[llvm] r304209 - This patch closes PR28513: an optimization of multiplication by different constants.

Vedant Kumar via llvm-commits llvm-commits at lists.llvm.org
Tue May 30 12:22:13 PDT 2017


Hi Andrew,

I think this change is responsible for a tablgen failure in stage2 builds:

  http://green.lab.llvm.org/green/job/clang-stage2-configure-Rthinlto_build/2171/ <http://green.lab.llvm.org/green/job/clang-stage2-configure-Rthinlto_build/2171/>

I reproduced the failure locally (without ThinLTO), reverted your commit, rebuilt the stage1 clang, rebuilt the stage2 llvm-tblgen tool, and found that the crash disappears when your commit is reverted.

Could you take a look?

FAILED: lib/Target/ARM/ARMGenRegisterBank.inc.tmp
cd /Volumes/Builds/pz-master-stage2-RA/lib/Target/ARM && /Volumes/Builds/pz-master-stage2-RA/bin/llvm-tblgen -gen-register-bank -I /Users/vk/llvm/lib/Target/ARM -I /Users/vk/llvm/include -I /Users/vk/llvm/lib/Target /Users/vk/llvm/lib/Target/ARM/ARM.td -o /Volumes
/Builds/pz-master-stage2-RA/lib/Target/ARM/ARMGenRegisterBank.inc.tmp
0  llvm-tblgen              0x0000000106fc9568 llvm::sys::PrintStackTrace(llvm::raw_ostream&) + 40
1  llvm-tblgen              0x0000000106fc9be6 SignalHandler(int) + 422
2  libsystem_platform.dylib 0x00000001076a7fba _sigtramp + 26
3  libsystem_platform.dylib 0x00007fff58deb468 _sigtramp + 1366570184
4  llvm-tblgen              0x0000000106e89cc7 llvm::CodeGenRegBank::getCompositeSubRegIndex(llvm::CodeGenSubRegIndex*, llvm::CodeGenSubRegIndex*) + 615
5  llvm-tblgen              0x0000000106e88be6 llvm::CodeGenRegister::computeSubRegs(llvm::CodeGenRegBank&) + 2182
6  llvm-tblgen              0x0000000106e8e9f0 llvm::CodeGenRegBank::CodeGenRegBank(llvm::RecordKeeper&) + 2192
7  llvm-tblgen              0x0000000106f384a1 llvm::EmitRegisterBank(llvm::RecordKeeper&, llvm::raw_ostream&) + 65
8  llvm-tblgen              0x0000000106f72c64 (anonymous namespace)::LLVMTableGenMain(llvm::raw_ostream&, llvm::RecordKeeper&) + 1172
9  llvm-tblgen              0x0000000106fcb15f llvm::TableGenMain(char*, bool (*)(llvm::raw_ostream&, llvm::RecordKeeper&)) + 3599
10 llvm-tblgen              0x0000000106f727a6 main + 134
11 libdyld.dylib            0x000000010733c6a5 start + 1
Stack dump:
0.      Program arguments: /Volumes/Builds/pz-master-stage2-RA/bin/llvm-tblgen -gen-register-bank -I /Users/vk/llvm/lib/Target/ARM -I /Users/vk/llvm/include -I /Users/vk/llvm/lib/Target /Users/vk/llvm/lib/Target/ARM/ARM.td -o /Volumes/Builds/pz-master-stage2-RA/lib/Target/ARM/ARMGenRegisterBank.inc.tmp
/bin/sh: line 1: 41986 Segmentation fault: 11  /Volumes/Builds/pz-master-stage2-RA/bin/llvm-tblgen -gen-register-bank -I /Users/vk/llvm/lib/Target/ARM -I /Users/vk/llvm/include -I /Users/vk/llvm/lib/Target /Users/vk/llvm/lib/Target/ARM/ARM.td -o /Volumes/Builds/pz
-master-stage2-RA/lib/Target/ARM/ARMGenRegisterBank.inc.tmp

thanks,
vedant

> On May 30, 2017, at 6:00 AM, Andrew V. Tischenko via llvm-commits <llvm-commits at lists.llvm.org> wrote:
> 
> Author: avt77
> Date: Tue May 30 08:00:44 2017
> New Revision: 304209
> 
> URL: http://llvm.org/viewvc/llvm-project?rev=304209&view=rev
> Log:
> This patch closes PR28513: an optimization of multiplication by different constants.
> It's implemented on DAG combiner level.
> 
> Modified:
>    llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
>    llvm/trunk/test/CodeGen/X86/mul-constant-i16.ll
>    llvm/trunk/test/CodeGen/X86/mul-constant-i32.ll
>    llvm/trunk/test/CodeGen/X86/mul-constant-i64.ll
> 
> Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=304209&r1=304208&r2=304209&view=diff
> ==============================================================================
> --- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
> +++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Tue May 30 08:00:44 2017
> @@ -80,6 +80,12 @@ static cl::opt<int> ExperimentalPrefLoop
>              " of the loop header PC will be 0)."),
>     cl::Hidden);
> 
> +static cl::opt<bool> MulConstantOptimization(
> +    "mul-constant-optimization", cl::init(true),
> +    cl::desc("Replace 'mul x, Const' with more effective instructions like "
> +             "SHIFT, LEA, etc."),
> +    cl::Hidden);
> +
> /// Call this when the user attempts to do something unsupported, like
> /// returning a double without SSE2 enabled on x86_64. This is not fatal, unlike
> /// report_fatal_error, so calling code should attempt to recover without
> @@ -30928,6 +30934,75 @@ static SDValue reduceVMULWidth(SDNode *N
>   }
> }
> 
> +static SDValue combineMulSpecial(uint64_t MulAmt, SDNode *N, SelectionDAG &DAG,
> +                                 EVT VT, SDLoc DL) {
> +
> +  auto combineMulShlAddOrSub = [&](int Mult, int Shift, bool isAdd) {
> +    SDValue Result = DAG.getNode(X86ISD::MUL_IMM, DL, VT, N->getOperand(0),
> +                                 DAG.getConstant(Mult, DL, VT));
> +    Result = DAG.getNode(ISD::SHL, DL, VT, Result,
> +                         DAG.getConstant(Shift, DL, MVT::i8));
> +    Result = DAG.getNode(isAdd ? ISD::ADD : ISD::SUB, DL, VT, N->getOperand(0),
> +                         Result);
> +    return Result;
> +  };
> +
> +  auto combineMulMulAddOrSub = [&](bool isAdd) {
> +    SDValue Result = DAG.getNode(X86ISD::MUL_IMM, DL, VT, N->getOperand(0),
> +                                 DAG.getConstant(9, DL, VT));
> +    Result = DAG.getNode(ISD::MUL, DL, VT, Result, DAG.getConstant(3, DL, VT));
> +    Result = DAG.getNode(isAdd ? ISD::ADD : ISD::SUB, DL, VT, N->getOperand(0),
> +                         Result);
> +    return Result;
> +  };
> +
> +  switch (MulAmt) {
> +  default:
> +    break;
> +  case 11:
> +    // mul x, 11 => add ((shl (mul x, 5), 1), x)
> +    return combineMulShlAddOrSub(5, 1, /*isAdd*/ true);
> +  case 21:
> +    // mul x, 21 => add ((shl (mul x, 5), 2), x)
> +    return combineMulShlAddOrSub(5, 2, /*isAdd*/ true);
> +  case 22:
> +    // mul x, 22 => add (add ((shl (mul x, 5), 2), x), x)
> +    return DAG.getNode(ISD::ADD, DL, VT, N->getOperand(0),
> +                       combineMulShlAddOrSub(5, 2, /*isAdd*/ true));
> +  case 19:
> +    // mul x, 19 => sub ((shl (mul x, 5), 2), x)
> +    return combineMulShlAddOrSub(5, 2, /*isAdd*/ false);
> +  case 13:
> +    // mul x, 13 => add ((shl (mul x, 3), 2), x)
> +    return combineMulShlAddOrSub(3, 2, /*isAdd*/ true);
> +  case 23:
> +    // mul x, 13 => sub ((shl (mul x, 3), 3), x)
> +    return combineMulShlAddOrSub(3, 3, /*isAdd*/ false);
> +  case 14:
> +    // mul x, 14 => add (add ((shl (mul x, 3), 2), x), x)
> +    return DAG.getNode(ISD::ADD, DL, VT, N->getOperand(0),
> +                       combineMulShlAddOrSub(3, 2, /*isAdd*/ true));
> +  case 26:
> +    // mul x, 26 => sub ((mul (mul x, 9), 3), x)
> +    return combineMulMulAddOrSub(/*isAdd*/ false);
> +  case 28:
> +    // mul x, 28 => add ((mul (mul x, 9), 3), x)
> +    return combineMulMulAddOrSub(/*isAdd*/ true);
> +  case 29:
> +    // mul x, 29 => add (add ((mul (mul x, 9), 3), x), x)
> +    return DAG.getNode(ISD::ADD, DL, VT, N->getOperand(0),
> +                       combineMulMulAddOrSub(/*isAdd*/ true));
> +  case 30:
> +    // mul x, 30 => sub (sub ((shl x, 5), x), x)
> +    return DAG.getNode(
> +        ISD::SUB, DL, VT, N->getOperand(0),
> +        DAG.getNode(ISD::SUB, DL, VT, N->getOperand(0),
> +                    DAG.getNode(ISD::SHL, DL, VT, N->getOperand(0),
> +                                DAG.getConstant(5, DL, MVT::i8))));
> +  }
> +  return SDValue();
> +}
> +
> /// Optimize a single multiply with constant into two operations in order to
> /// implement it with two cheaper instructions, e.g. LEA + SHL, LEA + LEA.
> static SDValue combineMul(SDNode *N, SelectionDAG &DAG,
> @@ -30937,6 +31012,8 @@ static SDValue combineMul(SDNode *N, Sel
>   if (DCI.isBeforeLegalize() && VT.isVector())
>     return reduceVMULWidth(N, DAG, Subtarget);
> 
> +  if (!MulConstantOptimization)
> +    return SDValue();
>   // An imul is usually smaller than the alternative sequence.
