32-SSE-NEXT: retl
%out = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %in, i1 -1)
ret <2 x i64> %out
@@ -415,130 +386,124 @@ define <2 x i64> @testv2i64u(<2 x i64> %
define <4 x i32> @testv4i32(<4 x i32> %in) nounwind {
; SSE2-LABEL: testv4i32:
; SSE2: # %bb.0:
-; SSE2-NEXT: pxor %xmm1, %xmm1
-; SSE2-NEXT: pxor %xmm2, %xmm2
-; SSE2-NEXT: psubd %xmm0, %xmm2
-; SSE2-NEXT: pand %xmm0, %xmm2
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm3
-; SSE2-NEXT: paddd %xmm2, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: psrlw $1, %xmm0
-; SSE2-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE2-NEXT: psubb %xmm0, %xmm3
-; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE2-NEXT: movdqa %xmm3, %xmm2
-; SSE2-NEXT: pand %xmm0, %xmm2
-; SSE2-NEXT: psrlw $2, %xmm3
-; SSE2-NEXT: pand %xmm0, %xmm3
-; SSE2-NEXT: paddb %xmm2, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: psrlw $4, %xmm0
-; SSE2-NEXT: paddb %xmm3, %xmm0
-; SSE2-NEXT: pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE2-NEXT: paddd %xmm0, %xmm1
+; SSE2-NEXT: pandn %xmm1, %xmm0
+; SSE2-NEXT: movdqa %xmm0, %xmm1
+; SSE2-NEXT: psrlw $1, %xmm1
+; SSE2-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT: psubb %xmm1, %xmm0
+; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
; SSE2-NEXT: movdqa %xmm0, %xmm2
-; SSE2-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
-; SSE2-NEXT: psadbw %xmm1, %xmm2
-; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE2-NEXT: psadbw %xmm1, %xmm0
-; SSE2-NEXT: packuswb %xmm2, %xmm0
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: psrlw $2, %xmm0
+; SSE2-NEXT: pand %xmm1, %xmm0
+; SSE2-NEXT: paddb %xmm2, %xmm0
+; SSE2-NEXT: movdqa %xmm0, %xmm1
+; SSE2-NEXT: psrlw $4, %xmm1
+; SSE2-NEXT: paddb %xmm0, %xmm1
+; SSE2-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT: pxor %xmm0, %xmm0
+; SSE2-NEXT: movdqa %xmm1, %xmm2
+; SSE2-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
+; SSE2-NEXT: psadbw %xmm0, %xmm2
+; SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; SSE2-NEXT: psadbw %xmm0, %xmm1
+; SSE2-NEXT: packuswb %xmm2, %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE3-LABEL: testv4i32:
; SSE3: # %bb.0:
-; SSE3-NEXT: pxor %xmm1, %xmm1
-; SSE3-NEXT: pxor %xmm2, %xmm2
-; SSE3-NEXT: psubd %xmm0, %xmm2
-; SSE3-NEXT: pand %xmm0, %xmm2
-; SSE3-NEXT: pcmpeqd %xmm3, %xmm3
-; SSE3-NEXT: paddd %xmm2, %xmm3
-; SSE3-NEXT: movdqa %xmm3, %xmm0
-; SSE3-NEXT: psrlw $1, %xmm0
-; SSE3-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE3-NEXT: psubb %xmm0, %xmm3
-; SSE3-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE3-NEXT: movdqa %xmm3, %xmm2
-; SSE3-NEXT: pand %xmm0, %xmm2
-; SSE3-NEXT: psrlw $2, %xmm3
-; SSE3-NEXT: pand %xmm0, %xmm3
-; SSE3-NEXT: paddb %xmm2, %xmm3
-; SSE3-NEXT: movdqa %xmm3, %xmm0
-; SSE3-NEXT: psrlw $4, %xmm0
-; SSE3-NEXT: paddb %xmm3, %xmm0
-; SSE3-NEXT: pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE3-NEXT: paddd %xmm0, %xmm1
+; SSE3-NEXT: pandn %xmm1, %xmm0
+; SSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE3-NEXT: psrlw $1, %xmm1
+; SSE3-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE3-NEXT: psubb %xmm1, %xmm0
+; SSE3-NEXT: movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
; SSE3-NEXT: movdqa %xmm0, %xmm2
-; SSE3-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
-; SSE3-NEXT: psadbw %xmm1, %xmm2
-; SSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE3-NEXT: psadbw %xmm1, %xmm0
-; SSE3-NEXT: packuswb %xmm2, %xmm0
+; SSE3-NEXT: pand %xmm1, %xmm2
+; SSE3-NEXT: psrlw $2, %xmm0
+; SSE3-NEXT: pand %xmm1, %xmm0
+; SSE3-NEXT: paddb %xmm2, %xmm0
+; SSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE3-NEXT: psrlw $4, %xmm1
+; SSE3-NEXT: paddb %xmm0, %xmm1
+; SSE3-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE3-NEXT: pxor %xmm0, %xmm0
+; SSE3-NEXT: movdqa %xmm1, %xmm2
+; SSE3-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
+; SSE3-NEXT: psadbw %xmm0, %xmm2
+; SSE3-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; SSE3-NEXT: psadbw %xmm0, %xmm1
+; SSE3-NEXT: packuswb %xmm2, %xmm1
+; SSE3-NEXT: movdqa %xmm1, %xmm0
; SSE3-NEXT: retq
;
; SSSE3-LABEL: testv4i32:
; SSSE3: # %bb.0:
-; SSSE3-NEXT: pxor %xmm1, %xmm1
-; SSSE3-NEXT: pxor %xmm2, %xmm2
-; SSSE3-NEXT: psubd %xmm0, %xmm2
-; SSSE3-NEXT: pand %xmm0, %xmm2
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm3
-; SSSE3-NEXT: paddd %xmm2, %xmm3
+; SSSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSSE3-NEXT: paddd %xmm0, %xmm1
+; SSSE3-NEXT: pandn %xmm1, %xmm0
; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT: movdqa %xmm3, %xmm4
-; SSSE3-NEXT: pand %xmm2, %xmm4
-; SSSE3-NEXT: movdqa {{.*#+}} xmm0 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSSE3-NEXT: movdqa %xmm0, %xmm5
-; SSSE3-NEXT: pshufb %xmm4, %xmm5
-; SSSE3-NEXT: psrlw $4, %xmm3
+; SSSE3-NEXT: movdqa %xmm0, %xmm3
; SSSE3-NEXT: pand %xmm2, %xmm3
-; SSSE3-NEXT: pshufb %xmm3, %xmm0
-; SSSE3-NEXT: paddb %xmm5, %xmm0
-; SSSE3-NEXT: movdqa %xmm0, %xmm2
-; SSSE3-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
-; SSSE3-NEXT: psadbw %xmm1, %xmm2
-; SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSSE3-NEXT: psadbw %xmm1, %xmm0
-; SSSE3-NEXT: packuswb %xmm2, %xmm0
+; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; SSSE3-NEXT: movdqa %xmm1, %xmm4
+; SSSE3-NEXT: pshufb %xmm3, %xmm4
+; SSSE3-NEXT: psrlw $4, %xmm0
+; SSSE3-NEXT: pand %xmm2, %xmm0
+; SSSE3-NEXT: pshufb %xmm0, %xmm1
+; SSSE3-NEXT: paddb %xmm4, %xmm1
+; SSSE3-NEXT: pxor %xmm0, %xmm0
+; SSSE3-NEXT: movdqa %xmm1, %xmm2
+; SSSE3-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
+; SSSE3-NEXT: psadbw %xmm0, %xmm2
+; SSSE3-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; SSSE3-NEXT: psadbw %xmm0, %xmm1
+; SSSE3-NEXT: packuswb %xmm2, %xmm1
+; SSSE3-NEXT: movdqa %xmm1, %xmm0
; SSSE3-NEXT: retq
;
; SSE41-LABEL: testv4i32:
; SSE41: # %bb.0:
-; SSE41-NEXT: pxor %xmm1, %xmm1
-; SSE41-NEXT: pxor %xmm2, %xmm2
-; SSE41-NEXT: psubd %xmm0, %xmm2
-; SSE41-NEXT: pand %xmm0, %xmm2
-; SSE41-NEXT: pcmpeqd %xmm0, %xmm0
-; SSE41-NEXT: paddd %xmm2, %xmm0
-; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT: movdqa %xmm0, %xmm3
-; SSE41-NEXT: pand %xmm2, %xmm3
-; SSE41-NEXT: movdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSE41-NEXT: movdqa %xmm4, %xmm5
-; SSE41-NEXT: pshufb %xmm3, %xmm5
+; SSE41-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT: paddd %xmm0, %xmm1
+; SSE41-NEXT: pandn %xmm1, %xmm0
+; SSE41-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSE41-NEXT: movdqa %xmm0, %xmm2
+; SSE41-NEXT: pand %xmm1, %xmm2
+; SSE41-NEXT: movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; SSE41-NEXT: movdqa %xmm3, %xmm4
+; SSE41-NEXT: pshufb %xmm2, %xmm4
; SSE41-NEXT: psrlw $4, %xmm0
-; SSE41-NEXT: pand %xmm2, %xmm0
-; SSE41-NEXT: pshufb %xmm0, %xmm4
-; SSE41-NEXT: paddb %xmm5, %xmm4
-; SSE41-NEXT: pmovzxdq {{.*#+}} xmm0 = xmm4[0],zero,xmm4[1],zero
-; SSE41-NEXT: punpckhdq {{.*#+}} xmm4 = xmm4[2],xmm1[2],xmm4[3],xmm1[3]
-; SSE41-NEXT: psadbw %xmm1, %xmm4
+; SSE41-NEXT: pand %xmm1, %xmm0
+; SSE41-NEXT: pshufb %xmm0, %xmm3
+; SSE41-NEXT: paddb %xmm4, %xmm3
+; SSE41-NEXT: pxor %xmm1, %xmm1
+; SSE41-NEXT: pmovzxdq {{.*#+}} xmm0 = xmm3[0],zero,xmm3[1],zero
+; SSE41-NEXT: punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm1[2],xmm3[3],xmm1[3]
+; SSE41-NEXT: psadbw %xmm1, %xmm3
; SSE41-NEXT: psadbw %xmm1, %xmm0
-; SSE41-NEXT: packuswb %xmm4, %xmm0
+; SSE41-NEXT: packuswb %xmm3, %xmm0
; SSE41-NEXT: retq
;
; AVX1-LABEL: testv4i32:
; AVX1: # %bb.0:
-; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT: vpsubd %xmm0, %xmm1, %xmm2
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; AVX1-NEXT: vpaddd %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm3
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT: vpshufb %xmm3, %xmm4, %xmm3
+; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT: vpaddd %xmm1, %xmm0, %xmm1
+; AVX1-NEXT: vpandn %xmm1, %xmm0, %xmm0
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm2
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX1-NEXT: vpshufb %xmm2, %xmm3, %xmm2
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vpshufb %xmm0, %xmm4, %xmm0
-; AVX1-NEXT: vpaddb %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm0
+; AVX1-NEXT: vpshufb %xmm0, %xmm3, %xmm0
+; AVX1-NEXT: vpaddb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; AVX1-NEXT: vpsadbw %xmm1, %xmm2, %xmm2
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
@@ -548,19 +513,18 @@ define <4 x i32> @testv4i32(<4 x i32> %i
;
; AVX2-LABEL: testv4i32:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT: vpsubd %xmm0, %xmm1, %xmm2
-; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX2-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; AVX2-NEXT: vpaddd %xmm2, %xmm0, %xmm0
-; AVX2-NEXT: vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm3
-; AVX2-NEXT: vmovdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT: vpshufb %xmm3, %xmm4, %xmm3
+; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT: vpaddd %xmm1, %xmm0, %xmm1
+; AVX2-NEXT: vpandn %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm2
+; AVX2-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX2-NEXT: vpshufb %xmm2, %xmm3, %xmm2
; AVX2-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX2-NEXT: vpshufb %xmm0, %xmm4, %xmm0
-; AVX2-NEXT: vpaddb %xmm3, %xmm0, %xmm0
+; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vpshufb %xmm0, %xmm3, %xmm0
+; AVX2-NEXT: vpaddb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; AVX2-NEXT: vpsadbw %xmm1, %xmm2, %xmm2
; AVX2-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
@@ -570,19 +534,18 @@ define <4 x i32> @testv4i32(<4 x i32> %i
;
; AVX512CDVL-LABEL: testv4i32:
; AVX512CDVL: # %bb.0:
-; AVX512CDVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CDVL-NEXT: vpsubd %xmm0, %xmm1, %xmm2
-; AVX512CDVL-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX512CDVL-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; AVX512CDVL-NEXT: vpaddd %xmm2, %xmm0, %xmm0
-; AVX512CDVL-NEXT: vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512CDVL-NEXT: vpand %xmm2, %xmm0, %xmm3
-; AVX512CDVL-NEXT: vmovdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512CDVL-NEXT: vpshufb %xmm3, %xmm4, %xmm3
+; AVX512CDVL-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX512CDVL-NEXT: vpaddd %xmm1, %xmm0, %xmm1
+; AVX512CDVL-NEXT: vpandn %xmm1, %xmm0, %xmm0
+; AVX512CDVL-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512CDVL-NEXT: vpand %xmm1, %xmm0, %xmm2
+; AVX512CDVL-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512CDVL-NEXT: vpshufb %xmm2, %xmm3, %xmm2
; AVX512CDVL-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX512CDVL-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX512CDVL-NEXT: vpshufb %xmm0, %xmm4, %xmm0
-; AVX512CDVL-NEXT: vpaddb %xmm3, %xmm0, %xmm0
+; AVX512CDVL-NEXT: vpand %xmm1, %xmm0, %xmm0
+; AVX512CDVL-NEXT: vpshufb %xmm0, %xmm3, %xmm0
+; AVX512CDVL-NEXT: vpaddb %xmm2, %xmm0, %xmm0
+; AVX512CDVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512CDVL-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; AVX512CDVL-NEXT: vpsadbw %xmm1, %xmm2, %xmm2
; AVX512CDVL-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
@@ -592,19 +555,18 @@ define <4 x i32> @testv4i32(<4 x i32> %i
;
; AVX512CD-LABEL: testv4i32:
; AVX512CD: # %bb.0:
-; AVX512CD-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CD-NEXT: vpsubd %xmm0, %xmm1, %xmm2
-; AVX512CD-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX512CD-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; AVX512CD-NEXT: vpaddd %xmm2, %xmm0, %xmm0
-; AVX512CD-NEXT: vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512CD-NEXT: vpand %xmm2, %xmm0, %xmm3
-; AVX512CD-NEXT: vmovdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512CD-NEXT: vpshufb %xmm3, %xmm4, %xmm3
+; AVX512CD-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX512CD-NEXT: vpaddd %xmm1, %xmm0, %xmm1
+; AVX512CD-NEXT: vpandn %xmm1, %xmm0, %xmm0
+; AVX512CD-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512CD-NEXT: vpand %xmm1, %xmm0, %xmm2
+; AVX512CD-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512CD-NEXT: vpshufb %xmm2, %xmm3, %xmm2
; AVX512CD-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX512CD-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX512CD-NEXT: vpshufb %xmm0, %xmm4, %xmm0
-; AVX512CD-NEXT: vpaddb %xmm3, %xmm0, %xmm0
+; AVX512CD-NEXT: vpand %xmm1, %xmm0, %xmm0
+; AVX512CD-NEXT: vpshufb %xmm0, %xmm3, %xmm0
+; AVX512CD-NEXT: vpaddb %xmm2, %xmm0, %xmm0
+; AVX512CD-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512CD-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; AVX512CD-NEXT: vpsadbw %xmm1, %xmm2, %xmm2
; AVX512CD-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
@@ -614,11 +576,9 @@ define <4 x i32> @testv4i32(<4 x i32> %i
;
; AVX512VPOPCNTDQ-LABEL: testv4i32:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubd %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpaddd %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQ-NEXT: vpaddd %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512VPOPCNTDQ-NEXT: vzeroupper
@@ -626,22 +586,19 @@ define <4 x i32> @testv4i32(<4 x i32> %i
;
; AVX512VPOPCNTDQVL-LABEL: testv4i32:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubd %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddd %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddd %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpopcntd %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: retq
;
; BITALG_NOVLX-LABEL: testv4i32:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubd %xmm0, %xmm1, %xmm2
-; BITALG_NOVLX-NEXT: vpand %xmm2, %xmm0, %xmm0
-; BITALG_NOVLX-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; BITALG_NOVLX-NEXT: vpaddd %xmm2, %xmm0, %xmm0
+; BITALG_NOVLX-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; BITALG_NOVLX-NEXT: vpaddd %xmm1, %xmm0, %xmm1
+; BITALG_NOVLX-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG_NOVLX-NEXT: vpopcntb %zmm0, %zmm0
+; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG_NOVLX-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; BITALG_NOVLX-NEXT: vpsadbw %xmm1, %xmm2, %xmm2
; BITALG_NOVLX-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
@@ -652,12 +609,11 @@ define <4 x i32> @testv4i32(<4 x i32> %i
;
; BITALG-LABEL: testv4i32:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubd %xmm0, %xmm1, %xmm2
-; BITALG-NEXT: vpand %xmm2, %xmm0, %xmm0
-; BITALG-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; BITALG-NEXT: vpaddd %xmm2, %xmm0, %xmm0
+; BITALG-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; BITALG-NEXT: vpaddd %xmm1, %xmm0, %xmm1
+; BITALG-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG-NEXT: vpopcntb %xmm0, %xmm0
+; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; BITALG-NEXT: vpsadbw %xmm1, %xmm2, %xmm2
; BITALG-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
@@ -667,27 +623,25 @@ define <4 x i32> @testv4i32(<4 x i32> %i
;
; X32-SSE-LABEL: testv4i32:
; X32-SSE: # %bb.0:
-; X32-SSE-NEXT: pxor %xmm1, %xmm1
-; X32-SSE-NEXT: pxor %xmm2, %xmm2
-; X32-SSE-NEXT: psubd %xmm0, %xmm2
-; X32-SSE-NEXT: pand %xmm0, %xmm2
-; X32-SSE-NEXT: pcmpeqd %xmm0, %xmm0
-; X32-SSE-NEXT: paddd %xmm2, %xmm0
-; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-SSE-NEXT: movdqa %xmm0, %xmm3
-; X32-SSE-NEXT: pand %xmm2, %xmm3
-; X32-SSE-NEXT: movdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-SSE-NEXT: movdqa %xmm4, %xmm5
-; X32-SSE-NEXT: pshufb %xmm3, %xmm5
+; X32-SSE-NEXT: pcmpeqd %xmm1, %xmm1
+; X32-SSE-NEXT: paddd %xmm0, %xmm1
+; X32-SSE-NEXT: pandn %xmm1, %xmm0
+; X32-SSE-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X32-SSE-NEXT: movdqa %xmm0, %xmm2
+; X32-SSE-NEXT: pand %xmm1, %xmm2
+; X32-SSE-NEXT: movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X32-SSE-NEXT: movdqa %xmm3, %xmm4
+; X32-SSE-NEXT: pshufb %xmm2, %xmm4
; X32-SSE-NEXT: psrlw $4, %xmm0
-; X32-SSE-NEXT: pand %xmm2, %xmm0
-; X32-SSE-NEXT: pshufb %xmm0, %xmm4
-; X32-SSE-NEXT: paddb %xmm5, %xmm4
-; X32-SSE-NEXT: pmovzxdq {{.*#+}} xmm0 = xmm4[0],zero,xmm4[1],zero
-; X32-SSE-NEXT: punpckhdq {{.*#+}} xmm4 = xmm4[2],xmm1[2],xmm4[3],xmm1[3]
-; X32-SSE-NEXT: psadbw %xmm1, %xmm4
+; X32-SSE-NEXT: pand %xmm1, %xmm0
+; X32-SSE-NEXT: pshufb %xmm0, %xmm3
+; X32-SSE-NEXT: paddb %xmm4, %xmm3
+; X32-SSE-NEXT: pxor %xmm1, %xmm1
+; X32-SSE-NEXT: pmovzxdq {{.*#+}} xmm0 = xmm3[0],zero,xmm3[1],zero
+; X32-SSE-NEXT: punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm1[2],xmm3[3],xmm1[3]
+; X32-SSE-NEXT: psadbw %xmm1, %xmm3
; X32-SSE-NEXT: psadbw %xmm1, %xmm0
-; X32-SSE-NEXT: packuswb %xmm4, %xmm0
+; X32-SSE-NEXT: packuswb %xmm3, %xmm0
; X32-SSE-NEXT: retl
%out = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %in, i1 0)
ret <4 x i32> %out
@@ -696,130 +650,124 @@ define <4 x i32> @testv4i32(<4 x i32> %i
define <4 x i32> @testv4i32u(<4 x i32> %in) nounwind {
; SSE2-LABEL: testv4i32u:
; SSE2: # %bb.0:
-; SSE2-NEXT: pxor %xmm1, %xmm1
-; SSE2-NEXT: pxor %xmm2, %xmm2
-; SSE2-NEXT: psubd %xmm0, %xmm2
-; SSE2-NEXT: pand %xmm0, %xmm2
-; SSE2-NEXT: pcmpeqd %xmm3, %xmm3
-; SSE2-NEXT: paddd %xmm2, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: psrlw $1, %xmm0
-; SSE2-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE2-NEXT: psubb %xmm0, %xmm3
-; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE2-NEXT: movdqa %xmm3, %xmm2
-; SSE2-NEXT: pand %xmm0, %xmm2
-; SSE2-NEXT: psrlw $2, %xmm3
-; SSE2-NEXT: pand %xmm0, %xmm3
-; SSE2-NEXT: paddb %xmm2, %xmm3
-; SSE2-NEXT: movdqa %xmm3, %xmm0
-; SSE2-NEXT: psrlw $4, %xmm0
-; SSE2-NEXT: paddb %xmm3, %xmm0
-; SSE2-NEXT: pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE2-NEXT: paddd %xmm0, %xmm1
+; SSE2-NEXT: pandn %xmm1, %xmm0
+; SSE2-NEXT: movdqa %xmm0, %xmm1
+; SSE2-NEXT: psrlw $1, %xmm1
+; SSE2-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT: psubb %xmm1, %xmm0
+; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
; SSE2-NEXT: movdqa %xmm0, %xmm2
-; SSE2-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
-; SSE2-NEXT: psadbw %xmm1, %xmm2
-; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE2-NEXT: psadbw %xmm1, %xmm0
-; SSE2-NEXT: packuswb %xmm2, %xmm0
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: psrlw $2, %xmm0
+; SSE2-NEXT: pand %xmm1, %xmm0
+; SSE2-NEXT: paddb %xmm2, %xmm0
+; SSE2-NEXT: movdqa %xmm0, %xmm1
+; SSE2-NEXT: psrlw $4, %xmm1
+; SSE2-NEXT: paddb %xmm0, %xmm1
+; SSE2-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT: pxor %xmm0, %xmm0
+; SSE2-NEXT: movdqa %xmm1, %xmm2
+; SSE2-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
+; SSE2-NEXT: psadbw %xmm0, %xmm2
+; SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; SSE2-NEXT: psadbw %xmm0, %xmm1
+; SSE2-NEXT: packuswb %xmm2, %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE3-LABEL: testv4i32u:
; SSE3: # %bb.0:
-; SSE3-NEXT: pxor %xmm1, %xmm1
-; SSE3-NEXT: pxor %xmm2, %xmm2
-; SSE3-NEXT: psubd %xmm0, %xmm2
-; SSE3-NEXT: pand %xmm0, %xmm2
-; SSE3-NEXT: pcmpeqd %xmm3, %xmm3
-; SSE3-NEXT: paddd %xmm2, %xmm3
-; SSE3-NEXT: movdqa %xmm3, %xmm0
-; SSE3-NEXT: psrlw $1, %xmm0
-; SSE3-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE3-NEXT: psubb %xmm0, %xmm3
-; SSE3-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE3-NEXT: movdqa %xmm3, %xmm2
-; SSE3-NEXT: pand %xmm0, %xmm2
-; SSE3-NEXT: psrlw $2, %xmm3
-; SSE3-NEXT: pand %xmm0, %xmm3
-; SSE3-NEXT: paddb %xmm2, %xmm3
-; SSE3-NEXT: movdqa %xmm3, %xmm0
-; SSE3-NEXT: psrlw $4, %xmm0
-; SSE3-NEXT: paddb %xmm3, %xmm0
-; SSE3-NEXT: pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE3-NEXT: paddd %xmm0, %xmm1
+; SSE3-NEXT: pandn %xmm1, %xmm0
+; SSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE3-NEXT: psrlw $1, %xmm1
+; SSE3-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE3-NEXT: psubb %xmm1, %xmm0
+; SSE3-NEXT: movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
; SSE3-NEXT: movdqa %xmm0, %xmm2
-; SSE3-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
-; SSE3-NEXT: psadbw %xmm1, %xmm2
-; SSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE3-NEXT: psadbw %xmm1, %xmm0
-; SSE3-NEXT: packuswb %xmm2, %xmm0
+; SSE3-NEXT: pand %xmm1, %xmm2
+; SSE3-NEXT: psrlw $2, %xmm0
+; SSE3-NEXT: pand %xmm1, %xmm0
+; SSE3-NEXT: paddb %xmm2, %xmm0
+; SSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE3-NEXT: psrlw $4, %xmm1
+; SSE3-NEXT: paddb %xmm0, %xmm1
+; SSE3-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE3-NEXT: pxor %xmm0, %xmm0
+; SSE3-NEXT: movdqa %xmm1, %xmm2
+; SSE3-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
+; SSE3-NEXT: psadbw %xmm0, %xmm2
+; SSE3-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; SSE3-NEXT: psadbw %xmm0, %xmm1
+; SSE3-NEXT: packuswb %xmm2, %xmm1
+; SSE3-NEXT: movdqa %xmm1, %xmm0
; SSE3-NEXT: retq
;
; SSSE3-LABEL: testv4i32u:
; SSSE3: # %bb.0:
-; SSSE3-NEXT: pxor %xmm1, %xmm1
-; SSSE3-NEXT: pxor %xmm2, %xmm2
-; SSSE3-NEXT: psubd %xmm0, %xmm2
-; SSSE3-NEXT: pand %xmm0, %xmm2
-; SSSE3-NEXT: pcmpeqd %xmm3, %xmm3
-; SSSE3-NEXT: paddd %xmm2, %xmm3
+; SSSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSSE3-NEXT: paddd %xmm0, %xmm1
+; SSSE3-NEXT: pandn %xmm1, %xmm0
; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT: movdqa %xmm3, %xmm4
-; SSSE3-NEXT: pand %xmm2, %xmm4
-; SSSE3-NEXT: movdqa {{.*#+}} xmm0 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSSE3-NEXT: movdqa %xmm0, %xmm5
-; SSSE3-NEXT: pshufb %xmm4, %xmm5
-; SSSE3-NEXT: psrlw $4, %xmm3
+; SSSE3-NEXT: movdqa %xmm0, %xmm3
; SSSE3-NEXT: pand %xmm2, %xmm3
-; SSSE3-NEXT: pshufb %xmm3, %xmm0
-; SSSE3-NEXT: paddb %xmm5, %xmm0
-; SSSE3-NEXT: movdqa %xmm0, %xmm2
-; SSSE3-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
-; SSSE3-NEXT: psadbw %xmm1, %xmm2
-; SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSSE3-NEXT: psadbw %xmm1, %xmm0
-; SSSE3-NEXT: packuswb %xmm2, %xmm0
+; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; SSSE3-NEXT: movdqa %xmm1, %xmm4
+; SSSE3-NEXT: pshufb %xmm3, %xmm4
+; SSSE3-NEXT: psrlw $4, %xmm0
+; SSSE3-NEXT: pand %xmm2, %xmm0
+; SSSE3-NEXT: pshufb %xmm0, %xmm1
+; SSSE3-NEXT: paddb %xmm4, %xmm1
+; SSSE3-NEXT: pxor %xmm0, %xmm0
+; SSSE3-NEXT: movdqa %xmm1, %xmm2
+; SSSE3-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
+; SSSE3-NEXT: psadbw %xmm0, %xmm2
+; SSSE3-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; SSSE3-NEXT: psadbw %xmm0, %xmm1
+; SSSE3-NEXT: packuswb %xmm2, %xmm1
+; SSSE3-NEXT: movdqa %xmm1, %xmm0
; SSSE3-NEXT: retq
;
; SSE41-LABEL: testv4i32u:
; SSE41: # %bb.0:
-; SSE41-NEXT: pxor %xmm1, %xmm1
-; SSE41-NEXT: pxor %xmm2, %xmm2
-; SSE41-NEXT: psubd %xmm0, %xmm2
-; SSE41-NEXT: pand %xmm0, %xmm2
-; SSE41-NEXT: pcmpeqd %xmm0, %xmm0
-; SSE41-NEXT: paddd %xmm2, %xmm0
-; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT: movdqa %xmm0, %xmm3
-; SSE41-NEXT: pand %xmm2, %xmm3
-; SSE41-NEXT: movdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSE41-NEXT: movdqa %xmm4, %xmm5
-; SSE41-NEXT: pshufb %xmm3, %xmm5
+; SSE41-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT: paddd %xmm0, %xmm1
+; SSE41-NEXT: pandn %xmm1, %xmm0
+; SSE41-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSE41-NEXT: movdqa %xmm0, %xmm2
+; SSE41-NEXT: pand %xmm1, %xmm2
+; SSE41-NEXT: movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; SSE41-NEXT: movdqa %xmm3, %xmm4
+; SSE41-NEXT: pshufb %xmm2, %xmm4
; SSE41-NEXT: psrlw $4, %xmm0
-; SSE41-NEXT: pand %xmm2, %xmm0
-; SSE41-NEXT: pshufb %xmm0, %xmm4
-; SSE41-NEXT: paddb %xmm5, %xmm4
-; SSE41-NEXT: pmovzxdq {{.*#+}} xmm0 = xmm4[0],zero,xmm4[1],zero
-; SSE41-NEXT: punpckhdq {{.*#+}} xmm4 = xmm4[2],xmm1[2],xmm4[3],xmm1[3]
-; SSE41-NEXT: psadbw %xmm1, %xmm4
+; SSE41-NEXT: pand %xmm1, %xmm0
+; SSE41-NEXT: pshufb %xmm0, %xmm3
+; SSE41-NEXT: paddb %xmm4, %xmm3
+; SSE41-NEXT: pxor %xmm1, %xmm1
+; SSE41-NEXT: pmovzxdq {{.*#+}} xmm0 = xmm3[0],zero,xmm3[1],zero
+; SSE41-NEXT: punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm1[2],xmm3[3],xmm1[3]
+; SSE41-NEXT: psadbw %xmm1, %xmm3
; SSE41-NEXT: psadbw %xmm1, %xmm0
-; SSE41-NEXT: packuswb %xmm4, %xmm0
+; SSE41-NEXT: packuswb %xmm3, %xmm0
; SSE41-NEXT: retq
;
; AVX1-LABEL: testv4i32u:
; AVX1: # %bb.0:
-; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT: vpsubd %xmm0, %xmm1, %xmm2
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; AVX1-NEXT: vpaddd %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm3
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT: vpshufb %xmm3, %xmm4, %xmm3
+; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT: vpaddd %xmm1, %xmm0, %xmm1
+; AVX1-NEXT: vpandn %xmm1, %xmm0, %xmm0
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm2
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX1-NEXT: vpshufb %xmm2, %xmm3, %xmm2
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vpshufb %xmm0, %xmm4, %xmm0
-; AVX1-NEXT: vpaddb %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm0
+; AVX1-NEXT: vpshufb %xmm0, %xmm3, %xmm0
+; AVX1-NEXT: vpaddb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; AVX1-NEXT: vpsadbw %xmm1, %xmm2, %xmm2
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
@@ -829,19 +777,18 @@ define <4 x i32> @testv4i32u(<4 x i32> %
;
; AVX2-LABEL: testv4i32u:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT: vpsubd %xmm0, %xmm1, %xmm2
-; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX2-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; AVX2-NEXT: vpaddd %xmm2, %xmm0, %xmm0
-; AVX2-NEXT: vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm3
-; AVX2-NEXT: vmovdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT: vpshufb %xmm3, %xmm4, %xmm3
+; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT: vpaddd %xmm1, %xmm0, %xmm1
+; AVX2-NEXT: vpandn %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm2
+; AVX2-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX2-NEXT: vpshufb %xmm2, %xmm3, %xmm2
; AVX2-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX2-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX2-NEXT: vpshufb %xmm0, %xmm4, %xmm0
-; AVX2-NEXT: vpaddb %xmm3, %xmm0, %xmm0
+; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vpshufb %xmm0, %xmm3, %xmm0
+; AVX2-NEXT: vpaddb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; AVX2-NEXT: vpsadbw %xmm1, %xmm2, %xmm2
; AVX2-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
@@ -872,11 +819,9 @@ define <4 x i32> @testv4i32u(<4 x i32> %
;
; AVX512VPOPCNTDQ-LABEL: testv4i32u:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubd %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpaddd %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQ-NEXT: vpaddd %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512VPOPCNTDQ-NEXT: vzeroupper
@@ -884,22 +829,19 @@ define <4 x i32> @testv4i32u(<4 x i32> %
;
; AVX512VPOPCNTDQVL-LABEL: testv4i32u:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubd %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddd %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddd %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpopcntd %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: retq
;
; BITALG_NOVLX-LABEL: testv4i32u:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubd %xmm0, %xmm1, %xmm2
-; BITALG_NOVLX-NEXT: vpand %xmm2, %xmm0, %xmm0
-; BITALG_NOVLX-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; BITALG_NOVLX-NEXT: vpaddd %xmm2, %xmm0, %xmm0
+; BITALG_NOVLX-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; BITALG_NOVLX-NEXT: vpaddd %xmm1, %xmm0, %xmm1
+; BITALG_NOVLX-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG_NOVLX-NEXT: vpopcntb %zmm0, %zmm0
+; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG_NOVLX-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; BITALG_NOVLX-NEXT: vpsadbw %xmm1, %xmm2, %xmm2
; BITALG_NOVLX-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
@@ -910,12 +852,11 @@ define <4 x i32> @testv4i32u(<4 x i32> %
;
; BITALG-LABEL: testv4i32u:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubd %xmm0, %xmm1, %xmm2
-; BITALG-NEXT: vpand %xmm2, %xmm0, %xmm0
-; BITALG-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
-; BITALG-NEXT: vpaddd %xmm2, %xmm0, %xmm0
+; BITALG-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; BITALG-NEXT: vpaddd %xmm1, %xmm0, %xmm1
+; BITALG-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG-NEXT: vpopcntb %xmm0, %xmm0
+; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; BITALG-NEXT: vpsadbw %xmm1, %xmm2, %xmm2
; BITALG-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
@@ -925,27 +866,25 @@ define <4 x i32> @testv4i32u(<4 x i32> %
;
; X32-SSE-LABEL: testv4i32u:
; X32-SSE: # %bb.0:
-; X32-SSE-NEXT: pxor %xmm1, %xmm1
-; X32-SSE-NEXT: pxor %xmm2, %xmm2
-; X32-SSE-NEXT: psubd %xmm0, %xmm2
-; X32-SSE-NEXT: pand %xmm0, %xmm2
-; X32-SSE-NEXT: pcmpeqd %xmm0, %xmm0
-; X32-SSE-NEXT: paddd %xmm2, %xmm0
-; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-SSE-NEXT: movdqa %xmm0, %xmm3
-; X32-SSE-NEXT: pand %xmm2, %xmm3
-; X32-SSE-NEXT: movdqa {{.*#+}} xmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-SSE-NEXT: movdqa %xmm4, %xmm5
-; X32-SSE-NEXT: pshufb %xmm3, %xmm5
+; X32-SSE-NEXT: pcmpeqd %xmm1, %xmm1
+; X32-SSE-NEXT: paddd %xmm0, %xmm1
+; X32-SSE-NEXT: pandn %xmm1, %xmm0
+; X32-SSE-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X32-SSE-NEXT: movdqa %xmm0, %xmm2
+; X32-SSE-NEXT: pand %xmm1, %xmm2
+; X32-SSE-NEXT: movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X32-SSE-NEXT: movdqa %xmm3, %xmm4
+; X32-SSE-NEXT: pshufb %xmm2, %xmm4
; X32-SSE-NEXT: psrlw $4, %xmm0
-; X32-SSE-NEXT: pand %xmm2, %xmm0
-; X32-SSE-NEXT: pshufb %xmm0, %xmm4
-; X32-SSE-NEXT: paddb %xmm5, %xmm4
-; X32-SSE-NEXT: pmovzxdq {{.*#+}} xmm0 = xmm4[0],zero,xmm4[1],zero
-; X32-SSE-NEXT: punpckhdq {{.*#+}} xmm4 = xmm4[2],xmm1[2],xmm4[3],xmm1[3]
-; X32-SSE-NEXT: psadbw %xmm1, %xmm4
+; X32-SSE-NEXT: pand %xmm1, %xmm0
+; X32-SSE-NEXT: pshufb %xmm0, %xmm3
+; X32-SSE-NEXT: paddb %xmm4, %xmm3
+; X32-SSE-NEXT: pxor %xmm1, %xmm1
+; X32-SSE-NEXT: pmovzxdq {{.*#+}} xmm0 = xmm3[0],zero,xmm3[1],zero
+; X32-SSE-NEXT: punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm1[2],xmm3[3],xmm1[3]
+; X32-SSE-NEXT: psadbw %xmm1, %xmm3
; X32-SSE-NEXT: psadbw %xmm1, %xmm0
-; X32-SSE-NEXT: packuswb %xmm4, %xmm0
+; X32-SSE-NEXT: packuswb %xmm3, %xmm0
; X32-SSE-NEXT: retl
%out = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %in, i1 -1)
ret <4 x i32> %out
@@ -954,11 +893,9 @@ define <4 x i32> @testv4i32u(<4 x i32> %
define <8 x i16> @testv8i16(<8 x i16> %in) nounwind {
; SSE2-LABEL: testv8i16:
; SSE2: # %bb.0:
-; SSE2-NEXT: pxor %xmm1, %xmm1
-; SSE2-NEXT: psubw %xmm0, %xmm1
-; SSE2-NEXT: pand %xmm0, %xmm1
-; SSE2-NEXT: pcmpeqd %xmm0, %xmm0
-; SSE2-NEXT: paddw %xmm1, %xmm0
+; SSE2-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE2-NEXT: paddw %xmm0, %xmm1
+; SSE2-NEXT: pandn %xmm1, %xmm0
; SSE2-NEXT: movdqa %xmm0, %xmm1
; SSE2-NEXT: psrlw $1, %xmm1
; SSE2-NEXT: pand {{.*}}(%rip), %xmm1
@@ -981,11 +918,9 @@ define <8 x i16> @testv8i16(<8 x i16> %i
;
; SSE3-LABEL: testv8i16:
; SSE3: # %bb.0:
-; SSE3-NEXT: pxor %xmm1, %xmm1
-; SSE3-NEXT: psubw %xmm0, %xmm1
-; SSE3-NEXT: pand %xmm0, %xmm1
-; SSE3-NEXT: pcmpeqd %xmm0, %xmm0
-; SSE3-NEXT: paddw %xmm1, %xmm0
+; SSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE3-NEXT: paddw %xmm0, %xmm1
+; SSE3-NEXT: pandn %xmm1, %xmm0
; SSE3-NEXT: movdqa %xmm0, %xmm1
; SSE3-NEXT: psrlw $1, %xmm1
; SSE3-NEXT: pand {{.*}}(%rip), %xmm1
@@ -1008,11 +943,9 @@ define <8 x i16> @testv8i16(<8 x i16> %i
;
; SSSE3-LABEL: testv8i16:
; SSSE3: # %bb.0:
-; SSSE3-NEXT: pxor %xmm1, %xmm1
-; SSSE3-NEXT: psubw %xmm0, %xmm1
-; SSSE3-NEXT: pand %xmm0, %xmm1
-; SSSE3-NEXT: pcmpeqd %xmm0, %xmm0
-; SSSE3-NEXT: paddw %xmm1, %xmm0
+; SSSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSSE3-NEXT: paddw %xmm0, %xmm1
+; SSSE3-NEXT: pandn %xmm1, %xmm0
; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; SSSE3-NEXT: movdqa %xmm0, %xmm2
; SSSE3-NEXT: pand %xmm1, %xmm2
@@ -1031,11 +964,9 @@ define <8 x i16> @testv8i16(<8 x i16> %i
;
; SSE41-LABEL: testv8i16:
; SSE41: # %bb.0:
-; SSE41-NEXT: pxor %xmm1, %xmm1
-; SSE41-NEXT: psubw %xmm0, %xmm1
-; SSE41-NEXT: pand %xmm0, %xmm1
-; SSE41-NEXT: pcmpeqd %xmm0, %xmm0
-; SSE41-NEXT: paddw %xmm1, %xmm0
+; SSE41-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT: paddw %xmm0, %xmm1
+; SSE41-NEXT: pandn %xmm1, %xmm0
; SSE41-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; SSE41-NEXT: movdqa %xmm0, %xmm2
; SSE41-NEXT: pand %xmm1, %xmm2
@@ -1054,11 +985,9 @@ define <8 x i16> @testv8i16(<8 x i16> %i
;
; AVX-LABEL: testv8i16:
; AVX: # %bb.0:
-; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT: vpsubw %xmm0, %xmm1, %xmm1
-; AVX-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX-NEXT: vpaddw %xmm1, %xmm0, %xmm0
+; AVX-NEXT: vpaddw %xmm1, %xmm0, %xmm1
+; AVX-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX-NEXT: vpand %xmm1, %xmm0, %xmm2
; AVX-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -1074,11 +1003,9 @@ define <8 x i16> @testv8i16(<8 x i16> %i
;
; AVX512VPOPCNTDQ-LABEL: testv8i16:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubw %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpaddw %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQ-NEXT: vpaddw %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpmovdw %zmm0, %ymm0
@@ -1088,11 +1015,9 @@ define <8 x i16> @testv8i16(<8 x i16> %i
;
; AVX512VPOPCNTDQVL-LABEL: testv8i16:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubw %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddw %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddw %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; AVX512VPOPCNTDQVL-NEXT: vpopcntd %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpmovdw %ymm0, %xmm0
@@ -1101,11 +1026,9 @@ define <8 x i16> @testv8i16(<8 x i16> %i
;
; BITALG_NOVLX-LABEL: testv8i16:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubw %xmm0, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpand %xmm1, %xmm0, %xmm0
; BITALG_NOVLX-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpaddw %xmm1, %xmm0, %xmm0
+; BITALG_NOVLX-NEXT: vpaddw %xmm1, %xmm0, %xmm1
+; BITALG_NOVLX-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG_NOVLX-NEXT: vpopcntw %zmm0, %zmm0
; BITALG_NOVLX-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; BITALG_NOVLX-NEXT: vzeroupper
@@ -1113,21 +1036,17 @@ define <8 x i16> @testv8i16(<8 x i16> %i
;
; BITALG-LABEL: testv8i16:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubw %xmm0, %xmm1, %xmm1
-; BITALG-NEXT: vpand %xmm1, %xmm0, %xmm0
; BITALG-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpaddw %xmm1, %xmm0, %xmm0
+; BITALG-NEXT: vpaddw %xmm1, %xmm0, %xmm1
+; BITALG-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG-NEXT: vpopcntw %xmm0, %xmm0
; BITALG-NEXT: retq
;
; X32-SSE-LABEL: testv8i16:
; X32-SSE: # %bb.0:
-; X32-SSE-NEXT: pxor %xmm1, %xmm1
-; X32-SSE-NEXT: psubw %xmm0, %xmm1
-; X32-SSE-NEXT: pand %xmm0, %xmm1
-; X32-SSE-NEXT: pcmpeqd %xmm0, %xmm0
-; X32-SSE-NEXT: paddw %xmm1, %xmm0
+; X32-SSE-NEXT: pcmpeqd %xmm1, %xmm1
+; X32-SSE-NEXT: paddw %xmm0, %xmm1
+; X32-SSE-NEXT: pandn %xmm1, %xmm0
; X32-SSE-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; X32-SSE-NEXT: movdqa %xmm0, %xmm2
; X32-SSE-NEXT: pand %xmm1, %xmm2
@@ -1150,11 +1069,9 @@ define <8 x i16> @testv8i16(<8 x i16> %i
define <8 x i16> @testv8i16u(<8 x i16> %in) nounwind {
; SSE2-LABEL: testv8i16u:
; SSE2: # %bb.0:
-; SSE2-NEXT: pxor %xmm1, %xmm1
-; SSE2-NEXT: psubw %xmm0, %xmm1
-; SSE2-NEXT: pand %xmm0, %xmm1
-; SSE2-NEXT: pcmpeqd %xmm0, %xmm0
-; SSE2-NEXT: paddw %xmm1, %xmm0
+; SSE2-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE2-NEXT: paddw %xmm0, %xmm1
+; SSE2-NEXT: pandn %xmm1, %xmm0
; SSE2-NEXT: movdqa %xmm0, %xmm1
; SSE2-NEXT: psrlw $1, %xmm1
; SSE2-NEXT: pand {{.*}}(%rip), %xmm1
@@ -1177,11 +1094,9 @@ define <8 x i16> @testv8i16u(<8 x i16> %
;
; SSE3-LABEL: testv8i16u:
; SSE3: # %bb.0:
-; SSE3-NEXT: pxor %xmm1, %xmm1
-; SSE3-NEXT: psubw %xmm0, %xmm1
-; SSE3-NEXT: pand %xmm0, %xmm1
-; SSE3-NEXT: pcmpeqd %xmm0, %xmm0
-; SSE3-NEXT: paddw %xmm1, %xmm0
+; SSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE3-NEXT: paddw %xmm0, %xmm1
+; SSE3-NEXT: pandn %xmm1, %xmm0
; SSE3-NEXT: movdqa %xmm0, %xmm1
; SSE3-NEXT: psrlw $1, %xmm1
; SSE3-NEXT: pand {{.*}}(%rip), %xmm1
@@ -1204,11 +1119,9 @@ define <8 x i16> @testv8i16u(<8 x i16> %
;
; SSSE3-LABEL: testv8i16u:
; SSSE3: # %bb.0:
-; SSSE3-NEXT: pxor %xmm1, %xmm1
-; SSSE3-NEXT: psubw %xmm0, %xmm1
-; SSSE3-NEXT: pand %xmm0, %xmm1
-; SSSE3-NEXT: pcmpeqd %xmm0, %xmm0
-; SSSE3-NEXT: paddw %xmm1, %xmm0
+; SSSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSSE3-NEXT: paddw %xmm0, %xmm1
+; SSSE3-NEXT: pandn %xmm1, %xmm0
; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; SSSE3-NEXT: movdqa %xmm0, %xmm2
; SSSE3-NEXT: pand %xmm1, %xmm2
@@ -1227,11 +1140,9 @@ define <8 x i16> @testv8i16u(<8 x i16> %
;
; SSE41-LABEL: testv8i16u:
; SSE41: # %bb.0:
-; SSE41-NEXT: pxor %xmm1, %xmm1
-; SSE41-NEXT: psubw %xmm0, %xmm1
-; SSE41-NEXT: pand %xmm0, %xmm1
-; SSE41-NEXT: pcmpeqd %xmm0, %xmm0
-; SSE41-NEXT: paddw %xmm1, %xmm0
+; SSE41-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT: paddw %xmm0, %xmm1
+; SSE41-NEXT: pandn %xmm1, %xmm0
; SSE41-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; SSE41-NEXT: movdqa %xmm0, %xmm2
