This patch closes PR28513: an optimization of multiplication by different constants.

The initial patch was rejected: I fixed the issue and re-apply it.

llvm-svn: 304972

4 files changed, 4253 insertions, 360 deletions

diff --git a/llvm/test/CodeGen/X86/mul-constant-i16.ll b/llvm/test/CodeGen/X86/mul-constant-i16.ll
index e3e2737cf3e..7b39bfe1c48 100644
--- a/llvm/test/CodeGen/X86/mul-constant-i16.ll
+++ b/llvm/test/CodeGen/X86/mul-constant-i16.ll
@@ -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
@@ -337,14 +348,19 @@ define i16 @test_mul_by_18(i16 %x) {
 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:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    leal (%ecx,%ecx,4), %eax
+; X86-NEXT:    shll $2, %eax
+; 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: %EDI<def> %EDI<kill> %RDI<def>
+; X64-NEXT:    leal (%rdi,%rdi,4), %eax
+; X64-NEXT:    shll $2, %eax
+; X64-NEXT:    subl %edi, %eax
 ; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
 ; X64-NEXT:    retq
   %mul = mul nsw i16 %x, 19
@@ -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
@@ -408,14 +432,19 @@ define i16 @test_mul_by_22(i16 %x) {
 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:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    leal (%ecx,%ecx,2), %eax
+; X86-NEXT:    shll $3, %eax
+; 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: %EDI<def> %EDI<kill> %RDI<def>
+; X64-NEXT:    leal (%rdi,%rdi,2), %eax
+; X64-NEXT:    shll $3, %eax
+; X64-NEXT:    subl %edi, %eax
 ; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
 ; X64-NEXT:    retq
   %mul = mul nsw i16 %x, 23
@@ -465,14 +494,19 @@ define i16 @test_mul_by_25(i16 %x) {
 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:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    leal (%ecx,%ecx,8), %eax
+; X86-NEXT:    leal (%eax,%eax,2), %eax
+; 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: %EDI<def> %EDI<kill> %RDI<def>
+; X64-NEXT:    leal (%rdi,%rdi,8), %eax
+; X64-NEXT:    leal (%rax,%rax,2), %eax
+; X64-NEXT:    subl %edi, %eax
 ; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
 ; X64-NEXT:    retq
   %mul = mul nsw i16 %x, 26
@@ -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
@@ -536,14 +582,20 @@ define i16 @test_mul_by_29(i16 %x) {
 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:    movzwl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl %ecx, %eax
+; X86-NEXT:    shll $5, %eax
+; X86-NEXT:    subl %ecx, %eax
+; X86-NEXT:    subl %ecx, %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:    movl %edi, %eax
+; X64-NEXT:    shll $5, %eax
+; X64-NEXT:    subl %edi, %eax
+; X64-NEXT:    subl %edi, %eax
 ; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
 ; X64-NEXT:    retq
   %mul = mul nsw i16 %x, 30
@@ -587,3 +639,30 @@ 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
+}
diff --git a/llvm/test/CodeGen/X86/mul-constant-i32.ll b/llvm/test/CodeGen/X86/mul-constant-i32.ll
index 76e46e1f1b0..d545b477e10 100644
--- a/llvm/test/CodeGen/X86/mul-constant-i32.ll
+++ b/llvm/test/CodeGen/X86/mul-constant-i32.ll
@@ -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 +934,52 @@ 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), %ecx
+; X86-NEXT:    leal (%ecx,%ecx,4), %eax
+; X86-NEXT:    shll $2, %eax
+; 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 %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 %edi, %eax # sched: [1:0.50]
+; 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 +992,46 @@ 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 +1039,49 @@ 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 +1089,52 @@ 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 +1142,52 @@ 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), %ecx
+; X86-NEXT:    leal (%ecx,%ecx,2), %eax
+; X86-NEXT:    shll $3, %eax
+; 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 %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 %edi, %eax # sched: [1:0.50]
+; 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 +1200,46 @@ 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 +1252,46 @@ 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 +1299,52 @@ 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), %ecx
+; X86-NEXT:    leal (%ecx,%ecx,8), %eax
+; X86-NEXT:    leal (%eax,%eax,2), %eax
+; 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 %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 %edi, %eax # sched: [1:0.50]
+; 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 +1357,46 @@ 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 +1404,52 @@ 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 +1457,55 @@ 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 +1513,53 @@ 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:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl %ecx, %eax
+; X86-NEXT:    shll $5, %eax
+; X86-NEXT:    subl %ecx, %eax
+; X86-NEXT:    subl %ecx, %eax
 ; X86-NEXT:    retl
 ;
-; X64-LABEL: test_mul_by_30:
-; X64:       # BB#0:
-; X64-NEXT:    imull $30, %edi, %eax
-; X64-NEXT:    retq
+; 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:    subl %edi, %eax # sched: [1:0.25]
+; X64-HSW-NEXT:    subl %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:    shll $5, %eax # sched: [1:0.50]
+; X64-JAG-NEXT:    subl %edi, %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_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 +1573,46 @@ 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 +1624,124 @@ 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