>   if (DAG.getMachineFunction().getFunction()->optForMinSize())
>     return SDValue();
> @@ -30992,7 +31069,8 @@ static SDValue combineMul(SDNode *N, Sel
>     else
>       NewMul = DAG.getNode(X86ISD::MUL_IMM, DL, VT, NewMul,
>                            DAG.getConstant(MulAmt2, DL, VT));
> -  }
> +  } else if (!Subtarget.slowLEA())
> +    NewMul = combineMulSpecial(MulAmt, N, DAG, VT, DL);
> 
>   if (!NewMul) {
>     assert(MulAmt != 0 &&
> 
> Modified: llvm/trunk/test/CodeGen/X86/mul-constant-i16.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/mul-constant-i16.ll?rev=304209&r1=304208&r2=304209&view=diff
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/mul-constant-i16.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/mul-constant-i16.ll Tue May 30 08:00:44 2017
> @@ -188,13 +188,16 @@ define i16 @test_mul_by_11(i16 %x) {
> ; X86-LABEL: test_mul_by_11:
> ; X86:       # BB#0:
> ; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
> -; X86-NEXT:    imull $11, %eax, %eax
> +; X86-NEXT:    leal (%eax,%eax,4), %ecx
> +; X86-NEXT:    leal (%eax,%ecx,2), %eax
> ; X86-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X86-NEXT:    retl
> ;
> ; X64-LABEL: test_mul_by_11:
> ; X64:       # BB#0:
> -; X64-NEXT:    imull $11, %edi, %eax
> +; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-NEXT:    leal (%rdi,%rdi,4), %eax
> +; X64-NEXT:    leal (%rdi,%rax,2), %eax
> ; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X64-NEXT:    retq
>   %mul = mul nsw i16 %x, 11
> @@ -225,13 +228,16 @@ define i16 @test_mul_by_13(i16 %x) {
> ; X86-LABEL: test_mul_by_13:
> ; X86:       # BB#0:
> ; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
> -; X86-NEXT:    imull $13, %eax, %eax
> +; X86-NEXT:    leal (%eax,%eax,2), %ecx
> +; X86-NEXT:    leal (%eax,%ecx,4), %eax
> ; X86-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X86-NEXT:    retl
> ;
> ; X64-LABEL: test_mul_by_13:
> ; X64:       # BB#0:
> -; X64-NEXT:    imull $13, %edi, %eax
> +; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-NEXT:    leal (%rdi,%rdi,2), %eax
> +; X64-NEXT:    leal (%rdi,%rax,4), %eax
> ; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X64-NEXT:    retq
>   %mul = mul nsw i16 %x, 13
> @@ -241,14 +247,19 @@ define i16 @test_mul_by_13(i16 %x) {
> define i16 @test_mul_by_14(i16 %x) {
> ; X86-LABEL: test_mul_by_14:
> ; X86:       # BB#0:
> -; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
> -; X86-NEXT:    imull $14, %eax, %eax
> +; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
> +; X86-NEXT:    leal (%ecx,%ecx,2), %eax
> +; X86-NEXT:    leal (%ecx,%eax,4), %eax
> +; X86-NEXT:    addl %ecx, %eax
> ; X86-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X86-NEXT:    retl
> ;
> ; X64-LABEL: test_mul_by_14:
> ; X64:       # BB#0:
> -; X64-NEXT:    imull $14, %edi, %eax
> +; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-NEXT:    leal (%rdi,%rdi,2), %eax
> +; X64-NEXT:    leal (%rdi,%rax,4), %eax
> +; X64-NEXT:    addl %edi, %eax
> ; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X64-NEXT:    retq
>   %mul = mul nsw i16 %x, 14
> @@ -338,14 +349,19 @@ define i16 @test_mul_by_19(i16 %x) {
> ; X86-LABEL: test_mul_by_19:
> ; X86:       # BB#0:
> ; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
> -; X86-NEXT:    imull $19, %eax, %eax
> +; X86-NEXT:    leal (%eax,%eax,4), %ecx
> +; X86-NEXT:    shll $2, %ecx
> +; X86-NEXT:    subl %ecx, %eax
> ; X86-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X86-NEXT:    retl
> ;
> ; X64-LABEL: test_mul_by_19:
> ; X64:       # BB#0:
> -; X64-NEXT:    imull $19, %edi, %eax
> -; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> +; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-NEXT:    leal (%rdi,%rdi,4), %eax
> +; X64-NEXT:    shll $2, %eax
> +; X64-NEXT:    subl %eax, %edi
> +; X64-NEXT:    movl %edi, %eax
> ; X64-NEXT:    retq
>   %mul = mul nsw i16 %x, 19
>   ret i16 %mul
> @@ -375,13 +391,16 @@ define i16 @test_mul_by_21(i16 %x) {
> ; X86-LABEL: test_mul_by_21:
> ; X86:       # BB#0:
> ; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
> -; X86-NEXT:    imull $21, %eax, %eax
> +; X86-NEXT:    leal (%eax,%eax,4), %ecx
> +; X86-NEXT:    leal (%eax,%ecx,4), %eax
> ; X86-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X86-NEXT:    retl
> ;
> ; X64-LABEL: test_mul_by_21:
> ; X64:       # BB#0:
> -; X64-NEXT:    imull $21, %edi, %eax
> +; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-NEXT:    leal (%rdi,%rdi,4), %eax
> +; X64-NEXT:    leal (%rdi,%rax,4), %eax
> ; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X64-NEXT:    retq
>   %mul = mul nsw i16 %x, 21
> @@ -391,14 +410,19 @@ define i16 @test_mul_by_21(i16 %x) {
> define i16 @test_mul_by_22(i16 %x) {
> ; X86-LABEL: test_mul_by_22:
> ; X86:       # BB#0:
> -; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
> -; X86-NEXT:    imull $22, %eax, %eax
> +; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
> +; X86-NEXT:    leal (%ecx,%ecx,4), %eax
> +; X86-NEXT:    leal (%ecx,%eax,4), %eax
> +; X86-NEXT:    addl %ecx, %eax
> ; X86-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X86-NEXT:    retl
> ;
> ; X64-LABEL: test_mul_by_22:
> ; X64:       # BB#0:
> -; X64-NEXT:    imull $22, %edi, %eax
> +; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-NEXT:    leal (%rdi,%rdi,4), %eax
> +; X64-NEXT:    leal (%rdi,%rax,4), %eax
> +; X64-NEXT:    addl %edi, %eax
> ; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X64-NEXT:    retq
>   %mul = mul nsw i16 %x, 22
> @@ -409,14 +433,19 @@ define i16 @test_mul_by_23(i16 %x) {
> ; X86-LABEL: test_mul_by_23:
> ; X86:       # BB#0:
> ; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
> -; X86-NEXT:    imull $23, %eax, %eax
> +; X86-NEXT:    leal (%eax,%eax,2), %ecx
> +; X86-NEXT:    shll $3, %ecx
> +; X86-NEXT:    subl %ecx, %eax
> ; X86-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X86-NEXT:    retl
> ;
> ; X64-LABEL: test_mul_by_23:
> ; X64:       # BB#0:
> -; X64-NEXT:    imull $23, %edi, %eax
> -; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> +; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-NEXT:    leal (%rdi,%rdi,2), %eax
> +; X64-NEXT:    shll $3, %eax
> +; X64-NEXT:    subl %eax, %edi
> +; X64-NEXT:    movl %edi, %eax
> ; X64-NEXT:    retq
>   %mul = mul nsw i16 %x, 23
>   ret i16 %mul
> @@ -466,14 +495,19 @@ define i16 @test_mul_by_26(i16 %x) {
> ; X86-LABEL: test_mul_by_26:
> ; X86:       # BB#0:
> ; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
> -; X86-NEXT:    imull $26, %eax, %eax
> +; X86-NEXT:    leal (%eax,%eax,8), %ecx
> +; X86-NEXT:    leal (%ecx,%ecx,2), %ecx
> +; X86-NEXT:    subl %ecx, %eax
> ; X86-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X86-NEXT:    retl
> ;
> ; X64-LABEL: test_mul_by_26:
> ; X64:       # BB#0:
> -; X64-NEXT:    imull $26, %edi, %eax
> -; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> +; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-NEXT:    leal (%rdi,%rdi,8), %eax
> +; X64-NEXT:    leal (%rax,%rax,2), %eax
> +; X64-NEXT:    subl %eax, %edi
> +; X64-NEXT:    movl %edi, %eax
> ; X64-NEXT:    retq
>   %mul = mul nsw i16 %x, 26
>   ret i16 %mul
> @@ -502,14 +536,19 @@ define i16 @test_mul_by_27(i16 %x) {
> define i16 @test_mul_by_28(i16 %x) {
> ; X86-LABEL: test_mul_by_28:
> ; X86:       # BB#0:
> -; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
> -; X86-NEXT:    imull $28, %eax, %eax
> +; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
> +; X86-NEXT:    leal (%ecx,%ecx,8), %eax
> +; X86-NEXT:    leal (%eax,%eax,2), %eax
> +; X86-NEXT:    addl %ecx, %eax
> ; X86-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X86-NEXT:    retl
> ;
> ; X64-LABEL: test_mul_by_28:
> ; X64:       # BB#0:
> -; X64-NEXT:    imull $28, %edi, %eax
> +; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-NEXT:    leal (%rdi,%rdi,8), %eax
> +; X64-NEXT:    leal (%rax,%rax,2), %eax
> +; X64-NEXT:    addl %edi, %eax
> ; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X64-NEXT:    retq
>   %mul = mul nsw i16 %x, 28
> @@ -519,14 +558,21 @@ define i16 @test_mul_by_28(i16 %x) {
> define i16 @test_mul_by_29(i16 %x) {
> ; X86-LABEL: test_mul_by_29:
> ; X86:       # BB#0:
> -; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
> -; X86-NEXT:    imull $29, %eax, %eax
> +; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
> +; X86-NEXT:    leal (%ecx,%ecx,8), %eax
> +; X86-NEXT:    leal (%eax,%eax,2), %eax
> +; X86-NEXT:    addl %ecx, %eax
> +; X86-NEXT:    addl %ecx, %eax
> ; X86-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X86-NEXT:    retl
> ;
> ; X64-LABEL: test_mul_by_29:
> ; X64:       # BB#0:
> -; X64-NEXT:    imull $29, %edi, %eax
> +; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-NEXT:    leal (%rdi,%rdi,8), %eax
> +; X64-NEXT:    leal (%rax,%rax,2), %eax
> +; X64-NEXT:    addl %edi, %eax
> +; X64-NEXT:    addl %edi, %eax
> ; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X64-NEXT:    retq
>   %mul = mul nsw i16 %x, 29
> @@ -537,14 +583,22 @@ define i16 @test_mul_by_30(i16 %x) {
> ; X86-LABEL: test_mul_by_30:
> ; X86:       # BB#0:
> ; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
> -; X86-NEXT:    imull $30, %eax, %eax
> +; X86-NEXT:    movl %eax, %ecx
> +; X86-NEXT:    shll $5, %ecx
> +; X86-NEXT:    movl %eax, %edx
> +; X86-NEXT:    subl %ecx, %edx
> +; X86-NEXT:    subl %edx, %eax
> ; X86-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> ; X86-NEXT:    retl
> ;
> ; X64-LABEL: test_mul_by_30:
> ; X64:       # BB#0:
> -; X64-NEXT:    imull $30, %edi, %eax
> -; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> +; X64-NEXT:    movl %edi, %eax
> +; X64-NEXT:    shll $5, %eax
> +; X64-NEXT:    movl %edi, %ecx
> +; X64-NEXT:    subl %eax, %ecx
> +; X64-NEXT:    subl %ecx, %edi
> +; X64-NEXT:    movl %edi, %eax
> ; X64-NEXT:    retq
>   %mul = mul nsw i16 %x, 30
>   ret i16 %mul
> @@ -587,3 +641,30 @@ define i16 @test_mul_by_32(i16 %x) {
>   %mul = mul nsw i16 %x, 32
>   ret i16 %mul
> }
> +
> +; (x*9+42)*(x*5+2)
> +define i16 @test_mul_spec(i16 %x) nounwind {
> +; X86-LABEL: test_mul_spec:
> +; X86:       # BB#0:
> +; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    leal 42(%eax,%eax,8), %ecx
> +; X86-NEXT:    leal 2(%eax,%eax,4), %eax
> +; X86-NEXT:    imull %ecx, %eax
> +; X86-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> +; X86-NEXT:    retl
> +;
> +; X64-LABEL: test_mul_spec:
> +; X64:       # BB#0:
> +; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-NEXT:    leal 42(%rdi,%rdi,8), %ecx
> +; X64-NEXT:    leal 2(%rdi,%rdi,4), %eax
> +; X64-NEXT:    imull %ecx, %eax
> +; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
> +; X64-NEXT:    retq
> +  %mul = mul nsw i16 %x, 9
> +  %add = add nsw i16 %mul, 42
> +  %mul2 = mul nsw i16 %x, 5
> +  %add2 = add nsw i16 %mul2, 2
> +  %mul3 = mul nsw i16 %add, %add2
> +  ret i16 %mul3
> +}
> 
> Modified: llvm/trunk/test/CodeGen/X86/mul-constant-i32.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/mul-constant-i32.ll?rev=304209&r1=304208&r2=304209&view=diff
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/mul-constant-i32.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/mul-constant-i32.ll Tue May 30 08:00:44 2017
> @@ -1,6 +1,12 @@
> ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
> ; RUN: llc < %s -mtriple=i686-unknown | FileCheck %s --check-prefix=X86
> -; RUN: llc < %s -mtriple=x86_64-unknown | FileCheck %s --check-prefix=X64
> +; RUN: llc < %s -mtriple=x86_64-unknown -print-schedule=true -mcpu=haswell| FileCheck %s --check-prefix=X64-HSW
> +; RUN: llc < %s -mtriple=x86_64-unknown -print-schedule=true -mcpu=btver2| FileCheck %s --check-prefix=X64-JAG
> +; RUN: llc < %s -mtriple=i686-unknown -mul-constant-optimization=false | FileCheck %s --check-prefix=X86-NOOPT
> +; RUN: llc < %s -mtriple=x86_64-unknown -mul-constant-optimization=false -print-schedule=true -mcpu=haswell| FileCheck %s --check-prefix=HSW-NOOPT
> +; RUN: llc < %s -mtriple=x86_64-unknown -mul-constant-optimization=false -print-schedule=true -mcpu=btver2| FileCheck %s --check-prefix=JAG-NOOPT
> +; RUN: llc < %s -mtriple=x86_64-unknown -print-schedule=true -mcpu=slm| FileCheck %s --check-prefix=X64-SLM
> +; RUN: llc < %s -mtriple=x86_64-unknown -mul-constant-optimization=false -print-schedule=true -mcpu=slm| FileCheck %s --check-prefix=SLM-NOOPT
> 
> define i32 @test_mul_by_1(i32 %x) {
> ; X86-LABEL: test_mul_by_1:
> @@ -8,10 +14,40 @@ define i32 @test_mul_by_1(i32 %x) {
> ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_1:
> -; X64:       # BB#0:
> -; X64-NEXT:    movl %edi, %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_1:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    movl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_1:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    movl %edi, %eax # sched: [1:0.17]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_1:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_1:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    movl %edi, %eax # sched: [1:0.25]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_1:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    movl %edi, %eax # sched: [1:0.17]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_1:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    movl %edi, %eax # sched: [1:0.50]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_1:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    movl %edi, %eax # sched: [1:0.50]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 1
>   ret i32 %mul
> }
> @@ -23,11 +59,47 @@ define i32 @test_mul_by_2(i32 %x) {
> ; X86-NEXT:    addl %eax, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_2:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    leal (%rdi,%rdi), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_2:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_2:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_2:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    addl %eax, %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_2:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; HSW-NOOPT-NEXT:    leal (%rdi,%rdi), %eax # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_2:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; JAG-NOOPT-NEXT:    leal (%rdi,%rdi), %eax # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_2:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    leal (%rdi,%rdi), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_2:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; SLM-NOOPT-NEXT:    leal (%rdi,%rdi), %eax # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 2
>   ret i32 %mul
> }
> @@ -38,11 +110,46 @@ define i32 @test_mul_by_3(i32 %x) {
> ; X86-NEXT:    imull $3, {{[0-9]+}}(%esp), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_3:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    leal (%rdi,%rdi,2), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_3:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_3:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_3:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $3, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_3:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; HSW-NOOPT-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_3:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; JAG-NOOPT-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_3:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_3:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; SLM-NOOPT-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 3
>   ret i32 %mul
> }
> @@ -54,11 +161,47 @@ define i32 @test_mul_by_4(i32 %x) {