; SSE41-NEXT: pand %xmm1, %xmm2
@@ -1250,11 +1161,9 @@ define <8 x i16> @testv8i16u(<8 x i16> %
;
; AVX-LABEL: testv8i16u:
; AVX: # %bb.0:
-; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT: vpsubw %xmm0, %xmm1, %xmm1
-; AVX-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX-NEXT: vpaddw %xmm1, %xmm0, %xmm0
+; AVX-NEXT: vpaddw %xmm1, %xmm0, %xmm1
+; AVX-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX-NEXT: vpand %xmm1, %xmm0, %xmm2
; AVX-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -1270,11 +1179,9 @@ define <8 x i16> @testv8i16u(<8 x i16> %
;
; AVX512VPOPCNTDQ-LABEL: testv8i16u:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubw %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpaddw %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQ-NEXT: vpaddw %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpmovdw %zmm0, %ymm0
@@ -1284,11 +1191,9 @@ define <8 x i16> @testv8i16u(<8 x i16> %
;
; AVX512VPOPCNTDQVL-LABEL: testv8i16u:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubw %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddw %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddw %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; AVX512VPOPCNTDQVL-NEXT: vpopcntd %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpmovdw %ymm0, %xmm0
@@ -1297,11 +1202,9 @@ define <8 x i16> @testv8i16u(<8 x i16> %
;
; BITALG_NOVLX-LABEL: testv8i16u:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubw %xmm0, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpand %xmm1, %xmm0, %xmm0
; BITALG_NOVLX-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpaddw %xmm1, %xmm0, %xmm0
+; BITALG_NOVLX-NEXT: vpaddw %xmm1, %xmm0, %xmm1
+; BITALG_NOVLX-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG_NOVLX-NEXT: vpopcntw %zmm0, %zmm0
; BITALG_NOVLX-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; BITALG_NOVLX-NEXT: vzeroupper
@@ -1309,21 +1212,17 @@ define <8 x i16> @testv8i16u(<8 x i16> %
;
; BITALG-LABEL: testv8i16u:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubw %xmm0, %xmm1, %xmm1
-; BITALG-NEXT: vpand %xmm1, %xmm0, %xmm0
; BITALG-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpaddw %xmm1, %xmm0, %xmm0
+; BITALG-NEXT: vpaddw %xmm1, %xmm0, %xmm1
+; BITALG-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG-NEXT: vpopcntw %xmm0, %xmm0
; BITALG-NEXT: retq
;
; X32-SSE-LABEL: testv8i16u:
; X32-SSE: # %bb.0:
-; X32-SSE-NEXT: pxor %xmm1, %xmm1
-; X32-SSE-NEXT: psubw %xmm0, %xmm1
-; X32-SSE-NEXT: pand %xmm0, %xmm1
-; X32-SSE-NEXT: pcmpeqd %xmm0, %xmm0
-; X32-SSE-NEXT: paddw %xmm1, %xmm0
+; X32-SSE-NEXT: pcmpeqd %xmm1, %xmm1
+; X32-SSE-NEXT: paddw %xmm0, %xmm1
+; X32-SSE-NEXT: pandn %xmm1, %xmm0
; X32-SSE-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; X32-SSE-NEXT: movdqa %xmm0, %xmm2
; X32-SSE-NEXT: pand %xmm1, %xmm2
@@ -1346,95 +1245,89 @@ define <8 x i16> @testv8i16u(<8 x i16> %
define <16 x i8> @testv16i8(<16 x i8> %in) nounwind {
; SSE2-LABEL: testv16i8:
; SSE2: # %bb.0:
-; SSE2-NEXT: pxor %xmm1, %xmm1
-; SSE2-NEXT: psubb %xmm0, %xmm1
-; SSE2-NEXT: pand %xmm0, %xmm1
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm2
-; SSE2-NEXT: paddb %xmm1, %xmm2
-; SSE2-NEXT: movdqa %xmm2, %xmm0
-; SSE2-NEXT: psrlw $1, %xmm0
-; SSE2-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE2-NEXT: psubb %xmm0, %xmm2
-; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE2-NEXT: movdqa %xmm2, %xmm1
-; SSE2-NEXT: pand %xmm0, %xmm1
-; SSE2-NEXT: psrlw $2, %xmm2
-; SSE2-NEXT: pand %xmm0, %xmm2
-; SSE2-NEXT: paddb %xmm1, %xmm2
-; SSE2-NEXT: movdqa %xmm2, %xmm0
-; SSE2-NEXT: psrlw $4, %xmm0
+; SSE2-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE2-NEXT: paddb %xmm0, %xmm1
+; SSE2-NEXT: pandn %xmm1, %xmm0
+; SSE2-NEXT: movdqa %xmm0, %xmm1
+; SSE2-NEXT: psrlw $1, %xmm1
+; SSE2-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT: psubb %xmm1, %xmm0
+; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; SSE2-NEXT: movdqa %xmm0, %xmm2
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: psrlw $2, %xmm0
+; SSE2-NEXT: pand %xmm1, %xmm0
; SSE2-NEXT: paddb %xmm2, %xmm0
-; SSE2-NEXT: pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT: movdqa %xmm0, %xmm1
+; SSE2-NEXT: psrlw $4, %xmm1
+; SSE2-NEXT: paddb %xmm0, %xmm1
+; SSE2-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE3-LABEL: testv16i8:
; SSE3: # %bb.0:
-; SSE3-NEXT: pxor %xmm1, %xmm1
-; SSE3-NEXT: psubb %xmm0, %xmm1
-; SSE3-NEXT: pand %xmm0, %xmm1
-; SSE3-NEXT: pcmpeqd %xmm2, %xmm2
-; SSE3-NEXT: paddb %xmm1, %xmm2
-; SSE3-NEXT: movdqa %xmm2, %xmm0
-; SSE3-NEXT: psrlw $1, %xmm0
-; SSE3-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE3-NEXT: psubb %xmm0, %xmm2
-; SSE3-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE3-NEXT: movdqa %xmm2, %xmm1
-; SSE3-NEXT: pand %xmm0, %xmm1
-; SSE3-NEXT: psrlw $2, %xmm2
-; SSE3-NEXT: pand %xmm0, %xmm2
-; SSE3-NEXT: paddb %xmm1, %xmm2
-; SSE3-NEXT: movdqa %xmm2, %xmm0
-; SSE3-NEXT: psrlw $4, %xmm0
+; SSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE3-NEXT: paddb %xmm0, %xmm1
+; SSE3-NEXT: pandn %xmm1, %xmm0
+; SSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE3-NEXT: psrlw $1, %xmm1
+; SSE3-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE3-NEXT: psubb %xmm1, %xmm0
+; SSE3-NEXT: movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; SSE3-NEXT: movdqa %xmm0, %xmm2
+; SSE3-NEXT: pand %xmm1, %xmm2
+; SSE3-NEXT: psrlw $2, %xmm0
+; SSE3-NEXT: pand %xmm1, %xmm0
; SSE3-NEXT: paddb %xmm2, %xmm0
-; SSE3-NEXT: pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE3-NEXT: psrlw $4, %xmm1
+; SSE3-NEXT: paddb %xmm0, %xmm1
+; SSE3-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE3-NEXT: movdqa %xmm1, %xmm0
; SSE3-NEXT: retq
;
; SSSE3-LABEL: testv16i8:
; SSSE3: # %bb.0:
-; SSSE3-NEXT: pxor %xmm1, %xmm1
-; SSSE3-NEXT: psubb %xmm0, %xmm1
-; SSSE3-NEXT: pand %xmm0, %xmm1
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm2
-; SSSE3-NEXT: paddb %xmm1, %xmm2
-; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT: movdqa %xmm2, %xmm3
-; SSSE3-NEXT: pand %xmm1, %xmm3
-; SSSE3-NEXT: movdqa {{.*#+}} xmm0 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSSE3-NEXT: movdqa %xmm0, %xmm4
+; SSSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSSE3-NEXT: paddb %xmm0, %xmm1
+; SSSE3-NEXT: pandn %xmm1, %xmm0
+; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSSE3-NEXT: movdqa %xmm0, %xmm3
+; SSSE3-NEXT: pand %xmm2, %xmm3
+; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; SSSE3-NEXT: movdqa %xmm1, %xmm4
; SSSE3-NEXT: pshufb %xmm3, %xmm4
-; SSSE3-NEXT: psrlw $4, %xmm2
-; SSSE3-NEXT: pand %xmm1, %xmm2
-; SSSE3-NEXT: pshufb %xmm2, %xmm0
-; SSSE3-NEXT: paddb %xmm4, %xmm0
+; SSSE3-NEXT: psrlw $4, %xmm0
+; SSSE3-NEXT: pand %xmm2, %xmm0
+; SSSE3-NEXT: pshufb %xmm0, %xmm1
+; SSSE3-NEXT: paddb %xmm4, %xmm1
+; SSSE3-NEXT: movdqa %xmm1, %xmm0
; SSSE3-NEXT: retq
;
; SSE41-LABEL: testv16i8:
; SSE41: # %bb.0:
-; SSE41-NEXT: pxor %xmm1, %xmm1
-; SSE41-NEXT: psubb %xmm0, %xmm1
-; SSE41-NEXT: pand %xmm0, %xmm1
-; SSE41-NEXT: pcmpeqd %xmm2, %xmm2
-; SSE41-NEXT: paddb %xmm1, %xmm2
-; SSE41-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT: movdqa %xmm2, %xmm3
-; SSE41-NEXT: pand %xmm1, %xmm3
-; SSE41-NEXT: movdqa {{.*#+}} xmm0 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSE41-NEXT: movdqa %xmm0, %xmm4
+; SSE41-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT: paddb %xmm0, %xmm1
+; SSE41-NEXT: pandn %xmm1, %xmm0
+; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSE41-NEXT: movdqa %xmm0, %xmm3
+; SSE41-NEXT: pand %xmm2, %xmm3
+; SSE41-NEXT: movdqa {{.*#+}} xmm1 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; SSE41-NEXT: movdqa %xmm1, %xmm4
; SSE41-NEXT: pshufb %xmm3, %xmm4
-; SSE41-NEXT: psrlw $4, %xmm2
-; SSE41-NEXT: pand %xmm1, %xmm2
-; SSE41-NEXT: pshufb %xmm2, %xmm0
-; SSE41-NEXT: paddb %xmm4, %xmm0
+; SSE41-NEXT: psrlw $4, %xmm0
+; SSE41-NEXT: pand %xmm2, %xmm0
+; SSE41-NEXT: pshufb %xmm0, %xmm1
+; SSE41-NEXT: paddb %xmm4, %xmm1
+; SSE41-NEXT: movdqa %xmm1, %xmm0
; SSE41-NEXT: retq
;
; AVX-LABEL: testv16i8:
; AVX: # %bb.0:
-; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT: vpsubb %xmm0, %xmm1, %xmm1
-; AVX-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX-NEXT: vpaddb %xmm1, %xmm0, %xmm0
+; AVX-NEXT: vpaddb %xmm1, %xmm0, %xmm1
+; AVX-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX-NEXT: vpand %xmm1, %xmm0, %xmm2
; AVX-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -1447,11 +1340,9 @@ define <16 x i8> @testv16i8(<16 x i8> %i
;
; AVX512VPOPCNTDQ-LABEL: testv16i8:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubb %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpaddb %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQ-NEXT: vpaddb %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpmovdb %zmm0, %xmm0
@@ -1460,11 +1351,9 @@ define <16 x i8> @testv16i8(<16 x i8> %i
;
; AVX512VPOPCNTDQVL-LABEL: testv16i8:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubb %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddb %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddb %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
; AVX512VPOPCNTDQVL-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQVL-NEXT: vpmovdb %zmm0, %xmm0
@@ -1473,11 +1362,9 @@ define <16 x i8> @testv16i8(<16 x i8> %i
;
; BITALG_NOVLX-LABEL: testv16i8:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubb %xmm0, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpand %xmm1, %xmm0, %xmm0
; BITALG_NOVLX-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpaddb %xmm1, %xmm0, %xmm0
+; BITALG_NOVLX-NEXT: vpaddb %xmm1, %xmm0, %xmm1
+; BITALG_NOVLX-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG_NOVLX-NEXT: vpopcntb %zmm0, %zmm0
; BITALG_NOVLX-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; BITALG_NOVLX-NEXT: vzeroupper
@@ -1485,31 +1372,28 @@ define <16 x i8> @testv16i8(<16 x i8> %i
;
; BITALG-LABEL: testv16i8:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubb %xmm0, %xmm1, %xmm1
-; BITALG-NEXT: vpand %xmm1, %xmm0, %xmm0
; BITALG-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpaddb %xmm1, %xmm0, %xmm0
+; BITALG-NEXT: vpaddb %xmm1, %xmm0, %xmm1
+; BITALG-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG-NEXT: vpopcntb %xmm0, %xmm0
; BITALG-NEXT: retq
;
; X32-SSE-LABEL: testv16i8:
; X32-SSE: # %bb.0:
-; X32-SSE-NEXT: pxor %xmm1, %xmm1
-; X32-SSE-NEXT: psubb %xmm0, %xmm1
-; X32-SSE-NEXT: pand %xmm0, %xmm1
-; X32-SSE-NEXT: pcmpeqd %xmm2, %xmm2
-; X32-SSE-NEXT: paddb %xmm1, %xmm2
-; X32-SSE-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-SSE-NEXT: movdqa %xmm2, %xmm3
-; X32-SSE-NEXT: pand %xmm1, %xmm3
-; X32-SSE-NEXT: movdqa {{.*#+}} xmm0 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-SSE-NEXT: movdqa %xmm0, %xmm4
+; X32-SSE-NEXT: pcmpeqd %xmm1, %xmm1
+; X32-SSE-NEXT: paddb %xmm0, %xmm1
+; X32-SSE-NEXT: pandn %xmm1, %xmm0
+; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X32-SSE-NEXT: movdqa %xmm0, %xmm3
+; X32-SSE-NEXT: pand %xmm2, %xmm3
+; X32-SSE-NEXT: movdqa {{.*#+}} xmm1 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X32-SSE-NEXT: movdqa %xmm1, %xmm4
; X32-SSE-NEXT: pshufb %xmm3, %xmm4
-; X32-SSE-NEXT: psrlw $4, %xmm2
-; X32-SSE-NEXT: pand %xmm1, %xmm2
-; X32-SSE-NEXT: pshufb %xmm2, %xmm0
-; X32-SSE-NEXT: paddb %xmm4, %xmm0
+; X32-SSE-NEXT: psrlw $4, %xmm0
+; X32-SSE-NEXT: pand %xmm2, %xmm0
+; X32-SSE-NEXT: pshufb %xmm0, %xmm1
+; X32-SSE-NEXT: paddb %xmm4, %xmm1
+; X32-SSE-NEXT: movdqa %xmm1, %xmm0
; X32-SSE-NEXT: retl
%out = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %in, i1 0)
ret <16 x i8> %out
@@ -1518,95 +1402,89 @@ define <16 x i8> @testv16i8(<16 x i8> %i
define <16 x i8> @testv16i8u(<16 x i8> %in) nounwind {
; SSE2-LABEL: testv16i8u:
; SSE2: # %bb.0:
-; SSE2-NEXT: pxor %xmm1, %xmm1
-; SSE2-NEXT: psubb %xmm0, %xmm1
-; SSE2-NEXT: pand %xmm0, %xmm1
-; SSE2-NEXT: pcmpeqd %xmm2, %xmm2
-; SSE2-NEXT: paddb %xmm1, %xmm2
-; SSE2-NEXT: movdqa %xmm2, %xmm0
-; SSE2-NEXT: psrlw $1, %xmm0
-; SSE2-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE2-NEXT: psubb %xmm0, %xmm2
-; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE2-NEXT: movdqa %xmm2, %xmm1
-; SSE2-NEXT: pand %xmm0, %xmm1
-; SSE2-NEXT: psrlw $2, %xmm2
-; SSE2-NEXT: pand %xmm0, %xmm2
-; SSE2-NEXT: paddb %xmm1, %xmm2
-; SSE2-NEXT: movdqa %xmm2, %xmm0
-; SSE2-NEXT: psrlw $4, %xmm0
+; SSE2-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE2-NEXT: paddb %xmm0, %xmm1
+; SSE2-NEXT: pandn %xmm1, %xmm0
+; SSE2-NEXT: movdqa %xmm0, %xmm1
+; SSE2-NEXT: psrlw $1, %xmm1
+; SSE2-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT: psubb %xmm1, %xmm0
+; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; SSE2-NEXT: movdqa %xmm0, %xmm2
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: psrlw $2, %xmm0
+; SSE2-NEXT: pand %xmm1, %xmm0
; SSE2-NEXT: paddb %xmm2, %xmm0
-; SSE2-NEXT: pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT: movdqa %xmm0, %xmm1
+; SSE2-NEXT: psrlw $4, %xmm1
+; SSE2-NEXT: paddb %xmm0, %xmm1
+; SSE2-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE3-LABEL: testv16i8u:
; SSE3: # %bb.0:
-; SSE3-NEXT: pxor %xmm1, %xmm1
-; SSE3-NEXT: psubb %xmm0, %xmm1
-; SSE3-NEXT: pand %xmm0, %xmm1
-; SSE3-NEXT: pcmpeqd %xmm2, %xmm2
-; SSE3-NEXT: paddb %xmm1, %xmm2
-; SSE3-NEXT: movdqa %xmm2, %xmm0
-; SSE3-NEXT: psrlw $1, %xmm0
-; SSE3-NEXT: pand {{.*}}(%rip), %xmm0
-; SSE3-NEXT: psubb %xmm0, %xmm2
-; SSE3-NEXT: movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE3-NEXT: movdqa %xmm2, %xmm1
-; SSE3-NEXT: pand %xmm0, %xmm1
-; SSE3-NEXT: psrlw $2, %xmm2
-; SSE3-NEXT: pand %xmm0, %xmm2
-; SSE3-NEXT: paddb %xmm1, %xmm2
-; SSE3-NEXT: movdqa %xmm2, %xmm0
-; SSE3-NEXT: psrlw $4, %xmm0
+; SSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE3-NEXT: paddb %xmm0, %xmm1
+; SSE3-NEXT: pandn %xmm1, %xmm0
+; SSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE3-NEXT: psrlw $1, %xmm1
+; SSE3-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE3-NEXT: psubb %xmm1, %xmm0
+; SSE3-NEXT: movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; SSE3-NEXT: movdqa %xmm0, %xmm2
+; SSE3-NEXT: pand %xmm1, %xmm2
+; SSE3-NEXT: psrlw $2, %xmm0
+; SSE3-NEXT: pand %xmm1, %xmm0
; SSE3-NEXT: paddb %xmm2, %xmm0
-; SSE3-NEXT: pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT: movdqa %xmm0, %xmm1
+; SSE3-NEXT: psrlw $4, %xmm1
+; SSE3-NEXT: paddb %xmm0, %xmm1
+; SSE3-NEXT: pand {{.*}}(%rip), %xmm1
+; SSE3-NEXT: movdqa %xmm1, %xmm0
; SSE3-NEXT: retq
;
; SSSE3-LABEL: testv16i8u:
; SSSE3: # %bb.0:
-; SSSE3-NEXT: pxor %xmm1, %xmm1
-; SSSE3-NEXT: psubb %xmm0, %xmm1
-; SSSE3-NEXT: pand %xmm0, %xmm1
-; SSSE3-NEXT: pcmpeqd %xmm2, %xmm2
-; SSSE3-NEXT: paddb %xmm1, %xmm2
-; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT: movdqa %xmm2, %xmm3
-; SSSE3-NEXT: pand %xmm1, %xmm3
-; SSSE3-NEXT: movdqa {{.*#+}} xmm0 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSSE3-NEXT: movdqa %xmm0, %xmm4
+; SSSE3-NEXT: pcmpeqd %xmm1, %xmm1
+; SSSE3-NEXT: paddb %xmm0, %xmm1
+; SSSE3-NEXT: pandn %xmm1, %xmm0
+; SSSE3-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSSE3-NEXT: movdqa %xmm0, %xmm3
+; SSSE3-NEXT: pand %xmm2, %xmm3
+; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; SSSE3-NEXT: movdqa %xmm1, %xmm4
; SSSE3-NEXT: pshufb %xmm3, %xmm4
-; SSSE3-NEXT: psrlw $4, %xmm2
-; SSSE3-NEXT: pand %xmm1, %xmm2
-; SSSE3-NEXT: pshufb %xmm2, %xmm0
-; SSSE3-NEXT: paddb %xmm4, %xmm0
+; SSSE3-NEXT: psrlw $4, %xmm0
+; SSSE3-NEXT: pand %xmm2, %xmm0
+; SSSE3-NEXT: pshufb %xmm0, %xmm1
+; SSSE3-NEXT: paddb %xmm4, %xmm1
+; SSSE3-NEXT: movdqa %xmm1, %xmm0
; SSSE3-NEXT: retq
;
; SSE41-LABEL: testv16i8u:
; SSE41: # %bb.0:
-; SSE41-NEXT: pxor %xmm1, %xmm1
-; SSE41-NEXT: psubb %xmm0, %xmm1
-; SSE41-NEXT: pand %xmm0, %xmm1
-; SSE41-NEXT: pcmpeqd %xmm2, %xmm2
-; SSE41-NEXT: paddb %xmm1, %xmm2
-; SSE41-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT: movdqa %xmm2, %xmm3
-; SSE41-NEXT: pand %xmm1, %xmm3
-; SSE41-NEXT: movdqa {{.*#+}} xmm0 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSE41-NEXT: movdqa %xmm0, %xmm4
+; SSE41-NEXT: pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT: paddb %xmm0, %xmm1
+; SSE41-NEXT: pandn %xmm1, %xmm0
+; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSE41-NEXT: movdqa %xmm0, %xmm3
+; SSE41-NEXT: pand %xmm2, %xmm3
+; SSE41-NEXT: movdqa {{.*#+}} xmm1 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; SSE41-NEXT: movdqa %xmm1, %xmm4
; SSE41-NEXT: pshufb %xmm3, %xmm4
-; SSE41-NEXT: psrlw $4, %xmm2
-; SSE41-NEXT: pand %xmm1, %xmm2
-; SSE41-NEXT: pshufb %xmm2, %xmm0
-; SSE41-NEXT: paddb %xmm4, %xmm0
+; SSE41-NEXT: psrlw $4, %xmm0
+; SSE41-NEXT: pand %xmm2, %xmm0
+; SSE41-NEXT: pshufb %xmm0, %xmm1
+; SSE41-NEXT: paddb %xmm4, %xmm1
+; SSE41-NEXT: movdqa %xmm1, %xmm0
; SSE41-NEXT: retq
;
; AVX-LABEL: testv16i8u:
; AVX: # %bb.0:
-; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT: vpsubb %xmm0, %xmm1, %xmm1
-; AVX-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX-NEXT: vpaddb %xmm1, %xmm0, %xmm0
+; AVX-NEXT: vpaddb %xmm1, %xmm0, %xmm1
+; AVX-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX-NEXT: vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX-NEXT: vpand %xmm1, %xmm0, %xmm2
; AVX-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -1619,11 +1497,9 @@ define <16 x i8> @testv16i8u(<16 x i8> %
;
; AVX512VPOPCNTDQ-LABEL: testv16i8u:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubb %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpaddb %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQ-NEXT: vpaddb %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQ-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpmovdb %zmm0, %xmm0
@@ -1632,11 +1508,9 @@ define <16 x i8> @testv16i8u(<16 x i8> %
;
; AVX512VPOPCNTDQVL-LABEL: testv16i8u:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubb %xmm0, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddb %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddb %xmm1, %xmm0, %xmm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %xmm1, %xmm0, %xmm0
; AVX512VPOPCNTDQVL-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
; AVX512VPOPCNTDQVL-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQVL-NEXT: vpmovdb %zmm0, %xmm0
@@ -1645,11 +1519,9 @@ define <16 x i8> @testv16i8u(<16 x i8> %
;
; BITALG_NOVLX-LABEL: testv16i8u:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubb %xmm0, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpand %xmm1, %xmm0, %xmm0
; BITALG_NOVLX-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpaddb %xmm1, %xmm0, %xmm0
+; BITALG_NOVLX-NEXT: vpaddb %xmm1, %xmm0, %xmm1
+; BITALG_NOVLX-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG_NOVLX-NEXT: vpopcntb %zmm0, %zmm0