+}
diff --git a/llvm/test/CodeGen/X86/mul-constant-i64.ll b/llvm/test/CodeGen/X86/mul-constant-i64.ll
index 8579179a823..ea841c761c7 100644
--- a/llvm/test/CodeGen/X86/mul-constant-i64.ll
+++ b/llvm/test/CodeGen/X86/mul-constant-i64.ll
@@ -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 +963,56 @@ 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), %eax
+; X86-NEXT:    leal (%eax,%eax,4), %ecx
+; X86-NEXT:    shll $2, %ecx
+; 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 %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 %rdi, %rax # sched: [1:0.50]
+; 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 +1027,46 @@ 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 +1074,53 @@ 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 +1128,56 @@ 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 +1185,56 @@ 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), %eax
+; X86-NEXT:    leal (%eax,%eax,2), %ecx
+; X86-NEXT:    shll $3, %ecx
+; 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 %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 %rdi, %rax # sched: [1:0.50]
+; 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 +1249,46 @@ 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 +1304,46 @@ 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 +1351,56 @@ 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), %eax
+; X86-NEXT:    leal (%eax,%eax,8), %ecx
+; X86-NEXT:    leal (%ecx,%ecx,2), %ecx
+; 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 %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 %rdi, %rax # sched: [1:0.50]
+; 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 +1416,46 @@ 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 +1463,56 @@ 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 +1520,59 @@ 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 +1580,59 @@ 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), %eax
+; X86-NEXT:    movl %eax, %ecx
+; X86-NEXT:    shll $5, %ecx
+; X86-NEXT:    subl %eax, %ecx
+; X86-NEXT:    subl %eax, %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:    subq %rdi, %rax # sched: [1:0.25]
+; X64-HSW-NEXT:    subq %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:    shlq $5, %rax # sched: [1:0.50]
+; X64-JAG-NEXT:    subq %rdi, %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_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 +1649,49 @@ 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 +1705,168 @@ 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
+}
diff --git a/llvm/test/CodeGen/X86/mul-constant-result.ll b/llvm/test/CodeGen/X86/mul-constant-result.ll
new file mode 100644
index 00000000000..65d80a699e2
--- /dev/null
+++ b/llvm/test/CodeGen/X86/mul-constant-result.ll
@@ -0,0 +1,1291 @@
+; 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 -mcpu=haswell| FileCheck %s --check-prefix=X64-HSW
+
+; Function Attrs: norecurse nounwind readnone uwtable
+define i32 @mult(i32, i32) local_unnamed_addr #0 {
+; X86-LABEL: mult:
+; X86:       # BB#0:
+; X86-NEXT:    pushl %esi
+; X86-NEXT:  .Lcfi0:
+; X86-NEXT:    .cfi_def_cfa_offset 8
+; X86-NEXT:  .Lcfi1:
+; X86-NEXT:    .cfi_offset %esi, -8
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    cmpl $1, %edx
+; X86-NEXT:    movl $1, %eax
+; X86-NEXT:    movl $1, %esi
+; X86-NEXT:    jg .LBB0_2
+; X86-NEXT:  # BB#1:
+; X86-NEXT:    movl %edx, %esi
+; X86-NEXT:  .LBB0_2:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    testl %edx, %edx
+; X86-NEXT:    je .LBB0_4
+; X86-NEXT:  # BB#3:
+; X86-NEXT:    movl %esi, %eax
+; X86-NEXT:  .LBB0_4:
+; X86-NEXT:    decl %ecx
+; X86-NEXT:    cmpl $31, %ecx
+; X86-NEXT:    ja .LBB0_39
+; X86-NEXT:  # BB#5:
+; X86-NEXT:    jmpl *.LJTI0_0(,%ecx,4)
+; X86-NEXT:  .LBB0_6:
+; X86-NEXT:    addl %eax, %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_39:
+; X86-NEXT:    xorl %eax, %eax
+; X86-NEXT:  .LBB0_40:
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_7:
+; X86-NEXT:    leal (%eax,%eax,2), %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_8:
+; X86-NEXT:    shll $2, %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_9:
+; X86-NEXT:    leal (%eax,%eax,4), %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_10:
+; X86-NEXT:    addl %eax, %eax
+; X86-NEXT:    leal (%eax,%eax,2), %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_11:
+; X86-NEXT:    leal (,%eax,8), %ecx
+; X86-NEXT:    jmp .LBB0_12
+; X86-NEXT:  .LBB0_13:
+; X86-NEXT:    shll $3, %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_14:
+; X86-NEXT:    leal (%eax,%eax,8), %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_15:
+; X86-NEXT:    addl %eax, %eax
+; X86-NEXT:    leal (%eax,%eax,4), %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_16:
+; X86-NEXT:    leal (%eax,%eax,4), %ecx
+; X86-NEXT:    leal (%eax,%ecx,2), %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_17:
+; X86-NEXT:    shll $2, %eax
+; X86-NEXT:    leal (%eax,%eax,2), %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_18:
+; X86-NEXT:    leal (%eax,%eax,2), %ecx
+; X86-NEXT:    leal (%eax,%ecx,4), %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_19:
+; X86-NEXT:    leal (%eax,%eax,2), %ecx
+; X86-NEXT:    jmp .LBB0_20
+; X86-NEXT:  .LBB0_21:
+; X86-NEXT:    leal (%eax,%eax,4), %eax