> ; X86-NEXT:    shll $2, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_4:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    leal (,%rdi,4), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_4:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (,%rdi,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_4:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (,%rdi,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_4:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    shll $2, %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_4:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; HSW-NOOPT-NEXT:    leal (,%rdi,4), %eax # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_4:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; JAG-NOOPT-NEXT:    leal (,%rdi,4), %eax # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_4:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    leal (,%rdi,4), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_4:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; SLM-NOOPT-NEXT:    leal (,%rdi,4), %eax # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 4
>   ret i32 %mul
> }
> @@ -69,11 +212,46 @@ define i32 @test_mul_by_5(i32 %x) {
> ; X86-NEXT:    imull $5, {{[0-9]+}}(%esp), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_5:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    leal (%rdi,%rdi,4), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_5:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_5:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_5:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $5, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_5:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; HSW-NOOPT-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_5:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; JAG-NOOPT-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_5:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_5:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; SLM-NOOPT-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 5
>   ret i32 %mul
> }
> @@ -86,12 +264,46 @@ define i32 @test_mul_by_6(i32 %x) {
> ; X86-NEXT:    leal (%eax,%eax,2), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_6:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    addl %edi, %edi
> -; X64-NEXT:    leal (%rdi,%rdi,2), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_6:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    addl %edi, %edi # sched: [1:0.25]
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_6:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    addl %edi, %edi # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_6:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $6, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_6:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $6, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_6:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $6, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_6:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    addl %edi, %edi # sched: [1:0.50]
> +; X64-SLM-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_6:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $6, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 6
>   ret i32 %mul
> }
> @@ -104,12 +316,46 @@ define i32 @test_mul_by_7(i32 %x) {
> ; X86-NEXT:    subl %ecx, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_7:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    leal (,%rdi,8), %eax
> -; X64-NEXT:    subl %edi, %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_7:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (,%rdi,8), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    subl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_7:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (,%rdi,8), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    subl %edi, %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_7:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $7, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_7:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $7, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_7:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $7, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_7:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    leal (,%rdi,8), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    subl %edi, %eax # sched: [1:0.50]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_7:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $7, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 7
>   ret i32 %mul
> }
> @@ -121,11 +367,47 @@ define i32 @test_mul_by_8(i32 %x) {
> ; X86-NEXT:    shll $3, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_8:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    leal (,%rdi,8), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_8:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (,%rdi,8), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_8:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (,%rdi,8), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_8:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    shll $3, %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_8:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; HSW-NOOPT-NEXT:    leal (,%rdi,8), %eax # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_8:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; JAG-NOOPT-NEXT:    leal (,%rdi,8), %eax # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_8:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    leal (,%rdi,8), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_8:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; SLM-NOOPT-NEXT:    leal (,%rdi,8), %eax # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 8
>   ret i32 %mul
> }
> @@ -136,11 +418,46 @@ define i32 @test_mul_by_9(i32 %x) {
> ; X86-NEXT:    imull $9, {{[0-9]+}}(%esp), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_9:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    leal (%rdi,%rdi,8), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_9:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_9:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_9:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $9, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_9:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; HSW-NOOPT-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_9:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; JAG-NOOPT-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_9:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_9:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; SLM-NOOPT-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 9
>   ret i32 %mul
> }
> @@ -153,12 +470,46 @@ define i32 @test_mul_by_10(i32 %x) {
> ; X86-NEXT:    leal (%eax,%eax,4), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_10:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    addl %edi, %edi
> -; X64-NEXT:    leal (%rdi,%rdi,4), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_10:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    addl %edi, %edi # sched: [1:0.25]
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_10:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    addl %edi, %edi # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_10:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $10, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_10:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $10, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_10:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $10, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_10:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    addl %edi, %edi # sched: [1:0.50]
> +; X64-SLM-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_10:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $10, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 10
>   ret i32 %mul
> }
> @@ -166,13 +517,49 @@ define i32 @test_mul_by_10(i32 %x) {
> define i32 @test_mul_by_11(i32 %x) {
> ; X86-LABEL: test_mul_by_11:
> ; X86:       # BB#0:
> -; X86-NEXT:    imull $11, {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    leal (%eax,%eax,4), %ecx
> +; X86-NEXT:    leal (%eax,%ecx,2), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_11:
> -; X64:       # BB#0:
> -; X64-NEXT:    imull $11, %edi, %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_11:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leal (%rdi,%rax,2), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_11:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rdi,%rax,2), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_11:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $11, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_11:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $11, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_11:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $11, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_11:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imull $11, %edi, %eax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_11:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $11, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 11
>   ret i32 %mul
> }
> @@ -185,12 +572,46 @@ define i32 @test_mul_by_12(i32 %x) {
> ; X86-NEXT:    leal (%eax,%eax,2), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_12:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    shll $2, %edi
> -; X64-NEXT:    leal (%rdi,%rdi,2), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_12:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    shll $2, %edi # sched: [1:0.50]
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_12:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    shll $2, %edi # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_12:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $12, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_12:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $12, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_12:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $12, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_12:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    shll $2, %edi # sched: [1:1.00]
> +; X64-SLM-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_12:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $12, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 12
>   ret i32 %mul
> }
> @@ -198,13 +619,49 @@ define i32 @test_mul_by_12(i32 %x) {
> define i32 @test_mul_by_13(i32 %x) {
> ; X86-LABEL: test_mul_by_13:
> ; X86:       # BB#0:
> -; X86-NEXT:    imull $13, {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    leal (%eax,%eax,2), %ecx
> +; X86-NEXT:    leal (%eax,%ecx,4), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_13:
> -; X64:       # BB#0:
> -; X64-NEXT:    imull $13, %edi, %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_13:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leal (%rdi,%rax,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_13:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rdi,%rax,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_13:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $13, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_13:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $13, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_13:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $13, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_13:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imull $13, %edi, %eax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_13:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $13, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 13
>   ret i32 %mul
> }
> @@ -212,13 +669,52 @@ define i32 @test_mul_by_13(i32 %x) {
> define i32 @test_mul_by_14(i32 %x) {
> ; X86-LABEL: test_mul_by_14:
> ; X86:       # BB#0:
> -; X86-NEXT:    imull $14, {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
> +; X86-NEXT:    leal (%ecx,%ecx,2), %eax
> +; X86-NEXT:    leal (%ecx,%eax,4), %eax
> +; X86-NEXT:    addl %ecx, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_14:
> -; X64:       # BB#0:
> -; X64-NEXT:    imull $14, %edi, %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_14:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leal (%rdi,%rax,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    addl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_14:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rdi,%rax,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    addl %edi, %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_14:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $14, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_14:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $14, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_14:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $14, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_14:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imull $14, %edi, %eax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_14:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $14, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 14
>   ret i32 %mul
> }
> @@ -231,12 +727,46 @@ define i32 @test_mul_by_15(i32 %x) {
> ; X86-NEXT:    leal (%eax,%eax,2), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_15:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    leal (%rdi,%rdi,4), %eax
> -; X64-NEXT:    leal (%rax,%rax,2), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_15:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leal (%rax,%rax,2), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_15:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rax,%rax,2), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_15:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $15, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_15:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $15, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_15:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $15, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_15:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    leal (%rax,%rax,2), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_15:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $15, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 15
>   ret i32 %mul
> }
> @@ -248,11 +778,47 @@ define i32 @test_mul_by_16(i32 %x) {
> ; X86-NEXT:    shll $4, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_16:
> -; X64:       # BB#0:
> -; X64-NEXT:    shll $4, %edi
> -; X64-NEXT:    movl %edi, %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_16:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    shll $4, %edi # sched: [1:0.50]
> +; X64-HSW-NEXT:    movl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_16:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    shll $4, %edi # sched: [1:0.50]
> +; X64-JAG-NEXT:    movl %edi, %eax # sched: [1:0.17]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_16:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    shll $4, %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_16:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    shll $4, %edi # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    movl %edi, %eax # sched: [1:0.25]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_16:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    shll $4, %edi # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    movl %edi, %eax # sched: [1:0.17]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_16:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    shll $4, %edi # sched: [1:1.00]