; BITALG_NOVLX-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; BITALG_NOVLX-NEXT: vzeroupper
@@ -1657,31 +1529,28 @@ define <16 x i8> @testv16i8u(<16 x i8> %
;
; BITALG-LABEL: testv16i8u:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubb %xmm0, %xmm1, %xmm1
-; BITALG-NEXT: vpand %xmm1, %xmm0, %xmm0
; BITALG-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpaddb %xmm1, %xmm0, %xmm0
+; BITALG-NEXT: vpaddb %xmm1, %xmm0, %xmm1
+; BITALG-NEXT: vpandn %xmm1, %xmm0, %xmm0
; BITALG-NEXT: vpopcntb %xmm0, %xmm0
; BITALG-NEXT: retq
;
; X32-SSE-LABEL: testv16i8u:
; X32-SSE: # %bb.0:
-; X32-SSE-NEXT: pxor %xmm1, %xmm1
-; X32-SSE-NEXT: psubb %xmm0, %xmm1
-; X32-SSE-NEXT: pand %xmm0, %xmm1
-; X32-SSE-NEXT: pcmpeqd %xmm2, %xmm2
-; X32-SSE-NEXT: paddb %xmm1, %xmm2
-; X32-SSE-NEXT: movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-SSE-NEXT: movdqa %xmm2, %xmm3
-; X32-SSE-NEXT: pand %xmm1, %xmm3
-; X32-SSE-NEXT: movdqa {{.*#+}} xmm0 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-SSE-NEXT: movdqa %xmm0, %xmm4
+; X32-SSE-NEXT: pcmpeqd %xmm1, %xmm1
+; X32-SSE-NEXT: paddb %xmm0, %xmm1
+; X32-SSE-NEXT: pandn %xmm1, %xmm0
+; X32-SSE-NEXT: movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X32-SSE-NEXT: movdqa %xmm0, %xmm3
+; X32-SSE-NEXT: pand %xmm2, %xmm3
+; X32-SSE-NEXT: movdqa {{.*#+}} xmm1 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X32-SSE-NEXT: movdqa %xmm1, %xmm4
; X32-SSE-NEXT: pshufb %xmm3, %xmm4
-; X32-SSE-NEXT: psrlw $4, %xmm2
-; X32-SSE-NEXT: pand %xmm1, %xmm2
-; X32-SSE-NEXT: pshufb %xmm2, %xmm0
-; X32-SSE-NEXT: paddb %xmm4, %xmm0
+; X32-SSE-NEXT: psrlw $4, %xmm0
+; X32-SSE-NEXT: pand %xmm2, %xmm0
+; X32-SSE-NEXT: pshufb %xmm0, %xmm1
+; X32-SSE-NEXT: paddb %xmm4, %xmm1
+; X32-SSE-NEXT: movdqa %xmm1, %xmm0
; X32-SSE-NEXT: retl
%out = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %in, i1 -1)
ret <16 x i8> %out
Modified: llvm/trunk/test/CodeGen/X86/vector-tzcnt-256.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-tzcnt-256.ll?rev=344447&r1=344446&r2=344447&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vector-tzcnt-256.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vector-tzcnt-256.ll Sat Oct 13 05:12:06 2018
@@ -15,144 +15,132 @@ define <4 x i64> @testv4i64(<4 x i64> %i
; AVX1-LABEL: testv4i64:
; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT: vpsubq %xmm1, %xmm2, %xmm3
-; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vpcmpeqd %xmm3, %xmm3, %xmm3
-; AVX1-NEXT: vpaddq %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT: vpand %xmm4, %xmm1, %xmm5
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT: vpshufb %xmm5, %xmm6, %xmm5
+; AVX1-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
+; AVX1-NEXT: vpaddq %xmm2, %xmm1, %xmm3
+; AVX1-NEXT: vpandn %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm4
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX1-NEXT: vpshufb %xmm4, %xmm5, %xmm4
; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm1
-; AVX1-NEXT: vpand %xmm4, %xmm1, %xmm1
-; AVX1-NEXT: vpshufb %xmm1, %xmm6, %xmm1
-; AVX1-NEXT: vpaddb %xmm5, %xmm1, %xmm1
-; AVX1-NEXT: vpsadbw %xmm2, %xmm1, %xmm1
-; AVX1-NEXT: vpsubq %xmm0, %xmm2, %xmm5
-; AVX1-NEXT: vpand %xmm5, %xmm0, %xmm0
-; AVX1-NEXT: vpaddq %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm3
-; AVX1-NEXT: vpshufb %xmm3, %xmm6, %xmm3
+; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vpshufb %xmm1, %xmm5, %xmm1
+; AVX1-NEXT: vpaddb %xmm4, %xmm1, %xmm1
+; AVX1-NEXT: vpxor %xmm4, %xmm4, %xmm4
+; AVX1-NEXT: vpsadbw %xmm4, %xmm1, %xmm1
+; AVX1-NEXT: vpaddq %xmm2, %xmm0, %xmm2
+; AVX1-NEXT: vpandn %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm2
+; AVX1-NEXT: vpshufb %xmm2, %xmm5, %xmm2
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm0
-; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
-; AVX1-NEXT: vpaddb %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vpsadbw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vpshufb %xmm0, %xmm5, %xmm0
+; AVX1-NEXT: vpaddb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vpsadbw %xmm4, %xmm0, %xmm0
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: testv4i64:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT: vpsubq %ymm0, %ymm1, %ymm2
-; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX2-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; AVX2-NEXT: vpaddq %ymm2, %ymm0, %ymm0
-; AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm3
-; AVX2-NEXT: vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT: vpshufb %ymm3, %ymm4, %ymm3
+; AVX2-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; AVX2-NEXT: vpaddq %ymm1, %ymm0, %ymm1
+; AVX2-NEXT: vpandn %ymm1, %ymm0, %ymm0
+; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX2-NEXT: vpshufb %ymm2, %ymm3, %ymm2
; AVX2-NEXT: vpsrlw $4, %ymm0, %ymm0
-; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX2-NEXT: vpshufb %ymm0, %ymm4, %ymm0
-; AVX2-NEXT: vpaddb %ymm3, %ymm0, %ymm0
+; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm0
+; AVX2-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; AVX2-NEXT: vpaddb %ymm2, %ymm0, %ymm0
+; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpsadbw %ymm1, %ymm0, %ymm0
; AVX2-NEXT: retq
;
; AVX512CDVL-LABEL: testv4i64:
; AVX512CDVL: # %bb.0:
-; AVX512CDVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CDVL-NEXT: vpsubq %ymm0, %ymm1, %ymm2
-; AVX512CDVL-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX512CDVL-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; AVX512CDVL-NEXT: vpaddq %ymm2, %ymm0, %ymm0
-; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512CDVL-NEXT: vpand %ymm2, %ymm0, %ymm3
-; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512CDVL-NEXT: vpshufb %ymm3, %ymm4, %ymm3
+; AVX512CDVL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; AVX512CDVL-NEXT: vpaddq %ymm1, %ymm0, %ymm1
+; AVX512CDVL-NEXT: vpandn %ymm1, %ymm0, %ymm0
+; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512CDVL-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512CDVL-NEXT: vpshufb %ymm2, %ymm3, %ymm2
; AVX512CDVL-NEXT: vpsrlw $4, %ymm0, %ymm0
-; AVX512CDVL-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX512CDVL-NEXT: vpshufb %ymm0, %ymm4, %ymm0
-; AVX512CDVL-NEXT: vpaddb %ymm3, %ymm0, %ymm0
+; AVX512CDVL-NEXT: vpand %ymm1, %ymm0, %ymm0
+; AVX512CDVL-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; AVX512CDVL-NEXT: vpaddb %ymm2, %ymm0, %ymm0
+; AVX512CDVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512CDVL-NEXT: vpsadbw %ymm1, %ymm0, %ymm0
; AVX512CDVL-NEXT: retq
;
; AVX512CD-LABEL: testv4i64:
; AVX512CD: # %bb.0:
-; AVX512CD-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CD-NEXT: vpsubq %ymm0, %ymm1, %ymm2
-; AVX512CD-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX512CD-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; AVX512CD-NEXT: vpaddq %ymm2, %ymm0, %ymm0
-; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512CD-NEXT: vpand %ymm2, %ymm0, %ymm3
-; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512CD-NEXT: vpshufb %ymm3, %ymm4, %ymm3
+; AVX512CD-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; AVX512CD-NEXT: vpaddq %ymm1, %ymm0, %ymm1
+; AVX512CD-NEXT: vpandn %ymm1, %ymm0, %ymm0
+; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512CD-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512CD-NEXT: vpshufb %ymm2, %ymm3, %ymm2
; AVX512CD-NEXT: vpsrlw $4, %ymm0, %ymm0
-; AVX512CD-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX512CD-NEXT: vpshufb %ymm0, %ymm4, %ymm0
-; AVX512CD-NEXT: vpaddb %ymm3, %ymm0, %ymm0
+; AVX512CD-NEXT: vpand %ymm1, %ymm0, %ymm0
+; AVX512CD-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; AVX512CD-NEXT: vpaddb %ymm2, %ymm0, %ymm0
+; AVX512CD-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512CD-NEXT: vpsadbw %ymm1, %ymm0, %ymm0
; AVX512CD-NEXT: retq
;
; AVX512VPOPCNTDQ-LABEL: testv4i64:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubq %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpaddq %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpaddq %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpopcntq %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0
; AVX512VPOPCNTDQ-NEXT: retq
;
; AVX512VPOPCNTDQVL-LABEL: testv4i64:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubq %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddq %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddq %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpopcntq %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: retq
;
; BITALG_NOVLX-LABEL: testv4i64:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubq %ymm0, %ymm1, %ymm2
-; BITALG_NOVLX-NEXT: vpand %ymm2, %ymm0, %ymm0
-; BITALG_NOVLX-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; BITALG_NOVLX-NEXT: vpaddq %ymm2, %ymm0, %ymm0
+; BITALG_NOVLX-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; BITALG_NOVLX-NEXT: vpaddq %ymm1, %ymm0, %ymm1
+; BITALG_NOVLX-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG_NOVLX-NEXT: vpopcntb %zmm0, %zmm0
+; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG_NOVLX-NEXT: vpsadbw %ymm1, %ymm0, %ymm0
; BITALG_NOVLX-NEXT: retq
;
; BITALG-LABEL: testv4i64:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubq %ymm0, %ymm1, %ymm2
-; BITALG-NEXT: vpand %ymm2, %ymm0, %ymm0
-; BITALG-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; BITALG-NEXT: vpaddq %ymm2, %ymm0, %ymm0
+; BITALG-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; BITALG-NEXT: vpaddq %ymm1, %ymm0, %ymm1
+; BITALG-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG-NEXT: vpopcntb %ymm0, %ymm0
+; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG-NEXT: vpsadbw %ymm1, %ymm0, %ymm0
; BITALG-NEXT: retq
;
; X32-AVX-LABEL: testv4i64:
; X32-AVX: # %bb.0:
-; X32-AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT: vpsubq %ymm0, %ymm1, %ymm2
-; X32-AVX-NEXT: vpand %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT: vpsubq {{\.LCPI.*}}, %ymm0, %ymm0
-; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT: vpand %ymm2, %ymm0, %ymm3
-; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-AVX-NEXT: vpshufb %ymm3, %ymm4, %ymm3
+; X32-AVX-NEXT: vpsubq {{\.LCPI.*}}, %ymm0, %ymm1
+; X32-AVX-NEXT: vpandn %ymm1, %ymm0, %ymm0
+; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm2
+; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X32-AVX-NEXT: vpshufb %ymm2, %ymm3, %ymm2
; X32-AVX-NEXT: vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT: vpand %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT: vpshufb %ymm0, %ymm4, %ymm0
-; X32-AVX-NEXT: vpaddb %ymm3, %ymm0, %ymm0
+; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm0
+; X32-AVX-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; X32-AVX-NEXT: vpaddb %ymm2, %ymm0, %ymm0
+; X32-AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; X32-AVX-NEXT: vpsadbw %ymm1, %ymm0, %ymm0
; X32-AVX-NEXT: retl
%out = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %in, i1 0)
@@ -163,48 +151,45 @@ define <4 x i64> @testv4i64u(<4 x i64> %
; AVX1-LABEL: testv4i64u:
; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT: vpsubq %xmm1, %xmm2, %xmm3
-; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vpcmpeqd %xmm3, %xmm3, %xmm3
-; AVX1-NEXT: vpaddq %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT: vpand %xmm4, %xmm1, %xmm5
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT: vpshufb %xmm5, %xmm6, %xmm5
+; AVX1-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
+; AVX1-NEXT: vpaddq %xmm2, %xmm1, %xmm3
+; AVX1-NEXT: vpandn %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm4
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX1-NEXT: vpshufb %xmm4, %xmm5, %xmm4
; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm1
-; AVX1-NEXT: vpand %xmm4, %xmm1, %xmm1
-; AVX1-NEXT: vpshufb %xmm1, %xmm6, %xmm1
-; AVX1-NEXT: vpaddb %xmm5, %xmm1, %xmm1
-; AVX1-NEXT: vpsadbw %xmm2, %xmm1, %xmm1
-; AVX1-NEXT: vpsubq %xmm0, %xmm2, %xmm5
-; AVX1-NEXT: vpand %xmm5, %xmm0, %xmm0
-; AVX1-NEXT: vpaddq %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm3
-; AVX1-NEXT: vpshufb %xmm3, %xmm6, %xmm3
+; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vpshufb %xmm1, %xmm5, %xmm1
+; AVX1-NEXT: vpaddb %xmm4, %xmm1, %xmm1
+; AVX1-NEXT: vpxor %xmm4, %xmm4, %xmm4
+; AVX1-NEXT: vpsadbw %xmm4, %xmm1, %xmm1
+; AVX1-NEXT: vpaddq %xmm2, %xmm0, %xmm2
+; AVX1-NEXT: vpandn %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm2
+; AVX1-NEXT: vpshufb %xmm2, %xmm5, %xmm2
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm0
-; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
-; AVX1-NEXT: vpaddb %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vpsadbw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vpshufb %xmm0, %xmm5, %xmm0
+; AVX1-NEXT: vpaddb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vpsadbw %xmm4, %xmm0, %xmm0
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: testv4i64u:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT: vpsubq %ymm0, %ymm1, %ymm2
-; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX2-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; AVX2-NEXT: vpaddq %ymm2, %ymm0, %ymm0
-; AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm3
-; AVX2-NEXT: vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT: vpshufb %ymm3, %ymm4, %ymm3
+; AVX2-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; AVX2-NEXT: vpaddq %ymm1, %ymm0, %ymm1
+; AVX2-NEXT: vpandn %ymm1, %ymm0, %ymm0
+; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX2-NEXT: vpshufb %ymm2, %ymm3, %ymm2
; AVX2-NEXT: vpsrlw $4, %ymm0, %ymm0
-; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX2-NEXT: vpshufb %ymm0, %ymm4, %ymm0
-; AVX2-NEXT: vpaddb %ymm3, %ymm0, %ymm0
+; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm0
+; AVX2-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; AVX2-NEXT: vpaddb %ymm2, %ymm0, %ymm0
+; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpsadbw %ymm1, %ymm0, %ymm0
; AVX2-NEXT: retq
;
@@ -230,61 +215,54 @@ define <4 x i64> @testv4i64u(<4 x i64> %
;
; AVX512VPOPCNTDQ-LABEL: testv4i64u:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubq %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpaddq %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpaddq %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpopcntq %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0
; AVX512VPOPCNTDQ-NEXT: retq
;
; AVX512VPOPCNTDQVL-LABEL: testv4i64u:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubq %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddq %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddq %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpopcntq %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: retq
;
; BITALG_NOVLX-LABEL: testv4i64u:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubq %ymm0, %ymm1, %ymm2
-; BITALG_NOVLX-NEXT: vpand %ymm2, %ymm0, %ymm0
-; BITALG_NOVLX-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; BITALG_NOVLX-NEXT: vpaddq %ymm2, %ymm0, %ymm0
+; BITALG_NOVLX-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; BITALG_NOVLX-NEXT: vpaddq %ymm1, %ymm0, %ymm1
+; BITALG_NOVLX-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG_NOVLX-NEXT: vpopcntb %zmm0, %zmm0
+; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG_NOVLX-NEXT: vpsadbw %ymm1, %ymm0, %ymm0
; BITALG_NOVLX-NEXT: retq
;
; BITALG-LABEL: testv4i64u:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubq %ymm0, %ymm1, %ymm2
-; BITALG-NEXT: vpand %ymm2, %ymm0, %ymm0
-; BITALG-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; BITALG-NEXT: vpaddq %ymm2, %ymm0, %ymm0
+; BITALG-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; BITALG-NEXT: vpaddq %ymm1, %ymm0, %ymm1
+; BITALG-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG-NEXT: vpopcntb %ymm0, %ymm0
+; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG-NEXT: vpsadbw %ymm1, %ymm0, %ymm0
; BITALG-NEXT: retq
;
; X32-AVX-LABEL: testv4i64u:
; X32-AVX: # %bb.0:
-; X32-AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT: vpsubq %ymm0, %ymm1, %ymm2
-; X32-AVX-NEXT: vpand %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT: vpsubq {{\.LCPI.*}}, %ymm0, %ymm0
-; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT: vpand %ymm2, %ymm0, %ymm3
-; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-AVX-NEXT: vpshufb %ymm3, %ymm4, %ymm3
+; X32-AVX-NEXT: vpsubq {{\.LCPI.*}}, %ymm0, %ymm1
+; X32-AVX-NEXT: vpandn %ymm1, %ymm0, %ymm0
+; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm2
+; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X32-AVX-NEXT: vpshufb %ymm2, %ymm3, %ymm2
; X32-AVX-NEXT: vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT: vpand %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT: vpshufb %ymm0, %ymm4, %ymm0
-; X32-AVX-NEXT: vpaddb %ymm3, %ymm0, %ymm0
+; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm0
+; X32-AVX-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; X32-AVX-NEXT: vpaddb %ymm2, %ymm0, %ymm0
+; X32-AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; X32-AVX-NEXT: vpsadbw %ymm1, %ymm0, %ymm0
; X32-AVX-NEXT: retl
%out = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %in, i1 -1)
@@ -295,56 +273,53 @@ define <8 x i32> @testv8i32(<8 x i32> %i
; AVX1-LABEL: testv8i32:
; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT: vpsubd %xmm1, %xmm2, %xmm3
-; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vpcmpeqd %xmm3, %xmm3, %xmm3
-; AVX1-NEXT: vpaddd %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT: vpand %xmm4, %xmm1, %xmm5
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT: vpshufb %xmm5, %xmm6, %xmm5
+; AVX1-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
+; AVX1-NEXT: vpaddd %xmm2, %xmm1, %xmm3
+; AVX1-NEXT: vpandn %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm4
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX1-NEXT: vpshufb %xmm4, %xmm5, %xmm4
; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm1
-; AVX1-NEXT: vpand %xmm4, %xmm1, %xmm1
-; AVX1-NEXT: vpshufb %xmm1, %xmm6, %xmm1
-; AVX1-NEXT: vpaddb %xmm5, %xmm1, %xmm1
-; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm5 = xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; AVX1-NEXT: vpsadbw %xmm2, %xmm5, %xmm5
+; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vpshufb %xmm1, %xmm5, %xmm1
+; AVX1-NEXT: vpaddb %xmm4, %xmm1, %xmm1
+; AVX1-NEXT: vpxor %xmm4, %xmm4, %xmm4
+; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm6 = xmm1[2],xmm4[2],xmm1[3],xmm4[3]
+; AVX1-NEXT: vpsadbw %xmm4, %xmm6, %xmm6
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
-; AVX1-NEXT: vpsadbw %xmm2, %xmm1, %xmm1