+; X86-NEXT:    leal (%eax,%eax,2), %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_22:
+; X86-NEXT:    shll $4, %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_23:
+; X86-NEXT:    movl %eax, %ecx
+; X86-NEXT:    shll $4, %ecx
+; X86-NEXT:    addl %ecx, %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_24:
+; X86-NEXT:    addl %eax, %eax
+; X86-NEXT:    leal (%eax,%eax,8), %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_25:
+; X86-NEXT:    leal (%eax,%eax,4), %ecx
+; X86-NEXT:    shll $2, %ecx
+; X86-NEXT:    jmp .LBB0_12
+; X86-NEXT:  .LBB0_26:
+; X86-NEXT:    shll $2, %eax
+; X86-NEXT:    leal (%eax,%eax,4), %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_27:
+; X86-NEXT:    leal (%eax,%eax,4), %ecx
+; X86-NEXT:    leal (%eax,%ecx,4), %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_28:
+; X86-NEXT:    leal (%eax,%eax,4), %ecx
+; X86-NEXT:  .LBB0_20:
+; X86-NEXT:    leal (%eax,%ecx,4), %ecx
+; X86-NEXT:    addl %ecx, %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_29:
+; X86-NEXT:    leal (%eax,%eax,2), %ecx
+; X86-NEXT:    shll $3, %ecx
+; X86-NEXT:    jmp .LBB0_12
+; X86-NEXT:  .LBB0_30:
+; X86-NEXT:    shll $3, %eax
+; X86-NEXT:    leal (%eax,%eax,2), %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_31:
+; X86-NEXT:    leal (%eax,%eax,4), %eax
+; X86-NEXT:    leal (%eax,%eax,4), %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_32:
+; X86-NEXT:    leal (%eax,%eax,8), %ecx
+; X86-NEXT:    leal (%ecx,%ecx,2), %ecx
+; X86-NEXT:    jmp .LBB0_12
+; X86-NEXT:  .LBB0_33:
+; X86-NEXT:    leal (%eax,%eax,8), %eax
+; X86-NEXT:    leal (%eax,%eax,2), %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_34:
+; X86-NEXT:    leal (%eax,%eax,8), %ecx
+; X86-NEXT:    leal (%ecx,%ecx,2), %ecx
+; X86-NEXT:    addl %ecx, %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_35:
+; X86-NEXT:    leal (%eax,%eax,8), %ecx
+; X86-NEXT:    leal (%ecx,%ecx,2), %ecx
+; X86-NEXT:    addl %eax, %ecx
+; X86-NEXT:    addl %ecx, %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_36:
+; X86-NEXT:    movl %eax, %ecx
+; X86-NEXT:    shll $5, %ecx
+; X86-NEXT:    subl %eax, %ecx
+; X86-NEXT:    jmp .LBB0_12
+; X86-NEXT:  .LBB0_37:
+; X86-NEXT:    movl %eax, %ecx
+; X86-NEXT:    shll $5, %ecx
+; X86-NEXT:  .LBB0_12:
+; X86-NEXT:    subl %eax, %ecx
+; X86-NEXT:    movl %ecx, %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+; X86-NEXT:  .LBB0_38:
+; X86-NEXT:    shll $5, %eax
+; X86-NEXT:    popl %esi
+; X86-NEXT:    retl
+;
+; X64-HSW-LABEL: mult:
+; X64-HSW:       # BB#0:
+; X64-HSW-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
+; X64-HSW-NEXT:    cmpl $1, %esi
+; X64-HSW-NEXT:    movl $1, %ecx
+; X64-HSW-NEXT:    movl %esi, %eax
+; X64-HSW-NEXT:    cmovgl %ecx, %eax
+; X64-HSW-NEXT:    testl %esi, %esi
+; X64-HSW-NEXT:    cmovel %ecx, %eax
+; X64-HSW-NEXT:    addl $-1, %edi
+; X64-HSW-NEXT:    cmpl $31, %edi
+; X64-HSW-NEXT:    ja .LBB0_36
+; X64-HSW-NEXT:  # BB#1:
+; X64-HSW-NEXT:    jmpq *.LJTI0_0(,%rdi,8)
+; X64-HSW-NEXT:  .LBB0_2:
+; X64-HSW-NEXT:    addl %eax, %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_36:
+; X64-HSW-NEXT:    xorl %eax, %eax
+; X64-HSW-NEXT:  .LBB0_37:
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_3:
+; X64-HSW-NEXT:    leal (%rax,%rax,2), %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_4:
+; X64-HSW-NEXT:    shll $2, %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_5:
+; X64-HSW-NEXT:    leal (%rax,%rax,4), %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_6:
+; X64-HSW-NEXT:    addl %eax, %eax
+; X64-HSW-NEXT:    leal (%rax,%rax,2), %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_7:
+; X64-HSW-NEXT:    leal (,%rax,8), %ecx
+; X64-HSW-NEXT:    jmp .LBB0_8
+; X64-HSW-NEXT:  .LBB0_9:
+; X64-HSW-NEXT:    shll $3, %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_10:
+; X64-HSW-NEXT:    leal (%rax,%rax,8), %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_11:
+; X64-HSW-NEXT:    addl %eax, %eax
+; X64-HSW-NEXT:    leal (%rax,%rax,4), %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_12:
+; X64-HSW-NEXT:    leal (%rax,%rax,4), %ecx
+; X64-HSW-NEXT:    leal (%rax,%rcx,2), %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_13:
+; X64-HSW-NEXT:    shll $2, %eax
+; X64-HSW-NEXT:    leal (%rax,%rax,2), %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_14:
+; X64-HSW-NEXT:    leal (%rax,%rax,2), %ecx
+; X64-HSW-NEXT:    leal (%rax,%rcx,4), %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_15:
+; X64-HSW-NEXT:    leal (%rax,%rax,2), %ecx
+; X64-HSW-NEXT:    jmp .LBB0_16
+; X64-HSW-NEXT:  .LBB0_18:
+; X64-HSW-NEXT:    leal (%rax,%rax,4), %eax
+; X64-HSW-NEXT:    leal (%rax,%rax,2), %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_19:
+; X64-HSW-NEXT:    shll $4, %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_20:
+; X64-HSW-NEXT:    movl %eax, %ecx
+; X64-HSW-NEXT:    shll $4, %ecx
+; X64-HSW-NEXT:    jmp .LBB0_17
+; X64-HSW-NEXT:  .LBB0_21:
+; X64-HSW-NEXT:    addl %eax, %eax
+; X64-HSW-NEXT:    leal (%rax,%rax,8), %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_22:
+; X64-HSW-NEXT:    leal (%rax,%rax,4), %ecx
+; X64-HSW-NEXT:    shll $2, %ecx
+; X64-HSW-NEXT:    jmp .LBB0_8
+; X64-HSW-NEXT:  .LBB0_23:
+; X64-HSW-NEXT:    shll $2, %eax
+; X64-HSW-NEXT:    leal (%rax,%rax,4), %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_24:
+; X64-HSW-NEXT:    leal (%rax,%rax,4), %ecx
+; X64-HSW-NEXT:    leal (%rax,%rcx,4), %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_25:
+; X64-HSW-NEXT:    leal (%rax,%rax,4), %ecx
+; X64-HSW-NEXT:  .LBB0_16:
+; X64-HSW-NEXT:    leal (%rax,%rcx,4), %ecx
+; X64-HSW-NEXT:    jmp .LBB0_17
+; X64-HSW-NEXT:  .LBB0_26:
+; X64-HSW-NEXT:    leal (%rax,%rax,2), %ecx
+; X64-HSW-NEXT:    shll $3, %ecx
+; X64-HSW-NEXT:    jmp .LBB0_8
+; X64-HSW-NEXT:  .LBB0_27:
+; X64-HSW-NEXT:    shll $3, %eax
+; X64-HSW-NEXT:    leal (%rax,%rax,2), %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_28:
+; X64-HSW-NEXT:    leal (%rax,%rax,4), %eax
+; X64-HSW-NEXT:    leal (%rax,%rax,4), %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_29:
+; X64-HSW-NEXT:    leal (%rax,%rax,8), %ecx
+; X64-HSW-NEXT:    leal (%rcx,%rcx,2), %ecx
+; X64-HSW-NEXT:    jmp .LBB0_8
+; X64-HSW-NEXT:  .LBB0_30:
+; X64-HSW-NEXT:    leal (%rax,%rax,8), %eax
+; X64-HSW-NEXT:    leal (%rax,%rax,2), %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_31:
+; X64-HSW-NEXT:    leal (%rax,%rax,8), %ecx
+; X64-HSW-NEXT:    leal (%rcx,%rcx,2), %ecx
+; X64-HSW-NEXT:    jmp .LBB0_17
+; X64-HSW-NEXT:  .LBB0_32:
+; X64-HSW-NEXT:    leal (%rax,%rax,8), %ecx
+; X64-HSW-NEXT:    leal (%rcx,%rcx,2), %ecx
+; X64-HSW-NEXT:    addl %eax, %ecx
+; X64-HSW-NEXT:  .LBB0_17:
+; X64-HSW-NEXT:    addl %eax, %ecx
+; X64-HSW-NEXT:    movl %ecx, %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_33:
+; X64-HSW-NEXT:    movl %eax, %ecx
+; X64-HSW-NEXT:    shll $5, %ecx
+; X64-HSW-NEXT:    subl %eax, %ecx
+; X64-HSW-NEXT:    jmp .LBB0_8
+; X64-HSW-NEXT:  .LBB0_34:
+; X64-HSW-NEXT:    movl %eax, %ecx
+; X64-HSW-NEXT:    shll $5, %ecx
+; X64-HSW-NEXT:  .LBB0_8:
+; X64-HSW-NEXT:    subl %eax, %ecx
+; X64-HSW-NEXT:    movl %ecx, %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+; X64-HSW-NEXT:  .LBB0_35:
+; X64-HSW-NEXT:    shll $5, %eax
+; X64-HSW-NEXT:    # kill: %EAX<def> %EAX<kill> %RAX<kill>
+; X64-HSW-NEXT:    retq
+  %3 = icmp eq i32 %1, 0
+  %4 = icmp sgt i32 %1, 1
+  %5 = or i1 %3, %4
+  %6 = select i1 %5, i32 1, i32 %1
+  switch i32 %0, label %69 [
+    i32 1, label %70
+    i32 2, label %7
+    i32 3, label %9
+    i32 4, label %11
+    i32 5, label %13
+    i32 6, label %15
+    i32 7, label %17
+    i32 8, label %19
+    i32 9, label %21
+    i32 10, label %23
+    i32 11, label %25
+    i32 12, label %27
+    i32 13, label %29
+    i32 14, label %31
+    i32 15, label %33
+    i32 16, label %35
+    i32 17, label %37
+    i32 18, label %39
+    i32 19, label %41
+    i32 20, label %43
+    i32 21, label %45
+    i32 22, label %47
+    i32 23, label %49
+    i32 24, label %51
+    i32 25, label %53
+    i32 26, label %55
+    i32 27, label %57
+    i32 28, label %59
+    i32 29, label %61
+    i32 30, label %63
+    i32 31, label %65
+    i32 32, label %67
+  ]
+
+; <label>:7:                                      ; preds = %2
+  %8 = shl nsw i32 %6, 1
+  br label %70
+
+; <label>:9:                                      ; preds = %2
+  %10 = mul nsw i32 %6, 3
+  br label %70
+
+; <label>:11:                                     ; preds = %2
+  %12 = shl nsw i32 %6, 2
+  br label %70
+
+; <label>:13:                                     ; preds = %2
+  %14 = mul nsw i32 %6, 5
+  br label %70
+
+; <label>:15:                                     ; preds = %2
+  %16 = mul nsw i32 %6, 6
+  br label %70
+
+; <label>:17:                                     ; preds = %2
+  %18 = mul nsw i32 %6, 7
+  br label %70
+
+; <label>:19:                                     ; preds = %2
+  %20 = shl nsw i32 %6, 3
+  br label %70
+
+; <label>:21:                                     ; preds = %2
+  %22 = mul nsw i32 %6, 9
+  br label %70
+
+; <label>:23:                                     ; preds = %2
+  %24 = mul nsw i32 %6, 10
+  br label %70
+
+; <label>:25:                                     ; preds = %2
+  %26 = mul nsw i32 %6, 11
+  br label %70
+
+; <label>:27:                                     ; preds = %2
+  %28 = mul nsw i32 %6, 12
+  br label %70
+
+; <label>:29:                                     ; preds = %2
+  %30 = mul nsw i32 %6, 13
+  br label %70
+
+; <label>:31:                                     ; preds = %2
+  %32 = mul nsw i32 %6, 14
+  br label %70
+
+; <label>:33:                                     ; preds = %2
+  %34 = mul nsw i32 %6, 15
+  br label %70
+
+; <label>:35:                                     ; preds = %2
+  %36 = shl nsw i32 %6, 4
+  br label %70
+
+; <label>:37:                                     ; preds = %2
+  %38 = mul nsw i32 %6, 17
+  br label %70
+
+; <label>:39:                                     ; preds = %2
+  %40 = mul nsw i32 %6, 18
+  br label %70
+
+; <label>:41:                                     ; preds = %2
+  %42 = mul nsw i32 %6, 19
+  br label %70
+
+; <label>:43:                                     ; preds = %2
+  %44 = mul nsw i32 %6, 20
+  br label %70
+
+; <label>:45:                                     ; preds = %2
+  %46 = mul nsw i32 %6, 21
+  br label %70
+
+; <label>:47:                                     ; preds = %2
+  %48 = mul nsw i32 %6, 22
+  br label %70
+
+; <label>:49:                                     ; preds = %2
+  %50 = mul nsw i32 %6, 23
+  br label %70
+
+; <label>:51:                                     ; preds = %2
+  %52 = mul nsw i32 %6, 24
+  br label %70
+
+; <label>:53:                                     ; preds = %2
+  %54 = mul nsw i32 %6, 25
+  br label %70
+
+; <label>:55:                                     ; preds = %2