> +; X64-SLM-NEXT:    movl %edi, %eax # sched: [1:0.50]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_16:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    shll $4, %edi # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    movl %edi, %eax # sched: [1:0.50]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 16
>   ret i32 %mul
> }
> @@ -266,13 +832,49 @@ define i32 @test_mul_by_17(i32 %x) {
> ; X86-NEXT:    addl %ecx, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_17:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    movl %edi, %eax
> -; X64-NEXT:    shll $4, %eax
> -; X64-NEXT:    leal (%rax,%rdi), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_17:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    movl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    shll $4, %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leal (%rax,%rdi), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_17:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    movl %edi, %eax # sched: [1:0.17]
> +; X64-JAG-NEXT:    shll $4, %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rax,%rdi), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_17:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $17, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_17:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $17, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_17:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $17, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_17:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    movl %edi, %eax # sched: [1:0.50]
> +; X64-SLM-NEXT:    shll $4, %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    leal (%rax,%rdi), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_17:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $17, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 17
>   ret i32 %mul
> }
> @@ -285,12 +887,46 @@ define i32 @test_mul_by_18(i32 %x) {
> ; X86-NEXT:    leal (%eax,%eax,8), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_18:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    addl %edi, %edi
> -; X64-NEXT:    leal (%rdi,%rdi,8), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_18:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    addl %edi, %edi # sched: [1:0.25]
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_18:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    addl %edi, %edi # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_18:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $18, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_18:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $18, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_18:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $18, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_18:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    addl %edi, %edi # sched: [1:0.50]
> +; X64-SLM-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_18:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $18, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 18
>   ret i32 %mul
> }
> @@ -298,13 +934,54 @@ define i32 @test_mul_by_18(i32 %x) {
> define i32 @test_mul_by_19(i32 %x) {
> ; X86-LABEL: test_mul_by_19:
> ; X86:       # BB#0:
> -; X86-NEXT:    imull $19, {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    leal (%eax,%eax,4), %ecx
> +; X86-NEXT:    shll $2, %ecx
> +; X86-NEXT:    subl %ecx, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_19:
> -; X64:       # BB#0:
> -; X64-NEXT:    imull $19, %edi, %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_19:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    shll $2, %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    subl %eax, %edi # sched: [1:0.25]
> +; X64-HSW-NEXT:    movl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_19:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    shll $2, %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    subl %eax, %edi # sched: [1:0.50]
> +; X64-JAG-NEXT:    movl %edi, %eax # sched: [1:0.17]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_19:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $19, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_19:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $19, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_19:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $19, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_19:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imull $19, %edi, %eax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_19:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $19, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 19
>   ret i32 %mul
> }
> @@ -317,12 +994,46 @@ define i32 @test_mul_by_20(i32 %x) {
> ; X86-NEXT:    leal (%eax,%eax,4), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_20:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    shll $2, %edi
> -; X64-NEXT:    leal (%rdi,%rdi,4), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_20:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    shll $2, %edi # sched: [1:0.50]
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_20:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    shll $2, %edi # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_20:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $20, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_20:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $20, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_20:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $20, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_20:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    shll $2, %edi # sched: [1:1.00]
> +; X64-SLM-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_20:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $20, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 20
>   ret i32 %mul
> }
> @@ -330,13 +1041,49 @@ define i32 @test_mul_by_20(i32 %x) {
> define i32 @test_mul_by_21(i32 %x) {
> ; X86-LABEL: test_mul_by_21:
> ; X86:       # BB#0:
> -; X86-NEXT:    imull $21, {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    leal (%eax,%eax,4), %ecx
> +; X86-NEXT:    leal (%eax,%ecx,4), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_21:
> -; X64:       # BB#0:
> -; X64-NEXT:    imull $21, %edi, %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_21:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leal (%rdi,%rax,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_21:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rdi,%rax,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_21:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $21, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_21:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $21, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_21:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $21, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_21:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imull $21, %edi, %eax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_21:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $21, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 21
>   ret i32 %mul
> }
> @@ -344,13 +1091,52 @@ define i32 @test_mul_by_21(i32 %x) {
> define i32 @test_mul_by_22(i32 %x) {
> ; X86-LABEL: test_mul_by_22:
> ; X86:       # BB#0:
> -; X86-NEXT:    imull $22, {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
> +; X86-NEXT:    leal (%ecx,%ecx,4), %eax
> +; X86-NEXT:    leal (%ecx,%eax,4), %eax
> +; X86-NEXT:    addl %ecx, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_22:
> -; X64:       # BB#0:
> -; X64-NEXT:    imull $22, %edi, %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_22:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leal (%rdi,%rax,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    addl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_22:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rdi,%rax,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    addl %edi, %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_22:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $22, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_22:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $22, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_22:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $22, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_22:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imull $22, %edi, %eax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_22:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $22, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 22
>   ret i32 %mul
> }
> @@ -358,13 +1144,54 @@ define i32 @test_mul_by_22(i32 %x) {
> define i32 @test_mul_by_23(i32 %x) {
> ; X86-LABEL: test_mul_by_23:
> ; X86:       # BB#0:
> -; X86-NEXT:    imull $23, {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    leal (%eax,%eax,2), %ecx
> +; X86-NEXT:    shll $3, %ecx
> +; X86-NEXT:    subl %ecx, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_23:
> -; X64:       # BB#0:
> -; X64-NEXT:    imull $23, %edi, %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_23:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    shll $3, %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    subl %eax, %edi # sched: [1:0.25]
> +; X64-HSW-NEXT:    movl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_23:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    shll $3, %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    subl %eax, %edi # sched: [1:0.50]
> +; X64-JAG-NEXT:    movl %edi, %eax # sched: [1:0.17]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_23:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $23, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_23:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $23, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_23:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $23, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_23:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imull $23, %edi, %eax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_23:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $23, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 23
>   ret i32 %mul
> }
> @@ -377,12 +1204,46 @@ define i32 @test_mul_by_24(i32 %x) {
> ; X86-NEXT:    leal (%eax,%eax,2), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_24:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    shll $3, %edi
> -; X64-NEXT:    leal (%rdi,%rdi,2), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_24:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    shll $3, %edi # sched: [1:0.50]
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_24:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    shll $3, %edi # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_24:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $24, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_24:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $24, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_24:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $24, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_24:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    shll $3, %edi # sched: [1:1.00]
> +; X64-SLM-NEXT:    leal (%rdi,%rdi,2), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_24:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $24, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 24
>   ret i32 %mul
> }
> @@ -395,12 +1256,46 @@ define i32 @test_mul_by_25(i32 %x) {
> ; X86-NEXT:    leal (%eax,%eax,4), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_25:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    leal (%rdi,%rdi,4), %eax
> -; X64-NEXT:    leal (%rax,%rax,4), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_25:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leal (%rax,%rax,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_25:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rax,%rax,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_25:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $25, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_25:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $25, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_25:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $25, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_25:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    leal (%rax,%rax,4), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_25:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $25, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 25
>   ret i32 %mul
> }
> @@ -408,13 +1303,54 @@ define i32 @test_mul_by_25(i32 %x) {
> define i32 @test_mul_by_26(i32 %x) {
> ; X86-LABEL: test_mul_by_26:
> ; X86:       # BB#0:
> -; X86-NEXT:    imull $26, {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    leal (%eax,%eax,8), %ecx
> +; X86-NEXT:    leal (%ecx,%ecx,2), %ecx
> +; X86-NEXT:    subl %ecx, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_26:
> -; X64:       # BB#0:
> -; X64-NEXT:    imull $26, %edi, %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_26:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leal (%rax,%rax,2), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    subl %eax, %edi # sched: [1:0.25]
> +; X64-HSW-NEXT:    movl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_26:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rax,%rax,2), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    subl %eax, %edi # sched: [1:0.50]
> +; X64-JAG-NEXT:    movl %edi, %eax # sched: [1:0.17]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_26:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $26, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_26:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $26, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_26:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $26, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_26:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imull $26, %edi, %eax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_26:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $26, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 26
>   ret i32 %mul
> }
> @@ -427,12 +1363,46 @@ define i32 @test_mul_by_27(i32 %x) {
> ; X86-NEXT:    leal (%eax,%eax,2), %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_27:
> -; X64:       # BB#0:
> -; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> -; X64-NEXT:    leal (%rdi,%rdi,8), %eax
> -; X64-NEXT:    leal (%rax,%rax,2), %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_27:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leal (%rax,%rax,2), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_27:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rax,%rax,2), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_27:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $27, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_27:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $27, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_27:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $27, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_27:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    leal (%rax,%rax,2), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_27:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $27, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 27