-; AVX1-NEXT: vpackuswb %xmm5, %xmm1, %xmm1
-; AVX1-NEXT: vpsubd %xmm0, %xmm2, %xmm5
-; AVX1-NEXT: vpand %xmm5, %xmm0, %xmm0
-; AVX1-NEXT: vpaddd %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm3
-; AVX1-NEXT: vpshufb %xmm3, %xmm6, %xmm3
+; AVX1-NEXT: vpsadbw %xmm4, %xmm1, %xmm1
+; AVX1-NEXT: vpackuswb %xmm6, %xmm1, %xmm1
+; AVX1-NEXT: vpaddd %xmm2, %xmm0, %xmm2
+; AVX1-NEXT: vpandn %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm2
+; AVX1-NEXT: vpshufb %xmm2, %xmm5, %xmm2
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm0
-; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
-; AVX1-NEXT: vpaddb %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm3 = xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; AVX1-NEXT: vpsadbw %xmm2, %xmm3, %xmm3
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vpshufb %xmm0, %xmm5, %xmm0
+; AVX1-NEXT: vpaddb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm4[2],xmm0[3],xmm4[3]
+; AVX1-NEXT: vpsadbw %xmm4, %xmm2, %xmm2
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; AVX1-NEXT: vpsadbw %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vpackuswb %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vpsadbw %xmm4, %xmm0, %xmm0
+; AVX1-NEXT: vpackuswb %xmm2, %xmm0, %xmm0
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: testv8i32:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT: vpsubd %ymm0, %ymm1, %ymm2
-; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX2-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; AVX2-NEXT: vpaddd %ymm2, %ymm0, %ymm0
-; AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm3
-; AVX2-NEXT: vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT: vpshufb %ymm3, %ymm4, %ymm3
+; AVX2-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; AVX2-NEXT: vpaddd %ymm1, %ymm0, %ymm1
+; AVX2-NEXT: vpandn %ymm1, %ymm0, %ymm0
+; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX2-NEXT: vpshufb %ymm2, %ymm3, %ymm2
; AVX2-NEXT: vpsrlw $4, %ymm0, %ymm0
-; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX2-NEXT: vpshufb %ymm0, %ymm4, %ymm0
-; AVX2-NEXT: vpaddb %ymm3, %ymm0, %ymm0
+; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm0
+; AVX2-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; AVX2-NEXT: vpaddb %ymm2, %ymm0, %ymm0
+; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; AVX2-NEXT: vpsadbw %ymm1, %ymm2, %ymm2
; AVX2-NEXT: vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
@@ -354,19 +329,18 @@ define <8 x i32> @testv8i32(<8 x i32> %i
;
; AVX512CDVL-LABEL: testv8i32:
; AVX512CDVL: # %bb.0:
-; AVX512CDVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CDVL-NEXT: vpsubd %ymm0, %ymm1, %ymm2
-; AVX512CDVL-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX512CDVL-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; AVX512CDVL-NEXT: vpaddd %ymm2, %ymm0, %ymm0
-; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512CDVL-NEXT: vpand %ymm2, %ymm0, %ymm3
-; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512CDVL-NEXT: vpshufb %ymm3, %ymm4, %ymm3
+; AVX512CDVL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; AVX512CDVL-NEXT: vpaddd %ymm1, %ymm0, %ymm1
+; AVX512CDVL-NEXT: vpandn %ymm1, %ymm0, %ymm0
+; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512CDVL-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512CDVL-NEXT: vpshufb %ymm2, %ymm3, %ymm2
; AVX512CDVL-NEXT: vpsrlw $4, %ymm0, %ymm0
-; AVX512CDVL-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX512CDVL-NEXT: vpshufb %ymm0, %ymm4, %ymm0
-; AVX512CDVL-NEXT: vpaddb %ymm3, %ymm0, %ymm0
+; AVX512CDVL-NEXT: vpand %ymm1, %ymm0, %ymm0
+; AVX512CDVL-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; AVX512CDVL-NEXT: vpaddb %ymm2, %ymm0, %ymm0
+; AVX512CDVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512CDVL-NEXT: vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; AVX512CDVL-NEXT: vpsadbw %ymm1, %ymm2, %ymm2
; AVX512CDVL-NEXT: vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
@@ -376,19 +350,18 @@ define <8 x i32> @testv8i32(<8 x i32> %i
;
; AVX512CD-LABEL: testv8i32:
; AVX512CD: # %bb.0:
-; AVX512CD-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CD-NEXT: vpsubd %ymm0, %ymm1, %ymm2
-; AVX512CD-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX512CD-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; AVX512CD-NEXT: vpaddd %ymm2, %ymm0, %ymm0
-; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512CD-NEXT: vpand %ymm2, %ymm0, %ymm3
-; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512CD-NEXT: vpshufb %ymm3, %ymm4, %ymm3
+; AVX512CD-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; AVX512CD-NEXT: vpaddd %ymm1, %ymm0, %ymm1
+; AVX512CD-NEXT: vpandn %ymm1, %ymm0, %ymm0
+; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512CD-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512CD-NEXT: vpshufb %ymm2, %ymm3, %ymm2
; AVX512CD-NEXT: vpsrlw $4, %ymm0, %ymm0
-; AVX512CD-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX512CD-NEXT: vpshufb %ymm0, %ymm4, %ymm0
-; AVX512CD-NEXT: vpaddb %ymm3, %ymm0, %ymm0
+; AVX512CD-NEXT: vpand %ymm1, %ymm0, %ymm0
+; AVX512CD-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; AVX512CD-NEXT: vpaddb %ymm2, %ymm0, %ymm0
+; AVX512CD-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512CD-NEXT: vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; AVX512CD-NEXT: vpsadbw %ymm1, %ymm2, %ymm2
; AVX512CD-NEXT: vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
@@ -398,33 +371,28 @@ define <8 x i32> @testv8i32(<8 x i32> %i
;
; AVX512VPOPCNTDQ-LABEL: testv8i32:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubd %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpaddd %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpaddd %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0
; AVX512VPOPCNTDQ-NEXT: retq
;
; AVX512VPOPCNTDQVL-LABEL: testv8i32:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubd %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddd %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddd %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpopcntd %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: retq
;
; BITALG_NOVLX-LABEL: testv8i32:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubd %ymm0, %ymm1, %ymm2
-; BITALG_NOVLX-NEXT: vpand %ymm2, %ymm0, %ymm0
-; BITALG_NOVLX-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; BITALG_NOVLX-NEXT: vpaddd %ymm2, %ymm0, %ymm0
+; BITALG_NOVLX-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; BITALG_NOVLX-NEXT: vpaddd %ymm1, %ymm0, %ymm1
+; BITALG_NOVLX-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG_NOVLX-NEXT: vpopcntb %zmm0, %zmm0
+; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG_NOVLX-NEXT: vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; BITALG_NOVLX-NEXT: vpsadbw %ymm1, %ymm2, %ymm2
; BITALG_NOVLX-NEXT: vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
@@ -434,12 +402,11 @@ define <8 x i32> @testv8i32(<8 x i32> %i
;
; BITALG-LABEL: testv8i32:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubd %ymm0, %ymm1, %ymm2
-; BITALG-NEXT: vpand %ymm2, %ymm0, %ymm0
-; BITALG-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; BITALG-NEXT: vpaddd %ymm2, %ymm0, %ymm0
+; BITALG-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; BITALG-NEXT: vpaddd %ymm1, %ymm0, %ymm1
+; BITALG-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG-NEXT: vpopcntb %ymm0, %ymm0
+; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG-NEXT: vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; BITALG-NEXT: vpsadbw %ymm1, %ymm2, %ymm2
; BITALG-NEXT: vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
@@ -449,19 +416,18 @@ define <8 x i32> @testv8i32(<8 x i32> %i
;
; X32-AVX-LABEL: testv8i32:
; X32-AVX: # %bb.0:
-; X32-AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT: vpsubd %ymm0, %ymm1, %ymm2
-; X32-AVX-NEXT: vpand %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; X32-AVX-NEXT: vpaddd %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT: vpand %ymm2, %ymm0, %ymm3
-; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-AVX-NEXT: vpshufb %ymm3, %ymm4, %ymm3
+; X32-AVX-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; X32-AVX-NEXT: vpaddd %ymm1, %ymm0, %ymm1
+; X32-AVX-NEXT: vpandn %ymm1, %ymm0, %ymm0
+; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm2
+; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X32-AVX-NEXT: vpshufb %ymm2, %ymm3, %ymm2
; X32-AVX-NEXT: vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT: vpand %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT: vpshufb %ymm0, %ymm4, %ymm0
-; X32-AVX-NEXT: vpaddb %ymm3, %ymm0, %ymm0
+; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm0
+; X32-AVX-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; X32-AVX-NEXT: vpaddb %ymm2, %ymm0, %ymm0
+; X32-AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; X32-AVX-NEXT: vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; X32-AVX-NEXT: vpsadbw %ymm1, %ymm2, %ymm2
; X32-AVX-NEXT: vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
@@ -476,56 +442,53 @@ define <8 x i32> @testv8i32u(<8 x i32> %
; AVX1-LABEL: testv8i32u:
; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT: vpsubd %xmm1, %xmm2, %xmm3
-; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vpcmpeqd %xmm3, %xmm3, %xmm3
-; AVX1-NEXT: vpaddd %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT: vpand %xmm4, %xmm1, %xmm5
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT: vpshufb %xmm5, %xmm6, %xmm5
+; AVX1-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
+; AVX1-NEXT: vpaddd %xmm2, %xmm1, %xmm3
+; AVX1-NEXT: vpandn %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm4
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX1-NEXT: vpshufb %xmm4, %xmm5, %xmm4
; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm1
-; AVX1-NEXT: vpand %xmm4, %xmm1, %xmm1
-; AVX1-NEXT: vpshufb %xmm1, %xmm6, %xmm1
-; AVX1-NEXT: vpaddb %xmm5, %xmm1, %xmm1
-; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm5 = xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; AVX1-NEXT: vpsadbw %xmm2, %xmm5, %xmm5
+; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vpshufb %xmm1, %xmm5, %xmm1
+; AVX1-NEXT: vpaddb %xmm4, %xmm1, %xmm1
+; AVX1-NEXT: vpxor %xmm4, %xmm4, %xmm4
+; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm6 = xmm1[2],xmm4[2],xmm1[3],xmm4[3]
+; AVX1-NEXT: vpsadbw %xmm4, %xmm6, %xmm6
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
-; AVX1-NEXT: vpsadbw %xmm2, %xmm1, %xmm1
-; AVX1-NEXT: vpackuswb %xmm5, %xmm1, %xmm1
-; AVX1-NEXT: vpsubd %xmm0, %xmm2, %xmm5
-; AVX1-NEXT: vpand %xmm5, %xmm0, %xmm0
-; AVX1-NEXT: vpaddd %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm3
-; AVX1-NEXT: vpshufb %xmm3, %xmm6, %xmm3
+; AVX1-NEXT: vpsadbw %xmm4, %xmm1, %xmm1
+; AVX1-NEXT: vpackuswb %xmm6, %xmm1, %xmm1
+; AVX1-NEXT: vpaddd %xmm2, %xmm0, %xmm2
+; AVX1-NEXT: vpandn %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm2
+; AVX1-NEXT: vpshufb %xmm2, %xmm5, %xmm2
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm0
-; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
-; AVX1-NEXT: vpaddb %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm3 = xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; AVX1-NEXT: vpsadbw %xmm2, %xmm3, %xmm3
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vpshufb %xmm0, %xmm5, %xmm0
+; AVX1-NEXT: vpaddb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm4[2],xmm0[3],xmm4[3]
+; AVX1-NEXT: vpsadbw %xmm4, %xmm2, %xmm2
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; AVX1-NEXT: vpsadbw %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vpackuswb %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vpsadbw %xmm4, %xmm0, %xmm0
+; AVX1-NEXT: vpackuswb %xmm2, %xmm0, %xmm0
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: testv8i32u:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT: vpsubd %ymm0, %ymm1, %ymm2
-; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX2-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; AVX2-NEXT: vpaddd %ymm2, %ymm0, %ymm0
-; AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm3
-; AVX2-NEXT: vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT: vpshufb %ymm3, %ymm4, %ymm3
+; AVX2-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; AVX2-NEXT: vpaddd %ymm1, %ymm0, %ymm1
+; AVX2-NEXT: vpandn %ymm1, %ymm0, %ymm0
+; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm2
+; AVX2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX2-NEXT: vpshufb %ymm2, %ymm3, %ymm2
; AVX2-NEXT: vpsrlw $4, %ymm0, %ymm0
-; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm0
-; AVX2-NEXT: vpshufb %ymm0, %ymm4, %ymm0
-; AVX2-NEXT: vpaddb %ymm3, %ymm0, %ymm0
+; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm0
+; AVX2-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; AVX2-NEXT: vpaddb %ymm2, %ymm0, %ymm0
+; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; AVX2-NEXT: vpsadbw %ymm1, %ymm2, %ymm2
; AVX2-NEXT: vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
@@ -555,33 +518,28 @@ define <8 x i32> @testv8i32u(<8 x i32> %
;
; AVX512VPOPCNTDQ-LABEL: testv8i32u:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubd %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpaddd %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpaddd %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0
; AVX512VPOPCNTDQ-NEXT: retq
;
; AVX512VPOPCNTDQVL-LABEL: testv8i32u:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubd %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddd %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddd %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpopcntd %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: retq
;
; BITALG_NOVLX-LABEL: testv8i32u:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubd %ymm0, %ymm1, %ymm2
-; BITALG_NOVLX-NEXT: vpand %ymm2, %ymm0, %ymm0
-; BITALG_NOVLX-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; BITALG_NOVLX-NEXT: vpaddd %ymm2, %ymm0, %ymm0
+; BITALG_NOVLX-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; BITALG_NOVLX-NEXT: vpaddd %ymm1, %ymm0, %ymm1
+; BITALG_NOVLX-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG_NOVLX-NEXT: vpopcntb %zmm0, %zmm0
+; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG_NOVLX-NEXT: vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; BITALG_NOVLX-NEXT: vpsadbw %ymm1, %ymm2, %ymm2
; BITALG_NOVLX-NEXT: vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
@@ -591,12 +549,11 @@ define <8 x i32> @testv8i32u(<8 x i32> %
;
; BITALG-LABEL: testv8i32u:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubd %ymm0, %ymm1, %ymm2
-; BITALG-NEXT: vpand %ymm2, %ymm0, %ymm0
-; BITALG-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; BITALG-NEXT: vpaddd %ymm2, %ymm0, %ymm0
+; BITALG-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; BITALG-NEXT: vpaddd %ymm1, %ymm0, %ymm1
+; BITALG-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG-NEXT: vpopcntb %ymm0, %ymm0
+; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG-NEXT: vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; BITALG-NEXT: vpsadbw %ymm1, %ymm2, %ymm2
; BITALG-NEXT: vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
@@ -606,19 +563,18 @@ define <8 x i32> @testv8i32u(<8 x i32> %
;
; X32-AVX-LABEL: testv8i32u:
; X32-AVX: # %bb.0:
-; X32-AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT: vpsubd %ymm0, %ymm1, %ymm2
-; X32-AVX-NEXT: vpand %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
-; X32-AVX-NEXT: vpaddd %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT: vpand %ymm2, %ymm0, %ymm3
-; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-AVX-NEXT: vpshufb %ymm3, %ymm4, %ymm3
+; X32-AVX-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; X32-AVX-NEXT: vpaddd %ymm1, %ymm0, %ymm1
+; X32-AVX-NEXT: vpandn %ymm1, %ymm0, %ymm0
+; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm2
+; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X32-AVX-NEXT: vpshufb %ymm2, %ymm3, %ymm2
; X32-AVX-NEXT: vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT: vpand %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT: vpshufb %ymm0, %ymm4, %ymm0
-; X32-AVX-NEXT: vpaddb %ymm3, %ymm0, %ymm0
+; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm0
+; X32-AVX-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; X32-AVX-NEXT: vpaddb %ymm2, %ymm0, %ymm0
+; X32-AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; X32-AVX-NEXT: vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; X32-AVX-NEXT: vpsadbw %ymm1, %ymm2, %ymm2
; X32-AVX-NEXT: vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
@@ -632,31 +588,28 @@ define <8 x i32> @testv8i32u(<8 x i32> %
define <16 x i16> @testv16i16(<16 x i16> %in) nounwind {
; AVX1-LABEL: testv16i16:
; AVX1: # %bb.0:
-; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT: vpsubw %xmm0, %xmm1, %xmm2
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm2
-; AVX1-NEXT: vpcmpeqd %xmm3, %xmm3, %xmm3
-; AVX1-NEXT: vpaddw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT: vpand %xmm4, %xmm2, %xmm5
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT: vpshufb %xmm5, %xmm6, %xmm5
+; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT: vpaddw %xmm1, %xmm0, %xmm2
+; AVX1-NEXT: vpandn %xmm2, %xmm0, %xmm2
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT: vpand %xmm3, %xmm2, %xmm4
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX1-NEXT: vpshufb %xmm4, %xmm5, %xmm4
; AVX1-NEXT: vpsrlw $4, %xmm2, %xmm2
-; AVX1-NEXT: vpand %xmm4, %xmm2, %xmm2
-; AVX1-NEXT: vpshufb %xmm2, %xmm6, %xmm2
-; AVX1-NEXT: vpaddb %xmm5, %xmm2, %xmm2
-; AVX1-NEXT: vpsllw $8, %xmm2, %xmm5
-; AVX1-NEXT: vpaddb %xmm2, %xmm5, %xmm2
+; AVX1-NEXT: vpand %xmm3, %xmm2, %xmm2
+; AVX1-NEXT: vpshufb %xmm2, %xmm5, %xmm2
+; AVX1-NEXT: vpaddb %xmm4, %xmm2, %xmm2
+; AVX1-NEXT: vpsllw $8, %xmm2, %xmm4
+; AVX1-NEXT: vpaddb %xmm2, %xmm4, %xmm2
; AVX1-NEXT: vpsrlw $8, %xmm2, %xmm2
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT: vpsubw %xmm0, %xmm1, %xmm1
-; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT: vpaddw %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm1
-; AVX1-NEXT: vpshufb %xmm1, %xmm6, %xmm1
+; AVX1-NEXT: vpaddw %xmm1, %xmm0, %xmm1
+; AVX1-NEXT: vpandn %xmm1, %xmm0, %xmm0
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm1
+; AVX1-NEXT: vpshufb %xmm1, %xmm5, %xmm1
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm0
-; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vpshufb %xmm0, %xmm5, %xmm0
; AVX1-NEXT: vpaddb %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpsllw $8, %xmm0, %xmm1
; AVX1-NEXT: vpaddb %xmm0, %xmm1, %xmm0
@@ -666,11 +619,9 @@ define <16 x i16> @testv16i16(<16 x i16>