+  %56 = mul nsw i32 %6, 26
+  br label %70
+
+; <label>:57:                                     ; preds = %2
+  %58 = mul nsw i32 %6, 27
+  br label %70
+
+; <label>:59:                                     ; preds = %2
+  %60 = mul nsw i32 %6, 28
+  br label %70
+
+; <label>:61:                                     ; preds = %2
+  %62 = mul nsw i32 %6, 29
+  br label %70
+
+; <label>:63:                                     ; preds = %2
+  %64 = mul nsw i32 %6, 30
+  br label %70
+
+; <label>:65:                                     ; preds = %2
+  %66 = mul nsw i32 %6, 31
+  br label %70
+
+; <label>:67:                                     ; preds = %2
+  %68 = shl nsw i32 %6, 5
+  br label %70
+
+; <label>:69:                                     ; preds = %2
+  br label %70
+
+; <label>:70:                                     ; preds = %2, %69, %67, %65, %63, %61, %59, %57, %55, %53, %51, %49, %47, %45, %43, %41, %39, %37, %35, %33, %31, %29, %27, %25, %23, %21, %19, %17, %15, %13, %11, %9, %7
+  %71 = phi i32 [ %8, %7 ], [ %10, %9 ], [ %12, %11 ], [ %14, %13 ], [ %16, %15 ], [ %18, %17 ], [ %20, %19 ], [ %22, %21 ], [ %24, %23 ], [ %26, %25 ], [ %28, %27 ], [ %30, %29 ], [ %32, %31 ], [ %34, %33 ], [ %36, %35 ], [ %38, %37 ], [ %40, %39 ], [ %42, %41 ], [ %44, %43 ], [ %46, %45 ], [ %48, %47 ], [ %50, %49 ], [ %52, %51 ], [ %54, %53 ], [ %56, %55 ], [ %58, %57 ], [ %60, %59 ], [ %62, %61 ], [ %64, %63 ], [ %66, %65 ], [ %68, %67 ], [ 0, %69 ], [ %6, %2 ]
+  ret i32 %71
+}
+
+; Function Attrs: norecurse nounwind readnone uwtable
+define i32 @foo() local_unnamed_addr #0 {
+; X86-LABEL: foo:
+; X86:       # BB#0:
+; X86-NEXT:    pushl %ebx
+; X86-NEXT:  .Lcfi2:
+; X86-NEXT:    .cfi_def_cfa_offset 8
+; X86-NEXT:    pushl %edi
+; X86-NEXT:  .Lcfi3:
+; X86-NEXT:    .cfi_def_cfa_offset 12
+; X86-NEXT:    pushl %esi
+; X86-NEXT:  .Lcfi4:
+; X86-NEXT:    .cfi_def_cfa_offset 16
+; X86-NEXT:  .Lcfi5:
+; X86-NEXT:    .cfi_offset %esi, -16
+; X86-NEXT:  .Lcfi6:
+; X86-NEXT:    .cfi_offset %edi, -12
+; X86-NEXT:  .Lcfi7:
+; X86-NEXT:    .cfi_offset %ebx, -8
+; X86-NEXT:    pushl $0
+; X86-NEXT:  .Lcfi8:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $1
+; X86-NEXT:  .Lcfi9:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi10:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %esi
+; X86-NEXT:    xorl $1, %esi
+; X86-NEXT:    pushl $1
+; X86-NEXT:  .Lcfi11:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $2
+; X86-NEXT:  .Lcfi12:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi13:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl $2, %edi
+; X86-NEXT:    pushl $1
+; X86-NEXT:  .Lcfi14:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $3
+; X86-NEXT:  .Lcfi15:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi16:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl $3, %ebx
+; X86-NEXT:    orl %edi, %ebx
+; X86-NEXT:    pushl $2
+; X86-NEXT:  .Lcfi17:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $4
+; X86-NEXT:  .Lcfi18:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi19:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl $4, %edi
+; X86-NEXT:    orl %ebx, %edi
+; X86-NEXT:    pushl $2
+; X86-NEXT:  .Lcfi20:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $5
+; X86-NEXT:  .Lcfi21:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi22:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl $5, %ebx
+; X86-NEXT:    orl %edi, %ebx
+; X86-NEXT:    pushl $3
+; X86-NEXT:  .Lcfi23:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $6
+; X86-NEXT:  .Lcfi24:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi25:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl $6, %edi
+; X86-NEXT:    orl %ebx, %edi
+; X86-NEXT:    pushl $3
+; X86-NEXT:  .Lcfi26:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $7
+; X86-NEXT:  .Lcfi27:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi28:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl $7, %ebx
+; X86-NEXT:    orl %edi, %ebx
+; X86-NEXT:    pushl $4
+; X86-NEXT:  .Lcfi29:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $8
+; X86-NEXT:  .Lcfi30:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi31:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl $8, %edi
+; X86-NEXT:    orl %ebx, %edi
+; X86-NEXT:    pushl $4
+; X86-NEXT:  .Lcfi32:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $9
+; X86-NEXT:  .Lcfi33:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi34:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl $9, %ebx
+; X86-NEXT:    orl %edi, %ebx
+; X86-NEXT:    pushl $5
+; X86-NEXT:  .Lcfi35:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $10
+; X86-NEXT:  .Lcfi36:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi37:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl $10, %edi
+; X86-NEXT:    orl %ebx, %edi
+; X86-NEXT:    pushl $5
+; X86-NEXT:  .Lcfi38:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $11
+; X86-NEXT:  .Lcfi39:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi40:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl $11, %ebx
+; X86-NEXT:    orl %edi, %ebx
+; X86-NEXT:    pushl $6
+; X86-NEXT:  .Lcfi41:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $12
+; X86-NEXT:  .Lcfi42:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi43:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl $12, %edi
+; X86-NEXT:    orl %ebx, %edi
+; X86-NEXT:    pushl $6