>   ret i32 %mul
> }
> @@ -440,13 +1410,52 @@ define i32 @test_mul_by_27(i32 %x) {
> define i32 @test_mul_by_28(i32 %x) {
> ; X86-LABEL: test_mul_by_28:
> ; X86:       # BB#0:
> -; X86-NEXT:    imull $28, {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
> +; X86-NEXT:    leal (%ecx,%ecx,8), %eax
> +; X86-NEXT:    leal (%eax,%eax,2), %eax
> +; X86-NEXT:    addl %ecx, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_28:
> -; X64:       # BB#0:
> -; X64-NEXT:    imull $28, %edi, %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_28:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leal (%rax,%rax,2), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    addl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_28:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rax,%rax,2), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    addl %edi, %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_28:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $28, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_28:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $28, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_28:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $28, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_28:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imull $28, %edi, %eax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_28:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $28, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 28
>   ret i32 %mul
> }
> @@ -454,13 +1463,55 @@ define i32 @test_mul_by_28(i32 %x) {
> define i32 @test_mul_by_29(i32 %x) {
> ; X86-LABEL: test_mul_by_29:
> ; X86:       # BB#0:
> -; X86-NEXT:    imull $29, {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
> +; X86-NEXT:    leal (%ecx,%ecx,8), %eax
> +; X86-NEXT:    leal (%eax,%eax,2), %eax
> +; X86-NEXT:    addl %ecx, %eax
> +; X86-NEXT:    addl %ecx, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_29:
> -; X64:       # BB#0:
> -; X64-NEXT:    imull $29, %edi, %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_29:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leal (%rax,%rax,2), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    addl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    addl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_29:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal (%rdi,%rdi,8), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal (%rax,%rax,2), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    addl %edi, %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    addl %edi, %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_29:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $29, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_29:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $29, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_29:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $29, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_29:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imull $29, %edi, %eax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_29:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $29, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 29
>   ret i32 %mul
> }
> @@ -468,13 +1519,58 @@ define i32 @test_mul_by_29(i32 %x) {
> define i32 @test_mul_by_30(i32 %x) {
> ; X86-LABEL: test_mul_by_30:
> ; X86:       # BB#0:
> -; X86-NEXT:    imull $30, {{[0-9]+}}(%esp), %eax
> -; X86-NEXT:    retl
> -;
> -; X64-LABEL: test_mul_by_30:
> -; X64:       # BB#0:
> -; X64-NEXT:    imull $30, %edi, %eax
> -; X64-NEXT:    retq
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    movl %eax, %ecx
> +; X86-NEXT:    shll $5, %ecx
> +; X86-NEXT:    movl %eax, %edx
> +; X86-NEXT:    subl %ecx, %edx
> +; X86-NEXT:    subl %edx, %eax
> +; X86-NEXT:    retl
> +;
> +; X64-HSW-LABEL: test_mul_by_30:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    movl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    shll $5, %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    movl %edi, %ecx # sched: [1:0.25]
> +; X64-HSW-NEXT:    subl %eax, %ecx # sched: [1:0.25]
> +; X64-HSW-NEXT:    subl %ecx, %edi # sched: [1:0.25]
> +; X64-HSW-NEXT:    movl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_30:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    movl %edi, %eax # sched: [1:0.17]
> +; X64-JAG-NEXT:    movl %edi, %ecx # sched: [1:0.17]
> +; X64-JAG-NEXT:    shll $5, %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    subl %eax, %ecx # sched: [1:0.50]
> +; X64-JAG-NEXT:    subl %ecx, %edi # sched: [1:0.50]
> +; X64-JAG-NEXT:    movl %edi, %eax # sched: [1:0.17]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_30:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $30, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_30:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $30, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_30:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $30, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_30:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imull $30, %edi, %eax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_30:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $30, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 30
>   ret i32 %mul
> }
> @@ -488,12 +1584,46 @@ define i32 @test_mul_by_31(i32 %x) {
> ; X86-NEXT:    subl %ecx, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_31:
> -; X64:       # BB#0:
> -; X64-NEXT:    movl %edi, %eax
> -; X64-NEXT:    shll $5, %eax
> -; X64-NEXT:    subl %edi, %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_31:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    movl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    shll $5, %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    subl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_31:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    movl %edi, %eax # sched: [1:0.17]
> +; X64-JAG-NEXT:    shll $5, %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    subl %edi, %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_31:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    imull $31, {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_31:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imull $31, %edi, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_31:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imull $31, %edi, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_31:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    movl %edi, %eax # sched: [1:0.50]
> +; X64-SLM-NEXT:    shll $5, %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    subl %edi, %eax # sched: [1:0.50]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_31:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imull $31, %edi, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 31
>   ret i32 %mul
> }
> @@ -505,11 +1635,124 @@ define i32 @test_mul_by_32(i32 %x) {
> ; X86-NEXT:    shll $5, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_32:
> -; X64:       # BB#0:
> -; X64-NEXT:    shll $5, %edi
> -; X64-NEXT:    movl %edi, %eax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_32:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    shll $5, %edi # sched: [1:0.50]
> +; X64-HSW-NEXT:    movl %edi, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_32:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    shll $5, %edi # sched: [1:0.50]
> +; X64-JAG-NEXT:    movl %edi, %eax # sched: [1:0.17]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_32:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    shll $5, %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_32:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    shll $5, %edi # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    movl %edi, %eax # sched: [1:0.25]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_32:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    shll $5, %edi # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    movl %edi, %eax # sched: [1:0.17]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_32:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    shll $5, %edi # sched: [1:1.00]
> +; X64-SLM-NEXT:    movl %edi, %eax # sched: [1:0.50]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_32:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    shll $5, %edi # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    movl %edi, %eax # sched: [1:0.50]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i32 %x, 32
>   ret i32 %mul
> }
> +
> +; (x*9+42)*(x*5+2)
> +define i32 @test_mul_spec(i32 %x) nounwind {
> +; X86-LABEL: test_mul_spec:
> +; X86:       # BB#0:
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    leal 42(%eax,%eax,8), %ecx
> +; X86-NEXT:    leal 2(%eax,%eax,4), %eax
> +; X86-NEXT:    imull %ecx, %eax
> +; X86-NEXT:    retl
> +;
> +; X64-HSW-LABEL: test_mul_spec:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,8), %ecx # sched: [1:0.50]
> +; X64-HSW-NEXT:    addl $42, %ecx # sched: [1:0.25]
> +; X64-HSW-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-HSW-NEXT:    addl $2, %eax # sched: [1:0.25]
> +; X64-HSW-NEXT:    imull %ecx, %eax # sched: [4:1.00]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_spec:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-JAG-NEXT:    leal 42(%rdi,%rdi,8), %ecx # sched: [1:0.50]
> +; X64-JAG-NEXT:    leal 2(%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; X64-JAG-NEXT:    imull %ecx, %eax # sched: [3:1.00]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_spec:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    leal 42(%eax,%eax,8), %ecx
> +; X86-NOOPT-NEXT:    leal 2(%eax,%eax,4), %eax
> +; X86-NOOPT-NEXT:    imull %ecx, %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_spec:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; HSW-NOOPT-NEXT:    leal (%rdi,%rdi,8), %ecx # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    addl $42, %ecx # sched: [1:0.25]
> +; HSW-NOOPT-NEXT:    leal (%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    addl $2, %eax # sched: [1:0.25]
> +; HSW-NOOPT-NEXT:    imull %ecx, %eax # sched: [4:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_spec:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; JAG-NOOPT-NEXT:    leal 42(%rdi,%rdi,8), %ecx # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    leal 2(%rdi,%rdi,4), %eax # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    imull %ecx, %eax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_spec:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; X64-SLM-NEXT:    leal 42(%rdi,%rdi,8), %ecx # sched: [1:1.00]
> +; X64-SLM-NEXT:    leal 2(%rdi,%rdi,4), %eax # sched: [1:1.00]
> +; X64-SLM-NEXT:    imull %ecx, %eax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_spec:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
> +; SLM-NOOPT-NEXT:    leal 42(%rdi,%rdi,8), %ecx # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    leal 2(%rdi,%rdi,4), %eax # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    imull %ecx, %eax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
> +  %mul = mul nsw i32 %x, 9
> +  %add = add nsw i32 %mul, 42
> +  %mul2 = mul nsw i32 %x, 5
> +  %add2 = add nsw i32 %mul2, 2
> +  %mul3 = mul nsw i32 %add, %add2
> +  ret i32 %mul3
> +}
> 
> Modified: llvm/trunk/test/CodeGen/X86/mul-constant-i64.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/mul-constant-i64.ll?rev=304209&r1=304208&r2=304209&view=diff
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/mul-constant-i64.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/mul-constant-i64.ll Tue May 30 08:00:44 2017
> @@ -1,18 +1,55 @@
> ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
> ; RUN: llc < %s -mtriple=i686-unknown | FileCheck %s --check-prefix=X86
> -; RUN: llc < %s -mtriple=x86_64-unknown | FileCheck %s --check-prefix=X64
> +; RUN: llc < %s -mtriple=x86_64-unknown -print-schedule=true -mcpu=haswell| FileCheck %s --check-prefix=X64-HSW
> +; RUN: llc < %s -mtriple=x86_64-unknown -print-schedule=true -mcpu=btver2| FileCheck %s --check-prefix=X64-JAG
> +; RUN: llc < %s -mtriple=i686-unknown -mul-constant-optimization=false | FileCheck %s --check-prefix=X86-NOOPT
> +; RUN: llc < %s -mtriple=x86_64-unknown -mul-constant-optimization=false -print-schedule=true -mcpu=haswell| FileCheck %s --check-prefix=HSW-NOOPT
> +; RUN: llc < %s -mtriple=x86_64-unknown -mul-constant-optimization=false -print-schedule=true -mcpu=btver2| FileCheck %s --check-prefix=JAG-NOOPT
> +; RUN: llc < %s -mtriple=x86_64-unknown -print-schedule=true -mcpu=slm| FileCheck %s --check-prefix=X64-SLM
> +; RUN: llc < %s -mtriple=x86_64-unknown -mul-constant-optimization=false -print-schedule=true -mcpu=slm| FileCheck %s --check-prefix=SLM-NOOPT
> 
> -define i64 @test_mul_by_1(i64 %x) {
> +define i64 @test_mul_by_1(i64 %x) nounwind {
> ; X86-LABEL: test_mul_by_1:
> ; X86:       # BB#0:
> ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> ; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_1:
> -; X64:       # BB#0:
> -; X64-NEXT:    movq %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_1:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    movq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_1:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    movq %rdi, %rax # sched: [1:0.17]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_1:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_1:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    movq %rdi, %rax # sched: [1:0.25]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_1:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    movq %rdi, %rax # sched: [1:0.17]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_1:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    movq %rdi, %rax # sched: [1:0.50]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_1:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    movq %rdi, %rax # sched: [1:0.50]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 1
>   ret i64 %mul
> }
> @@ -26,10 +63,43 @@ define i64 @test_mul_by_2(i64 %x) {
> ; X86-NEXT:    addl %eax, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_2:
> -; X64:       # BB#0:
> -; X64-NEXT:    leaq (%rdi,%rdi), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_2:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_2:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_2:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %edx
> +; X86-NOOPT-NEXT:    shldl $1, %eax, %edx
> +; X86-NOOPT-NEXT:    addl %eax, %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_2:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    leaq (%rdi,%rdi), %rax # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_2:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    leaq (%rdi,%rdi), %rax # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_2:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    leaq (%rdi,%rdi), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_2:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    leaq (%rdi,%rdi), %rax # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 2
>   ret i64 %mul
> }
> @@ -43,10 +113,43 @@ define i64 @test_mul_by_3(i64 %x) {
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_3:
> -; X64:       # BB#0:
