Index: lib/CodeGen/SelectionDAG/DAGCombiner.cpp =================================================================== --- lib/CodeGen/SelectionDAG/DAGCombiner.cpp +++ lib/CodeGen/SelectionDAG/DAGCombiner.cpp @@ -487,6 +487,10 @@ const SDLoc &DL); SDValue unfoldMaskedMerge(SDNode *N); SDValue unfoldExtremeBitClearingToShifts(SDNode *N); + SDValue simplifyREM(SDNode *N); + SDValue simplifyUREMWithConstant(EVT VT, SDValue REMNode, SDValue CompNode, + ISD::CondCode Cond, SelectionDAG &DAG, + const SDLoc &DL); SDValue SimplifySetCC(EVT VT, SDValue N0, SDValue N1, ISD::CondCode Cond, const SDLoc &DL, bool foldBooleans); SDValue rebuildSetCC(SDValue N); @@ -3785,8 +3789,7 @@ return SDValue(); } -// handles ISD::SREM and ISD::UREM -SDValue DAGCombiner::visitREM(SDNode *N) { +SDValue DAGCombiner::simplifyREM(SDNode *N) { unsigned Opcode = N->getOpcode(); SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); @@ -3835,6 +3838,26 @@ } } + // sdiv, srem -> sdivrem + if (SDValue DivRem = useDivRem(N)) + return DivRem.getValue(1); + + return SDValue(); +} + +// handles ISD::SREM and ISD::UREM +SDValue DAGCombiner::visitREM(SDNode *N) { + unsigned Opcode = N->getOpcode(); + SDValue N0 = N->getOperand(0); + SDValue N1 = N->getOperand(1); + EVT VT = N->getValueType(0); + + bool isSigned = (Opcode == ISD::SREM); + SDLoc DL(N); + + if (SDValue V = simplifyREM(N)) + return V; + AttributeList Attr = DAG.getMachineFunction().getFunction().getAttributes(); // If X/C can be simplified by the division-by-constant logic, lower @@ -3861,10 +3884,6 @@ } } - // sdiv, srem -> sdivrem - if (SDValue DivRem = useDivRem(N)) - return DivRem.getValue(1); - return SDValue(); } @@ -7783,6 +7802,86 @@ return SDValue(); } +// See "Test for Zero Remainder after Division by a Constant" in +// "Hacker's Delight" by Henry Warren +SDValue DAGCombiner::simplifyUREMWithConstant(EVT VT, SDValue REMNode, + SDValue CompNode, + ISD::CondCode Cond, + SelectionDAG &DAG, + const SDLoc &DL) { + // fold (seteq/ne (urem N, D), 0) -> (setule/ugt (rotr (mul N, P), K), Q) + // - D must be constant with D = D0 * 2^K where D0 is odd and D0 != 1 + // - P is the multiplicative inverse of D0 modulo 2^W + // - Q = floor((2^W - 1) / D0) + // where W is the width of the common type of N and D + assert((Cond == ISD::SETEQ || Cond == ISD::SETNE) && + "Only applicable for (in)equality comparisons."); + + if (SDValue V = simplifyREM(REMNode.getNode())) { + if (V.getOpcode() == ISD::UREM) + REMNode = V; + else + return DAG.getSetCC(DL, VT, V, CompNode, Cond); + } + + EVT REMVT = REMNode->getValueType(0); + if (!isTypeLegal(REMVT)) + return SDValue(); + + // If MUL is unavailable, we cannot proceed in any case + if (!TLI.isOperationLegalOrCustom(ISD::MUL, REMVT)) + return SDValue(); + + // TODO: Add non-uniform constant support + ConstantSDNode *Divisor = isConstOrConstSplat(REMNode->getOperand(1)); + ConstantSDNode *CompTarget = isConstOrConstSplat(CompNode); + if (!Divisor || !CompTarget || Divisor->isNullValue() || + !CompTarget->isNullValue()) + return SDValue(); + + const APInt &D = Divisor->getAPIntValue(); + + // Decompose D into D0 * 2^K + unsigned K = D.countTrailingZeros(); + bool DivisorIsEven = (K != 0); + APInt D0 = D.lshr(K); + + // P = inv(D0, 2^W) + // 2^W requires W + 1 bits, so we have to extend and then truncate + unsigned W = D.getBitWidth(); + APInt P = D0.zext(W + 1) + .multiplicativeInverse(APInt::getHighBitsSet(W + 1, 1)) + .trunc(W); + + // Q = floor((2^W - 1) / D0) + APInt Q = APInt::getAllOnesValue(W).udiv(D0); + + SDValue PVal = DAG.getConstant(P, DL, REMVT); + SDValue QVal = DAG.getConstant(Q, DL, REMVT); + // (mul N, P) + SDValue Mul = DAG.getNode(ISD::MUL, DL, REMVT, REMNode->getOperand(0), PVal); + AddToWorklist(Mul.getNode()); + + ISD::CondCode NewCC = (Cond == ISD::SETEQ) ? ISD::SETULE : ISD::SETUGT; + + // Rotate right only if D was even + if (DivisorIsEven) { + // We need ROTR to do this + if (!TLI.isOperationLegalOrCustom(ISD::ROTR, REMVT)) + return SDValue(); + SDValue ShAmt = + DAG.getConstant(K, DL, TLI.getShiftAmountTy(REMVT, DAG.getDataLayout())); + SDNodeFlags Flags; + Flags.setExact(true); + // UREM: (rotr (mul N, P), K) + SDValue Rotr = DAG.getNode(ISD::ROTR, DL, REMVT, Mul, ShAmt, Flags); + AddToWorklist(Rotr.getNode()); + return DAG.getSetCC(DL, VT, Rotr, QVal, NewCC); + } else { + return DAG.getSetCC(DL, VT, Mul, QVal, NewCC); + } +} + SDValue DAGCombiner::visitSELECT(SDNode *N) { SDValue N0 = N->getOperand(0); SDValue N1 = N->getOperand(1); @@ -8551,15 +8650,34 @@ } SDValue DAGCombiner::visitSETCC(SDNode *N) { + EVT VT = N->getValueType(0); + SDValue N0 = N->getOperand(0); + SDValue N1 = N->getOperand(1); + SDValue N2 = N->getOperand(2); + SDLoc DL(N); + + ISD::CondCode Cond = cast(N2)->get(); + // Fold remainder of division by a a constant + if (N0.getOpcode() == ISD::UREM && N0.hasOneUse() && + (Cond == ISD::SETEQ || Cond == ISD::SETNE)) { + AttributeList Attr = DAG.getMachineFunction().getFunction().getAttributes(); + + // When division is cheap or optimizing for minimum size, + // fall through to DIVREM creation by skipping this fold. + if (!TLI.isIntDivCheap(VT, Attr) && + !Attr.hasFnAttribute(Attribute::MinSize)) + if (SDValue V = simplifyUREMWithConstant(VT, N0, N1, Cond, DAG, DL)) + return V; + } + // setcc is very commonly used as an argument to brcond. This pattern // also lend itself to numerous combines and, as a result, it is desired // we keep the argument to a brcond as a setcc as much as possible. bool PreferSetCC = N->hasOneUse() && N->use_begin()->getOpcode() == ISD::BRCOND; - SDValue Combined = SimplifySetCC( - N->getValueType(0), N->getOperand(0), N->getOperand(1), - cast(N->getOperand(2))->get(), SDLoc(N), !PreferSetCC); + SDValue Combined = SimplifySetCC(VT, N0, N1, cast(N2)->get(), + SDLoc(N), !PreferSetCC); if (!Combined) return SDValue(); Index: test/CodeGen/AArch64/urem-seteq-vec-splat.ll =================================================================== --- test/CodeGen/AArch64/urem-seteq-vec-splat.ll +++ test/CodeGen/AArch64/urem-seteq-vec-splat.ll @@ -9,13 +9,9 @@ ; CHECK-NEXT: mov w8, #52429 ; CHECK-NEXT: movk w8, #52428, lsl #16 ; CHECK-NEXT: dup v2.4s, w8 -; CHECK-NEXT: umull2 v3.2d, v0.4s, v2.4s -; CHECK-NEXT: umull v2.2d, v0.2s, v2.2s -; CHECK-NEXT: uzp2 v2.4s, v2.4s, v3.4s -; CHECK-NEXT: movi v1.4s, #5 -; CHECK-NEXT: ushr v2.4s, v2.4s, #2 -; CHECK-NEXT: mls v0.4s, v2.4s, v1.4s -; CHECK-NEXT: cmeq v0.4s, v0.4s, #0 +; CHECK-NEXT: movi v1.16b, #51 +; CHECK-NEXT: mul v0.4s, v0.4s, v2.4s +; CHECK-NEXT: cmhs v0.4s, v1.4s, v0.4s ; CHECK-NEXT: movi v1.4s, #1 ; CHECK-NEXT: and v0.16b, v0.16b, v1.16b ; CHECK-NEXT: ret @@ -31,13 +27,9 @@ ; CHECK: // %bb.0: ; CHECK-NEXT: mov w8, #52429 ; CHECK-NEXT: dup v2.8h, w8 -; CHECK-NEXT: umull2 v3.4s, v0.8h, v2.8h -; CHECK-NEXT: umull v2.4s, v0.4h, v2.4h -; CHECK-NEXT: uzp2 v2.8h, v2.8h, v3.8h -; CHECK-NEXT: movi v1.8h, #5 -; CHECK-NEXT: ushr v2.8h, v2.8h, #2 -; CHECK-NEXT: mls v0.8h, v2.8h, v1.8h -; CHECK-NEXT: cmeq v0.8h, v0.8h, #0 +; CHECK-NEXT: movi v1.16b, #51 +; CHECK-NEXT: mul v0.8h, v0.8h, v2.8h +; CHECK-NEXT: cmhs v0.8h, v1.8h, v0.8h ; CHECK-NEXT: movi v1.8h, #1 ; CHECK-NEXT: and v0.16b, v0.16b, v1.16b ; CHECK-NEXT: ret Index: test/CodeGen/AArch64/urem-seteq.ll =================================================================== --- test/CodeGen/AArch64/urem-seteq.ll +++ test/CodeGen/AArch64/urem-seteq.ll @@ -10,11 +10,11 @@ ; CHECK: // %bb.0: ; CHECK-NEXT: mov w8, #52429 ; CHECK-NEXT: movk w8, #52428, lsl #16 -; CHECK-NEXT: umull x8, w0, w8 -; CHECK-NEXT: lsr x8, x8, #34 -; CHECK-NEXT: add w8, w8, w8, lsl #2 -; CHECK-NEXT: cmp w0, w8 -; CHECK-NEXT: cset w0, eq +; CHECK-NEXT: mov w9, #13108 +; CHECK-NEXT: mul w8, w0, w8 +; CHECK-NEXT: movk w9, #13107, lsl #16 +; CHECK-NEXT: cmp w8, w9 +; CHECK-NEXT: cset w0, lo ; CHECK-NEXT: ret %urem = urem i32 %X, 5 %cmp = icmp eq i32 %urem, 0 @@ -26,14 +26,11 @@ define i32 @test_urem_odd_bit30(i32 %X) nounwind readnone { ; CHECK-LABEL: test_urem_odd_bit30: ; CHECK: // %bb.0: -; CHECK-NEXT: mov w8, #-11 -; CHECK-NEXT: umull x8, w0, w8 -; CHECK-NEXT: mov w9, #3 -; CHECK-NEXT: lsr x8, x8, #62 -; CHECK-NEXT: movk w9, #16384, lsl #16 -; CHECK-NEXT: msub w8, w8, w9, w0 -; CHECK-NEXT: cmp w8, #0 // =0 -; CHECK-NEXT: cset w0, eq +; CHECK-NEXT: mov w8, #43691 +; CHECK-NEXT: movk w8, #27306, lsl #16 +; CHECK-NEXT: mul w8, w0, w8 +; CHECK-NEXT: cmp w8, #4 // =4 +; CHECK-NEXT: cset w0, lo ; CHECK-NEXT: ret %urem = urem i32 %X, 1073741827 %cmp = icmp eq i32 %urem, 0 @@ -45,14 +42,11 @@ define i32 @test_urem_odd_bit31(i32 %X) nounwind readnone { ; CHECK-LABEL: test_urem_odd_bit31: ; CHECK: // %bb.0: -; CHECK-NEXT: mov w8, w0 -; CHECK-NEXT: lsl x9, x8, #30 -; CHECK-NEXT: sub x8, x9, x8 -; CHECK-NEXT: lsr x8, x8, #61 -; CHECK-NEXT: mov w9, #-2147483645 -; CHECK-NEXT: msub w8, w8, w9, w0 -; CHECK-NEXT: cmp w8, #0 // =0 -; CHECK-NEXT: cset w0, eq +; CHECK-NEXT: mov w8, #43691 +; CHECK-NEXT: movk w8, #10922, lsl #16 +; CHECK-NEXT: mul w8, w0, w8 +; CHECK-NEXT: cmp w8, #2 // =2 +; CHECK-NEXT: cset w0, lo ; CHECK-NEXT: ret %urem = urem i32 %X, 2147483651 %cmp = icmp eq i32 %urem, 0 @@ -69,16 +63,15 @@ define i16 @test_urem_even(i16 %X) nounwind readnone { ; CHECK-LABEL: test_urem_even: ; CHECK: // %bb.0: -; CHECK-NEXT: mov w10, #9363 -; CHECK-NEXT: ubfx w9, w0, #1, #15 -; CHECK-NEXT: movk w10, #37449, lsl #16 -; CHECK-NEXT: umull x9, w9, w10 +; CHECK-NEXT: mov w9, #28087 ; CHECK-NEXT: and w8, w0, #0xffff -; CHECK-NEXT: lsr x9, x9, #34 -; CHECK-NEXT: mov w10, #14 -; CHECK-NEXT: msub w8, w9, w10, w8 -; CHECK-NEXT: cmp w8, #0 // =0 -; CHECK-NEXT: cset w0, ne +; CHECK-NEXT: movk w9, #46811, lsl #16 +; CHECK-NEXT: mul w8, w8, w9 +; CHECK-NEXT: mov w9, #18724 +; CHECK-NEXT: ror w8, w8, #1 +; CHECK-NEXT: movk w9, #9362, lsl #16 +; CHECK-NEXT: cmp w8, w9 +; CHECK-NEXT: cset w0, hi ; CHECK-NEXT: ret %urem = urem i16 %X, 14 %cmp = icmp ne i16 %urem, 0 @@ -90,14 +83,12 @@ define i32 @test_urem_even_bit30(i32 %X) nounwind readnone { ; CHECK-LABEL: test_urem_even_bit30: ; CHECK: // %bb.0: -; CHECK-NEXT: mov w8, #-415 -; CHECK-NEXT: umull x8, w0, w8 -; CHECK-NEXT: mov w9, #104 -; CHECK-NEXT: lsr x8, x8, #62 -; CHECK-NEXT: movk w9, #16384, lsl #16 -; CHECK-NEXT: msub w8, w8, w9, w0 -; CHECK-NEXT: cmp w8, #0 // =0 -; CHECK-NEXT: cset w0, eq +; CHECK-NEXT: mov w8, #20165 +; CHECK-NEXT: movk w8, #64748, lsl #16 +; CHECK-NEXT: mul w8, w0, w8 +; CHECK-NEXT: ror w8, w8, #3 +; CHECK-NEXT: cmp w8, #32 // =32 +; CHECK-NEXT: cset w0, lo ; CHECK-NEXT: ret %urem = urem i32 %X, 1073741928 %cmp = icmp eq i32 %urem, 0 @@ -109,15 +100,12 @@ define i32 @test_urem_even_bit31(i32 %X) nounwind readnone { ; CHECK-LABEL: test_urem_even_bit31: ; CHECK: // %bb.0: -; CHECK-NEXT: mov w8, #65435 -; CHECK-NEXT: movk w8, #32767, lsl #16 -; CHECK-NEXT: umull x8, w0, w8 -; CHECK-NEXT: mov w9, #102 -; CHECK-NEXT: lsr x8, x8, #62 -; CHECK-NEXT: movk w9, #32768, lsl #16 -; CHECK-NEXT: msub w8, w8, w9, w0 -; CHECK-NEXT: cmp w8, #0 // =0 -; CHECK-NEXT: cset w0, eq +; CHECK-NEXT: mov w8, #64251 +; CHECK-NEXT: movk w8, #47866, lsl #16 +; CHECK-NEXT: mul w8, w0, w8 +; CHECK-NEXT: ror w8, w8, #1 +; CHECK-NEXT: cmp w8, #4 // =4 +; CHECK-NEXT: cset w0, lo ; CHECK-NEXT: ret %urem = urem i32 %X, 2147483750 %cmp = icmp eq i32 %urem, 0 Index: test/CodeGen/X86/jump_sign.ll =================================================================== --- test/CodeGen/X86/jump_sign.ll +++ test/CodeGen/X86/jump_sign.ll @@ -236,13 +236,11 @@ ; CHECK-NEXT: jne .LBB12_8 ; CHECK-NEXT: # %bb.4: # %if.end29 ; CHECK-NEXT: movzwl (%eax), %eax +; CHECK-NEXT: imull $-13107, %eax, %eax # imm = 0xCCCD +; CHECK-NEXT: rorw %ax ; CHECK-NEXT: movzwl %ax, %eax -; CHECK-NEXT: imull $52429, %eax, %ecx # imm = 0xCCCD -; CHECK-NEXT: shrl $18, %ecx -; CHECK-NEXT: andl $-2, %ecx -; CHECK-NEXT: leal (%ecx,%ecx,4), %ecx -; CHECK-NEXT: cmpw %cx, %ax -; CHECK-NEXT: jne .LBB12_5 +; CHECK-NEXT: cmpl $13108, %eax # imm = 0x3334 +; CHECK-NEXT: jae .LBB12_5 ; CHECK-NEXT: .LBB12_8: # %if.then44 ; CHECK-NEXT: xorl %eax, %eax ; CHECK-NEXT: testb %al, %al @@ -399,11 +397,14 @@ ; CHECK-LABEL: func_test1: ; CHECK: # %bb.0: # %entry ; CHECK-NEXT: movl b, %eax +; CHECK-NEXT: xorl %ecx, %ecx ; CHECK-NEXT: cmpl {{[0-9]+}}(%esp), %eax ; CHECK-NEXT: setb %cl ; CHECK-NEXT: movl a, %eax -; CHECK-NEXT: testb %al, %cl -; CHECK-NEXT: je .LBB18_2 +; CHECK-NEXT: andl %eax, %ecx +; CHECK-NEXT: imull $-85, %ecx, %ecx +; CHECK-NEXT: cmpb $86, %cl +; CHECK-NEXT: jb .LBB18_2 ; CHECK-NEXT: # %bb.1: # %if.then ; CHECK-NEXT: decl %eax ; CHECK-NEXT: movl %eax, a Index: test/CodeGen/X86/urem-seteq-vec-splat.ll =================================================================== --- test/CodeGen/X86/urem-seteq-vec-splat.ll +++ test/CodeGen/X86/urem-seteq-vec-splat.ll @@ -11,84 +11,48 @@ ; CHECK-SSE2-LABEL: test_urem_odd_vec_i32: ; CHECK-SSE2: # %bb.0: ; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,3435973837,3435973837,3435973837] -; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2 +; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3] +; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm0 +; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] ; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3] -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm3 