+; X86-NEXT:  .Lcfi44:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $13
+; X86-NEXT:  .Lcfi45:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi46:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl $13, %ebx
+; X86-NEXT:    orl %edi, %ebx
+; X86-NEXT:    pushl $7
+; X86-NEXT:  .Lcfi47:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $14
+; X86-NEXT:  .Lcfi48:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi49:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl $14, %edi
+; X86-NEXT:    orl %ebx, %edi
+; X86-NEXT:    pushl $7
+; X86-NEXT:  .Lcfi50:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $15
+; X86-NEXT:  .Lcfi51:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi52:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl $15, %ebx
+; X86-NEXT:    orl %edi, %ebx
+; X86-NEXT:    pushl $8
+; X86-NEXT:  .Lcfi53:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $16
+; X86-NEXT:  .Lcfi54:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi55:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl $16, %edi
+; X86-NEXT:    orl %ebx, %edi
+; X86-NEXT:    pushl $8
+; X86-NEXT:  .Lcfi56:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $17
+; X86-NEXT:  .Lcfi57:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi58:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl $17, %ebx
+; X86-NEXT:    orl %edi, %ebx
+; X86-NEXT:    pushl $9
+; X86-NEXT:  .Lcfi59:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $18
+; X86-NEXT:  .Lcfi60:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi61:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl $18, %edi
+; X86-NEXT:    orl %ebx, %edi
+; X86-NEXT:    pushl $9
+; X86-NEXT:  .Lcfi62:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $19
+; X86-NEXT:  .Lcfi63:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi64:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl $19, %ebx
+; X86-NEXT:    orl %edi, %ebx
+; X86-NEXT:    pushl $10
+; X86-NEXT:  .Lcfi65:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $20
+; X86-NEXT:  .Lcfi66:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi67:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl $20, %edi
+; X86-NEXT:    orl %ebx, %edi
+; X86-NEXT:    pushl $10
+; X86-NEXT:  .Lcfi68:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $21
+; X86-NEXT:  .Lcfi69:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi70:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl $21, %ebx
+; X86-NEXT:    orl %edi, %ebx
+; X86-NEXT:    pushl $11
+; X86-NEXT:  .Lcfi71:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $22
+; X86-NEXT:  .Lcfi72:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi73:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl $22, %edi
+; X86-NEXT:    orl %ebx, %edi
+; X86-NEXT:    pushl $11
+; X86-NEXT:  .Lcfi74:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $23
+; X86-NEXT:  .Lcfi75:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi76:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl $23, %ebx
+; X86-NEXT:    orl %edi, %ebx
+; X86-NEXT:    pushl $12
+; X86-NEXT:  .Lcfi77:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $24
+; X86-NEXT:  .Lcfi78:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi79:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl $24, %edi
+; X86-NEXT:    orl %ebx, %edi
+; X86-NEXT:    pushl $12
+; X86-NEXT:  .Lcfi80:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $25
+; X86-NEXT:  .Lcfi81:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi82:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl $25, %ebx
+; X86-NEXT:    orl %edi, %ebx
+; X86-NEXT:    pushl $13
+; X86-NEXT:  .Lcfi83:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $26
+; X86-NEXT:  .Lcfi84:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi85:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl $26, %edi
+; X86-NEXT:    orl %ebx, %edi
+; X86-NEXT:    pushl $13
+; X86-NEXT:  .Lcfi86:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $27
+; X86-NEXT:  .Lcfi87:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi88:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl $27, %ebx
+; X86-NEXT:    orl %edi, %ebx
+; X86-NEXT:    pushl $14
+; X86-NEXT:  .Lcfi89:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $28
+; X86-NEXT:  .Lcfi90:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi91:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl $28, %edi
+; X86-NEXT:    orl %ebx, %edi
+; X86-NEXT:    pushl $14
+; X86-NEXT:  .Lcfi92:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $29
+; X86-NEXT:  .Lcfi93:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi94:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl $29, %ebx
+; X86-NEXT:    orl %edi, %ebx
+; X86-NEXT:    pushl $15
+; X86-NEXT:  .Lcfi95:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $30
+; X86-NEXT:  .Lcfi96:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi97:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    xorl $30, %edi
+; X86-NEXT:    orl %ebx, %edi
+; X86-NEXT:    pushl $15
+; X86-NEXT:  .Lcfi98:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $31
+; X86-NEXT:  .Lcfi99:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi100:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    xorl $31, %ebx
+; X86-NEXT:    orl %edi, %ebx
+; X86-NEXT:    orl %esi, %ebx
+; X86-NEXT:    pushl $16
+; X86-NEXT:  .Lcfi101:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    pushl $32
+; X86-NEXT:  .Lcfi102:
+; X86-NEXT:    .cfi_adjust_cfa_offset 4
+; X86-NEXT:    calll mult
+; X86-NEXT:    addl $8, %esp
+; X86-NEXT:  .Lcfi103:
+; X86-NEXT:    .cfi_adjust_cfa_offset -8
+; X86-NEXT:    xorl $32, %eax
+; X86-NEXT:    orl %ebx, %eax
+; X86-NEXT:    movl $-1, %eax
+; X86-NEXT:    jne .LBB1_2
+; X86-NEXT:  # BB#1:
+; X86-NEXT:    xorl %eax, %eax
+; X86-NEXT:  .LBB1_2:
+; X86-NEXT:    popl %esi
+; X86-NEXT:    popl %edi
+; X86-NEXT:    popl %ebx
+; X86-NEXT:    retl
+;
+; X64-HSW-LABEL: foo:
+; X64-HSW:       # BB#0:
+; X64-HSW-NEXT:    pushq %rbp
+; X64-HSW-NEXT:  .Lcfi0:
+; X64-HSW-NEXT:    .cfi_def_cfa_offset 16
+; X64-HSW-NEXT:    pushq %r15
+; X64-HSW-NEXT:  .Lcfi1:
+; X64-HSW-NEXT:    .cfi_def_cfa_offset 24
+; X64-HSW-NEXT:    pushq %r14
+; X64-HSW-NEXT:  .Lcfi2:
+; X64-HSW-NEXT:    .cfi_def_cfa_offset 32
+; X64-HSW-NEXT:    pushq %r12
+; X64-HSW-NEXT:  .Lcfi3:
+; X64-HSW-NEXT:    .cfi_def_cfa_offset 40
+; X64-HSW-NEXT:    pushq %rbx
+; X64-HSW-NEXT:  .Lcfi4:
+; X64-HSW-NEXT:    .cfi_def_cfa_offset 48
+; X64-HSW-NEXT:  .Lcfi5:
+; X64-HSW-NEXT:    .cfi_offset %rbx, -48
+; X64-HSW-NEXT:  .Lcfi6:
+; X64-HSW-NEXT:    .cfi_offset %r12, -40
+; X64-HSW-NEXT:  .Lcfi7:
+; X64-HSW-NEXT:    .cfi_offset %r14, -32
+; X64-HSW-NEXT:  .Lcfi8:
+; X64-HSW-NEXT:    .cfi_offset %r15, -24
+; X64-HSW-NEXT:  .Lcfi9:
+; X64-HSW-NEXT:    .cfi_offset %rbp, -16
+; X64-HSW-NEXT:    xorl %r12d, %r12d
+; X64-HSW-NEXT:    movl $1, %edi
+; X64-HSW-NEXT:    xorl %esi, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebx
+; X64-HSW-NEXT:    xorl $1, %ebx
+; X64-HSW-NEXT:    movl $2, %edi
+; X64-HSW-NEXT:    movl $1, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebp
+; X64-HSW-NEXT:    xorl $2, %ebp
+; X64-HSW-NEXT:    orl %ebx, %ebp
+; X64-HSW-NEXT:    movl $3, %edi
+; X64-HSW-NEXT:    movl $1, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %r14d
+; X64-HSW-NEXT:    xorl $3, %r14d
+; X64-HSW-NEXT:    movl $4, %edi
+; X64-HSW-NEXT:    movl $2, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebx
+; X64-HSW-NEXT:    xorl $4, %ebx
+; X64-HSW-NEXT:    orl %r14d, %ebx
+; X64-HSW-NEXT:    orl %ebp, %ebx
+; X64-HSW-NEXT:    movl $5, %edi
+; X64-HSW-NEXT:    movl $2, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %r14d
+; X64-HSW-NEXT:    xorl $5, %r14d
+; X64-HSW-NEXT:    movl $6, %edi
+; X64-HSW-NEXT:    movl $3, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebp
+; X64-HSW-NEXT:    xorl $6, %ebp
+; X64-HSW-NEXT:    orl %r14d, %ebp
+; X64-HSW-NEXT:    movl $7, %edi
+; X64-HSW-NEXT:    movl $3, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %r14d
+; X64-HSW-NEXT:    xorl $7, %r14d
+; X64-HSW-NEXT:    orl %ebp, %r14d
+; X64-HSW-NEXT:    orl %ebx, %r14d
+; X64-HSW-NEXT:    movl $8, %edi
+; X64-HSW-NEXT:    movl $4, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebx
+; X64-HSW-NEXT:    xorl $8, %ebx
+; X64-HSW-NEXT:    movl $9, %edi
+; X64-HSW-NEXT:    movl $4, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebp
+; X64-HSW-NEXT:    xorl $9, %ebp
+; X64-HSW-NEXT:    orl %ebx, %ebp
+; X64-HSW-NEXT:    movl $10, %edi
+; X64-HSW-NEXT:    movl $5, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebx
+; X64-HSW-NEXT:    xorl $10, %ebx
+; X64-HSW-NEXT:    orl %ebp, %ebx
+; X64-HSW-NEXT:    movl $11, %edi
+; X64-HSW-NEXT:    movl $5, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %r15d
+; X64-HSW-NEXT:    xorl $11, %r15d
+; X64-HSW-NEXT:    orl %ebx, %r15d
+; X64-HSW-NEXT:    orl %r14d, %r15d
+; X64-HSW-NEXT:    movl $12, %edi
+; X64-HSW-NEXT:    movl $6, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebx
+; X64-HSW-NEXT:    xorl $12, %ebx
+; X64-HSW-NEXT:    movl $13, %edi
+; X64-HSW-NEXT:    movl $6, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebp
+; X64-HSW-NEXT:    xorl $13, %ebp
+; X64-HSW-NEXT:    orl %ebx, %ebp
+; X64-HSW-NEXT:    movl $14, %edi
+; X64-HSW-NEXT:    movl $7, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebx
+; X64-HSW-NEXT:    xorl $14, %ebx
+; X64-HSW-NEXT:    orl %ebp, %ebx
+; X64-HSW-NEXT:    movl $15, %edi
+; X64-HSW-NEXT:    movl $7, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebp
+; X64-HSW-NEXT:    xorl $15, %ebp
+; X64-HSW-NEXT:    orl %ebx, %ebp
+; X64-HSW-NEXT:    movl $16, %edi
+; X64-HSW-NEXT:    movl $8, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %r14d
+; X64-HSW-NEXT:    xorl $16, %r14d
+; X64-HSW-NEXT:    orl %ebp, %r14d
+; X64-HSW-NEXT:    orl %r15d, %r14d
+; X64-HSW-NEXT:    movl $17, %edi
+; X64-HSW-NEXT:    movl $8, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebp
+; X64-HSW-NEXT:    xorl $17, %ebp
+; X64-HSW-NEXT:    movl $18, %edi
+; X64-HSW-NEXT:    movl $9, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebx
+; X64-HSW-NEXT:    xorl $18, %ebx
+; X64-HSW-NEXT:    orl %ebp, %ebx
+; X64-HSW-NEXT:    movl $19, %edi
+; X64-HSW-NEXT:    movl $9, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebp
+; X64-HSW-NEXT:    xorl $19, %ebp
+; X64-HSW-NEXT:    orl %ebx, %ebp
+; X64-HSW-NEXT:    movl $20, %edi
+; X64-HSW-NEXT:    movl $10, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebx
+; X64-HSW-NEXT:    xorl $20, %ebx
+; X64-HSW-NEXT:    orl %ebp, %ebx
+; X64-HSW-NEXT:    movl $21, %edi
+; X64-HSW-NEXT:    movl $10, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebp
+; X64-HSW-NEXT:    xorl $21, %ebp
+; X64-HSW-NEXT:    orl %ebx, %ebp
+; X64-HSW-NEXT:    movl $22, %edi
+; X64-HSW-NEXT:    movl $11, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %r15d