> -; X64-NEXT:    leaq (%rdi,%rdi,2), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_3:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_3:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_3:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $3, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $3, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_3:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_3:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_3:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_3:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 3
>   ret i64 %mul
> }
> @@ -60,10 +163,43 @@ define i64 @test_mul_by_4(i64 %x) {
> ; X86-NEXT:    shll $2, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_4:
> -; X64:       # BB#0:
> -; X64-NEXT:    leaq (,%rdi,4), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_4:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (,%rdi,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_4:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (,%rdi,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_4:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %edx
> +; X86-NOOPT-NEXT:    shldl $2, %eax, %edx
> +; X86-NOOPT-NEXT:    shll $2, %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_4:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    leaq (,%rdi,4), %rax # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_4:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    leaq (,%rdi,4), %rax # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_4:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    leaq (,%rdi,4), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_4:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    leaq (,%rdi,4), %rax # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 4
>   ret i64 %mul
> }
> @@ -77,10 +213,43 @@ define i64 @test_mul_by_5(i64 %x) {
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_5:
> -; X64:       # BB#0:
> -; X64-NEXT:    leaq (%rdi,%rdi,4), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_5:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_5:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_5:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $5, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $5, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_5:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_5:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_5:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_5:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 5
>   ret i64 %mul
> }
> @@ -95,11 +264,46 @@ define i64 @test_mul_by_6(i64 %x) {
> ; X86-NEXT:    leal (%edx,%ecx,2), %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_6:
> -; X64:       # BB#0:
> -; X64-NEXT:    addq %rdi, %rdi
> -; X64-NEXT:    leaq (%rdi,%rdi,2), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_6:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    addq %rdi, %rdi # sched: [1:0.25]
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_6:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    addq %rdi, %rdi # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_6:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $6, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $6, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_6:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $6, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_6:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $6, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_6:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    addq %rdi, %rdi # sched: [1:0.50]
> +; X64-SLM-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_6:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $6, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 6
>   ret i64 %mul
> }
> @@ -115,11 +319,46 @@ define i64 @test_mul_by_7(i64 %x) {
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_7:
> -; X64:       # BB#0:
> -; X64-NEXT:    leaq (,%rdi,8), %rax
> -; X64-NEXT:    subq %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_7:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (,%rdi,8), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    subq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_7:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (,%rdi,8), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    subq %rdi, %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_7:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $7, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $7, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_7:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $7, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_7:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $7, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_7:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    leaq (,%rdi,8), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    subq %rdi, %rax # sched: [1:0.50]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_7:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $7, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 7
>   ret i64 %mul
> }
> @@ -133,10 +372,43 @@ define i64 @test_mul_by_8(i64 %x) {
> ; X86-NEXT:    shll $3, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_8:
> -; X64:       # BB#0:
> -; X64-NEXT:    leaq (,%rdi,8), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_8:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (,%rdi,8), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_8:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (,%rdi,8), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_8:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %edx
> +; X86-NOOPT-NEXT:    shldl $3, %eax, %edx
> +; X86-NOOPT-NEXT:    shll $3, %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_8:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    leaq (,%rdi,8), %rax # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_8:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    leaq (,%rdi,8), %rax # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_8:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    leaq (,%rdi,8), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_8:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    leaq (,%rdi,8), %rax # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 8
>   ret i64 %mul
> }
> @@ -150,10 +422,43 @@ define i64 @test_mul_by_9(i64 %x) {
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_9:
> -; X64:       # BB#0:
> -; X64-NEXT:    leaq (%rdi,%rdi,8), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_9:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_9:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_9:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $9, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $9, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_9:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_9:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_9:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_9:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 9
>   ret i64 %mul
> }
> @@ -168,11 +473,46 @@ define i64 @test_mul_by_10(i64 %x) {
> ; X86-NEXT:    leal (%edx,%ecx,2), %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_10:
> -; X64:       # BB#0:
> -; X64-NEXT:    addq %rdi, %rdi
> -; X64-NEXT:    leaq (%rdi,%rdi,4), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_10:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    addq %rdi, %rdi # sched: [1:0.25]
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_10:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    addq %rdi, %rdi # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_10:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $10, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $10, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_10:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $10, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_10:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $10, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_10:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    addq %rdi, %rdi # sched: [1:0.50]
> +; X64-SLM-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_10:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $10, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 10
>   ret i64 %mul
> }
> @@ -180,16 +520,53 @@ define i64 @test_mul_by_10(i64 %x) {
> define i64 @test_mul_by_11(i64 %x) {
> ; X86-LABEL: test_mul_by_11:
> ; X86:       # BB#0:
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    leal (%eax,%eax,4), %ecx
> +; X86-NEXT:    leal (%eax,%ecx,2), %ecx
> ; X86-NEXT:    movl $11, %eax
> ; X86-NEXT:    mull {{[0-9]+}}(%esp)
> -; X86-NEXT:    imull $11, {{[0-9]+}}(%esp), %ecx
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_11:
> -; X64:       # BB#0:
> -; X64-NEXT:    imulq $11, %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_11:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leaq (%rdi,%rax,2), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_11:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rdi,%rax,2), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_11:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $11, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $11, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_11:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $11, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_11:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $11, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_11:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imulq $11, %rdi, %rax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_11:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $11, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 11
>   ret i64 %mul
> }
> @@ -204,11 +581,46 @@ define i64 @test_mul_by_12(i64 %x) {
> ; X86-NEXT:    leal (%edx,%ecx,4), %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_12:
> -; X64:       # BB#0:
> -; X64-NEXT:    shlq $2, %rdi
> -; X64-NEXT:    leaq (%rdi,%rdi,2), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_12:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    shlq $2, %rdi # sched: [1:0.50]
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_12:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    shlq $2, %rdi # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_12:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $12, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $12, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_12:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $12, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_12:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $12, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_12:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    shlq $2, %rdi # sched: [1:1.00]
> +; X64-SLM-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_12:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $12, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 12
>   ret i64 %mul
> }
> @@ -216,16 +628,53 @@ define i64 @test_mul_by_12(i64 %x) {
> define i64 @test_mul_by_13(i64 %x) {
> ; X86-LABEL: test_mul_by_13:
> ; X86:       # BB#0:
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    leal (%eax,%eax,2), %ecx
> +; X86-NEXT:    leal (%eax,%ecx,4), %ecx
> ; X86-NEXT:    movl $13, %eax
> ; X86-NEXT:    mull {{[0-9]+}}(%esp)
> -; X86-NEXT:    imull $13, {{[0-9]+}}(%esp), %ecx
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_13:
> -; X64:       # BB#0:
> -; X64-NEXT:    imulq $13, %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_13:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leaq (%rdi,%rax,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_13:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rdi,%rax,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_13:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $13, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $13, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_13:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $13, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_13:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $13, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_13:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imulq $13, %rdi, %rax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_13:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $13, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 13
>   ret i64 %mul
> }
> @@ -233,16 +682,56 @@ define i64 @test_mul_by_13(i64 %x) {
> define i64 @test_mul_by_14(i64 %x) {
> ; X86-LABEL: test_mul_by_14:
> ; X86:       # BB#0:
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    leal (%eax,%eax,2), %ecx
> +; X86-NEXT:    leal (%eax,%ecx,4), %ecx
> +; X86-NEXT:    addl %eax, %ecx
> ; X86-NEXT:    movl $14, %eax
> ; X86-NEXT:    mull {{[0-9]+}}(%esp)
> -; X86-NEXT:    imull $14, {{[0-9]+}}(%esp), %ecx
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_14:
> -; X64:       # BB#0:
> -; X64-NEXT:    imulq $14, %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_14:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leaq (%rdi,%rax,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    addq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_14:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rdi,%rax,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    addq %rdi, %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_14:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $14, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $14, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_14:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $14, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_14:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $14, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_14:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imulq $14, %rdi, %rax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_14:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $14, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 14
>   ret i64 %mul
> }
> @@ -258,11 +747,46 @@ define i64 @test_mul_by_15(i64 %x) {
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_15:
> -; X64:       # BB#0:
> -; X64-NEXT:    leaq (%rdi,%rdi,4), %rax
> -; X64-NEXT:    leaq (%rax,%rax,2), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_15:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leaq (%rax,%rax,2), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_15:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rax,%rax,2), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_15:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $15, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $15, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_15:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $15, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_15:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $15, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_15:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    leaq (%rax,%rax,2), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_15:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $15, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 15
>   ret i64 %mul
> }
> @@ -276,11 +800,49 @@ define i64 @test_mul_by_16(i64 %x) {
> ; X86-NEXT:    shll $4, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_16:
> -; X64:       # BB#0:
> -; X64-NEXT:    shlq $4, %rdi
> -; X64-NEXT:    movq %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_16:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    shlq $4, %rdi # sched: [1:0.50]