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3] -; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1] -; CHECK-SSE2-NEXT: psrld $2, %xmm2 -; CHECK-SSE2-NEXT: movdqa %xmm2, %xmm1 -; CHECK-SSE2-NEXT: pslld $2, %xmm1 -; CHECK-SSE2-NEXT: paddd %xmm2, %xmm1 -; CHECK-SSE2-NEXT: psubd %xmm1, %xmm0 -; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1 -; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0 -; CHECK-SSE2-NEXT: psrld $31, %xmm0 +; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3] +; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] +; CHECK-SSE2-NEXT: pxor {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: pcmpgtd {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: pandn {{.*}}(%rip), %xmm0 ; CHECK-SSE2-NEXT: retq ; ; CHECK-SSE41-LABEL: test_urem_odd_vec_i32: ; CHECK-SSE41: # %bb.0: -; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] -; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837] -; CHECK-SSE41-NEXT: pmuludq %xmm2, %xmm1 -; CHECK-SSE41-NEXT: pmuludq %xmm0, %xmm2 -; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7] -; CHECK-SSE41-NEXT: psrld $2, %xmm2 -; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm2 -; CHECK-SSE41-NEXT: psubd %xmm2, %xmm0 -; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1 +; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm0 +; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [858993459,858993459,858993459,858993459] +; CHECK-SSE41-NEXT: pminud %xmm0, %xmm1 ; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0 ; CHECK-SSE41-NEXT: psrld $31, %xmm0 ; CHECK-SSE41-NEXT: retq ; ; CHECK-AVX1-LABEL: test_urem_odd_vec_i32: ; CHECK-AVX1: # %bb.0: -; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] -; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837] -; CHECK-AVX1-NEXT: vpmuludq %xmm2, %xmm1, %xmm1 -; CHECK-AVX1-NEXT: vpmuludq %xmm2, %xmm0, %xmm2 -; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7] -; CHECK-AVX1-NEXT: vpsrld $2, %xmm1, %xmm1 -; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX1-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 ; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX1-NEXT: retq ; ; CHECK-AVX2-LABEL: test_urem_odd_vec_i32: ; CHECK-AVX2: # %bb.0: -; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] -; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837] -; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm0, %xmm2 -; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3] -; CHECK-AVX2-NEXT: vpsrld $2, %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm2 = [5,5,5,5] -; CHECK-AVX2-NEXT: vpmulld %xmm2, %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [3435973837,3435973837,3435973837,3435973837] +; CHECK-AVX2-NEXT: vpmulld %xmm1, %xmm0, %xmm0 +; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [858993459,858993459,858993459,858993459] +; CHECK-AVX2-NEXT: vpminud %xmm1, %xmm0, %xmm1 ; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX2-NEXT: retq ; ; CHECK-AVX512VL-LABEL: test_urem_odd_vec_i32: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpbroadcastd {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837] -; CHECK-AVX512VL-NEXT: vpmuludq %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmuludq %xmm2, %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3] -; CHECK-AVX512VL-NEXT: vpsrld $2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip){1to4}, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip){1to4}, %xmm0, %xmm1 ; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq @@ -100,25 +64,28 @@ ; Like test_urem_odd_vec_i32, but with 8 x i16 vectors. define <8 x i16> @test_urem_odd_vec_i16(<8 x i16> %X) nounwind readnone { -; CHECK-SSE-LABEL: test_urem_odd_vec_i16: -; CHECK-SSE: # %bb.0: -; CHECK-SSE-NEXT: movdqa {{.