;
; AVX2-LABEL: testv16i16:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX2-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX2-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; AVX2-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; AVX2-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -686,11 +637,9 @@ define <16 x i16> @testv16i16(<16 x i16>
;
; AVX512CDVL-LABEL: testv16i16:
; AVX512CDVL: # %bb.0:
-; AVX512CDVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CDVL-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; AVX512CDVL-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512CDVL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512CDVL-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; AVX512CDVL-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; AVX512CDVL-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512CDVL-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -706,11 +655,9 @@ define <16 x i16> @testv16i16(<16 x i16>
;
; AVX512CD-LABEL: testv16i16:
; AVX512CD: # %bb.0:
-; AVX512CD-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CD-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; AVX512CD-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512CD-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512CD-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; AVX512CD-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; AVX512CD-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512CD-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -726,11 +673,9 @@ define <16 x i16> @testv16i16(<16 x i16>
;
; AVX512VPOPCNTDQ-LABEL: testv16i16:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpmovdw %zmm0, %ymm0
@@ -738,11 +683,9 @@ define <16 x i16> @testv16i16(<16 x i16>
;
; AVX512VPOPCNTDQVL-LABEL: testv16i16:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512VPOPCNTDQVL-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQVL-NEXT: vpmovdw %zmm0, %ymm0
@@ -750,32 +693,26 @@ define <16 x i16> @testv16i16(<16 x i16>
;
; BITALG_NOVLX-LABEL: testv16i16:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; BITALG_NOVLX-NEXT: vpand %ymm1, %ymm0, %ymm0
; BITALG_NOVLX-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; BITALG_NOVLX-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; BITALG_NOVLX-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; BITALG_NOVLX-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG_NOVLX-NEXT: vpopcntw %zmm0, %zmm0
; BITALG_NOVLX-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0
; BITALG_NOVLX-NEXT: retq
;
; BITALG-LABEL: testv16i16:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; BITALG-NEXT: vpand %ymm1, %ymm0, %ymm0
; BITALG-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; BITALG-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; BITALG-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; BITALG-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG-NEXT: vpopcntw %ymm0, %ymm0
; BITALG-NEXT: retq
;
; X32-AVX-LABEL: testv16i16:
; X32-AVX: # %bb.0:
-; X32-AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm0
; X32-AVX-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; X32-AVX-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; X32-AVX-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; X32-AVX-NEXT: vpandn %ymm1, %ymm0, %ymm0
; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm2
; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -795,31 +732,28 @@ define <16 x i16> @testv16i16(<16 x i16>
define <16 x i16> @testv16i16u(<16 x i16> %in) nounwind {
; AVX1-LABEL: testv16i16u:
; AVX1: # %bb.0:
-; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT: vpsubw %xmm0, %xmm1, %xmm2
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm2
-; AVX1-NEXT: vpcmpeqd %xmm3, %xmm3, %xmm3
-; AVX1-NEXT: vpaddw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT: vpand %xmm4, %xmm2, %xmm5
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT: vpshufb %xmm5, %xmm6, %xmm5
+; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT: vpaddw %xmm1, %xmm0, %xmm2
+; AVX1-NEXT: vpandn %xmm2, %xmm0, %xmm2
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT: vpand %xmm3, %xmm2, %xmm4
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX1-NEXT: vpshufb %xmm4, %xmm5, %xmm4
; AVX1-NEXT: vpsrlw $4, %xmm2, %xmm2
-; AVX1-NEXT: vpand %xmm4, %xmm2, %xmm2
-; AVX1-NEXT: vpshufb %xmm2, %xmm6, %xmm2
-; AVX1-NEXT: vpaddb %xmm5, %xmm2, %xmm2
-; AVX1-NEXT: vpsllw $8, %xmm2, %xmm5
-; AVX1-NEXT: vpaddb %xmm2, %xmm5, %xmm2
+; AVX1-NEXT: vpand %xmm3, %xmm2, %xmm2
+; AVX1-NEXT: vpshufb %xmm2, %xmm5, %xmm2
+; AVX1-NEXT: vpaddb %xmm4, %xmm2, %xmm2
+; AVX1-NEXT: vpsllw $8, %xmm2, %xmm4
+; AVX1-NEXT: vpaddb %xmm2, %xmm4, %xmm2
; AVX1-NEXT: vpsrlw $8, %xmm2, %xmm2
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT: vpsubw %xmm0, %xmm1, %xmm1
-; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT: vpaddw %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm1
-; AVX1-NEXT: vpshufb %xmm1, %xmm6, %xmm1
+; AVX1-NEXT: vpaddw %xmm1, %xmm0, %xmm1
+; AVX1-NEXT: vpandn %xmm1, %xmm0, %xmm0
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm1
+; AVX1-NEXT: vpshufb %xmm1, %xmm5, %xmm1
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm0
-; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vpshufb %xmm0, %xmm5, %xmm0
; AVX1-NEXT: vpaddb %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpsllw $8, %xmm0, %xmm1
; AVX1-NEXT: vpaddb %xmm0, %xmm1, %xmm0
@@ -829,11 +763,9 @@ define <16 x i16> @testv16i16u(<16 x i16
;
; AVX2-LABEL: testv16i16u:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX2-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX2-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; AVX2-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; AVX2-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -849,11 +781,9 @@ define <16 x i16> @testv16i16u(<16 x i16
;
; AVX512CDVL-LABEL: testv16i16u:
; AVX512CDVL: # %bb.0:
-; AVX512CDVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CDVL-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; AVX512CDVL-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512CDVL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512CDVL-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; AVX512CDVL-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; AVX512CDVL-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512CDVL-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -869,11 +799,9 @@ define <16 x i16> @testv16i16u(<16 x i16
;
; AVX512CD-LABEL: testv16i16u:
; AVX512CD: # %bb.0:
-; AVX512CD-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CD-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; AVX512CD-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512CD-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512CD-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; AVX512CD-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; AVX512CD-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512CD-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -889,11 +817,9 @@ define <16 x i16> @testv16i16u(<16 x i16
;
; AVX512VPOPCNTDQ-LABEL: testv16i16u:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpmovdw %zmm0, %ymm0
@@ -901,11 +827,9 @@ define <16 x i16> @testv16i16u(<16 x i16
;
; AVX512VPOPCNTDQVL-LABEL: testv16i16u:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512VPOPCNTDQVL-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQVL-NEXT: vpmovdw %zmm0, %ymm0
@@ -913,32 +837,26 @@ define <16 x i16> @testv16i16u(<16 x i16
;
; BITALG_NOVLX-LABEL: testv16i16u:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; BITALG_NOVLX-NEXT: vpand %ymm1, %ymm0, %ymm0
; BITALG_NOVLX-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; BITALG_NOVLX-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; BITALG_NOVLX-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; BITALG_NOVLX-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG_NOVLX-NEXT: vpopcntw %zmm0, %zmm0
; BITALG_NOVLX-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0
; BITALG_NOVLX-NEXT: retq
;
; BITALG-LABEL: testv16i16u:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; BITALG-NEXT: vpand %ymm1, %ymm0, %ymm0
; BITALG-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; BITALG-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; BITALG-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; BITALG-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG-NEXT: vpopcntw %ymm0, %ymm0
; BITALG-NEXT: retq
;
; X32-AVX-LABEL: testv16i16u:
; X32-AVX: # %bb.0:
-; X32-AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT: vpsubw %ymm0, %ymm1, %ymm1
-; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm0
; X32-AVX-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; X32-AVX-NEXT: vpaddw %ymm1, %ymm0, %ymm0
+; X32-AVX-NEXT: vpaddw %ymm1, %ymm0, %ymm1
+; X32-AVX-NEXT: vpandn %ymm1, %ymm0, %ymm0
; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm2
; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -959,38 +877,33 @@ define <32 x i8> @testv32i8(<32 x i8> %i
; AVX1-LABEL: testv32i8:
; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT: vpsubb %xmm1, %xmm2, %xmm3
-; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vpcmpeqd %xmm3, %xmm3, %xmm3
-; AVX1-NEXT: vpaddb %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT: vpand %xmm4, %xmm1, %xmm5
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT: vpshufb %xmm5, %xmm6, %xmm5
+; AVX1-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
+; AVX1-NEXT: vpaddb %xmm2, %xmm1, %xmm3
+; AVX1-NEXT: vpandn %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm4
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX1-NEXT: vpshufb %xmm4, %xmm5, %xmm4
; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm1
-; AVX1-NEXT: vpand %xmm4, %xmm1, %xmm1
-; AVX1-NEXT: vpshufb %xmm1, %xmm6, %xmm1
-; AVX1-NEXT: vpaddb %xmm5, %xmm1, %xmm1
-; AVX1-NEXT: vpsubb %xmm0, %xmm2, %xmm2
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vpaddb %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm2
-; AVX1-NEXT: vpshufb %xmm2, %xmm6, %xmm2
+; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vpshufb %xmm1, %xmm5, %xmm1
+; AVX1-NEXT: vpaddb %xmm4, %xmm1, %xmm1
+; AVX1-NEXT: vpaddb %xmm2, %xmm0, %xmm2
+; AVX1-NEXT: vpandn %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm2
+; AVX1-NEXT: vpshufb %xmm2, %xmm5, %xmm2
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm0
-; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vpshufb %xmm0, %xmm5, %xmm0
; AVX1-NEXT: vpaddb %xmm2, %xmm0, %xmm0
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: testv32i8:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX2-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX2-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; AVX2-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; AVX2-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -1003,11 +916,9 @@ define <32 x i8> @testv32i8(<32 x i8> %i
;
; AVX512CDVL-LABEL: testv32i8:
; AVX512CDVL: # %bb.0:
-; AVX512CDVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CDVL-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; AVX512CDVL-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512CDVL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512CDVL-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; AVX512CDVL-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; AVX512CDVL-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512CDVL-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -1020,11 +931,9 @@ define <32 x i8> @testv32i8(<32 x i8> %i
;
; AVX512CD-LABEL: testv32i8:
; AVX512CD: # %bb.0:
-; AVX512CD-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CD-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; AVX512CD-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512CD-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512CD-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; AVX512CD-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; AVX512CD-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512CD-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -1037,11 +946,9 @@ define <32 x i8> @testv32i8(<32 x i8> %i
;
; AVX512VPOPCNTDQ-LABEL: testv32i8:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512VPOPCNTDQ-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX512VPOPCNTDQ-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -1054,11 +961,9 @@ define <32 x i8> @testv32i8(<32 x i8> %i
;
; AVX512VPOPCNTDQVL-LABEL: testv32i8:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512VPOPCNTDQVL-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX512VPOPCNTDQVL-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -1071,32 +976,26 @@ define <32 x i8> @testv32i8(<32 x i8> %i
;
; BITALG_NOVLX-LABEL: testv32i8:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; BITALG_NOVLX-NEXT: vpand %ymm1, %ymm0, %ymm0
; BITALG_NOVLX-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; BITALG_NOVLX-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; BITALG_NOVLX-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; BITALG_NOVLX-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG_NOVLX-NEXT: vpopcntb %zmm0, %zmm0
; BITALG_NOVLX-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0
; BITALG_NOVLX-NEXT: retq
;
; BITALG-LABEL: testv32i8:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; BITALG-NEXT: vpand %ymm1, %ymm0, %ymm0
; BITALG-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; BITALG-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; BITALG-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; BITALG-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG-NEXT: vpopcntb %ymm0, %ymm0
; BITALG-NEXT: retq
;
; X32-AVX-LABEL: testv32i8:
; X32-AVX: # %bb.0:
-; X32-AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm0
; X32-AVX-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; X32-AVX-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; X32-AVX-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; X32-AVX-NEXT: vpandn %ymm1, %ymm0, %ymm0
; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm2
; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -1114,38 +1013,33 @@ define <32 x i8> @testv32i8u(<32 x i8> %
; AVX1-LABEL: testv32i8u:
; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT: vpsubb %xmm1, %xmm2, %xmm3
-; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vpcmpeqd %xmm3, %xmm3, %xmm3
-; AVX1-NEXT: vpaddb %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT: vpand %xmm4, %xmm1, %xmm5
-; AVX1-NEXT: vmovdqa {{.*#+}} xmm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT: vpshufb %xmm5, %xmm6, %xmm5
+; AVX1-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
+; AVX1-NEXT: vpaddb %xmm2, %xmm1, %xmm3
+; AVX1-NEXT: vpandn %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm4
+; AVX1-NEXT: vmovdqa {{.*#+}} xmm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX1-NEXT: vpshufb %xmm4, %xmm5, %xmm4
; AVX1-NEXT: vpsrlw $4, %xmm1, %xmm1
-; AVX1-NEXT: vpand %xmm4, %xmm1, %xmm1
-; AVX1-NEXT: vpshufb %xmm1, %xmm6, %xmm1
-; AVX1-NEXT: vpaddb %xmm5, %xmm1, %xmm1
-; AVX1-NEXT: vpsubb %xmm0, %xmm2, %xmm2
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vpaddb %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm2
-; AVX1-NEXT: vpshufb %xmm2, %xmm6, %xmm2
+; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1
+; AVX1-NEXT: vpshufb %xmm1, %xmm5, %xmm1
+; AVX1-NEXT: vpaddb %xmm4, %xmm1, %xmm1
+; AVX1-NEXT: vpaddb %xmm2, %xmm0, %xmm2
+; AVX1-NEXT: vpandn %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm2
+; AVX1-NEXT: vpshufb %xmm2, %xmm5, %xmm2
; AVX1-NEXT: vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm0
-; AVX1-NEXT: vpshufb %xmm0, %xmm6, %xmm0
+; AVX1-NEXT: vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT: vpshufb %xmm0, %xmm5, %xmm0
; AVX1-NEXT: vpaddb %xmm2, %xmm0, %xmm0
; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: testv32i8u:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX2-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX2-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; AVX2-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; AVX2-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -1158,11 +1052,9 @@ define <32 x i8> @testv32i8u(<32 x i8> %
;
; AVX512CDVL-LABEL: testv32i8u:
; AVX512CDVL: # %bb.0:
-; AVX512CDVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CDVL-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; AVX512CDVL-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512CDVL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512CDVL-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; AVX512CDVL-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; AVX512CDVL-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512CDVL-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX512CDVL-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -1175,11 +1067,9 @@ define <32 x i8> @testv32i8u(<32 x i8> %
;
; AVX512CD-LABEL: testv32i8u:
; AVX512CD: # %bb.0:
-; AVX512CD-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CD-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; AVX512CD-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512CD-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512CD-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; AVX512CD-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; AVX512CD-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512CD-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -1192,11 +1082,9 @@ define <32 x i8> @testv32i8u(<32 x i8> %
;
; AVX512VPOPCNTDQ-LABEL: testv32i8u:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512VPOPCNTDQ-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX512VPOPCNTDQ-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -1209,11 +1097,9 @@ define <32 x i8> @testv32i8u(<32 x i8> %
;
; AVX512VPOPCNTDQVL-LABEL: testv32i8u:
; AVX512VPOPCNTDQVL: # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VPOPCNTDQVL-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; AVX512VPOPCNTDQVL-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; AVX512VPOPCNTDQVL-NEXT: vpandn %ymm1, %ymm0, %ymm0
; AVX512VPOPCNTDQVL-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512VPOPCNTDQVL-NEXT: vpand %ymm1, %ymm0, %ymm2
; AVX512VPOPCNTDQVL-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -1226,32 +1112,26 @@ define <32 x i8> @testv32i8u(<32 x i8> %
;
; BITALG_NOVLX-LABEL: testv32i8u:
; BITALG_NOVLX: # %bb.0:
-; BITALG_NOVLX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; BITALG_NOVLX-NEXT: vpand %ymm1, %ymm0, %ymm0