> +; X64-HSW-NEXT:    movq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_16:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    shlq $4, %rdi # sched: [1:0.50]
> +; X64-JAG-NEXT:    movq %rdi, %rax # sched: [1:0.17]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_16:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %edx
> +; X86-NOOPT-NEXT:    shldl $4, %eax, %edx
> +; X86-NOOPT-NEXT:    shll $4, %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_16:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    shlq $4, %rdi # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    movq %rdi, %rax # sched: [1:0.25]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_16:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    shlq $4, %rdi # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    movq %rdi, %rax # sched: [1:0.17]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_16:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    shlq $4, %rdi # sched: [1:1.00]
> +; X64-SLM-NEXT:    movq %rdi, %rax # sched: [1:0.50]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_16:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    shlq $4, %rdi # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    movq %rdi, %rax # sched: [1:0.50]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 16
>   ret i64 %mul
> }
> @@ -297,12 +859,49 @@ define i64 @test_mul_by_17(i64 %x) {
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_17:
> -; X64:       # BB#0:
> -; X64-NEXT:    movq %rdi, %rax
> -; X64-NEXT:    shlq $4, %rax
> -; X64-NEXT:    leaq (%rax,%rdi), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_17:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    movq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    shlq $4, %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leaq (%rax,%rdi), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_17:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    movq %rdi, %rax # sched: [1:0.17]
> +; X64-JAG-NEXT:    shlq $4, %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rax,%rdi), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_17:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $17, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $17, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_17:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $17, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_17:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $17, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_17:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    movq %rdi, %rax # sched: [1:0.50]
> +; X64-SLM-NEXT:    shlq $4, %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    addq %rdi, %rax # sched: [1:0.50]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_17:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $17, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 17
>   ret i64 %mul
> }
> @@ -317,11 +916,46 @@ define i64 @test_mul_by_18(i64 %x) {
> ; X86-NEXT:    leal (%edx,%ecx,2), %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_18:
> -; X64:       # BB#0:
> -; X64-NEXT:    addq %rdi, %rdi
> -; X64-NEXT:    leaq (%rdi,%rdi,8), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_18:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    addq %rdi, %rdi # sched: [1:0.25]
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_18:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    addq %rdi, %rdi # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_18:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $18, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $18, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_18:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $18, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_18:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $18, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_18:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    addq %rdi, %rdi # sched: [1:0.50]
> +; X64-SLM-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_18:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $18, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 18
>   ret i64 %mul
> }
> @@ -329,16 +963,58 @@ define i64 @test_mul_by_18(i64 %x) {
> define i64 @test_mul_by_19(i64 %x) {
> ; X86-LABEL: test_mul_by_19:
> ; X86:       # BB#0:
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
> +; X86-NEXT:    leal (%ecx,%ecx,4), %eax
> +; X86-NEXT:    shll $2, %eax
> +; X86-NEXT:    subl %eax, %ecx
> ; X86-NEXT:    movl $19, %eax
> ; X86-NEXT:    mull {{[0-9]+}}(%esp)
> -; X86-NEXT:    imull $19, {{[0-9]+}}(%esp), %ecx
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_19:
> -; X64:       # BB#0:
> -; X64-NEXT:    imulq $19, %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_19:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    shlq $2, %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    subq %rax, %rdi # sched: [1:0.25]
> +; X64-HSW-NEXT:    movq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_19:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    shlq $2, %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    subq %rax, %rdi # sched: [1:0.50]
> +; X64-JAG-NEXT:    movq %rdi, %rax # sched: [1:0.17]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_19:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $19, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $19, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_19:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $19, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_19:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $19, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_19:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imulq $19, %rdi, %rax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_19:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $19, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 19
>   ret i64 %mul
> }
> @@ -353,11 +1029,46 @@ define i64 @test_mul_by_20(i64 %x) {
> ; X86-NEXT:    leal (%edx,%ecx,4), %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_20:
> -; X64:       # BB#0:
> -; X64-NEXT:    shlq $2, %rdi
> -; X64-NEXT:    leaq (%rdi,%rdi,4), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_20:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    shlq $2, %rdi # sched: [1:0.50]
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_20:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    shlq $2, %rdi # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_20:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $20, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $20, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_20:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $20, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_20:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $20, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_20:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    shlq $2, %rdi # sched: [1:1.00]
> +; X64-SLM-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_20:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $20, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 20
>   ret i64 %mul
> }
> @@ -365,16 +1076,53 @@ define i64 @test_mul_by_20(i64 %x) {
> define i64 @test_mul_by_21(i64 %x) {
> ; X86-LABEL: test_mul_by_21:
> ; X86:       # BB#0:
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    leal (%eax,%eax,4), %ecx
> +; X86-NEXT:    leal (%eax,%ecx,4), %ecx
> ; X86-NEXT:    movl $21, %eax
> ; X86-NEXT:    mull {{[0-9]+}}(%esp)
> -; X86-NEXT:    imull $21, {{[0-9]+}}(%esp), %ecx
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_21:
> -; X64:       # BB#0:
> -; X64-NEXT:    imulq $21, %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_21:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leaq (%rdi,%rax,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_21:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rdi,%rax,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_21:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $21, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $21, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_21:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $21, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_21:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $21, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_21:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imulq $21, %rdi, %rax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_21:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $21, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 21
>   ret i64 %mul
> }
> @@ -382,16 +1130,56 @@ define i64 @test_mul_by_21(i64 %x) {
> define i64 @test_mul_by_22(i64 %x) {
> ; X86-LABEL: test_mul_by_22:
> ; X86:       # BB#0:
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    leal (%eax,%eax,4), %ecx
> +; X86-NEXT:    leal (%eax,%ecx,4), %ecx
> +; X86-NEXT:    addl %eax, %ecx
> ; X86-NEXT:    movl $22, %eax
> ; X86-NEXT:    mull {{[0-9]+}}(%esp)
> -; X86-NEXT:    imull $22, {{[0-9]+}}(%esp), %ecx
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_22:
> -; X64:       # BB#0:
> -; X64-NEXT:    imulq $22, %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_22:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leaq (%rdi,%rax,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    addq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_22:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rdi,%rax,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    addq %rdi, %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_22:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $22, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $22, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_22:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $22, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_22:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $22, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_22:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imulq $22, %rdi, %rax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_22:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $22, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 22
>   ret i64 %mul
> }
> @@ -399,16 +1187,58 @@ define i64 @test_mul_by_22(i64 %x) {
> define i64 @test_mul_by_23(i64 %x) {
> ; X86-LABEL: test_mul_by_23:
> ; X86:       # BB#0:
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
> +; X86-NEXT:    leal (%ecx,%ecx,2), %eax
> +; X86-NEXT:    shll $3, %eax
> +; X86-NEXT:    subl %eax, %ecx
> ; X86-NEXT:    movl $23, %eax
> ; X86-NEXT:    mull {{[0-9]+}}(%esp)
> -; X86-NEXT:    imull $23, {{[0-9]+}}(%esp), %ecx
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_23:
> -; X64:       # BB#0:
> -; X64-NEXT:    imulq $23, %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_23:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    shlq $3, %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    subq %rax, %rdi # sched: [1:0.25]
> +; X64-HSW-NEXT:    movq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_23:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    shlq $3, %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    subq %rax, %rdi # sched: [1:0.50]
> +; X64-JAG-NEXT:    movq %rdi, %rax # sched: [1:0.17]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_23:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $23, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $23, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_23:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $23, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_23:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $23, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_23:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imulq $23, %rdi, %rax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_23:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $23, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 23
>   ret i64 %mul
> }
> @@ -423,11 +1253,46 @@ define i64 @test_mul_by_24(i64 %x) {
> ; X86-NEXT:    leal (%edx,%ecx,8), %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_24:
> -; X64:       # BB#0:
> -; X64-NEXT:    shlq $3, %rdi
> -; X64-NEXT:    leaq (%rdi,%rdi,2), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_24:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    shlq $3, %rdi # sched: [1:0.50]
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_24:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    shlq $3, %rdi # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_24:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $24, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $24, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_24:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $24, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_24:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $24, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_24:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    shlq $3, %rdi # sched: [1:1.00]
> +; X64-SLM-NEXT:    leaq (%rdi,%rdi,2), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_24:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $24, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 24
>   ret i64 %mul
> }
> @@ -443,11 +1308,46 @@ define i64 @test_mul_by_25(i64 %x) {
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_25:
> -; X64:       # BB#0:
> -; X64-NEXT:    leaq (%rdi,%rdi,4), %rax
> -; X64-NEXT:    leaq (%rax,%rax,4), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_25:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leaq (%rax,%rax,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_25:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rax,%rax,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_25:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $25, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $25, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_25:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $25, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_25:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $25, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_25:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    leaq (%rax,%rax,4), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_25:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $25, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 25
>   ret i64 %mul
> }
> @@ -455,16 +1355,58 @@ define i64 @test_mul_by_25(i64 %x) {
> define i64 @test_mul_by_26(i64 %x) {
> ; X86-LABEL: test_mul_by_26:
> ; X86:       # BB#0:
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
> +; X86-NEXT:    leal (%ecx,%ecx,8), %eax
> +; X86-NEXT:    leal (%eax,%eax,2), %eax
> +; X86-NEXT:    subl %eax, %ecx
> ; X86-NEXT:    movl $26, %eax
> ; X86-NEXT:    mull {{[0-9]+}}(%esp)
> -; X86-NEXT:    imull $26, {{[0-9]+}}(%esp), %ecx
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_26:
> -; X64:       # BB#0:
> -; X64-NEXT:    imulq $26, %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_26:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leaq (%rax,%rax,2), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    subq %rax, %rdi # sched: [1:0.25]
> +; X64-HSW-NEXT:    movq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_26:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rax,%rax,2), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    subq %rax, %rdi # sched: [1:0.50]