+; X64-HSW-NEXT:    xorl $22, %r15d
+; X64-HSW-NEXT:    orl %ebp, %r15d
+; X64-HSW-NEXT:    orl %r14d, %r15d
+; X64-HSW-NEXT:    movl $23, %edi
+; X64-HSW-NEXT:    movl $11, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebp
+; X64-HSW-NEXT:    xorl $23, %ebp
+; X64-HSW-NEXT:    movl $24, %edi
+; X64-HSW-NEXT:    movl $12, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebx
+; X64-HSW-NEXT:    xorl $24, %ebx
+; X64-HSW-NEXT:    orl %ebp, %ebx
+; X64-HSW-NEXT:    movl $25, %edi
+; X64-HSW-NEXT:    movl $12, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebp
+; X64-HSW-NEXT:    xorl $25, %ebp
+; X64-HSW-NEXT:    orl %ebx, %ebp
+; X64-HSW-NEXT:    movl $26, %edi
+; X64-HSW-NEXT:    movl $13, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebx
+; X64-HSW-NEXT:    xorl $26, %ebx
+; X64-HSW-NEXT:    orl %ebp, %ebx
+; X64-HSW-NEXT:    movl $27, %edi
+; X64-HSW-NEXT:    movl $13, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebp
+; X64-HSW-NEXT:    xorl $27, %ebp
+; X64-HSW-NEXT:    orl %ebx, %ebp
+; X64-HSW-NEXT:    movl $28, %edi
+; X64-HSW-NEXT:    movl $14, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebx
+; X64-HSW-NEXT:    xorl $28, %ebx
+; X64-HSW-NEXT:    orl %ebp, %ebx
+; X64-HSW-NEXT:    movl $29, %edi
+; X64-HSW-NEXT:    movl $14, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebp
+; X64-HSW-NEXT:    xorl $29, %ebp
+; X64-HSW-NEXT:    orl %ebx, %ebp
+; X64-HSW-NEXT:    orl %r15d, %ebp
+; X64-HSW-NEXT:    movl $30, %edi
+; X64-HSW-NEXT:    movl $15, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %r14d
+; X64-HSW-NEXT:    xorl $30, %r14d
+; X64-HSW-NEXT:    movl $31, %edi
+; X64-HSW-NEXT:    movl $15, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    movl %eax, %ebx
+; X64-HSW-NEXT:    xorl $31, %ebx
+; X64-HSW-NEXT:    orl %r14d, %ebx
+; X64-HSW-NEXT:    orl %ebp, %ebx
+; X64-HSW-NEXT:    movl $32, %edi
+; X64-HSW-NEXT:    movl $16, %esi
+; X64-HSW-NEXT:    callq mult
+; X64-HSW-NEXT:    xorl $32, %eax
+; X64-HSW-NEXT:    orl %ebx, %eax
+; X64-HSW-NEXT:    movl $-1, %eax
+; X64-HSW-NEXT:    cmovel %r12d, %eax
+; X64-HSW-NEXT:    popq %rbx
+; X64-HSW-NEXT:    popq %r12
+; X64-HSW-NEXT:    popq %r14
+; X64-HSW-NEXT:    popq %r15
+; X64-HSW-NEXT:    popq %rbp
+; X64-HSW-NEXT:    retq
+  %1 = tail call i32 @mult(i32 1, i32 0)
+  %2 = icmp ne i32 %1, 1
+  %3 = tail call i32 @mult(i32 2, i32 1)
+  %4 = icmp ne i32 %3, 2
+  %5 = or i1 %2, %4
+  %6 = tail call i32 @mult(i32 3, i32 1)
+  %7 = icmp ne i32 %6, 3
+  %8 = or i1 %5, %7
+  %9 = tail call i32 @mult(i32 4, i32 2)
+  %10 = icmp ne i32 %9, 4
+  %11 = or i1 %8, %10
+  %12 = tail call i32 @mult(i32 5, i32 2)
+  %13 = icmp ne i32 %12, 5
+  %14 = or i1 %11, %13
+  %15 = tail call i32 @mult(i32 6, i32 3)
+  %16 = icmp ne i32 %15, 6
+  %17 = or i1 %14, %16
+  %18 = tail call i32 @mult(i32 7, i32 3)
+  %19 = icmp ne i32 %18, 7
+  %20 = or i1 %17, %19
+  %21 = tail call i32 @mult(i32 8, i32 4)
+  %22 = icmp ne i32 %21, 8
+  %23 = or i1 %20, %22
+  %24 = tail call i32 @mult(i32 9, i32 4)
+  %25 = icmp ne i32 %24, 9
+  %26 = or i1 %23, %25
+  %27 = tail call i32 @mult(i32 10, i32 5)
+  %28 = icmp ne i32 %27, 10
+  %29 = or i1 %26, %28
+  %30 = tail call i32 @mult(i32 11, i32 5)
+  %31 = icmp ne i32 %30, 11
+  %32 = or i1 %29, %31
+  %33 = tail call i32 @mult(i32 12, i32 6)
+  %34 = icmp ne i32 %33, 12
+  %35 = or i1 %32, %34
+  %36 = tail call i32 @mult(i32 13, i32 6)
+  %37 = icmp ne i32 %36, 13
+  %38 = or i1 %35, %37
+  %39 = tail call i32 @mult(i32 14, i32 7)
+  %40 = icmp ne i32 %39, 14
+  %41 = or i1 %38, %40
+  %42 = tail call i32 @mult(i32 15, i32 7)
+  %43 = icmp ne i32 %42, 15
+  %44 = or i1 %41, %43
+  %45 = tail call i32 @mult(i32 16, i32 8)
+  %46 = icmp ne i32 %45, 16
+  %47 = or i1 %44, %46
+  %48 = tail call i32 @mult(i32 17, i32 8)
+  %49 = icmp ne i32 %48, 17
+  %50 = or i1 %47, %49
+  %51 = tail call i32 @mult(i32 18, i32 9)
+  %52 = icmp ne i32 %51, 18
+  %53 = or i1 %50, %52
+  %54 = tail call i32 @mult(i32 19, i32 9)
+  %55 = icmp ne i32 %54, 19
+  %56 = or i1 %53, %55
+  %57 = tail call i32 @mult(i32 20, i32 10)
+  %58 = icmp ne i32 %57, 20
+  %59 = or i1 %56, %58
+  %60 = tail call i32 @mult(i32 21, i32 10)
+  %61 = icmp ne i32 %60, 21
+  %62 = or i1 %59, %61
+  %63 = tail call i32 @mult(i32 22, i32 11)
+  %64 = icmp ne i32 %63, 22
+  %65 = or i1 %62, %64
+  %66 = tail call i32 @mult(i32 23, i32 11)
+  %67 = icmp ne i32 %66, 23
+  %68 = or i1 %65, %67
+  %69 = tail call i32 @mult(i32 24, i32 12)
+  %70 = icmp ne i32 %69, 24
+  %71 = or i1 %68, %70
+  %72 = tail call i32 @mult(i32 25, i32 12)
+  %73 = icmp ne i32 %72, 25
+  %74 = or i1 %71, %73
+  %75 = tail call i32 @mult(i32 26, i32 13)
+  %76 = icmp ne i32 %75, 26
+  %77 = or i1 %74, %76
+  %78 = tail call i32 @mult(i32 27, i32 13)
+  %79 = icmp ne i32 %78, 27
+  %80 = or i1 %77, %79
+  %81 = tail call i32 @mult(i32 28, i32 14)
+  %82 = icmp ne i32 %81, 28
+  %83 = or i1 %80, %82
+  %84 = tail call i32 @mult(i32 29, i32 14)
+  %85 = icmp ne i32 %84, 29
+  %86 = or i1 %83, %85
+  %87 = tail call i32 @mult(i32 30, i32 15)
+  %88 = icmp ne i32 %87, 30
+  %89 = or i1 %86, %88
+  %90 = tail call i32 @mult(i32 31, i32 15)
+  %91 = icmp ne i32 %90, 31
+  %92 = or i1 %89, %91
+  %93 = tail call i32 @mult(i32 32, i32 16)
+  %94 = icmp ne i32 %93, 32
+  %95 = or i1 %92, %94
+  %96 = sext i1 %95 to i32
+  ret i32 %96
+}
+
+attributes #0 = { norecurse nounwind readnone uwtable  }