; BITALG_NOVLX-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; BITALG_NOVLX-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; BITALG_NOVLX-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; BITALG_NOVLX-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG_NOVLX-NEXT: vpopcntb %zmm0, %zmm0
; BITALG_NOVLX-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0
; BITALG_NOVLX-NEXT: retq
;
; BITALG-LABEL: testv32i8u:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; BITALG-NEXT: vpand %ymm1, %ymm0, %ymm0
; BITALG-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; BITALG-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; BITALG-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; BITALG-NEXT: vpandn %ymm1, %ymm0, %ymm0
; BITALG-NEXT: vpopcntb %ymm0, %ymm0
; BITALG-NEXT: retq
;
; X32-AVX-LABEL: testv32i8u:
; X32-AVX: # %bb.0:
-; X32-AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT: vpsubb %ymm0, %ymm1, %ymm1
-; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm0
; X32-AVX-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
-; X32-AVX-NEXT: vpaddb %ymm1, %ymm0, %ymm0
+; X32-AVX-NEXT: vpaddb %ymm1, %ymm0, %ymm1
+; X32-AVX-NEXT: vpandn %ymm1, %ymm0, %ymm0
; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; X32-AVX-NEXT: vpand %ymm1, %ymm0, %ymm2
; X32-AVX-NEXT: vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
Modified: llvm/trunk/test/CodeGen/X86/vector-tzcnt-512.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-tzcnt-512.ll?rev=344447&r1=344446&r2=344447&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vector-tzcnt-512.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vector-tzcnt-512.ll Sat Oct 13 05:12:06 2018
@@ -8,11 +8,9 @@
define <8 x i64> @testv8i64(<8 x i64> %in) nounwind {
; AVX512CD-LABEL: testv8i64:
; AVX512CD: # %bb.0:
-; AVX512CD-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CD-NEXT: vpsubq %zmm0, %zmm1, %zmm1
-; AVX512CD-NEXT: vpandq %zmm1, %zmm0, %zmm0
; AVX512CD-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; AVX512CD-NEXT: vpaddq %zmm1, %zmm0, %zmm0
+; AVX512CD-NEXT: vpaddq %zmm1, %zmm0, %zmm1
+; AVX512CD-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; AVX512CD-NEXT: vextracti64x4 $1, %zmm0, %ymm1
; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512CD-NEXT: vpand %ymm2, %ymm1, %ymm3
@@ -36,58 +34,53 @@ define <8 x i64> @testv8i64(<8 x i64> %i
;
; AVX512CDBW-LABEL: testv8i64:
; AVX512CDBW: # %bb.0:
-; AVX512CDBW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CDBW-NEXT: vpsubq %zmm0, %zmm1, %zmm2
-; AVX512CDBW-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; AVX512CDBW-NEXT: vpternlogd $255, %zmm2, %zmm2, %zmm2
-; AVX512CDBW-NEXT: vpaddq %zmm2, %zmm0, %zmm0
-; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512CDBW-NEXT: vpandq %zmm2, %zmm0, %zmm3
-; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512CDBW-NEXT: vpshufb %zmm3, %zmm4, %zmm3
+; AVX512CDBW-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
+; AVX512CDBW-NEXT: vpaddq %zmm1, %zmm0, %zmm1
+; AVX512CDBW-NEXT: vpandnq %zmm1, %zmm0, %zmm0
+; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512CDBW-NEXT: vpandq %zmm1, %zmm0, %zmm2
+; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512CDBW-NEXT: vpshufb %zmm2, %zmm3, %zmm2
; AVX512CDBW-NEXT: vpsrlw $4, %zmm0, %zmm0
-; AVX512CDBW-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; AVX512CDBW-NEXT: vpshufb %zmm0, %zmm4, %zmm0
-; AVX512CDBW-NEXT: vpaddb %zmm3, %zmm0, %zmm0
+; AVX512CDBW-NEXT: vpandq %zmm1, %zmm0, %zmm0
+; AVX512CDBW-NEXT: vpshufb %zmm0, %zmm3, %zmm0
+; AVX512CDBW-NEXT: vpaddb %zmm2, %zmm0, %zmm0
+; AVX512CDBW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512CDBW-NEXT: vpsadbw %zmm1, %zmm0, %zmm0
; AVX512CDBW-NEXT: retq
;
; AVX512BW-LABEL: testv8i64:
; AVX512BW: # %bb.0:
-; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT: vpsubq %zmm0, %zmm1, %zmm2
-; AVX512BW-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT: vpternlogd $255, %zmm2, %zmm2, %zmm2
-; AVX512BW-NEXT: vpaddq %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT: vpandq %zmm2, %zmm0, %zmm3
-; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512BW-NEXT: vpshufb %zmm3, %zmm4, %zmm3
+; AVX512BW-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
+; AVX512BW-NEXT: vpaddq %zmm1, %zmm0, %zmm1
+; AVX512BW-NEXT: vpandnq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm2
+; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512BW-NEXT: vpshufb %zmm2, %zmm3, %zmm2
; AVX512BW-NEXT: vpsrlw $4, %zmm0, %zmm0
-; AVX512BW-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT: vpshufb %zmm0, %zmm4, %zmm0
-; AVX512BW-NEXT: vpaddb %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT: vpshufb %zmm0, %zmm3, %zmm0
+; AVX512BW-NEXT: vpaddb %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpsadbw %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VPOPCNTDQ-LABEL: testv8i64:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubq %zmm0, %zmm1, %zmm1
-; AVX512VPOPCNTDQ-NEXT: vpandq %zmm1, %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; AVX512VPOPCNTDQ-NEXT: vpaddq %zmm1, %zmm0, %zmm0
+; AVX512VPOPCNTDQ-NEXT: vpaddq %zmm1, %zmm0, %zmm1
+; AVX512VPOPCNTDQ-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpopcntq %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: retq
;
; BITALG-LABEL: testv8i64:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubq %zmm0, %zmm1, %zmm2
-; BITALG-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; BITALG-NEXT: vpternlogd $255, %zmm2, %zmm2, %zmm2
-; BITALG-NEXT: vpaddq %zmm2, %zmm0, %zmm0
+; BITALG-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
+; BITALG-NEXT: vpaddq %zmm1, %zmm0, %zmm1
+; BITALG-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; BITALG-NEXT: vpopcntb %zmm0, %zmm0
+; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG-NEXT: vpsadbw %zmm1, %zmm0, %zmm0
; BITALG-NEXT: retq
%out = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %in, i1 0)
@@ -117,40 +110,36 @@ define <8 x i64> @testv8i64u(<8 x i64> %
;
; AVX512BW-LABEL: testv8i64u:
; AVX512BW: # %bb.0:
-; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT: vpsubq %zmm0, %zmm1, %zmm2
-; AVX512BW-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT: vpternlogd $255, %zmm2, %zmm2, %zmm2
-; AVX512BW-NEXT: vpaddq %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT: vpandq %zmm2, %zmm0, %zmm3
-; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512BW-NEXT: vpshufb %zmm3, %zmm4, %zmm3
+; AVX512BW-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
+; AVX512BW-NEXT: vpaddq %zmm1, %zmm0, %zmm1
+; AVX512BW-NEXT: vpandnq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm2
+; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512BW-NEXT: vpshufb %zmm2, %zmm3, %zmm2
; AVX512BW-NEXT: vpsrlw $4, %zmm0, %zmm0
-; AVX512BW-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT: vpshufb %zmm0, %zmm4, %zmm0
-; AVX512BW-NEXT: vpaddb %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT: vpshufb %zmm0, %zmm3, %zmm0
+; AVX512BW-NEXT: vpaddb %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpsadbw %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VPOPCNTDQ-LABEL: testv8i64u:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubq %zmm0, %zmm1, %zmm1
-; AVX512VPOPCNTDQ-NEXT: vpandq %zmm1, %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; AVX512VPOPCNTDQ-NEXT: vpaddq %zmm1, %zmm0, %zmm0
+; AVX512VPOPCNTDQ-NEXT: vpaddq %zmm1, %zmm0, %zmm1
+; AVX512VPOPCNTDQ-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpopcntq %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: retq
;
; BITALG-LABEL: testv8i64u:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubq %zmm0, %zmm1, %zmm2
-; BITALG-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; BITALG-NEXT: vpternlogd $255, %zmm2, %zmm2, %zmm2
-; BITALG-NEXT: vpaddq %zmm2, %zmm0, %zmm0
+; BITALG-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
+; BITALG-NEXT: vpaddq %zmm1, %zmm0, %zmm1
+; BITALG-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; BITALG-NEXT: vpopcntb %zmm0, %zmm0
+; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG-NEXT: vpsadbw %zmm1, %zmm0, %zmm0
; BITALG-NEXT: retq
%out = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %in, i1 -1)
@@ -160,11 +149,9 @@ define <8 x i64> @testv8i64u(<8 x i64> %
define <16 x i32> @testv16i32(<16 x i32> %in) nounwind {
; AVX512CD-LABEL: testv16i32:
; AVX512CD: # %bb.0:
-; AVX512CD-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CD-NEXT: vpsubd %zmm0, %zmm1, %zmm1
-; AVX512CD-NEXT: vpandq %zmm1, %zmm0, %zmm0
; AVX512CD-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; AVX512CD-NEXT: vpaddd %zmm1, %zmm0, %zmm0
+; AVX512CD-NEXT: vpaddd %zmm1, %zmm0, %zmm1
+; AVX512CD-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; AVX512CD-NEXT: vextracti64x4 $1, %zmm0, %ymm1
; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512CD-NEXT: vpand %ymm2, %ymm1, %ymm3
@@ -196,19 +183,18 @@ define <16 x i32> @testv16i32(<16 x i32>
;
; AVX512CDBW-LABEL: testv16i32:
; AVX512CDBW: # %bb.0:
-; AVX512CDBW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CDBW-NEXT: vpsubd %zmm0, %zmm1, %zmm2
-; AVX512CDBW-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; AVX512CDBW-NEXT: vpternlogd $255, %zmm2, %zmm2, %zmm2
-; AVX512CDBW-NEXT: vpaddd %zmm2, %zmm0, %zmm0
-; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512CDBW-NEXT: vpandq %zmm2, %zmm0, %zmm3
-; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512CDBW-NEXT: vpshufb %zmm3, %zmm4, %zmm3
+; AVX512CDBW-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
+; AVX512CDBW-NEXT: vpaddd %zmm1, %zmm0, %zmm1
+; AVX512CDBW-NEXT: vpandnq %zmm1, %zmm0, %zmm0
+; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512CDBW-NEXT: vpandq %zmm1, %zmm0, %zmm2
+; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512CDBW-NEXT: vpshufb %zmm2, %zmm3, %zmm2
; AVX512CDBW-NEXT: vpsrlw $4, %zmm0, %zmm0
-; AVX512CDBW-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; AVX512CDBW-NEXT: vpshufb %zmm0, %zmm4, %zmm0
-; AVX512CDBW-NEXT: vpaddb %zmm3, %zmm0, %zmm0
+; AVX512CDBW-NEXT: vpandq %zmm1, %zmm0, %zmm0
+; AVX512CDBW-NEXT: vpshufb %zmm0, %zmm3, %zmm0
+; AVX512CDBW-NEXT: vpaddb %zmm2, %zmm0, %zmm0
+; AVX512CDBW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512CDBW-NEXT: vpunpckhdq {{.*#+}} zmm2 = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
; AVX512CDBW-NEXT: vpsadbw %zmm1, %zmm2, %zmm2
; AVX512CDBW-NEXT: vpunpckldq {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
@@ -218,19 +204,18 @@ define <16 x i32> @testv16i32(<16 x i32>
;
; AVX512BW-LABEL: testv16i32:
; AVX512BW: # %bb.0:
-; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT: vpsubd %zmm0, %zmm1, %zmm2
-; AVX512BW-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT: vpternlogd $255, %zmm2, %zmm2, %zmm2
-; AVX512BW-NEXT: vpaddd %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT: vpandq %zmm2, %zmm0, %zmm3
-; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512BW-NEXT: vpshufb %zmm3, %zmm4, %zmm3
+; AVX512BW-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
+; AVX512BW-NEXT: vpaddd %zmm1, %zmm0, %zmm1
+; AVX512BW-NEXT: vpandnq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm2
+; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512BW-NEXT: vpshufb %zmm2, %zmm3, %zmm2
; AVX512BW-NEXT: vpsrlw $4, %zmm0, %zmm0
-; AVX512BW-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT: vpshufb %zmm0, %zmm4, %zmm0
-; AVX512BW-NEXT: vpaddb %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT: vpshufb %zmm0, %zmm3, %zmm0
+; AVX512BW-NEXT: vpaddb %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpunpckhdq {{.*#+}} zmm2 = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
; AVX512BW-NEXT: vpsadbw %zmm1, %zmm2, %zmm2
; AVX512BW-NEXT: vpunpckldq {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
@@ -240,22 +225,19 @@ define <16 x i32> @testv16i32(<16 x i32>
;
; AVX512VPOPCNTDQ-LABEL: testv16i32:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubd %zmm0, %zmm1, %zmm1
-; AVX512VPOPCNTDQ-NEXT: vpandq %zmm1, %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; AVX512VPOPCNTDQ-NEXT: vpaddd %zmm1, %zmm0, %zmm0
+; AVX512VPOPCNTDQ-NEXT: vpaddd %zmm1, %zmm0, %zmm1
+; AVX512VPOPCNTDQ-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: retq
;
; BITALG-LABEL: testv16i32:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubd %zmm0, %zmm1, %zmm2
-; BITALG-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; BITALG-NEXT: vpternlogd $255, %zmm2, %zmm2, %zmm2
-; BITALG-NEXT: vpaddd %zmm2, %zmm0, %zmm0
+; BITALG-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
+; BITALG-NEXT: vpaddd %zmm1, %zmm0, %zmm1
+; BITALG-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; BITALG-NEXT: vpopcntb %zmm0, %zmm0
+; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG-NEXT: vpunpckhdq {{.*#+}} zmm2 = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
; BITALG-NEXT: vpsadbw %zmm1, %zmm2, %zmm2
; BITALG-NEXT: vpunpckldq {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
@@ -289,19 +271,18 @@ define <16 x i32> @testv16i32u(<16 x i32
;
; AVX512BW-LABEL: testv16i32u:
; AVX512BW: # %bb.0:
-; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT: vpsubd %zmm0, %zmm1, %zmm2
-; AVX512BW-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT: vpternlogd $255, %zmm2, %zmm2, %zmm2
-; AVX512BW-NEXT: vpaddd %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT: vpandq %zmm2, %zmm0, %zmm3
-; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512BW-NEXT: vpshufb %zmm3, %zmm4, %zmm3
+; AVX512BW-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
+; AVX512BW-NEXT: vpaddd %zmm1, %zmm0, %zmm1
+; AVX512BW-NEXT: vpandnq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm2
+; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512BW-NEXT: vpshufb %zmm2, %zmm3, %zmm2
; AVX512BW-NEXT: vpsrlw $4, %zmm0, %zmm0
-; AVX512BW-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT: vpshufb %zmm0, %zmm4, %zmm0
-; AVX512BW-NEXT: vpaddb %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT: vpshufb %zmm0, %zmm3, %zmm0
+; AVX512BW-NEXT: vpaddb %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512BW-NEXT: vpunpckhdq {{.*#+}} zmm2 = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
; AVX512BW-NEXT: vpsadbw %zmm1, %zmm2, %zmm2
; AVX512BW-NEXT: vpunpckldq {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
@@ -311,22 +292,19 @@ define <16 x i32> @testv16i32u(<16 x i32
;
; AVX512VPOPCNTDQ-LABEL: testv16i32u:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT: vpsubd %zmm0, %zmm1, %zmm1
-; AVX512VPOPCNTDQ-NEXT: vpandq %zmm1, %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; AVX512VPOPCNTDQ-NEXT: vpaddd %zmm1, %zmm0, %zmm0
+; AVX512VPOPCNTDQ-NEXT: vpaddd %zmm1, %zmm0, %zmm1
+; AVX512VPOPCNTDQ-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: retq
;
; BITALG-LABEL: testv16i32u:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubd %zmm0, %zmm1, %zmm2
-; BITALG-NEXT: vpandq %zmm2, %zmm0, %zmm0
-; BITALG-NEXT: vpternlogd $255, %zmm2, %zmm2, %zmm2
-; BITALG-NEXT: vpaddd %zmm2, %zmm0, %zmm0
+; BITALG-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
+; BITALG-NEXT: vpaddd %zmm1, %zmm0, %zmm1
+; BITALG-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; BITALG-NEXT: vpopcntb %zmm0, %zmm0
+; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
; BITALG-NEXT: vpunpckhdq {{.*#+}} zmm2 = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
; BITALG-NEXT: vpsadbw %zmm1, %zmm2, %zmm2
; BITALG-NEXT: vpunpckldq {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
@@ -340,30 +318,27 @@ define <16 x i32> @testv16i32u(<16 x i32
define <32 x i16> @testv32i16(<32 x i16> %in) nounwind {
; AVX512CD-LABEL: testv32i16:
; AVX512CD: # %bb.0:
-; AVX512CD-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX512CD-NEXT: vpsubw %ymm0, %ymm2, %ymm3
-; AVX512CD-NEXT: vpand %ymm3, %ymm0, %ymm0
-; AVX512CD-NEXT: vpcmpeqd %ymm3, %ymm3, %ymm3
-; AVX512CD-NEXT: vpaddw %ymm3, %ymm0, %ymm0
-; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512CD-NEXT: vpand %ymm4, %ymm0, %ymm5
-; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512CD-NEXT: vpshufb %ymm5, %ymm6, %ymm5
+; AVX512CD-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
+; AVX512CD-NEXT: vpaddw %ymm2, %ymm0, %ymm3
+; AVX512CD-NEXT: vpandn %ymm3, %ymm0, %ymm0
+; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512CD-NEXT: vpand %ymm3, %ymm0, %ymm4
+; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512CD-NEXT: vpshufb %ymm4, %ymm5, %ymm4
; AVX512CD-NEXT: vpsrlw $4, %ymm0, %ymm0
-; AVX512CD-NEXT: vpand %ymm4, %ymm0, %ymm0
-; AVX512CD-NEXT: vpshufb %ymm0, %ymm6, %ymm0
-; AVX512CD-NEXT: vpaddb %ymm5, %ymm0, %ymm0
-; AVX512CD-NEXT: vpsllw $8, %ymm0, %ymm5
-; AVX512CD-NEXT: vpaddb %ymm0, %ymm5, %ymm0
+; AVX512CD-NEXT: vpand %ymm3, %ymm0, %ymm0
+; AVX512CD-NEXT: vpshufb %ymm0, %ymm5, %ymm0
+; AVX512CD-NEXT: vpaddb %ymm4, %ymm0, %ymm0
+; AVX512CD-NEXT: vpsllw $8, %ymm0, %ymm4
+; AVX512CD-NEXT: vpaddb %ymm0, %ymm4, %ymm0
; AVX512CD-NEXT: vpsrlw $8, %ymm0, %ymm0
-; AVX512CD-NEXT: vpsubw %ymm1, %ymm2, %ymm2
-; AVX512CD-NEXT: vpand %ymm2, %ymm1, %ymm1
-; AVX512CD-NEXT: vpaddw %ymm3, %ymm1, %ymm1
-; AVX512CD-NEXT: vpand %ymm4, %ymm1, %ymm2
-; AVX512CD-NEXT: vpshufb %ymm2, %ymm6, %ymm2
+; AVX512CD-NEXT: vpaddw %ymm2, %ymm1, %ymm2
+; AVX512CD-NEXT: vpandn %ymm2, %ymm1, %ymm1
+; AVX512CD-NEXT: vpand %ymm3, %ymm1, %ymm2
+; AVX512CD-NEXT: vpshufb %ymm2, %ymm5, %ymm2
; AVX512CD-NEXT: vpsrlw $4, %ymm1, %ymm1
-; AVX512CD-NEXT: vpand %ymm4, %ymm1, %ymm1
-; AVX512CD-NEXT: vpshufb %ymm1, %ymm6, %ymm1
+; AVX512CD-NEXT: vpand %ymm3, %ymm1, %ymm1
+; AVX512CD-NEXT: vpshufb %ymm1, %ymm5, %ymm1
; AVX512CD-NEXT: vpaddb %ymm2, %ymm1, %ymm1
; AVX512CD-NEXT: vpsllw $8, %ymm1, %ymm2
; AVX512CD-NEXT: vpaddb %ymm1, %ymm2, %ymm1
@@ -372,11 +347,9 @@ define <32 x i16> @testv32i16(<32 x i16>
;
; AVX512CDBW-LABEL: testv32i16:
; AVX512CDBW: # %bb.0:
-; AVX512CDBW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CDBW-NEXT: vpsubw %zmm0, %zmm1, %zmm1
-; AVX512CDBW-NEXT: vpandq %zmm1, %zmm0, %zmm0
; AVX512CDBW-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; AVX512CDBW-NEXT: vpaddw %zmm1, %zmm0, %zmm0
+; AVX512CDBW-NEXT: vpaddw %zmm1, %zmm0, %zmm1
+; AVX512CDBW-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512CDBW-NEXT: vpandq %zmm1, %zmm0, %zmm2
; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -392,11 +365,9 @@ define <32 x i16> @testv32i16(<32 x i16>
;
; AVX512BW-LABEL: testv32i16:
; AVX512BW: # %bb.0:
-; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT: vpsubw %zmm0, %zmm1, %zmm1
-; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; AVX512BW-NEXT: vpaddw %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT: vpaddw %zmm1, %zmm0, %zmm1