> +; X64-JAG-NEXT:    movq %rdi, %rax # sched: [1:0.17]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_26:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $26, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $26, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_26:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $26, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_26:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $26, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_26:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imulq $26, %rdi, %rax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_26:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $26, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 26
>   ret i64 %mul
> }
> @@ -480,11 +1422,46 @@ define i64 @test_mul_by_27(i64 %x) {
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_27:
> -; X64:       # BB#0:
> -; X64-NEXT:    leaq (%rdi,%rdi,8), %rax
> -; X64-NEXT:    leaq (%rax,%rax,2), %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_27:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leaq (%rax,%rax,2), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_27:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rax,%rax,2), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_27:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $27, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $27, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_27:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $27, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_27:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $27, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_27:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    leaq (%rax,%rax,2), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_27:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $27, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 27
>   ret i64 %mul
> }
> @@ -492,16 +1469,56 @@ define i64 @test_mul_by_27(i64 %x) {
> define i64 @test_mul_by_28(i64 %x) {
> ; X86-LABEL: test_mul_by_28:
> ; X86:       # BB#0:
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    leal (%eax,%eax,8), %ecx
> +; X86-NEXT:    leal (%ecx,%ecx,2), %ecx
> +; X86-NEXT:    addl %eax, %ecx
> ; X86-NEXT:    movl $28, %eax
> ; X86-NEXT:    mull {{[0-9]+}}(%esp)
> -; X86-NEXT:    imull $28, {{[0-9]+}}(%esp), %ecx
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_28:
> -; X64:       # BB#0:
> -; X64-NEXT:    imulq $28, %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_28:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leaq (%rax,%rax,2), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    addq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_28:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rax,%rax,2), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    addq %rdi, %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_28:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $28, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $28, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_28:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $28, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_28:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $28, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_28:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imulq $28, %rdi, %rax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_28:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $28, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 28
>   ret i64 %mul
> }
> @@ -509,16 +1526,59 @@ define i64 @test_mul_by_28(i64 %x) {
> define i64 @test_mul_by_29(i64 %x) {
> ; X86-LABEL: test_mul_by_29:
> ; X86:       # BB#0:
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NEXT:    leal (%eax,%eax,8), %ecx
> +; X86-NEXT:    leal (%ecx,%ecx,2), %ecx
> +; X86-NEXT:    addl %eax, %ecx
> +; X86-NEXT:    addl %eax, %ecx
> ; X86-NEXT:    movl $29, %eax
> ; X86-NEXT:    mull {{[0-9]+}}(%esp)
> -; X86-NEXT:    imull $29, {{[0-9]+}}(%esp), %ecx
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_29:
> -; X64:       # BB#0:
> -; X64-NEXT:    imulq $29, %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_29:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    leaq (%rax,%rax,2), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    addq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    addq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_29:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq (%rdi,%rdi,8), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq (%rax,%rax,2), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    addq %rdi, %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    addq %rdi, %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_29:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $29, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $29, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_29:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $29, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_29:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $29, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_29:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imulq $29, %rdi, %rax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_29:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $29, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 29
>   ret i64 %mul
> }
> @@ -526,16 +1586,60 @@ define i64 @test_mul_by_29(i64 %x) {
> define i64 @test_mul_by_30(i64 %x) {
> ; X86-LABEL: test_mul_by_30:
> ; X86:       # BB#0:
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
> +; X86-NEXT:    shll $5, %ecx
> ; X86-NEXT:    movl $30, %eax
> ; X86-NEXT:    mull {{[0-9]+}}(%esp)
> -; X86-NEXT:    imull $30, {{[0-9]+}}(%esp), %ecx
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_30:
> -; X64:       # BB#0:
> -; X64-NEXT:    imulq $30, %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_30:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    movq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    shlq $5, %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    movq %rdi, %rcx # sched: [1:0.25]
> +; X64-HSW-NEXT:    subq %rax, %rcx # sched: [1:0.25]
> +; X64-HSW-NEXT:    subq %rcx, %rdi # sched: [1:0.25]
> +; X64-HSW-NEXT:    movq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_30:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    movq %rdi, %rax # sched: [1:0.17]
> +; X64-JAG-NEXT:    movq %rdi, %rcx # sched: [1:0.17]
> +; X64-JAG-NEXT:    shlq $5, %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    subq %rax, %rcx # sched: [1:0.50]
> +; X64-JAG-NEXT:    subq %rcx, %rdi # sched: [1:0.50]
> +; X64-JAG-NEXT:    movq %rdi, %rax # sched: [1:0.17]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_30:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $30, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $30, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_30:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $30, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_30:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $30, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_30:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    imulq $30, %rdi, %rax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_30:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $30, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 30
>   ret i64 %mul
> }
> @@ -552,12 +1656,49 @@ define i64 @test_mul_by_31(i64 %x) {
> ; X86-NEXT:    addl %ecx, %edx
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_31:
> -; X64:       # BB#0:
> -; X64-NEXT:    movq %rdi, %rax
> -; X64-NEXT:    shlq $5, %rax
> -; X64-NEXT:    subq %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_31:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    movq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    shlq $5, %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    subq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_31:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    movq %rdi, %rax # sched: [1:0.17]
> +; X64-JAG-NEXT:    shlq $5, %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    subq %rdi, %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_31:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl $31, %eax
> +; X86-NOOPT-NEXT:    mull {{[0-9]+}}(%esp)
> +; X86-NOOPT-NEXT:    imull $31, {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_31:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    imulq $31, %rdi, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_31:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    imulq $31, %rdi, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_31:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    movq %rdi, %rax # sched: [1:0.50]
> +; X64-SLM-NEXT:    shlq $5, %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    subq %rdi, %rax # sched: [1:0.50]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_31:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    imulq $31, %rdi, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 31
>   ret i64 %mul
> }
> @@ -571,11 +1712,168 @@ define i64 @test_mul_by_32(i64 %x) {
> ; X86-NEXT:    shll $5, %eax
> ; X86-NEXT:    retl
> ;
> -; X64-LABEL: test_mul_by_32:
> -; X64:       # BB#0:
> -; X64-NEXT:    shlq $5, %rdi
> -; X64-NEXT:    movq %rdi, %rax
> -; X64-NEXT:    retq
> +; X64-HSW-LABEL: test_mul_by_32:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    shlq $5, %rdi # sched: [1:0.50]
> +; X64-HSW-NEXT:    movq %rdi, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_by_32:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    shlq $5, %rdi # sched: [1:0.50]
> +; X64-JAG-NEXT:    movq %rdi, %rax # sched: [1:0.17]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_by_32:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %eax
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %edx
> +; X86-NOOPT-NEXT:    shldl $5, %eax, %edx
> +; X86-NOOPT-NEXT:    shll $5, %eax
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_by_32:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    shlq $5, %rdi # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    movq %rdi, %rax # sched: [1:0.25]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_by_32:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    shlq $5, %rdi # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    movq %rdi, %rax # sched: [1:0.17]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_by_32:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    shlq $5, %rdi # sched: [1:1.00]
> +; X64-SLM-NEXT:    movq %rdi, %rax # sched: [1:0.50]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_by_32:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    shlq $5, %rdi # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    movq %rdi, %rax # sched: [1:0.50]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
>   %mul = mul nsw i64 %x, 32
>   ret i64 %mul
> }
> +
> +; (x*9+42)*(x*5+2)
> +define i64 @test_mul_spec(i64 %x) nounwind {
> +; X86-LABEL: test_mul_spec:
> +; X86:       # BB#0:
> +; X86-NEXT:    pushl %ebx
> +; X86-NEXT:    pushl %edi
> +; X86-NEXT:    pushl %esi
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
> +; X86-NEXT:    movl {{[0-9]+}}(%esp), %edi
> +; X86-NEXT:    movl $9, %edx
> +; X86-NEXT:    movl %ecx, %eax
> +; X86-NEXT:    mull %edx
> +; X86-NEXT:    movl %eax, %esi
> +; X86-NEXT:    leal (%edi,%edi,8), %ebx
> +; X86-NEXT:    addl $42, %esi
> +; X86-NEXT:    adcl %edx, %ebx
> +; X86-NEXT:    movl $5, %edx
> +; X86-NEXT:    movl %ecx, %eax
> +; X86-NEXT:    mull %edx
> +; X86-NEXT:    movl %eax, %ecx
> +; X86-NEXT:    leal (%edi,%edi,4), %edi
> +; X86-NEXT:    addl $2, %ecx
> +; X86-NEXT:    adcl %edx, %edi
> +; X86-NEXT:    movl %esi, %eax
> +; X86-NEXT:    mull %ecx
> +; X86-NEXT:    imull %esi, %edi
> +; X86-NEXT:    addl %edi, %edx
> +; X86-NEXT:    imull %ebx, %ecx
> +; X86-NEXT:    addl %ecx, %edx
> +; X86-NEXT:    popl %esi
> +; X86-NEXT:    popl %edi
> +; X86-NEXT:    popl %ebx
> +; X86-NEXT:    retl
> +;
> +; X64-HSW-LABEL: test_mul_spec:
> +; X64-HSW:       # BB#0:
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,8), %rcx # sched: [1:0.50]
> +; X64-HSW-NEXT:    addq $42, %rcx # sched: [1:0.25]
> +; X64-HSW-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-HSW-NEXT:    addq $2, %rax # sched: [1:0.25]
> +; X64-HSW-NEXT:    imulq %rcx, %rax # sched: [3:1.00]
> +; X64-HSW-NEXT:    retq # sched: [1:1.00]
> +;
> +; X64-JAG-LABEL: test_mul_spec:
> +; X64-JAG:       # BB#0:
> +; X64-JAG-NEXT:    leaq 42(%rdi,%rdi,8), %rcx # sched: [1:0.50]
> +; X64-JAG-NEXT:    leaq 2(%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; X64-JAG-NEXT:    imulq %rcx, %rax # sched: [3:1.00]
> +; X64-JAG-NEXT:    retq # sched: [4:1.00]
> +;
> +; X86-NOOPT-LABEL: test_mul_spec:
> +; X86-NOOPT:       # BB#0:
> +; X86-NOOPT-NEXT:    pushl %ebx
> +; X86-NOOPT-NEXT:    pushl %edi
> +; X86-NOOPT-NEXT:    pushl %esi
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %ecx
> +; X86-NOOPT-NEXT:    movl {{[0-9]+}}(%esp), %edi
> +; X86-NOOPT-NEXT:    movl $9, %edx
> +; X86-NOOPT-NEXT:    movl %ecx, %eax
> +; X86-NOOPT-NEXT:    mull %edx
> +; X86-NOOPT-NEXT:    movl %eax, %esi
> +; X86-NOOPT-NEXT:    leal (%edi,%edi,8), %ebx
> +; X86-NOOPT-NEXT:    addl $42, %esi
> +; X86-NOOPT-NEXT:    adcl %edx, %ebx
> +; X86-NOOPT-NEXT:    movl $5, %edx
> +; X86-NOOPT-NEXT:    movl %ecx, %eax
> +; X86-NOOPT-NEXT:    mull %edx
> +; X86-NOOPT-NEXT:    movl %eax, %ecx
> +; X86-NOOPT-NEXT:    leal (%edi,%edi,4), %edi
> +; X86-NOOPT-NEXT:    addl $2, %ecx
> +; X86-NOOPT-NEXT:    adcl %edx, %edi
> +; X86-NOOPT-NEXT:    movl %esi, %eax
> +; X86-NOOPT-NEXT:    mull %ecx
> +; X86-NOOPT-NEXT:    imull %esi, %edi
> +; X86-NOOPT-NEXT:    addl %edi, %edx
> +; X86-NOOPT-NEXT:    imull %ebx, %ecx
> +; X86-NOOPT-NEXT:    addl %ecx, %edx
> +; X86-NOOPT-NEXT:    popl %esi
> +; X86-NOOPT-NEXT:    popl %edi
> +; X86-NOOPT-NEXT:    popl %ebx
> +; X86-NOOPT-NEXT:    retl
> +;
> +; HSW-NOOPT-LABEL: test_mul_spec:
> +; HSW-NOOPT:       # BB#0:
> +; HSW-NOOPT-NEXT:    leaq (%rdi,%rdi,8), %rcx # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    addq $42, %rcx # sched: [1:0.25]
> +; HSW-NOOPT-NEXT:    leaq (%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; HSW-NOOPT-NEXT:    addq $2, %rax # sched: [1:0.25]
> +; HSW-NOOPT-NEXT:    imulq %rcx, %rax # sched: [3:1.00]
> +; HSW-NOOPT-NEXT:    retq # sched: [1:1.00]
> +;
> +; JAG-NOOPT-LABEL: test_mul_spec:
> +; JAG-NOOPT:       # BB#0:
> +; JAG-NOOPT-NEXT:    leaq 42(%rdi,%rdi,8), %rcx # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    leaq 2(%rdi,%rdi,4), %rax # sched: [1:0.50]
> +; JAG-NOOPT-NEXT:    imulq %rcx, %rax # sched: [3:1.00]
> +; JAG-NOOPT-NEXT:    retq # sched: [4:1.00]
> +;
> +; X64-SLM-LABEL: test_mul_spec:
> +; X64-SLM:       # BB#0:
> +; X64-SLM-NEXT:    leaq 42(%rdi,%rdi,8), %rcx # sched: [1:1.00]
> +; X64-SLM-NEXT:    leaq 2(%rdi,%rdi,4), %rax # sched: [1:1.00]
> +; X64-SLM-NEXT:    imulq %rcx, %rax # sched: [3:1.00]
> +; X64-SLM-NEXT:    retq # sched: [4:1.00]
> +;
> +; SLM-NOOPT-LABEL: test_mul_spec:
> +; SLM-NOOPT:       # BB#0:
> +; SLM-NOOPT-NEXT:    leaq 42(%rdi,%rdi,8), %rcx # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    leaq 2(%rdi,%rdi,4), %rax # sched: [1:1.00]
> +; SLM-NOOPT-NEXT:    imulq %rcx, %rax # sched: [3:1.00]
> +; SLM-NOOPT-NEXT:    retq # sched: [4:1.00]
> +  %mul = mul nsw i64 %x, 9
> +  %add = add nsw i64 %mul, 42
> +  %mul2 = mul nsw i64 %x, 5
> +  %add2 = add nsw i64 %mul2, 2
> +  %mul3 = mul nsw i64 %add, %add2
> +  ret i64 %mul3
> +}
> 
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170530/7892d0cf/attachment-0001.html>


More information about the llvm-commits mailing list