*#+}} xmm1 = [52429,52429,52429,52429,52429,52429,52429,52429] -; CHECK-SSE-NEXT: pmulhuw %xmm0, %xmm1 -; CHECK-SSE-NEXT: psrlw $2, %xmm1 -; CHECK-SSE-NEXT: pmullw {{.*}}(%rip), %xmm1 -; CHECK-SSE-NEXT: psubw %xmm1, %xmm0 -; CHECK-SSE-NEXT: pxor %xmm1, %xmm1 -; CHECK-SSE-NEXT: pcmpeqw %xmm1, %xmm0 -; CHECK-SSE-NEXT: psrlw $15, %xmm0 -; CHECK-SSE-NEXT: retq +; CHECK-SSE2-LABEL: test_urem_odd_vec_i16: +; CHECK-SSE2: # %bb.0: +; CHECK-SSE2-NEXT: pmullw {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: psubusw {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1 +; CHECK-SSE2-NEXT: pcmpeqw %xmm1, %xmm0 +; CHECK-SSE2-NEXT: psrlw $15, %xmm0 +; CHECK-SSE2-NEXT: retq +; +; CHECK-SSE41-LABEL: test_urem_odd_vec_i16: +; CHECK-SSE41: # %bb.0: +; CHECK-SSE41-NEXT: pmullw {{.*}}(%rip), %xmm0 +; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [13107,13107,13107,13107,13107,13107,13107,13107] +; CHECK-SSE41-NEXT: pminuw %xmm0, %xmm1 +; CHECK-SSE41-NEXT: pcmpeqw %xmm1, %xmm0 +; CHECK-SSE41-NEXT: psrlw $15, %xmm0 +; CHECK-SSE41-NEXT: retq ; ; CHECK-AVX-LABEL: test_urem_odd_vec_i16: ; CHECK-AVX: # %bb.0: -; CHECK-AVX-NEXT: vpmulhuw {{.*}}(%rip), %xmm0, %xmm1 -; CHECK-AVX-NEXT: vpsrlw $2, %xmm1, %xmm1 -; CHECK-AVX-NEXT: vpmullw {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX-NEXT: vpsubw %xmm1, %xmm0, %xmm0 -; CHECK-AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; CHECK-AVX-NEXT: vpmullw {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX-NEXT: vpminuw {{.*}}(%rip), %xmm0, %xmm1 ; CHECK-AVX-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0 ; CHECK-AVX-NEXT: vpsrlw $15, %xmm0, %xmm0 ; CHECK-AVX-NEXT: retq @@ -214,17 +181,9 @@ ; ; CHECK-AVX512VL-LABEL: test_urem_even_vec_i32: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vpsrld $1, %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpbroadcastd {{.*#+}} xmm3 = [2454267027,2454267027,2454267027,2454267027] -; CHECK-AVX512VL-NEXT: vpmuludq %xmm3, %xmm2, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuludq %xmm3, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsrld $2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip){1to4}, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprord $1, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip){1to4}, %xmm0, %xmm1 ; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq Index: test/CodeGen/X86/urem-seteq.ll =================================================================== --- test/CodeGen/X86/urem-seteq.ll +++ test/CodeGen/X86/urem-seteq.ll @@ -9,27 +9,18 @@ define i32 @test_urem_odd(i32 %X) nounwind readnone { ; X86-LABEL: test_urem_odd: ; X86: # %bb.0: -; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X86-NEXT: movl $-858993459, %edx # imm = 0xCCCCCCCD -; X86-NEXT: movl %ecx, %eax -; X86-NEXT: mull %edx -; X86-NEXT: shrl $2, %edx -; X86-NEXT: leal (%edx,%edx,4), %edx +; X86-NEXT: imull $-858993459, {{[0-9]+}}(%esp), %ecx # imm = 0xCCCCCCCD ; X86-NEXT: xorl %eax, %eax -; X86-NEXT: cmpl %edx, %ecx -; X86-NEXT: sete %al +; X86-NEXT: cmpl $858993460, %ecx # imm = 0x33333334 +; X86-NEXT: setb %al ; X86-NEXT: retl ; ; X64-LABEL: test_urem_odd: ; X64: # %bb.0: -; X64-NEXT: movl %edi, %eax -; X64-NEXT: movl $3435973837, %ecx # imm = 0xCCCCCCCD -; X64-NEXT: imulq %rax, %rcx -; X64-NEXT: shrq $34, %rcx -; X64-NEXT: leal (%rcx,%rcx,4), %ecx +; X64-NEXT: imull $-858993459, %edi, %ecx # imm = 0xCCCCCCCD ; X64-NEXT: xorl %eax, %eax -; X64-NEXT: cmpl %ecx, %edi -; X64-NEXT: sete %al +; X64-NEXT: cmpl $858993460, %ecx # imm = 0x33333334 +; X64-NEXT: setb %al ; X64-NEXT: retq %urem = urem i32 %X, 5 %cmp = icmp eq i32 %urem, 0 @@ -41,27 +32,18 @@ define i32 @test_urem_odd_bit30(i32 %X) nounwind readnone { ; X86-LABEL: test_urem_odd_bit30: ; X86: # %bb.0: -; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X86-NEXT: movl $-11, %edx -; X86-NEXT: movl %ecx, %eax -; X86-NEXT: mull %edx -; X86-NEXT: shrl $30, %edx -; X86-NEXT: imull $1073741827, %edx, %edx # imm = 0x40000003 +; X86-NEXT: imull $1789569707, {{[0-9]+}}(%esp), %ecx # imm = 0x6AAAAAAB ; X86-NEXT: xorl %eax, %eax -; X86-NEXT: cmpl %edx, %ecx -; X86-NEXT: sete %al +; X86-NEXT: cmpl $4, %ecx +; X86-NEXT: setb %al ; X86-NEXT: retl ; ; X64-LABEL: test_urem_odd_bit30: ; X64: # %bb.0: -; X64-NEXT: movl %edi, %eax -; X64-NEXT: movl $4294967285, %ecx # imm = 0xFFFFFFF5 -; X64-NEXT: imulq %rax, %rcx -; X64-NEXT: shrq $62, %rcx -; X64-NEXT: imull $1073741827, %ecx, %ecx # imm = 0x40000003 +; X64-NEXT: imull $1789569707, %edi, %ecx # imm = 0x6AAAAAAB ; X64-NEXT: xorl %eax, %eax -; X64-NEXT: cmpl %ecx, %edi -; X64-NEXT: sete %al +; X64-NEXT: cmpl $4, %ecx +; X64-NEXT: setb %al ; X64-NEXT: retq %urem = urem i32 %X, 1073741827 %cmp = icmp eq i32 %urem, 0 @@ -73,28 +55,18 @@ define i32 @test_urem_odd_bit31(i32 %X) nounwind readnone { ; X86-LABEL: test_urem_odd_bit31: ; X86: # %bb.0: -; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X86-NEXT: movl $1073741823, %edx # imm = 0x3FFFFFFF -; X86-NEXT: movl %ecx, %eax -; X86-NEXT: mull %edx -; X86-NEXT: shrl $29, %edx -; X86-NEXT: imull $-2147483645, %edx, %edx # imm = 0x80000003 +; X86-NEXT: imull $715827883, {{[0-9]+}}(%esp), %ecx # imm = 0x2AAAAAAB ; X86-NEXT: xorl %eax, %eax -; X86-NEXT: cmpl %edx, %ecx -; X86-NEXT: sete %al +; X86-NEXT: cmpl $2, %ecx +; X86-NEXT: setb %al ; X86-NEXT: retl ; ; X64-LABEL: test_urem_odd_bit31: ; X64: # %bb.0: -; X64-NEXT: movl %edi, %eax -; X64-NEXT: movq %rax, %rcx -; X64-NEXT: shlq $30, %rcx -; X64-NEXT: subq %rax, %rcx -; X64-NEXT: shrq $61, %rcx -; X64-NEXT: imull $-2147483645, %ecx, %ecx # imm = 0x80000003 +; X64-NEXT: imull $715827883, %edi, %ecx # imm = 0x2AAAAAAB ; X64-NEXT: xorl %eax, %eax -; X64-NEXT: cmpl %ecx, %edi -; X64-NEXT: sete %al +; X64-NEXT: cmpl $2, %ecx +; X64-NEXT: setb %al ; X64-NEXT: retq %urem = urem i32 %X, 2147483651 %cmp = icmp eq i32 %urem, 0 @@ -111,35 +83,23 @@ define i16 @test_urem_even(i16 %X) nounwind readnone { ; X86-LABEL: test_urem_even: ; X86: # %bb.0: -; X86-NEXT: movzwl {{[0-9]+}}(%esp), %ecx -; X86-NEXT: movl %ecx, %eax -; X86-NEXT: shrl %eax -; X86-NEXT: imull $18725, %eax, %eax # imm = 0x4925 -; X86-NEXT: shrl $17, %eax -; X86-NEXT: movl %eax, %edx -; X86-NEXT: shll $4, %edx -; X86-NEXT: subl %eax, %edx -; X86-NEXT: subl %eax, %edx +; X86-NEXT: imull $28087, {{[0-9]+}}(%esp), %eax # imm = 0x6DB7 +; X86-NEXT: rorw %ax +; X86-NEXT: movzwl %ax, %ecx ; X86-NEXT: xorl %eax, %eax -; X86-NEXT: cmpw %dx, %cx -; X86-NEXT: setne %al +; X86-NEXT: cmpl $9362, %ecx # imm = 0x2492 +; X86-NEXT: seta %al ; X86-NEXT: # kill: def $ax killed $ax killed $eax ; X86-NEXT: retl ; ; X64-LABEL: test_urem_even: ; X64: # %bb.0: -; X64-NEXT: movzwl %di, %ecx -; X64-NEXT: movl %ecx, %eax -; X64-NEXT: shrl %eax -; X64-NEXT: imull $18725, %eax, %eax # imm = 0x4925 -; X64-NEXT: shrl $17, %eax -; X64-NEXT: movl %eax, %edx -; X64-NEXT: shll $4, %edx -; X64-NEXT: subl %eax, %edx -; X64-NEXT: subl %eax, %edx +; X64-NEXT: imull $28087, %edi, %eax # imm = 0x6DB7 +; X64-NEXT: rorw %ax +; X64-NEXT: movzwl %ax, %ecx ; X64-NEXT: xorl %eax, %eax -; X64-NEXT: cmpw %dx, %cx -; X64-NEXT: setne %al +; X64-NEXT: cmpl $9362, %ecx # imm = 0x2492 +; X64-NEXT: seta %al ; X64-NEXT: # kill: def $ax killed $ax killed $eax ; X64-NEXT: retq %urem = urem i16 %X, 14 @@ -152,27 +112,20 @@ define i32 @test_urem_even_bit30(i32 %X) nounwind readnone { ; X86-LABEL: test_urem_even_bit30: ; X86: # %bb.0: -; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X86-NEXT: movl $-415, %edx # imm = 0xFE61 -; X86-NEXT: movl %ecx, %eax -; X86-NEXT: mull %edx -; X86-NEXT: shrl $30, %edx -; X86-NEXT: imull $1073741928, %edx, %edx # imm = 0x40000068 +; X86-NEXT: imull $-51622203, {{[0-9]+}}(%esp), %ecx # imm = 0xFCEC4EC5 +; X86-NEXT: rorl $3, %ecx ; X86-NEXT: xorl %eax, %eax -; X86-NEXT: cmpl %edx, %ecx -; X86-NEXT: sete %al +; X86-NEXT: cmpl $32, %ecx +; X86-NEXT: setb %al ; X86-NEXT: retl ; ; X64-LABEL: test_urem_even_bit30: ; X64: # %bb.0: -; X64-NEXT: movl %edi, %eax -; X64-NEXT: movl $4294966881, %ecx # imm = 0xFFFFFE61 -; X64-NEXT: imulq %rax, %rcx -; X64-NEXT: shrq $62, %rcx -; X64-NEXT: imull $1073741928, %ecx, %ecx # imm = 0x40000068 +; X64-NEXT: imull $-51622203, %edi, %ecx # imm = 0xFCEC4EC5 +; X64-NEXT: rorl $3, %ecx ; X64-NEXT: xorl %eax, %eax -; X64-NEXT: cmpl %ecx, %edi -; X64-NEXT: sete %al +; X64-NEXT: cmpl $32, %ecx +; X64-NEXT: setb %al ; X64-NEXT: retq %urem = urem i32 %X, 1073741928 %cmp = icmp eq i32 %urem, 0 @@ -184,26 +137,20 @@ define i32 @test_urem_even_bit31(i32 %X) nounwind readnone { ; X86-LABEL: test_urem_even_bit31: ; X86: # %bb.0: -; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X86-NEXT: movl $2147483547, %edx # imm = 0x7FFFFF9B -; X86-NEXT: movl %ecx, %eax -; X86-NEXT: mull %edx -; X86-NEXT: shrl $30, %edx -; X86-NEXT: imull $-2147483546, %edx, %edx # imm = 0x80000066 +; X86-NEXT: imull $-1157956869, {{[0-9]+}}(%esp), %ecx # imm = 0xBAFAFAFB +; X86-NEXT: rorl %ecx ; X86-NEXT: xorl %eax, %eax -; X86-NEXT: cmpl %edx, %ecx -; X86-NEXT: sete %al +; X86-NEXT: cmpl $4, %ecx +; X86-NEXT: setb %al ; X86-NEXT: retl ; ; X64-LABEL: test_urem_even_bit31: ; X64: # %bb.0: -; X64-NEXT: movl %edi, %eax -; X64-NEXT: imulq $2147483547, %rax, %rax # imm = 0x7FFFFF9B -; X64-NEXT: shrq $62, %rax -; X64-NEXT: imull $-2147483546, %eax, %ecx # imm = 0x80000066 +; X64-NEXT: imull $-1157956869, %edi, %ecx # imm = 0xBAFAFAFB +; X64-NEXT: rorl %ecx ; X64-NEXT: xorl %eax, %eax -; X64-NEXT: cmpl %ecx, %edi -; X64-NEXT: sete %al +; X64-NEXT: cmpl $4, %ecx +; X64-NEXT: setb %al ; X64-NEXT: retq %urem = urem i32 %X, 2147483750 %cmp = icmp eq i32 %urem, 0