+; AVX512BW-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm2
; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -412,17 +383,14 @@ define <32 x i16> @testv32i16(<32 x i16>
;
; AVX512VPOPCNTDQ-LABEL: testv32i16:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX512VPOPCNTDQ-NEXT: vpsubw %ymm0, %ymm2, %ymm3
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm3, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm3, %ymm3, %ymm3
-; AVX512VPOPCNTDQ-NEXT: vpaddw %ymm3, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
+; AVX512VPOPCNTDQ-NEXT: vpaddw %ymm2, %ymm0, %ymm3
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm3, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpmovdw %zmm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vpsubw %ymm1, %ymm2, %ymm2
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm2, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpaddw %ymm3, %ymm1, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpaddw %ymm2, %ymm1, %ymm2
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm2, %ymm1, %ymm1
; AVX512VPOPCNTDQ-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm1, %zmm1
; AVX512VPOPCNTDQ-NEXT: vpmovdw %zmm1, %ymm1
@@ -430,11 +398,9 @@ define <32 x i16> @testv32i16(<32 x i16>
;
; BITALG-LABEL: testv32i16:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubw %zmm0, %zmm1, %zmm1
-; BITALG-NEXT: vpandq %zmm1, %zmm0, %zmm0
; BITALG-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; BITALG-NEXT: vpaddw %zmm1, %zmm0, %zmm0
+; BITALG-NEXT: vpaddw %zmm1, %zmm0, %zmm1
+; BITALG-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; BITALG-NEXT: vpopcntw %zmm0, %zmm0
; BITALG-NEXT: retq
%out = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %in, i1 0)
@@ -444,30 +410,27 @@ define <32 x i16> @testv32i16(<32 x i16>
define <32 x i16> @testv32i16u(<32 x i16> %in) nounwind {
; AVX512CD-LABEL: testv32i16u:
; AVX512CD: # %bb.0:
-; AVX512CD-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX512CD-NEXT: vpsubw %ymm0, %ymm2, %ymm3
-; AVX512CD-NEXT: vpand %ymm3, %ymm0, %ymm0
-; AVX512CD-NEXT: vpcmpeqd %ymm3, %ymm3, %ymm3
-; AVX512CD-NEXT: vpaddw %ymm3, %ymm0, %ymm0
-; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512CD-NEXT: vpand %ymm4, %ymm0, %ymm5
-; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512CD-NEXT: vpshufb %ymm5, %ymm6, %ymm5
+; AVX512CD-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
+; AVX512CD-NEXT: vpaddw %ymm2, %ymm0, %ymm3
+; AVX512CD-NEXT: vpandn %ymm3, %ymm0, %ymm0
+; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512CD-NEXT: vpand %ymm3, %ymm0, %ymm4
+; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512CD-NEXT: vpshufb %ymm4, %ymm5, %ymm4
; AVX512CD-NEXT: vpsrlw $4, %ymm0, %ymm0
-; AVX512CD-NEXT: vpand %ymm4, %ymm0, %ymm0
-; AVX512CD-NEXT: vpshufb %ymm0, %ymm6, %ymm0
-; AVX512CD-NEXT: vpaddb %ymm5, %ymm0, %ymm0
-; AVX512CD-NEXT: vpsllw $8, %ymm0, %ymm5
-; AVX512CD-NEXT: vpaddb %ymm0, %ymm5, %ymm0
+; AVX512CD-NEXT: vpand %ymm3, %ymm0, %ymm0
+; AVX512CD-NEXT: vpshufb %ymm0, %ymm5, %ymm0
+; AVX512CD-NEXT: vpaddb %ymm4, %ymm0, %ymm0
+; AVX512CD-NEXT: vpsllw $8, %ymm0, %ymm4
+; AVX512CD-NEXT: vpaddb %ymm0, %ymm4, %ymm0
; AVX512CD-NEXT: vpsrlw $8, %ymm0, %ymm0
-; AVX512CD-NEXT: vpsubw %ymm1, %ymm2, %ymm2
-; AVX512CD-NEXT: vpand %ymm2, %ymm1, %ymm1
-; AVX512CD-NEXT: vpaddw %ymm3, %ymm1, %ymm1
-; AVX512CD-NEXT: vpand %ymm4, %ymm1, %ymm2
-; AVX512CD-NEXT: vpshufb %ymm2, %ymm6, %ymm2
+; AVX512CD-NEXT: vpaddw %ymm2, %ymm1, %ymm2
+; AVX512CD-NEXT: vpandn %ymm2, %ymm1, %ymm1
+; AVX512CD-NEXT: vpand %ymm3, %ymm1, %ymm2
+; AVX512CD-NEXT: vpshufb %ymm2, %ymm5, %ymm2
; AVX512CD-NEXT: vpsrlw $4, %ymm1, %ymm1
-; AVX512CD-NEXT: vpand %ymm4, %ymm1, %ymm1
-; AVX512CD-NEXT: vpshufb %ymm1, %ymm6, %ymm1
+; AVX512CD-NEXT: vpand %ymm3, %ymm1, %ymm1
+; AVX512CD-NEXT: vpshufb %ymm1, %ymm5, %ymm1
; AVX512CD-NEXT: vpaddb %ymm2, %ymm1, %ymm1
; AVX512CD-NEXT: vpsllw $8, %ymm1, %ymm2
; AVX512CD-NEXT: vpaddb %ymm1, %ymm2, %ymm1
@@ -476,11 +439,9 @@ define <32 x i16> @testv32i16u(<32 x i16
;
; AVX512CDBW-LABEL: testv32i16u:
; AVX512CDBW: # %bb.0:
-; AVX512CDBW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CDBW-NEXT: vpsubw %zmm0, %zmm1, %zmm1
-; AVX512CDBW-NEXT: vpandq %zmm1, %zmm0, %zmm0
; AVX512CDBW-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; AVX512CDBW-NEXT: vpaddw %zmm1, %zmm0, %zmm0
+; AVX512CDBW-NEXT: vpaddw %zmm1, %zmm0, %zmm1
+; AVX512CDBW-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512CDBW-NEXT: vpandq %zmm1, %zmm0, %zmm2
; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -496,11 +457,9 @@ define <32 x i16> @testv32i16u(<32 x i16
;
; AVX512BW-LABEL: testv32i16u:
; AVX512BW: # %bb.0:
-; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT: vpsubw %zmm0, %zmm1, %zmm1
-; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; AVX512BW-NEXT: vpaddw %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT: vpaddw %zmm1, %zmm0, %zmm1
+; AVX512BW-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm2
; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -516,17 +475,14 @@ define <32 x i16> @testv32i16u(<32 x i16
;
; AVX512VPOPCNTDQ-LABEL: testv32i16u:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX512VPOPCNTDQ-NEXT: vpsubw %ymm0, %ymm2, %ymm3
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm3, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm3, %ymm3, %ymm3
-; AVX512VPOPCNTDQ-NEXT: vpaddw %ymm3, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
+; AVX512VPOPCNTDQ-NEXT: vpaddw %ymm2, %ymm0, %ymm3
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm3, %ymm0, %ymm0
; AVX512VPOPCNTDQ-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm0, %zmm0
; AVX512VPOPCNTDQ-NEXT: vpmovdw %zmm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vpsubw %ymm1, %ymm2, %ymm2
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm2, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpaddw %ymm3, %ymm1, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpaddw %ymm2, %ymm1, %ymm2
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm2, %ymm1, %ymm1
; AVX512VPOPCNTDQ-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
; AVX512VPOPCNTDQ-NEXT: vpopcntd %zmm1, %zmm1
; AVX512VPOPCNTDQ-NEXT: vpmovdw %zmm1, %ymm1
@@ -534,11 +490,9 @@ define <32 x i16> @testv32i16u(<32 x i16
;
; BITALG-LABEL: testv32i16u:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubw %zmm0, %zmm1, %zmm1
-; BITALG-NEXT: vpandq %zmm1, %zmm0, %zmm0
; BITALG-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; BITALG-NEXT: vpaddw %zmm1, %zmm0, %zmm0
+; BITALG-NEXT: vpaddw %zmm1, %zmm0, %zmm1
+; BITALG-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; BITALG-NEXT: vpopcntw %zmm0, %zmm0
; BITALG-NEXT: retq
%out = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %in, i1 -1)
@@ -548,37 +502,32 @@ define <32 x i16> @testv32i16u(<32 x i16
define <64 x i8> @testv64i8(<64 x i8> %in) nounwind {
; AVX512CD-LABEL: testv64i8:
; AVX512CD: # %bb.0:
-; AVX512CD-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX512CD-NEXT: vpsubb %ymm0, %ymm2, %ymm3
-; AVX512CD-NEXT: vpand %ymm3, %ymm0, %ymm0
-; AVX512CD-NEXT: vpcmpeqd %ymm3, %ymm3, %ymm3
-; AVX512CD-NEXT: vpaddb %ymm3, %ymm0, %ymm0
-; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512CD-NEXT: vpand %ymm4, %ymm0, %ymm5
-; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512CD-NEXT: vpshufb %ymm5, %ymm6, %ymm5
+; AVX512CD-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
+; AVX512CD-NEXT: vpaddb %ymm2, %ymm0, %ymm3
+; AVX512CD-NEXT: vpandn %ymm3, %ymm0, %ymm0
+; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512CD-NEXT: vpand %ymm3, %ymm0, %ymm4
+; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512CD-NEXT: vpshufb %ymm4, %ymm5, %ymm4
; AVX512CD-NEXT: vpsrlw $4, %ymm0, %ymm0
-; AVX512CD-NEXT: vpand %ymm4, %ymm0, %ymm0
-; AVX512CD-NEXT: vpshufb %ymm0, %ymm6, %ymm0
-; AVX512CD-NEXT: vpaddb %ymm5, %ymm0, %ymm0
-; AVX512CD-NEXT: vpsubb %ymm1, %ymm2, %ymm2
-; AVX512CD-NEXT: vpand %ymm2, %ymm1, %ymm1
-; AVX512CD-NEXT: vpaddb %ymm3, %ymm1, %ymm1
-; AVX512CD-NEXT: vpand %ymm4, %ymm1, %ymm2
-; AVX512CD-NEXT: vpshufb %ymm2, %ymm6, %ymm2
+; AVX512CD-NEXT: vpand %ymm3, %ymm0, %ymm0
+; AVX512CD-NEXT: vpshufb %ymm0, %ymm5, %ymm0
+; AVX512CD-NEXT: vpaddb %ymm4, %ymm0, %ymm0
+; AVX512CD-NEXT: vpaddb %ymm2, %ymm1, %ymm2
+; AVX512CD-NEXT: vpandn %ymm2, %ymm1, %ymm1
+; AVX512CD-NEXT: vpand %ymm3, %ymm1, %ymm2
+; AVX512CD-NEXT: vpshufb %ymm2, %ymm5, %ymm2
; AVX512CD-NEXT: vpsrlw $4, %ymm1, %ymm1
-; AVX512CD-NEXT: vpand %ymm4, %ymm1, %ymm1
-; AVX512CD-NEXT: vpshufb %ymm1, %ymm6, %ymm1
+; AVX512CD-NEXT: vpand %ymm3, %ymm1, %ymm1
+; AVX512CD-NEXT: vpshufb %ymm1, %ymm5, %ymm1
; AVX512CD-NEXT: vpaddb %ymm2, %ymm1, %ymm1
; AVX512CD-NEXT: retq
;
; AVX512CDBW-LABEL: testv64i8:
; AVX512CDBW: # %bb.0:
-; AVX512CDBW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CDBW-NEXT: vpsubb %zmm0, %zmm1, %zmm1
-; AVX512CDBW-NEXT: vpandq %zmm1, %zmm0, %zmm0
; AVX512CDBW-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; AVX512CDBW-NEXT: vpaddb %zmm1, %zmm0, %zmm0
+; AVX512CDBW-NEXT: vpaddb %zmm1, %zmm0, %zmm1
+; AVX512CDBW-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512CDBW-NEXT: vpandq %zmm1, %zmm0, %zmm2
; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -591,11 +540,9 @@ define <64 x i8> @testv64i8(<64 x i8> %i
;
; AVX512BW-LABEL: testv64i8:
; AVX512BW: # %bb.0:
-; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT: vpsubb %zmm0, %zmm1, %zmm1
-; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; AVX512BW-NEXT: vpaddb %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT: vpaddb %zmm1, %zmm0, %zmm1
+; AVX512BW-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm2
; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -608,37 +555,32 @@ define <64 x i8> @testv64i8(<64 x i8> %i
;
; AVX512VPOPCNTDQ-LABEL: testv64i8:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX512VPOPCNTDQ-NEXT: vpsubb %ymm0, %ymm2, %ymm3
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm3, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm3, %ymm3, %ymm3
-; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm3, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vmovdqa {{.*#+}} ymm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm4, %ymm0, %ymm5
-; AVX512VPOPCNTDQ-NEXT: vmovdqa {{.*#+}} ymm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm5, %ymm6, %ymm5
+; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
+; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm2, %ymm0, %ymm3
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm3, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VPOPCNTDQ-NEXT: vpand %ymm3, %ymm0, %ymm4
+; AVX512VPOPCNTDQ-NEXT: vmovdqa {{.*#+}} ymm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm4, %ymm5, %ymm4
; AVX512VPOPCNTDQ-NEXT: vpsrlw $4, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm4, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm0, %ymm6, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm5, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vpsubb %ymm1, %ymm2, %ymm2
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm2, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm3, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm4, %ymm1, %ymm2
-; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm2, %ymm6, %ymm2
+; AVX512VPOPCNTDQ-NEXT: vpand %ymm3, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm0, %ymm5, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm4, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm2, %ymm1, %ymm2
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm2, %ymm1, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpand %ymm3, %ymm1, %ymm2
+; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm2, %ymm5, %ymm2
; AVX512VPOPCNTDQ-NEXT: vpsrlw $4, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm4, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm1, %ymm6, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpand %ymm3, %ymm1, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm1, %ymm5, %ymm1
; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm2, %ymm1, %ymm1
; AVX512VPOPCNTDQ-NEXT: retq
;
; BITALG-LABEL: testv64i8:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubb %zmm0, %zmm1, %zmm1
-; BITALG-NEXT: vpandq %zmm1, %zmm0, %zmm0
; BITALG-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; BITALG-NEXT: vpaddb %zmm1, %zmm0, %zmm0
+; BITALG-NEXT: vpaddb %zmm1, %zmm0, %zmm1
+; BITALG-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; BITALG-NEXT: vpopcntb %zmm0, %zmm0
; BITALG-NEXT: retq
%out = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %in, i1 0)
@@ -648,37 +590,32 @@ define <64 x i8> @testv64i8(<64 x i8> %i
define <64 x i8> @testv64i8u(<64 x i8> %in) nounwind {
; AVX512CD-LABEL: testv64i8u:
; AVX512CD: # %bb.0:
-; AVX512CD-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX512CD-NEXT: vpsubb %ymm0, %ymm2, %ymm3
-; AVX512CD-NEXT: vpand %ymm3, %ymm0, %ymm0
-; AVX512CD-NEXT: vpcmpeqd %ymm3, %ymm3, %ymm3
-; AVX512CD-NEXT: vpaddb %ymm3, %ymm0, %ymm0
-; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512CD-NEXT: vpand %ymm4, %ymm0, %ymm5
-; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512CD-NEXT: vpshufb %ymm5, %ymm6, %ymm5
+; AVX512CD-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
+; AVX512CD-NEXT: vpaddb %ymm2, %ymm0, %ymm3
+; AVX512CD-NEXT: vpandn %ymm3, %ymm0, %ymm0
+; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512CD-NEXT: vpand %ymm3, %ymm0, %ymm4
+; AVX512CD-NEXT: vmovdqa {{.*#+}} ymm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512CD-NEXT: vpshufb %ymm4, %ymm5, %ymm4
; AVX512CD-NEXT: vpsrlw $4, %ymm0, %ymm0
-; AVX512CD-NEXT: vpand %ymm4, %ymm0, %ymm0
-; AVX512CD-NEXT: vpshufb %ymm0, %ymm6, %ymm0
-; AVX512CD-NEXT: vpaddb %ymm5, %ymm0, %ymm0
-; AVX512CD-NEXT: vpsubb %ymm1, %ymm2, %ymm2
-; AVX512CD-NEXT: vpand %ymm2, %ymm1, %ymm1
-; AVX512CD-NEXT: vpaddb %ymm3, %ymm1, %ymm1
-; AVX512CD-NEXT: vpand %ymm4, %ymm1, %ymm2
-; AVX512CD-NEXT: vpshufb %ymm2, %ymm6, %ymm2
+; AVX512CD-NEXT: vpand %ymm3, %ymm0, %ymm0
+; AVX512CD-NEXT: vpshufb %ymm0, %ymm5, %ymm0
+; AVX512CD-NEXT: vpaddb %ymm4, %ymm0, %ymm0
+; AVX512CD-NEXT: vpaddb %ymm2, %ymm1, %ymm2
+; AVX512CD-NEXT: vpandn %ymm2, %ymm1, %ymm1
+; AVX512CD-NEXT: vpand %ymm3, %ymm1, %ymm2
+; AVX512CD-NEXT: vpshufb %ymm2, %ymm5, %ymm2
; AVX512CD-NEXT: vpsrlw $4, %ymm1, %ymm1
-; AVX512CD-NEXT: vpand %ymm4, %ymm1, %ymm1
-; AVX512CD-NEXT: vpshufb %ymm1, %ymm6, %ymm1
+; AVX512CD-NEXT: vpand %ymm3, %ymm1, %ymm1
+; AVX512CD-NEXT: vpshufb %ymm1, %ymm5, %ymm1
; AVX512CD-NEXT: vpaddb %ymm2, %ymm1, %ymm1
; AVX512CD-NEXT: retq
;
; AVX512CDBW-LABEL: testv64i8u:
; AVX512CDBW: # %bb.0:
-; AVX512CDBW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512CDBW-NEXT: vpsubb %zmm0, %zmm1, %zmm1
-; AVX512CDBW-NEXT: vpandq %zmm1, %zmm0, %zmm0
; AVX512CDBW-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; AVX512CDBW-NEXT: vpaddb %zmm1, %zmm0, %zmm0
+; AVX512CDBW-NEXT: vpaddb %zmm1, %zmm0, %zmm1
+; AVX512CDBW-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512CDBW-NEXT: vpandq %zmm1, %zmm0, %zmm2
; AVX512CDBW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -691,11 +628,9 @@ define <64 x i8> @testv64i8u(<64 x i8> %
;
; AVX512BW-LABEL: testv64i8u:
; AVX512BW: # %bb.0:
-; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT: vpsubb %zmm0, %zmm1, %zmm1
-; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; AVX512BW-NEXT: vpaddb %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT: vpaddb %zmm1, %zmm0, %zmm1
+; AVX512BW-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm2
; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
@@ -708,37 +643,32 @@ define <64 x i8> @testv64i8u(<64 x i8> %
;
; AVX512VPOPCNTDQ-LABEL: testv64i8u:
; AVX512VPOPCNTDQ: # %bb.0:
-; AVX512VPOPCNTDQ-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX512VPOPCNTDQ-NEXT: vpsubb %ymm0, %ymm2, %ymm3
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm3, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm3, %ymm3, %ymm3
-; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm3, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vmovdqa {{.*#+}} ymm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm4, %ymm0, %ymm5
-; AVX512VPOPCNTDQ-NEXT: vmovdqa {{.*#+}} ymm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm5, %ymm6, %ymm5
+; AVX512VPOPCNTDQ-NEXT: vpcmpeqd %ymm2, %ymm2, %ymm2
+; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm2, %ymm0, %ymm3
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm3, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VPOPCNTDQ-NEXT: vpand %ymm3, %ymm0, %ymm4
+; AVX512VPOPCNTDQ-NEXT: vmovdqa {{.*#+}} ymm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm4, %ymm5, %ymm4
; AVX512VPOPCNTDQ-NEXT: vpsrlw $4, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm4, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm0, %ymm6, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm5, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT: vpsubb %ymm1, %ymm2, %ymm2
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm2, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm3, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm4, %ymm1, %ymm2
-; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm2, %ymm6, %ymm2
+; AVX512VPOPCNTDQ-NEXT: vpand %ymm3, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm0, %ymm5, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm4, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm2, %ymm1, %ymm2
+; AVX512VPOPCNTDQ-NEXT: vpandn %ymm2, %ymm1, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpand %ymm3, %ymm1, %ymm2
+; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm2, %ymm5, %ymm2
; AVX512VPOPCNTDQ-NEXT: vpsrlw $4, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpand %ymm4, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm1, %ymm6, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpand %ymm3, %ymm1, %ymm1
+; AVX512VPOPCNTDQ-NEXT: vpshufb %ymm1, %ymm5, %ymm1
; AVX512VPOPCNTDQ-NEXT: vpaddb %ymm2, %ymm1, %ymm1
; AVX512VPOPCNTDQ-NEXT: retq
;
; BITALG-LABEL: testv64i8u:
; BITALG: # %bb.0:
-; BITALG-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT: vpsubb %zmm0, %zmm1, %zmm1
-; BITALG-NEXT: vpandq %zmm1, %zmm0, %zmm0
; BITALG-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1
-; BITALG-NEXT: vpaddb %zmm1, %zmm0, %zmm0
+; BITALG-NEXT: vpaddb %zmm1, %zmm0, %zmm1
+; BITALG-NEXT: vpandnq %zmm1, %zmm0, %zmm0
; BITALG-NEXT: vpopcntb %zmm0, %zmm0
; BITALG-NEXT: retq
%out = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %in, i1 -1)
More information about the llvm-commits
mailing list